On the Relationship between Naturalistic Semantics and Individuation Criteria for Terms in a Language of Thought Author(s): Robert D. Rupert Reviewed work(s): Source: Synthese, Vol. 117, No. 1 (1998/1999), pp. 95-131 Published by: Springer Stable URL: http://www.jstor.org/stable/20118099 . Accessed: 23/02/2012 20:48 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Springer is collaborating with JSTOR to digitize, preserve and extend access to Synthese. http://www.jstor.org ROBERT D. RUPERT ON THE RELATIONSHIP BETWEEN NATURALISTIC SEMANTICS AND INDIVIDUARON CRITERIA FOR TERMS IN A LANGUAGE OF THOUGHT* Naturalistically minded philosophers hope to identify a privileged nonse ABSTRACT. a a mental m and that which m represents, that holds between representation to m. The natural status underwrites the assignment of reference privileged for individuating criterion ist can accomplish this task only if she has in hand a nonsemantic to for the naturalist characterize it would be mental m, representations: question-begging as 'the representation with If of assigning such and such content'. for the purpose content, mantic relation relation whose individuate mental systems theory, we find using the tools of dynamical representations in the cognitive characterized that a given mental emerges nonsemantically, representation, between the subject and her environment. At least interactions system as the result of causal we for the most approach function basic of our mental to individuation of the subject's I argue that the dynamical representations, of a theory that assigns increases the plausibility causal history. systems-based as a reference 1. INTRODUCTION states and their role in the In an attempt to explain the nature of mental often begin with assumptions of the lives of human beings, philosophers a sort: (1) there exists in the human mind/brain system of rep following resentations in LOT')1 (sometimes among which connections 'terms in a language of thought', or 'terms and (2) the causal connections; mental representations partly explain human called there are causal holding between behavior. In order that assumptions least the two following questions at (1) and (2) be wholly legitimate, should be answered: How are mental individuated? and How do these representations acquire representations them as representa intentional content,2 the possession of which qualifies to the claim that the presence of to give substance tions and is supposed representations I aim to show helps explain human thought and behavior? that what appears to be the most promising first question, about individuation, count as an acceptable what might content. determinants of mental ^ has In this paper, to the answer important implications regarding answer to the second, concerning the Synthese 111: 95-131, 1999. W% ? 1999 Kluwer Academic Publishers. Printed in theNetherlands. 96 ROBERT D. RUPERT of other (1) and (2) are often made by philosophers in this paper specifically the arguments address to inclined mental understand attempts by naturalistically philosophers Naturalism in the philosophy of mind holds that the human representation. While assumptions orientations, theoretical mind makes is a completely organic, naturalism a distinctive to the methods what Furthermore, phenomenon. orientation is its commitment philosophical biological in investigating any and by the natural sciences all parts of reality, including, of course, the human mind. So naturalism on a more not which it also below, implies only methodology, implies that our theories of individuation and content for LOT terms should be spelled employed out in nonsemantic terms.3 as follows. of the paper, I proceed on naturalistic methodology, I outline a view In the remainder remarks individuate mantically from research After of how some brief to nonse a view mental representations, currently emerging that treats the mind as a dynamical system. Here I note the view to provide a better picture promise of the dynamical systems-based than we have had to date of how a cognitive rep system, one that employs resentations, perceptual I present can be integrated with systems or the muscular some the other component systems (e.g., the human of the system) subject. While reasons for thinking that we should adopt the dynamical to the study of cognition, it is not my intention systems-based approach to provide detailed in and weighty arguments support of this approach. to is view to understand of the present enough My goal dynamics-based a what its implications might be for theory of content for terms in LOT. In particular, I wish to bring tation to bear on the debate for LOT mental terms. I argue the dynamics-based view of mental over the viability of a naturalistic that given the dynamical systems-based represen semantics view of a certain and given view of the naturalistic philo representations, a terms of the defender naturalistic semantics for LOT sophical project, often believed should question the relevance of many thought experiments to be respectable coin in discussions of mental content. The philosophical striking result here is that a subject's actual causal history may play a larger role in determining the content of her LOT terms than has been widely thought. While be helpful a thought I present in more detail in later sections, the argument it may at this point to preview my basic line of reasoning. Typically, a counterfactual situation, experiment requires our imagining alone that constrains the fram and normally, it is metaphysical possibility this hypothetical In that describe circumstance. ing of the counterfactuals view of the nature and the acqui contrast, I argue that the dynamics-based a thoroughgoing with sition of mental representations, naturalism, together NATURALISTICSEMANTICSAND INDIVIDUATION 97 on the acceptable framing conditions places more stringent requirements that are the very stuff of philosophical for those counterfactuals thought If a dynamical about meaning. theory of indi systems-based experiments tells us that a given LOT for mental conditions viduation representations speaking, be tokened under circumstances physically to ask, as part of a philosophical for a naturalist our attribution of content would intuition-based what c, then term t cannot, it is fruitless experiment, t to be tokened we should reevaluate the role of thought in c. Accordingly, theories for of naturalistic semantic in critical discussions experiments LOT terms. Upon r??valuation, to theories of content according by the subject's actual history, counterexamples legitimate thought be were we find an increased an LOT to which this because attached plausibility is fixed term's reference it ismuch more to actual history-based difficult to give theories.4 2. REMARKS ON METHODOLOGY In basic that when agreement with Michael faced with (what we Devitt I assume (1994, pp. 565-569),5 our a natural is) phenomenon, suspect of should begin with an examination of that phenomenon investigation we or most obvious wish instances. Thus, if its most widely recognized we as a should to understand natural begin by phenomenon, representation the best examples we have of beings who use full-fledged repre inspecting we the beings in whose use of representations should examine sentations: we are most confident, i.e., human beings.6 As Devitt rightly points out, even our strongest intuitions as to which cases best exemplify being an F = a are in F that this case, defeasible, (in system employs representations) as our intu Devitt also But notes, strongest (Devitt 1994, p. 568). principle items have property F should serve as the starting itions regarding which point of our attempt is that the naturalist to figure out what should endeavor it is to have F. The upshot, to understand then, as it representation any general, species-independent actually exists in humans before making or intentionality. claims about representation (Of course, results in ethol our can inform and animal repre ogy study of human mental physiology this While the field takes is that latter sentations; my point precedence. it seems more strike some as unnecessarily chauvinistic, may that the very human abilities [e.g., flexible plausible when we recognize use of language, creative recombination of concepts, thoughts about the and to or that led scientists future] early cognitive reject behaviorism past position take internal mental lacking for the most seriously are abilities representations part in nonhuman species.) that seem to be 98 ROBERTD. RUPERT cannot We avoid uralistic the question of individuation when if we wish of mental a nat developing to say that a term of its nonsemantic theory representation: semantic value in virtue of certain acquires some characterization of what one of relations, we should have in mind those terms is and what sets it apart from the others (i.e., some characteri in LOT zation other than 'the term with to examine the relation such and such content'). Before moving on between dynamical models of the mind and LOT term individuation, note the limitations of Devitt's suggested methodology were we to attempt to apply it to the question of individuation. As Devitt as describes naturalistic semantic should take initial data it, theory points our semantic intuitions as to which items have specific semantic properties that p, where p's value is is, for example, meaning (A semantic property to be some specific content). However, to construct a complete naturalistic semantic theory for LOT terms, we need a detailed theory of the nature of the items that are supposed to bear these semantic properties. Here humans seem to have nothing analogous would to their semantic intuitions, i.e., that informs them of the nonsemantic individua nothing pretheoretically a human thinks of water today, how is tion criteria for LOT terms. When she to know the same LOT that she is tokening about water she thought individuation yesterday? Contrast term she tokened this with in a natural the case about when of term the meaning language. Knowing nothing of two, token, spoken words, a person can make a fairly reliable guess as to or not, the two are instances of the same type, especially if these whether, two words are from a language she speaks. With respect to terms in LOT, to the capacity for auditory though, there seems to be nothing analogous discrimination words that underlies whose meanings theory of individuation as the initial data points about the similarity of spoken judgements are not known. Thus, in developing a nonsemantic for LOT terms, we cannot take thinkers' intuitions for our theorizing (the way, for example, a linguist the range of acoustical variation speakers of a given to allow in the production of a single phoneme).7 We should look elsewhere, criteria for LOT terms, and then, for individuation systems theory appears to have much to offer in this regard. dynamical in identifying might language are willing 3. THE DYNAMICAL SYSTEMS-BASED APPROACH like to be able to explain how the various LOT terms differ from reference to their content. We might reasonably making the relevant free parameters of the human physical begin by cataloguing the LOT terms appear. By 'free parameters' I mean the system in which state of dimensions which the relevant the independent, noncognitive along We would each other without NATURALISTICSEMANTICSAND INDIVIDUATION 99 portion of the system can vary; of these parameters, we should be particu to larly interested in those whose variations we suspect make a difference the individuation of LOT terms. If we were to take the idea of a language of as the identify the relevant free parameters thought too literally, we might then 'cat' in LOT would letters out of which LOT terms are constructed; be different from 'bat' in LOT in virtue of the terms' having distinct initial however, given our general approach does not seem promising, in the brain. More plausibly, we might failure to find such letters written to locate properties of the brain on the consult contemporary neuroscience letters. This basis of which we can individuate of relevant would seem a subject's LOT term types. Here the set to consist of the firing rates of neurons, parameters such as the brain's temperature, together with values of other parameters, the nervous system's components that affect the way in which operate. that Note that it is incredibly difficult to construct a model of cognition are takes into account all of what we might relevant suspect parameters. For this reason most constructs meant theorists offer simplified cognitive at to capture something the risk of excluding about cognition important some pertinent details. Consider one the fact that there are approximately trillion neurons in a human (Stillings et al., 1987, p. 267; Churchland and Sejnowski, 1992, p. 51). If we begin our cognitive theorizing treating rates of neural firing as our free parameters, then idealization and sim seem unavoidable.8 This is true even of connectionist models, plification are to which be anatomically realistic in a way that other models supposed brain are not: such networks of cognition typically consist units relative to the number of neurons in the brain. of a small number of the free parameters of (the relevant decided how to characterize a we can the tools of dynamical part of) systems employ given system, that system. Begin with the idea of a state space. The theory to analyze state space is an abstract collection of possible overall states of the system, Having one state corresponding to each possible combination of values of the a standard light switch: to keep things simple, free parameters. Consider up (on) or down imagine that the switch can be in only two positions, with state space consists of the two possible the system's (off); accordingly, states of the system, one where the switch is up, and one where it is down. to a system's In addition state space exists a mathematical space called a 'phase space'. as to this is also the referred 'state (Frequently space', when necessary.) The phase space is an leaving context to disambiguate n -dimensional n mathematical space, where equals the number of free in the system, the various combinations of values of which parameters state of the system; the phase represents a possible means a vector of field, how the system will evolve by each space specifies, from any given 100 ROBERT D. RUPERT state. of vectors in the field is often referred (The pattern of distribution to as the 'shape' of the phase space or as the 'phase portrait'.) The phase (or difference space's shape is determined by a set of differential equations in systems that move states) that define what through discrete the 'dynamics' of the system. The light switch's phase space consists of one dimension that can assume only two values, the phase por trait of which will likely bore us to tears: as I have described it, the switch equations are called a static system with two discrete is essentially states; in the absence of disturbance from outside the system, the switch simply stays in the state in which it begins. The equation governing the light switch's behavior takes as input the initial state and yields as output that same state (although if it is to accurately capture the situation of a real light switch, the equation must also include an expression for influence from the outside, allowing a as closed which complicates matters); thus, system, its phase space is no more than two fixed numerical Van Gelder and Port of dynamical the application points.9 reasons for thinking that to will the systems theory system cognitive into the nature of human thought; in doing (1995) lay out many engender important insights so, Van Gelder and Port argue for a dynamical (DS)10 view systems-based of cognition.11 They are at pains, and for good reason, to explain exactly what advantage is gained by employing the tools of dynamical systems rather than those of the "old-fashioned approach". computational are in (That they give such reasons is important, for many philosophers as describing how the central nervous system clined to think of DS models that DS have much to do with models instantiates amind, without thinking theory the nature of cognition.) Of Van Gelder and Port's concerns, most germane is their worry that the computational theorist cannot for present purposes and the stimuli at the sensory periphery bridge the gulf between physical over that transductions and entities abstract, sensory produce symbolic are defined (Van Gelder and Port, 1995, pp. processes computational The 26-30). activity of sensing the world around us begins as a dynamical there is no need to iden interaction with that world. On the DS approach, nervous at in the the which electrical system become impulses tify point which to firing neurons at the assigned of the (or groups thereof) are values along dimensions sensory perimeter same phase space in which one finds LOT terms. Thus, on the DS view, there is no "theoretical gap between cognitive systems and their surrounds" abstract symbols, for the numerical values (Van Gelder and Port, 1995, p. 28), as there is on the computationalist's are contin of cognition the mathematics view. Taking the DS approach, uous with For this reason, DS theory of transduction. the mathematics yields a deep understanding of how LOT terms are introduced into the 101 NATURALISTICSEMANTICSAND INDIVIDUATION system by treating them as modeled cognitive brute sensory signals that are often the causes it follows from the methodology Furthermore, if we are to discover the nature after or continuous with we of mental the terms. of tokenings of LOT in Section endorsed 2 that should seek representation, first to understand how LOT terms appear in the human physical system, are for physically embodied the the best, people perhaps only, examples we have of full-blooded systems. representational To some readers, itmay seem odd that I talk approvingly about the DS to speak of mental representations, this because approach while continuing a as scientists who advocate the mind many cognitive treating dynamical bent with regard to mental system have an eliminativist representation. some eliminativist I shall consider in more detail in the closing claims what I take to be the relevance of the DS section, after first explaining to naturalistic For the time being, however, semantics. I hope approach not will suffice: it should the two following be noted that all first, points in the DS camp embrace eliminativism;12 scientists furthermore, cognitive of the sweeping eliminativist there is good reason to be suspicious claims that sometimes tinued issue from those DS talk of mental theorists who Thus, representations.13 view that the DS approach can tell us something other than that there are none. representations, 4. DS VIEWS, the value of con doubt I provisionally important adopt the about mental CAUSAL HISTORY, AND THE INTRODUCTION TERMS OF LOT I now indulge in a of the discussion, prevent a variety of possible misunderstand set the stage minor digression, for the remainder To which may and between what Iwill call consequent ings objections. Let us distinguish - terms that Fodor terms basic' LOT (1990a, p. 92) describes 'semantically as having their content fixed in an atomistic fashion - and those LOT terms whose content mental states. Two is fixed partly by the content-laden contribution serve to illustrate of other this distinction: contrasting examples one hand, it's plausible that the infant's LOT term 'object' emerges and acquires its reference independently of any very early in development content-laden contribution of other mental states.14 On the other hand, On to use the English term intentions example of speakers' same to refer to all stuff of the liquid kind as the samples con them during the period when the term 'water' is being introduced recall Putnam's 'water' fronting into the language plays an essential to H2O. Assuming (Putnam, 1975). Here the content role in limiting the representational that a roughly analogous process of the term content sometimes of 'liquid' 'water' fixes the 102 ROBERT content of terms determinate D. RUPERT in LOT, we will have cases where LOT in virtue of the contributions made extension terms acquire by other rep that already have their extensions fixed. (In fact, this would seem to be the case for most LOT terms.) It is beyond the scope of this to this distinction between make basic mental precise semantically piece contents and those whose estab depend upon previously representations resentations 1996, p. 424, for a more rigorous (see Bartsch, description seem to be a similar distinction). two comments of what would However, are in order: First, I do not mean to be suggesting that semantically basic lished content mental content representations is constructed. semantically other LOT contribute basic serve as definitional My point representations and that once terms, to the fixation simples out of which all other is only that the intentional content of the is fixed independently of the content of their content of content is fixed, such content can terms. Second, it is of of other LOT to bear inmind that the points Imake in the remain paramount importance basic LOT terms, even der of this paper apply most clearly to semantically chosen for ease of exposition when examples (as with suggest otherwise the discussion of 'horse' in Section 6). of natural language processing, Jeffrey regions in a state space (Elman, 1995). LOT This way of putting however.15 Which things is a bit misleading, on more a terms a subject can be said to possess than depends simple of the state space into quadrants or some such; it depends also partitioning of DS models In his development identifies LOT terms with Elman on the phase portrait. For a subject to possess an LOT term, a certain point (or region) in the subject's phase space must have the property of being an attractor in the phase space: it should be a point (or region) p in the phase to space such that when the system passes through states that correspond near p (i.e., points in what is called /?'s 'attractor basin'), the system of toward p. Generally speaking, we should limit our identification a given subject's LOT terms to attractors because the property of being points heads of the upon a point a special status in the dynamics no ever that the will reach There is system guarantee system: cognitive a state that corresponds to a given attractor; however, there is a much an attractor greater remain confers that the system will come close to an attractor at a state corresponding likelihood fixed to, pass through, or than there is that the close to, pass through, or remain fixed at a state that a to to corresponds randomly chosen point in the phase space. According one moves attractor to from the DS view, the cognitive system typically in between). It would another (perhaps with periods of chaotic behavior system will seem come ill founded, then, to identify a specific set of coordinates in the phase NATURALISTICSEMANTICSAND INDIVIDUATION 103 to an LOT term simply because such coordinates space as corresponding are there to be identified. Even were we to associate one of these points with an LOT term, we would not want to say that the subject possesses the concept associated with that that term or that the subject understands term. A dynamical model of speech comprehension can, for example, as a a state of the 'electron' space identify subject's region. But region if the subject has never heard of electrons or even entertained the idea of LOT then even if an experimenter says to the subject "think particles, about electrons", the subject's cognitive system will not pass through what we might want to call the 'electron' region in the subject's state space. The subatomic like her phonological rep subject might pass through a state in something we can of the English resentation space that identify with her recognition word this would not be the same as having acquired 'electron'; however, term the LOT of an electron. is no attractor as it's normally conceived of, i.e., as the concept terms the in of phase space again, there simply Thinking we as to the LOT term that would identify corresponding 'electron' term is missing from the terrain of the phase space.16 in actual history-based semantic theories for LOT on we to attractors want know should how the scene, i.e., terms, appear how the phase space that models the cognitive system of a particular sub 'electron'. Given This our interest ject takes on its shape. The research carried out by Esther Thelen and her associates in Thelen, 1995) (see the various studies cited and described offers one DS view of the development of the phase portrait, a view which to cogni takes the development of motor skills in infants as fundamental that the child's development tion. Thelen begins with the assumption of motor of certain attractors in skills can be identified with the emergence to the range of the child's physical move the phase space that corresponds as responses to develop infants' concrete interactions with the world.17 Natural patterns of move ment are slowly altered allowing the infant to achieve her goal (e.g., the a of Thelen further toy). grasping explains how these concrete interactions ments. Thelen then shows how these attractors serve as the basis for the development of more abstract concepts, might such as that of acting in general (Thelen, 1995, pp. 95-98). A related DS comes from Jean Petitot. While the emphasis here is not on perspective illustrates the way in which abstract struc issues, Petitot developmental tures (for example, concepts of semantic roles) can be derived from visual scenes (Petitot, 1995). Petitot's work bears on current concerns because certain tools necessary for human cognition, indi i.e., the nonsemantically can serve as viduated LOT terms that of abstract properties, representations seem again to be emerging from the actual physical of the experiences subject.18 104 ROBERT that we Assume LOT D. RUPERT terms, e.g., acquire many of our most 'object', in the ways described abstract and fundamental by Thelen of a given, semantically basic likely, then, that the emergence on the subject's having had certain kinds term t is highly dependent i.e., the subject's having had a certain kind of actual history. periences, t there has to have is not to say that during the process of acquiring LOT of ex This been a perfect covariance t and the members between to t. Yet, in the absence think should be assigned anything approaching the extension that we It and Petitot. seems of of itmay be that the only way for the subject to acquire t perfect covariance, for the (the only way subject's phase space to take on a shape such that we are willing to say that the subject has acquired t) is for the person to have had reference- fixing contact with 1.1 do not here offer an account of this relation.19 My focus on the subject's actual history is only the following: we are much more likely to be able to find a t and the reference class that historical causal relation between reference-fixing meant to show privileged, our intuitions assign to t (or, better the needs of the relevant that should be assigned - see note 25) if the theory empirical an of members of the assumed reference class for is, humans, presence t. of the acquisition of Under such circumstances, objections integral part to an actual history-based approach would seem harder to come by than has to t given still, the class we cannot proceed thought. For in offering objections, simply t if tokened without "What the by saying, subject having the right history, a member of class cl" In for example, without, having ever encountered often been to be relevant, the proposed counterfactual situ order for such an objection how that describes ation must be consistent with the empirical theory(ies) and acquired. And in this case, the relevant LOT terms are individuated theory, DS theory, seems to place substantial empirical range of causal histories consistent with the acquisition constraints on the of t. 5. THE NATURE OF THE CONSTRAINT and the one that follows, that a naturalistic semantic In this section I attempt tomake clear the effect of theory for LOT be firmly grounded I outline In the present section, a general constraint on thought experiments. Section 6 illustrates the way can bear on our theorizing to the constraint this constraint by applying demanding on a theory a specific asymmetric Consider of LOT theory of term individuation. intentional theory. dependence a simple, naturalistic Fodor I will following 99). The Crude Causal call Theory the content currently on offer, Jerry Fodor's theory of reference for LOT terms, which 'Crude Causal Theory' (Fodor, 1987, p. says that an LOT term t refers to whatever 105 NATURALISTICSEMANTICSAND INDIVIDUATION the tokening of t in the subject in question. Difficulties with this are term In LOT the 'horse'. the obvious. Take typical subject, the theory a causes an term that we can the of LOT horse of tokening perception causes label 'horse'; however, many other things can cause the (by stipulation) saddles or cows on dark nights (although tokening of 'horse', for example, we cannot be sure of this a priori - see notes 3 and 7). Some of these causes are members of kinds other than the kind horse. How are we to limit the the other causes of 'horse' of 'horse' to just the horses, excluding semantics needs to locate a tokenings? Generally speaking, a naturalistic can R that be characterized relation and that privileged nonsemantically extension holds kind a given LOT term t and the individual of t.20 that constitute the extension between In our quest replace the one to find a suitable identified naturalistic the Crude or members relation Causal R with of the which it would to seem by Theory, to appeal to counterfactual considerations. appropriate Accepted philo a proposed that we first calculate the extension suggests sophical method R assigns to t under some imagined circumstances and then check to see our intuitive assignment to t of extension whether that assignment matches If, in the imagined circumstances, to t an extension that conflicts with our intuitive in those circumstances. assigns conflict Here is taken to be a (possibly decisive) strike against I give no general argument against the method our candidate R verdict, then the the proposed R. of testing philo to our intuitive responses their implications sophical theses by comparing to counterfactual cases; however, naturalistic methodology imposes certain constraints on this approach as itmight be applied to test a semantic theory. asked to respect to the evaluation of a proposed R, we are commonly a t term LOT S tokens under such and such that conditions imagine subject without being given any reason to believe that it is nomologically possible for a human to token t under the conditions described. either Typically a critic or proponent of a proposed R points to an LOT term t and tells With us to imagine S tokening that very term under such and such conditions; and upon her doing so, we are asked what intuitive assignment of con tent we would make under those conditions R makes and whether that For the reasons set out in Section 2 above, we should be wary assignment. of conclusions based on such acts of imagining and our intuition-based to them. Ifmy explication of naturalistic methodology is correct, responses we are obliged, first and foremost to develop a theory of content that fits the actual facts of the human case; the theory, so developed, would then be to unusual cases, nonhuman tentatively extended species, etc., with an eye on what similarities might exist between cases and what the nonstandard case (although has been identified as theoretically in the human important 106 ROBERT D. RUPERT reflective is, of course, room for give and take here, for so-called in of At least the and the early stages theory development, equilibrium). seem as our now to count in we're would such, stage highest priority is to there in humans, not identify the ground of reference for mental representations to be able to token t where, in some other beings who are hypothesized so far as we can tell, it is impossible for a human to do so (or incred for further discussion that a human would do so of this ibly unlikely we our see construct best theories of human Section 7). Say that caveat, and they explain human cognitive skills by invoking LOT terms cognition, terms implies if this way of individuating in a certain way; individuated even for a that it is nomologically (or impossible astronomically unlikely) a t to term in thought token LOT under the circumstances described human is irrelevant (or largely irrelevant) then the thought experiment experiment, to the development and evaluation of a naturalistic semantic theory. I have just argued that our best theory of LOT term individuation should our use to evaluate naturalistic of thought experiments semantic constrain con 'NatCon' (for naturalistic theories; call the constraint I have described of three straint). Perhaps it is worth briefly inquiring after the implications other than the DS-based of LOT one, terms, to individuating approaches, see whether there is a way to escape, or at least render slight, NatCon's force. off common way of individuating LOT terms is are considering how thought experiments might or criticize a naturalistic to either motivate semantic theory, a First recall be used given know ralistic t should whether, relation that one that we limits. Given not be picked out according or not, the content assigned R is the correct assignment; an independently term (1) to which to its content: we want to to a given term by a natu thus, we must have in hand a content is an open question, term whose identifiable we can make an intuitive content assignment (or, better, an assignment motivated needs of a particular theory) by the explanatory in and (2) the tokening of which we can then inspect for its participation to see whether, relation R, so that (3) we can compare the two assignments or not, they match. The naturalistic by appeal the matter suggest a second tack: we identify t as they occur in of mental processes theory that a given It is sometimes proposed, for example, orientation might to the best current of the brain. to some specific I this against approach,21 t with a specific type of neural structure wish only to note that identifying t can be in which places a severe limitation on the range of situations it is nomologically tokened. Such situations are limited to those in which LOT (at least for a particular I press no complaint Though term is identical neural possible structure. for the subject to instantiate subject) the neurological structure in question. NATURALISTICSEMANTICSAND INDIVIDUATION 107 cases where should also exclude there is an incredibly (Such situations as small chance of the structure's the result of chaotic say, appearing, I assume, neural firing, for such a structure would, lack the causal powers in relations that would and participation incline us toward thinking of it as an LOT remains substantive and in effect. term.) Thus, NatCon a third alternative, consider the view that mental representations are as I rests view will realizable abbreviate This view (which 'MR'). multiply on the observation that a perfectly natural kind, say, moun respectable As tain, can take many stuffs out of which a given LOT ture, because different respect to its shape or size or the to LOT terms, MR implies that Applied not be identified with a specific neural struc forms with it is made.22 term t should can play the role of t on different a single subject. According occasions to the of its tokening, even within common functionalist elaboration of MR, mental possess representations their identities in virtue of their functional roles, i.e., their connections with each other and to inputs and outputs, and thus, facts about the physical con different neural structures stitution of a given token of amental representation lie beside the point. On a to to it would be MR, then, wrong-headed point specific neural structure on to it is and then that claim that the s, say t, go subject in question only s in the subject's brain. Furthermore, is present tokens t in cases where seem that amental state or representation since itwould could, at different times, be realized by different attractors in a single subject's phase space, seem to imply not only the irrelevance of neuroscience MR might to LOT term individuation, but also the irrelevance of any constraint that might emerge from the DS approach. MR faces metaphysical that threaten its very coherence. (For problems see the pertinent essays in Kim, of these problems, 1993.) aside general difficulties with MR, it does not seem that accepting an elaboration Setting MR will allow us to escape the force of NatCon. First off, we should, on methodological our to limit claims the grounds, multiple realizability to the sorts of realizing of a mental states in which we representation reason to believe; MR was conceived have positive of partly as a retort to type physicalism, without a human's that a being i.e., as a way to leave open the possibility constitution states. could have mental While it physical that such beings exist, and that our psychological theories will have to address their mental tell lives, our naturalistic methods eventually us to interpret MR more narrowly, as the claim that many different neural is possible (or many different attractors in phase space) can instantiate the term in a human; however, this is not the same as claiming that just any old neural structure can be (or any old attractor can count as) t in whatever one can dream up. If talking about terms in circumstances structures same LOT 108 ROBERT D. RUPERT is to yield any empirical power for an MR theorist, there will have on which to be substantial structures constraints of the various physical LOT as a given t in a given human.23 Without such con t structure at could be straints, any physical any time, rendering vacuous that invoke the tokening of t. For what explanatory any explanations power at what times count the tokening of / have if, in a single subject at a single time, any structure can count as a tokening of any LOT term? So long as one hypothesized t the right combination of terms in the subject's mind, might neural could LOT to explain current behavior, as would any other to find convenient the subject's for explaining theorist might attempt to secure the explanatory value always be 'present' term one happened behavior. The MR of talk about by citing those factors that are often thought by to individuate state types or determine content, factors such i's occurrence functionalists as relations to inputs and outputs. However, note the substantial on structures and constraints which inputoutput-based physical as / for the functionalist. state type (In the case of the mental Putnam the of that suggests signal possession example, inputs nature of can count pain, for "damage on what can count constraint body" as an input-based as pain [Putnam, In an individual 1967, p. 227].) subject, and perhaps on which neural in the species as a whole, there are physical limitations to the Machine's can play the role of terms for immediate sensory input (such as there are also physio that could signal damage at the periphery); on can structures constraints neural be which used to give motor logical even the great flexibility and adaptability commands; thus, acknowledging structures those and organization of our sen constitution system, the physical on can count muscular substantive constraints what and sory systems place as an LOT term for a given input or output. constraints (Physiological or motor as well - see Lewis, 1972, may extend to nonsensory concepts includes internal relations 1980.) I am well aware that the functionalist of nervous states and terms as part of her characterization state of mental some must term there LOT but be and types, types, always presumably, at the periphery to LOT terms for inputs and outputs. Given the connection limitations on what can count as input and output terms, physically-based between there would seem to be a limited number states that could play of physical LOT terms for inputs and outputs. does not liberate the semantic theorist roles vis-?-vis the appropriate mediating of MR For these reasons, acceptance to the extent that she is free to stipulate the tokening of t in subject S under seem interesting or convenient for the purposes of circumstances whatever cases test remains for naturalistic semantic theories; NatCon constructing in effect. NATURALISTICSEMANTICSAND INDIVIDUATION 109 Whichever way we individuate LOT terms, once we have done this in a naturalistically (i.e., nonintentional, nonsemantic) fashion,24 respectable we have introduced a substantive constraint on what will count as a rele a naturalistic se for the purposes of evaluating thought experiment to a for LOT terms. NatCon mantics implies that if we have appealed our T in terms individuation of LOT naturalistic theory particular (say, DS we are us as to to to T to look inform then the conditions theory), obliged vant a given LOT under which term can or cannot be tokened.25 6. AN EXAMPLE In this section, I illustrate one way in which NatCon might bear on the evaluation of naturalistic theories of intentional content. Consider Fodor's discussed asymmetric dependence theory of content (ADT hereafter) distances ADT from any (Fodor, 1987, 1990a). Granted Fodor sometimes claims about human psychology; if the naturalistic methodology however, in Section 2 is sound, then Fodor should be more advocated interested in as the way his theory applies to humans representational systems.26 Note much e.g., when discussing supporting examples and poten Fodor considers humans as test cases (see the various in Fodor, 1990a and Loewer and Rey, 1991). That examples he considers and Summerfield, Fodor, as well as his critics (e.g., Cram, 1992; Manfredi also that sometimes, tial counterexamples, 1992; Seager, 1993), find it hard to discuss ADT without using humans as test cases seems to bolster the reasoning of Section 2 of the present work; for it seems to show how difficult it is to argue in a persuasive manner about to abstract com attempting once we apply ADT to the human the conditions for representation from the human case. Regardless, while pletely situations that Fodor case, for whatever reason, the abstract, hypothetical as a means run a serious risk of violating describes of explaining ADT NatCon. - on Take the LOT term 'horse' as an example. Sometimes dark nights, - cows for instance have the property of being causes of 'horse'-tokenings the Crude Causal Theory would auto (Fodor, 1990a, pp. 121-122). While cows in the extension and incorrectly, include of 'horse', ADT matically, to give us a way to avoid this undesirable result. ADT's central term to 'horse' that the LOT refers horses rather than cows if says is supposed principle (a) in possible worlds the nomic where and being a cause of 'horse' tokenings the same as in the actual world), cows sible worlds cause of where 'horse' the nomic tokenings connection is severed being a horse has been broken (all else remaining do not cause 'horse', but (b) in pos connection between between being (all else remaining a cow and being a the same as in the 110 ROBERT horses D. RUPERT still cause 'horse' tokenings. are to says imagine what would principle were nomically of being-a-cause-of-'horse'-tokens actual world), that we The second half of the happen if the property dissociated from cows. we are to ask ourselves what other items, if any, such dissociation, to token 'horse'. If horses still still cause the subject in question then the second half of the asymmetric condition is would, dependence are we once to but how know what will the satisfied, happen pertinent nomic alterations are made? Upon would as a theory of human LOT term content, ADT depends on in about what would happen in a human cognitive system, described in nonintentional worlds where the causal that terms, possible regularities the system and its environment have been altered. normally hold between Understood claims to change the nomic structure of the world in the ways ADT pre so that cows (even on dark nights) no longer cause the subject in In order scribes, seem that specific changes would have to token 'horse', it would question structure to be made in the nomic of the world: some observable properties their of cows, most of and would likely properties being big four-legged, to be nomically from 'horse'. Thus we see that in order to dissociated as a out content for humans, Fodor must ADT of intentional theory spell account of LOT term indi underwrite the theory with some nonintentional have Without having given the details of such an account, Fodor runs a greater risk of violating NatCon, for we should be concerned that without a detailed term the effects of the LOT of individuation, theory suggested are unknown to us. If an in the nomic structure of the universe changes viduation. were to no longer cause the tokening being big and four-legged in me, what concomitant changes would occur? How am I to tell? One may well think that if the properties of being big and four-legged were to no longer cause the tokening of 'horse' in me, then horses would animal's of 'horse' no longer cause the tokening of 'horse' either (a serious problem for ADT). is a very important part of my After all, horses' being big and four-legged as not I horses. do them actually know what would happen if recognizing were no to longer cause me to token 'horse', being big and four-legged in the absence of a that constitutes my complaint: but it is this ignorance are terms of how LOT individuated, introduced, and retokened, we theory not in will the counterfactual situations ADT do know what happen simply directs that we imagine. it be that Fodor has inmind, theory might can be used to individuate terms in LOT? Fodor's What the machinery of which a the picture or one at at level with the the neural formal, ory symbolic computational, in good standing. I sup level. "Tokens of symbols are physical particulars which pose this to be true, inter alia, of tokens of mental representations combines 111 NATURALISTICSEMANTICSAND INDIVIDUATION to neural objects" are, presumably, (Fodor, 1990b, p. 315). According structures instan of neural this now familiar picture, a human's system tiates a system of symbols and rules. Such rules are algorithms, perhaps computed by way of sensitivity only implicitly, represented tional, nonsemantic properties of the symbols. Fodor's views term individuation, to both considered nonsemantic approaches is of a given type in virtue of its possessing 5: a symbol is of a given type or, a symbol properties; rophysiological its functional/computational stantively constrain claims suggest two in Section certain neu in virtue of theory Fodor chooses will sub term tokening under counterfactual role. Either regarding of Section this was the primary conclusion conditions; cern is not so much is false, but that we that ADT evaluate ADT to noninten 5. Again, my con cannot confidently requisite detail. would need complications lacking to be addressed if we (or Fodor) to try to fill in a picture of LOT term individuation in anything that sufficient detail. For example, were we to decide on a neuro approaches and were we to have a subject approach to term individuation physiological Numerous were still have tokening of 'horse' we are confident, we would is to in the subject's overall neural profile that portion which be identified as 'horse'. It is not my intent here to spell out such details; of ADT as a theory of content for human mental the viability however, current in whose to isolate depends on those details. In order to make representation ments as to which LOT terms will be tokened in nomically we must have a clear idea of the form those terms informed judge altered worlds, take in the cognitive system. a particular Once we have characterized someone as one possible characterization), a case can where form: "I think of following stands relation R (and ADT can raise an objection of the t and some R holds between take to be the c, but where c is not what we would intuitively of this form 'CER' (for 'counterexample of t". Call objections to CERs to /?') objections. Fodor introduces ADT largely as a response assumes that the Crude pressed against the Crude Causal Theory. Fodor collection extension Causal uations it is so easy to describe sit needs fixing largely because on cause dark 'horse' the relevant R holds (cows nights the resulting extension is incorrect (be but where assignment Theory where tokenings) cause 'horse' doesn't refer to cows, on dark nights or otherwise) (Fodor, the Crude Causal Theory, or before dismissing 1987, p. 101). However, any other naturalistic theory, on the basis of a CER objection, we should bear in mind the relativization by NatCon. We should not take to be an effective criticism of objection demanded just any logically consistent CER the theory of reference built on R. When evaluating a naturalistic semantic 112 ROBERT D. RUPERT at least until of CER objections, suspicious theory, we should be generally or to relativized the group of theories they're particular empirical theory the individuation criteria for LOT terms (a that nonsemantically specify in about which, recall, we have very few, relevant, pretheoretical schema into the tuitions). Proper relativization changes the CER objection a can case is "I describe that consistent with the relevant empir following: matter of LOT term individuation and where ical theory(ies) c not class c, but where is the extension and extension tions which fit this modified schema of f\ objections, where a reference to the relevant 'CER*' between t (Call objec '*' is to be case with relativizing we have Once theory(ies).) adopted the DS approach to individuating LOT to a naturalistic semantic terms, for example, any CER objection theory must take the form of a CER-DS for mental objection. representations replaced 7. in an individual R holds SWAMPMAN AND THE ROLE OF CAUSAL HISTORY DETERMINATION IN THE OF CONTENT the appearance of a particular LOT term in an actual human seems to depend on the subject's causal history, many philosophers as a red herring in discussions are tempted to view this consideration of na in the metaphysical theories of intentional content. "We're interested No matter how much ture of content", the cry rings out, "not the embodiment of content within one particular nomic setting". In response to this intellectual tendency, I reiterate the claim of Section central cases directs us to investigate 2: sound naturalistic methodology a over the as we actually encounter them. Without full debate into entering approach, once we have seen fit to take naturalism of human concepts the recent history of empirical investigations seriously, seems to support the secondary to a priori medita status here assigned merits of the naturalistic tions on the nature structure data on concept empirical to support the idea that humans are in touch of content. The do not seem and acquisition abstract concepts (say, Platonic across extensions determine cisely with suggest that human that pre forms or Fregean intensions) all possible worlds. The data instead are more like rough and ready identification concepts may serve us well in our dealings with the surrounding not determine in unusual or nomically extensions altered to the point, it seems that the developing child con More which procedures, world, but may environments.28 structs for the and representation the very concepts meaning expressly the of behavior and of systems purposes cognitive predicting explaining the thinker herself) in the actual world (Astington, 1993; Flavell (including et al., 1995; Gopnick of these and Wellman, 1992); given the imprecision 113 NATURALISTICSEMANTICSAND INDIVIDUATION concepts itwould seem they were developed, to intuitions regarding theoretical weight of the concepts of meaning and representation under and the limited use for which to accord much wrong-headed the correct application conditions. nonstandard, perhaps even nomologically impossible, All of this is perfectly consistent with the existence of a well-delineated natural relation of reference, which science, with what aid it cognitive from should intuitions, may get pretheoretical investigate. This does not, scientists should try to find, or will be able however, imply that cognitive across all possible to find, a reference relation that is fully determinate worlds. From the standpoint of the naturalist, the motivation behind such a desideratum If the goal is to provide a full analysis of our the naturalist then, as I have already suggested, concept of representation, has no reason to think that the concept will resolve into necessary and is unclear. sufficient conditions. relation If, instead, the goal of locating a reference across all possible worlds a is motivated desire that is fully determinate by to find out what relation we're actually referring to when we talk about in Section 2 suggests we re then the methodology described reference, the legitimacy of the motivation, without spond by acknowledging thereby to the Rather than and suffi certifying goal. identify necessary attempting in cient conditions for reference, that determine reference i.e., conditions the reference all possible worlds, we should investigate in this world, starting with what seem its most obvious concern To make more concrete the philosophical relation that exists instances. to which the pre a a is consider due to response, particular thought ceding experiment, Donald Davidson, often thought to have important philosophical impli in a swamp, as it cations. Imagine a being who appears out of nowhere, acts just like a person. to Davidson's original (According the Swampman is amolecule for molecule description, replica of Davidson himself, amazingly brought into existence by a bolt of lightning that simul were, and who but these details are unimportant for present destroys Davidson, see or a not, Davidson, 1987, p. 443.) Whether, purposes; being that ap a as a out of nowhere qualifies is greatly disputed matter (see person pears to the Swampman the contributions forum inMind and Language, 1996). taneously the behavior of Davidson's However, given that, ex hypothesi, Swampman a is indistinguishable from that of human, it is thought that our theories ad intentional should issue a definite and correct pronounce dressing matters ment regarding the nature and content, if any, of Swampman's intentional states. For Swampman's to bear directly appearance possible of Swampman the possibility should provide discussion, to CER-DS the of causal type objection theory of reference on the current the basis for LOT of a terms 114 ROBERT D. RUPERT to make room. I have not filled in the I have been attempting of such a theory of reference, but for present purposes, there is one an matters: that trait actual causal history-based only theory assigns reference to a semantically basic LOT term as a function of past causal in for which details teraction between has not interacted the subject and her environment. however, Swampman, in any way at the time when he his environment with first appears on the scene; yet by DS standards, Swampman would seem to possess LOT terms: the appropriate phase portrait would seem to contain a pattern of attractors similar to the patterns found in phase spaces for typical to the extent subjects. This opens the door for a CER-DS objection: are willing to any of Swampman's to attribute extensions LOT terms will causal theories fail, for Swampman's terms, actual history-based have reference without any causal history. human that we can two in at least the following respond legitimately of lack with of ways. First, given experience anything Swampman's nature or ilk, we should simply ignore the example, and any of our reac advocated tions to it, on methodological grounds. Given the methodology The naturalist our in Section 2, there is no reason to take seriously anyone's intuitions about our capacities; primary goal is to develop representational Swampman's one that (a) accom a naturalistically theory of representation, respectable our reactions to the real representational modates systems with which we and (b) explains the behavior of such actual sys have had actual experience tems. Given this goal, we should not accept intuitions about a hypothetical Swampman an otherwise as data points well-founded point made above about or count against to motivate of sufficient weight Recall the naturalistic of theory representation. of the nature of humans' and concepts meaning from an early age, as develop, representation: given that these concepts to ways systems, we should explain the behavior of actual representational lack firm intuitions about the content of expect that many persons will to be the naturalist states; we should also expect to of intuitions of those who claim have the and rightly so, suspicious, life. mental reliable, pretheoretical insight into the nature of Swampman's Swampman's Second, internal the naturalist should be concerned about the discussion of even as a hypothetical being: so far as we know, it is impos Swampman sible for a fully formed representational system to appear out of nowhere; so much such spontaneous goes against what we know about generation the genesis of representational irrelevant from the naturalist's systems that its alleged seems possibility the naturalistic standpoint. Accordingly, no more to provide a theory of con is of obliged psychology philosopher to say to Swampman tent applicable than is the nuclear chemist obliged atoms table would fall an element each of whose where on the periodic NATURALISTICSEMANTICSAND INDIVIDUATION 115 of 4,000 protons, 1 neutron, and 5 million electrons (cf. Dennett, the 1996, pp. 76-77). To defend this approach to the Swampman example, naturalist would seem to need amore detailed view of what is possible, and consists more a theoretical to the point, which possibilities count when evaluating in the natural sciences. It may be that both Swampman and my proposal are to odd atom, although appear, incredibly unlikely nomically possible in some sense do not violate fundamental (i.e., their appearances equations or it seems that there is a in quantum physics general relativity); however, sense in which irrelevant, and we they are methodologically significant I will not would like some theoretical apparatus to explain why. Although any such theory here, the actual practice of natural attempt to elaborate as its basis: we should be quite surprised, I think, serve scientists should to find working scientists refusing an otherwise perfectly good theory on the ground that it is inconsistent with an unactualized that is possibility so a case is in (This incredibly unlikely especially even if the example such as Swampman's, is allowed as relevant, where, the force of the thought experiment depends on intuitions of which we have - see reason to be suspicious the naturalist's first response independent of Faced with this the situation, response given above.) type physicist's to ever be actualized. seems to be to accept necessary. reference based tems, we the otherwise if we when Similarly, on our attention perfectly good theory, but renormalize can construct a perfectly good theory of to the actual histories of cognitive sys stand firm against those who claim that the unactualized of Swampman refutes our theory. Part of what is at issue here is should possibility the burden of proof. My claim is that from an examination of scientific the should when the is naturalist emerge: practice, following guideline as asked to take an incredibly far-fetched possibility seriously, providing a possible refutation of an otherwise well-motivated theory, the burden of proof is on the person doing the asking; and in the case at hand, this means that she must will there to be a significant that Swampman possibility and she must do so in a more convincing manner and persistence is appearance saying that Swampman's show and persist, appear than by simply not impossible. Natural ones; merely possible for LOT terms. These seem blanket naturalistic a bit extreme. be a molecule not science's job is to explain actual phenomena, and so it goes for a naturalistic of reference theory Given for molecule to Swampman-related worries may responses that Swampman to is described Davidson by we of Davidson should be himself, duplicate able to say something sensible about the meaning, if any, of Swampman's we we not? not should want to do rule out the possibility thoughts, Surely that Swampman has mental states (or can token mental representations) 116 ROBERT D. RUPERT that he (it?) is not a normal human. Here is a con solely on the grounds to make. cession I think that the naturalist should be willing Insofar as we on the successful theories of content developed have in hand empirically to apply these basis of our study of human capacities, we should be willing For exam theories to Swampman Neander, 1996, (cf. p. 127). by analogy we to want of that has course) say (hypothetically, may Swampman ple, states with a certain content because his (hypothetical) intentional phase is in a shape that is exactly like ones that we know to have been a The point here is that that allows for process shaped by reference-fixing. while we should be able to make some sense of Swampman's "behavior", space the Swampman should not be the tail that wags the theoreti example to philosophical cal canine:29 intuitively decided upon, "right" responses existence should not be dilemmas hypothetical posed by Swampman's an application a natu to disprove of modus tollens meant states content. If has intentional sound theory of ralistically Swampman to the content identified at all, it's because he has something comparable to drive used sound theory of content. Even this suggests too much by a naturalistically were to the points made above, if Swampman of a concession. Returning a to exist, he wouldn't real be Swampman, for he would possess history. all, so far as we can tell, there is no such thing as a swamp-person. What bearing, then, does the naturalistic defense against swamp-people for of causal history-based theories of reference have on our evaluation After LOT terms? Can still formulate theories not put the critic fundamental of reference? What a superscientist of example, the whole through childhood, sensory systems (and whatever acquires a wide aside fantastical and swamp-people to actual history-based objections in-principle discussed if, to take another commonly type on raises a human from in vitro fertilization range of LOT actual the input to this human's time manipulating so that the human other systems necessary) terms under conditions that would theory of reference, whatever to the captive human's LOT lead our it might be, history-based terms? Here the wrong extensions of a person who has no learning the mystery the objector has removed at it all and with the intuitively more plausible example of history replaced a system a person whose to create has been freakishly manipulated input favored, to assign history. by a standard phase portrait, despite a nonstandard for ethical standards research using existing, quite appropriate best modeled Given subjects, it would be unfair for me to respond to the critic by chal and prove that the example, as described, lenging her to play superscientist for our taking the supersci is nomologically possible. But as a condition to of proof to the critic: is fair the burden it shift entist example seriously, human we should require her to show that the superscientist story coheres, prima NATURALISTICSEMANTICSAND INDIVIDUATION 117 science to which our theory of content facie, with the theories in cognitive If our best theory of human LOT term acquisition and is to be relativized. as at I is have the odds with DS individuation, theory, suggested, supersci entist story, and itwould seem to be, then we can dismiss the superscientist schema. story as a violation of the constraint built into the CER* objection 4 between in Section the distinction introduced LOT terms Recall whose extensions representations in mind are fixed independently of the content of other mental are not so fixed. and those LOT terms whose extensions that our present focus is on terms of the former type, the terms, we can see more easily why the superscientist semantically success as a counterexample not meet to an actual does automatic story terms. of reference for LOT Consider history-based theory again the in Bearing basic basic LOT term. 'object' as an example of a semantically of the is version present in the infant object concept recognizable seem to be a fundamental term in the from a very early age and would fant's LOT term Some in the world will the superscientist infant's LOT. How shape the infant's the infant to acquire the 'object', without phase space causing exposing of the relevant the possibility infant to objects as stimuli (thus introducing 7?'s holding between It's easy to say that the su 'object' and objects)? if Thelen will just 'stimulate the right neurons'. However, is perscientist correct in claiming in the development of motor control that experiences are the rudiments of cognition, then the superscientist has her work cut out even more daunting given that it is very becomes for her. The challenge set that fully describes the workings of differential likely that the equations system is nonlinear and, at least by present lights, cognitive solved. (This is part of the reason why much of the research into human cognition that applies the tools of dynamical systems theory on rather than quantitatively depends qualitative modeling, precise mod of the human cannot be The scientist would have to be quite super after all, perhaps to the of herself. More point being a nomological impossibility likely than not, a to to master the for the scientist infant then, super get only way plausibly the LOT term 'object' is for the scientist to do what other parents do, i.e., eling.) a histori show the infant some objects. This, of course, would establish between objects and the infant's LOT term 'object', cal, causal connection which can be exploited by an actual history-based theory of reference for terms to explain how 'object' achieves its correct extension, LOT the class of objects. 118 ROBERT on 8. objections: the D. RUPERT incompatibility of lot and THE DS APPROACH In this section, two objections to my I address that the DS-based view of LOT use of the DS view term individuation of does first, com for meaningful interpersonal (possibly even intrapersonal) to the DS-based view of cognitive that according and second, parisons, not like attractors do behave linguistic units enough repeatable, processing, think of them as terms in LOT.30 for us to reasonably cognition: not allow runs as follows: Assume that the DS detail, the first objection a in of is that interactions correct, and, particular, theory cognition subject's with her environment system significantly shape her developing cognitive In more in the area of developmental by DS theorists working Given the large degree of variation in subjects' individual his psychology. even one not at different tories, no two persons person points in (perhaps as is her own life) will share the same cognitive this characterized profile in the way described by the appropriate phase portrait; the set of attractors will differ from one 1995, pp. 86-90). Thus, given that I have subject to the next (cf. Thelen, identified LOT terms with attractors, there seems to be no point in talking content of elements two different about the comparative subjects' LOT terms. I can assuage this first concern by reiterating what I take to in this paper. By invoking of DS theory to my argument LOT terms nonsemantically, the DS view, I intend to characterize i.e., by or content; we are to attribute reference no mention of their reference I also claim that given the way to such terms on independent grounds. interaction with the environment shapes attractors, a theory that assigns I believe be the relevance reference at success, on the basis this because chance of past causal interactions has a reasonable the it seems possible that as interaction molds causal relations are given a toehold; at the very attractors, reference-fixing as the route to attractor least, the need for interaction with the environment secures the opportunity for reference-fixing causal relations emergence no see results from the fact to enter the picture. As I things, problem one from differ that abstract mathematical profiles subject to the next; across for the purpose of making meaningful all that matters comparisons subjects a term with the relevant referential subject possess terms with the same reference that possessing however, the same attractor: Assume that that two subjects possess is that each properties. Note, does not require subject ?z's experience objects shaped an attractor in a; and because the causal interactions occurred between reference-fixing the appropriate attractor developing with and objects, the attractor in question refers to objects. 119 NATURALISTICSEMANTICSAND INDIVIDUATION for another subject, b: assume that in b, there exists an attrac Similarly tor that emerged as a result of ??'s interaction with objects; furthermore, assume that enough of this interaction was of the right sort to have fixed of Z?'sattractor the extension as the collection of objects in the world. This attractors are identical. From in no way implies that a's and ?'s relevant the standpoint of a naturalistic theory of reference for LOT terms, and any all that matters is externalist explanations, psychological accompanying as same attractors two attractors the have those that the reference, different might be. This situation can be roughly, but I think instruc I have just described as to in the linguistic context, where, the it situation stands tively, compared so to 'cat' refers does 'gato'. From the semanticist's for example, cats, and are different in form; at least it is irrelevant that the words standpoint, as I am concerned with here, if her interest is in referential semantics, refer to the same type of the linguist will care only that the two words note that reference might have been fixed for both Furthermore, terms by the same general reference-fixing In the historical devel process. we can occurrence the of similar types of each opment language, imagine animal. of events the reference of the terms was fixed: the appropriate by which " 'cat' will refer to all things of the pointed to extant cats and said speakers same natural kind as those animals throughout this: when when we over there" (substituting we turn to 'gato') (Putnam, construct theories of reference Spanish words 1975). My point is simply for natural languages, we worry not that the forms of words may differ from language to language; so long as the relevant terms refer to the same thing or group of things, we - and from can make at least that meaningful the standpoint comparison a of the philosopher naturalistic of for LOT reference developing theory is the one that counts.31 terms, the reference-related comparison The second objection charts a course to controversial territory, to the in connectionist net and compositionality lively debate over constituency can I to and do little resolve now, dynamical works, cognitive systems.32 the debate here, but we should keep in mind how the worry about con stituent structure relates to the view I have put forward. Some parties to the debate over constituent individual terms do not combine structure to claim that attractors corresponding to yield composite in the representations way terms from a public language do: a system of representation by at nor a tractors has neither a combinatorial semantics. syntax compositional a gap between the DS view and any talk of LOT: a combinatorial of thought possesses it is the very idea that the medium semantics that inspires the idea of a language syntax and a compositional This seems of thought to introduce in the first place; if these comparisons between natural language 120 ROBERT D. RUPERT then talk of LOT seems out place: LOT and the mental medium degenerate, no longer possesses to be distinctive of it. Fundamental what is supposed a mental state as a whole, here is the claim that an attractor representing of atomic parts; concatenation e.g., a belief, does not consist of a physical stands at odds with the idea of combi the lack of identifiable constituents reasons: If for the following semantics syntax and compositional a language is governed by a combinatorial in the well-formedness syntax, language should be defined by (perhaps implicit) syntactic rules; such rules natorial state legal patterns of combinations of atomic elements and do so by refer noun to the and verb, for instance, to which the syntactic categories, ring some to made here for reference atomic elements belong (allowances being is a legal string in English, 'John loves Mary' elements). syncategorematic 'loves' is a transitive verb flanked and it is so because, roughly speaking, to nouns if but the sentence as a whole were 'John' and the 'Mary'; by (as an attractor representing of the well-formedness thought is said to), the explanation seem to make no sense: such an explana sentence would lack the identifiable the complete the whole of nominal element 'John' appeal to the fact that the verb is preceded by a noun, when no noun there to be found. Similar remarks apply to compositional there is if has a compositional semantics semantics: A scheme of representation tion cannot and only if the semantic value of an entire legal string of that system is a function of the semantic values of the string's constituent parts. If the contains no identifiable element 'John' to entire thought 'John loves Mary' the semantic value value we can appeal when calculating can semantics it is unclear how a compositional sentence, attractors if to loves 'John complete Mary'. Therefore, representing apply seem pointless states lack constituent for us to mental structure, it would to on to assign reference identify certain attractors as terms in LOT and go whose of semantic the entire to individual terms do not appear as parts of corresponding thus the reference we assign to them (as well and complete representations, as their assigned remains inert: if the individual terms syntactic properties) them; attractors do not appear as parts of the whole mental state, they cannot contribute content of the to the representational their individual referential meanings their thought (and, likewise, alleged syntactic properties can have complete no effect What attractors on the subject's thought processes). seems to be missing from my account so that it will make sense of how is an explanation to say that the subject's to partly in the tokening of an combine, in LOT consists kening of 'John loves Mary' attractor identical to 'John'. Here I can do little more but Iwill attempt than point elsewhere, take in the at least suggest some directions the DS theorist might to identify LOT terms as recurring constituents of larger cognitive NATURALISTICSEMANTICSAND INDIVIDUATION 121 a beginning, consider Elman's use of principal components areas of the phase space (Elman, to (PCA) identify privileged analysis to analyze a system's behavior, When PCA 1995, pp. 10-15). applying we plot values of certain parameters of interest while others; suppressing and by doing so, we highlight patterns in the behavior of the system that structures. As we might not notice otherwise. Elman applies PCA to characterize syntac dimensions rather than to identify privileged tic categories, along which certain values are to be associated with the tokening of an individual LOT as means the for achieving term; however, PCA may also hold promise Petitot applies the morphodynamical latter goal. Furthermore, approach to describe both the attractors that represent syntactic categories (1995, pp. and passim) and to identify individual terms (1995, pp. 250-251). 233-234 (1998) may be of some interest Lastly, recent work by Paul Churchland one can how describes Churchland here; apply Gutman point alienation, measure to meaningfully certain statistical analyses, useful for performing more an across or two of structural similarity important subvariety of dy networks; note that we can also apply such systems, connectionist of similarity to compare the same system at different times in its history. The work I have cited is, to a large extent, in its early stages; how and variety of such work, it does not seem overly ever, given the richness namical measures to think that we will uncover LOT terms that reappear, although optimistic of the complete mental not in an immediately obvious way, as constituents states found in cognitive dynamical systems.33 toward which I have no more than gestured may The analytical methods robust LOT terms within syntactically provide the tools for individuating one might worry that if the structure of a DS theory of cognition. However, to DS theorists realize this promise, they may thereby relegate DS models the status of mere thus rendering DS mod models', 'implementation-level theorist (Horgan and from the standpoint of the cognitive els uninteresting theorists be interested Tienson, 1994, p. 327), for shouldn't true cognitive in thought itself, rather than how cognitive by one activity is implemented it were so simple a matter. Alas, particular type of cognitive being? Would or if we take naturalistic methodology seriously, we have no conceptual a priori grounds for cleanly separating essentially from activity cognitive seem to be the case, humans are the mere If, as would implementation. systems to which we have experimental only full-blown representational we carefully how humans represent the world; their develop capacity to represent partly via the emergence of attractors in the individual's by phase space, then properties possessed - at as mere details these attractors cannot be dismissed implementation access, had best and if humans examine 122 ROBERT D. RUPERT least not until we have a much more out thoroughly worked we at than have cognition present.34 I have left a number of issues unresolved; in closing I briefly on two of these. An issue of great importance to many cognitive is that of the causal the constituents of constituent efficacy identified using theory of comment theorists or not, structure, i.e., whether, methods have causal effects on analytical the system, qua constituents; many of the works cited in note 32 take up this issue, some authors claiming that even if constituent structure can be in dynamical uncovered structure is not causally such models, cognitive or explanatorily efficacious relevant (Garson, 1997; Ramsey, 1997). This is a dispute worthy of our interest, but deserves far more attention than space allows. about the value of assigning referents as se Also, we might wonder mantic the concern being that DS theorists themselves tend to contents, endorse functional role theories of content. For example, I have although to Petitot's LOT appealed approach as a possible means of characterizing seems inclined to assign content on the basis of internal terms, Petitot present relations and Tienson 1995, p. 243). Horgan among attractors (Petitot, an take a similar view: although of DS mod they argue for interpretation to which are causally efficacious, els according syntactic properties they states internally, characterize the content of mental in terms of mathe matical and structural relations among attractors (Horgan and Tienson, seems to acknowledge the 1996, pp. 155, 164). Paul Churchland tance of reference-determining to the outside world, connections the question of representational content, he remains a two-factor is that regardless of the 1998, (Churchland, p. 29). My position impor but on theorist impor attractors, we should take seriously and to which attempt to develop a theory of reference for LOT terms according at least at the basic level, by our commerce with reference is determined, tance of internal relations between our thoughts refer. Limitations to which world of me from this view it is be here; nevertheless, space prevent defending cause of what I take to be the advantages or of a causal, covariational, informational of terms reference for basic LOT that theory semantically I have appropriated the tools of DS theory in the way that I have, setting items aside in the external the internalist, functionalist DS theorists embrace. approach to the determination of content that many NOTES * A shorter and West version Texas of this paper Philosophical was Society; to the 1997 meeting of the New Mexico presented thanks to the audience for stimulating questions my NATURALISTICSEMANTICSAND INDIVIDUATION 123 I would Thalos and anonymous referees for and observations. also like to thank Mariam on earlier drafts of this paper. comments helpful 1 as 'terms in LOT', I often refer to these representations the arguments Although presented as in this paper do not imply the truth of the full-blown of thought language hypothesis advanced 1975, and Fodor, 1987, Appendix). (see Fodor, by Jerry Fodor 2 a theory of content. to develop I am not concerned Here I do not attempt with the Thus, terms: between the meanings of the following differences 'extension', 'reference', con 'intentional and 'extensional 'referentialcontent', content', content', 'representational . to tent' While the shades of meaning used may vary, all of these terms are commonly content in the sense that each of the terms is used describe the semantic of representations, subtle a mind-world or a type of content that presumes such where context however, interchangeably; clearly to employ I am careful that term. that we strike off in search of a brute physical description a nonsemantic at this, we then look to discover succeeded to denote either a relation. I use these terms Generally, the use of one over the others, demands 3 What's 'refers to' relation said here may suggest of a given LOT term t and having t refers. in virtue of which relation as a description This is misleading of how we might we will t takes. In scientific discover the physical forms actually practice, surely have to terms in of For example, look beyond couched the relevant sciences. descriptions physical we may as to when a term with to use intuitions to be the same content is likely have to help us identify tokened the range of physical variation that tokens of t can exhibit, the that if the same content assumption being to be tokening the same term. This would is being then represented seem to work especially the well subject when is likely we can to the subject under conditions stimuli such that the stimuli are very likely to cause a certain in cases where content for example, is directly content, thoughts with thought n. 10). Here we to immediate related observations 1990b, especially sensory (see Fodor, present reason from content rides. causes at the sensory will lead to similar periphery as well as in terms of what is represented vehicle that a similar to like cause, we might like effect pursue strategy that similar the principle of what content in terms effects, from Reasoning to motor commands; if the subject in movement x, and x engages repeatedly a we term is always infer that is neural LOT with y y, might preceded by activity repeated content to this bootstrapping 'do jc'. Although there are some obvious practical advantages we attribute contents in order to identity note the vehicles of content, method, by which with that respect must have content-independent individuation criteria (cf. ultimately we and (b) from a practical must have some 41^42), pp. 34-35, standpoint, the individuation criteria are before we begin the bootstrapping for we process, (a) the vehicles Pessin, 1995, idea what to know what sort of common element (certain types of patterns to look for in the physical for example) of the profile subject when we a term with on two different tokened the same content occasions. need While few on offer make theories a simple content function of neural think of the subject's activity, that she has past history, the theories of content offered in Dretske (1981), Dretske (1988), andMaloney assign ticular) LOT an essential in the fixation term content to the role of individual content that assigns (I have reference subject's tried my on the basis of (1994) in par history, a theory of at developing - see a subject's actual history past history own hand (learning their discussion is set in a much different Fodor context, Rupert, forthcoming). Although some sense and Lepore to be an obvious, of what is thought (1992, p. 157) give general with an actual history-based causal the worry content, problem theory of intentional being never has to engage that a person in direct causal interactions with, say, tigers in order to token an LOT term that refers to them. ROBERTD. RUPERT 124 5 Certain differences concerned exist the present and Devitt's in that (a) Devitt is project to separate semantics and (b) Devitt is careful language from more abstract about the nature of meaning it questions as the that Devitt believes truth and reference will ultimately emerge with primarily about reference questions self (although fundamental note semantic between natural once properties, the appropriate naturalistic methodology has been applied in semantics [Devitt, 1994, p. 572]). 6 See Churchland ology. 7 While it seems or not, is now she and Sejnowski (1992, p. 61), where the authors endorse similar method can tell, with a fair degree of reliability, human whether, average as she had on a the same thought this skill occasion, having previous in our development term individuation. of a theory of LOT This way of that your is of limited help term individuation LOT the horse: it individuates LOT puts the cart before approaching terms according to their reference. If the naturalist is to locate content in the natural, i.e., order of nature, then it seems that both terms of the 'refers to' relation will nonsemantic, to be nonsemantically have We should bear in mind that in cognitive science, been of as a medium for the expression typically thought as the content of a thought, while regarding perhaps fairly reliable as indicators cannot of thought be taken of the presence of specific content, see note 3 above for discussion in the mental medium of the potential (although mental indicators characters specified. have representations of content: intuitions value of such intuitions). practical 8 Localized of the brain portions the space to which we might look seem to have that specific jobs to do, and this suggests of a given LOT term could be of fewer dimen sions than the one trillion or more in the text. In any case, the number of free suggested nature that result from the localized will be quite large. Ignoring parameters complications of cognitive should not materially affect the arguments that follow. processing in search A contrasting arises if one takes seriously the idea that cognitive process complication in the brain alone (Rockwell, less localized 1994), and thus is much ing does not take place than is often of "noncranial" would alter not, however, parameters thought. The addition in the text. the arguments 9 on Van Gelder The preceding of dynamical systems theory draws explication heavily and Port (1995), Norton and Giunti See these sources for a more (1995) (1995). complete of the nature of dynamical and the mathematics used to analyze their systems explanation behavior. 10 I will use it has been thought theory 11 Also the abbreviation worked into a certain 'DS' systems only when talking about dynamical theory view of the mind, is most i.e., that the mind productively to talk about dynamical I mean of as a dynamical instead system. When as an area of mathematics, I refer to it as such. in general, see Horgan to this effect. and Tienson's work (1994, 1996) for arguments as systems Horgan differ from Van Gelder and Port's, in that Horgan and arguments though, on difficulties of their emphasis Tienson from the frame problem and place most arising of the computational the related problem of cognitive intractability processing. 12 are not abstract For example, but rather says, "Representations symbols Jeffrey Elman on are a not state but rather in the Rules in embedded space. operations symbols regions and Tienson's dynamics ers while mental Elman have of the system, other making a dynamics which transitions difficult" follow and it does representations, it's just that the representations conceived them to be" (ibid., p. 195). says, movement permits (Elman, rules and 1995, when the rules from p. certain 196). The mind regions does to oth employ these representations. processing from what we "may be different NATURALISTICSEMANTICSAND INDIVIDUATION 13 For why from of eliminativist claims coming examples of them, see Clark should be suspicious we the DS and Toribio camp, (1994); 125 reasons together with are Clark also relevant (1991) and Fodor and Pylyshyn (1988), which offer criticisms of connectionist-inspired, views eliminativist that would to develop attempt see Churchland representation, 14 to imply here I do not mean seem also some for a recent to hold against DS-inspired of a noneliminativist, aspects And eliminativism. connectionist of view (1998). in no that the infant's use of the LOT term 'object' is guided I merely that the content of these other representa suggest by other representations; in fixing the extension of 'object' for the infant. The idea is that even tions is not operative if there is some degree of content-based in the determination of reference, interdependence way some terms have of others; realist to have about their content fixed independently of the content-laden contribution at least if one is a of content determination would, process not seem to ever be able to get off the ground. for the (This demand terms is similar to that made of content for at least some LOT by the entire otherwise content, fixation independent in their criticism Fodor and Lepore of Paul Churchland's state-space semantics [Fodor and Lepore, 1996, pp. 152-155].) See Spelke (1990, 1991) and Bower (1989) for descriptions of experimental results that reveal the infant's early of a strikingly development rich object concept. 15 am I not is confused that Elman about this point, only that this way of talking saying can be misleading. 16 we have some way of terms as The reader may wonder whether ruling out certain LOT can content attractors other than to LOT the that 'electron', bearing saying only correspond a systematic, to justify our judgements terms. One way be to provide in this regard would structures distinction between that would be appropriate grounded physical physiologically candidates for representing theoretical and those that would be appro concepts complex for As acknowledged mental of phonemes. in note candidates being representations priate 3 above, this kind of distinction is difficult tomake at the physical level without employing some semantic hunches, by questioning only) can help patients although experimental, physiological whose brains are directly stimulated data while (such under as those local collected anaesthetic as to what structures would conclusions types of physical justify our general to play what kinds of semantic be appropriate roles in LOT. From a practical standpoint, we may be able to justify in the text by observing the judgement that however, suggested in question and fails to exhibit other the subject fails to say sensible things about electrons appropriately 17 We should electron-directed not ignore behaviors. the possibility and cognitive velopmental psychology, nativism has many advocates of innate science contributions more generally 1992 includes to the phase portrait. In de (e.g., in psycholinguistics), a largely approving review (Karmiloff-Smith, the developmental data supporting One might think that the more nativism). is governed and constraints, the less plausible cognitive development by innate principles we should find an actual history-based to one of for LOT terms, contrary theory of content of much of the main Dretske's much theses theory as cognitive if not of the present [Cummins, science work. 1991, (This pp. has demonstrated of cognitive all, aspects development is one of Karmiloff-Smith's primary points the development of the wetware itself. The phase LOT space emerge is especially of worry to has been raised with respect to note, however, It is important that as in the way of innate cognitive biases, many, are shaped by environmental interaction (this type 105-106]). in discussing the nativist literature), including even if the newborn here is that infant's point some innate shape, the attractors count as terms in that are to eventually interaction the environment. with This point only as the result of copious one considers clear when on processing, the distinction between constraints has 126 ROBERTD. RUPERT of which is supported and specific work, by the nativist developmental at birth. Given the many forms an implicit rule or constraint represented could possess much in the way of innate (Cummins, 1986), it seems that newborns 1990 for biases, while few, if any, innate, explicit mental (See Keil, having representations. a survey of the various researchers have in mind when types of constraints developmental the innate presence concepts can take explicitly and biases.) they talk about innate constraints 18 a Bartsch offers of conceptual amenable (1996) dynamics-based analysis development on the to the DS view outlined In Bartsch's above. model the stability of a concept depends on of structure depends of structure in the cognitive emergence emergence system, which the subject's direct experience of the world. Until the subject cannot be said to possess the concept a certain of structure stability in question, emerges, While to Bartsch. according Bartsch's analysis is largely in keeping with the DS view briefly described in the text. as structures to consciousness between and sub concepts present so in a way seems to give full that are not, and she does that only status to structures In of the former 1996, pp. 430-431). type (Bartsch, no special weight I ignore this distinction, of LOT terms, however, placing Bartsch distinguishes structures cognitive representational my discussion or lack thereof. on conscious accessibility 19 a terms that is consistent for LOT naturalistic semantics with DS theory Constructing a we will to In much addition work. relation, locating satisfactory content-fixing requires to DS to include at least two following have elements, (a) relating specifically theory: a principle of terms for LOT individuation that those separates attractors that are to be identified with LOT terms from those which should not be, and (b) a principle of ancestry over as LOT that tells us how attractors, terms, are to be identified time, as the phase some distance Bartsch's model (Bartsch, 1996) goes space in which they appear changes. a in phase toward characterizing the sort of stability space we like to see before we attribute a a to that semantic naturalistic extensions concept subject. However, theory assigns given a way to identify also needs stable attractors of a subject's currently history of a current LOT term can be said to in order that the content their early ancestors its relevant ancestors. interactions have been determined involving by past causal 20 no so far as I can tell. As noted the Crude Causal has real above, advocates, Theory on the basis with in of the causal theories cited the theoretical complexity history-based given to use the Crude Causal to illustrate the need for a it will be convenient Theory However, note 4, a suitable to identify relation worries about the mismatch naturalistic theory of content 21 see Andrew Pessin's But between the causal of powers states (Pessin, 1995, p. 39 and passim). types and those of psychological see Putnam statements and Fodor of MR, classic (1974). (1967) neural 22 For 23 R. Horgan and Tienson (1996, Chap. 9, n. 5) give a brief explanation of why it is empiri to keep the 'multiples' important their argument in terms of mental seem loss of empirical power would of multiply cally put realizability 24 The text in contrast, favor 1996, some of a more be susceptible naturalistically liberal, of universe. While I agree desirable, not reduction inevitable" to some (1993, states fairly minded such philosophers reject naturalism (Baker, promiscuous as part of the basic properties to or unification extent p. 94), with with Baker I think small. While other, when there they their points about the interpreted, to the multiple limit the discussion that all legitimate entities, require in nonintentional, nonsemantic to characterization ontologically that can take semantic p. 541) in no need verse, semantically to hold when we of uninterpreted LOT terms. a whose strictures naturalism suggests and relations erties, realizable states, a restrictive are good theories that "Unity which reasons, says in view, p. 94; Shapiro, furniture of the uni 1993, nonsemantic she prop terms; of the is merely I cannot NATURALISTICSEMANTICSAND INDIVIDUATION for a naturalist not just "merely", to find unity highly, desirable: thus, the an awareness as much theoretical pursue though with unity as possible, more so. Even over what if is decidedly have to settle for the less desirable into here, go 127 should naturalist that we may the liberal naturalists were it's not correct, in the text: even were we take point we would still somehow the universe, to attribute trivializing a semantically of natural graphic 25 The have properties to characterize extent undermine that would my as part of the basic furniture of we wish to which the entities we to individuate properties: to of semantic its simply by referring possession to be done by our much work of the explanatory those p property to what clear semantic laden of criterion should bear property semantic one be that I have semantic would risk words according to their employed or set of connected to which relativization by hinted So not only pp. 12-13). to the theory(ies) used our evaluation of a text), theories 1989, (Cummins, a given theory of content to individuate in the the relevant units (as argued representing semantic should also be constrained naturalistic theory by the or intentional content: of reference be served by our attribution the discussion relativize the naturalist at already in the text. A naturalistic as part of an attempt to understand representation a particular is a (and folk theory empirical theory seen construct here) p, we theory. In addition, the apparent facts in the case goes against part, we can individuate properties. sort of another in mind be especially sensitive, semantic theory should always possibility should we that has the entity i.e., nonsemantic, should as a theoretical identification for the most where, language, and phonemic, reader were semantic of explanatory we should purposes not criticize to a to do what to do in the first place it was not supposed naturalistic semantic theory for failing to explain for its failure the source of each and every semantic for example, intuition we have about our thoughts. might 26 Sometimes Fodor claims only to be giving sufficient conditions for intentionality, in the to to claim is wrong is irreducible show that Brentano that intentionality (Fodor, attempt 1990a, p. 96). But to evaluate from the discussion of Section 2, the worry to Brentano's emerges that we are in no to applies cases we know of intentional/representational If ADT i.e., human beings. systems, cases of representational does not apply to the best (only?) that we know of, what systems to could be a naturalist's for claiming that asymmetric is one way justification dependence position the best fix 27 sentations, processes more, raised LOT alleged solution problem unless ADT a solution to Brentano's and thus provides problem? about the specific characteristics of these repre say much physical can that to exhibit formal which beyond they claiming computational properties are sensitive Fodor does say Fodor, 1994, Lecture (see, for example, 1). When intentional Rarely Fodor's content, does Fodor an objection For example, in discussing to ADT terribly enlightening. the Fodor of that for Ned about the of Block, says purpose by talking tokening we should understand terms in counterfactual the term 'cow' as a "phono situations, it is not that terms in LOT 1990a, p. 111). However, (Fodor, sequence" given seems as if it and are not written, this talk of phonology and orthography to be metaphorical. in the text is that Fodor does not offer a theory My concern logical/orthographic cannot be heard would have of LOT term individuation thatmight legitimate thismetaphor. The readermay be put off by my use of a quotation from Fodor (1990b), given that has openly renounced the views there. While Fodor has given up on the expressed to LOT semantics described in 1990b, he has not given up his view teleological approach of the physical constitution summarized in the passage of LOT terms, so neatly in quoted the text (cf. Fodor, 1990a, p. 159). 28 on family I have in mind The empirical work is the work resemblance and the pro structure of concepts. The early, ground-breaking work in this vein is totype/stereotype Fodor 128 in Smith summarized Keil, and Medin the more of mitigation structure extreme et al., (Armstrong the basic 1989), point ROBERTD. RUPERT the past fifteen years have seen a although the prototype/stereotype-based view of concept and 1985; Murphy, 1989; Murphy 1988; Medin, seems to stand: humans rely largely on nondefini And (1981). claims 1983; of Smith, of this work For a recent discussion in applying of the bearing of the nature concepts. on philosophical somewhat in keeping with the view expressed method, here, Putnam and Tienson, 1996, pp. 142-143. (1975) and Stich (1983) also provide to philosophical of a stereotype-based view of concepts provocative applications questions, in Putnam's in Stich's about reference, case, and the nature of mental states, (although by heuristics tional of concepts see Horgan or Putnam's endorse without qualification of concepts and meanings his throughout seem to be an intellectual 1953) for what would to these applications, I do not thereby referring Stich's See discussions results). Wittgenstein's Investigations Philosophical (Wittgenstein, ancestor views of the nature and structure of concepts. of many contemporary a limited endorsement of the stereotype/prototype view of concept structure, By giving are stereotypes. to imply that word meanings I do mean 1981, Chap. 10, and Fodor, (See Fodor and Lepore, be word meanings, with can to doubt that stereotypes/prototypes 1992, Chap. 6, for good reasons useful of 'meanings'.) Even if concepts interpretation given one highly some of our LOT or natural to attach terms to structure language help prototype determinate external themselves, concepts lack such determinacy. because the concepts in nature, complex of meanings as guides (Note discussed whereas the sort Fodor favor (Fodor, might of LOT or natural to the application that I have kept 1990a, are such atomic ones, 29 Cf. David - terms of LOT the separate concepts as stereotypes are typically in the psychological literature are atomic terms I have terms. This in mind the LOT leaves open the possibilities (1) that the typical concept discussed in the psychological is a collection 6), Chap. terms, may language and LOT this is terms, that each often called 'features' to one is identical atomic - and that (2) LOT some literature the concepts, term.) in a different, of the Lewis's similar, context, dismissal, importantly although of a hypothetical theoretical relevance mad, 1980, p. 221). (Lewis, unique Martian 30 to my attention are the fact that both of these points A referee for this journal brought concern context. in the present of particular 31 not with I have dealt only with of interpersonal the question comparison, intrapersonal an additional The latter should be handled although complication analogously, comparison. a single of attractors within about the stability when we wonder subject over time: we must can content to fully explain be how meaningful made, intrapersonal comparisons on the appropriate to characterize in a dynamical the structural continuities decide way over time; see notes 18 and 19. space changes system whose phase 32 For a start, see the essays that addresses these questions. There is a wealth of literature arises inPart I ofMacdonald andMacdonald (1995); also valuable areVan Gelder (1990), Clark (1991), Clark and Toribio (1994), Horgan and Tienson (1996), Garson (1997), Ramsey (1997) and theworks of DS theorists discussed in Section 4 above. 33 There exist the uncovering within cognitive 1989, pp. other of mathematical methods structure syntactic viewed systems and 192-193) tensor or that, when properly to the characterization as dynamical product encoding of employed, individual may LOT lead to terms The use of cluster analysis (Clark, systems. schemes 1991) are two such (Smolensky, possibilities. 34Van Gelder and Port (1995), Clark (1995) andMcClamrock or another, as) when we implementation at the cognitive works (1995) argue that in one way of systems, cognitive knowledge study on our understanding can have great bearing details human level. (what are of how of thought the system NATURALISTICSEMANTICSAND INDIVIDUARON 129 REFERENCES Cognition Astington, J. W.: Mental 1993, R.: 1996, T. G. Bower, 1980, (ed.): MA. Harvard, R.: Concepts Might Press, and A. Mele (eds.), pp. 75-95. and a Dynamic Models Connectionist Data 421-454. 108, Synthese Vol. 1, Harvard Press, University W. in Infancy, Learning Infant: University H. and Freeman York. P. Churchland, in J. Heil Causation', Oxford, in the Philosophy, The Rational Harvard the Mind, of Discovery Formation', Readings 1989, New Company, Child's of Concept Theory N. Block, The and Mental 'Metaphysics Oxford Press, University Between 'The Relationship Causation, Oriented Some 'What 1983, 13, 263-308. MA. Cambridge, 1993, Baker, L. R.: Bartsch, and H. Gleitman: S. L., L. R. Gleitman Armstrong, Not Be', S. and T. J. Sejnowski: The 1992, Brain, Computational MIT Press, MA. Cambridge, across Sensory The and Neural Diversity: P. M.: 1998, Similarity 'Conceptual Journal 95, 5-32. Answered', of Philosophy Fodor/Lepore Challenge Distributed and Parallel Science, Clark, A.: 1989, Microcognition: Cognitive Philosophy, MA. MIT Press, Cambridge, Processing, Churchland, Clark, A.: (eds.), Philosophy NJ, pp. 115-128. 1995, 'Moving Minds: Hillsdale, Clark, 'In Defense 1991, Rumelhart A.: in J. Tomberlin Psychology, Ridgeview, A and J. Toribio: Clark, Cram, 42, H.: in the Service Atascadero, CA, of Real-Time Success', and Philosophical 9: AI Connectionism Perspectives 89-104. pp. Representing?', 'Doing Without of Representation', Causal Theory 1994, 'Fodor's 1992, Theory, Content Situating (ed.), Philosophical and D. E. S. P. Stich Ramsay, Elbaum Lawrence Associates, inW of Explicit Rules', and Connectionist 101, Synthese 401^31. The Philosophical Quarterly 56-70. R.: Cummins, 1986, and Belief, of Knowledge Representation 116-126. Cummins, R.: Cummins, R.: McLaughlin and His Brand and R. M. of Arizona University Critics, Blackwell, Oxford, Dretske, F.: Dennett, D.: 1987, 1988, Explaining 1996, MIT Behavior, 'Cow-Sharks, Magnets, The (eds.), Tucson, Cambridge, AZ, pp. MA. in B. Explanation', 102-117. pp. and One's Own Mind', Proceeding 'Knowing Association American 60, 441^458. Philosophical MIT and the Flow F.: 1981, Knowledge Dretske, of Information, D.: Davidson, Harnish Press, MIT Press, Representation, in Psychological Meaning and Mental 1989, Meaning 'The Role of Mental 1991, (ed.), Dretske inM. Information', 'Inexplicit Press, Addresses of Cambridge, MA. the MA. Press, Cambridge, Mind and Swampman', 11, 76 and Language 77. Devitt, M.: 1994, 'The Methodology 1995, 'Language of Naturalistic Semantics', Journal of Philosophy 91, and Van Gelder pp. 545-572. Elman, J.: as a Dynamical System', in Port 1995, 195-225. Flavell, J. H., F. L. Green and E. R. of Chicago Thinking, University J.: 1974, Fodor, 'Special Sciences', Fodor, J.: 1975, The Language Flavell: Press, Synthese of Thought, 1995, Chicago, 28, 77- Harvard Young IL. Children's Knowledge about 115. University Press, Cambridge, MA. ROBERTD. RUPERT 130 MIT Press, Cambridge, MA. MIT MA. J.: 1987, Psychosemantics, Press, Cambridge, J.: 1990a, A Theory and Other Essays, MIT Press, of Content or: Where do Truth Conditions J.: 1990b, 'Psychosemantics J.: 1981, Representations, Fodor, Fodor, Fodor, Fodor, MA. Cambridge, inW. Come From?', A Reader, G. Lycan and Cognition: Blackwell, Oxford, (ed.), Mind MA. J: 1994, The Elm and the Expert, MIT Press, Cambridge, A J. and E. Holism: 1992, Guide, Blackwell, Fodor, Shopper's Lepore: 312-337. pp. Fodor, Oxford. 'Paul Churchland and State Space J. and E. Lepore: 1996, Semantics', and Their Critics, Blackwell, Oxford, (ed.), The Churchlands pp. McCauley in R. N. Fodor, J., and Z. Pylyshyn: Fodor, Cognition in a Dynamic J.: 1997, 'Syntax Models M.: 1995, 'Dynamical Giunti, A Critical Architecture: 3-71. 28, Analysis', Garson, and Cognitive 'Connectionism 1988, 145-159. 355. Brain', 110, 343Synthese in Port and Van Gelder of Cognition', 1995,, pp. 549-571. and H. M. Wellman: A. Gopnick, Theory', and 101, 305-345. T. Horgan, Synthese T. and Horgan, Press, 1992, and Language 7, J. Tienson: 1994, Mind J. Tienson: the Child's 'Why 'A Nonclassical Framework Connectionism 1996, of Mind Theory Is a Really 145-171. and for Cognitive the Philosophy Science', MIT of Psychology, MA. Cambridge, on Cogni A Developmental Modularity: Perspective MA. Cambridge, MA. MIT Press, Cambridge, and Cognitive Kinds, Keil, F.: 1989, Concepts, Development, on Constraints: the Epigenetic 'Constraints Keil, F.: 1990, Cogni Landscape', Surveying tive Science 14, 135-168. A.: Karmiloff-Smith, tive Science, Kim, MIT J.: 1993, Lewis, D.: 1992, Beyond Press, and Mind, Supervenience 1972, Cambridge and Theoretical 'Psychophysical UK. Cambridge, Australasian Journal Press, University Identifications', of Philosophy 50, 249- 258, reprinted inBlock 1980. Lewis, D.: 'Mad Pain 1980, B and G. Rey Loewer, and Martian in Block Pain', inMind: 1991, Meaning (eds.): 1980, Fodor pp. 216-222. and his Critics, Blackwell, Oxford. and C. Macdonald, G. Macdonald: 1995, Connectionism: on Debates Psychological Oxford. Blackwell, Explanation, J. C: 'Content: 1994, Maloney, Covariation, and Contingency', Control, 100, Synthese 241-290. and D. M. P. A. Manfredi, in Fodor's McClamrock, Murphy, 562. G. Murphy, L.: G. L. 1988, 1996, 1995, Pessin, A.: cal Studies L. Medin: Review A.: 1995, 78, Cognition, 92, Studies 66, Concepts', 'The Role of Theories A Flaw Asymmetry: 261-283. of Chicago University Complex 1985, without 'Robustness 1992, Philosophical 'Comprehending and D. Psychological K.: Neander, Norton, Summerfield: of Content', Theory R.: 1995, Existential Cognitive Chicago, Science 12, in Conceptual 529 Coherence', 289-316. 'Swampman Meets An 'Dynamics: 'Mentalese Syntax: Between and Language 11, 118-129 and Van Gelder 1995, pp. 45-68. Mind Swampcow', Introduction', in Port a Rock and Two Hard Places', Philosophi 33-53. in Visual and Attractor J.: 1995, Syntax: Constituency 'Morphodynamics in Port and Van Gelder tion and Cognitive 1995, pp. 227-281. Grammar', as Motion, MIT Press, Cambridge, 1995, Mind Port, R. F. and T. Van Gelder (eds.): Petitot, IL. Press, Percep MA. NATURALISTICSEMANTICSAND INDIVIDUATION H.: 1967, Art, Mind, and Putnam, as reprinted in text and notes H.: Putnam, and D. D. Merrill H. Capit?n of Pittsburgh University in Block of Mental States' are to the Block 1975, Mind, inW. Predicates', 'Psychological Religion, 'The Nature Press, 1980, 131 PA, Pittsburgh, pp. 223 -231. (eds.), 37-48, pp. references Page volume. and Reality: Language, UK. Cambridge, 'Do Connectionist 1997, Philosophical Vol. Papers, 2, Cambridge Press, University W.: Ramsey, and Language 12, 34-66. W. T.: 1994, 'On What Rockwell, Their Earn Representations Explanatory Keep?', Mind Is Identical the Mind With', Philosophical Psychology 7, 307-323. R. D.: Rupert, 'The Best forthcoming, Test of Extension: Theory First to Principle(s)', appear inMind and Language 14:3 (September 1999). 'Fodor's 1993, Seager, W.: Science 60, 262-277. Shapiro, L.: 1996, 26, 523-542. E. Smith, E.: Smith, 1989, 'Concepts MIT Press, and D. Medin: E., Problems from Bottom 'Representation Science, Cognitive of Content: Theory and 1981, and Top', in M. Induction', MA, Cambridge, Categories and Stillings, S.E. Weisler in Port and L. Baker-Ward: and Van Gelder Van Gelder, Cognitive Van Gelder, Approach Macmillan, T.: of Harvard Press, University and in D. the Language S. Kosslyn Osherson, to Cognitive Invitation in of Thought', and T. J. Vol. Science, 1990, to Cognitive J. L. Garfield, 1987, Cognitive and Dynamics 69-100. pp. Science, E. MIT Press, L. Rissland, Science: An D. Theory', Associates, MA. Cambridge, A. Rosenbaum, MIT Introduction, of an Embodied the Development Press, Cognition', 1995, 'Compositionality: A Connectionist on a Classical Variation Theme', Science 14, pp. 355-384. and R. F. Port: 1995, 'It's About to Cognition', in Port and Van Gelder L.: 1953, Philosophical Investigations, New York. 13416 4th Ave. S. Seattle,WA 98168 USA robrupert Foundations Cambridge, MA. Cambridge, E.: 1995, 'Time-Scale Thelen, Wittgenstein, (ed.), S.1991, NJ, pp. 133-169. 1983, From Folk Psychology N. A. ,M. H. Feinstein, S.: I. Posner of Philosophy MA, pp. 99- 127. on Piaget's in Infancy: Reflections 'Physical Knowledge and R. Gelman Lawrence Erlbaum Mind, (eds.), The Epig?nesis of Press, Hillsdale, Journal Concepts, P.: 1991, 'Connectionism, Constituency, and Rey 1991, pp. 201-227. E. S.: 1990, of Visual Knowledge', Spelke, 'Origins An Hollerbach and Action: (eds.), Visual Cognition Stich, Canadian pp. 501-526. Smolensky, Loewer E. Spelke, S. Carey of Philosophy MA. Cambridge, 2, MIT and Objections', @ aol. com Time: 1995, An Overview pp. 1^4-3. translated by G. of the Dynamical E. M. Anscombe, in