Quilty-Dunn et al. put forth a strong thesis: That language-of-thought hypothesis (LOTH) is the “best game” in town as far as a computational theory-of-mind is concerned. They marshal evidence from object perception, deductive reasoning, and other domains to support this claim. I am sympathetic to the view that LOT continues to provide a philosophical and conceptual foundation for modern cognitive science. In this commentary, however, I submit, as I did in Xu (Reference Xu2019), that core knowledge systems in human infants do not satisfy the criteria for being in the format of LOT. Here I focus on the domain of object, in particular object individuation in human infants.
Inspired by an analysis of the logic of common nouns (Macnamara, Reference Macnamara1986; Wiggins, Reference Wiggins1980), we reported a series of experiments demonstrating that 10-month-old infants failed to use sortal-kind distinctions (e.g., between a duck and a ball) to establish a representation of two objects in an occlusion event; by 12 months, they can do so (the “is-it-one-or-two” task; Xu & Carey, Reference Xu and Carey1996; see Xu, Reference Xu1997, Reference Xu2007, for reviews). We argued that it is not until the end of the first year that infants represent basic-level sortal-kinds such as duck, ball, spoon, and cup, and learning a natural language – specifically words for these sortal-kinds – may play a causal role in acquiring these concepts. A lot has happened since then.
For the rest of our discussion, it is important to keep in mind that three pieces of evidence are needed to claim that infants represent sortal-kind concepts in an LOT format: (1) success in using between-kind distinctions in object individuation, (2) failure in using within-kind distinctions in object individuation at the same age, and (3) evidence showing that infants detect the perceptual distinctions between sequentially presented objects over occlusion. In support of the claim that infants younger than 12 months do not represent basic-level sortal-kinds, Xu and Carey (Reference Xu and Carey1996) and Xu, Carey, and Quint (Reference Xu, Carey and Quint2004) presented evidence for (1)–(3). Since then, many published studies have used similar methods (the “is-it-one-or-two” task) to further investigate the ontogenetic origin of sortal-kind concepts, focusing on three other superordinate-level concepts: Agent, animate, and artifact (see Croteau, Cheries, & Xu, Reference Croteau, Cheries and Xuforthcoming, for a review). For the concept of an agent, Bonatti, Frot, Zangl, and Mehler (Reference Bonatti, Frot, Zangl and Mehler2002) found that 10-month-old infants successfully individuated a doll head from an inanimate object (a between-kind distinction, agent vs. object), and a doll head from a dog head, but they failed to individuate a doll head from another doll head (a within-kind distinction). Recent studies by Bródy, Oláh, Király, and Biro (Reference Bródy, Oláh, Király and Biro2022), Taborda-Osorio, Lyons, and Cheries (Reference Taborda-Osorio, Lyons and Cheries2019), and Taborda-Osorio and Cheries (Reference Taborda-Osorio and Cheries2018) found that 10-, 11-, or 13-month-old infants used preferences, social-moral dispositions, and internal properties to individuate agents. For the concept of animacy, Surian and Caldi (Reference Surian and Caldi2010) found that 10-month-old infants successfully individuated an animate and an inanimate object (a dynamic caterpillar and a stationary cup; a between-kind distinction) but failed to individuate two animates (a rabbit and a bee; a within-kind distinction). Decarli, Franchin, Piazza, and Surian (Reference Decarli, Franchin, Piazza and Surian2020) provided converging evidence, further disentangling the use of sortal-kind versus featural information. Lastly, Futó, Téglás, Csibra, and Gergely (Reference Futó, Téglás, Csibra and Gergely2010) found that 10-month-old infants successfully individuated an object with a function and another object with a different function, although they did not demonstrate a difference in individuation contrasting between-kind versus within-kind distinctions. The studies on agent, animacy, and artifact did not present direct evidence that infants encoded the various relevant perceptual feature differences, but given what we know about infant perception in general, most would agree that not encoding perceptual differences between objects was an unlikely explanation for the failures in individuation tasks (though see Kibbe & Leslie, Reference Kibbe and Leslie2019). It is also important to note that Wilcox, Baillargeon, Lin, Stavans, and their colleagues have conducted many related experiments over the years, with a strong focus on investigating when infants use featural information in object individuation and the relationship between object files and physical reasoning. Their studies have not aimed to probe the development of sortal-kind concepts (e.g., Lin et al., Reference Lin, Li, Gertner, Ng, Fisher and Baillargeon2021; Stavans, Lin, Wu, & Baillargeon, Reference Stavans, Lin, Wu and Baillargeon2019; Wilcox & Baillargeon, Reference Wilcox and Baillargeon1998). A review of their studies and the various methodological differences between their methods and the “is-it-one-or-two” task is beyond the scope of this commentary; however, these details are important for interpreting this body of research.
The studies reviewed above support the view that toward the end of the first year of life, infants represent sortal-kind concepts: Object, agent, animate, and perhaps artifact. During the first year of life, infants also hear many, many words that refer to basic-level object kinds, people, animals, and artifact kinds. Given the evidence on how words facilitate object categorization, individuation, and inductive inference of nonobvious properties (see Perszyk & Waxman, Reference Perszyk and Waxman2018; Xu, Reference Xu2002, Reference Xu2007; and others for reviews), it remains a viable hypothesis that it is language learning that changes the format of early representations into an LOT.
The core knowledge view (Spelke, Reference Spelke2022) also argues for several other systems of early knowledge besides object. In particular, the number sense presents another strong case that these prelinguistic representations are incompatible with an LOT format. A rich body of research suggests that prelinguistic representations of number share very little with the conceptual representations needed for learning number words. It is an open question whether the prelinguistic representations of agents, places, or social beings are in the format of an LOT.
I applaud Quilty-Dunn et al. for drawing our attention, once again, to the significance of the LOTH. If core knowledge systems are indeed not in the format of an LOT, as I have argued here, cognitive scientists face a major challenge in understanding learning and development in many domains: How does language learning change the format of prelinguistic representations, or alternatively, how does language learning create new conceptual representations that are in the format of an LOT?
Quilty-Dunn et al. put forth a strong thesis: That language-of-thought hypothesis (LOTH) is the “best game” in town as far as a computational theory-of-mind is concerned. They marshal evidence from object perception, deductive reasoning, and other domains to support this claim. I am sympathetic to the view that LOT continues to provide a philosophical and conceptual foundation for modern cognitive science. In this commentary, however, I submit, as I did in Xu (Reference Xu2019), that core knowledge systems in human infants do not satisfy the criteria for being in the format of LOT. Here I focus on the domain of object, in particular object individuation in human infants.
Inspired by an analysis of the logic of common nouns (Macnamara, Reference Macnamara1986; Wiggins, Reference Wiggins1980), we reported a series of experiments demonstrating that 10-month-old infants failed to use sortal-kind distinctions (e.g., between a duck and a ball) to establish a representation of two objects in an occlusion event; by 12 months, they can do so (the “is-it-one-or-two” task; Xu & Carey, Reference Xu and Carey1996; see Xu, Reference Xu1997, Reference Xu2007, for reviews). We argued that it is not until the end of the first year that infants represent basic-level sortal-kinds such as duck, ball, spoon, and cup, and learning a natural language – specifically words for these sortal-kinds – may play a causal role in acquiring these concepts. A lot has happened since then.
For the rest of our discussion, it is important to keep in mind that three pieces of evidence are needed to claim that infants represent sortal-kind concepts in an LOT format: (1) success in using between-kind distinctions in object individuation, (2) failure in using within-kind distinctions in object individuation at the same age, and (3) evidence showing that infants detect the perceptual distinctions between sequentially presented objects over occlusion. In support of the claim that infants younger than 12 months do not represent basic-level sortal-kinds, Xu and Carey (Reference Xu and Carey1996) and Xu, Carey, and Quint (Reference Xu, Carey and Quint2004) presented evidence for (1)–(3). Since then, many published studies have used similar methods (the “is-it-one-or-two” task) to further investigate the ontogenetic origin of sortal-kind concepts, focusing on three other superordinate-level concepts: Agent, animate, and artifact (see Croteau, Cheries, & Xu, Reference Croteau, Cheries and Xuforthcoming, for a review). For the concept of an agent, Bonatti, Frot, Zangl, and Mehler (Reference Bonatti, Frot, Zangl and Mehler2002) found that 10-month-old infants successfully individuated a doll head from an inanimate object (a between-kind distinction, agent vs. object), and a doll head from a dog head, but they failed to individuate a doll head from another doll head (a within-kind distinction). Recent studies by Bródy, Oláh, Király, and Biro (Reference Bródy, Oláh, Király and Biro2022), Taborda-Osorio, Lyons, and Cheries (Reference Taborda-Osorio, Lyons and Cheries2019), and Taborda-Osorio and Cheries (Reference Taborda-Osorio and Cheries2018) found that 10-, 11-, or 13-month-old infants used preferences, social-moral dispositions, and internal properties to individuate agents. For the concept of animacy, Surian and Caldi (Reference Surian and Caldi2010) found that 10-month-old infants successfully individuated an animate and an inanimate object (a dynamic caterpillar and a stationary cup; a between-kind distinction) but failed to individuate two animates (a rabbit and a bee; a within-kind distinction). Decarli, Franchin, Piazza, and Surian (Reference Decarli, Franchin, Piazza and Surian2020) provided converging evidence, further disentangling the use of sortal-kind versus featural information. Lastly, Futó, Téglás, Csibra, and Gergely (Reference Futó, Téglás, Csibra and Gergely2010) found that 10-month-old infants successfully individuated an object with a function and another object with a different function, although they did not demonstrate a difference in individuation contrasting between-kind versus within-kind distinctions. The studies on agent, animacy, and artifact did not present direct evidence that infants encoded the various relevant perceptual feature differences, but given what we know about infant perception in general, most would agree that not encoding perceptual differences between objects was an unlikely explanation for the failures in individuation tasks (though see Kibbe & Leslie, Reference Kibbe and Leslie2019). It is also important to note that Wilcox, Baillargeon, Lin, Stavans, and their colleagues have conducted many related experiments over the years, with a strong focus on investigating when infants use featural information in object individuation and the relationship between object files and physical reasoning. Their studies have not aimed to probe the development of sortal-kind concepts (e.g., Lin et al., Reference Lin, Li, Gertner, Ng, Fisher and Baillargeon2021; Stavans, Lin, Wu, & Baillargeon, Reference Stavans, Lin, Wu and Baillargeon2019; Wilcox & Baillargeon, Reference Wilcox and Baillargeon1998). A review of their studies and the various methodological differences between their methods and the “is-it-one-or-two” task is beyond the scope of this commentary; however, these details are important for interpreting this body of research.
The studies reviewed above support the view that toward the end of the first year of life, infants represent sortal-kind concepts: Object, agent, animate, and perhaps artifact. During the first year of life, infants also hear many, many words that refer to basic-level object kinds, people, animals, and artifact kinds. Given the evidence on how words facilitate object categorization, individuation, and inductive inference of nonobvious properties (see Perszyk & Waxman, Reference Perszyk and Waxman2018; Xu, Reference Xu2002, Reference Xu2007; and others for reviews), it remains a viable hypothesis that it is language learning that changes the format of early representations into an LOT.
The core knowledge view (Spelke, Reference Spelke2022) also argues for several other systems of early knowledge besides object. In particular, the number sense presents another strong case that these prelinguistic representations are incompatible with an LOT format. A rich body of research suggests that prelinguistic representations of number share very little with the conceptual representations needed for learning number words. It is an open question whether the prelinguistic representations of agents, places, or social beings are in the format of an LOT.
I applaud Quilty-Dunn et al. for drawing our attention, once again, to the significance of the LOTH. If core knowledge systems are indeed not in the format of an LOT, as I have argued here, cognitive scientists face a major challenge in understanding learning and development in many domains: How does language learning change the format of prelinguistic representations, or alternatively, how does language learning create new conceptual representations that are in the format of an LOT?
Acknowledgments
I thank Jenna Croteau and Erik Cheries for helpful discussion.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.