The authors argue for a distinction between so-called “knowledge attributions” and belief attributions, where the former are defined by four unique features: (1) necessarily true content, (2) difference from true belief, (3) compatibility with egocentric ignorance, and (4) modality-generality. The authors suggest that research on theory of mind should shift toward investigating “knowledge attributions” because they are more basic. We agree that there may be two different kinds of representations underlying the authors' distinction, and that one kind of representation may be more basic (even if it may not be developmentally prior). However, we do not agree with the unique features which they take to characterize “knowledge attributions,” or with the claim that these are core representations for mental-state attribution.
The authors suggest that belief plays a central role in the field since Premack and Woodruff (Reference Premack and Woodruff1978) and associated commentaries, and position themselves in opposition to this focus on belief. Instead, we argue that metarepresentation is at the core of attributing representational mental states; focus fell on false belief because it provides a stringent diagnostic for metarepresentation. In our view, most cases of “knowledge attributions,” as defined by the authors, could be explained without appealing to metarepresentation or even to attribution in the traditional sense. Although the authors provide no details on what they take the underlying format to be, it may merely involve two simple representations: our own reality representation and a copy of it to be used for others (Phillips & Norby, Reference Phillips and Norby2019). This kind of non-metarepresentational format would allow one to encode the most up-to-date state of reality for oneself and for others, thereby enabling predictions about who will act in a reality-congruent way or be a reliable source of information. In case there are such representations, more research is needed to characterize their format and their role in cognition. Nevertheless, they should not become the main focus in theory of mind as they do not seem to explain a wide range of phenomena, which rely on tracking how other individuals represent the world without necessarily corresponding to reality.
Other cases of “knowledge attributions,” as defined by the authors, could truly be understood as mental-state attributions involving metarepresentation but we would argue that they are indistinguishable from belief attributions. In particular, we find the four features used by the authors to be insufficiently predictive. Specifically, one subset of these features is compatible with belief attributions (3 and 4); and the other subset seems ad-hoc, especially from the perspective of cognitive mechanisms (1 and 2). On the first point, studies have shown that false-belief attribution also allows for egocentric ignorance (3) in many populations (Biervoye, Meert, Apperly, & Samson, Reference Biervoye, Meert, Apperly and Samson2018; Call & Tomasello, Reference Call and Tomasello1999; Krachun, Carpenter, Call, & Tomasello, Reference Krachun, Carpenter, Call and Tomasello2009; Samson, Apperly, Kathirgamanathan, & Humphreys, Reference Samson, Apperly, Kathirgamanathan and Humphreys2005). In fact, attribution under egocentric ignorance may be one of the best illustrations of the metarepresentational format because attributed beliefs and their content can be manipulated independently (Leslie, Reference Leslie1987). Similarly, false-belief attribution also exhibits modality-generality (4) because it allows for integration of content from different perceptual sources (Song, Onishi, Baillargeon, & Fisher, Reference Song, Onishi, Baillargeon and Fisher2008; Tauzin & Gergely, Reference Tauzin and Gergely2018). On the second point, where the features are indeed incompatible with belief attribution, they merely stipulate the difference from true beliefs (2), or they seem to depend on facts in the external world to secure true contents (1), which may be better captured by epistemologists than psychologists (Ichikawa & Jenkins, Reference Ichikawa, Jenkins, Carter, Gordon and Jarvis2017).
As a general point, we see no way to distinguish these “knowledge attributions” from certain belief attributions on the basis of their truth, at least as these attributions are currently understood within cognitive science. It may be more fruitful to recast the distinction within perception instead of truth, for conceptual as well as empirical reasons. From a conceptual perspective, we cannot imagine how a non-verbal creature could attribute knowledge versus belief in a real-world scenario without appealing to physical cues to perceptual access as a proxy for knowledge. And from an empirical perspective, when working with non-verbal participants such as infants and nonhuman primates, we are limited to testing contexts where beliefs/knowledge are formed on the basis of perceptual access. Despite this, data suggest that uninterrupted perceptual access is the only relevant factor for nonhuman primates, although this does not seem to be the case for children: Independent of their success in false-belief conditions, children's performance is modulated by multiple factors (Kaminski, Call, & Tomasello, Reference Kaminski, Call and Tomasello2008). Future research should try to understand the nuanced factors that contribute to human mental-state attribution as opposed to collapsing it into the two-way distinction that may better explain primate findings.
Similar to the authors, we will end by addressing the role of mental-state attribution in learning. Unlike the authors, we argue that it is actually belief attributions which undergird learning for both conceptual and empirical reasons (Kampis, Somogyi, Itakura, & Király, Reference Kampis, Somogyi, Itakura and Király2013). Given that we cannot see how a young child could hope to distinguish knowledge from mere true belief unless they use uninterrupted perceptual access as a proxy, belief attributions should be equally good to motivate learning. Furthermore, experimental research supports that children and adults rely on false beliefs to learn about the physical world when they themselves lack perceptual access (to locate a target; Biervoye et al., Reference Biervoye, Meert, Apperly and Samson2018; Call and Tomasello, Reference Call and Tomasello1999; Krachun et al., Reference Krachun, Carpenter, Call and Tomasello2009; Samson et al., Reference Samson, Apperly, Kathirgamanathan and Humphreys2005); and even infants use false-belief attributions to acquire person-specific (“She prefers object A”; Luo, Reference Luo2011) or generalizable information (“Object A is preferable”; Kampis et al., Reference Kampis, Somogyi, Itakura and Király2013).
Differentiating representations of reality from mental-state attributions is important to furthering our understanding of the mind. We need to better understand each kind of representation, and its role in broader cognition. But this should not undermine or replace research into theory of mind, which should remain centered around metarepresentation and mental-state attribution. Within this domain, we need better theoretical constructs to distinguish flavors of mental-state attribution, if and when we wish to distinguish them at all. Even after decades, we barely understand the processes and representations involved in this central aspect of human cognition.
The authors argue for a distinction between so-called “knowledge attributions” and belief attributions, where the former are defined by four unique features: (1) necessarily true content, (2) difference from true belief, (3) compatibility with egocentric ignorance, and (4) modality-generality. The authors suggest that research on theory of mind should shift toward investigating “knowledge attributions” because they are more basic. We agree that there may be two different kinds of representations underlying the authors' distinction, and that one kind of representation may be more basic (even if it may not be developmentally prior). However, we do not agree with the unique features which they take to characterize “knowledge attributions,” or with the claim that these are core representations for mental-state attribution.
The authors suggest that belief plays a central role in the field since Premack and Woodruff (Reference Premack and Woodruff1978) and associated commentaries, and position themselves in opposition to this focus on belief. Instead, we argue that metarepresentation is at the core of attributing representational mental states; focus fell on false belief because it provides a stringent diagnostic for metarepresentation. In our view, most cases of “knowledge attributions,” as defined by the authors, could be explained without appealing to metarepresentation or even to attribution in the traditional sense. Although the authors provide no details on what they take the underlying format to be, it may merely involve two simple representations: our own reality representation and a copy of it to be used for others (Phillips & Norby, Reference Phillips and Norby2019). This kind of non-metarepresentational format would allow one to encode the most up-to-date state of reality for oneself and for others, thereby enabling predictions about who will act in a reality-congruent way or be a reliable source of information. In case there are such representations, more research is needed to characterize their format and their role in cognition. Nevertheless, they should not become the main focus in theory of mind as they do not seem to explain a wide range of phenomena, which rely on tracking how other individuals represent the world without necessarily corresponding to reality.
Other cases of “knowledge attributions,” as defined by the authors, could truly be understood as mental-state attributions involving metarepresentation but we would argue that they are indistinguishable from belief attributions. In particular, we find the four features used by the authors to be insufficiently predictive. Specifically, one subset of these features is compatible with belief attributions (3 and 4); and the other subset seems ad-hoc, especially from the perspective of cognitive mechanisms (1 and 2). On the first point, studies have shown that false-belief attribution also allows for egocentric ignorance (3) in many populations (Biervoye, Meert, Apperly, & Samson, Reference Biervoye, Meert, Apperly and Samson2018; Call & Tomasello, Reference Call and Tomasello1999; Krachun, Carpenter, Call, & Tomasello, Reference Krachun, Carpenter, Call and Tomasello2009; Samson, Apperly, Kathirgamanathan, & Humphreys, Reference Samson, Apperly, Kathirgamanathan and Humphreys2005). In fact, attribution under egocentric ignorance may be one of the best illustrations of the metarepresentational format because attributed beliefs and their content can be manipulated independently (Leslie, Reference Leslie1987). Similarly, false-belief attribution also exhibits modality-generality (4) because it allows for integration of content from different perceptual sources (Song, Onishi, Baillargeon, & Fisher, Reference Song, Onishi, Baillargeon and Fisher2008; Tauzin & Gergely, Reference Tauzin and Gergely2018). On the second point, where the features are indeed incompatible with belief attribution, they merely stipulate the difference from true beliefs (2), or they seem to depend on facts in the external world to secure true contents (1), which may be better captured by epistemologists than psychologists (Ichikawa & Jenkins, Reference Ichikawa, Jenkins, Carter, Gordon and Jarvis2017).
As a general point, we see no way to distinguish these “knowledge attributions” from certain belief attributions on the basis of their truth, at least as these attributions are currently understood within cognitive science. It may be more fruitful to recast the distinction within perception instead of truth, for conceptual as well as empirical reasons. From a conceptual perspective, we cannot imagine how a non-verbal creature could attribute knowledge versus belief in a real-world scenario without appealing to physical cues to perceptual access as a proxy for knowledge. And from an empirical perspective, when working with non-verbal participants such as infants and nonhuman primates, we are limited to testing contexts where beliefs/knowledge are formed on the basis of perceptual access. Despite this, data suggest that uninterrupted perceptual access is the only relevant factor for nonhuman primates, although this does not seem to be the case for children: Independent of their success in false-belief conditions, children's performance is modulated by multiple factors (Kaminski, Call, & Tomasello, Reference Kaminski, Call and Tomasello2008). Future research should try to understand the nuanced factors that contribute to human mental-state attribution as opposed to collapsing it into the two-way distinction that may better explain primate findings.
Similar to the authors, we will end by addressing the role of mental-state attribution in learning. Unlike the authors, we argue that it is actually belief attributions which undergird learning for both conceptual and empirical reasons (Kampis, Somogyi, Itakura, & Király, Reference Kampis, Somogyi, Itakura and Király2013). Given that we cannot see how a young child could hope to distinguish knowledge from mere true belief unless they use uninterrupted perceptual access as a proxy, belief attributions should be equally good to motivate learning. Furthermore, experimental research supports that children and adults rely on false beliefs to learn about the physical world when they themselves lack perceptual access (to locate a target; Biervoye et al., Reference Biervoye, Meert, Apperly and Samson2018; Call and Tomasello, Reference Call and Tomasello1999; Krachun et al., Reference Krachun, Carpenter, Call and Tomasello2009; Samson et al., Reference Samson, Apperly, Kathirgamanathan and Humphreys2005); and even infants use false-belief attributions to acquire person-specific (“She prefers object A”; Luo, Reference Luo2011) or generalizable information (“Object A is preferable”; Kampis et al., Reference Kampis, Somogyi, Itakura and Király2013).
Differentiating representations of reality from mental-state attributions is important to furthering our understanding of the mind. We need to better understand each kind of representation, and its role in broader cognition. But this should not undermine or replace research into theory of mind, which should remain centered around metarepresentation and mental-state attribution. Within this domain, we need better theoretical constructs to distinguish flavors of mental-state attribution, if and when we wish to distinguish them at all. Even after decades, we barely understand the processes and representations involved in this central aspect of human cognition.
Financial support
This research is supported by a McDonnell Foundation Network Grant: “The Ontogenetic Origins of Abstract Combinatorial Thought.”
Conflict of interest
None.