Mahr & Csibra (M&C) propose a new and provocative account of the content and function of episodic memory. Although a significant contribution to the ongoing debate about the nature and purpose of memory, we see important gaps in their framework.
The main problem is that this account says very little about the relationship between semantic and episodic memory, instead endorsing a sharp distinction between these two types of memory (see M&C Table 1). We outline a number of in-principle arguments and review relevant neuroimaging and neuropsychological evidence to challenge this distinction. By highlighting the close interactions between these memory systems, we place equal emphasis on the importance of semantic memory for the creation and communication of beliefs.
Some types of semantic memories have a strong experiential basis. Personal experiences that are repeated frequently (e.g., doing the weekly grocery shopping) can be stored as event scripts or schemas (Alba & Hasher Reference Alba and Hasher1983; Renoult et al. Reference Renoult, Davidson, Palombo, Moscovitch and Levine2012). Such schemas contain the most common elements of repeated experiences. We see at least three ways that these memories interact with episodic memories. First, semantic memories may begin as personal memories of individual episodes that become decontextualized and compiled into schemas over repetitions. Second, schema memories during retrieval can form the foundation for reconstruction of specific episodes. Imagine, for example, that your partner asks you whether an advertised sale has started at the clothing store near your usual supermarket. To answer, you could sample your supermarket-shopping schema to retrieve memories of nearby stores, using these as retrieval cues for details of your most recent visit to the mall.
Perhaps most significantly for the M&C account, semantic memories can serve as “grounds for believing” in social communication. For example, someone may ask which of two grocery stores you think provides the best value for money. Constructing an answer would likely require sampling schemas of shopping at each store (Stewart et al. Reference Stewart, Chater and Brown2006). Notably, these samples would involve retrieval of abstract impressions of “value” across episodes rather than detailed episodic information. This example illustrates that the decontextualized knowledge in semantic memory is no less important to social communication than episodic memory. In fact, semantic memory serves a crucial communicative function by virtue of its veracity. It strips away episodic memories to preserve the most important elements and hence is less malleable (and error-prone) than episodic memory.
The argument against a strict fractionation between episodic and semantic memory is supported by mounting evidence from the neuroimaging and neuropsychological literature (Greenberg & Verfaellie Reference Greenberg and Verfaellie2010). Neuroimaging findings of robust activation of semantic processing brain regions during so-called episodic memory tasks (Binder & Desai Reference Binder and Desai2011; Burianová et al. Reference Burianová, McIntosh and Grady2010) reveal a close correspondence between episodic and semantic memory on the neuroanatomical level, underscoring the unclear boundaries that exist between these memory systems.
Working from this premise, episodic constructions of the type proposed by M&C should contain a large proportion of semantic concepts (Binder et al. Reference Binder, Desai, Graves and Conant2009). Neuropsychological evidence from clinical populations supports this assertion. For example, patients with semantic dementia, in which the conceptual knowledge base is progressively eroded, display marked impairments in constructing novel future events, despite a relative preservation of episodic memory (Irish et al. Reference Irish, Addis, Hodges and Piguet2012a; Reference Irish, Addis, Hodges and Piguet2012b). These impairments in mental simulation are attributable to atrophy of specialized semantic processing regions in the anterior temporal lobes (reviewed by Irish & Piolino Reference Irish and Piolino2016). Indeed, semantic representations are posited to underlie a host of other higher-order cognitive processes including language, reasoning, planning, problem solving, and social interaction (Binder et al. Reference Binder, Desai, Graves and Conant2009).
As M&C note, an unlimited episodic memory store would be highly unfeasible, requiring us to store the minutiae of every event experienced in order to justify any and all inferences that we make. In this regard, the utility of semantic memory becomes apparent – by extracting and retaining the crux of past events, we avail ourselves of undifferentiated information that can be generalized across contexts and scenarios (Lambon Ralph et al. Reference Lambon Ralph, Sage, Jones and Mayberry2010). It is thus possible for us to engage in non-episodic forms of simulation to envisage what the future may hold, without having experienced these events previously (Irish et al. Reference Irish, Eyre, Dermody, O'Callaghan, Hodges, Hornberger and Piguet2016; Klein et al. Reference Klein, Loftus and Kihlstrom2002b).
Considering the communicative function of episodic memory endorsed by M&C, we propose that semantic memory provides the foundation for the veridical telling of events. The persuasive exchange of information will invariably depend upon the accessibility of information that is relevant to the speakers' goals. Rather than placing unwieldy demands upon episodic recollection to support our beliefs, we propose that semantic memory represents a more efficient system with which to flexibly represent and justify beliefs. For example, in reasoning why one should not walk down dark alleyways alone at night, one does not need to have experienced an actual mugging to persuasively communicate this information. Rather, the “computational bottleneck” of the episodic memory system is superseded by the abstracted conceptual knowledge that certain areas of a city may not be safe at night (see also Hegdé Reference Hegdé2007). Of course, episodic memory can also be utilized to instill credibility, however, we contend that it is not the limiting factor in this process.
In short, it seems prudent to prioritize the role of semantic memory in the communication of information. Semantic facts are stable; they are passed on through public records, family history, and collective discourse to represent the outcomes of inferences and experiences. The essential elements are extracted, providing a foundation that is not subject to the vagaries of episodic context or personal biases. A major challenge then, is to establish a consensus regarding the respective contributions of, and interactions between, episodic and semantic memory with regard to persuasive communication. Rather than positioning these entities as mutually exclusive systems, we conceptualize semantic memory as providing the essential scaffold for constructive endeavours across past, future, and social contexts (Irish & Piguet Reference Irish and Piguet2013), and worthy of inclusion in the conversation.
Mahr & Csibra (M&C) propose a new and provocative account of the content and function of episodic memory. Although a significant contribution to the ongoing debate about the nature and purpose of memory, we see important gaps in their framework.
The main problem is that this account says very little about the relationship between semantic and episodic memory, instead endorsing a sharp distinction between these two types of memory (see M&C Table 1). We outline a number of in-principle arguments and review relevant neuroimaging and neuropsychological evidence to challenge this distinction. By highlighting the close interactions between these memory systems, we place equal emphasis on the importance of semantic memory for the creation and communication of beliefs.
Some types of semantic memories have a strong experiential basis. Personal experiences that are repeated frequently (e.g., doing the weekly grocery shopping) can be stored as event scripts or schemas (Alba & Hasher Reference Alba and Hasher1983; Renoult et al. Reference Renoult, Davidson, Palombo, Moscovitch and Levine2012). Such schemas contain the most common elements of repeated experiences. We see at least three ways that these memories interact with episodic memories. First, semantic memories may begin as personal memories of individual episodes that become decontextualized and compiled into schemas over repetitions. Second, schema memories during retrieval can form the foundation for reconstruction of specific episodes. Imagine, for example, that your partner asks you whether an advertised sale has started at the clothing store near your usual supermarket. To answer, you could sample your supermarket-shopping schema to retrieve memories of nearby stores, using these as retrieval cues for details of your most recent visit to the mall.
Perhaps most significantly for the M&C account, semantic memories can serve as “grounds for believing” in social communication. For example, someone may ask which of two grocery stores you think provides the best value for money. Constructing an answer would likely require sampling schemas of shopping at each store (Stewart et al. Reference Stewart, Chater and Brown2006). Notably, these samples would involve retrieval of abstract impressions of “value” across episodes rather than detailed episodic information. This example illustrates that the decontextualized knowledge in semantic memory is no less important to social communication than episodic memory. In fact, semantic memory serves a crucial communicative function by virtue of its veracity. It strips away episodic memories to preserve the most important elements and hence is less malleable (and error-prone) than episodic memory.
The argument against a strict fractionation between episodic and semantic memory is supported by mounting evidence from the neuroimaging and neuropsychological literature (Greenberg & Verfaellie Reference Greenberg and Verfaellie2010). Neuroimaging findings of robust activation of semantic processing brain regions during so-called episodic memory tasks (Binder & Desai Reference Binder and Desai2011; Burianová et al. Reference Burianová, McIntosh and Grady2010) reveal a close correspondence between episodic and semantic memory on the neuroanatomical level, underscoring the unclear boundaries that exist between these memory systems.
Working from this premise, episodic constructions of the type proposed by M&C should contain a large proportion of semantic concepts (Binder et al. Reference Binder, Desai, Graves and Conant2009). Neuropsychological evidence from clinical populations supports this assertion. For example, patients with semantic dementia, in which the conceptual knowledge base is progressively eroded, display marked impairments in constructing novel future events, despite a relative preservation of episodic memory (Irish et al. Reference Irish, Addis, Hodges and Piguet2012a; Reference Irish, Addis, Hodges and Piguet2012b). These impairments in mental simulation are attributable to atrophy of specialized semantic processing regions in the anterior temporal lobes (reviewed by Irish & Piolino Reference Irish and Piolino2016). Indeed, semantic representations are posited to underlie a host of other higher-order cognitive processes including language, reasoning, planning, problem solving, and social interaction (Binder et al. Reference Binder, Desai, Graves and Conant2009).
As M&C note, an unlimited episodic memory store would be highly unfeasible, requiring us to store the minutiae of every event experienced in order to justify any and all inferences that we make. In this regard, the utility of semantic memory becomes apparent – by extracting and retaining the crux of past events, we avail ourselves of undifferentiated information that can be generalized across contexts and scenarios (Lambon Ralph et al. Reference Lambon Ralph, Sage, Jones and Mayberry2010). It is thus possible for us to engage in non-episodic forms of simulation to envisage what the future may hold, without having experienced these events previously (Irish et al. Reference Irish, Eyre, Dermody, O'Callaghan, Hodges, Hornberger and Piguet2016; Klein et al. Reference Klein, Loftus and Kihlstrom2002b).
Considering the communicative function of episodic memory endorsed by M&C, we propose that semantic memory provides the foundation for the veridical telling of events. The persuasive exchange of information will invariably depend upon the accessibility of information that is relevant to the speakers' goals. Rather than placing unwieldy demands upon episodic recollection to support our beliefs, we propose that semantic memory represents a more efficient system with which to flexibly represent and justify beliefs. For example, in reasoning why one should not walk down dark alleyways alone at night, one does not need to have experienced an actual mugging to persuasively communicate this information. Rather, the “computational bottleneck” of the episodic memory system is superseded by the abstracted conceptual knowledge that certain areas of a city may not be safe at night (see also Hegdé Reference Hegdé2007). Of course, episodic memory can also be utilized to instill credibility, however, we contend that it is not the limiting factor in this process.
In short, it seems prudent to prioritize the role of semantic memory in the communication of information. Semantic facts are stable; they are passed on through public records, family history, and collective discourse to represent the outcomes of inferences and experiences. The essential elements are extracted, providing a foundation that is not subject to the vagaries of episodic context or personal biases. A major challenge then, is to establish a consensus regarding the respective contributions of, and interactions between, episodic and semantic memory with regard to persuasive communication. Rather than positioning these entities as mutually exclusive systems, we conceptualize semantic memory as providing the essential scaffold for constructive endeavours across past, future, and social contexts (Irish & Piguet Reference Irish and Piguet2013), and worthy of inclusion in the conversation.