Two central questions for theories of concepts are: first, what is the nature of the developmentally primitive conceptual basis that humans are endowed with; and, second, what sorts of mechanisms are available for expanding this basis to capture the adult conceptual repertoire. I will focus on Carey's answer to the second question (Carey Reference Carey2009), which centers on a “Quinian bootstrapping” mechanism for concept learning.
The goal of bootstrapping is to arrive at a new set of primitive concepts that are incommensurable with the ones the learner now possesses; that is, whose content cannot be captured in terms of any of the concepts possessed initially. The first stage of bootstrapping occurs when a learner encounters a set of interrelated explicit public symbols, such as the sentences that compose a scientific theory or the formal notation of mathematics (p. 306). These public symbols are not initially mapped onto any already existing concepts. Rather, they are uninterpreted (or partially interpreted), hence largely meaningless to the learner. These placeholders are then taken up by various “modeling processes”: abstract forms of theoretical inference such as abduction, induction, and analogical reasoning that provide them with their content. Eventually, these symbols come to have conceptual content in virtue of acquiring a stable conceptual role in a new theoretical structure (p. 418).
From this account, it appears that Quinian bootstrapping requires language or another external representational medium. These give the vehicles that one learns to manipulate and that become endowed with content by the end of the process. But it is not clear why these representational media are essential. It is true that conceptual change is often driven by learning a theory in some social context, such as the lab or the classroom. Therefore, external symbols are necessary to convey the theory to new learners. Grasping the theory itself, however, involves constructing mental representations corresponding to its new theoretical terms and the propositions that they participate in. This process seems independent of language.
For example, Carey notes that in the case of Kepler's explanation for why the planets revolve around the sun, he initially entertained the hypothesis that “something in the sun causes planets to move” (p. 427), which contains a linguistic placeholder structure. This structure was given various labels by Kepler (“anima motrix,” “vis motrix”), but when introduced, it was just as a thing, whatever it might be, that produces the motion of the planets. Introducing this new concept involves only hypothesizing the existence of a certain type of entity. Therefore, in logical terms, this placeholder is just the expression of a bound variable of the form “∃x.” If central cognition has this minimal quantificational apparatus, then the equivalent thought containing a placeholder representation should be formable as well. Carey's own discussion of the emergence of quantification gives convincing reasons to think that it does (pp. 254–263). In addition, there is reason to think that creating new primitive representations occurs in other cognitive domains as well, such as perceptual categorization (Schyns et al. Reference Schyns, Goldstone and Thibaut1998).
If the actual act of coining a new placeholder representation is not language-dependent, what is the role of language in concept learning? Probably there is no one single role that it plays, but a major one is that it serves as a signal to the child of the presence of an important type of thing in the environment. As Carey notes, linguistic labels are treated as special in a way that other conventional and natural signs are not (Xu Reference Xu2002). But as a category indicator, language is not unique. The presence of a category of interest can be signaled in many ways. Perception is one: objects presenting a surprising or novel appearance, such as one's first coelacanth or kangaroo, may also belong to new categories. In addition, causal powers and relations are signals of the presence of interesting categories. If there is a set of phenomena that display the signature of belonging to a common system of causal relations, then there is reason to posit some underlying – but unperceived – cause tying them together. That children attend to such factors is attested by the range of essentialist reasoning that they display (Gelman Reference Gelman2003). It is also attested in more mundane ways by their use of causal properties to classify objects (Gopnik & Sobel Reference Gopnik and Sobel2000; Nazzi & Gopnik Reference Nazzi and Gopnik2003). There are undoubtedly other sources of evidence that can be brought to bear, but perception, language, and causation are prima facie reliable (if sometimes conflicting) indicators of categories. And coining new concepts is performed in response to the detection of categories that are likely to prove useful for social, practical, or theoretical purposes.
Learning a new concept for a kind, property, substance, event, or individual may be a piecemeal or atomistic affair; think here of adding a new animal or food concept to one's repertoire. Quinian bootstrapping as Carey describes it often involves acquiring a set of interrelated concepts, such as the rational numbers or the adult's weight/density concepts (p. 370). It is thus a locally holistic process. The relationship between the two processes is that coining concepts for new categories is an essential prerequisite to building larger knowledge structures that include them. Kepler needed to coin a concept for the force emitted by the sun in order to hypothesize about its nature. Röntgen needed to coin a concept of X-rays in order to describe how they produced their characteristic effects. We need to conceptualize a new species as a distinct grouping in order to begin theorizing about its ethology, evolutionary history, and so on. To engage in Quinian bootstrapping, one needs theories, even local ones, and theories need theoretical concepts. Coining is precisely an atomistic process that can produce these concepts (Weiskopf Reference Weiskopf2008).
Carey holds that Quinian bootstrapping is one mechanism among many for producing conceptual change and discontinuity, albeit an important one. This seems correct. At a general level, I suggest that we see the capacity to coin new mental representations in response to an open-ended range of evidence and epistemic conditions as the common capacity that underlies much of our concept learning. This mechanism is the bottleneck through which many other pathways to concept learning flow, including many of those involved in Quinian bootstrapping, such as differentiation and coalescence. The capacity for creating new representations that is involved in these processes is one that may use language as one cue among many, but which is not in and of itself dependent upon language.
You have
Access
Two central questions for theories of concepts are: first, what is the nature of the developmentally primitive conceptual basis that humans are endowed with; and, second, what sorts of mechanisms are available for expanding this basis to capture the adult conceptual repertoire. I will focus on Carey's answer to the second question (Carey Reference Carey2009), which centers on a “Quinian bootstrapping” mechanism for concept learning.
The goal of bootstrapping is to arrive at a new set of primitive concepts that are incommensurable with the ones the learner now possesses; that is, whose content cannot be captured in terms of any of the concepts possessed initially. The first stage of bootstrapping occurs when a learner encounters a set of interrelated explicit public symbols, such as the sentences that compose a scientific theory or the formal notation of mathematics (p. 306). These public symbols are not initially mapped onto any already existing concepts. Rather, they are uninterpreted (or partially interpreted), hence largely meaningless to the learner. These placeholders are then taken up by various “modeling processes”: abstract forms of theoretical inference such as abduction, induction, and analogical reasoning that provide them with their content. Eventually, these symbols come to have conceptual content in virtue of acquiring a stable conceptual role in a new theoretical structure (p. 418).
From this account, it appears that Quinian bootstrapping requires language or another external representational medium. These give the vehicles that one learns to manipulate and that become endowed with content by the end of the process. But it is not clear why these representational media are essential. It is true that conceptual change is often driven by learning a theory in some social context, such as the lab or the classroom. Therefore, external symbols are necessary to convey the theory to new learners. Grasping the theory itself, however, involves constructing mental representations corresponding to its new theoretical terms and the propositions that they participate in. This process seems independent of language.
For example, Carey notes that in the case of Kepler's explanation for why the planets revolve around the sun, he initially entertained the hypothesis that “something in the sun causes planets to move” (p. 427), which contains a linguistic placeholder structure. This structure was given various labels by Kepler (“anima motrix,” “vis motrix”), but when introduced, it was just as a thing, whatever it might be, that produces the motion of the planets. Introducing this new concept involves only hypothesizing the existence of a certain type of entity. Therefore, in logical terms, this placeholder is just the expression of a bound variable of the form “∃x.” If central cognition has this minimal quantificational apparatus, then the equivalent thought containing a placeholder representation should be formable as well. Carey's own discussion of the emergence of quantification gives convincing reasons to think that it does (pp. 254–263). In addition, there is reason to think that creating new primitive representations occurs in other cognitive domains as well, such as perceptual categorization (Schyns et al. Reference Schyns, Goldstone and Thibaut1998).
If the actual act of coining a new placeholder representation is not language-dependent, what is the role of language in concept learning? Probably there is no one single role that it plays, but a major one is that it serves as a signal to the child of the presence of an important type of thing in the environment. As Carey notes, linguistic labels are treated as special in a way that other conventional and natural signs are not (Xu Reference Xu2002). But as a category indicator, language is not unique. The presence of a category of interest can be signaled in many ways. Perception is one: objects presenting a surprising or novel appearance, such as one's first coelacanth or kangaroo, may also belong to new categories. In addition, causal powers and relations are signals of the presence of interesting categories. If there is a set of phenomena that display the signature of belonging to a common system of causal relations, then there is reason to posit some underlying – but unperceived – cause tying them together. That children attend to such factors is attested by the range of essentialist reasoning that they display (Gelman Reference Gelman2003). It is also attested in more mundane ways by their use of causal properties to classify objects (Gopnik & Sobel Reference Gopnik and Sobel2000; Nazzi & Gopnik Reference Nazzi and Gopnik2003). There are undoubtedly other sources of evidence that can be brought to bear, but perception, language, and causation are prima facie reliable (if sometimes conflicting) indicators of categories. And coining new concepts is performed in response to the detection of categories that are likely to prove useful for social, practical, or theoretical purposes.
Learning a new concept for a kind, property, substance, event, or individual may be a piecemeal or atomistic affair; think here of adding a new animal or food concept to one's repertoire. Quinian bootstrapping as Carey describes it often involves acquiring a set of interrelated concepts, such as the rational numbers or the adult's weight/density concepts (p. 370). It is thus a locally holistic process. The relationship between the two processes is that coining concepts for new categories is an essential prerequisite to building larger knowledge structures that include them. Kepler needed to coin a concept for the force emitted by the sun in order to hypothesize about its nature. Röntgen needed to coin a concept of X-rays in order to describe how they produced their characteristic effects. We need to conceptualize a new species as a distinct grouping in order to begin theorizing about its ethology, evolutionary history, and so on. To engage in Quinian bootstrapping, one needs theories, even local ones, and theories need theoretical concepts. Coining is precisely an atomistic process that can produce these concepts (Weiskopf Reference Weiskopf2008).
Carey holds that Quinian bootstrapping is one mechanism among many for producing conceptual change and discontinuity, albeit an important one. This seems correct. At a general level, I suggest that we see the capacity to coin new mental representations in response to an open-ended range of evidence and epistemic conditions as the common capacity that underlies much of our concept learning. This mechanism is the bottleneck through which many other pathways to concept learning flow, including many of those involved in Quinian bootstrapping, such as differentiation and coalescence. The capacity for creating new representations that is involved in these processes is one that may use language as one cue among many, but which is not in and of itself dependent upon language.