Representations involving causation play a special role in The Origin of Concepts, grounding Carey's view that there are “central representational systems with innate conceptual content that is distinct from that of core cognition systems” (Carey Reference Carey2009, p. 246). For infants' causal representations are held to be innate but not grounded in core cognition, unlike representations of objects, numbers, and agents, which do involve core cognition. After discussing the distinction between core cognition and central representational systems, I shall argue, contra Carey, that infants' causal cognition might depend on core cognition after all. This matters for two reasons. First, we are left without a clear case of innate representation outside core cognition. Second, it suggests that, as in the case of number (see Ch. 8), there may be a developmental discontinuity between infants' and adults' causal notions. If so, understanding “how the human capacity for causal representation arises” (Carey Reference Carey2009, p. 216) will require explaining how humans bootstrap themselves across the discontinuity.
Let me first outline part of what motivates the distinction between core and central cognition. Officially, core cognition is characterised by six properties (pp. 67–68), but for my purposes it is useful to start from motivation for one aspect of the distinction. At what age do infants typically first know that solid barriers stop rolling balls? This is a hard question because, as Carey explains, there is compelling evidence for apparently inconsistent answers. Infants' looking behaviours reveal that infants have different expectations about the trajectories of objects depending on the presence and positions of solid barriers (Baillargeon et al. Reference Baillargeon, Kotovsky, Needham, Sperber and Premack1995; Carey Reference Carey2009, pp. 76ff.; Spelke & Van de Walle Reference Spelke, Van de Walle, Eilan, McCarthy and Brewer1993; Wang et al. Reference Wang, Kaufman and Baillargeon2003). Yet at 2½ years, their reaching behaviours systematically indicate a failure to understand interactions (Berthier et al. Reference Berthier, De Blois, Poirier, Novak and Clifton2000; Carey Reference Carey2009, pp. 111–15; Hood et al. Reference Hood, Cole-Davies and Dias2003). Carey's view involves distinguishing “two kinds of knowledge” (p. 115) or “two types of conceptual representations” (p. 22). There are principles that are known (or, better, cognised [pp. 10–11]) in this sense: Abilities to individuate and track objects exploit the approximate truth of these principles. We can explain the sensitivity of infants to solid barriers on the hypothesis that they do know that solid barriers stop rolling balls in this sense of knowledge: Where the principle that solid barriers stop rolling balls is violated, infants cannot compute a continuous trajectory for an object, and their attention is drawn to it for the same sorts of reasons that cause them to attend to anomalous appearances and disappearances (p. 140). But in another sense, principles are known only if they can serve as reasons that explain and justify purposive actions, or only if they could in principle become elements in intuitive theories with explanatory potential (Ch. 10). We can explain the failure of 2-year-olds' reaching behaviours to deal with solid barriers on the hypothesis that they do not know that solid barriers stop rolling balls in this sense of knowledge.
If (as I think) Carey's interpretation of these apparently conflicting findings is right, we must be cautious when attributing knowledge, concepts, or representations. This applies to the case of causation. Some of the best evidence for causal representations in infancy comes from extensions of Michotte's launching paradigm (Carey Reference Carey2009, p. 218). What is the nature of these representations? Abilities to individuate and track objects involve sensitivity to cues to object identity such as continuity of movement and distinctness of surfaces. As Michotte noted, the causal representations he identified arise when there is a conflict between these cues (Michotte Reference Michotte and Miles1946/1963, p. 51). This and other evidence (see Butterfill Reference Butterfill2009, pp. 420–21, for a review) suggests that we can characterise Michottian causal phenomena in roughly the same way that Carey characterises infants' knowledge that solid barriers stop rolling balls: it is a side-effect (although perhaps a developmentally significant one) of the computations involved in core cognition of objects.
I have not yet suggested anything inconsistent with Carey's position. I also accept her arguments that (1) infants' causal representations are not limited to mechanical interactions and that (2) the Michottian phenomena are not “the [only] source of the human capacity for causal representations” (Carey Reference Carey2009, p. 243; Saxe & Carey Reference Saxe and Carey2006). However, I do object to Carey's argument for the further claims that (3) “not all of an infant's earliest causal representations are modular” (p. 240) and that (4) infants “make complex causal inferences” (p. 242). The argument for all four claims hinges on an impressive series of studies showing that infants can represent causal interactions involving changes of state as well as of location, that they are sensitive to colliding objects' size or weight, and that they expect animate and inanimate objects to occupy different causal roles. These findings support (1) and (2) but not (3) and (4). Why not? Carey is surely right that these findings cannot be explained by supposing that there is core cognition of (or a module for) causation. But this is compatible with the hypothesis that sensitivity to causal principles is a feature of several individual modules. That many possibly modular processes require sensitivity to causal principles does not imply that any such principles are central. For example, cognition of speech is sometimes thought to be modular (Liberman & Mattingly Reference Liberman, Mattingly, Hoffman and Palermo1991) and might even be characterised as a species of core cognition. Speech cognition involves sensitivity to causation in ways analogous to those discussed in The Origin of Concepts. For example, it requires categorising things by whether they are potential producers of speech, and identifying which bits of speech are coming from a single source. In fact, on one view, cognition of speech involves identifying the phonetic activities most likely to be causing observable sounds and movements (Liberman & Mattingly Reference Liberman and Mattingly1985). By itself, this is neither evidence for the existence of non-modular causal representations, nor for causal representations that are not aspects of core cognition.
But why think that causal representations are initially all embedded in core cognition?
As in the case of number, the answer concerns signature limits. Children in their second and third years of life show limited abilities to reason explicitly about the causal powers of solid objects and of agents. The principle considered earlier, that solid barriers stop rolling balls, is as much about causation as it is about objects. If infants' causal representations were nonmodular and if infants could make complex causal inferences, how could they fail to appreciate that solid barriers stop rolling balls?
If it is possible that infants' and adults' causal representations are discontinuous in the same sense that their numerical representations are (Ch. 4 and 8), how might humans bootstrap themselves across the discontinuity? In the case of number, numeral lists and counting routines play a key role. Tool use may play an analogous role in bootstrapping causal representations. Basic forms of tool use may not require understanding how objects interact (Barrett et al. Reference Barrett, Davis and Needham2007; Lockman Reference Lockman2000), and may depend upon core cognition of contact-mechanics (Goldenberg & Hagmann Reference Goldenberg and Hagmann1998; Johnson-Frey Reference Johnson-Frey2004). Experience of tool use may in turn assist children in understanding notions of manipulation, a key causal notion (Menzies & Price Reference Menzies and Price1993; Woodward Reference Woodward2003).
Perhaps, then, non-core capacities for causal representation are not innate, but originate with experiences of tool use.
You have
Access
Representations involving causation play a special role in The Origin of Concepts, grounding Carey's view that there are “central representational systems with innate conceptual content that is distinct from that of core cognition systems” (Carey Reference Carey2009, p. 246). For infants' causal representations are held to be innate but not grounded in core cognition, unlike representations of objects, numbers, and agents, which do involve core cognition. After discussing the distinction between core cognition and central representational systems, I shall argue, contra Carey, that infants' causal cognition might depend on core cognition after all. This matters for two reasons. First, we are left without a clear case of innate representation outside core cognition. Second, it suggests that, as in the case of number (see Ch. 8), there may be a developmental discontinuity between infants' and adults' causal notions. If so, understanding “how the human capacity for causal representation arises” (Carey Reference Carey2009, p. 216) will require explaining how humans bootstrap themselves across the discontinuity.
Let me first outline part of what motivates the distinction between core and central cognition. Officially, core cognition is characterised by six properties (pp. 67–68), but for my purposes it is useful to start from motivation for one aspect of the distinction. At what age do infants typically first know that solid barriers stop rolling balls? This is a hard question because, as Carey explains, there is compelling evidence for apparently inconsistent answers. Infants' looking behaviours reveal that infants have different expectations about the trajectories of objects depending on the presence and positions of solid barriers (Baillargeon et al. Reference Baillargeon, Kotovsky, Needham, Sperber and Premack1995; Carey Reference Carey2009, pp. 76ff.; Spelke & Van de Walle Reference Spelke, Van de Walle, Eilan, McCarthy and Brewer1993; Wang et al. Reference Wang, Kaufman and Baillargeon2003). Yet at 2½ years, their reaching behaviours systematically indicate a failure to understand interactions (Berthier et al. Reference Berthier, De Blois, Poirier, Novak and Clifton2000; Carey Reference Carey2009, pp. 111–15; Hood et al. Reference Hood, Cole-Davies and Dias2003). Carey's view involves distinguishing “two kinds of knowledge” (p. 115) or “two types of conceptual representations” (p. 22). There are principles that are known (or, better, cognised [pp. 10–11]) in this sense: Abilities to individuate and track objects exploit the approximate truth of these principles. We can explain the sensitivity of infants to solid barriers on the hypothesis that they do know that solid barriers stop rolling balls in this sense of knowledge: Where the principle that solid barriers stop rolling balls is violated, infants cannot compute a continuous trajectory for an object, and their attention is drawn to it for the same sorts of reasons that cause them to attend to anomalous appearances and disappearances (p. 140). But in another sense, principles are known only if they can serve as reasons that explain and justify purposive actions, or only if they could in principle become elements in intuitive theories with explanatory potential (Ch. 10). We can explain the failure of 2-year-olds' reaching behaviours to deal with solid barriers on the hypothesis that they do not know that solid barriers stop rolling balls in this sense of knowledge.
If (as I think) Carey's interpretation of these apparently conflicting findings is right, we must be cautious when attributing knowledge, concepts, or representations. This applies to the case of causation. Some of the best evidence for causal representations in infancy comes from extensions of Michotte's launching paradigm (Carey Reference Carey2009, p. 218). What is the nature of these representations? Abilities to individuate and track objects involve sensitivity to cues to object identity such as continuity of movement and distinctness of surfaces. As Michotte noted, the causal representations he identified arise when there is a conflict between these cues (Michotte Reference Michotte and Miles1946/1963, p. 51). This and other evidence (see Butterfill Reference Butterfill2009, pp. 420–21, for a review) suggests that we can characterise Michottian causal phenomena in roughly the same way that Carey characterises infants' knowledge that solid barriers stop rolling balls: it is a side-effect (although perhaps a developmentally significant one) of the computations involved in core cognition of objects.
I have not yet suggested anything inconsistent with Carey's position. I also accept her arguments that (1) infants' causal representations are not limited to mechanical interactions and that (2) the Michottian phenomena are not “the [only] source of the human capacity for causal representations” (Carey Reference Carey2009, p. 243; Saxe & Carey Reference Saxe and Carey2006). However, I do object to Carey's argument for the further claims that (3) “not all of an infant's earliest causal representations are modular” (p. 240) and that (4) infants “make complex causal inferences” (p. 242). The argument for all four claims hinges on an impressive series of studies showing that infants can represent causal interactions involving changes of state as well as of location, that they are sensitive to colliding objects' size or weight, and that they expect animate and inanimate objects to occupy different causal roles. These findings support (1) and (2) but not (3) and (4). Why not? Carey is surely right that these findings cannot be explained by supposing that there is core cognition of (or a module for) causation. But this is compatible with the hypothesis that sensitivity to causal principles is a feature of several individual modules. That many possibly modular processes require sensitivity to causal principles does not imply that any such principles are central. For example, cognition of speech is sometimes thought to be modular (Liberman & Mattingly Reference Liberman, Mattingly, Hoffman and Palermo1991) and might even be characterised as a species of core cognition. Speech cognition involves sensitivity to causation in ways analogous to those discussed in The Origin of Concepts. For example, it requires categorising things by whether they are potential producers of speech, and identifying which bits of speech are coming from a single source. In fact, on one view, cognition of speech involves identifying the phonetic activities most likely to be causing observable sounds and movements (Liberman & Mattingly Reference Liberman and Mattingly1985). By itself, this is neither evidence for the existence of non-modular causal representations, nor for causal representations that are not aspects of core cognition.
But why think that causal representations are initially all embedded in core cognition?
As in the case of number, the answer concerns signature limits. Children in their second and third years of life show limited abilities to reason explicitly about the causal powers of solid objects and of agents. The principle considered earlier, that solid barriers stop rolling balls, is as much about causation as it is about objects. If infants' causal representations were nonmodular and if infants could make complex causal inferences, how could they fail to appreciate that solid barriers stop rolling balls?
If it is possible that infants' and adults' causal representations are discontinuous in the same sense that their numerical representations are (Ch. 4 and 8), how might humans bootstrap themselves across the discontinuity? In the case of number, numeral lists and counting routines play a key role. Tool use may play an analogous role in bootstrapping causal representations. Basic forms of tool use may not require understanding how objects interact (Barrett et al. Reference Barrett, Davis and Needham2007; Lockman Reference Lockman2000), and may depend upon core cognition of contact-mechanics (Goldenberg & Hagmann Reference Goldenberg and Hagmann1998; Johnson-Frey Reference Johnson-Frey2004). Experience of tool use may in turn assist children in understanding notions of manipulation, a key causal notion (Menzies & Price Reference Menzies and Price1993; Woodward Reference Woodward2003).
Perhaps, then, non-core capacities for causal representation are not innate, but originate with experiences of tool use.