Concept Acquisition Without Representation William Dylan Sabo Abstract: Contemporary debates in concept acquisition presuppose that cognizers can only acquire concepts on the basis of concepts they already have, and thus requires that they have at least some innate concepts. I argue that this presupposition, which I call the Conceptual Mediation Thesis, should be rejected. I argue that distinguishing between indicating states and representing states of cognizers provides the basis for an alternative account of concept acquisition. On this account, concepts are acquired via indicating states of perceptual systems. This alternative shows how concepts can be acquired without using representations, and so how a cognizer with no concepts to begin with could go on to acquire some.
Concept Acquisition Without Representation William Dylan Sabo 1 1 Introduction Contemporary debates in concept acquisition concern the primitive concepts a cognizer needs to begin with in order to be able to acquire the rest of her conceptual repertoire. Contemporary empiricists hold that a relatively small number of primitive concepts is sufficient to explain the acquisition of all other concepts. Contemporary rationalists hold that the number of primitive concepts must be quite large. But both agree that a cognizer must have some primitive concepts in order to be able to acquire any concepts at all. 1 Both positions accept what I call the Conceptual Mediation Thesis (CMT): (CMT) All acquisition of new concepts by a cognizer must be mediated by concepts that cognizer already possesses. CMT entails that any system capable of acquiring concepts must have some unacquired concepts. If unacquired concepts are innate concepts, then both positions are committed to the existence of at least some innate concepts. But I think the common ground is unwarranted. I think we should reject CMT. In this paper I propose a strategy for avoiding commitment to CMT. I explain how a cognizer might acquire some concepts without having any to start with. I will not argue directly against CMT; instead, I argue that it is unnecessary. I think that a view of concept acquisition that does not require innate concepts to explain how concepts are acquired is, other things equal, a preferable view. Questions of the innateness or otherwise of a given concept are empirical questions; we should not be committed to answers to those questions as a precondition of explaining concept acquisition at all. 1 See Fodor (1981, 1998) for an explanation of the debate along these lines.
Concept Acquisition Without Representation William Dylan Sabo 2 2 Concepts and Mental Representations I will restrict my attention to situations where a concept is are acquired by perceiving members of the extension of the concept (in contrast, for example, to cases of concepts acquired by means of learning a scientific theory). This is not because this is the only way concepts can be acquired, but because it is a central case, and it is plausibly the first kind of conceptual development in which a developing cognizer engages. I will assume, following standard versions of the Representational Theory of Mind, 2 that concepts are mental representations (and I will use both terms interchangeably in what follows). A mental representation is a mental structure with conditions of semantic evaluation, that (1) exhibits intentional inexistence: what it represents need not actually exist; (2) can therefore misrepresent: what it represents as being the case need not in fact be the case; but (3) is not stimulus-bound: its tokens can be used correctly even when those tokens are not directly caused by what the representation represents. Why accept these conditions on mental representations? Because of the explanatory goals for which mental representations are invoked. Cognizers use mental representations to adapt their behavior to features of their environment they cannot currently perceive. For mental representations to play this role, they must be able to represent what is not presently the case. This requires they be potentially correct even when what they represent is not currently present to the cognizer. When I desire to buy an eggplant, there need be no present eggplant that causes this tokening of EGGPLANT, but neither is my desire in error. Such an internal state of a cognizer can be used in planning, hypothetical reasoning, and in the control of adaptive and appropriate behavior regarding its intentional object. Since explaining such behavior is the point of invoking mental representations, internal states of a system will count as mental representations when they can be used in this way. 2 see Fodor (1975, 1998), Sterelny (1990).
Concept Acquisition Without Representation William Dylan Sabo 3 Internal states that satisfy conditions (1) and (2) but not (3) I will call indicating states, in contrast to representing states. Indicating states can respond to the presence of what they indicate, much as a mousetrap responds to the presence of a mouse. They can even respond erroneously, much as a change in temperature might, by expanding or contracting its metal parts, cause a mousetrap to snap shut when no mouse is present. But they do not represent what they indicate, any more than the mousetrap represents a mouse before striking. I will assume that representational content is constituted by some kind of mind-world relation: that a concept has the content that it does because it stands in the right sort of relation to the feature of the world it represents. There are different views about what this relation is for some it is a causal relation, for others an informational relation, for still others a teleological relation but those differences will not matter here. 3 What is important is that, on anyone s view, a concept represents some feature of the world in part because instantiations of that feature are appropriately related to tokenings of the concept. If a concept is a mental structure that represents some aspect of the world by standing in some natural relation to what it represents, then a theory of concept acquisition has to explain how that mental structure comes to stand in that relation. According to CMT, having some concepts is required to set up this relation. 3 Conceptual Mediation and Sustaining Mechanisms The most common view of how concepts mediate the acquisition of new concepts is that they are parts of those new concepts. On this view, cognizers assemble new concepts out of concepts they already have. On definitional versions of this view, the content of this new 3 For causal theories of content, see Kripke (1972), Putnam (1975); for informational theories, see Dretske (1981), Fodor (1990); for teleological theories, see Dretske (1988), Millikan (1989).
Concept Acquisition Without Representation William Dylan Sabo 4 concept is determined by the content of its constituent concepts. To acquire the concept RED SQUARE is to form a representation composed of RED and SQUARE, which the cognizer already possesses. On nondefinitional versions, the constituent concepts represent salient or typical features of the concept s extension. Acquiring the concept BIRD might involve constructing it out of concepts such as FLIES, FEATHERED, and BEAKED, for example. 4 Jerry Fodor argues that definitional versions of this view are psychologically implausible, and nondefinitional versions face problems accounting for the compositionality of concepts. He concludes that concepts are not in general acquired via construction. 5 But CMT does not require that concepts are acquired by literally constructing them from other concepts. Even if acquiring a concept is not a matter of assembling it from others, one may need to use concepts to set up the content-constitutive relation between a new concept and what it represents. This is because these relations are complex, and require other conditions to obtain as well. Suppose, for example, that my EGGPLANT concept represents eggplants because eggplants normally cause me to token EGGPLANT. The fact that they do depends on a complicated story about my perceptual and cognitive systems. If my eyes are not working properly, or if the lighting is bad, or if I do not know what an eggplant looks like, then seeing an eggplant will not cause me to token EGGPLANT. For my EGGPLANT concept to be related to eggplants so as to represent them, my perceptual and cognitive systems must be appropriately configured. If other conditions have to obtain in order for a given concept to represent what it does, then it is plausible that acquiring a concept is a matter of setting up these conditions. Following 4 For reviews of various theories of concepts along these lines, see Prinz (2002), Laurence and Margolis (1999). 5 See Fodor (1981, 1998).
Concept Acquisition Without Representation William Dylan Sabo 5 Eric Margolis 6, I will call the mechanisms that set up content-constitutive relations sustaining mechanisms. Margolis proposes that acquiring a concept is acquiring a sustaining mechanism. Your sustaining mechanism for the concept EGGPLANT may include knowing that typical eggplants are purple, oblong, and roughly a foot in length. Learning these things allows tokenings of your EGGPLANT concept to be appropriately related to eggplants in the way that a theory of content requires for your EGGPLANT concept to represent eggplants. Margolis presents this to explain how new primitive concepts can be acquired. This blocks the inference from primitive to innate. But his view seems implicitly committed to CMT, since setting up sustaining mechanisms seems to involve the use of concepts one already has. Setting up a sustaining mechanism for EGGPLANT involves assembling knowledge about what eggplants look like, and having this knowledge apparently requires you already to have concepts of particular colors, shapes, and sizes. The question then arises of how those concepts could be acquired. And if there are no plausible candidates for mediating the acquisition of color, shape, or size concepts, then we appear to be committed to the innateness of those concepts. Thinking of sustaining mechanisms as conceptual constructions requires commitment to CMT. But nothing about the idea of a sustaining mechanism itself requires implies CMT. Are there any plausible candidates that do not require concepts? 4 Perceptual Isomorphisms and Conceptual Representation My proposal is that states of our perceptual systems mediate the acquisition of new concepts. This might seem compatible with CMT, if the role perception plays in concept 6 See Margolis (1998), Laurence and Margolis (2000).
Concept Acquisition Without Representation William Dylan Sabo 6 acquisition requires that perception use mental representations. In contrast, I argue that perception s role in concept acquisition requires no appeal to representation. I will explain the relevant notion of a perceptual state by focusing on the human visual system. The visual system enters into various states in response to various patterns of irradiation of the eyes photoreceptors caused by the visible features of the environment. These states of the visual system are systematically interrelated; there are the complex relations between the kind of state caused by a particular pattern of irradiation, and the related states that would be caused by certain transformations of the pattern of irradiation. For example, the state of a visual system when presented with a red, 1 radius circle differs in systematic ways from the state that it would be in when presented with a green 1 radius circle, or again when presented with a red 2 radius circle. Changes in the presented scene correspond systematically to changes in the states of the perceptual system of the organism, in ways determined on the one hand by the effect the environment has on the perceptual systems, and on the other hand by the structure of those perceptual systems. These states of perceptual systems, such as the human visual system, are indicating states (see section 1). States of perceptual systems respond to stimulation by indicating the source of that stimulation. But this is not all; these indicating states are systematically interrelated in ways that reflect the relations between features of the world. The indicating states for two different shades of color, for example, do not simply indicate the presence of one or the other of those shades; they also reflect the relations between the colors whose presence they indicate that one is lighter than the other, for example. They do so in virtue of there being a relation between the two states paralleling the relation obtaining between those features of the world.
Concept Acquisition Without Representation William Dylan Sabo 7 There is thus an isomorphism between the indicating states of the visual system and perceptible aspects of the environment. This isomorphism obtains in virtue of the systematic interconnections that obtain between indicating states of the perceptual system, and counterpart relations that obtain between the states of the environment that the states of the perceptual system indicate. So, perceptual systems respond to features of the environment by entering into indicating states that are isomorphic to the features of the environment they indicate. But these states are indicating states, not representing states. The states of perceptual systems are stimulus-bound; they are not states that the cognizer can spontaneously reproduce. This does not mean that these states cannot be produced by other causes; but it does mean that, when such a state is produced by something other than what it represents, that tokening counts as an error. As section 1 argued, this is reason to regard such states as nonrepresentational. The picture is, then, that we have a system of perceptual states isomorphic to the perceptually accessible properties of the environment. Although these states are not as such representational states, they can explain the acquisition of representational states. 5 How to Acquire Concepts from Perception I have argued that our perceptual systems do not respond to sensory stimuli by producing sensory representations. I will now argue that these perceptual states can explain the acquisition of concepts. Recall from section 2 that the problem is to explain how a cognizer can acquire a mental structure that stands in an appropriate mind-world relation to a feature of the environment. That mental structure must satisfy the conditions on mental representations set out in section 1. I will
Concept Acquisition Without Representation William Dylan Sabo 8 argue that a system of indicating states isomorphic to features of the environment provides the basis for a solution to this problem without commitment to CMT. I will suppose that, in addition to the capacity to have perceptual states of the kind I have described, any cognizer able to acquire concepts will have capacities to record its perceptual states, and to group together perceptual states, both current and recorded, which bear certain relations to each other. I will say that these relations are relations of Ximilarity. Ximilarity is a technical term introduced to denote the relations perceptual states bear to each other in virtue of which they are isomorphic to features of the environment. Our intuitive notions of similarity may provide epistemic guidance to identifying relations of Ximilarity (as when, for example, the fact that different subjects show activation in the same brain regions while performing the same tasks provides evidence that the subjects are in Ximilar psychological states), but I emphasize that what relations, precisely, are those in virtue of which perceptual states are Ximilar is an empirical question for continuing research in cognitive science. I will now argue that a cognizer equipped with these capacities can acquire a concept. Suppose a child not yet equipped with a SQUIRREL concept is visually presented with a squirrel. That squirrel will cause a complex set of perceptual states in the child. Now suppose that the squirrel runs from right to left across the child s field of vision. In such a case there is a single thing the squirrel that presents a temporally extended and evolving appearance to the child. The location and orientation of the squirrel will change as it moves, as will the shape it presents to the child. Thus the perceptual states the squirrel causes in the child will change from moment to moment. In order for the child to keep track of the squirrel, she will have to be able to treat this evolving appearance as an appearance, or series of appearances, of the same thing,
Concept Acquisition Without Representation William Dylan Sabo 9 despite the alterations in the child s perceptual states. She will have to be able to follow its path and reorient herself and her perceptual systems to the squirrel. Explaining how the child can do this will require appeal to the kinds of perceptual states I outlined above, along with mechanisms that enable the child to keep track of, orient toward, and attend to stimuli. But none of this requires that the child have any mental representations. Orienting towards and following stimuli are within the capabilities of relatively simple mechanical devices, such as those employing motion sensors. Now suppose that the child loses perceptual contact with the squirrel, and later comes into perceptual contact with a squirrel again. Since squirrels tend to look relatively similar, the child s perceptual systems will enter into states Ximilar to those entered into upon seeing the first squirrel. The fact that these perceptual states are Ximilar will reflect the fact that they are caused by interestingly similar things in the world. Recall that we have equipped the child with the capacity to record and group together various perceptual states that it has been in and is currently in, on the basis of their Ximilarity to each other. Once this child has actually entered into some perceptual states that it groups together as Ximilar, the mechanisms that subserve these capacities will be able to group further perceptual states, including states that the child has never actually been in, with these as Ximilar in the same way. So, multiple perceptual encounters with squirrels will provide these mechanisms with the initial perceptual states that they group together, allowing them to establish grouping relations to further potential perceptual states produced by perceptions of squirrels. Having grouped a family of Ximilar perceptual states together, the child can produce a new type of inner state, into which the child will enter whenever it enters into a perceptual state appropriately Similar to this group. Then these perceptual states will then mediate the relation
Concept Acquisition Without Representation William Dylan Sabo 10 between perceptions of a feature of the environment and productions of the same mental structure this new inner state-type in response. This new state has a kind of autonomy from the perceptual states that generate it, in that a token of it can be produced without a token of any particular perceptual state being produced as well. The state might, for example, be produced by a front view of a squirrel, or a squirrel in profile, or any number of different squirrel-produced perceptual states. This, in turn, raises the possibility that the state might also come to be produced by states other than those involved in current perception. If so, then this state is no longer stimulus-bound, in the sense of section 1. Then it is no longer a state that merely indicates the presence of squirrels, but as a state that represents squirrels. Thus the child has developed a SQUIRREL concept. 7 I have explained how a child might acquire a SQUIRREL concept without requiring that she have any representational states. The perceptual states to which I have appealed are merely indicating states, not representational states. And the mechanisms for manipulating those states for recording them and for grouping them as Ximilar are relatively simple mechanisms that operate automatically, and only when activated by the relevant environmental stimulation. No representational states need to be invoked to explain these operations. So the view I have offered allows explains the acquisition of concepts through perception without appealing to any representational resources the acquisition of which the view does not allow us to explain in turn. This result shows us how to explain concept acquisition without commitment to CMT. 7 Slight complications arise from the fact that SQUIRREL is a natural kind concept; there is the possibility of things that have the same overt perceptual signs as squirrels but which have different underlying essential properties. One possible way of accounting for how a SQUIRREL concept represents squirrels, given such possibilities, is that children are predisposed to treat natural kinds as having an underlying essence: see Margolis (1998) for discussion.
Concept Acquisition Without Representation William Dylan Sabo 11 References: Dretske, F. (1981) Knowledge and the Flow of Information (Cambridge, MA: MIT Press). ------------ (1988) Explaining Behavior: Reasons in a World of Causes (Cambridge, MA: MIT Press / Bradford). Fodor, J. (1975) The Language of Thought (Cambridge, MA: Harvard University Press). ---------- (1981) The Present Status of the Innateness Controversy in J. Fodor, RePresentations: Philosophical Essays on the Foundations of Cognitive Science (Cambridge, MA: MIT Press / Bradford). ---------- (1990) A Theory of Content in J. Fodor, A Theory of Content and Other Essays (Cambridge, MA: MIT Press / Bradford). ---------- (1998) Concepts: Where Cognitive Science Went Wrong (Oxford: Clarendon Press). Kripke, S. (1972) Naming and Necessity in D. Davidson and G. Harman (eds.) Semantics of Natural Language (Dordrecht: D. Reidel). Laurence, S. and Margolis, E. (1999) Concepts and Cognitive Science in S. Laurence and E. Margolis (eds.) Concepts: Core Readings (Cambridge, MA: MIT Press / Bradford). Laurence, S. and Margolis, E. (2002) Radical Concept Nativism Cognition 86, 25-55. Margolis, E. (1998) How To Acquire a Concept Mind & Language 13:3, 347-369. Millikan, R. (1989) Biosemantics The Journal of Philosophy 86:6, 281-297. Prinz, J. (2002) Furnishing the Mind: Concepts and Their Perceptual Basis (Cambridge, MA: MIT Press / Bradford). Putnam, H. (1975) The Meaning of Meaning in K. Gunderson (ed.) Language, Mind, and Knowledge, Minnesota Studies in the Philosophy of Science VII (Minneapolis, MN: University of Minnesota Press). Sterelny, Kim (1990) The Representational Theory of Mind: An Introduction (Oxford: Basil Blackwell).