Bastin et al.’s integrative memory model proposes that subjective memory experiences result from the interaction between core representational systems and an attribution system. The relevant core representational system in the case of familiarity concerns “entity” processing, the implementation of which involves perirhinal cortex, a region regarded to be at the apex of visual object processing. The core representation system for recollection concerns “relational” processing, and includes hippocampus, posterior cingulate, and retrosplenial cortex, which together are believed to coordinate the reactivation of stimulus- and context-specific representations distributed through cortex. So far, this picture is broadly consistent with our impressions of the cognitive, neuropsychological, and neuroimaging literatures on recollection and familiarity.
The “attribution system” is more nebulous, and perhaps more controversial. The term “attribution” implies that in addition to mnemonic signals, non-mnemonic signals are sometimes interpreted as having arisen due to prior exposure to the test item. In general, this could be a sensible heuristic, since many non-mnemonic effects, be they perceptual, emotional, aesthetic, and so forth, are known to result from prior exposure to an item, even sometimes in the absence of explicit memory for that prior exposure (e.g., Hamann & Squire Reference Hamann and Squire1997; Oppenheimer Reference Oppenheimer2008; Schwarz & Winkielman Reference Schwarz and Winkielman2004), so the attribution of such signals to memory is effectively a reverse inference.
In the case of familiarity, it is well established that the fluency with which a test item is processed can be (mis-)attributed to familiarity due to past exposure to the item. By experimentally manipulating the fluency with which a test item is processed (e.g., via masked repetition priming), researchers can induce a familiarity-like feeling which is then attributed to memory. This so-called Jacoby–Whitehouse illusion (Jacoby & Whitehouse Reference Jacoby and Whitehouse1989) has further been found to be specific to familiarity (Rajaram Reference Rajaram1993); a finding that has been replicated many times, including in several studies by one of the present authors (Li et al. Reference Li, Taylor, Wang, Gao and Guo2017; Taylor & Henson Reference Taylor and Henson2012a; Taylor et al. Reference Taylor, Buratto and Henson2013; Woollams et al. Reference Woollams, Taylor, Karayanidis and Henson2008). We therefore agree that familiarity judgements can be based on the attribution of a non-mnemonic signal (i.e., processing fluency) to memory, as stated in the model.
The integrative memory model further proposes that recollection-based memory judgements are also made based on the interaction between its core system and an attribution system. It is implied that the participant will experience a subjective feeling of recollection if the attribution system assesses the memory trace reactivated to be relevant and sufficient (in terms of amount of information retrieved). However, by the authors’ description of the core representational system subserving recollection, the relevant signals for recollection judgements are all memory signals, and therefore, we do not see why an attribution system needs to be posited to intervene between the core representational system and the decision module.
In several studies, we have found that masked conceptual primes (rather than repetition or lexically associated primes) presented before items in a recognition memory test increase correct recollection responses (Li et al. Reference Li, Taylor, Wang, Gao and Guo2017; Taylor & Henson Reference Taylor and Henson2012a; Taylor et al. Reference Taylor, Buratto and Henson2013). However, we do not agree that this constitutes evidence for an attribution system mediating recollection decisions. This is because one hallmark of the attribution framework is that it sometimes misfires, resulting in increased false alarms (primed unstudied items being endorsed as old more often than unprimed unstudied items). But in our studies, only correct recollection responses are increased by conceptual primes; false alarm recollection responses are not increased, and therefore, mis-attribution does not reliably occur. Instead, conceptual priming appears to facilitate veridical retrieval of the encoding event. This is consistent with the notion that the decision module that subserves a recollection judgement relies only on mnemonic signals and therefore, contrary to the target model, it does not require mediation by an attribution system.
It is noteworthy that the evidence for the attribution system comes exclusively from studies using single words as stimuli, whereas much of the rest of the model is mainly based on studies that use objects, pairs of objects, pairs of words, and/or visually rich scenes or environments. One crucial difference between these types of stimuli is that the very same word stimuli are commonly encountered outside of the laboratory, whereas object and scene images and their relations are often (though admittedly not always) unique to the experiment. A second crucial difference is that, at least in the priming and recognition memory studies discussed above, words are often presented in visually impoverished encoding conditions (e.g., one at a time on a blank screen), whereas in studies investigating relational memory, object images are often presented in trial-unique contexts, in pairs, or in well-defined locations in an environment. Thus, it is likely that recollection decisions in word-list memory experiments rely heavily on retrieval of contextual elements from the encoding episode (since the stimuli themselves are not distinctive), and these contextual elements are likely to be internal (e.g., what the participant thought of when they read the word in the study phase) rather than external (e.g., where in the environment, or with which other object or context the target object was presented).
We suggest that the mechanism by which conceptual primes increase correct recollection is related to a (partial) reactivation of the internal context participants activate during the encoding stage, and not to an attribution of a non-mnemonic signal to memory. Crucially, in the present model, when creating a memory trace during encoding, an item is bound to its external context (e.g., object-scene), and internal context (e.g., conceptual associations/personal experiences with a target word) is not accounted for. Perhaps the neural networks subserving the core representational system for recollection could be expanded to include regions supporting semantic associations (e.g., anterior temporal lobes).
In summary, we agree that familiarity judgements can be based on the attribution of a non-mnemonic signal to memory, as described in the proposed model, but we do not see evidence for an attributional basis of recollection judgements. In addition, we suggest expansion of the core representational system for recollection to include internal context.
You have
Access
Bastin et al.’s integrative memory model proposes that subjective memory experiences result from the interaction between core representational systems and an attribution system. The relevant core representational system in the case of familiarity concerns “entity” processing, the implementation of which involves perirhinal cortex, a region regarded to be at the apex of visual object processing. The core representation system for recollection concerns “relational” processing, and includes hippocampus, posterior cingulate, and retrosplenial cortex, which together are believed to coordinate the reactivation of stimulus- and context-specific representations distributed through cortex. So far, this picture is broadly consistent with our impressions of the cognitive, neuropsychological, and neuroimaging literatures on recollection and familiarity.
The “attribution system” is more nebulous, and perhaps more controversial. The term “attribution” implies that in addition to mnemonic signals, non-mnemonic signals are sometimes interpreted as having arisen due to prior exposure to the test item. In general, this could be a sensible heuristic, since many non-mnemonic effects, be they perceptual, emotional, aesthetic, and so forth, are known to result from prior exposure to an item, even sometimes in the absence of explicit memory for that prior exposure (e.g., Hamann & Squire Reference Hamann and Squire1997; Oppenheimer Reference Oppenheimer2008; Schwarz & Winkielman Reference Schwarz and Winkielman2004), so the attribution of such signals to memory is effectively a reverse inference.
In the case of familiarity, it is well established that the fluency with which a test item is processed can be (mis-)attributed to familiarity due to past exposure to the item. By experimentally manipulating the fluency with which a test item is processed (e.g., via masked repetition priming), researchers can induce a familiarity-like feeling which is then attributed to memory. This so-called Jacoby–Whitehouse illusion (Jacoby & Whitehouse Reference Jacoby and Whitehouse1989) has further been found to be specific to familiarity (Rajaram Reference Rajaram1993); a finding that has been replicated many times, including in several studies by one of the present authors (Li et al. Reference Li, Taylor, Wang, Gao and Guo2017; Taylor & Henson Reference Taylor and Henson2012a; Taylor et al. Reference Taylor, Buratto and Henson2013; Woollams et al. Reference Woollams, Taylor, Karayanidis and Henson2008). We therefore agree that familiarity judgements can be based on the attribution of a non-mnemonic signal (i.e., processing fluency) to memory, as stated in the model.
The integrative memory model further proposes that recollection-based memory judgements are also made based on the interaction between its core system and an attribution system. It is implied that the participant will experience a subjective feeling of recollection if the attribution system assesses the memory trace reactivated to be relevant and sufficient (in terms of amount of information retrieved). However, by the authors’ description of the core representational system subserving recollection, the relevant signals for recollection judgements are all memory signals, and therefore, we do not see why an attribution system needs to be posited to intervene between the core representational system and the decision module.
In several studies, we have found that masked conceptual primes (rather than repetition or lexically associated primes) presented before items in a recognition memory test increase correct recollection responses (Li et al. Reference Li, Taylor, Wang, Gao and Guo2017; Taylor & Henson Reference Taylor and Henson2012a; Taylor et al. Reference Taylor, Buratto and Henson2013). However, we do not agree that this constitutes evidence for an attribution system mediating recollection decisions. This is because one hallmark of the attribution framework is that it sometimes misfires, resulting in increased false alarms (primed unstudied items being endorsed as old more often than unprimed unstudied items). But in our studies, only correct recollection responses are increased by conceptual primes; false alarm recollection responses are not increased, and therefore, mis-attribution does not reliably occur. Instead, conceptual priming appears to facilitate veridical retrieval of the encoding event. This is consistent with the notion that the decision module that subserves a recollection judgement relies only on mnemonic signals and therefore, contrary to the target model, it does not require mediation by an attribution system.
It is noteworthy that the evidence for the attribution system comes exclusively from studies using single words as stimuli, whereas much of the rest of the model is mainly based on studies that use objects, pairs of objects, pairs of words, and/or visually rich scenes or environments. One crucial difference between these types of stimuli is that the very same word stimuli are commonly encountered outside of the laboratory, whereas object and scene images and their relations are often (though admittedly not always) unique to the experiment. A second crucial difference is that, at least in the priming and recognition memory studies discussed above, words are often presented in visually impoverished encoding conditions (e.g., one at a time on a blank screen), whereas in studies investigating relational memory, object images are often presented in trial-unique contexts, in pairs, or in well-defined locations in an environment. Thus, it is likely that recollection decisions in word-list memory experiments rely heavily on retrieval of contextual elements from the encoding episode (since the stimuli themselves are not distinctive), and these contextual elements are likely to be internal (e.g., what the participant thought of when they read the word in the study phase) rather than external (e.g., where in the environment, or with which other object or context the target object was presented).
We suggest that the mechanism by which conceptual primes increase correct recollection is related to a (partial) reactivation of the internal context participants activate during the encoding stage, and not to an attribution of a non-mnemonic signal to memory. Crucially, in the present model, when creating a memory trace during encoding, an item is bound to its external context (e.g., object-scene), and internal context (e.g., conceptual associations/personal experiences with a target word) is not accounted for. Perhaps the neural networks subserving the core representational system for recollection could be expanded to include regions supporting semantic associations (e.g., anterior temporal lobes).
In summary, we agree that familiarity judgements can be based on the attribution of a non-mnemonic signal to memory, as described in the proposed model, but we do not see evidence for an attributional basis of recollection judgements. In addition, we suggest expansion of the core representational system for recollection to include internal context.