What Babies Know (Spelke, Reference Spelke2022) summarizes and systematizes several decades of ingenious, influential research on infants' cognition. Its central thesis is that “infants’ learning rests on a set of cognitive systems that we share with animals and that evolved over hundreds of millions of years. At least six distinct systems serve to represent highly abstract properties of the unchanging navigable environment, of movable objects, of number, and of the living, animate, and social beings who populate our world” (Spelke, Reference Spelke2022, p. xvii). In this commentary, we ask whether there may be room for one more – a core domain of substances for nonsolid things like liquid or sand – and we examine some consequences of adding it to Spelke's “at least six.”
The key tenet of Spelke's core-knowledge approach is that underneath all the things that vary across humans, there exists a set of conceptual capacities common to everyone. The research motivated by core knowledge strives to characterize these abilities and their development. In the case of object knowledge, children never receive explicit instruction about how objects behave and interact, yet they draw universally similar expectations about them. For example, they universally expect that hidden objects do not cease to exist when they are hidden from view (Baillargeon, Spelke, & Wasserman, Reference Baillargeon, Spelke and Wasserman1985). Such expectations appear to be universal across age groups as well as across individuals of the same age. Expectations about objects are evident in other species as well. Rhesus macaques expect an object to stop when it comes in contact with a wall rather than pass through it (Santos & Hauser, Reference Santos and Hauser2002). Humans and chickens have similar expectations about partially occluded objects (Chiandetti & Vallortigara, Reference Chiandetti and Vallortigara2011; Kellman & Spelke, Reference Kellman and Spelke1983).
However, our world includes more than physical objects. Spelke has highlighted five other core-knowledge domains, arguing that core-knowledge systems are evolutionarily important abilities, each solving a limited set of problems. These systems are encapsulated, that is, they operate independently of other cognitive systems. An advantage of encapsulation is that core abilities are universal and are effortlessly acquired with little experience. A limitation of encapsulation is that these abilities lack flexibility and precision (e.g., Samuels, Reference Samuels, Margolis, Samuels and Stich2012). Evidence supporting the independence of core systems comes from neurological research identifying specialized brain areas for processing this information, as well as clinical studies showing that these domain-specific abilities can be lost while other cognitive abilities remain intact.
Spelke often cites nonobjects like sand as a contrast case to items in the core domain of objects. Our focus in this commentary is on arguing that instead of being merely a fringe contrast example, substances may be a core domain of their own. As adults, we automatically react differently when we upset a glass of beer than when we upset a bowl of pretzels. These reactions to spills result from our understanding that objects and liquids have different physical properties and so behave differently. These reactions may seem obvious, but when do we develop the notion that liquids deform to fill space whereas solids don't? This ontological categorical distinction has captivated linguists who trace differences in the meaning of quantitative expressions, such as “many pretzels” (vs. *“much pretzels”) and “much beer” (vs. *“many beer”) (e.g., Rothstein, Reference Rothstein2017; Wellwood, Reference Wellwood2019). In the philosophical domain of metaphysics, there are distinctions between entities that come in atomic units (like pretzels) and those that have no clear units (like beer; e.g., Frege, Reference Frege1980; Koslicki, Reference Koslicki1997). In the field of psychology, we look at the origins and development of knowledge about substances and how it compares to knowledge about objects (Hespos & VanMarle, Reference Hespos and VanMarle2012; Rips & Hespos, Reference Rips and Hespos2015, Reference Rips and Hespos2019). For example, unlike objects, liquids deform to fit a container and a solid object can pass through them. Yet, like objects, liquids are omnipresent, and knowledge of how substances behave is probably universal across cultures and species.
Early evidence suggested that infants had principled expectations about objects, but not about substances (Cheries, Mitroff, Wynn, & Scholl, Reference Cheries, Mitroff, Wynn and Scholl2008; Chiang & Wynn, Reference Chiang and Wynn2000; Huntley-Fenner, Carey, & Solimando, Reference Huntley-Fenner, Carey and Solimando2002). In one study, Huntley-Fenner et al. (Reference Huntley-Fenner, Carey and Solimando2002) showed infants a pile of sand poured on a stage; then the pile was concealed by a screen, and a second pile of sand was poured behind a nearby but separate screen. The test trials alternated between an expected (by adults) outcome and an unexpected outcome, and looking time was the dependent measure. The expected outcome was to reveal two piles of sand, one behind each screen. The unexpected outcome was to reveal only a single pile of sand behind one of the screens and nothing behind the other. Infants looked equally at the expected and unexpected outcomes, providing evidence that they did not detect the violation when one sand pile disappeared. In contrast, when the sand was replaced with solid objects that were shaped like sand piles, the infants looked significantly longer at the unexpected test trials. Infants' difficulties in tracking sand extended to collections of objects, like a disassembled pile of Legos (Chiang & Wynn, Reference Chiang and Wynn2000). Together, these findings were interpreted as evidence that infants have principled expectations for objects but not for substances (Spelke & Kinzler, Reference Spelke and Kinzler2007).
However, evidence for infants' knowledge of substances began to appear in later studies. Bourgeois, Khawar, Neal, and Lockman (Reference Bourgeois, Khawar, Neal and Lockman2005) introduced a different approach to ask if infants had concepts of distinct materials. They presented infants with entities that varied in whether they were rigid (particle board) versus flexible (sponge) versus liquid (water) versus discontinuous (netting) and found that 6- to 10-month-old infants adjusted their actions toward the entities based on their material-specific qualities. This finding was important because it demonstrated that infants applied different behaviors to objects and liquids.
Our first experiment on substance knowledge asked whether infants have material-specific ideas about liquids. Using a looking paradigm, we habituated infants to either a glass that contained a liquid or a glass that had solid contents but was otherwise perceptually similar. The glass was tipped back and forth, and the motion cues revealed whether the contents were liquid or solid. Next, in the test trials, all infants saw a straw lowered into the glass. On half the trials, the contents of the glass were liquid, and the straw penetrated the surface of the liquid, coming to rest at the bottom of the glass. On the other half of the trials, the contents of the glass were solid, and the straw stopped when it met the surface of the solid. Infants dishabituated (i.e., showed a significant increase in looking time compared to their last habituation trials) when there was a state change from liquid to solid or from solid to liquid. These studies show that infants have distinct ideas about how objects and liquids behave (Hespos, Ferry, & Rips, Reference Hespos, Ferry and Rips2009).
We started with a water-like substance because it is the most prevalent example of its kind. However, core principles go beyond information in the immediate environment. Our initial findings raised questions about how far infants' ideas about substances extend. Do infants develop expectations about liquids because of their experience in drinking and bathing or would they generalize such expectations to unfamiliar events with similar physical attributes? More specifically, would an infant who has never been to a beach know that the sand in a cup should pour out and not tumble? Our next study provided a positive answer (Hespos, Ferry, Anderson, Hollenbeck, & Rips, Reference Hespos, Ferry, Anderson, Hollenbeck and Rips2016). The events were like those of the liquid experiment except the liquid was replaced with sand. We again found that the infants distinguished sand from a solid object. Together these findings provide evidence that knowledge of substances emerges early, based on little or no experience.
The results we have presented suggest that infants can grasp simple physical properties that apply to nonsolid substances, and Spelke now acknowledges these findings (Spelke, Reference Spelke2022, pp. 62–63, footnote 8). However, the previous research showing success with objects and failure with substances in otherwise identical paradigms poses questions about the extent of infants' knowledge (Chiang & Wynn, Reference Chiang and Wynn2000; Huntley-Fenner et al., Reference Huntley-Fenner, Carey and Solimando2002; Rosenberg & Carey, Reference Rosenberg, Carey, Hood and Santos2009). Infants seem unable to predict the number of piles that result from pouring sand behind adjacent screens. But could this be because of the working memory demands of the pouring event rather than to lack of substance knowledge? We tested infants' expectations about a simplified pouring event. Although nonsolid substances can sometimes spread to fit the space allotted, constraints particular to sand limit its ability to do so. If infants see two cups of sand poured at opposite ends of a tray behind a screen, would it violate their expectations to reveal a single pile? What if just one cup was poured behind a screen? Would it be surprising if a single pour resulted in two separate piles? Our findings provide a yes answer to both questions (Anderson, Hespos, & Rips, Reference Anderson, Hespos and Rips2018).
The emerging picture is that infants have ideas about how substances behave, divide, and accumulate. These ideas are distinct from those governing how objects behave and interact, but they have at least some of the standard core properties (Spelke, Reference Spelke2022, Ch. 5): They appear as early as 4 months of age. They seem to operate automatically and unconsciously. They are abstract, in that motion cues cause infants to make inferences about later division and accumulation. They go beyond information in the immediate perceptual array.
Of course, many of the criteria for core domains remain open to investigation in the case of substances, but it is worth considering the implications of a possible substance domain for the general core-knowledge framework. The experiments described earlier show that infants understand some of the relations that hold between objects and substances. They know that solid objects can pass through nonsolid substances but not through other solid objects. What's unclear is how infants can know about these relations if objects and substances belong to distinct domains. Encapsulation of domains is a hallmark of core knowledge, as Spelke (Reference Spelke2022) emphasizes. So, if objects and substances belong to different core domains (that by definition can't talk to each other), then to capture the relations, the object domain must contain information about substances (e.g., that objects can pass through substances), or the substance domain must contain information about objects (ditto) or both. We're unsure how this overlap could be consistent with encapsulation. There are similar overlaps between object and number domains, and future research could work out these details.
Similarly, all objects consist of substances, and the causal properties of the objects are inherited in part from those substances. A toy's behavior during a collision will differ for a toy made of rubber than for an otherwise similar toy made of metal. Assuming that infants are sensitive to such differences in behavior, then if it's the object domain that predicts the nature of collisions (Spelke, Reference Spelke2022, sects. 2.3 and 2.5), it must have access to knowledge about the substances that compose the objects. But if objects and substances belong to separate encapsulated core domains, this kind of coordination is difficult to explain. Perhaps the correct domains are not objects and substances, but solid objects and nonsolid substances. However, that would leave wood, metal, and other solid substances and socks, raindrops, and other nonsolid objects in limbo until the knowledge is elaborated or refined later in development.
Issues like these reveal a tension between the core-knowledge program's expanding scope and its architectural restrictions. The experimental program that Spelke launched has successfully revealed an increasing range of knowledge that infants deploy. But this increase comes with problems of accommodating the information within core knowledge's separate fiefdoms.
You have
Access
What Babies Know (Spelke, Reference Spelke2022) summarizes and systematizes several decades of ingenious, influential research on infants' cognition. Its central thesis is that “infants’ learning rests on a set of cognitive systems that we share with animals and that evolved over hundreds of millions of years. At least six distinct systems serve to represent highly abstract properties of the unchanging navigable environment, of movable objects, of number, and of the living, animate, and social beings who populate our world” (Spelke, Reference Spelke2022, p. xvii). In this commentary, we ask whether there may be room for one more – a core domain of substances for nonsolid things like liquid or sand – and we examine some consequences of adding it to Spelke's “at least six.”
The key tenet of Spelke's core-knowledge approach is that underneath all the things that vary across humans, there exists a set of conceptual capacities common to everyone. The research motivated by core knowledge strives to characterize these abilities and their development. In the case of object knowledge, children never receive explicit instruction about how objects behave and interact, yet they draw universally similar expectations about them. For example, they universally expect that hidden objects do not cease to exist when they are hidden from view (Baillargeon, Spelke, & Wasserman, Reference Baillargeon, Spelke and Wasserman1985). Such expectations appear to be universal across age groups as well as across individuals of the same age. Expectations about objects are evident in other species as well. Rhesus macaques expect an object to stop when it comes in contact with a wall rather than pass through it (Santos & Hauser, Reference Santos and Hauser2002). Humans and chickens have similar expectations about partially occluded objects (Chiandetti & Vallortigara, Reference Chiandetti and Vallortigara2011; Kellman & Spelke, Reference Kellman and Spelke1983).
However, our world includes more than physical objects. Spelke has highlighted five other core-knowledge domains, arguing that core-knowledge systems are evolutionarily important abilities, each solving a limited set of problems. These systems are encapsulated, that is, they operate independently of other cognitive systems. An advantage of encapsulation is that core abilities are universal and are effortlessly acquired with little experience. A limitation of encapsulation is that these abilities lack flexibility and precision (e.g., Samuels, Reference Samuels, Margolis, Samuels and Stich2012). Evidence supporting the independence of core systems comes from neurological research identifying specialized brain areas for processing this information, as well as clinical studies showing that these domain-specific abilities can be lost while other cognitive abilities remain intact.
Spelke often cites nonobjects like sand as a contrast case to items in the core domain of objects. Our focus in this commentary is on arguing that instead of being merely a fringe contrast example, substances may be a core domain of their own. As adults, we automatically react differently when we upset a glass of beer than when we upset a bowl of pretzels. These reactions to spills result from our understanding that objects and liquids have different physical properties and so behave differently. These reactions may seem obvious, but when do we develop the notion that liquids deform to fill space whereas solids don't? This ontological categorical distinction has captivated linguists who trace differences in the meaning of quantitative expressions, such as “many pretzels” (vs. *“much pretzels”) and “much beer” (vs. *“many beer”) (e.g., Rothstein, Reference Rothstein2017; Wellwood, Reference Wellwood2019). In the philosophical domain of metaphysics, there are distinctions between entities that come in atomic units (like pretzels) and those that have no clear units (like beer; e.g., Frege, Reference Frege1980; Koslicki, Reference Koslicki1997). In the field of psychology, we look at the origins and development of knowledge about substances and how it compares to knowledge about objects (Hespos & VanMarle, Reference Hespos and VanMarle2012; Rips & Hespos, Reference Rips and Hespos2015, Reference Rips and Hespos2019). For example, unlike objects, liquids deform to fit a container and a solid object can pass through them. Yet, like objects, liquids are omnipresent, and knowledge of how substances behave is probably universal across cultures and species.
Early evidence suggested that infants had principled expectations about objects, but not about substances (Cheries, Mitroff, Wynn, & Scholl, Reference Cheries, Mitroff, Wynn and Scholl2008; Chiang & Wynn, Reference Chiang and Wynn2000; Huntley-Fenner, Carey, & Solimando, Reference Huntley-Fenner, Carey and Solimando2002). In one study, Huntley-Fenner et al. (Reference Huntley-Fenner, Carey and Solimando2002) showed infants a pile of sand poured on a stage; then the pile was concealed by a screen, and a second pile of sand was poured behind a nearby but separate screen. The test trials alternated between an expected (by adults) outcome and an unexpected outcome, and looking time was the dependent measure. The expected outcome was to reveal two piles of sand, one behind each screen. The unexpected outcome was to reveal only a single pile of sand behind one of the screens and nothing behind the other. Infants looked equally at the expected and unexpected outcomes, providing evidence that they did not detect the violation when one sand pile disappeared. In contrast, when the sand was replaced with solid objects that were shaped like sand piles, the infants looked significantly longer at the unexpected test trials. Infants' difficulties in tracking sand extended to collections of objects, like a disassembled pile of Legos (Chiang & Wynn, Reference Chiang and Wynn2000). Together, these findings were interpreted as evidence that infants have principled expectations for objects but not for substances (Spelke & Kinzler, Reference Spelke and Kinzler2007).
However, evidence for infants' knowledge of substances began to appear in later studies. Bourgeois, Khawar, Neal, and Lockman (Reference Bourgeois, Khawar, Neal and Lockman2005) introduced a different approach to ask if infants had concepts of distinct materials. They presented infants with entities that varied in whether they were rigid (particle board) versus flexible (sponge) versus liquid (water) versus discontinuous (netting) and found that 6- to 10-month-old infants adjusted their actions toward the entities based on their material-specific qualities. This finding was important because it demonstrated that infants applied different behaviors to objects and liquids.
Our first experiment on substance knowledge asked whether infants have material-specific ideas about liquids. Using a looking paradigm, we habituated infants to either a glass that contained a liquid or a glass that had solid contents but was otherwise perceptually similar. The glass was tipped back and forth, and the motion cues revealed whether the contents were liquid or solid. Next, in the test trials, all infants saw a straw lowered into the glass. On half the trials, the contents of the glass were liquid, and the straw penetrated the surface of the liquid, coming to rest at the bottom of the glass. On the other half of the trials, the contents of the glass were solid, and the straw stopped when it met the surface of the solid. Infants dishabituated (i.e., showed a significant increase in looking time compared to their last habituation trials) when there was a state change from liquid to solid or from solid to liquid. These studies show that infants have distinct ideas about how objects and liquids behave (Hespos, Ferry, & Rips, Reference Hespos, Ferry and Rips2009).
We started with a water-like substance because it is the most prevalent example of its kind. However, core principles go beyond information in the immediate environment. Our initial findings raised questions about how far infants' ideas about substances extend. Do infants develop expectations about liquids because of their experience in drinking and bathing or would they generalize such expectations to unfamiliar events with similar physical attributes? More specifically, would an infant who has never been to a beach know that the sand in a cup should pour out and not tumble? Our next study provided a positive answer (Hespos, Ferry, Anderson, Hollenbeck, & Rips, Reference Hespos, Ferry, Anderson, Hollenbeck and Rips2016). The events were like those of the liquid experiment except the liquid was replaced with sand. We again found that the infants distinguished sand from a solid object. Together these findings provide evidence that knowledge of substances emerges early, based on little or no experience.
The results we have presented suggest that infants can grasp simple physical properties that apply to nonsolid substances, and Spelke now acknowledges these findings (Spelke, Reference Spelke2022, pp. 62–63, footnote 8). However, the previous research showing success with objects and failure with substances in otherwise identical paradigms poses questions about the extent of infants' knowledge (Chiang & Wynn, Reference Chiang and Wynn2000; Huntley-Fenner et al., Reference Huntley-Fenner, Carey and Solimando2002; Rosenberg & Carey, Reference Rosenberg, Carey, Hood and Santos2009). Infants seem unable to predict the number of piles that result from pouring sand behind adjacent screens. But could this be because of the working memory demands of the pouring event rather than to lack of substance knowledge? We tested infants' expectations about a simplified pouring event. Although nonsolid substances can sometimes spread to fit the space allotted, constraints particular to sand limit its ability to do so. If infants see two cups of sand poured at opposite ends of a tray behind a screen, would it violate their expectations to reveal a single pile? What if just one cup was poured behind a screen? Would it be surprising if a single pour resulted in two separate piles? Our findings provide a yes answer to both questions (Anderson, Hespos, & Rips, Reference Anderson, Hespos and Rips2018).
The emerging picture is that infants have ideas about how substances behave, divide, and accumulate. These ideas are distinct from those governing how objects behave and interact, but they have at least some of the standard core properties (Spelke, Reference Spelke2022, Ch. 5): They appear as early as 4 months of age. They seem to operate automatically and unconsciously. They are abstract, in that motion cues cause infants to make inferences about later division and accumulation. They go beyond information in the immediate perceptual array.
Of course, many of the criteria for core domains remain open to investigation in the case of substances, but it is worth considering the implications of a possible substance domain for the general core-knowledge framework. The experiments described earlier show that infants understand some of the relations that hold between objects and substances. They know that solid objects can pass through nonsolid substances but not through other solid objects. What's unclear is how infants can know about these relations if objects and substances belong to distinct domains. Encapsulation of domains is a hallmark of core knowledge, as Spelke (Reference Spelke2022) emphasizes. So, if objects and substances belong to different core domains (that by definition can't talk to each other), then to capture the relations, the object domain must contain information about substances (e.g., that objects can pass through substances), or the substance domain must contain information about objects (ditto) or both. We're unsure how this overlap could be consistent with encapsulation. There are similar overlaps between object and number domains, and future research could work out these details.
Similarly, all objects consist of substances, and the causal properties of the objects are inherited in part from those substances. A toy's behavior during a collision will differ for a toy made of rubber than for an otherwise similar toy made of metal. Assuming that infants are sensitive to such differences in behavior, then if it's the object domain that predicts the nature of collisions (Spelke, Reference Spelke2022, sects. 2.3 and 2.5), it must have access to knowledge about the substances that compose the objects. But if objects and substances belong to separate encapsulated core domains, this kind of coordination is difficult to explain. Perhaps the correct domains are not objects and substances, but solid objects and nonsolid substances. However, that would leave wood, metal, and other solid substances and socks, raindrops, and other nonsolid objects in limbo until the knowledge is elaborated or refined later in development.
Issues like these reveal a tension between the core-knowledge program's expanding scope and its architectural restrictions. The experimental program that Spelke launched has successfully revealed an increasing range of knowledge that infants deploy. But this increase comes with problems of accommodating the information within core knowledge's separate fiefdoms.
Acknowledgement
The authors thank Grace Coram, Kexin Que, and Zixin Zeng for comments on an earlier version of this commentary.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.