Bastin et al. propose an integrative framework for episodic memory based on interactions between two representational subsystems and an attribution system. They suggest that an “entity representation core system” supports familiarity, a “relational representation core system” is recruited for recollection, and an “attribution system” determines subjective memory qualities (target article, sect. 3). I agree with many aspects of their proposal. Describing memory in terms of the representational formats in which prior experiences are reinstated, and linking these formats to specific brain structures and distinct subjective qualities, is a fruitful and innovative approach. The authors nicely describe that the common distinction between representational and process-based memory theories is not as clear-cut as it may seem, since specific types of representations lend themselves more naturally to specific computational processes.
Despite my overall agreement, I would like to make two critical comments. First, I believe the proposed framework could emphasize more the impact of memory representations in neocortical areas beyond the medial temporal lobe. Second, I am not convinced that representational systems only determine the content of memories whereas an attribution system defines their subjective quality. Instead, I suggest that the subjective qualities of memory can also be explained by specific representational formats, rather than by a separate system for metacognitive judgments.
With regard to the first aspect, Bastin et al. strongly focus their proposal on the contribution of the hippocampus and the perirhinal cortex. As demonstrated by decades of neuropsychological research, these structures are indeed central and indispensable for memory functioning. Nevertheless, the representational features of memories as well as their subjective properties (see also my second point below) are determined not only by properties of medial temporal areas, but crucially also by the representational formats in specific neocortical areas which they control during retrieval. The authors do acknowledge the role of neocortical regions such as the occipitotemporal and posterior parietal cortex for the entity representation core system and for integrative functions; however, they do not attempt to explain properties of memory representations by the representational features of these neocortical areas. In fact, it is unlikely that the full content of memories resides in the distribution of synaptic weights in medial temporal areas. Instead, as acknowledged by the authors, these regions act as “pointers” to neocortical areas (Pacheco et al. Reference Pacheco, Sánchez-Fibla, Duff, Principe, Rocamora, Zhang, Axmacher and Verschure2019; Teyler & Rudy Reference Teyler and Rudy2007).
I would like to give four examples where neocortical areas are relevant for understanding memory representations. First, episodic memories are characterized by either an egocentric or an allocentric visual perspective (Conway & Pleydell-Pearce Reference Conway and Pleydell-Pearce2000; Nigro & Neisser Reference Nigro and Neisser1983). These perspectives are putatively determined by egocentric representations in posterior parietal cortex (e.g., Byrne et al. Reference Byrne, Becker and Burgess2007) versus perspective-invariant (allocentric) representations in the temporal neocortex. Second, memory representations contain varying degrees of perceptual detail versus general semantic information, and in extreme cases may only consist of schematic event knowledge. The perceptual versus conceptual representational format of a memory likely maps onto neocortical processing steps (with schema representations depending strongly on prefrontal regions). Third, episodic memories rely on the construction of mental scenarios (Barry & Maguire Reference Barry and Maguire2019; Cheng et al. Reference Cheng, Werning and Suddendorf2016) whose level of detail can be flexibly adapted to fit situational demands. Their different degrees of representational detail may reflect the different processing steps in sensory areas as well as contributions from semantic expectations and schemas, which all rely on neocortical areas. Finally, autobiographical episodic memories in healthy subjects can, if they are sufficiently relevant, become central constituents of personal narratives, that is, of the stories that people tell about themselves (e.g., Renoult et al. Reference Renoult, Davidson, Palombo, Moscovitch and Levine2012). By contrast, intrusions and flashbacks in posttraumatic stress disorder patients are characterized by a lack of semantic processing and narrative integration; therapeutic concepts such as narrative exposure therapy attempt to transform these pathological representational formats into more context-dependent and semantically integrated ones (Schauer et al. Reference Schauer, Neuner and Elbert2011).
Now, egocentric versus allocentric visual perspectives, perceptual versus conceptual representations, flexible representations, and the amount of narrative integration are just some examples for those representational features of memories which can only be understood by taking into account the representational properties of neocortical areas – in particular, in ventral and dorsal visual streams and in the corresponding networks in other sensory modalities. This is emphasized in conceptual frameworks such as Brewin's “dual representation theory” (Brewin et al. Reference Brewin, Dalgleish and Joseph1996). These frameworks still assume a central role of medial temporal lobe structures (including the amygdala, which is not mentioned in the Bastin et al.’s proposal) for controlling these representations, whose properties are nevertheless essentially determined by the specific computations in neocortical areas.
With regard to my second point, I believe the authors underestimate the influence of representational properties on the subjective features of memories. Although I agree that fluency signals and context-dependent expectations may shape how we subjectively experience a reinstated memory representation, the subjective, or phenomenological, features of episodic memory are primarily determined by the specific representational formats of the memorized events. In addition to the examples described above, feelings of familiarity and recollection can also be best explained by the representational properties of an event – that is, whether individual items or relational information is being remembered. In fact, Bastin et al. acknowledge that these types of representations are predominantly associated with feelings of familiarity or recollections; however, as there are some cases where these feelings dissociate from their typical contents, they propose to explain the feelings by metacognitive mechanisms rather than representational features. I would actually find it more parsimonious to embrace the general heuristic value of representational formats for explaining subjective memory experiences and conceptualize possible dissociations as atypical effects of representations for which the overall system was not designed. In fact, it is not even clear whether feelings of familiarity that occur in atypical conditions – and in particular in patients with neurological disorders – are the same as those under common circumstances. In other words, if feelings of familiarity or recollection do not match with the typical representational format of the corresponding memories, are these feelings really the same as in more common cases?
Bastin et al. propose an integrative framework for episodic memory based on interactions between two representational subsystems and an attribution system. They suggest that an “entity representation core system” supports familiarity, a “relational representation core system” is recruited for recollection, and an “attribution system” determines subjective memory qualities (target article, sect. 3). I agree with many aspects of their proposal. Describing memory in terms of the representational formats in which prior experiences are reinstated, and linking these formats to specific brain structures and distinct subjective qualities, is a fruitful and innovative approach. The authors nicely describe that the common distinction between representational and process-based memory theories is not as clear-cut as it may seem, since specific types of representations lend themselves more naturally to specific computational processes.
Despite my overall agreement, I would like to make two critical comments. First, I believe the proposed framework could emphasize more the impact of memory representations in neocortical areas beyond the medial temporal lobe. Second, I am not convinced that representational systems only determine the content of memories whereas an attribution system defines their subjective quality. Instead, I suggest that the subjective qualities of memory can also be explained by specific representational formats, rather than by a separate system for metacognitive judgments.
With regard to the first aspect, Bastin et al. strongly focus their proposal on the contribution of the hippocampus and the perirhinal cortex. As demonstrated by decades of neuropsychological research, these structures are indeed central and indispensable for memory functioning. Nevertheless, the representational features of memories as well as their subjective properties (see also my second point below) are determined not only by properties of medial temporal areas, but crucially also by the representational formats in specific neocortical areas which they control during retrieval. The authors do acknowledge the role of neocortical regions such as the occipitotemporal and posterior parietal cortex for the entity representation core system and for integrative functions; however, they do not attempt to explain properties of memory representations by the representational features of these neocortical areas. In fact, it is unlikely that the full content of memories resides in the distribution of synaptic weights in medial temporal areas. Instead, as acknowledged by the authors, these regions act as “pointers” to neocortical areas (Pacheco et al. Reference Pacheco, Sánchez-Fibla, Duff, Principe, Rocamora, Zhang, Axmacher and Verschure2019; Teyler & Rudy Reference Teyler and Rudy2007).
I would like to give four examples where neocortical areas are relevant for understanding memory representations. First, episodic memories are characterized by either an egocentric or an allocentric visual perspective (Conway & Pleydell-Pearce Reference Conway and Pleydell-Pearce2000; Nigro & Neisser Reference Nigro and Neisser1983). These perspectives are putatively determined by egocentric representations in posterior parietal cortex (e.g., Byrne et al. Reference Byrne, Becker and Burgess2007) versus perspective-invariant (allocentric) representations in the temporal neocortex. Second, memory representations contain varying degrees of perceptual detail versus general semantic information, and in extreme cases may only consist of schematic event knowledge. The perceptual versus conceptual representational format of a memory likely maps onto neocortical processing steps (with schema representations depending strongly on prefrontal regions). Third, episodic memories rely on the construction of mental scenarios (Barry & Maguire Reference Barry and Maguire2019; Cheng et al. Reference Cheng, Werning and Suddendorf2016) whose level of detail can be flexibly adapted to fit situational demands. Their different degrees of representational detail may reflect the different processing steps in sensory areas as well as contributions from semantic expectations and schemas, which all rely on neocortical areas. Finally, autobiographical episodic memories in healthy subjects can, if they are sufficiently relevant, become central constituents of personal narratives, that is, of the stories that people tell about themselves (e.g., Renoult et al. Reference Renoult, Davidson, Palombo, Moscovitch and Levine2012). By contrast, intrusions and flashbacks in posttraumatic stress disorder patients are characterized by a lack of semantic processing and narrative integration; therapeutic concepts such as narrative exposure therapy attempt to transform these pathological representational formats into more context-dependent and semantically integrated ones (Schauer et al. Reference Schauer, Neuner and Elbert2011).
Now, egocentric versus allocentric visual perspectives, perceptual versus conceptual representations, flexible representations, and the amount of narrative integration are just some examples for those representational features of memories which can only be understood by taking into account the representational properties of neocortical areas – in particular, in ventral and dorsal visual streams and in the corresponding networks in other sensory modalities. This is emphasized in conceptual frameworks such as Brewin's “dual representation theory” (Brewin et al. Reference Brewin, Dalgleish and Joseph1996). These frameworks still assume a central role of medial temporal lobe structures (including the amygdala, which is not mentioned in the Bastin et al.’s proposal) for controlling these representations, whose properties are nevertheless essentially determined by the specific computations in neocortical areas.
With regard to my second point, I believe the authors underestimate the influence of representational properties on the subjective features of memories. Although I agree that fluency signals and context-dependent expectations may shape how we subjectively experience a reinstated memory representation, the subjective, or phenomenological, features of episodic memory are primarily determined by the specific representational formats of the memorized events. In addition to the examples described above, feelings of familiarity and recollection can also be best explained by the representational properties of an event – that is, whether individual items or relational information is being remembered. In fact, Bastin et al. acknowledge that these types of representations are predominantly associated with feelings of familiarity or recollections; however, as there are some cases where these feelings dissociate from their typical contents, they propose to explain the feelings by metacognitive mechanisms rather than representational features. I would actually find it more parsimonious to embrace the general heuristic value of representational formats for explaining subjective memory experiences and conceptualize possible dissociations as atypical effects of representations for which the overall system was not designed. In fact, it is not even clear whether feelings of familiarity that occur in atypical conditions – and in particular in patients with neurological disorders – are the same as those under common circumstances. In other words, if feelings of familiarity or recollection do not match with the typical representational format of the corresponding memories, are these feelings really the same as in more common cases?