The view that infants possess a propositional language-of-thought (LoT) appeals as a unifying account for precocious physical (Stahl & Feigenson, Reference Stahl and Feigenson2015), logical (Cesana-Arlotti et al., Reference Cesana-Arlotti, Martín, Téglás, Vorobyova, Cetnarski and Bonatti2018), probabilistic (Denison & Xu, Reference Denison and Xu2010; Téglás, Girotto, Gonzalez, & Bonatti, Reference Téglás, Girotto, Gonzalez and Bonatti2007), and social reasoning (Baillargeon, Scott, & He, Reference Baillargeon, Scott and He2010; Hamlin, Wynn, & Bloom, Reference Hamlin, Wynn and Bloom2007; Powell & Spelke, Reference Powell and Spelke2013). It suggests continuity along development in the format of human thought. But arguing for such continuity also raises questions. Most, if not all, of the cognitive skills of young infants are also documented in nonhuman species (Engelmann et al., Reference Engelmann, Haux, Völter, Schleihauf, Call, Rakoczy and Herrmann2022; Krupenye, Kano, Hirata, Call, & Tomasello, Reference Krupenye, Kano, Hirata, Call and Tomasello2016), suggesting continuity along evolution. We should thus attribute the same type of thoughts to nonhuman animals and human infants, to animals and human adults. How, then, do we account for animals’ failure to acquire human natural languages and develop unique human cognitive skills? Careful appraisal of the available data and careful experimental designs may instead highlight important discontinuities in the format of thought along both developmental and evolutionary scales, suggesting that a full-fledged LoT, involving all six properties identified by Quilty-Dunn et al. is not yet available to young infants (nor to animals).
I applaud the project of Quilty-Dunn et al. to list specific properties of a propositional LoT and evaluate the presence of these properties in various subdomains of cognitive science. The strength of the evidence for each property in all domains is however unequal. In particular, before concluding that infants possess a full-fledged LoT, we need to provide evidence for each property, individually, and also investigate the limits of each property. I will focus on the first property, “discrete constituents.” It is the most important, as it is presupposed by most other properties: Roles are attributed to discrete constituents; predication combines discrete constituents; logical operators are conceived as discrete constituents. Contrary to Quilty-Dunn et al., I will argue that, although both perception and infant cognition certainly possess discrete representations of objects and possibly of features, there is no evidence for discrete representations of relations in perception nor in prelexical infants.
Although experimental evidence suggests that perceptual representations of relational events and scenes are generalizable to a certain extent (e.g., Goupil, Papeo, & Hochmann, Reference Goupil, Papeo and Hochmann2022; Papeo, Reference Papeo2020; see Kominsky & Scholl, Reference Kominsky and Scholl2020, for the limits of those generalizations), there is no evidence that those representations are discrete, dissociated from the object representations. Rather, relations may well be represented by perceptual schema composed of discrete object representations. The generalizability can be obtained through the underspecification of object representations, a process we previously called “abstraction by impoverishment” (Hochmann & Papeo, Reference Hochmann and Papeo2021). For instance, in perception, a schematic social interaction would consist in two schematic bodies facing each other (Papeo, Reference Papeo2020), a schematic relation of support would consist in an empty object file on top of another empty object file, and so on. Similar representations, with object files possibly enriched with thematic roles, may account for the representation of many relational events in infancy (Leslie & Keeble, Reference Leslie and Keeble1987; Rochat, Striano, & Morgan, Reference Rochat, Striano and Morgan2004; Tatone, Geraci, & Csibra, Reference Tatone, Geraci and Csibra2015).
We recently provided direct evidence supporting the proposal that prelexical infants lack discrete representations for abstract relations (Hochmann, Reference Hochmann2022). We showed that infants can represent the relation same in a format that is abstract, as it can generalize to novel instances of the relation. However those representations are limited to four same individuals, suggesting that the format of infants’ representations is not something like S(A,B), where A and B would be object representations and S the representation of the relation between those objects, but rather (X X), where X is a variable for an object (see Hochmann, Reference Hochmann2022, for the full argumentation). The repetition of the variable carries the relational content same, but only symbols for objects are explicitly represented. This view is reinforced by the systematic failure of young children and other animal species in the relational match-to-sample task, where they need to match pairs of the same or different images (e.g., matching square–square to circle–circle and square-star to moon-triangle). If infants and young children possessed discrete symbols S and D for the relations same and different, they should activate S for both square–square and circle–circle, and D for both square-star and moon-triangle, and easily match S to S or D to D. Instead children fail until the age of 4, and only succeed when actively using the words “same” and “different” (Hochmann et al., Reference Hochmann, Tuerk, Sanborn, Zhu, Long, Dempster and Carey2017). Likewise, chimpanzees (and other animal species) fail the relational match-to-sample task, unless they previously acquired external unitary symbols that refer to the relations same and different (Premack, Reference Premack1983; Thompson, Oden, & Boysen, Reference Thompson, Oden and Boysen1997). These observations highlight a discontinuity along human development. They put forward the hypothesis that relations are initially represented in mental models, and that discrete representations of relations are related to the acquisition of words for those relations. The discrete symbols for abstract relations are possibly no other than the words that refer to those relations.
Finally, even granting infants the capacity to solve the disjunctive syllogism (Cesana-Arlotti et al., Reference Cesana-Arlotti, Martín, Téglás, Vorobyova, Cetnarski and Bonatti2018) or to compute negation (Hochmann & Toro, Reference Hochmann and Toro2021), more experimental work is necessary to describe the format of the representations that permit those performances. Although discrete logical operators could account for these data, other hypotheses are still on the table, including among others, probabilistic representations and inhibitory mechanisms.
In conclusion, the LoT hypothesis is a hypothesis about the format of mental representations. Despite the appeal of a unifying account of cognition and perception, from infancy to adulthood, from bees to humans, discontinuities in the format of thoughts deserve to be studied and highlighted. Quilty-Dunn et al. provide a framework to think about these issues in infants – as well as in nonhuman animals – and develop experimental approaches to decide whether each LoT property is present or absent in infancy, whether infants indeed possess a propositional LoT, or whether they still need to acquire some of the pieces before they can fully play the game.
The view that infants possess a propositional language-of-thought (LoT) appeals as a unifying account for precocious physical (Stahl & Feigenson, Reference Stahl and Feigenson2015), logical (Cesana-Arlotti et al., Reference Cesana-Arlotti, Martín, Téglás, Vorobyova, Cetnarski and Bonatti2018), probabilistic (Denison & Xu, Reference Denison and Xu2010; Téglás, Girotto, Gonzalez, & Bonatti, Reference Téglás, Girotto, Gonzalez and Bonatti2007), and social reasoning (Baillargeon, Scott, & He, Reference Baillargeon, Scott and He2010; Hamlin, Wynn, & Bloom, Reference Hamlin, Wynn and Bloom2007; Powell & Spelke, Reference Powell and Spelke2013). It suggests continuity along development in the format of human thought. But arguing for such continuity also raises questions. Most, if not all, of the cognitive skills of young infants are also documented in nonhuman species (Engelmann et al., Reference Engelmann, Haux, Völter, Schleihauf, Call, Rakoczy and Herrmann2022; Krupenye, Kano, Hirata, Call, & Tomasello, Reference Krupenye, Kano, Hirata, Call and Tomasello2016), suggesting continuity along evolution. We should thus attribute the same type of thoughts to nonhuman animals and human infants, to animals and human adults. How, then, do we account for animals’ failure to acquire human natural languages and develop unique human cognitive skills? Careful appraisal of the available data and careful experimental designs may instead highlight important discontinuities in the format of thought along both developmental and evolutionary scales, suggesting that a full-fledged LoT, involving all six properties identified by Quilty-Dunn et al. is not yet available to young infants (nor to animals).
I applaud the project of Quilty-Dunn et al. to list specific properties of a propositional LoT and evaluate the presence of these properties in various subdomains of cognitive science. The strength of the evidence for each property in all domains is however unequal. In particular, before concluding that infants possess a full-fledged LoT, we need to provide evidence for each property, individually, and also investigate the limits of each property. I will focus on the first property, “discrete constituents.” It is the most important, as it is presupposed by most other properties: Roles are attributed to discrete constituents; predication combines discrete constituents; logical operators are conceived as discrete constituents. Contrary to Quilty-Dunn et al., I will argue that, although both perception and infant cognition certainly possess discrete representations of objects and possibly of features, there is no evidence for discrete representations of relations in perception nor in prelexical infants.
Although experimental evidence suggests that perceptual representations of relational events and scenes are generalizable to a certain extent (e.g., Goupil, Papeo, & Hochmann, Reference Goupil, Papeo and Hochmann2022; Papeo, Reference Papeo2020; see Kominsky & Scholl, Reference Kominsky and Scholl2020, for the limits of those generalizations), there is no evidence that those representations are discrete, dissociated from the object representations. Rather, relations may well be represented by perceptual schema composed of discrete object representations. The generalizability can be obtained through the underspecification of object representations, a process we previously called “abstraction by impoverishment” (Hochmann & Papeo, Reference Hochmann and Papeo2021). For instance, in perception, a schematic social interaction would consist in two schematic bodies facing each other (Papeo, Reference Papeo2020), a schematic relation of support would consist in an empty object file on top of another empty object file, and so on. Similar representations, with object files possibly enriched with thematic roles, may account for the representation of many relational events in infancy (Leslie & Keeble, Reference Leslie and Keeble1987; Rochat, Striano, & Morgan, Reference Rochat, Striano and Morgan2004; Tatone, Geraci, & Csibra, Reference Tatone, Geraci and Csibra2015).
We recently provided direct evidence supporting the proposal that prelexical infants lack discrete representations for abstract relations (Hochmann, Reference Hochmann2022). We showed that infants can represent the relation same in a format that is abstract, as it can generalize to novel instances of the relation. However those representations are limited to four same individuals, suggesting that the format of infants’ representations is not something like S(A,B), where A and B would be object representations and S the representation of the relation between those objects, but rather (X X), where X is a variable for an object (see Hochmann, Reference Hochmann2022, for the full argumentation). The repetition of the variable carries the relational content same, but only symbols for objects are explicitly represented. This view is reinforced by the systematic failure of young children and other animal species in the relational match-to-sample task, where they need to match pairs of the same or different images (e.g., matching square–square to circle–circle and square-star to moon-triangle). If infants and young children possessed discrete symbols S and D for the relations same and different, they should activate S for both square–square and circle–circle, and D for both square-star and moon-triangle, and easily match S to S or D to D. Instead children fail until the age of 4, and only succeed when actively using the words “same” and “different” (Hochmann et al., Reference Hochmann, Tuerk, Sanborn, Zhu, Long, Dempster and Carey2017). Likewise, chimpanzees (and other animal species) fail the relational match-to-sample task, unless they previously acquired external unitary symbols that refer to the relations same and different (Premack, Reference Premack1983; Thompson, Oden, & Boysen, Reference Thompson, Oden and Boysen1997). These observations highlight a discontinuity along human development. They put forward the hypothesis that relations are initially represented in mental models, and that discrete representations of relations are related to the acquisition of words for those relations. The discrete symbols for abstract relations are possibly no other than the words that refer to those relations.
Finally, even granting infants the capacity to solve the disjunctive syllogism (Cesana-Arlotti et al., Reference Cesana-Arlotti, Martín, Téglás, Vorobyova, Cetnarski and Bonatti2018) or to compute negation (Hochmann & Toro, Reference Hochmann and Toro2021), more experimental work is necessary to describe the format of the representations that permit those performances. Although discrete logical operators could account for these data, other hypotheses are still on the table, including among others, probabilistic representations and inhibitory mechanisms.
In conclusion, the LoT hypothesis is a hypothesis about the format of mental representations. Despite the appeal of a unifying account of cognition and perception, from infancy to adulthood, from bees to humans, discontinuities in the format of thoughts deserve to be studied and highlighted. Quilty-Dunn et al. provide a framework to think about these issues in infants – as well as in nonhuman animals – and develop experimental approaches to decide whether each LoT property is present or absent in infancy, whether infants indeed possess a propositional LoT, or whether they still need to acquire some of the pieces before they can fully play the game.
Financial support
This work was supported by the Agence Nationale pour la Recherche grant ANR-16-CE28-0006 TACTIC and the collaborative McDonnell Foundation Grant 220020449.
Competing interest
None.