Phillips and colleagues propose that the ability to represent knowledge is separate from, and fundamental to, representing beliefs. We believe that this account lays the groundwork for future studies on theory of mind (ToM) but would benefit from consideration of memory theory and research. We suggest turning to semantic and episodic memory, not only because these forms of memory resemble, and potentially contribute to, knowledge and belief representations, but also because conceptual advances in understanding semantic versus episodic memory involved assessing the directionality of their relationship, areas of overlap, and how they are expressed in the real world. Addressing similar issues in relation to knowledge and belief representations could lead to further progress in understanding our capacity to represent others' minds.

Episodic memory is memory for detailed personal experiences that occurred at a specific time and place (Tulving, Reference Tulving, Tulving and Donalson1972, Reference Tulving1983). By contrast, semantic memory contains knowledge about the world and oneself in a context-free form that can be represented separately from the specific experiences in which the knowledge was acquired. In proposing his theory on episodic memory, Endel Tulving was cognizant of similarities with semantic memory but needed to demonstrate how episodic memory is distinct (Renoult & Rugg, Reference Renoult and Rugg2020). Similar to belief representations, episodic memory is experience-dependent and late-developing, both phylogenetically and ontogenetically. It is flexible and reconstructive, so that retrieval from episodic memory lays down new, unique episodic memories. Retrieval of knowledge from semantic memory should leave its content unaltered, although it provides content for new episodic memories. In this way, semantic memory represents an essential foundation for episodic memory: without semantic memory, the emergence of episodic memory should not be possible (Tulving, Reference Tulving1983).

Findings in child development are consistent with the proposed relationship between semantic and episodic memory. Semantic memory typically emerges before episodic memory (Nelson & Fivush, Reference Nelson and Fivush2004; Robertson & Köhler, Reference Robertson and Köhler2007). Individuals with developmental amnesia because of early compromise of the hippocampal memory system experience selective deficits in acquiring episodic memories but can acquire personal and general knowledge (Rosenbaum et al., Reference Rosenbaum, Carson, Abraham, Bowles, Kwan, Köhler and Richards2011; Vargha-Khadem et al., Reference Vargha-Khadem, Gadian, Watkins, Connelly, Van Paesschen and Mishkin1997). The idea that semantic memory is a prerequisite for episodic memory nevertheless has been challenged on the grounds that older adults with semantic dementia (temporal-variant frontotemporal dementia) show an opposite dissociation to that seen in developmental amnesia: impaired semantic memory with relatively intact episodic memory (Viard et al., Reference Viard, Desgranges, Matuszewski, Lebreton, Belliard, de La Sayette and Piolino2013; Westmacott, Leach, Freedman, & Moscovitch, Reference Westmacott, Leach, Freedman and Moscovitch2001). Moreover, in neurotypical populations, memory for specific experiences is the route by which at least some knowledge is acquired or extracted (Sommer, Reference Sommer2017; Yee, Chrysikou, & Thompson-Schill, Reference Yee, Chrysikou, Thompson-Schill, Ochsner and Kosslyn2013).

Even if semantic memory is fundamental to episodic memory, there is widespread agreement that they are intertwined, especially in the real world (Prebble, Addis, & Tippett, Reference Prebble, Addis and Tippett2013; Renoult, Irish, Moscovitch, & Rugg, Reference Renoult, Irish, Moscovitch and Rugg2019). For instance, the self, once viewed as a defining feature of episodic memory (Tulving, Reference Tulving1985), is recognized as an essential part of personal semantic memory (i.e., memory for self-related facts; Grilli & Verfaellie, Reference Grilli and Verfaellie2014). In semantic dementia, concepts that are personally significant (e.g., patient's own coffee mug) are less vulnerable to disruption than concepts that do not hold such meaning (e.g., doctor's coffee mug; Snowden, Griffiths, & Neary, Reference Snowden, Griffiths and Neary1996). More ecologically valid measures of semantic memory have revealed further areas of overlap, with deficits observed in patients with hippocampal damage on tasks involving construction of narratives (Rosenbaum, Gilboa, Levine, Winocur, & Moscovitch, Reference Rosenbaum, Gilboa, Levine, Winocur and Moscovitch2009; Verfaellie, Bousquet, & Keane, Reference Verfaellie, Bousquet and Keane2014) and scenes (Lynch, Keane, & Verfaellie, Reference Lynch, Keane and Verfaellie2020). This is not surprising given marked differences between lab-based, list-learning tests and real-world autobiographical tests of episodic memory in terms of their neural basis (Gilboa, Reference Gilboa2004; McDermott, Szpunar, & Christ, Reference McDermott, Szpunar and Christ2009) and developmental trajectory (Pathman, Samson, Dugas, Cabeza, & Bauer, Reference Pathman, Samson, Dugas, Cabeza and Bauer2011).

The methods used to assess knowledge versus belief attribution may similarly fall short of simulating reality, where the divide between knowledge and belief is not always so clear. It may be difficult to verify that “Suzy knows where to buy an Italian newspaper” if this scenario from the target article was set in the real world. Correspondence of lab-based tests to real-world situations has also been questioned in studies examining the relationship between episodic memory and ToM. These investigations were prompted by simulation theories that view the ability to recollect one's own past mental states as central to inferring the current mental states of others (Buckner & Carroll, Reference Buckner and Carroll2007; Shanton & Goldman, Reference Shanton and Goldman2010). Evidence that episodic memory and ToM emerge close in time in early development (Perner & Ruffman, Reference Perner and Ruffman1995), are impaired together in autism spectrum disorder (Ciaramelli et al., Reference Ciaramelli, Spoglianti, Bertossi, Generali, Telarucci, Tancredi and Igliozzi2018), and rely on an overlapping neural substrate (Spreng, Mar, & Kim, Reference Spreng, Mar and Kim2009) supports this idea. Contrary to this view, however, patients with severely impaired episodic memory because of hippocampal dysfunction perform at the same level as controls on a large number of tests known to be sensitive to ToM (Rabin, Braverman, Gilboa, Stuss, & Rosenbaum, Reference Rabin, Braverman, Gilboa, Stuss and Rosenbaum2012; Rosenbaum, Stuss, Levine, & Tulving, Reference Rosenbaum, Stuss, Levine and Tulving2007), including the kinds of measures described in the target article that assess the capacity to represent others' knowledge versus beliefs (cf. Krachun, Carpenter, Call, & Tomasello, Reference Krachun, Carpenter, Call and Tomasello2009; Stuss, Gallup, & Alexander, Reference Stuss, Gallup and Alexander2001). The findings seem at variance with the idea that ToM depends on episodic memory, but other studies in amnesia suggest that when familiar others are concerned, episodic memory is required for ToM (Rabin, Carson, Gilboa, Stuss, & Rosenbaum, Reference Rabin, Carson, Gilboa, Stuss and Rosenbaum2013; Rabin, Carson, Gilboa, Stuss, & Rosenbaum, Reference Rabin, Olsen, Gilboa, Buchsbaum and Rosenbaum2016). This research highlights another lesson that may be gleaned from research on semantic versus episodic memory: Extending the special population method to include patients with focal brain lesions to infer the causal structure of knowledge and belief attributions (Rosenbaum, Gilboa, & Moscovitch, Reference Rosenbaum, Gilboa and Moscovitch2014).

In sum, we describe several issues that have been raised in the study of semantic versus episodic memory that are relevant to knowledge versus belief attribution and how they have been addressed. A growing number of memory researchers have adopted a more fluid, process-oriented view of semantic and episodic memory in place of the more traditional systems approach. Although a consensus has not yet been reached, the process of re-evaluating these forms of memory has encouraged more rigorous testing of causal claims under naturalistic conditions. Future research on the relationship between knowledge and belief attribution might benefit from a similar approach.