PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. I DATE DUE DATE DUE DATE DUE I MSU Is An Affirmative ActiorVEqueI Opportunity Institution cWMt NATURALIZING SEMANTICS: FODOR AND DRETSKE ON THE CONTENT OF PSYCHOLOGICAL STATES BY Carol Winifred Slater A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of PhiIOSOphy 1990 Both Jert states of pation ll Fodor su; causal dl nation-p its auth succeed the kind invocati establis also has of misre Symbol c the Uncl and in 1 that it infelic; that th) CaUsal ‘ Versjal intenti( SUPPOSQ; Vith Ev. a CaUSaI NATURALIZING SEMANTICS: FODOR AND DRETSKE ON THE CONTENT OF PSYCHOLOGICAL STATES BY Carol Vinifred Slater Both Jerry Fodor and Fred Dretske have proposed that representational states of systems can have intrinsic content by virtue of their partici- pation in lav-governed or otherwise reliable relations with the world. Fodor suggests that the relevant relations are synchronic--asymmetric causal dependencies; Dretske invokes instead a genetic account of infor— mation-processing functions. I argue that neither proposal lives up to its author's hopes. Dretske's notion of functional meaning does not succeed in carrying us from purely objective or "natural" relations to the kind of intentionality characteristic of beliefs; in particular, invocation of learning and time—invariant natural meaning fails to establish the desired nomic possibility of misrepresentation. Fodor also has problems with error: I argue that his current characterization of misrepresentation in a labelling context as the misapplication of a symbol of the language of thought is unsatisfactory both with regard to the unclarity with which the concept of a labelling content is developed and in the dependence of the notion of symbol application upon concepts that it is supposed to explicate. In addition, I draw attention to infelicities of notation and reliance on analogies with natural language that threaten the intelligibility of Fodor's account of asymmetric causal dependency. Finally, I argue that it is by no means uncontro- versial that a semantics for psychological states that can defend intentional explanation from eliminativist threat must, as Fodor supposes, identify events individuated under psychological description with events as individuated by physics. I conclude that the failure of a causal semantics need not be fatal to the claims of psychology. Good ft gratefl care at of a d on thi. reassu them, i Discus HY col Especi Hertzo the Sc Shapel would ; Slater Acknowledgements Good fortune in one’s mentors is very good fortune indeed. I am grateful to Herb Hendry, the Chair of my Guidance Committee, for his care and counsel over the years and for affording me early on a glimpse of a different garden. Richard Hall kindly agreed to supervise my work on this thesis; he has been a skillful and enthusiastic guide, patient, reassuring and extraordinarily generous of his time and talent. With them, Barbara Abbott and Gene Cline made place for me in the Philosophy Discussion Group and encouraged me to join the dance. My colleagues at Alma College have been equally supportive, most especially members of the Cognitive Science Group, Wally Beagley and Ted Hertzog, and Ron Kapp, our late Provost. Susan Dinwoody, Secretary of the Science Division, has been of inestimable help in making teXt shapely in the face of deadlines. Finally, I cannot imagine what I. would have done without the affection and expertise of my husband, Hal Slater, who knew before I did that this was what I wanted to do. iii Chapte Chapte Chapte Chapte Chapte Chapte Refere TABLE OF CONTENTS Chapter 1. Introduction . . . . . . . . . . . . . . . . . . . . 1 Chapter 2. Jerry Fodor and the Language of Thought . . . . . . . 13 Chapter 3. A Critical Look at Fodor's Semantics for LOT . . . . 52 Chapter 4. Fred Dretske’s Information-Theoretic Account of Representation . . . . . . . . . . . 89 Chapter 5. Critical Comments on Dretske’s Doctrine . . . . . . 121 Chapter 6. Saving Life As We Know It . . . . . . . . . . . . . 160 References . . . . . . . . . . . . . . . . . . . . . . . . . . 213 iv The Wh: on occ; If the imposs thing. and ca: relati. room b. about that i than, ‘ Alice'. in the indivi rOle C. full e: giVEn I liever! are 391 a pig. and be] Chapter One Introduction The White Queen told Alice (sincerely and reflectively) that she had up- on occasion believed as many as six impossible things before breakfast. If the Queen was wrong-—if it is, in fact, impossible to believe six impossible things--then the Queen believed at least one impossible thing. Beliefs can, it seems, relate us to states of affairs that do not and cannot possibly exist. Sticks and stones do not stand in such queer relations; thinkers and believers apparently do. Alice nibbled the mush- room because she wanted to grow smaller. In company with Alice's beliefs about the mushroom, Alice’s desire to grow smaller--the state of Alice that is the state it is by virtue of being her desire to shrink (rather than, say, her desire to whistle "Dixie") provides an explanation of Alice's mushroom-nibbling. But if what actually causes our behavior is, in the last analysis, neurotransmitters squirted across synapses--events individuated by biochemical or electrical properties--what explanatory role could possibly be left for states picked out by their content? The full explanation of the motions of sticks and stones can, we think, be given without resort to content; only in the case of thinkers and be- lievers do we rely for relevant generalizations upon what their states are about. Alice suspected that the Duchess' ugly baby was turning into a pig. Her conjecture was right but it might have been wrong. Thinkers and believers can be mistaken; at least some of their states are semanti may be, neither observa desires drastic to dome to redu of the Vith tVI tific v: rapreser thought, theoreti tion is and scie PSYCholc Vhat mer It is, ] termed a half of PSYcholC meaning semantically evaluable. But truth and falsity, compelling though they may be, are not aspects of the world vouched for by physical theory: neither right nor wrong, sticks and stones simply age. These (and other) observations have suggested that the capacity to bear beliefs and desires separates minded systems from mindless ones in a special and drastic way. Those who reject this division have, therefore, struggled to douesticate mindedness. It has, however, stubbornly resisted efforts to reduce it to relations and properties adequate to our comprehension of the rest of the natural order. In what follows, we shall be concerned with two recent attempts to find a place for mindedness within a scien- tific worldview-—Fred Dretske's information-theoretic approach to representation and Jerry Fodor's causal semantics for the language of thought. In this chapter, I shall place their projects in a broader theoretical context and, in particular, show how their problem situa- tion is shaped by a shared commitment to the continuity of philosophical and scientific inquiry--by the conviction that philosophers of mind and psychologists are, alike, in the business of developing an account of what mental states are. It is, I think, helpful to look at the work with which we shall be con- cerned against the background of an earlier period-~roughly, the first half of this century--during which, by mutual consent, philosophers and psychologists went their separate ways. Concern about placing mind and meaning in the world was restricted (rather remarkably, it seems in ret- rospect) to the community of professional philosophers. To the extent that t1 America he pos: be tr0t tions I terms; than a ciplina mind as linguis cations attenti tle poi analysi from em regated This 1: that their attention could be attracted to questions of this sort, American psychologists tended to assume that one way or another it would be possible to avoid commitment to what were generally acknowledged to be troublesome "mental" entities. Behavioral or physiological transla- tions would eventually be forthcoming for at least some psychological terms; properly understood, the rest would turn out to be nothing more than a convenient route from one set of observables to another.1 As dis— ciplinary folklore had it, psychology had lost not only its soul but its mind as well. At the same time, philosophers of mind who had taken the linguistic turn settled into the project of providing analyses or expli— cations of everyday psychological concepts. Psychologists paid minimal attention to their results; philosophers bent on producing then saw lit— tle point in becoming familiar with psychological research. Conceptual analysis, however it was characterized, was seen as distinct in kind from empirical inquiry. The working life of the two communities was seg- regated both in fact and on principle.2 1This is, of course, a drastic oversimplification. Some psychologists-- e.g., Kurt Lewin-—insisted that the theoretical terms of psychology would, like those of chemistry, carve the world at its joints; some-- e.g., Huzafer Sherif--insisted that groups and group properties were every bit as real as individuals and individual properties. Chroniclers of the psychoanalytic movement insisted, for their part, on the crucial significance of Freud's "discovery" that psychological properties were not reducible to biological ones. I think it is fair to say, however, that insofar as it was characterized by any systematic views, mainstream academic psychology in this country at least paid lip service to some version of reductionism and/or instrumentalism. 2Psychologists who found out about it were not amused that philosophers were accusing them of misuse of everday language. Philosophers who took notice were scandalized by what they took to be the chronic confusion of psychologists. Each community warned its novices of the dangers of (Footnote continued) Academic ps; ical doctri of science. was the on] cal empiric of philoso; joining th. nethodolog The immedi recent peg challenge: Tie L332 "The Wind Vay the g nidsr of was marke Academic psychologists did, however, pay heed to then current philosoph- ical doctrine with regard to theory construction and the proper conduct of science. It has, in fact, been said that American academic psychology was the only scientific community whose practice was influenced by logi— cal empiricism.3 Unfortunately, shared respect for this particular body of philosophical doctrine had the effect of separating rather than joining the two communities since it mandated distinct objectives and methodologies for empirical and formal sciences. The immediate background of the work we shall be considering is the more recent period during which this odd division of intellectual labor was challenged. By the early 1970's, when Jerry Fodor chose as epigraphs for The Language gf Thought E.M. Forster's "Only connect" and P.L. Travers' ”The wind is changing", one did not need to be a weatherman to see which way the wind was blowing.4 Psychology and philosophy were both in the midst of a paradigm shift marked by an eclipse of the commitments that had legitimated the separation of their concerns and competencies. It was marked, as well, by the emergence of a theoretical perspective that 2(continued) lapsing into the other. Mary Douglas' comments on the role of "purity" in academic communities seem remarkably apt in this context. 31 was surprised to find a methods text published within the last decade still using the terminology of "protocol sentences”. Admittedly, the first edition had appeared in 1960 (F.J. McGuigan, Experimental Psychology: Methods 93 Research, 4th edition (Englewood Cliffs, N.J.: \ Prentice-Hall, Inc., 1983).. 1'Jerry A. Fodor, The Language of Thought (Cambridge: Harvard University “'Press, 1975). Hereafter, LOT.‘ drew the proceSSr A declt growing accepte science least 5 be conc scienti \ 5Harold associa tin Perc (Chm that pr inVOlVe SClence Philosd domains UDlVers drew the two communities together in shared seriousness about mental processes.5 A decline in confidence in the analytic/synthetic distinction and growing acceptance of Quinean wholism undermined quite generally the accepted division of labor between philosophy and the so-called "special sciences" and, in particular, that between philosophy of mind and (at least some sorts of) psychology. Both of the authors with whom we shall be concerned emphasize the extent to which their work is continuous with scientific inquiry.6 Fodor is particularly explicit on this score. 5Harold I. Brown advances the view that logical empiricism and its associated philosophy of science functioned as a paradigm in philosophy bin Perception, Theory and Commitment: the New Philosophy pf Science (Chicago: University of Chicago Press, 1977). Recent discussion suggests that professional interests as well as intellectual commitments were involved in the division of labor between philosophy and the "special sciences". Richard Rorty discusses possible advantages to professional philosophy of dividing cognitive turf between empirical and conceptual domains in Philosophy and the Mirror pf Nature (Princeton, NJ: Princetont University Press, 1979). Historians of psychology have suggested that in the United States, at least, psychologists also had a stake in drawing a line between psychology and philosophy in a way that emphasized the dis- tinctly ”scientific" character of their enterprise (and in pointing out the exquisite applicability of their new science to the needs of busi- ness and government). See, for example, Karl Danziger, "The social origins of modern psychology"§’in Alan R. Buss, ed. Psychology 12 Social; Context (New York, NY: Halsted Press, 1979) and Thomas M. Camfield, "The professionalization of American psychology, 1870-1917", Journal pf £52 History pf the Behavior Sciences, 9 (1973), pp. 66—75. 6They are by no means alone in this. See, for example, Michael Devitt and Kim Sterelny's recent manifesto. Philosophy, we say, is not an a priori discipline. It is not a subject that can be developed apart from other areas of human knowledge. Its results form no body of knowledge against which the lesser breeds are to be tested. It is not an intellectual police force. It is empirical and fallible. . . . Philosophy is continuous with science. (Language 29d Reality: Ag Introduction 39‘} the Philosophy 9f Language (Cambridhe, MA: The MIT (Footnote continued) Elsewhere project < ask for a belief. Fodor ar ‘8 know tifie c] “'higher Cogniti: inquiry guistjc them in \ 6 (Court The distinction between a philosophical and a psycholog- ical theory is heuristic: a quick way of indicating which kinds of constraints are operative in motivating a given move in theory construction. . . [In these essays] philosophical and psychological considerations are appealed to indiscriminately. Elsewhere, he specifically rejects the traditionally "philosophical" project of linguistic analysis, telling us that there is no reason to ask for an "explication” or an "analysis" of the "ordinary notion" of belief. Nor do I really think that hunting for analyses (in that, or maybe any other, sense) is a reasonable occupa- tion for grown-ups. What we do want, or so it seems t me, is just a workable theory about what beliefs E£S° Fodor argues that confirmation in science is "isotropic"--that anything we know or believe is potentially relevant to the acceptance of a scien— tific claim. More generally, he sees isotropy as characteristic of ”'higher', 'more intelligent', less reflexive, less routine exercises of cognitive capacities."9 It is thus what we would expect in philosophical inquiry as well: if it is impossible to segregate the logical and lin- guistic from the factual in science, it cannot be possible to segregate them in philosophy, either. 6(continued) Press/A Bradford Book, 1987), p. 225. Hereafter, L&R. Unsurprisingly, Devitt has also proposed a causal theory of meaning. 7Introduction; Something on the State of the Art, Representations: Philosophical Essays 99 the Foundations pf Cpgnitive Science (Cambridge, MA: A Bradford Book/The MIT Press, 1981), p. 19. Hereafter, Intro., Rep. 8"A Reply to Brian Loar's 'Must Beliefs Be Sentences?’ in PSA 1982,1 Volume 2, pp. 644-653, p. 649. 9The Modularity 2: Mind: Ag Essay 22 Faculty Psychology (Cambridge, MA:to A Bradford Book/MIT Press, 1983), pp. 105 ff. Henceforth, MM. % Dretske is and science lem” betwec the one ha -fl—anmwrc Dretske 5 been dism Place of are not E Dretske is equally committed to continuity between philosophical inquiry and science. He deplores what he sees as a "serious communication prob- lem" between the cognitive sciences-—psychology and computer science--on the one hand, and philosophy, on the other. Without a common vocabulary with which to address the issues. . . isolation is inevitable. The result, I think, is an impoverishment of both. . . . If contact is to be made between philosophy and the wealth of relevant material in the cognitive sciences, then some bridges must be built, if only terminological bridges, between philosophical treatments of knowledge, belief, and per- ception and those scientific disciplinesoconcerned with the same dimensions of our mental life. Dretske stresses the vulnerability of his theory to what might once have been dismissed as mere brute facts of psychology. His account of the place of mind in the natural order will, he tells us, fail if the facts are not as he takes them to be. There is so much relevant material about which I am ig— norant-—some of the work in developmental psychology, for instance—-that I am prepared to see portions of this analysis condemned as factually inadequate. Given stud— ies A, B, and C, that just is not, or could not be, the way things actually happen. At the same time that disciplinary boundaries were being redrawn between philosophy and the special sciences, psychologists and philosophers found themselves sharing a metaphor that encouraged members of both com- munities to take a newly realistic stance toward mental processes. In psychology, the computer metaphor was central to an emerging cognitive paradigm. Richard Boyd characterizes it as a "theory constitutive" meta— 10Fred I. Dretske, Knowledge and the Flow 9f Information (Cambridge, MA: n A Bradford Book/The MIT Press, 1981), p. viii. Hereafter, KFI. 11KFI, p. x. phor: '[A] men and con influencing of the bas: chronicler: have been i A..""LJ-.fi (AHMD—‘mo’." phor: "[A] concern with exploring analogies, or similarities, between men and computational devices has been the most important single factor influencing postbehaviorist cognitive psychology. . . [providing] much 12 of the basic theoretical vocabulary of contemporary psychology." Other chroniclers of the rise of cognitive science offer what they take to have been a widely current interpretation of this metaphor. [Clomputers and the human mind, in some respects, are both instantiations of a general-purpose machine. They achieve their excellence by their ability to symbolical- ly represent and manipulate anything that can be specified in symbolic form--from states of the world to their own functions--and to combine a relatively small set of basic operations into complex, high-speed se- quences guided by conditional decision making. The information-processing paradigm includes the idea that human cognition, too, involves the symbolic representa— tion and manipulation of events, and the concatenation of a finite set of basic capabilities into a potentially infinite range of behavigrs, in situations that are familiar or unfamiliar. The computer metaphor not only provided psychologists with a promising theoretical vocabulary for talking about mental processes: a more strin- gent reading of the analogy, taking account of results in mathematical logic, also lent support to wholehearted realism about these processes. If, as Turing had shown, an austere and ontologically unproblematic sys- tem can carry out any fully specified logical procedure--i.e., symbol 12"Metaphor and Theory Change: What is 'Metaphor' a Metaphor For?" in, Andrew Qrtony, ed., Metaphor and Thouggg, Cambridge: Cambridge n, University Press, 1979, pp. 360—61. Boyd cites Ulrich Neisser and George Miller as sources for this observation. 13Roy Lachman, Janet L. Lachman and Earl C. Butterfield, Coggitive 9 Psychology and Information Processing: gg Introduction (Hillsdale, NJ: Lawrence Erlbaum Associates, Publishers, 1979), p. 103. Hereafter, CP&IP. This discussion has been cited in texts in the history of psychology. It may not be true but it is becoming canonical. mmfipulatio symbol mani then (at 1e realizable. scientific: place as t] .._nmrrr1"r1 H‘H'w In Philos eel of a Mind turr iorism ar held, Phi $95 of Mt in fact I tional tl a EUDCti. line of . Yet, the 4 CP&IP’ manipulation--and if psychological processes can be viewed as species of symbol manipulations (at least some of) which are precisely specifiable, then (at least some) psychological processes are unproblematically realizable. Theorizing about how minds work became recognized as a scientifically respectable enterprise; computer simulation took its place as theoretical model building at a functional level. Because a Turing machine. . . [is] universal, it can imitate other machines. A program to accomplish this im- itation is a kind of theory of the imitated system. If a program can be written to imitate some aspect of human behavior, the programmer must have a theory of how that behavior is effected. Each run of the program is like an experiment . . . Computer science has suggested to psy- chologists the appropriateness of describing internal mental processes at the "program level." This is the level of the computer analogy. It concerns the logical capabilities of humans and computers, not the physical hardware of computing machinery and biological brains. In philosophy, as in psychology, the computer metaphor was part and par- cel of a new paradigm; as psychology became cognitive, philosophy of mind turned to machine functionalism as the successor to logical behav- iorism and central state identity theory. If, as the new naturalism held, philosophy was really continuous with science, looking for analy— ses of mentalistic idiom in behavioral terms was not only unsatisfying in fact but seriously beside the point in principle. If, as the computa- tional theorists held, psychological theory was appropriately pitched at a functional level, central state identity theory, while in the right line of work, had its sights fixed on entirely the wrong target. Better yet, the puzzle that had motivated these unrewarding enterprises ap— 14cpsrp, pp. 89-90. peared tr could, i and unpr As anott wanted. That th themsel natural argued inferer PTOCes: ”ith t1 intent: mers a1 a C0nd realiz °f how intent 10 peared to have been dissolved. The physical symbol system hypothesis could, it seemed, explain how states could be both irreducibly mental and unproblematically causal. In Fodor's words, functionalism offered behaviorism without reductionism. . . physical- ism without parochialism. . . . And the seriousness of the psychologist's undertaking was vindicated by provid- ing fOISa Realistic interpretation of his constructs. Bliss! As another theorist put it, "It seemed that we could have everything we wanted. The world was made one again."16 That the computer metaphor and a computational psychology would not, by themselves, suffice to establish mindedness intelligibly within the natural order soon became evident, however. Both Dretske and Fodor argued that what computers do can be construed as computation-—as inference, addition or anything else interestingly analogous to a mental process--only to the extent that computer states can be seen as endowed with the appropriate representational content. But, they continued, the intentionality of computer states depends upon that of computer program- mers and users. Computational domestication of mindedness thus rests on a conditional: mental states can be identified with unproblematically realizable computational states only if we can give an adequate account of how such states could come by an original, underived or "intrinsic" intentionality.17 15Intro., Rep., pp. 10-11. 16Stephen Stich, informal comment to NEH Summer Seminar, "Philosophical Implications of Cognitive Science," 1989. Both Dretsk Each beliew of mental s ported scir follow, I 1 and raise « doubt that tic semant concerned ion shoulc least, {hi the langu nation 80 Purchase In tufnir COncerne( OPhy. It ;;“--.. Proble shadow t that it "folk Ps rth Sci Daniel D SEqUenCe 11 Both Dretske and Fodor offer proposals designed to meet this demand. Each believes that in the absence of an adequate account of the content of mental states, psychological idiom has a place neither in any pur- ported science nor in our everyday practice. In the chapters that follow, I shall lay out, as clearly as I can, the doctrine being urged and raise critical questions about its success. I shall find reason to doubt that either Dretske or Fodor has delivered an adequate naturalis- tic semantics for mental states. The final chapter will, therefore, be concerned with the question of how seriously a friend of intentional id— iom should take such a verdict. I shall argue that, in Fodor's case, at least, the route from failure to provide a causal theory of content for the language of thought to elimination of everyday psychological expla— nation goes by way of a philosophy of science that has, itself, little purchase on our loyalties. In turning our attention to the two projects with which we shall be concerned, we shall be looking at a self-consciously new kind of philos— ophy. It may, however, be worth reminding ourselves that this view of 17Problems of intrinsic intentionality were by no means the only ones to shadow the "Panglossian paradigm". Another major line of objection was that it was very unlikely that the psychological states recognized by "folk psychology" would (or could) have analogues in any possible cor- rect science of the mind. This point was argued, in various ways, by Daniel Dennett, Patricia and Paul Churchland and Stephen Stich. One con- sequence of seeing philosophy and science as continuous was to demote everyday notions from their position as privileged targets of conceptual analysis to that of (possibly unredeemably poor) theoretical terms. It is for this reason that, as we shall see, both Fodor and Dretske feel obliged to defend their care for everyday idiom before launching into an explication designed to prevent its elimination. philosophy Leibniz puz "only the ; anything wt failure of of the men his task at matter tha Dretske, h inconsiste theories c 0f things in philosr work Vith doctrine 12 philos0phy stands at the very start of our modern enterprise. When Leibniz puzzled over the thought factory and concluded that, seeing "only the parts impinging upon one another. . . we should not see anything which would explain a perception," he was concerned with the failure of the best going physical theory to give a satisfactory account of the mental. Like Fodor and Dretske, he saw no discontinuity between his task and that of the physical theorist who described particles of matter that "knock, impel, and resist one another," and, like Fodor and Dretske, he was disinclined to attribute to a physical system properties 18 That our inconsistent with its nature as thus best understood. theories of mind are responsible to everything we know about the nature of things as well as to our logic and our language is not new doctrine in philosophy; it is a view to which we have, at length, returned. The work with which we shall be concerned is evidence of how difficult a doctrine it is to put into practice. 18Honadology (1714) 17. In Leibniz: Philosophical Writingg, ed. G.H.R.\A Parkinson, tr. Mary Morris and G.H.R. Parkinson (London: J.M. Dent & Sons, Everyman’s Library, 1973 (1934),\p. 181; New Essays 2g Human,5 Understanding, trans. and ed. by Peter Remnant and Jonathan Bennett (Cambridge: Cambridge University Press, 1981, 1985), p. 439. Almost a d work on tt i Ten years In his m 1Ogical d° by v:‘ Cerned r abOUt i of its Chapter Two Jerry Fodor and the Language of Thought Almost a decade ago, Jerry Fodor concluded a retrospective survey of work on the foundations of cognitive science with this observation: What we need now is a semantic theory for mental repre- sentations: a theory of how mental representations represent. Ten years later, Fodor finds our situation much the same. [Tlhe main joint business of the philosophy of language and the philosophy of mind is the problem of representa- tion itself: the metaphysical question of the place of meaning in the world ordgr. How can anything manage to be about anything. . In his most recent work, Fodor offers a tentative answer: primitive non- logical symbols of the language of thought express the properties they do by virtue of their possible causes. In this chapter I shall be con- cerned with straightforward exposition of Fodor's proposal. Questions about its success are reserved for Chapter Three; closer consideration of its claims to constitute, as Michael Devitt has put it, a defense of "Life As We Know It" is postponed until Chapter Six.3 1Intro., Rep., p. 31. 2Jerry A. Fodor, Psychosemantics: The Problem of Meaning” in the Philosophy of Mind (Cambridge, MA: A Bradford Book/The MIT Press, 1987), p. xi. Hereafter, Psysem. 3"Against Direct Reference," forthcoming in Midwest Studies in Philosophy, Volume XIV, Contemporary Perspectives in the Philosophy of Language II, n. , p. 35. 13 Fodor thin attack by r worries ab concerns st belief/des theory of of a seman is needed. Fodor end States as titulars’ by a reaj formulae Semantic leng p1 FodOr,S recent belief dire Q_ theta there ‘hink. 14 Fodor thinks that both scientific and "commonsense" psychology are under attack by eliminativists whose brief includes a battery of ontological worries about intentional and semantic properties. Fodor takes these concerns seriously. Unlike the eliminativists, however, he believes that belief/desire psychology can and should be defended. Fodor's preferred theory of mind leads him to believe that it is, in particular, provision of a semantics for linguistically structured mental representations that is needed. Fodor endorses a theory of mind that construes contentful psychological states as computational relations to internal formulae. As physical par— ticulars, such formulae can be endowed with the causal powers demanded by a realist understanding of commonsense psychological explanations. As formulae involved in computations, however, they have intentional and semantic properties as well and it is to the explication of these puz- zling properties that Fodor's causal theory is addressed. Fodor's championship of intentional idiom is longstanding. In his most recent work, his emphasis has been on "vindicating" ordinary, everyday belief/desire explanations,4 but he has earlier drawn attention to the dire consequences of its elimination for scientific psychology. "If there are no facts about mental states and processes," he warns, "then there is, quite literally, nothing for psychology to be about".5 Fodor thinks that there lg something for psychology to be about; he thinks 4Psysem., p. 10. 15 that there are generalizations about us (and other systems) that cannot be expressed without reliance on intentional idiom. "The paradigm situation--the grist for the cognitivist's mill--is one where proposi- tional attitudes interact causally and do so ig virtue pf their content."6 The laws of psychology are laws about the content of proposi- tional attitudes: we cannot formulate them without regard to what a belief is about, what a desire is Egg. [Sleeing that a is F is a normal cause of believing that a is F;. . . statements that P are normally caused by beliefs that P; observations that many of the gs are P often contribute to the etiology of the belief that all the gs are F; the belief that a thing is red is a normal cause of the inference that the thing is colored; and so on and on. Defending "real (viz. cognitive) psychology" therefore demands a defense of intentional locutions.8 To some degree, Fodor's defense of informal, everyday belief/desire psy- chology rests on a similar basis: he sees commonsense explanations as respectable protoscience. Just as they stand, everyday ascriptions of belief and desire have impressive predictive and explanatory power; suitably refined and extended, they will be capable of participating in a genuinely scientific psychology. Fodor's claim on their behalf is, 5"Three Cheers for Propositional Attitudes", in Rep., pp. 102-3. Hereafter, Three Cheers. 6"Propositional Attitudes", in Rep., p. 183. Hereafter PAs. 7Intro., Rep., p. 25. 8PAs, p. 184. 16 however, even stronger. Commonsense psychology is not only a good theory and therefore one that we ought not to give up; it is an indispensable part of our conceptual scheme, one that we are in no position to aban- don. Intentional explanation, he tells us, lies as close as can be to our "cognitive core," in two distinct but closely related ways.9 First, intentional explanation undergirds our thought and, perhaps even more important, our practice.10 "The idea that behavior is explicable in terms of the propositional attitudes of the agent plays . . . quite a special role in our world view: nobody seriously doubts it".11 We have no idea of how to explain ourselves to ourselves except in a vocabulary which is saturated with belief/ desire psychology. One is tempted to transcendental argument: What Kant said to Hume about physical objects holds, mutatis mutandis, for the propositional attitudfg; we can't give them up because yg don't know how to. Second, Fodor suspects that the de facto pervasiveness of intentional explanation reflects a deeper natural necessity. He thinks that the way our intellects are constituted makes it unlikely that £22111 doing with- out intentional explanation is even "a biologically viable option” for us.13 Fodor speculates that the ubiquity of intentional explanation is, in fact, best explained by an innate predisposition so to construe each other and that this genetic determination is, in turn, best explained by 9Psysem., p. xii. 10Psysem., p. xii. 11Three Cheers, p. 122. 12Psysem., pp. 9-10. 13Psysem., xii. 17 the selective advantage conferred by an (at least approximately true) "innately apprehended commonsense psychology".14 Fodor argues that the survival of creatures as intricately and deeply dependent upon each other as we are demands the ready availability of "predictively adequate views about how. . . conspecifics are likely to behave".15 We are so built that we have, as individuals, no alternative to belief/desire psychology because, as a species, we could not do without such a theory. It is, I think, a measure of the depth of Fodor's concern for belief/desire psychology that he is willing to offer on its behalf just the sort of argument that he elsewhere dismisses as "Pop—Darwinism".16 If both canons of rational theory choice and our neural hardwiring are on the side of intentional explanation, what philosophical work remains to be done? The answer lies in a cluster of commitments shared by Fodor and his eliminativist adversaries. 0n Fodor's view, making use of an explanatory apparatus puts us under an obligation to countenance the entities it features. At the same time, other of his views sharply limit his ontological hospitality. From Fodor's point of view, it is troubling that, although commonsense psy— chology can be made consistent with the demand that any entity with causal powers be a physical entity, it also gives every sign of commit— ting us to the possession by those entities of ontologically suspect 14Psysem., "Epilogue: Creation Myth", p. 132. 15Three Cheers, p. 122. 16Psysem., p. 106. 18 semantic and intentional properties. When, for example, we field explanations of the form 'Robin is going to Chicago because Robin thinks you can get good pizza in Chicago (and Robin wants good pizza)’, it cer- tainly gggggg for all the world as if we were talking about a relation between Robin and objects of Robin's believing and desiring. We are inclined to say that there is something that Robin believes and desires (from here on, I shall deal only with belief) and that it is on account of Robin's being in that relation to that something that Robin is Chicago-bound. We are also inclined to say of what Robin believes--that you can get good pizza in Chicago-~that it can be true or false, and, what is more, that whether it is true or false depends upon how things stand with regard to what it is about-~in this case, the goodness of pizza in Chicago. In sum, deployment of commonsense psychology in ex- planation commits us to the literal truth of (at least some) ascriptions of belief; ascriptions of belief, taken at face value, commit us to re— lations to objects of belief; the objects of belief to which we are committed have both causal ggg semantic and intentional properties. Fodor believes that functionalism, appropriately conjoined with what he calls "the computer metaphor", allows us to take ascriptions of causal interactions among psychological states seriously. It seems unlikely, however, that intentionality or truth evaluability will be recognized by a completed physics and from Fodor's viewpoint, a condition on the onto- logical acceptability of properties is that they be understandable in terms of properties acknowledged by basic physics. Thus, "If aboutness is real, it must be really something else."17 We shall return, in 19 Chapter Six, to closer consideration of the eliminativist/reductionist premise that lends urgency to Fodor's project. Here, we shall simply note that for Fodor, as for many other theorists, intentional properties pose a puzzle--one that, on his account, survives the successes of functionalism. Let us see how this works. First, the successes of functionalism and a "computational" theory of mind. Fodor thinks that functionalism provides a framework for assigning caus- al powers to mental states. Functionalism holds that "what determines the psychological type to which a mental particular belongs is. . .[its] causal role in the mental life of the organism".18 On this view, being in one rather than another psychological state is a matter of being in a state that occupies one rather than another position in a matrix of pos- sible causal relations. A system has, e.g., beliefs and desires insofar as it occupies states that are causally related in ways that (some good psychological theory says that) beliefs and desires are related, regardless of how these states happen to be realized. "Functional indi- viduation is individuation with respect to causal role; for purposes of psychological theory construction, only. . . causes and effects. . . count."19 Functionalist doctrine with regard to mental kinds allows (but does not require) that mental particulars be describable, one and all, in the vocabulary of physics. Functionalism is thus consistent with 17"Meaning and the World Order", Psysem., p. 97. '° 18"the Hind-Body Problem", Scientific American Jan. 1981, p. 118. 19Intro., Rep., p. 11. 20 token physicalism without imposing type physicalism. It is not sur- prising that Fodor is among the philosophers of mind who have endorsed functionalism. Viewed from one perspective, it offered behaviorism without reductionism; viewed from another, it offered physicalism without parochialism. The idea that mental particulars are physical, the idea that mental kinds are relational, the idea that mental processes are causal, and the idea that there could, at least in principle, be nonbiological bearers of mental properties were all harmonized. And the seriousness of the psychologist's undertaking was vindicated by providing a Realisti O interpretation of. . . [psychological] constructs. Turing machine functionalism is particularly appealing. [Tlhe inputs and outputs of Turing machines are extreme- ly restricted, and their elementary operations are extremely trivial. . . .[Ilt is a correspondingly simple problem to build machines that can realize them. . . . [A] psychological explanation that can be formulated as a program for a Turing mgfhine is ipso facto mechanically realizable. Since Turing machines are capable of carrying out any symbol manipula- tion that can be formally specified, a psychological theory that conceptualizes mental processes as such symbol manipulations--as "computations”-—can be cast in this form. Fodor enthusiastically endors— es computational functionalism as metatheory for cognitive psychology. He believes that it is the only philosophical position consistent with the practice of cognitive psychologists, whose goal, he tells us, is to fill in "explanation schema of, roughly, the form: having the attitude R 20Intro., Rep., pp. 10-11. 21Intro., Rep., p. 14. 21 £2 proposition 2 ig contingently identical £9 being ig computational 22 relation 9 £2 the formula[e]. L F." [Tlhe psychologist assumes that some organic processes satisfy descriptions like 'storing, accepting, rejectingz computing, etc. P' and that the organism learns, perceives, decides, remembers, believes, etc., whatever it does because 153stores, accepts, rejects, or computes whatever it does. From the philosopher's perspective, it is "at the computational level that the causal gap between mental representations and bodily goings-on 24 is bridged." On Fodor's account, what the system stores, accepts, rejects, computes, and so on is something that can be described as a sentence or a formula: the particulars to which causal powers are ascribed instantiate symbol . [Clomputational states. . . can [in turn] be directly explicated as relations between the organism and formulae. . . .[Ilnsofar as one can (loosely) say that the organism stores the information that P, one must be able (s strictly) to say that the organism is in a certain computational relation to the formula P. . [i. e.,] the relation of storing P. . .[Flor any propositional atti- tude of the organism. . . there will be a corresponding computational relation between the organism and some formula(e) of the internal code such that (EBB organism has the propositional attitude iff the orgggism lg lg that relation) is nomologically necessary. Functionalism allows us to be realistic about psychological states; machine functionalism--specification of psychological states in compu- 22Lo'r, p. 77. 23LOT, p. 75. ’\24J. Heil, Review, Representations, Philosophical Books, 23 (Oct. 1982), pp. 231-233. 25 LOT, p. 75. 22 tational terms --allows us (at least in principle) to satisfy criteria of mechanical realizability. To see psychological states this way is, however, to see the particulars involved (physical though they may be) as endowed with intentional properties that still demand explication. Fodor describes machine functionalism as requiring "a dual-aspect theory all the way down, with intensional characterization specifying one of the aspects and mechanical characterization specifying the other".26 Fodor's view is that just as we can view computer states as having con- tent--as being encodings, or representations of, e.g., numbers--as well as describe them in a physical vocabulary, so we can see mental states as bodily events (in the brain, we suppose) which also serve as encodings or "mental representations". Psychological processes-— inferences, judgments, decisions—~are, on this view, analogous to computations in that they are formal operations on 'shapes' which carry content. Just as we can explain why the computer prints out '12' when we offer it 'Sum(7,5)' by invoking the fact that, under the rules of addi- tion, 7+5=12, so we can explain why Mary fled the building on hearing that it was burning by invoking the observation that a rule of rational decision making mandates that if one believes that X is hazardous to one's health, one tries (ceteris paribus) to distance oneself from X. Just as there are rules which relate computer states by virtue of their numerical content, so there are rules which relate psychological states by virtue of their propositional content. In either case, explanation 26Intro., Rep., p. 22. 23 is a matter of displaying a sequence of states as rule—governed with regard to their content. Mental states (including, especially, token havings of propositional attitudes) interact causally. Such inter- actions constitute the mental processes that eventuate (inter alia) in the behaviors of organisms. . . . [Ilt is crucial to the whole program of explaining behavior by reference to mental states that the propositional attitudes belonging to these chains are typically7ggg- arbitrarily related in respect to their content. The computer metaphor relies on there being a relation between the con— tents of successive mental states that makes it appealing to talk about sequences of states as derivations rather than merely as formal transformations.28 As Fodor puts it, When we think of an organism as a computer, we attempt to assign formulae in the vocabulary of a psychological theory to physical states of the organism (e.g., to states of its nervous system). Ideally, the assignment should be carried through in such a fashion that (some, at least) of the sequences of states that are causally implicated in the production of behavior can be interpreted as computations which have appropriate des— criptions of behavior as their 'last line'. The idea is that, in the case of organisms as in the case of real computers, if we get the right way of assigning formulae to the states it will be feasible to interpret the sequence of events that causes tag output as a computa~ tional derivation of the output. The notion of computation, Fodor tells us, "is intrinsically connected to such semantic concepts as implication, confirmation, and logical consequence. Specifically, a computation is a transformation of repre— 27PAs, p. 182. $ 28David Hills, "Introduction: Mental Representations and Languages of Thought" in N. Block, ed. Readings 1g Philosophy gf Psychology (Cambridge, MA: Harvard U. Press, 1981), vol. 2, pp. 17-18. 29 LOT, pp. 73—74. 24 sentations which respects these sorts of semantic relations."3O Cognitive psychology has as its natural domain that subset of relations among psychological states that are at least in a "loose sense rational": each must be "analyzable as a relation to a representation, and. . . its causal antecedents (or consequents or both) should be analyzable as relations to semantically related representations."31 "If," says Fodor, "translation is that enterprise in which we do our best for the rationality of texts, cognitive psychology is the one in which we do our best for the rationality of mental processes at large."32 We can now appreciate the seriousness with which Fodor treats the slogan, "No computation without representation."33 The crucial place of mental representation in his theory leads him to offer a characteriza- tion of the system of representations to which he is committed. The token physical events to which we stand in computational relations are not only representations--they not only have semantic properties—~but belong to a system with a syntax or, as Fodor puts it, with "a combinatorial semantics: the kind of semantics in which there are (relatively) complex expressions whose content is determined, in some 34 regular way, by the content of their (relatively) simple parts." Fodor 3OMM, p. 5. 31LOT, p. 203. 32LOT, pp. 198-199. 33LOT, p. 34. 25 cites various "glaring facts" that, he believes, count toward explicat- ing propositional attitudes as relations between organisms and representations that are best construed as "formulae in an internal language. . . internal sentences, as it were,"35 arguing that cognitive states (and not just their intentional contents) have such constituent structure. For one thing, this hypothesis is the best (because the only) way to explain the generativity or productivity of the propositional attitudes: There is a (potentially) infinite set of. . . belief- state types, each with its distinctive intentional object and its distinctive causal role. This is immedi- ately explicable on the assumption that belief states have combinatorial structure; that they are somehow built up out of elements and that the intentional object and causal role of each such state depends on what 36 elements it contains and how they are put together. In response to the objection that we have no warrant to invoke such un- limited capacity, even as an idealization, Fodor has more recently formulated an argument that relies, instead, upon what he takes to be generally accepted empirical facts about what he terms the systematicity of linguistic and other behavioral capabilities: "[Tlhe ability to produce/understand some sentences is intrinsically connected to the ability to produce/understand others. . . . You don't. . . find native speakers who know how to say . . . that John loves the girl but don't 37 know how to say. . . that the girl loves John." The same goes for non— 34Psysem., ”Why There Still Has to Be a Language of Thought", p. 138. Henceforth, Language. 35Rep., p. 187. 36Language, p. 147. 26 linguistic competencies: we simply do not find in the animal kingdom "punctate" minds, "capable of learning to respond selectively to a situation such that aRb but quite unable to respond selectively to a situation such that bRa."38 This is just what would be predicted by "a theory that says that the sentence 'John loves the girl' is made out of the same parts as the sentence' the girl loves John', and made by appli— cations of the same rules of composition";39 it would, on the other hand, be monumentally accidental on a theory that denies constituent structure to the events that carry representational content.40 Fodor's theory of mind thus holds that mental states are functional--in particular, computational—-relations between organisms, on the one hand, and entities with sentence-like properties, on the other; these "formulae" are representations encoded in an inner, unlearned language 41 of thought. We can now understand why Fodor thinks that the single 37J. A. Fodor and Z. W. Pylyshyn, "Connectionism and Cognitive Architecture: A Critical Analysis" Cognition (1988), pp. 3-71, p. 37. Final stress added. Hereafter, C&CA. 38Language, p. 153. Fodor acknowledges the existence of isolated exceptions to this generalization. 39caca, pp. 48-49. 40"One man's affirmation of the consequent is another man's inference to the best explanation" (Psysem., p. 149). Elsewhere, discussing the fu- sion theory of the syntactic objects of verbs of propositional attitude, Fodor notes that a good theory does not leave the landscape littered with coincidences--e.g., the fact that we feel that there is a relation between 'John believes that it is raining' and 'John believes that it is raining hard’ (PAs, p. 179), both remarks are, I think, intended to sig— nal the belief that his project is governed by canons of scientific theory choice as well as by philosophical principles. 27 most pressing problem in philosophy of mind is giving a naturalistic account of intentionality, providing a theory that articulates, in non-semantic and non- intentional terms, sufficient condzgions for one bit of the world to be 2999; another bit. If we construe the content of complex mental representations as (in some sense) a logical function of constituent representations, the problem of legitimizing intentional idiom reduces to that of providing a naturalis- tic semantics for the basic nonlogical "vocabulary" of Mentalese.43 Defending commonsense (and scientific) psychology, then, is a matter of overcoming the objections of a philosophical adversary who says that 41The most obvious reason for postulating a "ggg-natural" language of thought is that it should be possible to ascribe beliefs to nonlinguis- tic beings, at least some of whose doings we think we can explain by belief ascription. Fodor thinks the language of thought must be un- learned because he believes that acquisition of a natural language, N, requires mental states such as hypothesizing that P, where 'P' is a characterization of truth—conditions for some expression of N. It is, Fodor notes, impossible for P always to be an expression of the language that is being acquired; at some stage, hypotheses about the semantics of N, and evidence for and against these hypotheses, will have to be formu— lated in something other than the not-yet-available N. I shall have nothing whatever to say about Fodor's innateness hypothesis, in parti- cular, about how it is to be harmonized with his causal semantics. 42Psysem., p. 2. Fodor is also committed to the "atomism" suggested by this formulation. See Chapter 3, "Meaning Holism" in Psysem. I shall not be taking up his arguments against non—atomistic theories. 43Fodor admits that there may be some puzzles about the mode of composi— tion, but does not pause to worry about them. One wonders. The logic of Mentalese could be problematic. For example, if some of the computations in question are those involved in decision theory, as Fodor suggests, it would seem that we need to be able to express possibility and necessity. Moreover, Fodor's Gricean tack requires that competent speaker/hearers compute over representations of (other speaker/hearers') beliefs and desires. Modal logics and logics of belief are, however, notably nonstandard. Fodor faults prototype-theory—plus-fuzzy-set—theory for not meeting demands of compositionality but it is not clear that he is really prepared to meet them, either. 28 "there are a priori, metaphysical reasons for supposing that semantics can't be naturalized." Fodor thinks that all he needs to win the argu- ment (and save Life As We Know It) is "a plausible, naturalistic, 44 sufficient condition for 'A's meaning g." The work to which we now turn is an attempt to provide such conditions. Fodor offers, first, a condition of adequacy for a naturalistic theory of representation. [Wlhat we want at a minimum is something of the form :3 represents §l lg true lll Q where the vocabulary in which condition C is couched contalgs neither inten- tional nor semantical expressions. Fodor takes 'represents' at its face value as a two-place relation and tells us that "only two sorts of naturalistic theories of the represen— tation relation. . . have ever been proposed. . . . [Alt least one of these [-—some sort of resemblance relation between R and §--] is cer- 46 tainly wrong". This leaves us with some variety of causal theory, a candidacy that Fodor takes to be "prima facie attractive": Causal relations are natural relations if anything is. You might wonder whether resemblance is part of the nat- ural order (or whether it's only, as it were, in the eye of the beholder). But to wonder that about sausation is to wonder whether there lg a natural order. 44Psysem., p. 124. \°45"Semantics Wisconsin Style," ynthese 52 (1984), p. 232. Hereafter, SWS. This was, it should be noted, an early formulation of the project. It is clear that 'iff’ is too strong. 46SWS, p. 233. 47sws, p. 233. \\ 29 Causation also has appealing formal properties: like representation, it is an asymmetric relation that holds between particulars. This latter is important because one of the advantages Fodor sees in functionalism is its ability to accommodate token physicalism: the objects of (tokenings of) propositional attitudes are tokens of physical as well as semantical types. Fodor does little to try to convince us that causation reconstructs (or is even relevant to) our pretheoretic notion of representation. Perhaps this is because causality is so clearly involved in paradigmatic cases of representation--in depictions and perceptions. What P is a picture (or a perceptual experience) of seems to depend on 8 what caused P (in the relevant way).4 Analogy with the causal theory of the reference of names and natural kind terms may also paly a role here: [I]f tokens of 'water' have a different interpretation on Twin-Earth. . . or, equivalently, if tokens of 'water’ are type-distinct from tokens of 'water2'. . [that is] because it's XYZ that bears to 'water2' tokens the sortégf causal relations that H20 bears to tokens of 'water'. Elsewhere, Fodor argues that semantic evaluation of beliefs is most rea— sonably done "with respect to the kind of . . . [stuff] that gave rise to them"--i.e., with respect to "relevantly local samples" of that kind of stuff, where "relevant localness is fundamentally an etiological . .,50,51 notion. Fodor sees his major task as one of coming up with a 48Although, as B. Abbott has pointed out, pictures of, e.g., Santa Claus present problems for a pure causal account of this sort. 49Psysem., p. 98. 50"Cognitive Science and the Twin-Earth Problem", Notre Dame Journal pl Formal Logic, 23 (1982), p. 113. Hereafter, Cog Sci. 30 version of causal theory that can meet standard challenges——in particu— lar, that of allowing the possibility of glgrepresentation. In Psychosemantics, therefore, he introduces what he takes to be an intui- tively appealing "Crude Causal Theory" of content (CCT), and then proceeds to refine it in the face of predictable objections. It will be helpful, in considering Fodor's proposal, to begin with some notational conventions. (1) Terms for Mentalese symbols--for types of mental representations or concepts in the language of thought-— are capitalized or, in the case of single letters, bolded. Thus: HORSE, RED, PROTON, A (2) Terms for properties are italicized [underlined]. Thus: horse, red, proton, g 51It seems likely that Fodor is also relying on the ”only game in town" strategy here. We may compare Devitt's "transcendental argument": What sort of relation between a speaker and the external world could determine reference [which is a relation to particular things outside the head]? From a naturalistic viewpoint, there is only one possibility: a causal relation. . . . To suppose that one's thoughts [i.e., descriptive beliefs] can reach out to particular objects outside the mind is to have magical theories of refer- ence and intentionality" ("Meanings Just Ain' t in the .L Head," in, Method, Reason, and Langga g: Essays in Honour of Hilary Putnam, ed. George Boolos, Cambridge University Press, forthcoming, mss. pp. 8- 9). Devitt distinguishes between the demand that intentional properties be reduced "in some strong old-fashioned sense" to physical ones, on the one hand, and that they be explained "in more basic physical terms," on the other. He notes, with admirable candor, that What that requirement amounts to is not perfectly clear to me nor, I suspect, to anyone else. Still, we are all fairly good at recognizing a satisfactory explanation when we see one (p. 39). The continuing debate between Fodor and Dretske with regard to criteria of adequacy for an account of intentionality suggests that this last may be a bit overoptimistic. 31 (3) Terms for tokens of symbols in the language of thought——for a particular mental representation-—are set off by single quotation marks. Thus: 'horse'(s), 'red’, 'proton'(s), 'A'(s) (4) Terms for instantiations of properties are left unadorned. Thus: horse(s), red, proton(s), A(s) Fodor's concern is to build on "the intuition that. . . semantic inter- pretations of mental symbols are determined by, and only by, . . nomological relations" between higher order properties-~between the property of being an instance of a property, 5, and the property of 52 being a tokening of a Mentalese symbol, A. "The Crude Causal Theory says, in effect, that a symbol expresses a property iff it's nomologi- cally necessary that gll and only instances of the property cause 53 tokenings of the symbol". The Psychosemantics account amends CCT in three ways. First, Fodor retreats from (i) a demand for necessary and sufficient conditions for A to represent 5 to (i') provision of "plausible sufficient conditions." Second, he softens the demand for (ii) a purely nomological link between symbol and property expressed to (ii') the less stringent requirement of reliable correlation between them. Finally, he attempts to make provision for error, which he charac- terizes as the possibility that particulars other than As can cause 'A's. This last requires a more elaborated characterization of the req— uisite nomological link between g instances and A tokenings. In recent 52Psysem., p. 99. 53Psysem., p. 100. 32 work, Fodor has, in addition, considerably amended the Psychosemantics approach to the problem of error. Let us consider the first round of refinements to CCT before picking up Fodor's afterthoughts. Fodor notes that the requirement that gll As cause 'A’s can hardly be what we want: it is an implausibly strong condition. The idea was that my Mentalese symbol . . . [HORSE] ex- presses the property of being a horse only if gll instantiations of the property cause the symbol to be tokened. . . . But that's preposterous on the face of it. What about Ancient Athenian horses? What about horses on Alpha Centauri? What about 40th Century horses? What about horses ig4Peking (where, unlike many horses), I have never been? What we need, Fodor says, is a satisfiable condition, "a condition such that it's plausible that, in at least some cases, 'A's express 5 because 55 that condition lg satisfied". Fodor's solution is to extend CCT to in— clude counterfactual causal properties. HORSE expresses horse even if no horse has ever elicited 'horse'--even if no horse ever does elicit ’horse’—-because there are "circumstances such that. . . [(a)] instanti- ations of horse would cause 'horse' to be tokened in my belief box. were the circumstances to obtain; and (b) 'horse' expresses the property horse (in my ideolect of Mentalese) in virtue of the truth of (a)".56’57 54Psysem., p. 111. 55Psysem., p. 111. 56Psysem., pp. 111-112, my stress. 571"odor extends this counterfactual strategy to symbols that express uninstantiated properties as well: UNICORN expresses unicorn if unicorns--were there to be any--reliably elicited ’unicorn's. "Perhaps (Footnote continued) 33 The first modification of CCT, then, is from the condition that all in- stantiations of 5 cause 'A's, to one that all instantiations of A would cause 'A's under relevant circumstances. Fodor notes that it is, in general, difficult to characterize the requisite contrary-to-fact con- ditions in naturalistic--i.e., nonintentional--terms. He believes that there is, however, at least a subset of Mentalese symbols the conditions for whose elicitation can be so specified and that this account can be extended to other sorts of symbols. He calls concepts capable of such straightforward explication "psychophysical" (or, alternatively, "sensory") concepts. Sensory concepts are "ones for which it appears most clearly possible to enumerate the conditions in which (reliable, causal) correlation makes content".58 Psychophysics is precisely in the business of telling us how much of the wall has to be painted red. . . and how close to the wall one has to be, and how bright the lights have to be, and so forth. . . such that if it's that much red, and that bright and that close. . . then you'll think 'red' if your eyes are polated toward the wall and your visual system is intact. Fodor notes with satisfaction that characterization of such optimal psy— chophysical conditions can be carried out entirely "in nonintentional, nonsemantic vocabulary: in the vocabulary of wavelengths, candlepowers, 6O retinal irradiations and the like". Thus, in at least some cases, "a 57 (continued) what CCT should say about unicorns is that they would be nomically sufficient for 'unicorn'-- tokenings if there were any" (Psysem., pp. 163-64, n.). He suggests that uninstantiable properties cannot be expressed by primitives of Mentalese. 58 Psysem., p. 114. 59Psysem., pp. 112—113. plat PYOI inst It: psy< thir argt pro] the Dep bes his tic con tal W01- 34 plausible sufficient condition for certain symbols to express certain properties. . . [is] that tokenings of those symbols are connected to instantiations of the properties they express gy psychophysical law".61 It is plausible that RED expresses Egg because, be there ever so many instances of redness that come and go without eliciting a tokening of 'red' in anybody's head, we can nevertheless say, without lapsing into intentional idiom, that they ygglg have elicited such tokenings gag the psychophysical conditions been such and such: "[Tlhe fact that red things force 'red' upon one in psychophysically optimal circumstances is arguably all there is to 'red' expressing Egg".62 "For the concepts and properties that it applies to," Fodor notes, ”psychophysics is just what the causal theory ordered".63 Deployed counterfactually, the deliverances of psychophysics can, at best, silence the metaphysical skeptic against whom Fodor is directing his philos0phical argument. They do not give us a full-spectrum seman- tics for the language of thought, however: "[Nlot all the organism's concepts are under the sort of local causal control that psychophysics 64 talks about". This bothers Fodor: "If the Causal Theory is going to ‘work at all, it's got to work for the so-called 'theoretical 60Psysem. , p. 113. 61Psysem., p. 113. 62Psysem., p. 116. 63Psysem., p. 113. 6l‘Psysem., p. 113. 35 vocabulary”.65 Psychophysics is not only inadequate to providing the appropriate counterfactuals for PROTON, however: it cannot handle even such 'observable' concepts as HORSE. "It's perfectly nomologlgally possible to be in a psychophysically optimal relation to a horse and yet not token the concept 'horse'," Fodor tells us.66 Fodor's suggestion is that we allow for longer causal chains, some of whose links are constituted by non-nomological but nonetheless reliable relations. Although horse is not a psychophysical property, [Ilnstantiations of horse are, very often, causally re- sponsible for instantiations of what Egg psychophysical properties. It is. . . because Dobbin is a horse that Dobbin has that horsy look. And it's entirely plausible that having that horsy look reduces to having some or other (maybg7quite disjunctive) bundle of psychophysical properties. In some creatures, with some histories, a computational process will take psychophysical concepts (lawfully caused by the psychophysical pro- perties whose instantiations are caused-—very often--by horses) as input and reliably deliver ’horse' as output. If there is a Mentalese symbol that is reliably tokened when members of this particular set of psycho- physical concepts are tokened, then that symbol expresses horse. HORSE means horse if 'horse' tokenings are reliably caused by tokenings of psychophysical concepts that are caused by instan— tiations of pyschophysical properties for which instantiations of horse are in fact causally respon- sible. The causal chain runs from horses in the world to horse-looks in the world to psychophysical concepts in 65Psysem., p. 112. 66Psysem., p. 116, my emphasis. 67Psysem., p. 118. US th an. ll: ‘. H813: 36 the yes-box to 'horse’ in the yes-box. . . 68[HORSE] means horse because that chain is reliable. Diagrammatically, Fodor's sufficient condition for HORSE to mean horse looks like this. Dobbin, ——> [P1,P2,P3. . .Pn] _-> ['Pl','P2','P3'. . .’Pn’} —_> horse' Old Paint, Silver horse instances of tokenings of tokens of instances psychophysical psychophysical HORSE are very properties concepts i.e., 'horse’s. often which lawfully which reliably causally cause cause responsible for Even though they are not pure psychophysical concepts, HORSE and CHAIR (and perhaps SMILE) are observational concepts. How so? For one thing, the psychophysical properties to which instances of ggggg or Egglg are very likely to give rise are conveniently detectable by our sensory ap- paratus. In addition, the going-togetherness of these psychophysical properties and the non-psychophysical properties in question-—ggggg, Eggl£--can be equally conveniently detected, and thus come to influence us. Some mechanism in us that is sensitive to such correlations realizes the function that takes disjunctions of psychophysical concepts as input and delivers an observational concept as output. Observational proper— ties are thus ones that we can see with the naked eye and respond to barebrained. 68Psysem., p. 122. 37 Not all properties fall into even so extended a causal net, however. Fodor points out that pgglgg is a different sort of property from ggggg (and thus that PROTON is a different sort of concept from HORSE) because "there isn't [as in the case of horses] a lggk (taste, smell. . . etc.) that being a proton is typically causally responsible for a thing's having".69 Protons, he notes, are too small to see "even with the lights 70 Fodor thinks that we can accommodate nonobservational con— turned up". cepts like PROTON by adding a few more links to the causal chain. Let us look at this further extension of CCT. First, protons, unlike horses, are not ordinarily a reliable cause of psychophysical properties to which our senses respond unaided. We have problems of amplification and translation that we do not face with Dobbin. To hear the song the proton sings or see the track it leaves requires knowing (or at least taking advantage of) physical laws that enable us to harness protons in contrived causal sequences that even- tuate in for-us-audible (or visible) events. Provision of the requisite laboratory equipment inserts additional links in the causal chain be- tween instantiations of nonobservable properties and tokenings of nonobservable concepts. Moreover, the correlations involved may not be self-evident; we may need to arrange carefully "controlled" conditions in order to detect them against a background of prevailing noise. Finally, although the basic picture of what is going on is, Fodor 69Psysem., p. 118, my stress. 7OPsysem., p. 118. 38 assures us, the same in the case of PROTON as in that of HORSE, the com— putation that delivers a token of a nonobservable concept will typically involve considerably more complexity than would be required in the case of observables. Fodor begins with a statement of the similarities. The picture is that there's, as it were, a computer be- tween the sensorium and yes—box, and that the tokening of certain psychophysical concepts eventuates in the computer running through certain calculations that in turn lead to tokenings of 'proton' (or of 'horse', or whatever) on the appropriate occasions. . . . [i.e.,] when its inputs are tokenings of psychophysical concepts for which prpions [or horses] are in fact causally responsible. Our natural way of describing the complex mental sequences involved in tokenings of nonobservable concepts like PROTON is to say that inference from the sounds and sights of our laboratory equipment is mediated by theory-—that is, by a set of (appropriately interrelated) beliefs. Fodor points out that while this is a reasonable enough view, characterizing the process in this way reintroduces the intentional idiom we are trying to eliminate. His suggestion is that the naturalizing semanticist can construe the fact that 5 holds a theory, T, as nothing more than the fact that S's stored formulae and operating procedures are such that, given a particular set of psychophysical concepts, the output will, in fact, be a tokening of a particular symbol of Mentalese. For the pur- poses at hand, T is treated as nothing more than a mechanism for transforming one set of shapes into another set of shapes, a causal process that we may, with a clear conscience, include in a naturalistic account of meaning. 71Psysem., p. 123. 39 [Mlaybe only the computational properties of our theo- ries matter to their role in fixing the meanings of mental representations; for those purposes our theories are the formalism we use to calculate what to put in the yes-box when. Fodor notes that even if it is not the whole story about believing or using a theory, Quine's view of theories as associative networks that allow the ”interanimation of sentences" in response to stimulation may provide a picture sufficient for the purposes of naturalizing represen- tation.73 So far, we have looked at Fodor's strategy for dealing with the trouble- some likelihood that not gll instances of any property will eventuate in tokenings of a particular symbol of Mentalese. Invoking counterfactuals, ("They ygglg have, if only. . .), he moves from contrary-to-fact conditions specifiable in purely nomological terms (in the case of psychophysical or sensory concepts), to those whose specification involves increasingly elaborate causal chains, at least some of whose links are constituted by merely reliable correlations. Even if this were to be satisfactory, a stubborn challenge still confronts the causal theorist: how is error possible? The problem of error for a causal theory takes the following form, Fodor notes. I see a cow which, stupidly, I misidentify. I take it, say, to be a horse. So taking it causes me to effect the 72Psysem., p. 123. 73Psysem., pp. 122-123. 4O tokening of a symbol; viz. I say 'horse'. . . . [Wle want it to be that my utterance of 'horse' means horse in virtue of the causal relation between (some) ’horse' tokenings and horses; . . . we don't want it that my utterance of 'horse’ means ggy in virtue of the causal relation between (some) 'horse' tokenings and cows. But if the causal relations are the same, and if causation makes representation74how can the semantic connections not be the same too? We need a principled basis for disregarding occasional cow-caused tokenings in determining what the symbol HORSE expresses. Otherwise, HORSE will express the property horse-or-cow, in which case a tokening of HORSE elicited by a cow (which is, after all, a perfectly good in- stance of horse-or-cow) would not count as an error. In ngchosemantics, Fodor talks about this as the disjunction problem. What we need, he notes, is some way of distinguishing between horse-caused tokenings of HORSE and tokenings caused by cows, a difference that will allow us to deal the latter out of the game when it comes to fixing the property expressed by HORSE. Attempts to make this distinction, Fodor tells us, have typically taken the form of adverbial qualification of the requisite causal relation be- tween property instantiations and symbol tokenings: e.g., only tokenings caused under optimal epistemic circumstances or under circumstances such that their tokening under these circumstances confers g selective advan- tage lg the system count toward determining the content of the symbol being tokened.75 Fodor presents arguments against invocation of either selective advantage or epistemic optimality. 7“Psysem., p. 107. 41 First, suppose we consider as optimal those circumstances under which tokening 'X' would be most useful to the organism—-those under which such tokenings would confer an adaptive advantage. The problem is that circumstances of this sort need not coincide with times when an instance of P is actually present. Perceptual defense or wishful thinking or even downright hallucination might, in some cases, turn out to be adaptive. Since we want our naturalized semantics to provide truth conditions for 'X', this first interpretation of optimality will not do. We might, instead, take circumstances that are optimal for belief fixation--those under which instantiations of P and 'X' tokenings are most likely to be correlated. The problem here is that characterizations of such epistemi- cally optimal circumstances will vary widely--stand close up or far off, take advantage of the rods in peripheral vision or the cones in foveal vision, and so on. Specifying optimal conditions for tokening a specific symbol of Mentalese is, it seems, going to depend on what it is a symbol pf. And this means importing the notion of content into what is supposed to be a reductive explication of that notion. So much for considerations of optimality. Fodor also rejects Dretske's proposal that we distinguish between what has caused tokenings of 'X' during a learning period and what causes it later on, with only the former determining what property (or properties) 'X' expresses. Such a distinction, Fodor tells us, is not principled enough to bear the weight of determining truth conditions. Worse yet, it fails to take into ac- 75I&R, p. 8. 42 count relevant counterfactuals: the events that ygglg have elicited 'X' tokenings ggg they been presented during the training period. If, as Dretske maintains, nomological relations make for information conveyed, and information conveyed makes for representation, such counterfactuals surely must have import for what 'X' represents. Fodor thinks he can accommodate the possibility of error "without relying on notions of optimality or teleology; and if we can we should. 76 . . [Tlhe less Pop-Darwinism the better, surely". His suggestion is that we invoke counterfactual prOperties to distinguish among causal circumstances. There is, he urges, an asymmetry between such properties that nicely corresponds to a felt distinction between tokenings that matter with regard to what a symbol expresses and those whose causes we wish to disregard in fixing content. Here is his appeal to our intuitions on this score. [Mlisidentifying a cow as a horse wouldn't have led me to say 'horse' except that there was independently g semantic relation between 'horse' tokeningg and horses: But for the fact that the word 'horse' expresses the prOperty of being a horse, it would not have been that word that taking a cow to be a horse would have caused me to utter. Whereas, by contrast, since 'horse' does mean horse, the fact that horses cause me to say 'horse' does not depend on there being a semantic—-or, inde d, any--connection between 'horse' tokenings and cows. That there is a causal relationship between cows and 'horse' tokenings depends on there being a causal relationship between horses and 'horse’ 76Psysem., p. 106. 77Psysem., pp. 107-8. 43 tokenings; that there is a causal relationship between horses and 'horse’ tokenings, however, does ggl depend on there being a causal relationship between cows and 'horse' tokenings. Cow—caused 'horse' tokenings are, thus, in Fodor's term, 'wild'. In terms of a possible worlds analysis, HORSE expresses horse (and not horse-or—cow) because: (1) Horses cause 'horse's in this world. (2) At any nomologically accessible world at which horses do ggl cause 'horse' tokenings, neither do cows. There is no nomically accessible world at which horses do not cause ’horse' tokenings but cows do. (3) There is some nomologically accessible world in which cows do not7§ause 'horse' tokenings, but horses, nevertheless, gg. Fodor assures us that it does not matter whether we are, as here, deal- ing with tokens of English words or are considering tokens of Mentalese symbols. He thinks that misapplication of a symbol--misrepresentation-- can be analyzed the same way in either case. We shall return to this claim later. For the time being, the important thing about Fodor's account of wild tokenings is that it is designed to accommodate the pos- sibility of error. Let us take a moment to see how Fodor’s more general views on truth and falsity motivate such an account, and then go on to his second thoughts about the adequacy of this particular version. 78Psysem., p. 109. My paraphrase. ti 44 What it takes to give an account of error obviously depends, inter alia, upon our theoretical view of truth. In considering Fodor's causal semantics we shall want, therefore, to keep in mind that he is a cor- respondence theorist, construing truth along the lines of ita est sicut siggificat--that is, as requiring things to be as they are represented as being.79 Error, he tells us in a recent paper, is, first and foremost, a matter of misrepresentation. Applying 'platypus' to a duck is wrong because in so doing we represent the duck "as having the property which ’platypus' expresses. . . the property of being a platypus. Which, of course,. . .[it] doesn't; only platypuses do."80 On the other hand, "You get a truth when you say gl a platypus that it lg one."81 Naturalizing this view of truth requires Fodor to answer (in appropriately nonintentional language) two rather different questions. First, what makes it be one entity (rather than another or perhaps nothing at all) that lg represented? Why, for example, is it my duck Gladwin that a particular tokening of 'platypus' represents as a platypus? When we use a referring term, to what (if anything) are we referring? This issue has tended to take center stage in discussions of 23 791 owe the nifty Latin tag to Prior, "Correspondence Theory of Truth," infhPaul Edward, ed. The Encyclopgglg pl Philosophy, (New York: Macmillan Publishing Co., Inc. 8 The Free Press, 1967). Fodor endorses Tarski's approach and Tarski tells us he is trying to capture Aristotle's notion. According to Prior, Aristotle's notion is reflected in this phrase. 80, by transitivity. . . 80"Information and Representation", paper delivered at the conference "Information, Language and Cognition" 2/20/88, Simon Fraser University, Vancouver, BC. (Hereafter, I&R.) P. 12. My stress. Some of these obser— vations were also made in a talk at Central Michigan University in October of 1987. I have substituted a (somewhat more plausible) duck for the cow in Fodor's example. 45 causal theories of names but is treated rather off-handedly in ngchosemantics. Second, once we have pinned down what is being represented (referred to), we may ask, what makes it the case that it is being represented as having one rather than another property? What property is Gladwin (rightly or wrongly) represented as having? What makes the symbol of which 'platypus' is a token express the property platypus rather than the property platypus—or-dimly-illuminated—duck? When is 'water' true of H20 and when is it true of XYZ (or of HZO pg XYZ)? In Psychosemantics, Fodor focusses almost exclusively on this second issue. Fodor's theory should, then, have two parts—-one that associates symbols with types pl events (or properties), and another that associates symbol tokenings with particular events (or property instantiations). When the particular event with which a symbol-tokening is associated instantiates the type of event (property) with which the symbol is associated, we have truth; when it does not, we have error.82 To the extent to which he makes the distinction between these two questions, Fodor gives, in ngchosemantics, similar answers to both, invoking in each case a small set of causal relations among symbol tokenings and events. 81I&R 4 82 r P0 ' \5 For this way of looking at things, I am indebted to Susan Haack's dis- cussion of J.L. Austin's 1950 correspondence theory. Philosophy pl Logics, (Cambridge: Cambridge University Press, 1978), pp. 93 ff. In the case of natural languages, Austin relies on conventions to establish the required associations--"descriptive" conventions to achieve the first, "demonstrative" conventions to achieve the second. Although this cannot be the correct answer in the case of Mentalese, it seems useful to pose parallel questions. 46 Difficulties with the account of error advanced in Psychosemantics have, however, led Fodor to amend it. In particular, his account of reference --of what it is that associates symbol-tokenings with events-—has undergone notable change. Fodor draws attention to a problem with his original account: if error consists in a tokening of A being occasioned by something other than an instance of 5, then, not only do we get to count as errors tokenings of 'platypus' elicited by dimly illuminated ducks, which seems all to the well and good, but we are also constrained to count as errors 'platypus’-tokenings due to memories of Melbourne, which is not at all good since there is nothing apparently ygpgg with coming to think of a platypus as a result of thinking about Melbourne. Reflection on this difficulty has led Fodor to some new thoughts about the theory developed in Chapter 4 of Psychosemantics, although by no means to its wholesale rejection. What has survived revision is Fodor's account in terms of asymmetrical causal dependencies of how symbols are associated with properties. A expresses g, Fodor tells us, iff tokens of A that are not A-occasioned are causally dependent on those that are, but not vice-versa. Although 'platypus' can be elicited by a dimly illu— minated duck or by a memory of Melbourne, 'platypus' expresses platypus and not platypus—or-dimlyeilluminated-duck or platypus-or—thought—of- Melbourne by reason of causal asymmetry: were 'platypus'—tokenings not reliably elicited by platypuses, they would not be elicited by ducks (nor by reminiscences of Melbourne), whereas the reverse is not the case. Considerations of asymmetry of causal dependence are still relied upon to bar awkward disjunctive contents by uncoupling the property 47 expressed by a symbol from at least some of the instances causally responsible for its tokening. What has not survived Fodor's second thoughts is his confidence that the disjunction problem was all that stood in the way of a naturalistic ac- count of the truth and falsity of contentful states. "The disjunction problem,” he now tells us, "has nothing in particular to do with error."83 What we learn from a consideration of 'platypus'-tokenings elicited by memories of Melbourne is that the causal account of error proposed in Psychosemantics yields conspicuously counterintuitive results in an important set of cases--in what Fodor now calls representational uses of symbols, as opposed to their labelling function. Let us look at this distinction. There are, Fodor tells us, two ways in which a tokening of 'platypus' can be used to say or think something true. [Olne way to generate a true utterance (thought) is to apply a token of the symbol to something that is in its extension. You ggt a truth when you say pl a platypus that it lg one. 83I&R, p. 8. More precisely, ”[I]t has to do with error only insofar as the misapplications of symbols share with lots of other kinds of symbol tokenings the property of not being caused by things in the symbol's ex- tension." Pp. 8-9. How there could be symbol tokenings whose causes did not instantiate the property expressed by the symbol is, of course, the disjunction problem. Fodor's original problem was how, on a causal theory, we could ever be entitled to deliver a verdict of error; his current problem is how, on a causal account, we can avoid giving this verdict across an embarassingly broad range of cases. 8418B, p. 4, stress removed. 48 Fodor proposes to "call the uses of symbols that generate such truths ’labelling’ uses." There is, however, a second way in which symbol tokenings can feature in true thoughts or utterances. Somebody says: ’The platypus has webbed feet’. . . . [Wlhen I use ’platypus’ to say that platypai are web- footed, I’m not applying the term to anything; rather, I’magsing it to stand lpg the things that it applies to. Other cases where "the function of a symbol appears to be representation rather than labelling” include occurrences in existentials (’there are platypuses’) . . hypotheticals (’if there are platypuses, there may be anything’. . .) [and] occurrences embedded to verbs of propositional attitudes (’No one can really believe that there are platypuses’). . . . [Iln none of these cases is anything being called a platypus. Yet all these tokens of ’platypus’ are symbols; they are all paradigps of things that have bona fide intensional properties. Symbol tokenings put to representational use in true thoughts or utter- ances are not necessarily (or even typically) occasioned by instances of the property they express. This is serious, Fodor notes, because "The typical use of symbols lg thinking is representation, not labelling."87 No doubt, I do sometimes think ’Hello, platypus here’; thoughts of this kind are, I suppose, the usual product of perceptual processes. . . . [and] must sometimes play a role in reasoning. But it is as silly to think of thinking as primarily consisting in labelling as it would be to think of talking as consisting primarily in cries of ’GavagaiI’ Much. . . of what goes on in thinking is a movement from representation to represen- tation; we put to ourselves ways that thgsworld might be, and then we figure out what follows. 8518B, p. 4. My stress. 86Ian, p. 4. 87I&R, p. 6 49 Reliance on causal history yields the verdict that symbol tokenings used representationally are, by and large, cases of error. It thus appears that considerations of causal history, no matter how ingeniously embellished by reflections on asymmetry, cannot provide an adequate naturalistic reduction of truth and falsity. Fodor’s response to this perplexity is to offer a treatment of error that is substantially different from that advanced in Psychosemantics, one that, according to him, (1) does not yield anomolous verdicts of error in the case of represen- tational uses of symbols (because it yields no verdicts at all in such cases); (2) provides an intuitively acceptable and adequately naturalistic ac- count of error in the case of mislabellings (where it does deliver verdicts), and (3) appears capable of extension to the case of representational uses. Fodor argues for (1) at some length, treats (2) sketchily and gives us a bare IOU with regard to (3). What, then, does Fodor’s new (and more circumscribed) account of error look like? Error, in the case of labelling, at least, is now seen as the 88I&R, p. 6. 50 misapplication of a symbol--that is, as the application of a symbol to something that does not instantiate the property it expresses. Most importantly, whereas ngchosemantics identified 'P’s referring lp 9 pg pg occasion with Q’s causing p ’P’—tokening pg that occasion, Fodor’s new story explicitly distinguishes between the two. In Psychosemantics, it will be recalled, reduction of denotation to causation-~at least in paradigmatic cases—-was treated as unproblematic. Let’s start with the most rudimentary sort of example: the case where a predicative expression (’horse’, as it might be) is said of or thought of, an object of predi- cation (a horse, as it might be). Let the Crude Causal Theory of Content be the following: In such cases [1] the symbol tokeningg denote their causes, and [2] the symbol types express the properéy whose instantiations reliably cause their tokenings. On the new account, however, applying a symbol lp Q is explicitly dis- tinguished from being caused to token it py Q. |A|pplying ’platypus’ to a platypus doesn’t uncomplicat- edly reduce to having the platypus in question cause you to token ’platypus’. Think of the chap whose habit it is to use ’platypusI’ as an ejaculation expressing extreme surprise. . . . In this chap’s case, a token of ’platypus!’ that happens to be platypus-occasioned would nevertheless not count as an application of the term to the animal, and would not count as the saying of some- thing true. . . .[Alt best, only the right kind of platypus-cagaed tokens counts as applications of ’platypus’. Once application is distinguished from causation, tokenings of ’platypus’ caused by non-platypuses need no longer necessarily count as 89Psysem. p. 99, stress added. Fodor still endorses a suitably elaborated version of [2] but has jettisoned [1]. 90153, p. 5. 51 glgapplications, as the saying or thinking of something false. In parti- cular, insofar as symbols used representationally are by stipulation not being applied to anything, they are, p fortiori, not being applied to their causes. In such cases, discrepancy between the property expressed by a symbol and that instantiated by a cause of its tokening does not necessarily constitute misapplication. We are no longer stuck with the uncomfortable verdicts noted above: being caused by a thought of Melbourne does not in and of itself render a ’platypus’—tokening erroneous-—it would only do so were it the case that being caused by a thought of Melbourne constituted the application of ’platypus’ to a thought of Melbourne. Fodor thus manages to avoid the counterintuititive consequences of his earlier theory of error in the case of representational uses of symbols. In the next chapter, we shall inquire after the success of Fodor’s proposal as thus revised. Chapter Three A Critical Look at Fodor’s Semantics for LOT In an effort to justify realism about everyday intentional explanation, Fodor has proposed a causal semantics for at least some uses of the language of thought. In this chapter, I shall argue that even in its revised version, his theory of error has difficulties. The notion of a labelling context-~one in which the cause of a symbol’s tokening is relevant to evaluation of the formula in which it occurs—-is far from clear; there may or may not be such contexts. The explication of what it is to apply a symbol to something in such a context seems, moreover, to fall short of Fodor’s own criteria for naturalization. I shall also raise questions about Fodor's doctrine of asymmetric causal dependence, arguing first, that its truth is debatable, and then that its counterfactual claims may not even make sense. Let us begin with Fodor’s new view of error. If we combine Fodor’s computational theory of mind and the causal semantics introduced in "Meaning and the World Order," we get the following picture. For S to have the occurrent belief that a is F is for S to stand in a computa- tional relation (corresponding in its causal role to belief) to an internally structured state that is (under interpretation) a tokening of a sentence of the language of thought, ’Fa’. S’s belief that a is F is true iff ’Fa’ is true; ’Fa' is true iff whatever ’a’ denotes (refers to, 52 53 picks out) has the property expressed by ’F’. Most of Fodor’s attention has been dedicated to providing an account of what it is for a symbol of the language of thought to express one property rather than another--to express 2, say, rather than F;pg;§. By contrast, he has treated rather offhandedly the question of how reference is determined-—by virtue of what it is one thing rather than another whose instantiation of F makes ’Fa’ true. He originally assumed that when a symbol is tokened, the property it expresses is being predicated of whatever it is that caused the tokening. Thus, if it is a transaction with a horse that leads me to token ’horse’, it is of that horse that I have predicated horsehood and believe (truly) that it is a horse. If, on the other hand, it was a cow that stood in the relevant causal relation to my tokening ’horse’, it is of that cow that I predicated horsehood and believe (falsely) that it is a horse. Error, on this account, is a matter of being caused to token a symbol of the language of thought by the wrong sort of thing, i.e., by something that does not instantiate the property expressed by the symbol. One thing that is wrong with this story is, of course, the unargued identification of ’S predicates F of a’ with ’S is caused to token F by a'. As Fodor points out, a can be the cause of S’s tokening F without S’s thereby predicating F of anything: a hammer hitting my thumb can cause an emphatic but non-predicative tokening of MERDE. Moreover, a tokening of F may be used to predicate F of something quite other than its local cause. A tokening of ’Bossie is a cow’ may have as its local 54 causes a milk bottle, a dish of cottage cheese and a fond recollection of Bossie, without is-a—cow being predicated of any of these; is-a-cow is being predicated of Bossie, who is not necessarily in a local causal relation to the tokening of the symbol that expresses this property. Being the cause of a tokening of a symbol is clearly neither necessary nor sufficient to make something a subject of predication of the pro- perty expressed by the symbol. Considerations such as these have led Fodor to revise his account of denotation and predication. He now pro- poses, first, that it is only in special contexts that the local causal history of a constituent, ’F’, of a formula, ’Fa’, will be relevant to its evaluation--that causation determines denotation only in a re- stricted set of cases. Second, predication is now explicated in terms of the causal powers of a tokening of a formula to which S stands in an appropriate computational relation, rather than in terms of its local causal history. On the revised view, my predicating F of a is not simply a matter of a’s having been a local cause of my tokening F, but, rather, requires that my standing in the computational relation corresponding to belief to a tokening of ’Fa’ cause a family of relevant dispositions on my part to inference and action--dispositions to process tokens of ’F’ and ’a’ in certain ways, and to behave in certain ways (F-wise, one might say) toward a. Fodor calls the special contexts in which causation is relevant to reference "labelling contexts", and talks about predica— tion in such contexts as the "application" of a symbol to an object. 55 It is clear that Fodor has a stake in there being at least some cases where what we are talking or thinking about-—what, if anything, we are predicating a property pl, and, therefore, what it is whose properties render our predication true or false-—can be explicated in terms of purely causal relations. Even if it is not possible to do this in most cases of symbol use, not even in most interesting cases, the existence of gpgg such cases is important. (Fodor tells us that a theory adequate only to labelling uses of symbols cannot hope to provide a naturalistic account of "mental lives. . . richer than the frog’s".1 "[I]t is as silly to think of thinking as primarily consisting in labelling as it would be to think of talking as consisting primarily of cries of ’Gavagail’".2 ) If reference always has some descriptive component, naturalistic reduction will be impossible and Fodor’s philosophical adversary will have carried the day.3 Labelling contexts are thus theoretically important to Fodor’s overall project because they purportedly provide a domain for a pure causal theory of reference. In their case at least (unlike that of so-called "representational" use of 1I&R, p. 7. 2I&R, p. 6. 3Moreover, insofar as Fodor is a correspondence theorist, an ineliminable descriptive component in reference would be unsatisfactory because, as Devitt and Sterelny note, to the extent that we rely on description to determine reference, we make the reference of some words dependent on that of others, and thus leave reference internal to the lan- guage. We need an explanation of the external relation that the whole system of words bears to the world (L&R, p. 60). Devitt and Sterelny are, of course, themselves correspondence theorists. 56 symbols), truth and local causal history go hand in hand. "True labellings carry information about their causes," Fodor tells us.“ One condition upon the correct application of the term ’platypus’-—i.e., upon its use to express a truth in a labelling context—-is that "tokens of the symbol [’platypus’] are under the causal control of instances of the property platypus".5 The truth conditions of sentences of natural language are relevant to Fodor’s case insofar as they suggest by analogy the existence of labelling contexts in the language of thought. Fodor therefore begins by encouraging the intuition that there are, indeed, natural language contexts that meet his criteria for the labelling use of symbols. He is, however, not explicit about what such contexts might be, relying, rather, on examples. Fodor’s "paradigmatic" case of labelling is "platypus-spotting": "A platypus galumphs by and one says ’Platypus!’ (where. . . the utterance has the force: ’There goes a platypusI’)".6 The example is in English, but Fodor assures us that "[Flor most present purposes, we might as well assume that English lg Mentalese".7 This suggests that a labelling con— text is a demonstrative predication, that ’There goes a platypus!’ is to be understood as ’That’s a platypus!’ If this is what Fodor has in mind, natural language does not lend unambiguous support to his claims. First, I&R, p. 10. 18R, p. 10. I&R, pp. 3—4. 18R, p. 14, n. 57 even if we can conjure up situations in which a tokening of (English) ’There goes a platypus!’ is true only if it was caused by a platypus in the vicinity, it is not the case that the truth of English demonstrative predications requires, lg general, that they be caused by an instance of the property predicated. Second, if we look at the subset of natural language demonstrative predications in which the required coincidence holds, we find that it involves the application of conventional rules whose functional analogues in Mentalese are, to say the least, not instantly apparent. It is often enough the case that the truth of a tokening of a natural language demonstrative predication--’That’s a cat’, ’She’s six feet tall’, ’There’s a mouse’--seems to require that its local causal history include an instance of the property predicated--cathood, six foot tall womanhood, mousehood. It is, however, important to notice that this oc— curs because (1) the truth of a predication requires that the subject of predication instantiate the property being predicated of it; (2) the subject of a demonstrative predication is the denotation of the demon— strative; and, (3) pragmatic conventions may, in certain cases, dictate the selection of a local cause of an utterance as the relevant denota- tum. Thus, in the absence of other indication, a competent English speaker who hears ’There goes a platypus!’ or ’That’s a platypus!’ will take the item of which platypus-hood is taken to have been predicated (and which must, therefore, instantiate platypus—hood for the utterance 58 to be true) to be whatever item in the vicinity appears to have elicited the utterance. Truth conditions involve denotata and the denotata of demonstrative predications frequently turn out to be their causes. Contrary to what Fodor’s examples might suggest, however, they are not necessarily so. It is, at any rate, by no means uncontroversial that the reference of demonstrative ’that’ is determined by its local causal history. Kaplan’s well-known example suggests that the denotation of a demonstrative may, rather, be determined by an accompanying ostension. Kaplan tells us that a speaker who, not realizing that a picture of Rudolph Carnap has been replaced by one of Spiro Agnew, points behind him and says, ’That is a picture of the greatest philosopher of the twentieth century’, has, in so doing, "said of a picture of Spiro Agnew that it pictures one of the greatest philosophers of the twentieth century." Speech and gesture to- gether, he tells us, "suggest no other natural interpretation to the linguistically competent. . . observer".8 Where indication is explicit and unambiguous, convention appears to dictate that the item of which the property is predicated will be the item indicated by the speaker. When explicit indication is ambiguous--when, for example, the line of sight determined by a fingerpoint intersects a window, a gauze curtain, a virtual image or a milling crowd—-additional pragmatic rules will need to be invoked to determine what is being indicated (e.g., ’Given two possible targets of a point, exercise charity and choose the one that 8D. Kaplan, "Dthat", mss., p. 22. 59 yields a true or at least a plausible proposition’). The pragmatic rule, ’Identify the denotation of a demonstrative predication by identifying a local cause of the utterance’ is more reasonably seen as a default option, effective only in the absence of unambiguous indication: when there is no explicit indication or attempts to disambiguate such indi~ cation fail, a competent listener may try to figure out what in the vicinity is most likely to have attracted the speaker’s attention and, thus, to have caused the utterance.9 The likelihood that demonstrative reference is governed by such complex conventions suggests the implaus- ibility of identifying causes with denotata in demonstrative contexts in natural language.10 Understood as a demonstrative predication, then, English ’Platypus!’ (a’That’s a platypusI’) does not necessarily have the truth conditions 91f, however, this turns out to be something of which the listener has reason to believe the speaker has reason to believe the audience is un- aware, it may, in turn, be rejected. Discussing the Kaplan example, Wettstein notes that one account of the semantics of ’that’ is [S] "Uses of ’that’ refer to the individual that is indicated by the cues avail— able to the audience" (cited by John Perry in "Cognitive Significance and the New Theory of Reference", Npgg, 22 (1988), p. 11). One might argue that the rule that tends to pick out causes as denotata is best stated as ’Identify as the reference of ’that’ whatever in the vicinity the speaker has reason to believe the hearer will believe is the cause of the utterance’. If either of these more Gricean conjectures is correct, causality is never an essential condition on demonstrative reference but beliefs of speakers and hearers about each other may be. This is, in fact, my own hunch. 101 am grateful to R. Hall for his refusal to accept overquick and over- simplified accounts of demonstrative reference. It was also helpful to be reminded that it does not take a Completely Correct Theory of Demonstrative Reference to establish that an account under consideration is seriously incorrect. 60 Fodor associates with a labelling use of symbols if, indeed, it ever has them. The subset of cases where the truth of a demonstrative predication requires a coincidence of the property expressed by the symbol tokened and the property causally responsible for its tokening is governed by complex conventions, conventions that make essential reference to, e.g., hearer’s beliefs about probable causes of utterances and speaker’s beliefs about those beliefs. Mentalese cannot take its reference by convention. A naturalized semantics cannot make ineliminable use of notions of belief and belief about belief. It is hard to see how nomo- logical or causal relations could reconstruct the complex pragmatic rules used in natural language. Despite the initial attractiveness of the proposal, it seems unlikely that Fodor’s labelling contexts can be identified with the Mentalese equivalents of natural language demonstrative predications. What else could they possibly be? Fodor tells us that information—based semantic theories are right about true labellings and also that (at best) information-based semantic theories give us "a naturalistic theory of representation in perception".11 Are Fodor’s labelling contexts perceptual judgments? Perhaps. Fodor offers ’Hello, platypus here’ as an example of labelling and then says of it, "[Tlhoughts of this kind are, I suppose, the usual product of perceptual processes".12 Representational thought, by con— trast, "extends the reach of the mental life beyond what is locally 11I&R, p. 6. 1215R, p. 6. 61 given to perception; unlike labelling, we can do it with our eyes 13 shut". Perhaps labelling uses of symbols are cases where symbols are used to express perceptual judgments. In particular, we might wonder whether what Fodor has in mind might not be what have elsewhere been termed de re perceptual beliefs. On one account, at least, de re belief contents seem very much like the tokenings of Mentalese symbols to which Fodor wants to ascribe truth value: [Nlot only are their contents not complete representa- tions (in the same sense in which concepts, as expressed by predicates, are not complete representations), they represent their objects as having certain properties without containing any elfpent that identifies the object being represented. Bach takes perceptual belief to be the "fundamental case" of de re belief and suggests that what determines reference in such cases must be 15 "causal relations and causal chains". Perhaps ’Platypus!’ expresses a de re perceptual belief about its cause.16 13I&R, p. 6. 14K. Bach, "De re Belief and Methodological Solipsism," in, A. Woodfield, ed., Thought and Object: Essays pg Intentionality, Oxford: Clarendon Press, 1982, p. 138. 15 K. Bach, p. 139. 16Bach notes that Burge "likes to speak about pp ES belief contents as being contextually ’applied’ to an object" (p. 138). This, of course, is how Fodor talks about the predication in a labelling context of the property expressed by a symbol. 62 Some not entirely implausible premises about perceptual judgments seem, indeed, to deliver the desired result, namely, that a tokening of a per- ceptual judgment, ’Fa’, is true iff its cause instantiates F. (From here on, I shall continue to mean g tokening pl a perceptual judgment but drop out the qualification for brevity.) Let us begin with the premise that a perceptual judgment, ’Fa’, predicates l of a, where a is a per- ceptual object (roughly, what, if anything, in the surrounding world S sees, hears, etc.) Perceptual judgments, on this view, are about-~have as their denotations--objects of perception. This means that determining the denotation of a perceptual judgment is a matter of fixing a percep- tual object. How? Well, if perceptual judgments are caused by perceptual experiences, the perceptual object we are after is the object (if any) of the perceptual experience responsible for the perceptual judgment with which we are concerned. And on one persuasive account, that percep- tual object is the object in the world that causes, via an appropriate perceptual process, the perceptual experience with which we are concerned-—the perceptual experience that caused the tokening of ’Fa’. A perceptual judgment, ’Fa’, thus predicates l of the cause of the visual experience that caused its tokening. If we assume that the cause of a cause of X is a cause of X, then what a perceptual judgment ’Fa’ predi- cates l of stands in a local causal relation to that perceptual judgment: it predicates l of a (local, environmental) cause of its (internal, psychological) cause. Like any other predication, a percep— tual judgment, ’Fa’, is true iff the object of which it predicates l instantiates E. Which gives us the desired conclusion that a perceptual 63 judgment, ’Fa’ is true iff an instantiation of E stands in a local causal relation to that judgment. Suppose, then, that Fodor’s labelling contexts are construed along the lines of perceptual judgments: we read ’Platypus!’ now as [P] ’That (which I see) is a platypusI’ Such a formula would ascribe platypus-hood to an object of perception determined jointly by S’s visual experience and the causal context of that experience.17 The object of S’s perception would be whatever in S’s environment caused (in an appro- priate way--by means of reflected light and a visual system, say,) S’s visual experience. The object thus picked out would be the object of which the tokening of [P] would predicate platypus-hood. On this view, the truth of [P] would depend, just as Fodor would have it, on coincidence between the cause of [P]--a platypus—~and the property [P] expresses (ascribes, predicates)-- omething’s being p platypus. Home free? Not quite. This account assumes that the content of a percep— tual experience and its causal context together determine an object of perception and, thus, can, together, determine the denotation of a perceptual judgment finely enough to fix the truth value of the thought (i.e., the tokening of a sentence of Mentalese) that expresses that .judgment. One obvious problem with this view is that it assumes a * 17The account of perceptual objects on which this discussion leans (heavily) is presented in R. Hall, "A View About Visual Perception", lnss., and Hall, Richard,"Seeing Perfectly Dark Things and the Causal Conditions of Seeing", Theoria, 45 (1979). 64 perceptual experience; this might seem difficult to accommodate within a 18 theory that eschews qualia. But perhaps not. Talking about tokenings of 'red’ in the ’occurs-to—me’ box, Fodor says, "In the psychology I’m assuming, an organism need not be conscious of the thoughts that occur 19 to it". There is certainly no reason why such "occurrings" could not cause the tokenings of perceptual judgments: Fodor takes perceptual judgments to be typical outcomes of perceptual processes, of which, perhaps, being occurred to might be one. More serious is the question of whether perceptual experience and causal context really do determine perceptual objects (and thus the denotata of perceptual judgments) with the requisite degree of fineness. Martin Davies has argued that they do not, that competing claims relevant to semantic evaluation remain unresolved. To the extent that . . .[a perceptual experience], and the [causal] context of p’s thought at l, jointly deter- mine a man p to be the object of g’s thought they also jointly determine as the object of that thought a temporal slice of a man, a front surface of a man, a mereological union of parts of a man, an aggregate of molecules constituting a man, and doubtless various other objects of more or less recherche sorts, objects which are not identical with the man p. And in general these differences matter for truth evaluation. Perhaps Q will still exist tomorrow although the temporal slice of p will not exist tomorrow. . . . Perhaps Q will one day have a finger cease to exist and yet go on existing himself although the mereological union of p’s parts will n26 endure through that finger’s ceasing to exist. 18"If, as the inversion arguments. . . suggest, qualitative contents are in some sense ineffable, it would hardly seem reasonable to require a philosophy of qualia to eff them" (Intro., Rep., p. 18). 19Psysem., p. 165, n. 65 Davies concludes that even in conjunction with context, "bare demon- stratives" like ’that’ cannot determine the objects of perceptual demonstrative thought: "[I]n such thoughts the employment of [sortal] 21 concepts in individuative roles is ineliminable". If this is so, then the reference of terms of Mentalese has an irreducible descriptive component. Devitt and Sterelny make a point very similar to Davies’ in discussing their causal theory of the reference of natural language terms-- specifically, names. They admit what they call a "knowledge" requirement-—what we might want to call a descriptive component--to the perceptual—causal "grounding" of names on the basis of the following considerations. The name [’Nana’] was grounded in Nana in virtue of per- ceptual contact with her. But that contact is not with pll of Nana, either temporally or spatially. Temporally, the contact in any one grounding is only with her for a brief period of her life, with a "time-slice" of Nana. On the strength of such contacts, Nana, the sum of many time-slices sighted and unsighted, is the referent of ’Nana’. Spatially, the contact is only with an unde— tached part of her, perhaps a relatively small part of her face (she may be peering around a corner). In virtue of what was the grounding in the whole Nana not in a time-slice or undetached part of her? . . . [Tlhere must be something about the mental state of the grounder that makes it the case that the name is grounded in the cause of the perceptual experience gpa whole object. . . . It seems that the grounder must, at some level, "think of" the cause of his experience under some general cate- gorial term like ’animal’ or ’material object’. It is 20Martin Davies, "Individuation and the Semantics of Demonstratives", Journal pl Philosophical Logic 11 (1982), pp. 291-92. Hereafter, I&SD. 21 I&SD, p. 297. 66 because he does so that the grounding lg in Nana and not in a temporal and spatial part of her. Devitt and Sterelny admit that "The move away from a pure-causal theory 23 of names has a price," but are consoled by what they take to be provi- sion of a pure—causal theory for deictic uses of ’this’, ’that’ and 24 ’it’. "The simple demonstratives, ’this’ and ’that’, and the pronoun, ’it’, have near enough no descriptive content and seem amenable to a pure-causal theory (in their deictic use)".25 Devitt and Sterelny are, however, talking here about natural language, not the language of thought. And as we have seen, this means helping ourselves to consider— ations whose counterparts in Mentalese are obscure. Davies draws attention to the background against which deixis functions in natural language. Speakers of English do sometimes succeed in directing the attention of an audience to a particular object of a particular sort by an utterance of the bare demonstra- tive ’that’ accompanied, perhaps, by an act of pointing. It is not difficult to see how this might be achieved. We have a greater (human) interest in men than in tem- poral slices of men, in tomatoes than in front surfaces of tomatoes, in cars than in mereological unions of car parts, and in cats than in aggregates of molecules con- stituting cats. Once an audience’s attention has been directed towards the right feature the audience will naturally tend to fix his attention upon an object of a sort in which he has a greater interest rather than an object of a sort in which he has a lesser interest. ggp ll lg common knowledge that all this lg so. Such common knowledge pg this, lpgether with common knowledgp about, for example, the preceding discourse, gpy remove the 22L&R, pp. 64-65. 23L&R, p. 65. 24L&R, p. 84. 25L&R, p. 83. 67 need for use pl p complgx demonstrative [e.g., ’that man’l lg pg utterance. The common knowledge (and common knowledge pl common knowledge) that allows the use of ’bare’ demonstratives in natural language is just the domain from which a naturalizing theorist is barred on pain of circularity. Mentalese may be a lot like English in some ways but it also differs from it in ways that matter. If there are labelling contexts of the sort Fodor wishes to display, my hunch is that they are more likely to be perceptual judgments than anything else. It is, however, unclear that this is what he has in mind. If it is, he owes us more argument than he offers. To use a phrase Fodor likes, I think the verdict with regard to the existence of labelling contexts in natural language is ’Not proven’. Even if we were persuaded that there were cases in which the truth con- ditions of tokenings of Mentalese sentences essentially involved their immediate causes, we would still need to look at Fodor’s notion of the application of a symbol. That tokens of ’platypus’ are "under the causal control of instances of the property platypus," is, he tells us, only "part of the story" about [appropriately] applying ’platypus’ to platy— puses in labelling contexts"--i.e., about predicating platypus-hood of a 26I&SD, p. 297, My stress. 68 platypus (truly).27 Fodor offers an "old fashioned suggestion" with regard to what else is required for full-fledged predication. [A] Applying a symbol to Bossie [also] involves things like having a disposition to reason and to act in cer- tain ways, both in respect of the symbol and in respect of Bossie. So, for example, if you apply ’platypus’ to Bossie. . . you are prepared to act towards Bossie in whatever gays you are disposed to act towards platypai. Fodor emphasizes that the relevant dispositions include those toward inference as well as action: if you believe ’platypai lay eggs’, applying ’platypus’ to Bossie requires being prepared to accept ’Bossie lays eggs’ and acting toward Bossie "in whatever way you are prepared to act toward egg-layers." 29 Fodor is optimistic about naturalizing [A]. There are, after all, ideas floating around for func- tionalist reconstructions of . . .notions [like accepting, being prepared lp reason/act lg such and such p ypy, etc.], and I’mBBrepared to view these ideas as not altogether crazy. In particular, Fodor is hopeful of their reduction to talk of causal relations among symbol—tokenings and behaviors. 27I&R, 28 291&R, 3018B, . 10. I&R, p. 12. . 12. pp. 12-13. My italics. 69 One way to apply [the symbol] ’is a platypus’ to Bossie is to have a token of. . . [’Bossie is a platypus’] in your "accept-box". The accept-box is defined by reference to the causal roles of the symbols it contains;. . . a symbol is said to be in the accept box iff certain of lis tokens have certain causes and effects. . . . Is Fodor’s optimism justified? I think not. Let us, for convenience, use ’be disposed to treat 0 P-wise’ to abbreviate having the relevant clus- ter of dispositions toward action and reasoning that Fodor mentions. Then, S applies ’P’ to 0 iff (1) ’P’ is tokened in a labelling context-—that is, one where ’P’ can be used to express a truth only if the proximal cause of P’s being tokened is an instance of the property expressed by ’P’-—and Q is the proximal cause of ’P’; and (2) S is disposed to treat 0 P—wise. If I am not disposed to treat 0 P-wise--if I am not disposed to act toward 0 as I am disposed to act toward other instances of the property expressed by ’P’-—I am not applying ’P’ to 0. Suppose, now, that up to time l I have been unable correctly to identify a single one of the CIA agents on campus. I have, in fact, been enthusiastically inviting each and every one of them to committee meetings. Because they have been appreciative and available, I am, at l, disposed to continue inviting them to meetings. At l, I take a hard look at my officemate and token, ’Good grief--he is a campus CIA agent!’ We would, I think, like to say that in so doing I am applying ’is a campus CIA agent’ to my officemate, 311&R, p. 13. 70 that the application is appropriate if he instantiates the property ex- pressed by ’campus CIA agent’ and erroneous if he does not. According to (2), however--at least as we have been reading it——I cannot be applying ’is a campus CIA agent’ to my officemate unless I am disposed to act toward him the way I am disposed to act toward campus CIA agents. But the way I am disposed to act toward all of the other campus CIA agents is to invite them to meetings and ply them with coffee, which is pre- cisely what I am gpl now disposed to do with respect to my officemate. This suggests that (2) is not the straightforward extensional requirement that I be disposed to treat 0 as I am disposed to treat P-instances, but, rather, the thoroughly intensional requirement that I be disposed to treat 0 as I am disposed to treat whatever I lplg lp pp or represent lp gygpll pg a P-instance. But (2) is part of an explica- tion of what it is to apply ’P’ to Q, i.e., of what it is to represent 0 as P, or to take 0 to be an instance of P, and thus these notions are not available for use in (2) without circularity. If the relevant effects of having ’Bossie is a platypus’ in my accept-box include being caused to treat Bossie the way I am disposed to treat whatever I take to be a platypus, we have an unreduced intensional locution on our hands despite Fodor’s attempt to achieve naturalization by way of a functionalist reconstruction. There are other difficulties as well with a dispositional account of the application of a symbol, some of a very familiar sort. Suppose, for example, that we are concerned with the application of ’is red’ to some 71 Q. We are, on Fodor’s account, required to be disposed to act toward Q as we are disposed to act toward other things to which we apply ’red’. But how, precisely, lg that? Is there any shared disposition to respond toward everything we take to be red? Or that we take to be an instance of fire or water or the letter Z? Another sort of problem is presented by the student who, misunder- standing what has been going on, looks at the number displayed on the blackboard and tokens, ’WowI So lgpllg the largest prime number!’ We would like to say that this is an application of ’largest prime number’ to the number tokened on the blackboard, but for this to be the case, the student would have to be disposed to treat it the way she is dis- posed to treat other tokenings of the largest prime number. Not only aren’t there any such instantiations but there couldn’t be any. ’Unicorn’ and ’witch’ might yield to analysis in terms of counter- factuals--’S applies "unicorn" to 0 only if at the nearest world at which there are unicorns, S treats Q the way she treats unicorns’--but there is no world at which the largest prime number is instantiated. Fodor’s revised theory of error introduces the notions of a labelling concept and the application of a symbol. I have argued that neither of these will bear the weight it is required to bear: the notion of a labelling context is not well enough worked out to allow us to see whether any such uses exist in the language of thought, and the notion of applying a symbol seems to make essential use of intentional idiom. 72 Fodor’s notion of a special labelling context in which symbols are applied and whose truth conditions essentially involve causes raises more problems than it solves. In revising his theory of error, Fodor has retained, with little alteration, his explication of what it is for a symbol to express one rather than another property. What can we say about his explication of what it is for a symbol to express a property? I shall be arguing that it, too, fares badly, that asymmetry of causal dependence does not explain what gives symbols of Mentalese their meaning. According to Fodor, a symbol of Mentalese, R, expresses a property, P, if and only if the following conditions hold.32 (1) At least some instances of F actually cause R tokenings or, if F is uninstantiated, its instances would cause R tokenings were they to occur. (F is nomically related or at least reliably linked to R.) (2) The capacity of any other property, G, to cause R tokenings at the actual world is nomically dependent on a lawful relation between the properties of being F instances and being R tokenings. (The nomic relation between G and R is itself nomically dependent upon the nomic relation between F and R.) (3) The capacity of an F instance to cause an R tokening is independent of G’s capacity to cause R. (The nomic relation between F and R is nomically independent of the nomic relation between G and R.) frogether, (2) and (3) comprise what Fodor refers to as asymmetric causal «dependence [ACD]. Fodor believes that some such "purely formal" asymme- try of dependence between relations will have to characterize any ¥ 32][&R, p. 11; Psysem. pp. 108-9. 73 account, causal or otherwise, of what a symbol does (and does not) express.33 He has, therefore, a considerable theoretical stake in the joint plausibility of (2) and (3). It has been shown elsewhere that (2) is vulnerable to challenge.34 Contrary to Fodor’s claim, it appears en- tirely compatible with ’horse’ expressing gplgp that there be a nearby world at which cows, but not horses, cause ’horse’ tokenings. I shall argue that (3) is equally questionable: it is not the case that if R expresses F there is some nomically possible world where Fs cause Rs but Gs do not. Fodor urges (3) as intuitively compelling. "From a semantic point of view,” he tells us, "mistakes have to be accidents: if cows aren’t in the extension of ’horse’ [i.e., if ’horse’ expresses gplgp and not gplgg pl_ppy], then cows being called horses can’t be repuired for ’horse’ to mean what it does."35 For ’horse’ to express gplgp, it is not necessary that cows cause ’horse’ tokenings; consistent with ’horse’ expressing gplgg, there are, thus, possible worlds where cows do not cause ’horse’ tokenings. At first glance, (3) seems unexceptionable. Things are not, however, as simple as they seem. If ’from a semantic point of view’ means ’with regard to linguistic .necessity’, Fodor’s observation is correct but hardly germane. True ‘ 33Psysem. , p. 110. 34R. Hall, "Fodor’s Theory of Error is Erroneous", MS. SPsysem., p. 108. 74 enough, if what ’horse’ expresses is gplgp, there is no linguistic necessity that cows cause ’horse’ tokenings--there are linguistically possible worlds at which cows are not called ’horse’. But ACD deals in nomic, not linguistic necessities. That there are linguistically pos- sible worlds--i.e., worlds susceptible of coherent description--at which cows do not cause ’horse’ tokenings is neither here nor there with re- gard to the claim that if R expresses F, then, for any G such that G instances can cause R at our world, there is a nomically possible world at which Fs cause R tokenings but Gs do not. Nomic possibility con- strains more stringently than does linguistic possibiity.36 (3) says that for any relevant G, there exists a nomically accessible world at which Gs do not cause Rs but Fs do. (3) is persuasive because such worlds seem remarkably easy to construct. Their apparent avail- ability is, however, due largely to vagueness about what can stand in for G in (3). Once we narrow our focus to properties that can be featured in nomic relations, as Fodor requires, it becomes considerably more difficult to find the appropriate worlds. Even though Bossie the cow, Zeno the zebra and Willie the wildebeest have, in fact, tricked us into tokening ’horse’, it certainly seems that 36Even were linguistic necessity were what mattered to the truth of ACD, all that the argument above underwrites is the existence of a possible world at which Gs do not cause Rs. We would still have no assurance that at any such world Fs pp cause Rs, as (3) requires: even on the causal theory of meaning, R’s expressing F entails only that (at least some) Fs cause Rs at the actual world or would, were there to be any there. 75 they need not have done so. Surely, there are nomically possible worlds where stronger illumination, better night vision or a more cautious attitude effectively precludes such mistokening, and there seems no reason why, at some such world, gpggp instantiations should not cause ’horse’ tokenings. This does not, however, bear on the status of cow- or zebra- or wildebeest-caused ’horse' tokenings as determined by Fodor’s theory. ACD, as we have already noted, deals with asymmetry of gpglp relations between properties. Fodor drives this point home in a "parade version" of his doctrine of asymmetric causal dependency. (It will be recalled that a ’wild’ token of a symbol is one whose cause is irrelevant to what the symbol expresses.) B—caused ’A’ tokens are wild only if the nomic dependence of [(instantiations of the property of being an ’A’ tokening) upon (instantiations of the property of being a B)] is itself dependent upon [(the nomic dependence of the property of being an ’Aé7tpgening) upon (instantiations of some property other than B)]. ’ Even if we admit being an instantiation of Bossie or Zeno or Willie as a bona fide property, still, like being a left shoe or beingwithin 100 meters of the Eiffel Tower, it is not the sort of property that features 39 in natural laws. It is not what Fodor calls a nomic property. Nor is 37Psysem., p. 164, fn. 6. Parentheses added; one obvious error corrected. 38Fodor has stressed the centrality of nomic relations in informal remarks as well. In LOT, his discussion of the "private language" argument proposes that nomic relations bear the burden of guaranteeing the language of thought the "coherence" achieved by convention in natural language. 39"Why Paramecia Don’t Have Mental Representations", in Peter A. French, (Footnote continued) 76 it even plausible that a property of this sort might be reliably connected with psychophysical properties that are themselves featured in laws. Being Bossie or Zeno or Willie is clearly not the kind of property that reliably gives rise to, say, a particular pattern (or set of patterns) of light. Bossie can change shape or color and still be Bossie; the same goes for Zeno and Willie. Whatever it is that makes for the identity of such individuals does not seem likely to interface with properties featured in psychophysics or cognitive psychology. Focussing on the asymmetric dependence of ’horse’ tokenings caused by Bossie or Zeno or Willie on ’horse’ tokenings caused by Dobbin and Paint and Silver is, it seems, bothering with the wrong G. After what property ought we then to be asking? To pose the question is to notice how little guidance we are given in this regard. As we have seen, Fodor talks in Psychosemantics about cows as causes of mistaken tokenings of ’horse’ and elsewhere (rather less plausibly) about cows as causes of erroneous tokenings of ’platypus’.40 Perhaps what (3) demands, then, is a nomically accessible world at which instances of cow or zebra or wildebeest do not cause ’horse’ but at which instances of horse do (or at which instances of cow do not cause tokenings of ’platypus’ but in- stances of platypus do). But this cannot be right, either. At the actual world, error is not the result of a nomic connection between the property of beingZan instance of cow or zebra or wildebeest and the 39(continued) Theodore E. Uehling and Howard K. Wettstein, eds. Midwest Studies lg Philosophy X (Minneapolis: University of Minnesota Press, 1986), p. 10. 40I&R. 77 property of being an instance of ’horse’—-there simply is no such lawful dependence. Indeed, the proposed property is just the sort of gerrymandered and open-ended disjunction to which Fodor has elsewhere 41,42 denied natural kindhood. Nor does it help to deal with each of the disjuncts separately: cow, zebra and wildebeest may each be a natural kind, but no one of them, ppp kind, is lawfully linked to ’horse’ tokenings. Most cows (zebras, wildebeests) are innocent of any such causal complicity. So what property could stand in for G? The best candidate in sight is, I think, a property that Fodor invokes in tracing a semantically relevant causal sequence from horses to ’horse’s: the 43 property of having "that horsy look". If having a horsy look is, as he urges, part of the reliable causal sequence that connects gplgp instances with ’horse’ tokenings44 surely it is just the property we want when we are looking for a similarly lawful connection between ppy instances and (mis)tokenings of ’horse’. Following Fodor, then, let us say that Bossie and other non-horses are capable of causing ’horse’ tokenings because, like horses, they have a horsy look, a look similarly reducible to "some or other. . . bundle of psychophysical properties,"45 which properties can, in turn, serve as "links in a causal chain" that 41LOT. 42’Open-ended’ because we must take into consideration not only all Gs that do cause Rs at the actual world but all Gs that would do so were they to be instantiated. This covers a lot of territory. 43psysem., p. 118. 44Psysem., p. 119. 45Psysem., p. 118. 78 eventuates in ’horse’ tokenings.46 In so saying, we identify a property that is, at least, of the right sort to figure in ACD’s nomic hierarchy. Is there a world governed by natural laws relevantly similar to those of our own, where non-horses with "that horsy look" do gpl cause ’horse’ tokenings but ppgp llpp horses nevertheless do? It seems unlikely. A horsy look is the shape that horses take on at a world by virtue of their horsehood. Any creature that has lglg shape at a world has, at that world, a horsy look. How could there be a world at which non—horses with lglg property fail to cause ’horse’ tokenings but horses occasion them? What natural causal process could possibly have such an outcome? Surely not one that goes from DNA-in-an-environment to a resultant out— ward form with a capacity reliably to produce in viewers (via laws of psychophysics and cognitive psychology) a tokening of ’horse’. It is worth noting that we cannot satisfy (3) simply by constructing a world where non-horses have (upon occasion) the perceptually relevant properties that pplgpl horses have but where viewers have perceptual abilities that allow them to distinguish such non-horses from the real thing. "That horsy look" is, at a world, whatever perceptually relevant form characterizes horses pl that world; it is the set of features per- ceptually relevant to horse-identification at that world. Otherwise, it could not be part of a nomic chain connecting horse instances with ’horse’ tokenings. What is perceptually relevant at a world, however, 46Psysem., p. 119. 79 depends upon the perceptual capabilities of viewers at that world. If viewers at a world are capable of distinguishing cows from horses by their X-ray vision, then having a horsy look pl that world is having whatever features reliably cause ’horse’ tokenings in viewers with X-ray vision. It may appear unfair to Fodor to choose having that horsy look as the relevant property to stand in for G: it is obviously not going to do the job he has in mind. If he has a better candidate, it is up to him to identify it. Pending such help, (3) stands as considerably less intuitively appealing than it seemed at first blush. If either (2) or (3) proves unacceptable, ACD fails; at the moment there seems good reason to reject them both. We have been proceeding so far on the assumption that although parti- cular ones of Fodor’s counterfactual claims were open to criticism, his general strategy of talking about what symbols of Mentalese ypplp be tokened under contrary-to-fact conditions was itself unobjectionable. We are now ready to challenge this assumption. In what follows, I shall be raising questions about the extent to which we know what we are talking about when we make claims about tokenings of symbols of the language of thought at variously specified possible worlds. I shall be arguing that Fodor is dangerously unclear about typecasting such tokens. 80 Let me begin by drawing attention to a rather obvious point: to make sense of the counterfactual claims upon which Fodor’s project relies, we must be able to make sense of the notion that a particular event is or is not a tokening of a‘Mentalese symbol, R,--there must be some consid— eration in the light of which it counts (or fails to count) as such a tokening. The claim that under optimal psychophysical conditions any instance of Egg would cause a tokening of Mentalese RED is, for example, the claim that at any nomically accessible world at which certain cir- cumstances hold, there will be, among the events caused by any instance of lpp, a tokening of Mentalese RED. The claim that R expresses F only if the nomic dependence of R tokenings on G instances is itself dependent on the nomic dependence of R tokenings on F instances is, similarly, the claim that at every nomically accessible world where R tokenings are among the effects of G instances, R tokenings are also among the effects of F instances. The truth or falsity of such counter- factuals is determined by the way things go at the relevant worlds--by whether R tokenings stand in the specified relations. It thus behooves us to be reasonably clear about what makes an event be (or not be) an R tokening.47 47To avoid charges of verificationism, let me emphasize that what is at issue here is not the epistemological question, How are we to tell whether an event is a tokening of R? but, rather, the metaphysical one, What is it that makes an event pp a tokening of R? By virtue of what is an event a tokening of R rather than of R’ (or perhaps of no symbol of Mentalese whatsoever)? 81 We may, for convenience, divide the question, asking first, what makes an event be a tokening of any symbol of Mentalese whatsoever—-what makes it count as a mental representation rather than as a stage of a diges- tive process-—and, second, given that an event is a tokening of some or other symbol of Mentalese, what makes it count as a tokening of R rather than of R’? Let us assume for the time being that we have at least a line on the first of these. If, as Fodor tells us, being a mental repre- sentation amounts to being a type of brain state that stands in a certain nomic relation to having a belief, and if having a belief is a matter of causal relations and causal powers, then being a mental repre— sentation (rather than a stage of a digestive process) is a matter of 48 This does not tell us, however, what it is for any functional role. particular occurrent mental representation to count as a token of one or another symbol. We still need to know how Fodor proposes to assign tokens of Mentalese to their respective symbols. That this question poses a problem is, I suspect, rather considerably obscured by Fodor’s choice of notation and by his insistence on the expository inter- changeability of terms referring to symbols of English and Mentalese. The first renders it difficult to ask the question; the second makes it appear that we already know the answer. Fodor explains his "orthographic conventions" in Psychosemantics as follows. Whenever it matters, names of concepts will appear in caps; names of properties in italic; and names of words. . . in quotes. From time to time, I use quoted English formulas to refer to the corresponding expressions of Mentalese. . . . Thus Egg is a property (the one that red things qua red things share); RED is the concept 48LOT, p. 78; Psysem., p. 88; I&R, p. 13. 82 which denotes (or expresses) £393 and ’red' is a term (either of English or Mentalese) that encodes that con- cept. . . . [Since] RTH has it that concepts just plg expressions of Mentalese, it turns out that the two formulas ’the concept RED’9and ’the Mentalese expression "red"’ are coreferential. Similarly, he says elsewhere, Thus "’rain’" will. . . refer to the sypBol that encodes the . . . [property] rain in Mentalese. In addition, Fodor uses pluralized forms of the names of Mentalese symbols to refer to their tokenings: e.g., "[Tlhe causation of ’horse’s’ by nonhorses has to be asymmetrically dependent upon the causation of ’horse’s by horses. . ."51 Names of tokens are thus typographically linked with names of symbols--there is no way to separate the name of a token from the name of the symbol of which it is a token. This makes it unhandy, to say the least, to ask by virtue of what an event counts as a token of one rather than another symbol of Mentalese: an ’F’ is neces- sarily a token of the Mentalese symbol F, a ’G’ is necessarily a token of G and so on. Expressing a question about typing tokens requires an end run around Fodor’s notation, as, for example, ’What makes an event, e, count as a token of the Mentalese symbol R?’ A question that cannot be expressed conveniently in a theorist’s notation of choice is apt to be overlooked in the theory; as far as I can tell, Fodor has not addressed himself to this issue lately. agPsysem., p. 160, n. 50I&R, p. 14, n. 51Psysem., p. 165, n. 83 Even if we manage to pose the question, Fodor’s predeliction for down- playing differences between expressions of Mentalese and English words tends, I believe, further to obscure the issue. As competent speakers and readers, we know what auditory or visual features--what perceptible properties—-matter in making a spoken or written word of English count as a token of ’horse’; we have, what is more, a pretty good idea of what range of variation is permissible with respect to such features. Competent speakers and readers of a natural language type its tokens without giving the matter a second thought.52 Encouraged to assimilate Mentalese to English, we are likely to take it for granted that there is something analogous that makes an occurrent mental representation be a tokening of a particular Mentalese symbol. But what makes it the case that a set of marks or sounds is a tokening of the English word ’horse’ --features determined by linguistic convention and parameters determined by the shared quality spaces of human speakers and readers--can hardly be what makes it the case that a neural event is a tokening of the Mentalese symbol R. 52Or perhaps even a first thought. On the other hand, a sociologist once asked my husband how mathematicians ascertained that a mark on the first line of a proof and a similar looking one on the next line were both commas. And we find the following in a particularly painstaking approach to set theory: We do not propose to enter into a discussion of the psy- chological and metaphysical problems which underlie the use of ordinary language [to describe a formalized language]. . . for example, the possibility of recog- nizing that a letter of the alphabet is ’the same’ in two different places on the page, etc. . . . (Nicolas Bourbaki, Elements pl Mathematics: Theory pl Splg (Reading, MA: Addison—Wesley Publishing Company, 1968). I am indebted to Hal Slater for this. 84 It is, indeed, not clear that what tokens of a Mentalese symbol have in common should be characterized in terms of their form. At least one reading of Fodor’s orthographic conventions suggests that symbols of Mentalese are individuated, rather, by the properties they express. It will be recalled that a string of letters in single quotes refers to the Mentalese symbol that expresses the property whose name we get when we remove the quotation marks and italicize the string: e.g., "’rain’" refers to the Mentalese symbol that expresses the property lplg. On a de dicto reading, then, ’red’ is the name of whatever symbol of Mentalese expresses (at a world) the property lpp. We may call this mode of individuating symbols of Mentalese semantic typecasting. Semantic typecasting obviates questions about the sameness of shape of neural events and makes Fodor’s orthographic conventions appear natural. Unfortunately, semantic typecasting of symbols of Mentalese raises problems when we try to pose the question central to Fodor’s project: What makes a symbol of Mentalese express one rather than another property? The first of these problems is, by now, predictable. If, by explicit orthographic convention, ’horse’ is the name of whatever Mentalese symbol encodes the concept HORSE at a world, and is thus the name of whatever Mentalese symbol expresses the property gplgg at that world, then ’horse’ expresses gplgp necessarily and it is hard to see how we can even ask what makes the Mentalese symbol ’horsei express gplgp rather than horse or cow (or, for that matter, whale). Once again, we 85 have to evade Fodor’s notation and say something like, ’By virtue of what does some symbol of Mentalese, R, express gplgp rather than gplgg pl_ppy?’ Second, even if the relevant query can be expressed, the counterfactual claims with which Fodor attempts to answer them will, on this reading, be circular. Let us see why. If "’horse's" names instances of the property of being the symbol ’horse’, then, if we are dealing with ’horse’s, we are necessarily dealing with instances of the Mentalese symbol that expresses the property gplgp. But on Fodor’s account, what makes ’horse’ pp the Mentalese symbol that expresses gplgp is just the structure of causal relations in which llg tokenings fea— ture. Which leaves us with the uncomfortable finding that, in order for a Mentalese symbol to express gplgp, it must be tokenings of lgpl symbol that stand in appropriate nomic relations to horses; for a particular event to be a tokening of lgpl symbol, it must be an instance of the symbol that expresses horse. The troubles I have just canvassed arise when we read locutions like ’the symbol that expresses gplgp’ de dicto—-when we individuate symbols of Mentalese semantically. It is thus perhaps perverse to do so. Why not assume, instead, that such definite descriptions are to be read de re-- as picking out, at any world, the very same event type that just happens to express gplgg at the actual world? Would it not make better sense to typecast symbols of Mentalese syntactically, as we do the words and letters of natural language? In favor of this view is the fact that there are places where nothing but a de re reading will do. "But for the 86 fact that the word ’horse’ expresses the property of being p horse," Fodor argues, "it would not have been that word that taking a cow to be 53,54 a horse would have caused me to utter". Similarly, a bit later, Fodor says, "A causal theorist can acknowledge only one kind of world in which Xs don’t cause ’X’s; viz., the kind of world in which ’X’ doesn’t 55 mean l". There is no way that either of these remarks could sustain semantic typecasting. (There is no way the second of them can be made consistent with Fodor’s orthographic conventions, either. We are not the 56 only ones pushed to evade them.) Let us suppose, then, that Fodor’s 53Psysem., pp. 107-8. 541 am indebted to R. Hall for pointing this out and for being generally skeptical of the claim that symbols of Mentalese are consistently individuated semantically. 55Psysem., p. 110. 56Elsewhere, however, Fodor seems to require semantic typecasting. Con- sider, for example, his defense of the necessity of ACD. Suppose, he says, that we deny (2), the clause that says that R expresses F (and not G) only if it is the case that were Fs not to cause Rs, Gs wouldn’t either. So to do, he tells us, is to imagine a world at which ’horse' was not applied to horses but was applied to cows, but where 'horse’ still meant horse and applications of ’horse’ to cows were ipso facto false. We have, in other words, a world at which horses do not cause ’horse’s but cows do. He says, This doesn’t look like a possible case to me. What on earth would gplp ’horse’ mean horse in such a case? What would stop it from meaning ppy? (Psysem., p. 164, n.) But why should it be required that ’horse’ mean horse at this world? Only if it is assumed that ’horse’ refers to the symbol that expresses horse. As far as I can tell, Fodor’s argument relies on the assumption that for a world to be one at which Xs cause ’horse’ tokenings, it must be a world at which Xs cause tokenings of the Mentalese symbol that expresses horse. If all that were required for e to be a tokening of ’horse’ at the world in question was that e share the shape of the symbol type expressing horse at the actual world, Fodor’s objection could not be raised. At least in this context, Fodor seems to assume semantic typecasting-~a tokening of a symbol that does not express horse (Footnote continued) intent Then we token c actual tokenir appa re Ve car us: 1‘ 'hors that case Q8113] 87 intent is, throughout, to typecast symbols of Mentalese syntactically.57 Then we cannot avoid the problem of what it means for an event to be a token of the symbol (type) that expresses a particular property at the actual world. We need, that is, an analogue to the process by which tokenings are gathered into natural language types. It is not at all apparent, however, what this would be. We cannot rely on the actual causes of Mentalese tokens to type them for us: if we go this way, it is impossible for a cow to cause a tokening of ’horse’. We cannot rely on counterfactual causes, either. Suppose we say that an event caused by a cow counts as a tokening of ’horse’ just in case it would not have been caused by a cow were it not capable of being caused by a horse, but not vice—versa. The truth of claims of this sort, like those at which we looked earlier, depends on how things go at nomi— cally accessible worlds. In this case, it is relations between lglg yply tokening and ppy (or gplgp) instances that matter. This means that there is some way of identifying tokenings across worlds without reference to their causes. My hunch is that establishing transworld identity condi- tions for individual neural events is at least as formidable a puzzle as identifying at a possible world what counts as a token of a given type. 56(continued) does not count as a tokening of ’horse’. 57His informal comments in discussion suggest that this may well be his goal. Perhaps on the bi them. Do; content have, in so—calle< causal a and, fin an issue been the example: serioUs Unevide 88 Perhaps there is, finally, a way of sorting events into Mentalese types on the basis of what they cause rather than what causes (or would cause) them. Doing this without lapsing into the functional role account of the content of symbols that Fodor has rejected looks unpromising, however. I have, in turn, expressed dissatisfaction with Fodor’s treatment of so-called labelling contexts and the application of symbols, with his causal asymmetry condition for the meaning of Mentalese expressions, and, finally, with the lack of concern for how tokens are to be typed, an issue central to his theoretical proposals. Throughout, my worry has been that lack of detail and pervasive reliance on natural language examples tend to obscure significant difficulties. We may well see as a serious shortcoming in a theoretical proposal the fact that it makes unevident the costs of its adoption. Like Fodo mined ch as Worthy basis for nomicity easy to r FOdor. B Shall be theorist belief c shall l< First, day Vi; Chapter Four Fred Dretske’s Information-Theoretic Account of Representation Like Fodor, Fred Dretske finds that the materialism to which he is com- mitted challenges belief-desire idiom; like Fodor, Dretske sees this idiom as worthy of defense. Both theorists are concerned to find a naturalistic basis for semantic content; in attempting to do so, both rely on notions of nomicity and the causal structure of the world. These similarities make it easy to read Dretske as though he were aiming at the same objectives as Fodor. Because there are important differences between their projects, I shall begin this chapter by comparing some central commitments of the two theorists; I shall then take up Dretske’s initial attempts at analyzing belief content in information theoretic terms. In the next chapter, we shall look critically at his more recent work on the problem of misrepresentation. First, compared with Fodor’s intentional realism, Dretske’s attachment to psychological explanation has more to do with the legitimacy of our every— day view of ourselves as agents, acting on reasons, and rather less to do with the claims of cognitive science. Second, it follows from Dretske’s understanding of this everyday view and his thoughts about the nature of explanation that his criteria for the ontological respectability of pro— positional attitudes will also differ from Fodor’s. Dretske holds, as Fodor 89 most c quires ("best possib eXplan repres and, e Where constr inform invite “Don : it V01 line, at th. Drets Drets and h thjng 90 most certainly does not, that acknowledgement of beliefs and desires re- quires that their content be assigned a causal role by an ineliminable ("best") theory. Dretske (but not Fodor) must therefore worry about the possible "causal impotence" of representational content, the possible explanatory irrelevance of "ygpl a structure represents and gpy it represents it."1 Finally, Dretske’s strategy is notably more historical and, especially in his later work, more teleological, than is Fodor’s. Where Fodor strives to show us that puzzling intentional properties can be constructed out of causal ones in logical space, Dretske tells us that information is "the raw material out of which minds were manufactured"2 and invites us to understand such manufacture in terms of benefits conferred upon systems capable of utilizing information: "It is hard to see why. it would be. . . worth having a mind if it didn’t, somewhere along the line, help one get along better in the world."3 Let us take a closer look at these parallels and differences. Dretske and Fodor agree in finding intentionality deeply problematic. Dretske identifies the worry as follows: "The content of a cognitive state, and hence, the cognitive state itself, depends (for its identity) on some- thing beyond the extension or reference of the terms we use to express the 1"Why Thinking Helps", mss., 1988, p. 10. Henceforward, WTH. 2Dretske, Fred, "Precis of Knowledge and the Flow pl Information" in H. Kornblith, ed., Naturalizing Epistemology (Cambridge, Mass.: The MIT Press/A Bradford Book, 1985), p. 174. Hereafter, Precis. 3 WTH, p. 5. conter not tl is on my thc about Dretsk can be Yeast 4 Cognitt mGChan: 91 content."4 If, however, my thought that what is on the stove is burning is not the same as my thought that John’s lunch is burning, even though what is on the stove (and burning) is, in fact, John’s lunch, it would seem that my thoughts have, essentially, properties that are not dependent upon facts about the material world. A materialist [thus] confronts the task of explaining, or explaining away, this intentional feature of cognitive states. Some account must be given of how a purely physical system could occupy states having a content of this sort. Or, failing this, some explanation must be given of why we sys- tematically delude ogrselves into thinking that yp occupy states of this sort. Dretske does not think that we are deluded. He believes that intentionality can be naturalized, that we can "bake a mental cake using only physical yeast and flour."6 In particular, he thinks he can show that "genuine cognitive systems. . . can develop out of lower order, purely physical. mechanisms."7 He commends his theory to us as the best explanation of how this could happen. If physical systems are capable of developing concepts, with the consequent capacity for holding beliefs, both true and false. . . then the internal structures that qualify as be- liefs or representations must develop in something like the way I dgscribe. I see no other way for meaning. . . to evolve. 4Dretske, Fred, "The Intentionality of Cognitive States", Midwest Studies lg Philosophy, vol. 5, ed. Peter A. French, Theodore E. Uehling, Howard K. Wettstein (Minneapolis: University of Minnesota Press, 1980), p. 282. Hereafter, ICS. 5Ics, p. 282. 6KFI, p. xi. 7KFI, p. viii. 8KFI, pp. x~xi. The view that naturalization requires a genetic account is (Footnote continued) 92 Why does Dretske think it important to come up with such an account? Why not simply eliminate awkward intentional properties? Like Fodor, Dretske believes that commonsense psychology is a fundamental part of our view of ourselves. Dretske thinks there is a heavy cost involved in exchanging psychological explanations, phrased as they are in terms of reasons that are available to each of us, for biological accounts available only to those with neuroscientific expertise. We would, he thinks, thereby give up. . . authority. . . about why we do the things we do. . .[and thus] relinquish a conception of ourselves as human agents. Ehis is something that we human agents will not soon give up. 8 (continued) by no means unique to Dretske. Compare, for example, Dan Lloyd’s observa- tion that "To fit a mental entity into the natural order, without magic or handwaving" requires (among other things) a naturalistic account of the origin of the mental: "Ultimately, evolutionary biology must account for the beginnings of the basic cognitive equipment of our species. . . . Developmental psychology and learning theory must account for those capa- cities we acquire in life." According to Lloyd, a natural philosophy of mind builds on natural science, asks similar questions, and settles for the same sort of necessity. ("Mental Representation From the Bottom Up", Synthese 70 (1987) 23-78, pp. 23-24.) Dretske explicitly holds his theory to be vulnerable to empirical results in, e.g., developmental psychology (KFI, p. x.). Interestingly, Fodor himself respects this intuition. Criticizing the new associationists’ doctrine on concept formation, he observes that "the con- structibility lg lpgical princlple of arbitrarily complicated processes from elementary ones doesn’ t begin to imply that such processes are con- structible ino ontogeny by the operation of any . . . mechanism . . . that associationists would be prepared to live with" (HM, p. 34n). Bradford Book/The MIT Press, 1988) p. x. Henceforward, EB. Dretske phrases this as the worry that "it will be biology rather than psychology that explains why we do the things we do" (EB, p. x). If what bothers him is the prospect of being less than expert about why we do things,——if it is the demise of what has been called ’causal Cartesianism’ that bothers him--he should be equally worried about non—commonsense psychology, even one that uses belief—desire idiom. Freud endorsed beliefs (Footnote continued) Dretske's v nation is a saw his ta: would allo explanatio tional idi intention; such role 93 Dretske’s view of what it takes to save intentional explanation from elimi- nation is also somewhat different from Fodor’s. Fodor, it will be recalled, saw his task first, as finding a characterization of mental states that would allow us to ascribe causal powers to them, and, second, as finding an explanation of their semantic properties that would not make use of inten— tional idiom. At no point did he feel obliged to assign a causal role to intentional properties themselves. Indeed, he explicitly denies them any such role. I don’t believe that there are intentional mechanisms. That is, I don’t believe that contents per se determine causal roles. . . . [I]t’s got to be possible to tell the whole story about mental causation. . . without referring lp ; ; l intentional properties of. . . mental states. . . . [Wlhile I’m prepared to sign on for-chunterfactual-supporting inlan- tional generalizations, I balk at intentional causation. Dretske, on the other hand, believes that it is incumbent upon him to show that intentional properties are featured in psychological explanation. Citing the passage above, he says, 9(continued) and desires but notoriously rejected the claim that our own explanations of our actions were always correct. Freud may have believed that correct ex- planations were at least potentially available to us--that they were there but in some sense veiled; more recent theorists have challenged the notion that we have any form of privileged "observational" access to the correct explanation of our behavior. Nisbett and his colleagues have, for example, ground our faces in evidence that even when there is nothing especially threatening about self—knowledge, we may still fail (rather spectacularly) to come up with correct explanations of our choices, judgments and actions (Richard E. Nisbett and Timothy DeCamp Wilson, "Telling More Than We Can Know: Verbal Reports on Mental Processes", Psychological Review, 84, 1977, pp. 231—259). Unlike Dretske, I am not convinced that even our everyday view grants us epistemic "authority" with regard to explanations of our own actions. Dretske’s insistence on such authority may explain why he does not go to bat for scientific psychology with the same enthusiasm as Fodor. 10Language, Psysem., pp. 139-40, stress in the original. Dretske co takes it I fending. 1 says, DretSke 1 POWerles tax, the elemen I S because 94 I disagree. . . with Jerry Fodor. . .[that] an adequate story can be told about mental causation without making intentional properties. . . determine causal roles. It isn’t enough to have these. . . properties figure in counterfactual— supporting generalizations. That (alone) won’t show that people behavglthe way they do because of what they believe and desire." Dretske comes to this view by two somewhat different routes. First, he takes it to be part of the everday picture of ourselves that he is de— fending. Although our ordinary way of thinking may not be sacred, Dretske says, I do think it is a reasonable place to begin. And it is, other things being equal, and assuming a managgable philosophical cost, a desirable place to end. Dretske rejects Fodor’s doctrine that "[mleaning itself is causally inert, powerless to initiate, modify or influence behavior," that "it is the syn- tax, the shape and form, not the semantics, of. . . [the] representational elements [of the language of thought] that does all the explanatory work," because it flies in the face of our ordinary view that the belief that there is a beer in the refrigerator (together, of course, with a desire for a beer) causes Clyde to go to the refrigerator because of what this bellsf is a belief about and what this desire is a desire lpl. Dretske thinks this intuition reflects a more general fact about what is involved in giving an explanation: A causal explanation of an event is. . . more than a specifi— cation, under some description or other, of the event’s 11"Putting Information to Work", paper given at conference "Information, Language and Cognition", Simon Fraser University, Vancouver, BC, 1988, p. 12. Henceforth, PIW. 12Dretske, Fred, "Reasons and Causes", mss. 1986, p. 4, n. Henceforward, R&C 0 13sec, p. 4. C81 am( pm Um Thus, "the desire“ mus do."15 Fodor's so: fore, be D. for unders our doing against e1 Well. We n that we 31 eXplanati‘ that Play ThuS , Dre Perhaps ( inVentOry in Vhlch \ l4 ng’ 15 Rgc’ 16 R&C, l7 Rfic’ 95 cause. An explanation requires, in addition, and perhaps among other things, some indication of which of the pro- perties of the cause, by being law-instantiating properiles, underlie the cause’s efficacy in producing that effect. Thus, "the semantic aspect of reasons, the what-it—is we believe and desire" must "[figure] essentially in the causal explanation of what we do."15 Fodor’s solution to the salvage of belief-desire psychology cannot, there— fore, be Dretske’s solution. For one thing, it does not provide "resources for understanding the explanatory role of reasons, the why and wherefore of 16 our doing something because of what we want and believe." Fodor’s defense against eliminativism fails to meet another of Dretske’s requirements as well. We may recall that among Fodor’s central commitments was the tenet that we are obliged to admit into our ontology all entities featured in the explanations we use. Dretske holds, with equal emphasis, that pgly entities that play a role in bona fide causal explanations may be so acknowledged. Thus, Dretske can say, "If reasons aren’t causes, one of the chief motives, perhaps (for some people) the pgly motive, for including them in one’s 17 What is more, the causal explanations inventory of the mind, vanishes." in which reasons are featured must be ineliminable. Reasons must not only be causes, they must be, with respect to some class of phenomena, the only 14sec, pp. 1-2. 15mac, p. 3. 16R&C, p. 5. 17R&C, p. 1. 96 successful candidates for causal (and thus explanatory) status. For Fodor, the tension between materialistic monism and realism about the propositional attitudes lies in the awkward intentional properties to which use of commonsense psychological explanation commits us. For Dretske, by contrast, materialism poses two additional problems. First, how is it pos- sible to attribute causal powers to these intentional properties? Second, how can the resulting style of psychological explanation survive a com— pleted neuroscience? As a materialist, Dretske holds that all motions are susceptible of mechanistic explanation; the answer to ’Why did my fingers move that way just then?’ will, he assumes, sooner or later be given in terms of muscle twitches, neural firings, transmitter substances and the like. If this is so, however, it might appear that the extensionally equi- valent ’Why did I pick up the wineglass just now?’ will also, sooner or later, be answerable in the vocabulary of neurobiology. Answers like ’Because I believed that the glass had sherry in it and I wanted a sip of sherry just then' are, it appears, acceptable only for the time being: as soon as neuroscience gets its act together, psychology will wither away. Where Fodor was primarily concerned to show that the mental states to which we were committed by our reliance on commonsense psychology were not irre- trievably lumbered with properties alien to physics, Dretske finds need to convince us that psychological explanation is not hopelessly redundant as a causal account of our doings. Dretske’s project thus both overlaps with and differs from Fodor’s. Viewed in long per subgoals. A. Deva intr 97 in long perspective, it may be seen as having a structure of goals and subgoals. A. Develop a naturalistic theory of meaning: domesticate the intentionality of meanings and beliefs. 1. Show that intentionality ppl gp is a natural ("objective) property. 2. Show that the kind of intentionality we attribute to belief states, in particular, can be characterized naturalistically. a. With respect to degree (the problem of semantic content). b. With respect to semantic evaluability (the problem of misrepresentation). 3. Show that naturalistically characterized processes can eventuate in systems whose states have such intentionality. B. Defend psychological explanation (i.e., explanation featuring states individuated with respect to their content). 1. Show that intentional properties can play causal roles. Fodor's pc largely wi over the i tion of it shall be theory of mantle QC the Prob? tion bee. recent e able r01. ClOSely 98 2. Show that psychological explanation is ineliminable—- that what is explained by intentional causation cannot be (as well) explained otherwise. Fodor’s positive doctrine on the semantics of Mentalese has been presented largely within the compass of two relatively recent papers. Dretske has, over the last decade, produced an extensive body of work on the naturaliza- tion of intentionality and belief. In the remainder of this chapter, I shall be concerned with his initial efforts to develop a naturalistic theory of belief by way of an information-theoretic characterization of se- mantic content. In the chapter that follows, we shall take up later work on the problem of misrepresentation, where an explicitly teleological orienta— tion becomes more focal. Although we shall not be discussing Dretske’s most recent efforts to give psychological explanation a legitimate and inelimin— able role, we shall want to keep in mind that he sees this objective as closely related to the naturalization of meaning. [A] strategy that I find tempting. . . [is to choose] among otherwise plausible theories of meaning on the basis of how easy they make the job of exhibiting the explanatory role of meaning. Any theory of meaning that doesn’t enable a thing’s having meaning. . .[to be] a causally relevant property of the thing, and hence the fact that it has this meaning. . [to be] an explanatorily important fact about the thing, is a theory of meaning that can be rejected at the outset. It would be like having a theory of pain that made its occur rence something of which the sufferer could not be aware. How, then, does Dretske propose that we get from the rocks, the nebulae and the unbelieving worms to full-fledged cognitive systems capable of repre- 133ac, pp. 9_1o. 99 senting (and misrepresenting) the world to themselves? His central theoretical concept has been that of information. In a recent paper, he tells us that he is "convinced that belief and meaning are notions that ultimately derive from the information-carrying properties of living systems," and refers his reader to the particular characterization of information that I shall be discussing here.19 0n the one hand, Dretske characterizes carrying the information that l as an entirely natural, ’objective’ property of physical systems. On the other hand, he tells us that carrying the information that l is an intentional property. Because the "information a signal carries exhibits the same llgp of intentionality as what we believe or know," Dretske tells us, it is plausible that "the intentionality of our cognitive attitudes (the way they have a unique content), a feature that some philosophers take to be distinctive of the mental, is a manifestation of their underlying 20 It is thus understandable how purely information-theoretic structure." physical systems might have intentional states. Because a signal’s infor— mational content exhibits a different ppglpp of intentionality than does what we believe or know, it is, in addition, understandable how some, but not all, systems in nature come to have mental lives. "The higher-level accomplishments associated with intelligent life," Dretske tells us, can be seen as "manifestations of progressively more efficient ways of handling and coding information".21 19PIW, pp. 5,6. 20KFI, p. 76. First stress added. 100 For Dretske, the naturalization of belief is thus a matter of showing that beginning with a natural property and using only naturalness-preserving transformations, we can arrive at the property with which we are concerned --the semantic content of beliefs. Let us look more closely at how he goes about this. Dretske defines the informational content of an event as follows. Where l is the receiver’s prior knowledge about relevant probabilities, [I] That r is G carries the information that s is F a The conditional probability, p, of g’s beigg l giVen l_ande£'is 1 (and given 5 alone, p is less than 1). This, Dretske tells us, gives us the signal’s "pp 52 informational content . . [the signal--l’s being §--] carries the information pl or pppgl g that it is l."23 Three points about Dretske’s notion of information are particularly important to his account. The first is his claim that carrying informational content is a natural, objective property of the physical world; the second is that it is a bona fide intentional property; the last is that having genuine cognitive content cannot simply be a matter of having intentionality per se. On the one hand, emphasizing what our belief states share with the intentional states of thermostats makes it less puz- zling how there could be a place for meaning in the natural order; on the other hand, drawing attention to differences between the intentional states 21KFI, p. vii. 22KFI, p. 65. My wording. I am indebted to H. Hendry for pointing out problems in the original formulation. 23KFI, p. 66. 101 of thermostats and our beliefs allows Dretske to respect the intuition that thermostats do not, in fact, have beliefs. Finally, the way in which he characterizes full-fledged cognitive content guarantees that it can (because it must) play a causal, and hence, an explanatory, role in our doings. Let us take these points in order, beginning with the intentionality of informational states. Informally, [I] says that l’s being 9 carries the information that g is 2 just in case l’s being g is, in the light of relevant background knowledge, a sure sign of g’s being 2. There is, however, more to the concept of information than that. Dretske takes it that [I] requires (or perhaps implies) that the relation between l’s being 9 and g’s being 5 be a lawful one—-one guaranteed either by a law of nature or a principle of logic. Mere correlation, even perfect correlation, is not, on Dretske’s account, enough to yield informational content. [E]ven. . . a perfectly general truth expressing an excep— tionless uniformity, is not sufficient for . . . transmitting information. . . . Even if the properties l and g are per- fectly correlated. . . this does not mean that there is information in g’s being 2 about g’s being g. . . .[Tlhe correlation between E and G may be sheerest coincidence, a correlation whose persistencezls not assured by any law of nature or principle of logic. Acccordingly, "even though ’F’ and ’G’ are true of exactly the same things (have the same extension), the information that g is l is different from the information that s is g."25 The property carries the information that 2['KFI, p. 74. 25KFI, p. 75. . is H .' crea salve veri .I’ Ih sional con I 1 ¢ s Just as t can be se content, from mod; AS Dret: nature It is (as it 102 . is thus intentional: the predicate ’carries the information that . . .’ creates a context in which we cannot substitute coreferring terms salve veritate (in the same way that the mental predicates ’believes that . . .’, ’hopes that. . .’, ’desires that. . .’ and so on create nonexten- sional contexts). [Insofar as] our definition of information distinguishes be- tween extensionally equivalent pieces of information . . . . statements describing the information carried by a signal are intensional [i.e., non—extensional] descriptions and the 26 phenomena they describe qualify as intentional phenomena. Just as the "uniqueness" of belief contents--their characteristic opacity-— can be seen to have its origin in the intentionality of informational content, the intentionality of informational content derives in its turn from modal properties of nomic and logical dependencies. The ultimate source of the intentionality inherent in the transmission and receipt of information is. . . the nomic regularities on which the transmission of information depends. . . . [Slince a lawful connection between two properties or magnitudes is itself an intentional phenomenon, information inherits its intentiog’l feature from the nomic regularities on which it depends. As Dretske points out, it is generally agreed that ’is a natural law that .’ creates a non-extensional context: the truth of ’It is a law of nature that all A are 8’ and ’All B are C’ does not underwrite the truth of ’It is a law of nature that all A are C’ but, merely, that of ’All A are (as it happens) C’. Explanatory geneaology stops here, however. We do not have an agreed upon 26KFI, p. 76. Stress suppressed. 27KFI, p. 77. 103 account of how natural laws themselves come by the property that distinguishes (1) from (1’). Dretske acknowledges that ”the ultimate clarification of our idea of information (in particular its intentional or semantic aspect) rests with a clarification of the modal authority of natural laws."28 But, he notes, "For our purposes it is not important where natural laws acquire this puzzling property. What is important is that they have it."29 Dretske’s second point is that, even having established that carrying in— formational content is both a natural pgp an intentional phenomenon, we still do not have a characterization of genuine semantic content--the content of beliefs, for example. First we shall see why this is so and then we shall see how Dretske proposes to ’manufacture’ semantic content from informational content. The problem is that our notion of semantic content is of something a good deal more focussed and selective-—in Dretske’s words, of something more "exclusive"-—than informational content turns out to be. We expect to be able to talk about lgg meaning of a sentence that Mary uttered or lgp belief that John thereby acquired, but, as Dretske observes, "Generally speaking, it makes little sense to speak of lgp informational content of a signal." [I]f a signal carries the information that g is l, and g’s being E carries, in turn, the information that g _g Q (or l ZBKFI, p. 79. 29 KFI, p. 78. Stress suppressed. 104 lg H), then this same signal [also] carries the information that s is p (or t is H). . . .[I]f r carries the information that s is a squaEe, then it also caEries the information that g is a rectangle, . . . a quadrilateral, a parallelogram, not a circle, not a pentagon, a square or a circle, and so on. . . [I]f there is a natural law to the effect that whenever , l is p, . . . then no signal can bear the message is l without also conveying the information that l is Thus, if an event (the utterance on an occasion of a sentence, say) carries as its informational content the proposition [L] ’Susan is in the library’, it plgp carries among its informational contents the logical consequences of [L], ’Susan is either in the library or at the beach,’ ’Either Susan is in the library or it snowed last July 4th’, and any number of other pro- positions that we do not think of as being any part of the meaning of any utterance that (on any occasion) gppgg ’Susan is in the library’. In much the same way, we would like the belief state whose content is that Susan is in the library to be a belief state different from the one whose content is that Susan is either in the library or at the beach. Moreover, since it is a law of nature that dead people do not move their vocal cords, the utter— ance ’Susan is in the library’ also carries among its informational contents the nomic consequence that the speaker is (at the very least) alive but it seems counterintuitive that this is part of its meaning.31 The BOKFI, p. 71. 31Examples of this sort make it apparent that Dretske’s notion of informational content not only fails to coincide with our concepts of meaning and belief content, as he himself suggests, but is pretty far from our pretheoretic intuitions with regard to ’carries information about . . .’, as well. Even if we accept that the utterance of ’Susan is in the library’ carries information that the speaker is alive, surely we are not prepared to say that it carries information about the weather, English literature, and so on. To the extent that our intuitions about intentional— (Footnote continued) proble format ina t feren< Real ' Drets m tells are i infor Chap alit beli 1:00 Firs of, meat infc 105 problem, as Dretske tells us, is that "Signals. . . are pregnant with in- 32 Informational contents come nomically or logically "nested" formation." in a way that semantic content does not. How are we to deal with this dif- ference between a naturalistic property with appealing credentials and the Real Thing? Dretske begins by analyzing the discrepancy in terms of what he calls plpplg of intentionality. A candidate for the role of belief content, he tells us, must be characterized by a higher order of intentionality than are informational structures. He goes on to argue that there is an information—theoretic (and thus an irreproachably naturalistic) way to characterize structures that will display just such high order intention- ality, structures that are, therefore, plausible "analogues" of genuine beliefs. Moreover, he urges that the capacity of some (but not all) systems to occupy states of this sort has, itself, a naturalistic explanation. First, then, for Dretske’s diagnosis in terms of degrees of intentionality of the peculiar uniqueness of beliefs. We have seen that beliefs and meanings are more finely individuated than are events construed merely as information-carrying structures.33 Any structure with the informational content that l is F must also be a structure with the informational content 31 (continued) ity are intuitions about aboutness, Dretske’s characterization of ’carries the information that. . .’ does not seem to square with everyday intuitions about intentionality. 32KFI, p. 73. Dretske pursues the biological metaphor with a vengeance here and below. that i is to being 3 lief with consequen our hands in the gr sitional they wil' acceptin ence is, same in “‘4‘ a Only th< Vhose r. worlds_ StruCtu more Se intenti Corefer Same r£ tences all Do; intEnt 106 that l is E if being g stands in a relation of nomic or logical consequence to being l. By contrast, a pgllpl with the content that l is l is not a be- lief with the content that l is 9 even if being g is a nomic or logical consequence of being 3: we have, we think, two quite distinct beliefs on our hands. Dretske suggests that it is useful to think of this difference in the grain of individuation as corresponding to a difference among propo- sitional contexts with regard to the breadth of the substitution classes they will tolerate. Purely extensional contexts are the most permissive, accepting (salve veritate) the intersubstitution of any terms whose refer- ence is, as a matter of fact, the same--i.e., terms whose reference is the same in the actual world. By comparison, contexts with what Dretske calls first order intentionality are choosier, accepting as intersubstitutable only those terms whose coreference is a matter of natural law—-i.e., terms whose reference is (at minimum) the same in all nomically accessible worlds. Dretske notes that the intentionality of information-carrying structures is of this minimal, first order sort. There are, however, still more selective contexts--those with what Dretske calls second order intentionality--where intersubstitution salve veritate requires that coreference be a matter of logical necessity, i.e., that terms have the same reference in all possible worlds. If we think of the reference of sen— tences as their truth values, this means that sentences true (or false) at all possible worlds are intersubstitutable in contexts of second order intentionality. But we know that there are contexts in which even this is 33From here on in, I shall just mention beliefs. Dretske thinks the two are closely related but it gets tiresome to keep saying ’beliefs and meanings'. 107 inadmissable. That I believe the logical truth that either it is raining in New York or it is not raining in New York does not guarantee that I believe the logical truth that if the square root of two is a rational number then Oliver North has been up front with us, despite the fact that each of the exchanged sentences is true at all possible worlds and hence true at exactly the same possible worlds, which is to say that their reference is the same at all possible worlds. Contexts such that even coreference by logical necessity does not license substitution salve veritate are, in Dretske’s terminology, marked by third order intentionality. The ability to discriminate among logical truths and falsities, to reject substitutions of terms with necessarily identical referents, is a hallmark of belief con- tent, Dretske tells us. To have third order intentionality is thus to have content with the uniqueness characteristic of beliefs. The naturalizing theorist’s task is now to understand how higher—order intentional structures can be manufactured out of lower-order intentional states. . . . to describe the way structures having a semantic content. . . can be developed out of [merely] information-bearing struc— tures. . . . Such an account is needed in order to . clarify the difference between simple information—processing mechanisms (dictaphones, television sets, voltmeters) and genuine cognitive systems (frogs, humans, and perhaps some computers), and. . . to reveal the underlying naturalistic basis of that cluster of mentalistic attribgles that assign semantic contents to one’s internal states. How does Dretske propose to achieve this goal? The first step is conceptual -—provision of an information-theoretic formulation of the notion of seman— tic content. As we have seen, asking for ’the informational content of 8’ yields promiscuously nested sets of (nomically or logically) related 34KFI, p. 175. 108 propositions, incapable of achieving, on that account, the higher order intentionality characteristic of belief (or other cognitive) content. We can, however, identify content more narrowly by specifying the relation of a proposition to other propositional contents in a given nested set—-the relative position or location of a particular propositional content in that set. In particular, Dretske notes, we may focus attention on that proposi- tional content carried by S within which all other information carried by S is nested (either nomically or logically)--the outermost "shell" of S’s informational content. Dretske identifies such information with the semantic content of an information-carrying structure. [S] Structure § has the fact that l is l as its semantic content = (a) S carries the information that l is l and (b) S carries no other piece of information, 5 is g, which is such that the information that l is l isBBested (nomically or analytically) in 5’s being g. Structures individuated with regard to their semantic content are called semantic structures.36 Dretske suggests that semantic structures "have the same order of intentionality as beliefs and. . . . are [therefore] the 37 ideal information-theoretic analogs of belief." He notes that, unlike in- formational contents, semantic contents (and thus semantic structures) are 35KFI, p. 184. Dretske calls information encoded in this way "completely digitalized" information (in contrast with information nested within other content, which he describes as being carried in "analog" form). He makes no use of this terminology outside of KFI, however. 36KFI, p. 179. 37KFI, p. 179. Stress suppressed. 109 distinct in ways that closely resemble the ways in which belief contents (and thus beliefs) are individuated. If S carries the information that tis F, and the information that t is G is nested in l’s being£ F, then. . . S [also] car— ries the information that l is G. . .[But] 8 has the first as its semantic content without having the second as its semantic content. . .because S carries the information that l is F in digital form [1. e. , in its outermost shell; see fn 35] wléhout carrying the information that l is G in digital form. Contents thus defined have third order intentionality: the propositional context created by ’has the semantic content. . .’, like that created by ’believes. . .’, does not accept the intersubstitution salve veritate of terms whose reference is the same even in all possible worlds. From the truth of ’S has the semantic content that l is l’ and ’It is logically (nomically) necessary that l’s be Q’s’ we are not constrained to conclude to the truth of ’S has the semantic content that l is Q’.39 Dretske notes that this characterization does not distinguish between logically or nomically equivalent semantic contents--contents that are not asymmetrically "nested" relative to each other. He suggests, however, that cognitive structures can be individuated more finely than can their seman- tic contents. Since it is the content of a cognitive structure and not semantic content per se that corresponds to belief content, this would 38KFI,p. 178. Stress added. 39Indeed, except in the case where F’s and G’s are nomological or logical equivalents, a set of sentences of this form will be inconsistent. If l’s being G is a (nomic or logical) consequence of t’ s being F but not vice- versa, then the information that l is G is nested inside the information that l is l, in which case it cannot qualify as semantic content. 110 enable us to explain how an individual might, for example, have a cognitive structure, C, whose semantic content was P, while failing to have a cogni- tive structure, C’, whose semantic content was also P--e.g., have the belief that two is even but not the belief that the smallest prime is even 0 40 At this point, Dretske has wrung from his purely information—theoretic ap- proach all that it can yield by way of naturalizing the notion of belief. His definitions ensure that semantic content is determinate in a way that belief contents are and informational contents are not. He is, however, left with three pieces of unfinished business. First, it is not obvious that semantic content, in and of itself, can provide the causal explanation of a system’s doings that Dretske requires of belief contents. Second, since semantic content is informational content, it is unclear how it could ever be false. Finally, we do not yet know what natural processes might "manufacture" beliefs (or even semantic structures) out of merely informa- tional structures. I shall, in what follows, try to sort out what Dretske has to say about each of these matters. It will be helpful to keep in mind 40Dretske suggests, in particular, that we take compositionality into account when individuating cognitive structures. Thus, the cognitive structure whose semantic content is ’x is water’ is not identical with the cognitive structure whose semantic content is the nomologically equivalent ’x is H 0’ because the latter, but not the former, has the concepts ’H’ and ’O’ as constituents. Nomically or logically equivalent "primitive" concepts yield identical cognitive structures, however. "[Tlo try to. . . distinguish primitive equivalent concepts. . . would be to pursue a level of intentionality beyond that to be found in cognitive structures as ordinarily understood," i.e., it would lead us to distinguish among beliefs that are ordinarily taken to be type identical (KFI, p. 217). 111 that the treatment they receive in KFI is, on Dretske’s own account, preliminary and incomplete. Let us begin with the question of the explanatory role of belief content. Dretske’s way with this is brisk: he uses the notion of semantic content to define a new sort of structure, one that has the required causal powers. A semantic structure qualifies as a cognitive structure (and therefore. . . as a belief) insofar as its semantic content is a capgal determinant of output in the system in which it occurs. Dretske invokes his canon of explanatory adequacy to defend this stipula— tion. It is only when a structure's semantic content "defines the structure’s causal influence on output," he tells us, that we can say that "the system does 5 because it occupies an internal state with the content that g is l--because. . . it believes (or knows) that g is l."42 Assuming that we wish to make use of everyday psychological explanation, we must find a way to endow the naturalistic analogues of belief contents with causal powers; the definition of cognitive structure seems to do the trick. We may note, however, that this move postpones rather than solves the problem of giving causal powers to belief content. If cognitive content is, as Dretske would have it here, a special kind of semantic content, and if 4lrrr, p. 199. 42KFI, p. 199. By contrast, other semantic structures are not cognitive structures. [Olur sensory experience, the experience associated (say) with seeing a truck pass by, does not qualify as a cognitive structure. It has a semantic content (expressible by a com- plex sentence describing all the information carried in the visual experience) but this semantic content exercises no control over output (KFI, p. 201). 112 semantic content is a special kind of informational content, then, if cognitive content is to play a causal role, at least some informational content must also be able to do so. As Dretske has put it recently, [If information] can be provided a causal job to do, if it can be put to work, then the causal role of meaning 329 belief. . . being a derivative, will fall into place. A closer look at the way Dretske treats the notion of semantic content re— veals, indeed, an unargued assumption that information ppp information must play a causal role--must "matter"--to any system capable of manifesting such content. Let us take a moment to see how this happens. It will be recalled that according to Dretske, the semantic content of a structure is, by definition, simply its strongest propositional content. Semantic content is a matter of what an informational structure is "mute" about. In discussing such content, however, Dretske draws an additional distinction between the deletion of information by a system and the mere loss of information in a system. That a cheap tape recorder loses lots of information between input and output does not, on his account, make it a candidate for having semantic content; informational loss of this sort is "indiscriminate and unselective with respect to the information contained in the arriving signal."44 By contrast, what makes possible a state whose l“31>1w, p. 6. AAKFI, p. 260, n. Similarly, the fact that information is lost between receptors and sensory store does not, in and of itself, endow iconic memory, say, with semantic content. By contrast, the selective process that winnows the contents of sensory store preliminary to its storage in short-term memory is informationally sensitive and this allows us to say that short-term memory (Footnote continued) 113 semantic content is ’This is an apple’ is a systematic stripping away of components of information. . [e.g.,] size, color, orientation, surroundings. . . lg order lp feature. . . pgp component ofélhis information--the information that it is an apple. To have the sort of content characteristic of belief states--to have semantic structures-- a system must somehow discriminate among the various pieces of information embodied in a physical structure and select one. . . as the content of that higher ordgg intentional state. . . to be identified as the belief. To carry semantic content, then, the system must be able to respond to informational contents ppp informational contents--otherwise the develop- ment of its semantic structures can hardly be described in terms of discrimination among pieces pl information or selection pl p particular piece pl information to be deleted. Trimming a manuscript in the course of rebinding it may result in loss of the last line of text and thus in loss of the information carried in the last line of text but this loss does not constitute a deletion (or a discarding or a selection) pl this information unless the best explanation of the lines’ excision adverts to their content (rather than merely to, e.g., their unfortunate position on the page). The difference between systems capable of occupying states of higher order intentionality and those not so capable thus turns out, somewhat unexpec- tedly, to be not just a difference in the information a structure carries 44(continued) has semantic content (KFI,p. 147). 5Precis, p. 183. First and second stress added. 46KFI, p. 174, stress suppressed and stress added. 114 or fails to carry, but, rather, a matter of whether a system has the capacity to respond to informational content ppp informational content. We are, however, given no argument that a system must have such a capacity before it can meet the information theoretic definition of semantic structure--that a system cannot have structures that have an outermost shell of informational content unless the system has, in some sense, developed such structures because pl their informational content. Cognitive structures--semantic structures whose content plays a causal role in a system’s doings--are Dretske’s candidates for beliefs understood naturalistically. They still lack an essential property of beliefs, however: unlike semantic structures, beliefs are capable of having false content. As we have seen earlier, any proponent of "indicator semantics"47 is faced with the puzzle of how content that is determined by how things plp can manage to represent things as they are gpl. Dretske’s response is to make a distinction between cognitive structure types and their tokens. He calls our attention to the semantic relation between conventional symbol types and their tokens. A conventional symbol token, he tells us, "inherits 48 My utterance of its meaning from the symbol type of which it is a token." ’There’s tea in the pot’, for example, represents the pot as having tea in it by virtue of the meaning of the symbol types I token even if (because the pot is, in fact, empty) their tokening does not carry the information 47The phrase is Barry Loewer’s, "From Information to Intentionality", Synthese 29, (1987). Hereafter, II. 48KFI, p. 192. 115 that there is tea in the pot. Because we can drive a wedge between what I say about things and how things are-—between the meaning of my utterance and its truth--my tokening of an English sentence can misrepresent the teapot’s state. Similarly, Dretske urges, tokens of cognitive structure types are entitled to lay claim, by descent, as it were, to a representa- tional content that is independent of their informational content. A token of the cognitive structure type whose content is ’g is E’ can (under cer- tain circumstances) say of some individual, p, that it is l, purely on the strength of its semantic lineage, regardless of whether p is, in fact, l. Cognitive structure types correspond to concepts--e.g., to the concept ex- pressed by the open sentence ’g is l’. Tokenings of cognitive structures correspond to beliefs--e.g., to the belief that p is f. If p is not B, a tokening of the cognitive structure that we identify with the concept of F-ness misrepresents p as being E and we have a false belief.49 The question now arises, What determines the concept identified with a cog- nitive structure type, and thus, the belief identified with its tokening? Dretske draws a further parallel between conventional symbols and semantic structures. Both, he tells us, acquire their content by virtue of the in- formation that it is their function to carry. It is because they are tokens agDretske is talking, throughout, about pp lg beliefs and primitive con- cepts. It is thus the tokening of a cognitive structure caused in the relevant way by an individual, p, that counts as a belief about p. If beliefs are tokenings of cognitive structures, what of non-occurrent beliefs, like Hal’s belief that any gumdrop is smaller than some mountain? Are we going to have dispositions to token cognitive structures? On what occasions, do you suppose? Is this going to get us in the same trouble as Fodor’s dispositions? 116 of a cartographic symbol whose jpp it is to carry information about highway whereabouts that double red lines on a map represent the highway as going from A to B. In the case of conventional symbols, function (and, thereby, meaning) is conferred by arbitrary assignment: mapmakers could just as well have asigned the task of carrying information about highway location to double green lines or dotted black ones. This does not, however, answer the question with which the naturalizing semanticist must be concerned: "Who or what assigns meanings, or information—carrying roles" to "those neurologi- cal structures we mean to identify with the beliefs of living organisms?"50 To discover what information a cognitive structure has the function of carrying, Dretske tells us, we must inquire into its origins. That a system believes something depends, partially, on the effects (on system output) of these internal states, since to qualify for cognitive content an internal structure must have executive responsibilities. But the content [of the belief] is determined solely py the structure’s origin--by its information heritage. The content of a cognitive structure is determined by "what information 52 . was instrumental in the formation of the structure," what 53 What concept we say an organism has information "crystallized" it. acquired depends on "what information we believe was instrumental in the formation of the relevant internal structures,"54 what information was SOKFI, p. 193. 51KFI, pp. 201-2. Last emphasis added. SZKFI, p. 193. 53xr1, p. 194. 54KFI, p. 195. 117 responsible for "shaping. . . [the system’s] discriminatory and identificatory responses."55 If we identify beliefs with. . . tokens of these abstract semantic structures [i.e., with cognitive structure types], we solve the problem with which . . . [we] began, . . . how (in terms of informational structures) to account for the possible falsity of beliefs. . . . [The] problem is resolved . . by realizing that a type of structure. . . may have in- formational origins (in the sense that that type of structure developed as a system’s way of encoding certain information) without (subsequent) instancesspf that structure having similar informational origins. In the next chapter, we shall look more closely at Dretske’s proposal that we identify the content of a cognitive structure with the information whose provision explains its genesis and the closely related claim that (at least in the case of certain biological systems) the best explanation of how an information-carrying structure acquired its causal role in a system will advert to advantages to the system of utilizing the information it provided to regulate transactions with its environment. In conjunction, these claims give us Dretske’s doctrine of functional meaning. Here, we may simply note foreshadowings of this theme. We have, I believe, Dretske’s permission to do so: in a later paper, to which we shall be turning shortly, he tells us that in KFI he had defined the semantic content of a structure as "the information it was developed to carry (hence, acquired the function of carrying)."57 Here again, it appears that the formal definition of semantic content in KFI fails to deliver precisely what Dretske already had in mind. SSKFI, p. 194. 56rr1, pp. 196-197. 57"Misrepresentation", in Belief: Form, Content and Function, ed. Radu J. iBogdan (Oxford: Clarendon Press, 1986), p. 28. Hereafter, Misrep. 118 Even at this stage, semantic content is not just the information a struc- ture happens to carry as its strongest propositional content, but, rather, the information that a structure was developed so to carry: a semantic content is information that a structure carries as its strongest proposi- tional content because doing so is (or has been) of benefit to the system. I think the text supports this interpretation. We may, for example, notice locutions like "systematic stripping away of components of information. lg order lp feature. . . one component,"58 "a system must. . . select one [piece of information] as the content of that higher order intentional state";59 we hear Dretske say that "The essence of conceptualization is loss of. . . excess information."60 Discussing the content of innate concepts, he tells us that "[Tlhe question becomes: what information did natural selection design that structure to carry in completely digital form [i.e., as its outermost informational shelll?"61 [A] system’s internal states derive their content. . . from their information—carrying role-—from the sort of situation they were developed (or prewired) to represent. . . Internal states derive their content (in the first instance at least) from their informational origin, not from their effects, although (from an evolutionary standpoint) it is hard to see why or how structures of this type could develop unless they had some beneficial. . . effect on behavior. 58Precis, p. 183, stress added. 59KFI, p. 174, stress added. 6oPrecis, p. 183, stress added. 61KFI, p. 234, stress added. 62KFI, p. 209. It is hard semantic c cmnent th have it. I not involv content of sible for structure be coexte We haVe 5 SYStems 1 Semantic SYStems, argumen t strUCIUr a bearin With a I It has i 119 It is hard not to interpret this as (at the very least) a suggestion that semantic content is (not only the strongest propositional content but) content that structures have because it is useful to the system for them to have it. I cannot think of any way of reading "developed lp. . ." that does not involve a goal state. (There is a difference between saying that the content of a structure is determined by the information that was respon- sible for its development and saying that it is the information that the structure was developed to carry. In an evolutionarly context the two will be coextensive but the descriptions are not synonymous.) We have seen signs that Dretske takes semantic content as committing us to systems to which information matters as information. I think he also sees semantic content as committing us to information that matters to systems as systems. Just as we were given no argument for the first, we are given no argument for the second--that is, for the claim that no system can have structures that encode information in an outside shell unless doing so has a bearing on the fate of the system. Together, the two claims are consonant with a recent statement that cognitive processes are processes whose job it is to coordi- nate behavior. . . with the [generally external] conditions . . . on which satisfaclion of6geeds and desires depends, on which benefits are contingent. It has been suggested that Dretske’s current and explicitly teleological 64 account of misrepresentation deviates significantly from that of KFI. It 63WTH, p. 1. 64II, p. 316n. is my impr emphasis a One piece manufactu discrimir iS, I th: the dire in great 1aSt sec 120 is my impression that the differences are, to a large extent, those of emphasis and theoretical vocabulary. One piece of drastically unfinished business remains: how is mind manufactured from information in real time? Dretske alludes in KFI both to discrimination learning and to evolutionary processes, but the discussion is, I think, too fragmentary to hear much weight. It does, however, show us the direction he plans to take. Let us turn to later work, which presents in greater detail the doctrine of functional meaning roughly limned in the last sections of KFI. "Misrep Dretske natural constit meaning natural tinctiv falsity with th Possibl SurrOUn There c and bui “hich w fOr eXa watches Changin 1\ 2 . ”lsrep 3 . ”lsrep Chapter 5 Critical Comments on Dretske’s Doctrine "Misrepresentation" is the most recently published portion of Fred Dretske’s project of explicating "the baffling place of mind. . . in the natural order of things."1 Although, in Dretske’s view, misrepresentation constitutes "perhaps only a small part . . . of the general problem of meaning or intentionality,"2 it is a critical part. If we cannot give a naturalistic account of belief content, we have not domesticated the dis- tinctively mental; unless what we explain naturalistically is capable of falsity or error, it cannot be belief content. Dretske therefore begins with the question, "How do we manage to get things ylpgg? How is it possible for physical systems to misrepresent the state of their 3 surroundings?" There can be little doubt that states of some systems-~those that we design and build—-have the capacity to misrepresent by virtue of meanings with which we, their minded users, have endowed them. The hands on my watch can, for example, misrepresent the hour as 1:37 when it is really 2:48 because watches of this sort have been designed to tell time by means of the changing position of their hands. That my watch’s unbudging hands 1Precis, p. 169. 2Misrep., p. 17. 3Misrep., p. 17. First stress added. 121 misreprese cerned to could just of the mor is easy. cal syste which we "Once we intentiOt stance?" take," it biOngi C In phat "real 1 respons. fOr mis terms. SYstemS .\ HiSrep 5 Migrep 6 HiSrEp 122 misrepresent is clearly relative to my goals and interests. Had I been con- cerned to solve a murder, say, rather than to arrive in class on time, I could just as well have interpreted their position as a correct indication of the moment when the watch stopped. Explaining how artifacts say wrongly is easy. The hard part is to explain how we ourselves, construed as physi- cal systems without a Designer or Interpreter, come by that intentionality which we can, when it suits us, "lavish. . . on the systems we describe."4 "Once we have intentionality," Dretske observes, "we can . . . adopt the intentional stance. But what. . . gives us. . . the power to adopt this stance?"5 Until we can give an account of "nature’s way of making a mis- take," until we can identify "the place where the misrepresentational buck stops," Dretske tells us, "we do not have a clue how naturally-evolving biological systems could have acquired the capacity for belief."6 In what follows, I shall be arguing that despite his unswerving reliance on "real life" examples--fuel gauges, bacteria and classically conditioned responses--Dretske falls short of providing nomically possible conditions for misrepresentation on the part of systems construed in purely physical terms. At best, I shall claim, he shows that misrepresentation by such systems is logically possible. Given his goal of finding a place for mind in the natural order, this is not enough. 4Misrep.., p. 17. 5Misrep., p. 17. 6Misrep., p. 17. "Misrep light 0: thing 1: actually determir terms of fore, ea semantic relation 0f neste "Misrepr again, 1 Changes Cit in K their to the role c0ntent 123 "Misrepresentation" is, I think, continuous with and best understood in the light of Dretske’s earlier work. Once more, we encounter tokens with some— thing like informational content that depends upon the circumstances that actually occasion them, and a (distinct) semantically evaluable content determined by their type. Again, Dretske explicates misrepresentation in terms of possible discrepancy between these two kinds of content. As be- fore, each is taken to be naturalistically characterizable, with the semantically evaluable content of a symbol established by an "objective" relationship between its tokens and the world. Just as the non—specificity of nested informational contents created difficulties in KFI, so, in "Misrepresentation", does the indeterminacy of functional meaning, and once again, it is a system’s capacity for learning that saves the day. Two changes are worth noting. First, the functional orientation that was impli- cit in KFI is now made explicit: where, formerly, what types bequeathed to their tokens was conceptual content, it is now functional meaning. Second, the role played earlier by Dretske’s proprietary concept of informational content is performed in this essay by what he will be calling ’natural meaning’. Let me sketch the general shape of Dretske’s newest proposal before taking it up in critical detail. Dretske’s overall strategy is to explain the pos- sibility of misrepresentation on the part of physical systems by displaying instances of such systems that meet what he takes to be relevant criteria for error and then inferring from their actuality to their nomic possibil- ity.7 His first step is to establish that there are particular 124 circumstances under which we are perfectly willing to say of physical systems that they misrepresent the state of their surroundings. He then generalizes to what would seem to be sufficient conditions for such attri— bution. Roughly, he tells us, a tokening of a state of a physical system may be said to misrepresent when it fails to convey the information that tokens of that type are normally supposed to convey. Next, Dretske argues that ’supposed to convey’ can be applied without a presupposition of inten- tionality. A tokening of a state will be said to have natural functional meaning if we can tell an explanatory story in the language of physical and biological science about how the information normally conveyed by the state led to its current causal role in the system’s doings. Dretske then displays instances of actual systems to whose states we seem willing to attribute such natural functional meaning. Finally, Dretske tackles the task of convincing us that there are systems such that (1) their states have such natural functional meanings pgp (2) these natural functional meanings can deviate from the informational content of (at least some of) their tokenings. If there are such systems, then there are physical systems capable of underived misrepresentation. If there are physical systems capable of underived misrepresentation, then we know a way in which there could be such systems in a world whose laws are those of our world. Dretske’s candidates turn out to be biological systems capable of "associative learning."8 I shall be arguing that Dretske has problems every 7ICS, p. 282. He has described an earlier paper in similar terms, as "an attempt to sketch. . . an explanation for how purely physical systems could (because even the simplest mechanical systems pp) occupy intentional states of the appropriate kind" (ICS, p. 282). inch of th What, ther Dretske b natural m Just as actua113 only it in Unde iSay] f that w DretskE tween ‘ 125 inch of the way, most crucially with the last step. What, then, does Dretske’s case look like in detail? Dretske begins by introducing the central notions of natural signs and natural meaningg, modelled, he tells us, on Grice’s use of the terms.9 Natural signs are. . . more or less reliable indicators, and what they mean is what they indicate to be so. The power of a natural sign to mean something. . . is underwritten by. . objective constraints. . . . In most cases this relation is causal or lawful. . . . Sometimes there are merely regular- ities, non—lawful but none the less pervasive, that lp secure the connection between sign and significance. Just as in KFI the informational content of a token depends on how things actually are, so here, an event has the natural meaning--meansn--that 2 only if g. "The project," Dretske tells us, "is to see how far one can go in understanding glgrepresentation, the power of a . . . state l to . [say] falsely that y is E. . ., in terms of a natural sign’s meaningln] that y is g."11 Dretske’s first step is to show that we are willing to apply ’misrepresentation’ to physical systems on the basis of a discrepancy be— tween what tokenings of their states meann and the representational content 8Misrep., p. 35. 9Dretske cites P. Grice, "Meaning", Philosophical Review, 66 (1957), 377-88, on p. 20 of "Misrepresentations". 10Misrep., pp. 18-19. 11Misrep., p. 19. First stress added. 126 12 Before they carry by virtue of "identifiable functions" of such states. we take up this claim, it may be appropriate to pause and worry a bit about Dretske’s doctrine of natural signs. Pretty clearly, meaningn is supposed to play a role analogous to that of informational content in KFI. This means, first, that it is intended to supply the ’objective’ raw material out of which semantically evaluable representation can be constructed. Second, we may recall that informational content seemed to Dretske a par- ticularly promising kind of raw material because (among other things) it had built in, so to speak, a basis for understanding the intentionality of cognitive states. We may, therefore, want to ask how well meaningn meets these same demands. I shall be arguing that it is not successful on either score. Urging this point will require a preliminary detour in which we explore Dretske’s doctrine with regard to these new terms. How, for example, does Dretske think that natural meaning differs--if at all-—from informational content? We are left largely on our own here. In order to indicate the problems we face in disentangling the two concepts, let me quote (at some length) from the slightly earlier "Machines and the Mental", in which ’sign’, ’natural meaning’ and ’information' occur side by side for the first time. Discussing what would be required for a computer’s symbols to possess a non—derived semantic content, Dretske says that we would have to put the computer into. . . a larger system that has the kind of sensory capabilities. . . that enable what goes on inside the computer to mean something, in Paul Grice’s natural sense of meaning, about what goes on outside the computer. The ele- 12Misrep., p. 21. 127 ments over which the computer performs its operations will then have a meaning that is independent of the conventions of its users. They will then mean something in the same way the swing of a galvanometer needle means something regarding the electrical activity in the circuit to which it is connected, the way expanding mercury means something about the sur- rounding temperaturp3 . . . This kind of meaning is sometimes called information. This last sentence is followed by a reference to KFI, as is Dretske's brief discussion of natural meaning in "Misrepresentation", where we are told that we will find "a fuller account of it" in Chapters 1 and 2 of that book.14 Similarly, Dretske talks about "information, lppl information, the kind of meaning associated with natural signs," and "information, intrinsic natural meaning," and notes that even though we may playfully assign an ar- bitrary meaning to, say, the sound of a smoke alarm, "[Tlhat isn’t actually what the sound means. That isn’t what it is a sign of, not the information it carries."15 By repeatedly putting ’natural meaning’ and ’information’ in apposition, Dretske seems to suggest that the two terms may be used inter— changeably. Clearly, however, this is not the case, even if we look only at Dretkse’s own use of ’natural sign’ and ’natural meaning’ (leaving aside for the moment any question of what Grice may have had in mind), we notice an important distinction: information, according to Dretske, requires strictly lawful relations—-nomic or logical--between event types, whereas 13 Philosophical Association (Sept. 1985) vol. 59, no. 1, p. 28. Hereafter, M&M.) 1'The examples are presented in KFI. There is only the briefest allusion to Gricean doctrine, however. lsnsn, p. 29. 128 natural meaning does not. He is blessedly clear on both counts. Thus, in KFI, we have the following. Correlations, even pervasive correlations, are gpl to be confused with informational relations. Even if the properties l and G are perfectly correlated. . . this does not mean that there is information in g’s being 5 about g’s being 9. . [Tlhe correlation between E and 9 may be the sheerest coin- cidence. . . whose persistencp6is not assured by any law of nature or principle of logic. Similarly, in response to inquiry, he has said that "information can be de- fined in terms of conditional probabilities only in so far as the latter are understood as expressions of some form of nomic dependence," adding that he personally understands conditional probability to be "a nomic concept."17 In contrast, being a natural sign of some state of affairs l is, on Dretske’s account, merely a matter of being, locally and for the time being, "a reliable indicator" of 2. Thus, for example, It is partly the fact,. . . not itself lawful, that. . . squirrels or woodpeckers. . . do not regularly ring doorbells while foraging for food that makes the ringing bell mean n that someone. . . is at the door. If squirrels changed their habits (because, say, doorbells were made out of nuts), then a ringlng doorbell would no longer meanln] what it now does. Dretske points out that the "regularity" involved here is "not fully lawful." "Nonetheless," he says, "the doorbell retains its natural meaning 19 as long as this regularity persists." Similarly, in "Machines and the Mental", natural meaning is equated with "the kind of meaning we associate 16KFI, pp. 73—74. 17Personal communication to R. Hall. 18Misrep., p. 19. 19Misrep., p. 19. 129 with reliable signs and trustworthy indicators, the kind of meaning pos- sessed by dark clouds, shadows, prints, leaf patterns, smoke. . .".20 But, as Dretske himself observes (in "Misrepresentation"), there can be smoke without fire, and, as every gardener discovers, dark clouds can be tokened without rain being instantiated. That there is a difference between infor- mational content and natural meaning is equally clear when we look at Dretske’s analyses of each in terms of counterfactuals. The "lawful dependence" demanded for information is, he tells us, "a dependence that 21 The precludes the occurrence of one [condition] without the other." doorbell’s ringing qualifies as a natural sign, however, by virtue of sat- isfying a considerably weaker condition: "as things gpy stand, we can (usually) say that the hell would not be ringing unless someone was at the door."22 Where does this leave us? At the very least, it appears that we shall need to distinguish between informational content and natural meaning while remembering that Dretske himself has not done so very systematically. Moreover, if meaningn is not simply informational content under a different 20M&M, p. 28. If we want to be sticky about it, we may note that Dretske has, in the immediately preceding sentence, equated natural meaning with information. Given that equivalence is transitive, carrying the information that l is now equated with being a reliable but not necessarily infallible sign that l, which flatly contradicts the requirement that the conditional probability of one event type given the other be 1 for the former to carry the informational content that the latter holds. 21"Author’s Response", lgp Behavioral pgp Brain Sciences (1983) vol. 6, 1, p. 82. My stress. 22 Misrep., p. 19. 130 name, we shall also need to inquire a bit more closely into its nature. How, for example, is meaningn lgpl 3 related to being a natural sign of 2? Once again, Dretske is not very helpful. On the one hand, natural signs seem most reasonably construed as event lyppg--sign and significance would seem linked in that tokenings of the sign are (in some way or other) reli- ably associated with instances of the property it signifies. On the other hand, Dretske insists that "in speaking of signs and their natural meaning I should always be understood as referring to particular events, states or 23 conditions: lglg track, lgpgp clouds, and lgpl smoke." If natural signs are particular events, however, it is hard to see how there could be a reliable relationship between any pgp such event and anything else. We might do better to say that natural meaning—-meaningn--inheres only in individual events, just as information is carried only by particular events. A necessary condition for an event’s meaningn l, however, is that there be, generally, a reliable association between tokenings of events of that type and instances of 2. That is to say, for a particular event to meann lgpl l the event type must be a natural sign of g. This would parallel neatly the earlier condition that a tokening of a state carries the informational content lgpl 2 only if there is a lawful dependency between tokening of that type and instances of g. In what follows, I shall be reading Dretske as committed to something along these lines.24 23Misrep., p. 20. 24I suspect that this is a somewhat strenuous reconstruction, however. There are several places where Dretske seems to be using ’meansn' and ’is a natural sign of’ interchangeably. (Footnote continued) 131 Having done something toward fixing ideas, we may now return to the ques- tion with which we began: how successfully does Dretske’s new conceptual apparatus help us cope with the task of providing a naturalized account of intentionality? In particular, does it provide adequately ’objective’ raw material from which to construct intentionality? Does it give us a foothold in the straightforwardly physical from which we may attain to nonextension- ality? Let us take these issues up one at a time. Dretske insists, in several places, on the objectivity of natural signs. Naturally—occurring signs mean something," he assures us, "and they do so without any assistance from us." 24 (continued) Grice has next to nothing to say about natural signs. He mentions them only once, within quotation marks, as pretheoretic notions to which he gives short shrift. "What more can be said about the distinction between the cases where we should say that the word [’means’] is applied in a natural sense and the cases where we should say that the word is applied in an nonnatural sense?". . . This question . . is, I think, what people are getting at when they dis- play an interest in a distinction between "natural" and "conventional" signs. But I think my formulation is better. (Meaning, p. 438.) Going to Grice to find out what Dretske intends by ’natural sign’ turns out to be about as helpful as going to Shannon and Weaver to find out what he means by ’information’. Dretske appropriates Grice’ 3 "Those spots mean measles" as an example of meaningn and I take it that he would agree equally with Grice’ s remark, apropos of "The recent budget means that we shall have a hard year", that "some things which mean naturally are not signs of what they mean NN ." On the other hand, would Dretske accept equally as meaning in the atural sense "Those spots didn’t mean anything to me, but to the doctor they meant measles"? Given his emphasis on the 'objectivity’ of natural meaning, the answer is not entirely obvious. 132 [A] northerly-flowing river means that there is a downward gradient in that direction. Shadows to the east mean that the sun is in the west. . . . The power of these events. . . to mean what they do is independent of the way we interpret them —-or25indeed, of whether we interpret or recognize them at all. From Dretske’s perspective, we can settle for nothing less if we want a non-circular account of intentionality. But it is hard to see how meaningn can attain to this objective status. If we have understood Dretske cor- rectly, an event meansn that B only if it is a token of a type that is a natural sign of 2. But what, precisely, is required for event type X to be a natural sign of event type Y? How "reliable" does a "reliable" indicator have to be? It will be recalled that as presented in KFI, a crucial point about information was that §(l) carries the information that l(y) only if the conditional probability that y is l given that l is p is equal to 1. Dretske is adamant that he will not accept less; it seems equally clear that he is not demanding more. That is, it does not matter with regard to informational content whether or how often l fails to be 9 on occasions 6 Shall we say, similarly, that §(l) is a reliable when y is, in fact, {.2 indicator, and thus, a natural sign that l(y) under the same circumstances? The condition appears to be both too strong and too weak to meet Dretske’s needs. If the reading suggested above is correct, he wants dark clouds to 25Misrep., p. 18. 26Cf. Dretske's comment with regard to requirements for knowledge: "I know someone is at my door when the bell rings, even ll the bell sometimes mal- functions and doesn’t ring (causing me to have the false belief that no one is at the door). The decisive question is whether the bell sometimes rings when no one is at the door." (Response, p. 86.) This is relevant, I think, because Dretske sees an intimate connection between information and knowledge. 133 count as natural signs of rain, smoke of fire and spots of measles even though the relevant conditional probabilities are, in each case, less than 1. At the same time, if the doorbell fails to ring when someone is actually at the door in any appreciable number of cases, we are unlikely to regard it as a reliable indicator of there being someone there, even if someone is always there whenever it happens to ring. We are surely not likely to say that we have a reliable indicator of pregnancy in the case of a test that regularly yields false negatives even if it never gives us a false posi- tive. Nor is this merely a quibble about how we ordinarily use words. Dretske wants the "correlation” that constitutes natural meaning to play a causal role in symbol processing and the organization of behavior. The correlation between a ringing bell and someone’s presence at the door, the kind of correlation that confers on the ringing bell the meaning that someone is at the door, changes the way a (suitably exposed) nervous system processes the internal sign of a ringing bell. . . . [Tlhe very same phenomenon can be illustrated at every biological level. at which learning occurs. It is. . . an instance of what learning theorists describe as the contingencies modifying the way a system processes, and hence rggponds to, the internal signs for stimulus conditions. That content can have causal impact is crucial to Dretske’s realism about cognitive states. But if the relation of reliable indication is to be equated with the sort of correlation that underlies learning, it cannot be construed along the lines of the informational relation as Dretske defines it in KFI. In the case of classical conditioning, for example, it is not enough that the conditioned stimulus (bell, buzzer, light) never occur without the unconditioned stimulus being presented (meat powder, shock, 27M&M, pp. 31-32. 134 puff of air); if, in the same series, the unconditioned stimulus is some- times presented without the bell or buzzer, learning will fail to take place.28 It appears, therefore, that in determining what constitutes a reliable ind- icator we shall have to look at all four cells of the relevant contingency table and not just at the one that Dretske takes to be germane in the case 28Rescorla, for example, studied the conditions under which a tone that had been paired with shock would acquire the ability to interfere with subse- quent bar-pressing behavior: suppression of bar-pressing was taken to be a measure of an animal’s having learned a response to the tone. Two groups of rats were used. For both groups of rats, the tone never occurred without shock: (Pr(shock/tone)=1). In one group, however, shock was also presented without the tone. Although this group received more shock overall, suppres— sion of bar-pressing was not observed. Results of this sort have led learning theorists to emphasize the importance of "stimulus contingency" in classical conditioning. (Robert A. Rescorla, "Pavlovian conditioned inhibi- tion", Psychological Bulletin (1969) Vol. 72, no. 2, pp. 77—94.) The article and a brief discussion may be found in H. Rachlin, Behavior and Learning (San Francisco: W.H. Freeman and Company, 1976.) The term "stimulus contingency" and a presentation in terms of contingency tables and conditional probabilities may be found in J.E.R. Staddon, Adaptive Behavior and Learning (Cambridge: Cambridge University Press, 1983. Hereafter, Learning.) pp. 140 ff. Staddon makes the following obser- vation with respect to conclusions drawn from Rescorla’s research: [Alnimals show conditioning to a stimulus only when it predicts the US: CS and U08 must therefore be correlated for conditioning to occur. This conclusion is appealing and wide- ly accepted, but. . . one might say of the concept of "predictability," as of the bikini: What it reveals is sug- gestive, but what it conceals is vital. "Predictability" is something quite different from "contiguity." Contiguity is a perfectly unambiguous, quantitative time relation between two events. But the predictability of something depends upon the knowledge of the observer. . . . Rescorla’s explanation is better than Pavlov’s, but its success carries a cost: A gain in comprehensiveness has also meant a loss in precision. (p. 140.) I am indebted to Wally Beagley for these citations and for helpful discussion of Rescorla's work. 135 of informational content. Since we are not holding out for perfect associ— ation, we will also have to think about what proportion of false negatives and false positives are tolerable in particular cases. We can already see where this will lead us. Let us think back to Dretske’s example of door-bell-ringing as a natural sign of the presence of a person on our porch. Suppose there were a retired Skinnerian in the neighborhood, who, to while away the weary hours, trained pigeons to push doorbells. Little by little, as more and more trained pigeons were loosed on the subdivision, doorbellringing would become a less and less reliable indicator of person-on-the-porch. At what point would doorbellringing cease to be a natural sign of person-on—the-porch and door- bell rings cease, thereby, to meann person-on—the—porch in that shrinking proportion of cases in which somebody (not a pigeon) really was on the porch? Obviously, it depends upon the interests and beliefs involved. If I wanted a refill for my magic mascara marker very, very much, and believed that itinerant Avon persons were the only source of this commodity, then it might be worth my while to put up with rather a high ratio of pigeon- initiated summons to the door rather than miss out on a valued opportunity. For me, bellringing might be a sufficiently reliable indicator of visitors to count (in my life) as a natural sign of their presence. Moreover, whether it did or did not might well depend not only on my longing for mascara and my beliefs about its source but on the availability to me of other reliable news about the latter: bellringing might maintain its status as a natural sign of person—on—the-porch were I to have no better source of 136 information, but be abandoned once I figured out how to peer through the curtains at the sound of a footstep. If we are on anything like the right track here, ’natural sign’ and ’natural meaning’ are thoroughly pragmatic notions, scarcely the ’objective’ stuff out of which to compose a naturalized semantics. It might be suggested that the notion of a reliable indicator--and thus, of a natural sign and of meaningn-—could be naturalized by explicating it in terms of whatever environmental correlations actually serve to establish and maintain a learned association between two kinds of events. This, how- ever, will not work: Dretske explicates learning in terms of meaningn and natural signs, which makes the move circular. Second,’learns that’ turns out to be right in there with ’believes that’: substitution fails in such contexts and the degree of intentionality revealed by its failures is that of the problematic verbs of propositional attitude--we cannot substitute even necessarily equivalent expressions salve veritate into ’learns that ’, as anyone who teaches mathematics can testify. Finally, although there certainly is a normative sense of ’learns’ such that I cannot learn lgpl 2 unless 3, its everyday psychological sense-—the sense that naturalizes the concept--allows Mary Jane to learn that men are dangerous even if they are not: ’learns lgpl 2' here simply means ’comes to believe lgpl l on the basis of relevant experiences (rather than, say, as a result of hypnosis or the ingestion of purple pills).’ Although the psycho- logical sense of ’learn’ is not as neat as we might wish, what matters for 137 our purposes is that we are not going to get the required tie between world and symbol if we rely on the naturalistic sense of 'learn’: we will, rather, get some very odd truth-conditions.29 Natural meaning does not seem to meet the demand for objectivity. Does it offer us insight into the sources of intentionality? We may recall that Dretske was attracted to the notion of information because, on his con- strual, it relied on nomic relations and these, in turn, displayed a kind of intentionality: coextensive terms could not be substituted salve veritate in the context ’It is a natural law that. . .’. In both KFI and "The Intentionality of Cognitive States", Dretske argued from this that cognitive states might most properly be seen as differing in degree but not in kind from physical states. [If] the lawful dependence of one magnitude on another is part of the physicist’s picture of reality, then intention- ality is also part of that picture. Hence, to the extent to which the mentality of our cognitive states resides in their intentional structure, knowledga, perception, and memory are perfectly ’natural’ phenomena. As we have seen, however, the regularities constitutive of natural signs are not, on Dretske’s account, strictly lawful. Nor do statements of such 29A learning text, for example, gives us the following: "’Learning,’ like its complement, ’memory,’ is a concept with no generally agreed meaning. . Learning is a category defined largely by exclusion. There is . . . no point in attempting a neat definition for learning. What we need is some understanding of the ways in which an animal’s environment can produce long-lasting changes in its behavior. If we can understand how these effects come about and, especially, if we can find the general principles that underlie them, then learning will be one of the phenomena explained." (Learning, p. 395.) 30ms, pp. 287—288. 138 regularities create non-extensional contexts. If what we are talking about are merely the correlations that hold around here these days, coextension- ality is adequate to maintain the truth of our claims: if being a cordate is highly correlated with X, then so is being a renate. It may be argued that my being able to predict from cordate status to X does not guarantee that I can predict from renate status to X. But this is because prediction by an individual involves that individual’s knowledge of the regnant corre- lations. Contexts like ’John can predict. . .’ are intensional but not available to Dretske on pain of circularity; contexts like ’It can be pre— dicted from. . .’ do not involve circular presuppositions but are not intensional, either. It would appear, then, that meaningn does not bear within it the seeds of higher order intentionality because it does not offer us a picture of the natural world as intentional all the way down in the way that Dretske hoped the concept of informational content might do. Even if meaningn were an "objective commodity" of the sort Dretske is looking for, it is hard to see how it could provide a link between the vocabularies of the physicist and the cognitive psychologist. Meaningn seems to introduce new problems without contributing to the solu— tion of old ones. Let us set aside our misgivings on this score now and go on to Dretske’s treatment of functional meaning, the analogue of what was previously called ’conceptual content’. There is, he tells us, "a kind of meaning that attaches to. . .components of systems. . . for which there are identifiable functions,"31 a sort of meaning that is separable from their meaning“. Because it is the function of the needle on the fuel gauge to ‘J— 139 inform us, by its position, about the amount of gas in the tank, there is, Dretske notes, a sense in which the needle’s resting on FULL means that the tank is full, even when the tank is, in fact, empty: its functional meaning is that the tank is full even though it does not meann that the tank is full. The needle’s position can thus glgrepresent the state of the fuel tank because of the job it has been assigned to do in the enterprise of driving. Moreover, the explanatory story we tell about how it came to have this job includes the fact that the needle’s position is, under certain circumstances at least, a natural sign of the state of the fuel tank. The functional meaning of a tokening of the needle’s position-~what it meansf-- is thus dependent on what it meansn under ’normal’ circumstances. Dretske puts it this way. When p’s being 9 is, normally, a natural sign of y’s being 2, when this is what it normally meansn, then there is a sense in which it means this whether or not y is E if it is the function of p to indicate the Spndition of y. Let us call this kind of meaning meaningf. More formally, (Mf) p’s being g means that y is l = p’s function is to indicate the condition of y, and the way it performs this function is, in part, by inglcating that y is l by its (p’s) being g. Dretske’s claim is that there exists a schema for applying ’misrepresentation’ to tokenings of the states of systems that involves 31Misrep., p. 21. 2Misrep., p. 22. It is not entirely clear what we are to understand by ’normal’ here, and about this we receive no help. 33Misrep., p. 22. 140 contrasting meaningn and meaningf. When a system tokens a state with the functional meaning lgpl 2 under circumstances such that the token does not meann lgpl l, we say that we have a case of misrepresentation on our hands. His next move will be to show that we can have meaningf without the intru- sion of our intentionality-~we can have underived meaningf. It is, however, not as obvious as it might seem at first glance that Dretske has provided us with a criterion (or an analysis, or an account) of misrepresentation that really will be applicable in situations where no intentionality is already presumed. Notice that when we say that the fuel gauge is misrepre- senting the state of the fuel tank, we can also say that the fuel gauge is misrepresenting the state of the fuel tank lp somebody. My hunch is that the existence of somebody to whom a state of affairs is being misrepre- sented is, in fact, ordinarily understood and, moreover, that we would gpl want to use ’misrepresent’ in cases where there was no such ’recipient’ of the misrepresentation. ’Misrepresent’, like ’represent’ (or ’signal’ or ’inform’) would seem to be a three- rather than a two-place predicate: l (mis)represents l to l. (Perhaps a fourth place is needed: ’...as 3’.) When l (or the domain of ls) is obvious, we simply do not mention it. If this is correct, the applicability of ’(mis)represent’ is limited to cases where some ’target’ or ’recipient’ can be identified. But it also appears that anything that can fill this role will itself have to be an intentional system. To my ear, at least, it sounds peculiar to say that stimulation from the retina represents the world to the visual cortex, say, rather than to a person. We may not need another system to provide the intentionality, as we do in the case of artifacts, but some intentional system I think 141 there must be. If this is the case, functional meaning presumes rather than explicates intentionality. Let us put this qualm aside, however, and continue to pursue Dretske’s account. Dretske’s concern for giving an account of underived intentionality leads him to insist that providing a naturalized account of (mis)representation requires meaningsf that depend only upon "functions a thing has which are independent of our interpretative intentions and purposes."34 Only such "natural functions", he tells us, can endow tokenings of states with a 35 with a meaningf that is "not parasitic on the way we exploit them," meaningf that does not rely on unreduced intentions. The search for a naturalized account of (mis)representation thus becomes a quest for func- tions of systems themselves that might be served by states that carry meaningn about their world. Not surprisingly, Dretske’s prime candidates are information-gathering functions of biological systems. There are, he notes, biological systems to whose components we willingly grant a func- tional status that is, to all appearances, innocent of extrasystemic purposes. Dretske draws our attention, for example, to magnetic navigation in bacteria. Some marine bacteria have internal magnets (called magneto- somes) that function like compass needles, aligning. . . [the bacteria] parallel to the earth’s magnetic field. Since these magnetic lines incline downwards. . . in the northern hemi- sphere. . . bacteria in the northern hemisphere. . . propel themselves [downwards]. . . . [Ilt is reasonable to suppose that. . . [magnetotaxis] functions so as to enable the bac- teria to avoid surface water. Since these organisms are 34Misrep., p. 25, stress suppressed. 5Misrep., p. 25, stress suppressed. 142 capable of living only in the absence of oxygen, movement towards geomagnetic north will take the [northern hemisphere —dwelling] bacteria away from oxygen—rich surface water and towards the comparatively oxygen-free sediment at the bottom. Southern-hemispperic bacteria have their magnetosomes reversed. It is not anybody’s wish but the metabolism of the bacteria that makes oxy— gen lethal to them; it is not anybody’s purpose but the way their world works that renders thir magnetosomes indicators of the direction in which oxygen level increases. It is on the strength of evolutionary theory--a nonpurposive story if ever there was one--that we say that the bacteria we encounter come equipped with magnetosomes because locomotion so guided proved (in the long run) healthier for the bacteria (for reasons just canvassed) than did available alternatives. It is in the light of these (impeccably naturalistic) considerations that we say that magnetosomes have a function for the bacteria. There are, thus, systems to whose constituents we attribute a wholely natural function, and thus systems the tokenings of whose states are invested with a natural meaningf. Dretske now needs to show that there are systems that satisfy the condi- tions he has set-—systems whose states (like those of the magnetotaxic bacteria) are endowed with natural functional meaning, systems whose func- tionally meaningful states (like those of the fuel gauge) can be tokened on occasions where they fail to carry their usual meaning". If such systems exist, then, according to Dretske, tokenings of their states can misrepre- sent their world and can do so without benefit of extrinsic intentionality. 36Misrep., p. 26. 143 Looking for such systems, Dretske encounters a familiar problem for which he offers a novel but unsatisfactory solution. As we have seen, natural meaning is the raw material from which Dretske proposes to create semanti- cally evaluable content--natural meaningf. Meaning", however, is no more uniquely given than was the informational content with which he dealt in KFI. "Functional indeterminacy" seems thus to pose a major obstacle to Dretske’s program. To see how, let us consider once again his example of the magnetotaxic bacteria. Using a bar magnet, he notes, we can lure these organisms into oxygen—rich surface water where they promptly perish. Are they in this case victims of misrepresentation? Only if tokenings of their magnetosomatic states meantf something that was not the case. What their magnetosomatic states meantf, in turn, depends (at least in part) on what tokenings of these states normally meantn. And their normal meaningn would appear to include all of the following: (a) the direction of the prevailing local magnetic field is so-and-so; (b) the direction of the regional geomagnetic field is so—and-so (c) this way is up (d) this way lies oxygen-rich water (e) this way lie death and destruction. Did tokenings of the bacteria’s magnetosomatic states meanf falsely, then, as did the hands on my stopped watch or the pointer of the broken fuel gauge? Not, says Dretske, if (a) is what determines the relevant meaningf. On this view, what the states of the bacteria’s magnetosomes normally indi- cate is that the local magnetic field is so—and-so; the natural function of 144 the magnetosomes is, accordingly, to indicate that the local magnetic field is so-and-so; tokenings of magnetosome states therefore meanf that the local magnetic field is so-and-so and lgpl lg jpgl gpy ll ypg. One can simply decide, says Dretske, that the function of the magnetosome is "to align the organism with the prevailing magnetic field. It is. . . the job 37 On the other of magnetic north to be the direction of oxygen-free water." hand, taking any of (b), (c), (d) or (e) as the relevant meaningn will, under the same circumstances, yield a verdict of misrepresentation. If, as seems here to be the case, a tokening of a state has no specific meaningn and, thereby, no specific meaningf, then there is no determinate answer to the question whether it misrepresents the world in a given situation, in which case we can hardly say that it has semantically evaluable content. Worse yet, Dretske's example seems to show that there will always be legi- timate contenders for meaningf that make misrepresentation impossible under any circumstances. What we need, Dretske tells us, is "some principled way of saying what the natural function of a mechanism is, what its various states . . . meanf."38 More precisely, what we need is some principled way to set aside a certain troublesome set of meaningsf--those that advert to states of affairs so causally proximal that discrepancy between meaningf and meaningn is precluded. A brief backward look will, I think, put into broader perspective the difficulty Dretske faces here. Fodor, it will be recalled, forcefully 37Misrep., p. 30. 38Misrep., p. 32, stress suppressed. 145 characterized the challenge facing the causal theorist.39 Misrepresentation, for such a theorist, is a matter of discrepancy between the property expressed by a symbol type and the property whose instance is causally responsible for a tokening of that symbol. The causal theorist also holds, however, that the property expressed by a symbol type depends upon the causal circumstances of its tokenings. For discrepancy to be possible, then, there must be some tokenings whose causal circumstances do not count toward determination of symbol content, some second-class or "wild" tokenings whose causal circumstances provide, instead, occasions of possible error. If misrepresentation is to be possible, not all causal circumstances can be put on an equal footing with regard to the determination of type content. Dretske’s naturalizing strategy is similar in general outline to Fodor’s. The problems they face have, therefore, a similar shape. Because Dretske’s theory differs from Fodor’s, it is not some set of causal circumstances that has to be dealt out of the game, however, but rather a set of meanings“. In particular, Dretske has to find a reasonable basis for disre— garding the meaningn that a tokening of a state carries with regard to its most proximate cause. As we have seen, if lglg meaningn is allowed a role in determining meaningf, discrepancy becomes impossible and misrepresenta- tion (as Dretske analyzes it) equally impossible. Dretske has, therefore, to show us how we can legitimately ignore this set of meanings“. It may be argued that an obvious solution is available were Dretske to take his own 39SWS; Psysem., Ch. 4. 146 theory seriously. The route from meaningn to natural functional meaning runs, after all, ylp the role played by natural meaning in the genesis of a system’s sensory-motor coordinations. Magnetosomatic states have natural functional meanings for bacteria because in their environment of adaptation ical fact, responsible for their belng included lg the behavioral machinery pl the bacteria. This consideration-~central, one might think, to Dretske’s teleological approach--would seem capable of insulating meaningf from troublesome proximal meaningsn like (a). Surely, we cannot give an adequate causal account of the bacteria’s current locomotor arrangements solely on the basis of the fact that their magnetosomatic states are natural signs of the most proximate magnetic field, without mentioning anything about the direction of the prevailing geomagnetic field, oxygen-free water, or which way is up around here. Had the magnetosomes indicated the direction of the local magnetic field in a world where this was not reliably correlated with the direction of salubrious oxygen—free water, there is no reason to be— lieve that a mechanism for magnetotaxis would have evolved. (Certain of our own brain states surely are correlated with environmental states of affairs but have acquired no functional meaning for us.) Why, then, does Dretske think that proximal meaningf based on (a) is so serious a problem? His most direct answer is given in a footnote, where he insists that functional meaning must take into account not only a system’s informational needs but also the particular sensory mechanism by which these needs are met. [Enc] says that a photoreceptor in the fruit—fly has the function of enabling the fly to reach humid spots (in virtue of the correlation between dark spots and humid spots). I have no objection to describing things in this way. But the question remains: gpy does it perform this function? We can 147 answer this question without supposing that there is any mechanism of the fly whose function it is to indicate the degree of humidity. The sensory mechanism can perform this function if there is merely something to indicate the 1uminosity--i.e. a photoreceptor. That will enable the fly to reach humid spots. Likewise, the bacteria’s magnetotactic sense enables (and, let us say, has the function of enabling) the bacteria to avoid oxygen-rich water. But the ypy it does it. . . is by having a sensor that indicates, and has the functipa of indicating, the direction of the magnetic field. Elsewhere, Dretske offers two other, rather different defenses of (a). It is, he notes, the most "modest" of the alternatives, in that it runs the least risk of "inflating the natural functions" of the system under consi- deration. It "sounds a bit far—fetched" to him, he says, "to describe the bacteria’s sensory mechanism as. . . having the function of indicating. 41 He also observes that there is a sense of the whereabouts of oxygen." ’function’ in which we would say that "this primitive sensory mechanism is, after all, functioning perfectly well [even] when, under. . . [a] bar mag- 42 net’s influence, it leads its possessor into a toxic environment." Because these comments are made somewhat off—handedly, it is hard to see precisely what principle is at stake. I think Dretske wishes here, as elsewhere, to conform his theory to ordinary language, to what we are comfortable saying about bacteria bits. I suspect that he is also concerned to hew as closely as possible to a stringently naturalistic line, and thus wishes to give an account of functional meaning that relies on ’objective’ characteristics of a system’s sensory mechanism as far as possible. 40Misrep., p. 31, note. Last stress added. AlMisrep., p. 32, stress removed. 42 Misrep., p. 29, stress added. 148 Dretske’s response to the difficulty he has raised is, in any case, to conclude that organisms which, like bacteria, have only "primitive representational capacities" simply do not have "clear and unambiguous capacity for misrepresentation."43 How, then, can any natural system overcome functional indeterminacy and achieve semantically evaluable content? In "Misrepresentation", Dretske’s solution takes the form of (1) specifying conditions under which it would be conceptually impossible for the most causally proximate causes of tokenings of a state to figure in their func- tional meaning, and (2) attempting to show that these conditions are met by actual biological systems and, thus, by nomologically possible ones. His case for (1) goes like this. The natural function of tokenings of the states of a component of a system derives from what they meann under "normal" circumstances. If there is no one thing that they meann from one occasion to another-- if they have no time-invariant meaningn—-there is nothing that it could be the natural function of the component to indicate by its states. If there is nothing that it is their natural function to indicate, such states have no natural meaningf to bequeath to their tokens. 43Misrep., p. 32. Can a system be capable of (at least some sort of) repre- sentation without having a capacity for (genuine) glgrepresentation? I would not have thought so. 149 Therefore, meaningsn that vary from one occasion to another cannot be relevant to the meaningf of tokenings of states of components of systems; only time-invariant meaningsn can be so. With this in place, Dretske goes on to (2), reassuring us that there are organisms the meaningsn of whose behavior-regulating states fail the time- invariance test with respect to proximal states of affairs but pass it with regard to distal ones. This requires, Dretske says, "a certain threshold of complexity in the information-processing capabilities of a system," one that we cross "somewhere between the single cell and man," at the point 44 What he where we encounter creatures capable of "associative learning." has in mind here is usually called classical conditioning, a process in which a response originally elicited by one stimulus comes to be evoked by another stimulus as a result of the systematic pairing of the two stimuli. The new or "conditioned" stimulus is usually labelled the CS. For creatures at the requisite level of complexity, Dretske tells us, an adaptive behavior-causing internal state, R, initially correlated with a type of proximal stimulus, gi, can, under the right circumstances, be conditioned to another type of proximal stimulus, E§j' For example, an internal neural state, R, that controls some adaptive response, and that is originally A“Misrep., pp. 33 and 35. Dretske assumes, of course, that we can tell a naturalistic story about this increase in complexity. 150 elicited only by a bright light, can come to be elicited by, say, a tone, 45 if the tone occurs in conjunction with the light. For such a creature, tokenings of R will thus meann gi at one time but mean pgj at another time. Just what R means [on any occasion] will depend on the indi- vidual’s learning history . . . . There is no time-invariant meaningn for R; hence, nothing that, through time, could be its function to indicate. In terms of the...[proximal pyents] that produce R, R can have no time-invariant meaningf. At the same time, R will, however, maintain its distal meaningn--R, let us say—-since the new stimuli that elicit it do so only because they have been regularly paired with R. [Bly hypothesis, any new [proximal stimulus] g to which R becomes conditioned is a natural sign of R. Learning is a process in which stimuli that indicate [mean ] the presence of R . . . [come], in their turn, [to be] indicated by so relevant internal state of the organism (R in this case). We can, therefore, deal out meaningsn with respect to learnable proximal stimuli for R without sacrificing a distal meaningn from which meaningf can 45There is a certain amount of vagueness with regard to what Dretske is talking about here. He is using a notation--s , cs.—-associated with clas- sical conditioning, a paradigm generally applied td reflex or autonomic responses--more importantly, to responses that have a "prewired" eliciting stimulus. Elsewhere, he will deny that such innate responses are relevant to an individual’s having reasons for his or her behavior. His example of the kind of behavior controlled by the internal state R is an adaptive "avoidance" response. Although some reflexes (blinking) could be seen as avoidance responses, most avoidance responses are not reflexes. This and some of his later work suggests that g ’s are, rather, discriminative sti— muli for an operant response. (You put the light on before the shock and ll the animal performs the chosen response within some time interval after light-on, it avoids the shock that it would otherwise get. Or you arrange things so that a bar-press results in a drop of water only if it is per- formed when a light is on.) Discriminative stimuli are much closer to Dretske’s idea of events that carry information about when it would be beneficial to perform some bit of behavior. 46Misrep., p. 35. 47Misrep., p. 35. Stress added. 151 be constructed; we can still construe the natural function of R as one of "indicating the [distal] condition (R) for which [all] these diverse [proximal] stimuli are Signson48 [Tlhe occurrence of R means that R is present. It does not mean that [proximal] g or s obtains, even or . 3 though, at any given sfate 3f43evelopmeh¥, it will mean this . n for some def1nite value of g. Dretske concludes that the kind of "complexity" that allows a creature to transfer an adaptive response from one stimulus to another-—to one that has been, "in the ’experience’ of the organism," a natural sign of it50--endows the creature with "a genuine power of misrepresentation."51 When there is a breakdown in the normal chain of natural signs, when, say pg occurs (a learned sign of R) under cir- cumstances in which it does gpl meann that R is present (in the way that a broken clock does not mean that it is 3:30 a.m.), R still means . . . that R is present. It means this because that is what it is supposed to mean , what is ifs natural function to meaga, pgp there lg avallable gp other condition ll ppg meanf. On Dretske’s latest account, then, a system is capable of misrepresenting § as R just in case its capacity for associative learning makes it possible for the same functional internal state, R, that was originally elicited by a set of proximal natural signs of R-ness--gi-—to be triggered by a set of new and different natural signs of R-ness--pgi. Since we are, it would seem, familiar with actual creatures who have a capacity for such 48Misrep., p. 36. 49Misrep., p. 36. 50Misrep., p. 35. 51Misrep., p. 36. 52Misrep., p. 36. Stress added. 152 associative learning, their nomic possibility follows, giving us Nature’s way of making a mistake. Or does it? I have already expressed anticipatory misgivings about the last step in Dretske’s proposal. I should like now to argue that, contrary to his claim, it is not the case that we can count on a capacity for learning either to deal proximal meaningf out of the game or to leave distal meaningf securely in place. I shall be arguing, first, that the concept of time-invariance does not support the burden Dretske places on it with re- gard to the former and, second, that the concept of learning fails equally with regard to the latter. More particularly, I shall be saying that ’time-invariant’ is ambiguous between two readings, one of which makes time-invariance vulnerable to the capacity for learning, the other of which renders it relevant to meaningf, but neither of which does both. I shall then go on to say that ’learning’ is similarly ambiguous between a reading that guarantees the fixity of distal meaningn and one that provides us with instances in the actual world, but that again neither does both. Let us take up these objections in order. The first part of Dretske’s strategy is to show that once a capacity for learning has emerged, proximal meaningsn are, so to speak, neutralized, because, lacking time invariance, they cannot contribute to functional meanings. This amounts to two claims: (1) if there is a capacity for learning, proximal meaningn is not time-invariant 153 (2) if proximal meaningn is not time-invariant, it cannot give rise to meaningf. We shall, therefore, need to ask what, precisely, Dretske means by ’is not time-invariant’. One possibility is that the meaningn of a tokening of a state fails to be time-invariant just in case it ppg undergo change over time--nomic possibility is built into the concept. On this reading, (1) would seem true by definition: we cannot have necessary sameness of meaningn if a behavior-regulating internal state R can be correlated with g1 at l1, with g2 at l2 and so on, as a result of its encounters with the world. It is, however, less clear that this understanding of time-invari- ance entitles us to (2). Consider, for example, a creature in whose sheltered life some internal (behavior-regulating) state, R, has always been a natural sign of g1. Its neural fittings are sufficiently sophis- ticated, however, that had it encountered a richer and more varied environment, R would have been conditioned to (become correlated with) a whole series of new and different gi’s. Do we really want to say that for this creature R cannot currently have the function of indicating g1? More to the point, can Dretske consistently advance such a claim after telling us that the meaningf of door-bell ringings was determined by whatever regularities actually prevailed in the neighborhood, even though these were only contingent? It would seem not. The other possibility is, of course, that ’time-invariant’ is not a modal concept after all. On this reading, the meaningn of tokenings of a state 154 fails to be time-invariant just in case it has, in fact, changed from one time to another. This reading makes (2) a good deal more attractive, if not entirely a shoo-in. If, by definition, functional meaning is fixed by lgp normal meaningn of tokenings of a state with respect to some event type, 53 At the then the absence of any one such meaningn certainly precludes it. same time, however, (1) now appears open to challenge. A history of shifting proximal meaningsn may well undermine the possibility of proximal meaningf but a capacity for learning does not, in and of itself, guarantee such a history, and thus is no threat to proximal meaningf. Applied consis- tently, then, neither reading of ’time-invariant’ gives Dretske both of the premises he needs to show that a capacity for learning dissolves proximal meaningf. Even if (1) and (2) were both allowed, however, we would still need to show that the same learning capacity that has been invoked to block unwanted proximal meaningf does not simultaneously erode the distal functional mean- ings on which Dretske’s account of misrepresentation depends. If the neural plasticity of an organism permits new proximal stimuli to elicit R token- ings, what guarantee do we have that all such new stimuli will, in fact, be natural signs of R? If they are not, then R tokenings will not normally 53Not entirely a shoo-in because it is not clear what kind of change we are talking about. If the meaning of tokenings of R changes rapidly or often enough, what makes it the case that there lg a meaning to be had? How long does a correlation have to last for natural signhood to be declared? More importantly, if meaningn changes slowly or seldom, why can’t we say that it has, within each spacious interval, given rise to a different meaning ? Once again, pragmatic considerations seem to threaten a non-circular account of content. 155 meann R, the natural function of R cannot be to indicate R and R will therefore be unable to bequeath ’R’ to its tokenings as their functional meaning. Dretske’s answer is that we are talking about proximal stimuli that have acquired their ability to elicit R tokenings as a result of associative learning.54 Dretske is explicit about what ’learning’ means. Learning "is a process in which stimuli that indicate the presence of R are, in their turn, indicated by some relevant internal state of the organism."55 "Almost any g can become a pg," Dretske tells us, "by func- 56 R tioning (in the ’experience’ of the organism) as a sign of R." continues to have the content that R is R across episodes of learning because "_y hypothesis, any new gi to which. . . [it] becomes conditioned 57 is a natural sign of R." Invariance of distal meaningsf is thus ensured for systems for which learn— ing (in Dretske’s sense of the word) is the only way in which tokenings of new proximal stimulus types could come to be correlated with R tokenings 54Misrep., p. 35. As far as I can tell, the only place where Dretske uses the term ’plasticity’ is in KFI: "What endows some systems with the capacity to occupy states which have, as their semantic content, facts about some distant source is the plasticity of the system for extracting information about a source from a variety of physically different signals. The system, as it were, ignores the particular messenger in order to respond to the information delivered by that messenger" (p. 187). 55Misrep., p. 35. 56Misrep., p. 35. S7Misrep., p. 35. My emphasis. This view of learning is consistent with what we have in a slightly earlier publication: "Learning, in fact, is a process in which the meaning of internal signs (i.e., their correlation with external conditions). . . helps to determine how these signs are exploited for purposes of motor control" (M&M, p. 32. Stress deleted). 156 (or for which any other mechanism for altering the efficacy of stimuli somehow had the same end result). Such a system could not, for example, on the basis of its experience (or anything else) come to token R to a new proximal stimulus type that did gpl, in fact, signal R-ness dependably-- that did gpl indicate R. Dretske’s learner, in short, cannot, by defini— tion, learn anything that is not the case. There is, of course, nothing logically impossible about the existence of such a system. We might dub it The Perfect Learner.58 Dretske’s argument rests, however, on the rather more stringent requirement of nomological possibility. It requires showing that creatures actually exist at this (or some nomologically accessible) world such that the natural meaningf of their internal states can be discrepant from their meaning“. If it turns out that the only creatures for whom this discrepancy is possible are Perfect Learners, Dretske has not made his case, because the natural processes of our world do not, even under ’normal’ circumstances, eventuate in Perfect Learners. Indeed, learners in the everyday, psychological sense of the term —-systems capable of rewiring themselves in the ways we characterize as either classical or operant conditioning—-seem to be deeply Imperfect Learners. They acquire taste aversions to innocent but unfamiliar—tasting substances; they attack provocative-looking objects that happen along after a painful episode and ignore those that could have figured as causes of their distress; they stubbornly persist in avoidances long after stimulus type and aversive outcome have ceased to bear any correlation to each other. Dretske might 58Well, not absolutely Perfect: it might fail to learn something that was the case and thus not token R to a new stimulus type that actually was correlated with R-ness. 157 still salvage the existence of Perfect Learners by saying that actual organisms surely rewire themselves in a way that accurately reflects what has gone with what lg their egperience, where ’experience’ refers not to a stretch of their history, viewed from the outside, as it were, but, rather, to the way the organism represents the world to itself. This would make every learner a Perfect Learner. Such a move would, however, presuppose that we already had an account of representation available, as well as be- ing conspicuously out of keeping with the ’objective’ emphasis of Dretske’s program. Last but not least, it would surely yield some very odd truth conditions. If I am correct about the problems Dretske’s account runs into in its last and crucial step, his new account of misrepresentation provides us, at best, with a story about the logical possibility of natural systems falling into intrinsic error, but does not explain how actual ones do so. It may be helpful at this point to remind ourselves of the budget of problems with which Dretske set out. Even if we take into account only the goals of do- mesticating intentionality and giving an account of belief content we are, I think, quite far from their solution. Dretske has not yet explained (1) how physical systems could manifest intentionality--in particular, (2) how naturalistically characterized content can be either "exclusive" or truth- evaluable. With regard to the first, I have argued that replacing the notion of information by the concept of natural meaning, construed as a matter of local correlation, is unhelpful because ’meansn’, unlike ’carries the information that’, yields fully extensional contexts. Nomicity may 158 provide the basic material from which higher degrees of intentionality can be constructed; correlation does not. With a bit of interpretative ingenuity, the notion of a natural sign can be made to deliver intensional contexts--that I can (given my knowledge) or do (given my interests and alternatives) rely upon the presence of X to predict Y does not guarantee that I can (given my knowledge) or do (given my interests and alternatives) rely upon the presence of Z to predict Y, even where X and Z are coextensive properties. We pay for this intensionality, however, by sacrificing the objectivity or mind-independence that Dretske claims for information. One will not get intensionality out of ’is a natural sign of ’ unless one smuggles it in beforehand. Fodor argues that "What we do not have--what, for all we now know, we ppggpl have--is a notion of information that is both ’objective’ and appropriate to behavioral explanation."59 How about (2)? A second goal of Dretske’s naturalizing project was to endow naturalistically characterized content with the fine grain or exclusivity with which we individuate beliefs. In "Misrepresentation", we see Dretske struggle to exclude troublesome proximal functional meanings. He attempts to do this by stipulating that natural meanings be time—invariant and then deploying the capacity for associative learning to block such invariance relative to the proximal stimulus. Disambiguated, neither ’time invariant’ nor ’learning’ is adequate to the demands that Dretske makes on it. This 59Fodor, Jerry A.,"Information and Association", Notre Dame Journal pl Formal Logic 27 (1986), p. 316. The intentional notion of information is, he insists, "perspectival and receiver relative" (p. 314). 159 failure leads to a third shortfall. Exclusion of proximal meanings is, in general, a condition for the possibility of misrepresentation in a causal theory. The capacity for misrepresentation is a precondition for the possi- bility of false belief; the possibility of having false beliefs is a requirement for having beliefs simpliciter, and a system that cannot have beliefs cannot have a mind. So if the evolution of the capacity for learning does not explain the genesis of the possibility of error, it does not explain the making of mind in real time, either. Given Dretske’s com- mitment to providing such an account, this is bad news. What we do not know at the end of "Misrepresentation" is just what Dretske most wanted to re- veal to us at the start of KFI--how, in a mindless world, minds came to be. Chapter 6. Saving Life As We Know It Suppose nothing even remotely like Fodor’s causal theory can get off the ground. How bad would this be? On Fodor’s account, very bad indeed. [I]f commonsense intentional psychology really were to col— lapse, that would be, beyond comparison, the greatest intellectual catastrophe in the history of our species. We’ll be in deep, deep trouble if we have to give it up. Doing semantics for high stakes is, beyond doubt, exciting but thrills like these do not come cheap. To get from the unavailability of a causal theory to the crash of commonsense psychology requires a budget of premises. For 1Psysem., p. xii. More recently, he has upped the ante: "[I]f it isn’t lit- erally true that my wanting is causally responsible for my reaching, . . . and my believing is causally responsible for my saying . . . then practi- cally everything I believe about anything is false and it’s the end of the world." (Fodor, Jerry A., "Making Mind Matter More", Philosophical Topics 17 (1989), p. 77.) Other theorists regard the crash of folk psychology with an equally acute relish. Thus, for example, Paul Churchland: The magnitude of the conceptual revolution here suggested . . . would be enormous. And the benefits to humanity might be equally great. If each of us possessed an accurate neuro- scientific understanding of . . . what we now [only] conceive dimly. . . then the sum total of human misery might be much reduced. The simple increase in mutual understanding that the new framework made possible could contribute substantially toward a more peaceful and humane society. Paul M. Churchland, Matter and Consciousness (Cambridge, MA: A Bradford Book/MIT Press, 1984), p. 45. iInvocation of earlier "conceptual revolutions" is a hallmark of eliminati- vist optimism: "[Tlhe picture of the universe that has emerged from the rubble of the geocentric view is more beautiful, more powerful, and more satisfying than anything Aristotle or Dante could imagine. . .[Wle may be poised to begin a similar adventure. " (Stephen P. Stich, From Folk Ps cholo to _C_pgnitive Science: The Case _gainst Belief (Cambridge, MA: A Bradford Book/MIT Press, 1986), p.246.) 160 161 some of them, Fodor offers extensive arguments--e.g., for the necessity of a computational theory of mind and a genuine language of thought. In what follows, I shall draw attention to the extent to which Fodor's version of the eliminativist threat also relies upon less remarked notions of scien- tific respectability and adequacy of explanation. I shall argue that Fodor's interpretation of these demands is open to a line of criticism that closely parallels his own challenges to an earlier eliminativist program. Fodor takes seriously the claim that there is a state of affairs that could obligate us to give up commonsense psychological explanation. "Commonsense belief/desire psychology," he tells us, "is, in philosophically interesting ways, problematic."2 In particular, the semantic evaluability of beliefs and desires is ”metaphysically worrying."3 Like the eliminativists against whom he is arguing, Fodor thus endorses some or other instantiation of what I shall be calling the Eliminativist Schema. [E] There is some state of affairs, A, such that if A turns out to be the case, commonsense psychology will justifiably be relegated to the trash heap of intellectual history. The question is, what, precisely, stands in for A, and why, for a given A, should its truth be presumed to support the eliminativist consequent? We shall want to begin, therefore, by identifying the version of [E] that appears (or can, with a bit of work, be made to appear) in Psychosemantics, and then inquire whether there are grounds for accepting it. 2Psysem., p. x. 3Psysem., p. 156n. 162 For guidance as to the version of the Eliminativist Schema that informs Fodor's project we may, I think, reasonably look to his criteria for the vindication of the propositional attitudes. Holding onto the attitudes, Fodor tells us, requires "showing how you could have (or, at a minimum, showing that you could have) a respectable science whose ontology expli- citly acknowledges states that exhibit the sorts of properties that common sense attributes to the attitudes."4 We may extract from the "minimal" for- mulation and tentatively attribute to Fodor a condition for eliminating the propositional attitudes, [EF]. [EF]: Propositional attitudes deserve elimination just in case it is impossible that there be a respectable science, S, such that S explicitly acknowledges states exhibiting the sorts of properties that common sense attributes to the attitudes. Is [BF] what Fodor has in mind? Could he find [BF] plausible? He could were he to take to be persuasive a more general ontological principle, scienti- fic superrealism, [SSR], of which [EF] is a straightforward instance. Scientific Superrealism [SSR]: For any state (entity), P, P deserves elimination just in case it is impossible that there be a respectable science, S, such that S explicitly acknow- ledges states (entities) exhibiting the sorts of properties attributed to P. It is, however, difficult to imagine that Fodor would find [SSR] convin- cing: it offers neither plausible sufficient nor necessary conditions for consigning anything to inexistence. 0n the one hand, it debars what appear to be perfectly acceptable candidates for existential recognition. A theomist who takes [SSR] seriously and who, like Fodor, accepts the impos- sibility of there being a science that acknowledges states exhibiting the 4Psysem.,p. 10. 163 property of, say, being a genuine Chippendale, is stuck with the conclusion that there are no such things as chairs, since common sense willingly at— tributes this property to chairs.5 Nor does it help to amend [SSR] to de- mand for the ontological acceptability of P that there be a respectable science, S, such that S explicitly acknowledges states (entities) that have essentially the sorts of properties attributed to P essentially. Even if we ignore questions about what it means to have (or to attribute) properties essentially, it would still be the case that a theorist who acknowledged the impossibility of a science featuring entities exhibiting umbrellahood as an essential property must be prepared to eliminate entities to which umbrellahood is attributed essentially, which is to say, umbrellas. But a doctrine whose verdict is that there are neither chairs nor umbrellas is scarcely in a position to bequeath plausibility to its instances. [SSR] licenses ontological overkill, dispensing with all but a handful of "natural" properties such that they carve at the joints, they are intrinsic, they are highly specific, the sets of their instances are ipso facto not en- tirely miscellaneous, there are only just enough of th m to characterise things completely and without redundancy. 5 Fodor says that science taxonomizes by causal powers: being a genuine Chippendale is not an example of such individuation. See Psysem., pp. 44 ff. 6David Lewis, 93 the Plurality of worlds, (Oxford: Basil Blackwell, 1986), p. 60. A more austere version of [SSR] limits us to the handful of "sparse" properties acknowledged by physics, thus disenfranchising swans, elms and DNA. Lewis, of course, is not himself a scientific superrealist. He urges that "an adequate account of what there is ought to accommodate both [sparse and abundant propertiesl" (p. 60). 164 Less obviously, perhaps, [SSR] is also excessively permissive: as stated, it denies the apparently sensible claim that, e.g., unicorns do not exist. According to [SSR], inexistence requires that no possible science acknow- ledge the property in question; elimination is thus precluded in cases of nomically possible but unrealized properties.7 It would therefore seem un- charitable to presume that Fodor depends upon [SSR] to ground [BF]. Indeed, he appears explicitly to reject [SSR] when he tells us that differences between the categories of scientific psychology and those of common sense 8 Clearly, some- do not, in themselves, justify skepticism about the latter. thing is wrong with the suggested reconstruction. We can take care of the problem of over—permissiveness by rewriting [BF] as [BF]'. [BF]': Propositional attitudes deserve elimination just in case there is not some actual respectable science, 8, such that S explicitly acknowledges states exhibiting the sorta of properties that common sense attributes to the attitudes. Can we find grounds for accepting [BFl' that are less drastic than endorsement of scientific superrealism? 7Even if unicornhood is not nomically possible, surely some unactualized properties may be so. To deny that there could be any such properties is to identify nomic possibility with actuality--not in itself a particularly ap- pealing doctrine. If, as Richard Hall has suggested, we cannot even make sense of the notion of a possible science, we have yet another reason to reject [SSR]. 8Psysem., pp. 44-45. 9[BF]' is, in fact, what we would have arrived at had we gone with the more demanding of Fodor's conditions for the vindication of the attitudes: "[Vlindicating commonsense psychology. . . means showing how you could have . . . a respectable science. . .etc." I assume that this is a demand to (iisplay such a respectable science-—to show that it is actual. Otherwise, it does not differ from the minimal demand to "show that you could have n 165 Perhaps we can if we are willing to invoke more premises.10 Consider the following, beginning with a statement of explanatory scientific realism, [XSR]. Explanatory scientific realism [XSR]: For any state (entity), P, featured in a purportedly causal explanation, P deserves elimination just in case there is no respectable science, S, such that S explicitly acknowledges states (entities) with the sorts of properties attributed to P. To this, we add claims about the explanatory aspirations of commonsense psychology, [C], and the essential role of the propositional attitudes in that enterprise, [PA]. [C] Commonsense psychology offers purportedly causal explanations. [PA] The propositional attitudes are featured in commonsense psychology. Taken together, [XSR], [C] and [PA] entail [BF]'. All three premises are consistent with Fodor's explicit commitments. [C] and [PA] fall directly 10It may appear that we have missed a bet in not opting for a more modest garden—variety scientific realism, [SR]. Scientific realism [SR]: For any state (entity), P, P is on- tologically admissible if there is a respectable science, S, such that S explicitly acknowledges states (entities) exhibiting the sorts of properties attributed to P. Fodor is, without a doubt, a scientific realist. He has, in fact, expli- citly endorsed something very like [SR], telling us that "[T]he best kind of ontological argument is. . . we need this construct to do our science" (Intro., Rep., p. 29). We have, however, bought plausibility at the price of relevance. From [SR] we cannot derive [EF]', but only the weaker [BF]*: [BF]*: We are entitled to hold onto the propositional atti- tudes if there is some respectable science, S, such that S acknowledges states exhibiting the same sorts of properties that common sense attributes to the attitudes. [BF]* is not strong enough to render Fodor's project a defense of Life as ‘We Know It, allowing as it does that there are other routes to ontological respectability besides acknowledgment by good science. It is clear that in Fodor's view, the absence of a vindicating science spells deep trouble for psychological idiom. If [SR] is incapable of yielding a bona fide elimina— tivist conditional--a no-nonsense version of [E]--its plausibility does us no good whatsoever. 166 out of his characterization of commonsense psychology and the claim that "[Blither propositional-attitude psychology is in the business of causal explanation or it is out of work."11 [XSR] prunes the domain within which legitimation by a respectable science is required for ontological admiss- ability: unlike [SSR], it does not require us to disallow chairs and umbrellas just because they exhibit scientifically irrelevant properties. In so doing, [XSR] respects Fodor's explicit distinction between conceptual schemes that are in the business of causal explanation and those that are in some other line of work. As he puts it, "Causal explanation is just one human preoccupation among many. . . . Not all taxonomies have that end in view."12 We have come up with an eliminativist conditional that is appropriately menacing and with grounds for it that are consistent with Fodor's commitments. If there are other combinations that meet these constraints, they do not leap easily to the eye. I shall therefore take [BFl' as Fodor's version of the Bliminativist Schema and proceed on the 11Psysem., p. 157n. 12Psysem., p. 45. [XSR] is also consistent with Fodor's position on the se- mantics of natural kind terms. Discussing the language of thought, he tells us that if a term is "an intended kind term" and it turns out that no such kind exists, then it is a "failed kind term and. . . failed terms don't apply to anything." ("A Theory of Content. Part I: The Problem", mss., p. 49.) If we take 'intended kind term’ to mean 'candidate for deployment in causal explanations', and failure to denote a kind as failure to denote a property that can appropriately be invoked in causal explanation, and 'X's’ not applying to anything as equivalent to the inexistence of KS, then we have something like [XSR] in the formal mode. This is also how Ian Hacking interprets Putnam's doctrine on natural language kind terms: if, contrary to initial belief, 'X' turns out not to be a natural kind term, "the ques- tion of its extension does not arise. If it must arise, the extension is the empty set." (Representing and Intervening: Introductory Topics in the Philosophy of Natural Science, Cambridge: Cambridge University Press, 1983, p. 80.) Henceforth, R&I. 167 assumption that its plausibility is supposed to depend upon that of [XSR], [C] and [PA]. How compelling, then, are these premises? I shall be arguing that despite its initial plausibility, [XSR] has problems that become apparent when we try to unpack the unexplicated "respectable science".13 According to Fodor, the verdicts of "respectable science" play a crucial role in vindicating states and properties invoked in explanation. What, then, constitutes scientific respectability? What would it take, for ex- ample, for RTM to be a science of mind "good" enough to confer ontological legitimacy on the propositional attitudes? Fodor has told us elsewhere that any "working science" is "ipso facto in philosophical good repute."14 In ordinary parlance, an enterprise counts as a working science by virtue of receiving NSF funding, supporting an applied technology, publishing refer- eed journals, having an agreed-upon methodology, and the like. It seems unlikely, however, that Fodor would view existential import as following from such sociological facts.15 Cognitive considerations offer a more at- tractive alternative: we have a working (and therefore a philosophically 13If not earlier. Suppose we invoke the state of having typhoid fever to explain a death and, at the same time, attribute to that state the (second order) property of being dreadful, a property not likely to be explicitly acknowledged by respectable science on any understanding of the term. According to [XSR], typhoid fever would be apt for elimination. There may be ways of patching this up, but they are not immediately obvious. 14 PAs, in Rep., p. 200. 15A representative of NEH once distinguished the sciences from the humanities for me by saying that almost anything would be counted as a humanity as long as it didn't involve "quantitative methods". It seems equally doubtful that Fodor considers equations to be the differentia of good science, however. 168 reputable) science when we have an enterprise that delivers empirically adequate, counterfactual supporting generalizations over some specified domain. Somewhat surprisingly, this turns out not to be good enough for Fodor. He insists that everyday intentional psychology has just such gener- alizations at its disposal but obviously does not take this as settling the question of its scientific respectability. Otherwise, why should its states and properties stand in need of salvage by some "good science of mind"? p Even more surprisingly, it is not only the informal kind of psychology whose credentials are in question. Fodor notes that although cognitive psy— L chologists invoke representations in their explanatory accounts of relations between, e.g., word frequency and retrieval time, this does not automatically legitimate the construct in question: not, he tells us, when "the construct is mental representation and the science is psychology."16 We have, it seems, come up against a tension in Fodor's methodology. On the one hand, he embraces a thoroughly Quinean naturalizing epistemology17; on the other hand, he is sensitive to the demands of a more traditional philosophy of science.18 16Intro., Rep., p. 29. 17He has said, for example, that "[Wle ought to give up asking for analyses; psychology is all the philosophy of mind that we are likely to get" (PAs, p. 202). 18That Fodor's fealty to the claims of ”working science" is tempered by a 'willingness to take seriously more "philosophical" concerns is apparent in the following passage. [Ylou might think Aunty would argue like this: "When I was a girl, ontology was thought to be an a priori science; but now I'm told that view is out of fashion. If, therefore, psycho- (Footnote continued) 169 Let us turn, therefore, to his more traditionally philosophical treatment of the requirements of good science: Psychosemantics was, after all, writ— ten for philosophers. I shall be arguing that the notion of a respectable science on which Fodor relies is that developed in "Special Sciences (or: the Disunity of Science as a Working Hypothesis)"19, that arguments in Psychosemantics assume this picture, and that interpreting Fodor's demands on good science in the light of this proposal enables us to understand both his current uneasiness about psychology and his insistence on the need to "reduce" the intentional properties to which it is committed. I shall then go on to argue that this construal of scientific respectability (and thus, of Fodor’s conditions for the vindication of intentional idiom), rests on questionable assumptions. It is easy to see Fodor's current insistence on the need to "naturalize" semantic properties as flatly inconsistent with his well known advocacy of intentional psychology. A brief review of his work in psychological meta- 18(continued) logists say that there are mental representations, then I suppose that there probably are." . . . That is not, however, the way that Aunty actually does argue. Far from it. . . . If there is one thing that Aunty believes in her bones, it is the ontological promiscuity of psychologists. So. . although psycholinguists may talk as though they were pro- fessionally committed to mental representations, Aunty takes that to be loose talk" (Psysem., p. 144). You might think that having signalled his disapproval of Aunty's a priori methodology, Fodor would refuse to argue on her grounds. That is not, however, what happens: he settles in to show that Aunty's proposed paraphrase of the psychologists’ "loose talk" won’t work. 19§yg£hg§g, 28 (1974), 97-115. Subsequent page references are to the (somewhat revised) version included as the second half of the Introduction to LOT. 170 theory, however, allows us to see not only compatibility but continuity between these positions. In Psychological Explanation, Fodor argues that because psychological concepts are typically functional, they are unlikely to conform to requirements for their reductive elimination in favor of those of neuroanatomy, and, a fortiori, those of basic physics. To take the possibility of reductive elimination to follow from the regulative ideal of unity of science, and thus to be a hallmark of scientific respectability, is, he tells us, to subject psychological inquiry to Procrustean stric- tures. The "traditional approach to the unity of science," Fodor concludes, is "in need of liberalization." [I]t will have to require something less (or other) than re— ducibility as the relation between constructs in neurology and those in psychology. It seems. . . that scientific theories can fit together in more than one way, perhaps in many ways. If this is correct, then reduction is only one kind of. . . relation between scientific theories that satis— fies reasonable constraints on the unity of science. It ould be interesting to know what other kinds. . . there are. Fodor's answer is given in "Special Sciences (or: the Disunity of Science as a Working Hypothesis)", in which he proposes to show how, "on a revised account of the unity of science"21, it is possible for special sciences to be genuine sciences even though they are not strictly reducible to basic physics. For convenience, I shall be referring to this proposal as Fodor's New View. 20Psychological Explanation: An Introduction to the Philosophy of Psychology (New York: Random House, 1968), pp. —119- 20. Hereafter, PE. 21 LOT, p. 20. 171 [1] Law of special ------------- > 8 2y science [3] Generalizations ——:s/§\\ P4/2\\ *my [2] Disjunctive Plx v sz v. .P nx, P' x le v P*2y v. predicates of (ultimately) 1 physics *{LPQ 1 [4] Laws of sriLPZF— . (ultimately) physics *iLPpi’ Statements at [1] are laws of the special sciences-~counterfactua1 support- ing generalizations of the form 'All events which consist of 5's being 51 bring about events which consist of y's being 52," where S1 and 52 are predicates of a special science (1975, P. 10). Every event that satisfies a special predicate, Si’ is ultimately identified—-perhaps by way of a chain of predicates of progressively more "basic" special sciences-~with an event that satisfies some or other physical predicate, as indicated by the fanned—out disjunctions at [2]. Fodor refers to this condition as 'token physicalismf. [1] and [2] are related by [3]-—true empirical generaliza— tions (not laws) of the form "every event which consists of 5's satisfying Slil is identical with some event which consists of x's satisfying some or other predicate belonging to the disjunction P1 v P2 v. . .v Pn'"22 Lawful relations between special properties are made possible by a "cross-cutting" of those that hold between physical properties. Quite as much as classical reductive eliminativism, Fodor's New View respects "the generality of physics," its unique position as "basic science," and, most especially, its 22LOT, p. 20. 172 privileged role with respect to establishing the legitimacy of every other theory. This becomes particularly clear when we think of the New View as being, like its classical predecessor, not merely a descriptive and explanatory thesis but a set of regulative constraints as well. The only constraint made explicit in "Special Sciences" is token physicalism--the demand that every event to which the predicates of a respectable special science apply be a physical event; a second requirement is, however, easily discernible. On the New View, it is by virtue of their physical properties that events can participate in lawful special relations; it is by virtue of the lawful special relations in which events participate that they satisfy special de- scriptions. Events thus owe their special descriptions to their physical descriptions and cannot differ with regard to the first without differing with regard to the second. As we have seen, however, events can differ in their physical descriptions without differing in their special descrip- tions—~this is precisely the point of functional taxonomy. In the terminology that Fodor now uses to refer to this asymmetry among special and physical properties, special predicates (properties) must supervene on physical ones. Fodor believes that the New View avoids shortcomings of the reductionist interpretation of unity of science while preserving its virtues. Removing the demand for type physicalism removes barriers to psychology's character- istically functional taxonomy: as long as particular psychological events 173 can be identified with physical--e.g., neurological--events, the existence of lawful relations between psychological states can, presumably, be ex- plained by the existence of lawful relations between the neurological (chemical, electrical, etc.) properties whose instantiations are gathered to psychological types. At the same time, requirements of token physicalism and supervenience ensure that the special sciences meet crucial demands on genuine science. "[Flor any reasonable purpose," Fodor tells us, "the weak- 23 We shall return shortly to the question of what er doctrine will do." purposes, precisely, these constraints might be thought to serve. First, however, we shall take note of their contribution to the eliminativist threat resident in Psychosemantics. At the time it appeared, "Special Sciences" expressed a liberalizing view, legitimating the "autonomy" of the special sciences-~especially, psychology --by showing that scientific respectability did not require that their laws and properties reduce to those of physics. What was less remarked was that the special sciences are not on this account "autonomous” in any ordinary sense of the word. Despite Fodor's democratic assurance that ”Not all the classes of things about which there are important, counterfactual support- ing generalizations to make. . . are, or correspond to, physical kinds," 23LOT, p. 13. Fodor probably has another agenda as well. Noting that if we add to the premise of token physicalism—-that every event talked about by any science is a physical event--the assumption that every event is, in fact, describable in the vocabulary of some or other science, we get the stronger doctrine of materialism, he comments casually that while it is possible to be a token physicalist without being a materialist, it is hard to see why anybody would bother (LOT, p. 13). Support for (or at least con- sonance with) across—the-board materialism seems thus to be another of the virtues that Fodor saw in classical unity of science doctrine. 174 and the challenge, "Why. . . should not the kind predicates of the special 24 sciences cross-classify the physical natural kinds?", a respectable special science turns out, on the New View, to be one that knows its place: its lawlike generalizations must be legitimated by those of the next more basic science and may, in turn, serve to certify the explanatory generali- zations of special sciences above it. If this interpretation of Fodor's views about respectable science is correct, we should not be surprised that semantic or intentional properties present themselves to him as problematic nor that he should have misgivings about the "goodness" of a science that acknowledges states exhibiting such properties. We have noted that the New View places two constraints on respectable science-—token physicalism and supervenience. Bach poses problems for intentional idiom. That the demand for supervenience creates difficulties is, by now no news. Given the right scenario, common sense will, it seems, blithely assign be- liefs with different content to Twins who are, by hypothesis, molecule for molecule identical. That states with precisely the same physical descrip— tion should differ in their psychological description is clearly going to be troublesome to a New View theorist. Fodor takes this to be the least of the problems raised by intentional idiom, however, proposing that it can be overcome by introducing a suitably refined notion of "narrow" content. That this can be done is debatable; even if it were to prove possible, however, we are left with what Fodor himself takes to be a more serious incitement 24LOT, pp. 24-25. Cf. his more recent pronouncement: ”[Tlhings in Nature overlap in their causal powers to various degrees and in various respects; the sciences play these overlaps, each in its own way" (Psysem., p. 45). 175 to intentional irrealism, the "ontological intuition. . . that there is no place for intentional categories in a physicalistic view of the world," that "the semantic (and/or the intentional) will prove permanently recal- citrant to integration in the natural order".25 Having narrow content is, after all, still an intentional property: even if the truth conditions that they deliver-~are, as Fodor suggests, context relative, states narrowly in- dividuated are still, in this sense, abggg something.26 But (by the eliminativist premise, [XSR]), aboutness is apt for elimination unless there is some good science that acknowledges aboutness——a respectable science whose predicates include those of the form '5 is about y'. And (on the New View), a science cannot be respectable unless the events to which its predicates apply are token identifiable with events to which the predicates of physics apply. 80 unless there is some physical relation, Rp, such that instantiation of xpr is sufficient for instantiation of x is abggt y, intentional idiom is in deep trouble. If aboutness is not to be apt for elimination, each of its tokenings must be (ultimately) identifi- able with those of some or other physical property: if aboutness is to be acknowledged as (in this sense) "real", it must (in this sense) be 25Psysem., p. 97; SWS, p. 232. That there is more than one problem here can be missed. One reaction I have heard to Chapter Four's presentation of a causal semantics is, "Didn't he read Chapter Two?" 26Fodor notes that a narrow content account may seem to be a "no content" account, but insists on the intentionality of narrowly individuated states:"[A] narrow content is essentially a function from contexts onto truth conditions; different functions from contexts onto truth conditions are ipso facto different narrow contents. It’s hard to see what more you could want of an intentional state than that it should have semantic pro- perties that are intrinsic to its individuation" (Psysem., p. 53, second and third emphases added). 176 something else. Because the New View has it that if there were not physical properties, there would not be any other scientifically interesting pro- perties and, a fortiori, not any scientifically interesting intentional 27 ones, physical properties are ontologically the more basic. If aboutness is real, then, it must (in this sense) really be something else. Fodor's eliminativist premise holds intentional idiom hostage to the exis- tence of a respectable science that deploys it; New View constraints embedded in the premise hold hostage the respectability of a science that deploys intentional idiom to the possibility of there being available both (1) a (theoretically relevant) notion of content that respects superve- nience, and (2) a physical relation such that its instantiation is plausibly sufficient for the instantiating event's being 32223 something. These are notably stiff conditions. We may therefore want to ask, as Fodor has done in the case of classical reductionism, whether we are honor bound to accept them. In particular, we may want to ask what demands on scienti- fic respectability might make them obligatory. Our next step is, therefore, to turn critical attention to New View constraints on respectable science. Predictably, I shall find them less than binding. 27Another way to look at this might be to say that in the case of the requisite token identities we have essentially physical events that are also contingently, special events. This seems consistent both with Fodor’s view that tokenings of physical properties are gathered to special kinds by something less than nomological necessity and with Stephen Stich's inter- pretation of Fodor's computational theory of mind: "On this view, each cognitive state token is a brain state token--its essential type is deter- mined by some neurophysiological property or other."("Narrow Content Meets Fat Syntax", mss. Stress added.) If nothing else, the obscurity of the sug- gestion reflects the unclarity of the notion of token physicalism on which New View doctrine rests. 177 New View claims to normative force have two bases, corresponding to Fodor's twinned allegiances to naturalized epistemology and to a more traditional philosophy of science. Of the two, the latter is, I believe, the more cen— tral. I shall therefore begin by dispatching what might be taken to be empirically based claims to New View authority. It could be argued that the best descriptive account of science derives normative force with regard to scientific practice in the same way that a best account of how children learn to read derives normative force with regard to instructional practice or a good computer—assisted analysis of winning Olympic dives derives nor- mative force with regard to coaching. The best picture of how X happens (or can happen) tells us what we ought to do if X is what we want to achieve. To the extent that the New View tells us how it is that the generalizations of the special sciences can capture lawful regularities, the New View tells us what sorts of considerations ought to constrain the practice of special science and thus what constitutes "good" special science. To make a natur- alistic case for the New View as a set of normative constraints, then, is to display it as a theory that best explains relevant scientific phenomena. This is precisely what Fodor tries to do in "Special Sciences". He begins by drawing attention to the explanatory shortcomings of its rival, classi- cal reductionism, pointing out that as "an empirical thesis"28, it is 29 ”incompatible with probable results in the special sciences." In addition, Reductionism. . . flies in the face of facts about the scien— tific institution: the existence of a vast and interleaved 28LOT, p. 9. 291.01:, p. 19. 178 conglomerate of special scientific disciplines which often appear to proceed with only the most casual acknowledgment of the constraint that their theories must turn out to be physics 'in the long run'. . . .[T]he acceptance of this constraint often plays lbttle or no role in the practical validation of theories. By contrast, the New View can explain not only "why there are special 31 sciences at all," but also why there are (without exception) exceptions to their generalizations.32 I shall urge, against this ”best explanation" argument, first, that the New View falls considerably short of Fodor’s advertisement for its virtues as an explanatory theory, and, second, that even if it were as good an empirical theory as one might wish, this could not, in fact, ground claims to tell us what counts as good or respectable science. To begin with, some of Fodor's criticisms of classical reductionism as an empirical thesis seem equally applicable to the New View. If, in practice, special scientists pay little heed to the classical constraint that their theories turn out, in the end, to be physics, they appear to pay equally little heed to the New View dictum that the events to which their theories refer be physical events, or that there be in view a physical ”mechanism" that explains the existence of the laws on which they rely. As Fodor him- self points out, it is not generally the case that 30LOT, p. 24. Fodor's argument is actually a bit more complicated than this. He attributes an "epistemic" explanation of these phenomena to the reductionists and then proceeds to argue against that. 31LOT, p. 23. 32LOT, pp. 23—4. 179 one literally has. . . a [mechanical] specification in hand for each functionally individuated component of each psycho- logical theory that one is prepared to take seriously. In practice, the flow of argument usually goes in the opposite direction: . . . .[W]e infer from the--independently patent --existence of the psychological process to the possibility of its mechanical . . simulation, rather than going the other way around. And if classical reductionism runs counter to our best hunches about "the way the world is put together,"34 if there is little reason to believe that, as a matter of fact, all the counterfactual supporting regularities there are will turn out to be expressible in the language of physics--i.e., that physical type identity holds--there seems equally little reason to assume that, as a matter of fact, physical token identity holds--i.e., that events individuated under any and all theoretically successful description will turn out to correspond with events individuated under the description of basic physics. John Haugeland offers an example of such failure. "Robust" events of everyday discourse, he notes, do not necessarily correspond to events char- acterized mathematically at the level of physical theory. If, for example, two people initiate waves that hit the same floating cork simultaneously, there are, in everyday language, two distinct "robust" events: "At the level of description of wave-hits, there is ygggg and there is mine, and 35 (at that level of description) these are not the same." At the level of 33Intro., Rep., p. 15. 3‘L0T, p. 24. 35John Haugeland, "Weak Supervenience", Am. Phil. Quarterly, 19 (1982), (Footnote continued) 180 "mathematical" events, however--state descriptions in the language of physics—-we notice that when we turn up our microscope. . . and look at the positions and velocities of the water molecules, there isn’t a trace of either wave-hit to be found anywhere. . . . None of the par- ticular positions or velocities at the molecular level is determinately a constituent of one wave-hit (or the other, or neither)—-though each makes a contribution to the overall effect. . . . Wave—hits simply don't "hang together" as de- terminat§6individuals. . . at the level of particle physics. More generally, Haugeland notes, [Tlhe individuals, or "tokens", of which our sentences are true are just as "relative" to. . . description as are the kinds or "types" into which those sentences sort them. The world does not come metaphysically individuated any more than it comes metaphysically categorized, prior to and independent of any specific descriptive resources. . . . [Tlhe indivi- duals we can discuss in one way of talking need not be identified with the individuals we can discuss in another. 35(continued) hereafter, WS, p. 100. Stress added. 36vs, pp. 100-101. 37WS, p. 101. stress added. Frederick Stoutland makes a similar point in his discussion of the individuation of actions: "The notion that only physical descriptions are intrinsic and play the privileged role of identifying the particular which is the descriptum for other descriptions is, I believe, mistaken" ("Davidson on Intentional Behavior", in Ernest LePore and Brian P. McLaughlin, eds., Actions and Events: Perspectives 25 the Philosophy of Donald Davidson (Oxford: Basil Blackwell, 1985), p. 55). Hereafter, A&B. Fodor acknowledges that individuation is relative to a vocabulary in the course of making another point in Psychological Explanation. Perhaps. . . what is required in order for physics to be gen- eral is that, for each event, there be some description in terms of physical magnitudes. But this suggestion is possibly too strong and certainly unclear, since it inherits and car- ries forward the vagueness of presystematic criteria for individuating events. At 2:00 a.m. we go on daylight-saving time. Is what happens at 2:00 a.m. [identical with] a physi- (Footnote continued) 181 If it is a contingent matter whether the language of psychology and the language of physics carve the world into the same particulars, we may have reason to find ourselves as pessimistic about the token identities on which the New View insists as we are about the existence of the type identities required by classical reductionism. Given the passing of the neat "wet flip-flops” view of brain functioning, there is less reason to suppose that the meta- bolic constituents of mental events can be identified in the midst of all the irrelevant physiological housekeeping;. . there is more reason to suppose that distinct mental events will each "supervene on" (the activity in) extended brgén regions, which may largely, or even entirely coincide. We may also question Fodor's claim in "Special Sciences" that the New View explains not only how the special sciences can capture genuine lawful regu- larities but also why we may anticipate exceptions to their explanatory generalizations. Some of the tokens gathered to the predicates featured in a special law may fail to be linked to each other by more basic laws. The existence of such nomologically unmoored events accounts, we are told, for the "noise" that exists at every level above basic physics. But where, it might be asked, do these unmoored instances of special properties comefrom? Surely nothing can count as an instance of a property of a special science 37 (continued) cal event? And, if there is no clear answer to this sort of question, must we therefore abandon the search for physical theories for which generality must be claimed? (Pp. 11—12.) Where Fodor differs from Haugeland and Stoutland is in his insistence that any respectable science must individuate tokens in the same way that physics does. 38WS, p. 101. This point, currently urged by connectionists, is precisely parallel to Fodor's observation that despite "institutionalized gambles" on their existence, "There are no firm data for any but the grossest corres- pondencels] between types of psychological states and types of neurological states" (LOT, p. 17). 182 39 Even were we to fill in the that does not behave according to its laws. theoretical gaps, moreover, observation fails to support Fodor's claim that "one of the things that the hierarchical arrangement of our sciences buys for us" is the ability to explain exceptions to the generalizations of the special sciences by "golingl 'down' one or more levels and. . . [using] the 40 Fodor's own example of unsuc- vocabulary of a more 'basic' science." cessful prediction at the intentional level--someone's failure to arrive at the airport at the time promised--could easily have as its best explanation a story that adverted to such events as a strike by airline attendants or a thunderstorm over Dallas. In neither case do we explain the failure of pre— diction by going ggyg. (To neural mechanisms? Whose?) The relevant direction of explanation might better be described as up--the flight atten- dants were on strike because inflation had diminished their buying power and a stagnant economy militated against their being given better pay--or even sideways--it is hard to say where meteorology stands vis a vis inten- tional psychology. It may, perhaps, be worth noting that Fodor's own endorsement of the downward explicability of exceptions sounds a bit quali- fied these days: the following passage from Psychosemantics includes an 39One possibility is that a given special predicate, S , figures in a variety of special laws. The physical properties on which the regularities expressed in one such law depend may not overlap entirely with those underlying the regularities expressed in another. S would thus have the structure of a family resemblance concept, constitufed by a lot of partial overlaps. Could such a concept be a kind term? It is by no means clear. Another possibility is that the procedures whereby special events are identified are fallible and the unmoored events are not, in fact, really instances of the special property in question. On this account, however, it would be hard to understand how physics could be exceptionless. 40Psysem., p. 6. 183 interesting assortment of hedges. Exceptions to the generalizations of a special science are typically inexplicable from the point of view of. . . that science. That's one of the things that makes it a special science. But of course, it may nevertheless be perfectly pos- sible to explain the exceptions in the vocabulary of some other science. In the most familiar case, you go 'down' one or more levels and use the vocabulary of a more 'basic’ science. (The current failed to run through the circuit be- cause the terminals were oxidized; he no longer recognizes familiailobjects because of a cerebral accident. And so forth.) Nonetheless, Fodor is still willing to make the unlikely claim that when we encounter gaps in the explanatory story of commonsense psychology, we will be able to account for them "in the vocabulary of some lower—level science (neurology, say, or biochemistry; at worst, physics)."42 We may have yet more serious reservations about an empirical basis for New View normative claims. On the analogy we are using, earning authority over the practice of X-ing requires being the best available story about how X-ing occurs or is possible. So earning authority over the doing of science --getting to tell us what counts as respectable science and what does not—— requires being the best available explanation of how such good science happens (or is possible). But to put it this way is to see that there are problems in applying the analogy to the case at hand. For the New View to be the bggg explanation of how the special sciences are managing to capture genuine lawful regularities, it must be an explanation of the fast that they are doing so, and thus, presupposes that they are doing s9. For the 41Psysem., p. 6. Stress removed. 42Psysem., p. 6. 184 New View to be the best explanation of the scientific success of the spe— cial sciences, there must, therefore, be some logically prior determinant of scientific success. (For a computer analysis to reveal how winning Olympic dives are achieved there must already be winning dives; for a cog- nitive theory to explain how children learn to read there must already be something that distinguishes readers from non-readers. And so on.) But scientific success and scientific respectability are, in this context, the 43 So for the New View to tell us what counts as respectable same thing. science by virtue of its being the best explanation of the scientific suc- cess of the special sciences, there must be some way, independent of the application of New View standards, to characterize what counts as respect- able science. The New View cannot, therefore, claim both to be the best theory of how respectable special science is possible Egg (because it is the best theory) the arbitor of what counts as respectable science. I have been arguing that New View claims to normative authority based on its status as an empirical theory are open to challenge. Fodor has not put all his eggs in this basket, however. The virtues of classical reductionism that the New View is suppposed to preserve are philosophical ones. As advo- cate of a liberalizing position, Fodor is primarily concerned in "Special Sciences" to establish that New View constraints are sufficient to meet these philosophical demands. His qualms about intentional idiom, however, 43If what the New View claimed to explain was simply the sociological fact that there are more and more special sciences, this would not be relevant to its claims to tell us what counts as respectable science. The most that could be said on the basis of its success in this department would be that we ought to heed New View doctrine if we wanted to get funded. 185 spring from his conviction that they are also necessary. It is to this claim that I shall now turn critical attention. We shall want to take a careful look at the New View as normative doctrine and at a range of gener— ally acknowledged philosophical obligations that might seem to make its constraints binding. I shall argue that, appearances to the contrary, the New View is not rendered obligatory by commitments to minimal monism, faith in scientific progress or defense of its cognitive claims, deference to the special role of physics, the avoidance of blatant pseudoexplanation or con- sensual views about the nature of the scientific enterprise. What it takes to clinch New View constraints is, I shall argue, premises with regard to scientific theory that are far from uncontroversial. On the New View, a science is respectable only if token physicalism and supervenience hold. A theorist who endorses New View constraints may thus be seen as embracing a tripartite normative doctrine: a science is respec- table only if the laws that it invokes (1) stand in a hierarchical relation of explicability to (2) a unique body of laws which (3) is identical with the completed physics. More formally, the New View places the following conditions on respectable science. Hierarchy [H]: Say that for any law, L, of a theory, T, and laws L', L". . . of (nonidentical) theories T', T". . ., L', L". . . explicate L directly iff the existence of L', L",. . . figures essentially in an explanation of the possibility of L, and explicate L indirectly iff they explicate directly some law that explicates L. Then, T is scientifically respectable only if either T is basic physics, or, for any L in T, there exist L', L",. . . such that L', L",. . . explicate L (either directly or indirectly). 186 [H] says that a necessary condition on the scientific adequacy of a theory is that its laws either stand in a vertical relation of explicability to some more basic laws or are themselves laws of physics. Monopoly [M]: There is one and only one theory, T, whose laws explicate, either directly or indirectly, any law that requires explication. [M] says that there is a single theory, T, the existence of whose laws can explain the possibility of the laws of all other theories. The laws of T thus explicate, directly or indirectly, all other laws invoked in adequate explanations. [P]: T, the unique theory satisfying [H] & [M], is the completed physics. Together, [H], [M], and [P] entail the New View doctrine that no science is respectable unless the laws that it invokes either are those of basic physics or are explicable, directly or indirectly, by the laws of basic physics. It is in this sense that physics is, on the New View, "basic" science. Let us take up each of these conditions in turn, asking whether it is entailed by any philosophical commitment that we should be unwilling to give up. I shall begin by considering what might seem to be a quite plausible claim: that New View normative doctrine in general and [H], in particular, is rec— ommended to us because it protects functional explanation from explanatory defects to which it is peculiarly vulnerable, and that it does so by enforcing the existence of relevant physical realizations. In "Special Sciences", Fodor tells us that the pgipt of reductionism, either of the type-type or the token-token variety, is "to explicate the physical 187 mechanisms whereby events conform to the laws of the special sciences" (1975, p. 19). This is a bit cryptic as it stands, but elsewhere the need to give due regard to underlying physical mechanisms is linked directly with regard for the integrity of explanation. We are, for instance, offered two examples of failed functional explanation--first, reliance on the ”dormitive power" of morphine to explain why it puts people to sleep, and second, invocation of a Perfect Question Answerer to explain "the (occasional) human capacity to provide true answers to questions posed."44 We are also provided with a diagnosis and a remedy. The trouble, we are told, is one of "question-begging functional explanations [that] depend on postulating processes for which no mechanical realization is known, or for which none can be imagined, or--in the extreme case--for which none can exist."45 The solution is "to allow functional individuation only where there exists a mechanism that can carry out the function and only where we 46 Is [B] have at least some idea of what such a mechanism might be like." necessary to block obvious explanatory failures of the sort to which Fodor draws our attention here? The first thing to note is that we are, in fact, being confronted with a number of significantly different cases. At least one of them-~invocation of a process whose "mechanical realization" is currently unknown--is not, as far as I can see, an uncontroversial defect in scientific explanation. 44Intro., in Rep., p. 12 45Intro., in Rep., p. 15. 46Intro., in Rep., pp. 12-13. 188 Although it would surely be pigs to have on hand an account of all the reg- ularities presupposed in our explanations and, perhaps, to be able to write the bottom line in the language of basic physics, it is possible, as Hilary Putnam has argued, to have a perfectly satisfactory--indeed, perhaps the only possible-~explanation (of why a cubical peg fifteen—sixteenths of an inch on a side goes through a square hole one inch on a side but not through a circular hole one inch in diameter) that stops short of such 47 ultimate reduction. Haugeland notes in a similar vein that we could have a bona fide explanation of how a fiber optics bundle conveys images without 48 It is having an explanation of how its constituent fibers convey light. one thing to prefer among competing explanations the one for which there is (or promises to be) such a "deeper" account, and even to live in hope and faith that such explanation will eventually be not only available but expressible in the vocabulary of physics; it is quite another thing to make such availability a condition on acknowledgement of explanation as such. If Putnam and Haugeland are right, as I think they are, leaving unexplained regularities presupposed in a purported explanation is not, in and of itself, enough to warrant its rejection. An account that leaves such questions open—~perhaps even permanently--is not the same as a question-begging nonexplanation. 47"Philosophy and Our Mental Life", in Mind unguage and Reality: Philosophical Papers, Vol.2 (Cambridge: Cambridge University Press, 1975), pp. 295 ff. ‘snrhe Nature and plausibility of Cognitivism" in Mind Design: Philosophy, Psychology, Artificial Intelligence (Montgomery, VT: Bradford Books, Publishers, Inc. , 1981), pp. 246. 189 There is little question, however, that 'Because morphine has dormitive powers' is, at least in answer to 'Why does morphine put people to sleep?', an unsatisfactory explanation, and equally little doubt that, regardless of the query, 'Because we are equipped with a Perfect Question Answering Device' will be so. I should like to argue that we do not need New View constraints to protect against the defects that these cases exemplify. If we take explanation to be a special case of discourse governed by demands for cooperation, both cases can be seen as violations of broadly applicable conversational maxims and therefore as unsatisfactory on that ground alone."9 Let us begin with the impossible Perfect Question Answerer. The maxim of Quality prohibits generally contributions to a conversation that are likely to lead one's partner astray--ungrounded assertions, misleading implica- tions, and so on. False (or misleading) premises risk conclusions that are false (or misleading). The maxim of Quality therefore leads us to expect that, in particular, invocation of false premises in an explanation will be unsatisfactory. If I ask, "Why is the cream sour?" and you reply, "Well, Martha looked at it; Martha is pregnant, and for all cream and all lookers, if the locker is pregnant, the cream sours," you can predict that I will reject the explanation because the generalization being invoked is false. (Martha's being a virgin or in another city will have the same corrosive effect.) One particular way in which a premise can be false (or misleading 49H.P. Grice, "Logic and Conversation" in The Logic pf Grammar, ed. D. Davidson and G. Harman (Encino, CA: Dickenson Publishing Co., Inc., 1975). 190 --I shall, in the interests of brevity, omit this from now on) is for it to include nonreferring terms in referential positions. Thus, it would be equally unacceptable to say that the cream was left out overnight, that elves come by at night and bathe in available cream, and that for all elves and all cream, if an elf bathes in the cream it turns sour. Since there are no elves, this is an inadequate explanation. We may now offer an alterna— tive diagnosis of the deficiency of explanations featuring a Perfect Question Answerer: since a Perfect Question Answerer is a logically impos— sible device, it certainly isn't actual, and if it does not exist, any explanation that presupposes its existence invokes a false premise and is ipso facto unsatisfactory. We do not, it seems, need [H] to rule out func- tional explanations that rely on mechanisms known to be impossible of realization. What about Moliere's complaint against invoking the dormitive powers of morphine to explain why it puts people to sleep? Fodor notes that the ac- ceptability of such a response is context relative and my intuitions agree.so Consider: (1) Why does morphine put people to sleep? It has dormitive powers. (2) Why do doctors prescribe morphine? It has dormitive powers. 50"[Tlhere are. . . imaginable circumstances where an etiology that adverts to the dormative [sic] power of morphine might be. . . illuminating. (what with explanation being interest relative and all that)" (Intro., Rep., p. 12). 191 (3) Why do I feel so sleepy when I take this stuff? It contains morphine and morphine has dormitive powers. (1) is unacceptable under any circumstances; (2) and (3) might be accep- table under some conditions but not others. Such context relativity reflects, I would urge, the operation of the maxim of Quantity, which in— structs us to give a listener just as much information as is needed: since needs vary from one situation to another so will explanatory adequacy. Assuming that a request for an explanation is a request for at least some information not already available, allusion to dormitive powers in answer to ’Why does morphine put people to sleep?’ can never be acceptable because it can never provide any information. The inquirer already believes that morphine puts people to sleep; that is a presupposition of appropriately posing the question. To say that morphine has dormitive powers is to add nothing to what the questioner already has available. What Moliere and we object to here is not that a question about underlying mechanical realiza- tion that should always be answered has been left open, but, rather, that no information whatever has been provided.51 51Notice, however, how carefully (1) has to be worded and how much we have to assume about the situation for the response to be utterly uninformative. Had the question been, ’Why do people go to sleep after they take morphine?’, and our listener someone who took placebo effects (or Occasionalism) seriously, 'It has dormitive powers’ would have been a potentially informative and thus an explanatory answer. It is also clear that the deficit I am attributing to (1) can be shared by nonfunctional explanations as well. Consider Why is BettyJo so fat? Because she's obese. 192 It may be of interest to note that Fodor himself has sketched an account of what goes wrong when we invoke dormitive powers that is similar to the one I have been urging. "Flowers are red because of the presence of red- producing entities" is informative, he tells us, even though "the seemingly parallel remark in Moliere about morphine and ’dormitive power’ is not," first, because no one doubted that logically independent des- criptions of the cause and the effect must eventually be forthcoming and, second, because the data about ”directly observable" traits permitted theorists to establish the fact that quite subtle relations of dominance, recessivity52 connectedness, and so on must hold between the genes. Scientists routinely specify candidate kinds "by reference to precisely the appearances that they function to save;" theory construction is the task of "working out relations between the entities so specified in a way that ex- tends the range of phenomena their postulation explains."53 If the range of phenomena is adequately broad, the entities originally so specified feature in genuine explanations. Two things are worth noting here. First, the ade- quacy of a candidate explanation is explicitly taken to depend upon its informativeness--it is because logical dependence of cause and effect diminishes informativeness that we aspire to eliminate it. Second, infor- mativeness is characterized entirely in terms of horizontal liaisons. With regard to the cases under consideration, then, [H] seems, both on my account and this one of Fodor’s, to be unnecessary to explanatory adequacy. 52PE, p. 36. 53PB, p. 37. 193 Neither in the case of the Perfect Question Answerer nor in that of dormi- tive powers does our account of what goes wrong need to allude to a lack of appropriate vertical relationships between a regularity invoked in (or pre- supposed by) an explanation and any more basic laws that explain its possibility.5“ Let us put aside our objection to [H] and ask why conformity with [M] might be considered an additional condition on scientific respectability. Why might it be thought that one single theory grounds every other one? The ap— peal of unity of science is, I think, heavily overdetermined and difficult to unravel for critical examination. One strand may be the very general thematic dominance of unity over diversity in Western European thought and, in particular, a tendency to equate diversity with disorder. Discussing the appeal of classical reductionism, for example, Fodor notes that [I]t is tempting to suppose that there must be only one prin- ciple of sorting. . . on pain of there otherwise being chaos, that either there is 925 kind of sggentifically relevant similarity or there is every kind. Fodor takes care to reassure his reader that the existence of irreducibly diverse kinds is not a threat to cognitive order, that there is still a 54Similar pragmatic explanations of the unsatisfactoriness of accounts that involve us in infinite regress—-e.g., cognitive theories that rely on in- telligent homunculi and "inner eye" stories about perception—_are possible. Insofar as the question that is left open after the reply is precisely the one that was originally posed, nothing informative has been said. (We feel, I think, a particular sense of affront in cases like these, where violation of a maxim is unacknowledged, unobvious, and serves no communicative purpose.) I do not, of course, claim that the adequacy of all explanation will be amenable to Gricean analysis. 5592, p. 119. 194 principled basis for distinguishing legitimate taxonomies from illegitimate ones. What justifies a taxonomy, what makes a kind "natural," is the power and generality of the theories that we are enabled to formulate when we taxonomize in that way. . . . [I]f we can find. . . principles that permit simple and powerful accounts of the etiology of behavior, then that is itself an adequate juggification for sorting according to those principles. If, as Fodor urges here, it is the power and simplicity of a theory that justifies its vocabulary, and if, as Haugeland and others have pointed out, the way we individuate events depends upon our choice of a theoretical vo— cabulary, then it is the power and simplicity--the "goodness"--of a theory that should settle how tokens are to be individuated for scientific purposes rather than compliance with the demand that all events be indivi- duated according to a single theoretical vocabulary. It is not mandated monopoly upon which we rely to ward off cognitive chaos, rampant relativism and other unspeakable horrors, but, rather, the considered judgment of scientific communities with regard to the goodness of competing theories.57 Other considerations may also lend plausibility to the assumption that an explanatory hierarchy must have a single base. It may, for instance, appear that our faith in the progress of science entails something like [M]. Fodor notes that one way to think of reduction is as the subsumption of less 56PB, p. 119. The normative force of ’natural order’ is pretty obvious here. Passages like these two lend plausibility to neo—Durkheimian warnings that it is all too easy to select one’s epistemology on the basis of one’s social ideology. 57That it must be physical theory, in particular, that holds the monopoly is an issue to which we shall return. 195 general laws by more general ones and thus as the achievement of laws of greater and greater explanatory power. On this understanding, laws of the greatest possible explanatory power are possible only if there is some one ultimate set of laws that subsumes all others. Unwillingness to endorse [M] may thus appear to be a vote of no confidence in the future achievements of science. Two remarks are in order here. First, even a classical reduction- ist might want to distinguish between our highest hopes for science and what we would be well advised to accept for the time being. A concern that may legitimately be consulted in choosing between theories may be inappro- priate when applied as a do or die condition: as they used to say in the Sampling Section of the Survey Research Center, "The worst enemy of the good is the perfect." Second, and more to the point here, the New View’s rejection of the classical demand for subsumption of laws makes the whole question of faith in scientific progress, seen as such subsumption, irrelevant to acceptance of [M]. [M] may also appear to be a consequence of a minimal monism, [MM], that maintains that there is one and only one kind of stuff. This, however, is not the case. [MM] is an ontological doctrine; [M] is an epistemic one. [MM] may be a necessary condition for [M]’s possibility--if there are dras- tically different kinds of stuff, the hope that all laws will be explicable 58 by laws of a single theory seems dim --but [MM] surely does not entail [M]. There might well be only one kind of stuff without our being able to 58Especially if kinds of stuff are differentiated by virtue of the laws they obey. 196 formulate laws in such a way that the possibility of their existence could be explained by the existence of laws expressed in some one theoretical vocabulary. Fodor makes the point briskly: [I]t was a pervasive and characteristic error in positivistic thinking to infer the unity of science from the unity of the subject matter of science°9viz. the epistemological thesis from the ontological one. Here, as elsewhere, Fodor’s early observations deserve our careful attention. Finally, acceptance of pluralism of any sort may seem to undermine the cog— nitive claims of science. Ian Hacking has proposed that attachment to a doctrine of unity of science may have roots in the rhetorical role played by the metaphor of the Book of Nature, written by One Author in One Language, at a point in the history of science when the claims of empirical 60 inquiry needed support against more familiar textual authority. In the 59C&R, p. 324, n. 60Ian Hacking, The Emergence of Probability: A Philosophical Study of Early Ideas About Probability (Cambridge: Cambridge University Press, 1975), p. 45. Hereafter, EP. Writing to the Grand Duchess Christina, for example, Galileo addressed the resolution of conflict between Biblical and scien- tific views of the universe by observing that "the holy Bible and the phenomena of nature proceed alike from the divine Word," but noted pointedly that while Biblical truths may be obscured by the forms of words used to express them to imperfect understandings, Nature, "as the observant executrix of God's commands," necessarily reveals God’s laws directly. (Discoveries and Opinions of Galileo, Trans. Stillman Drake (Garden City, N. Y. : Doubleday Anchor Books, 1957),p pp 182- 3.) A century or so earlier, Paracelsus had written [Tlhe firmament is. . . a book containing all virtues and all propositions [. . .] the stars in heaven must be taken together in order that we may read the sentence in the firma- ment. It is like a letter that has been sent to us from a (Footnote continued) 197 period under consideration, unity was taken to be a perfection and thus a necessary attribute both of the Author and, derivatively, of the Author’s works. None of this, of course, seems even remotely relevant to the ques— tion of contemporary commitment to unity of science. Science itself is now the major source of cognitive authority; were its cognitive claims to be attacked, allusion to the Book of Nature would hardly seem a useful line of defense. It is nevertheless possible that commitment to some single source of truths--some single kind of facts to which all others are, in one way or another, beholden--still derives strength from an unexamined assumption that admitting diversity somehow weakens the claims of the scientific en- terprise.61 Vivid as they may be, none of the concerns at which we have looked so far 60 (continued) hundred miles off, and in which the writer’s mind speaks to us (EP, p. 41). Finding a place for intentionality in the natural order was not, it seems, a big problem in those days. 61Ian Hacking also made this point in his lecture, "The Disunity of Science", Nobel Conference XXV: The End of Science?, Gustavus Adolphus College, St. Peter, MN, October 3, 1989. Elsewhere, he has offered the observation that the history of science has, in fact, tended more toward diversity than conceptual unity, concluding, "The ideal end of science is not unity but absolute plethora." God did not write a book of Nature of the sort that the old Europeans imagined. He wrote a Borgesian library. . . . For every book, there is some humanly accessible bit of Nature such that that book, and no other, makes possible the comprehension, prediction and influencing of what is going on. . . [Tlhis is New World Leibnizianism. . . [Tlhe best way to maximize [the variety of] phenomena and have simplest laws is to have. . . [laws] each applying to this or that but none applying to all. (R&I, p. 218.) 198 oblige us to accept the conjunction of [H] and [M]. It may be objected at this point, however, that in considering [H] and [M] apart from [P], we do less than justice to the New View. [H] and [M], it might be argued, are not necessary because we need a relation of explicability to just some or other uniquely ultimate theory in order to meet requirements on scientific expla- nation but, rather, are entailed by the special role in the natural order played, in particular, by physical law. It is, perhaps, our respect for 62 what has been called "the primacy of the physical" that leads us to accept [H] & [M] & [P] as a package. How might this be thought to work? Suppose, for example, that we are not just minimal monists but share a 3 that "without 64 ”compelling sense that the universe is basically physica ,"6 any mggggg there wouldn’t be anything else contingent either." Unlike [MM], which is a purely metaphysical thesis, this sort of physicalism seems to have epistemic consequences. What could we mean by ’physical’ (or ’material’) if not ’that which obeys physical law’? And if what we want to endorse is the claim that everything there is obeys physical law (and that there is no other law of this scope), then surely physics, as the science whose theory expresses such law, deserves special deference. One way to characterize the special role of physics might then be to say that it has asymmetric veto power over the other sciences: what the laws of physics prohibit, no other science may license (or require); what physics ordains, 62vs, p. 96. 63WS, p. 96, stress added. 64John Haugeland dubs this last "vapid materialism" in "The Intentionality All-Stars", to appear in Philosophical Perspectives, IV: Philosophy pf Mind and Action Theory, James E. Tomberlin, ed., p. 3. 199 65 It might appear that physicalism so construed no other science may bar. entails acceptance of the New View ensemble even if minimal monism does not. Here is one way in which an illusion of such entailment might be created. A theorist who acknowledges the veto power of physics will want to reject explanations that presuppose entities or processes that violate laws of physics. The New View also demands such rejection. (If the laws of physics preclude the possibility of X, they cannot, a fortiori, explain X’s possibility: thus, what the laws of physics prohibit, they cannot explicate, and, according to the New View, what the laws of physics cannot explicate cannot legitimately be featured in explanation.) It might therefore appear that the veto power of physics entails New View constraints. This argument has, however, the form of a familiar logical fallacy.66 We have, moreover, already seen that we do not need doctrines of hierarchy or monopoly to bar explanations featuring perpetual motion or the intervention of Maxwell’s Demon. In company with conformity to ordinary conversational maxims, commitment to the veto power of physics is itself enough do the job. 65Other sciences may, of course, bar what physics permits: not every physi- cally possible system is a biologically possible one. Haugeland’s notion of "nomological supervenience" is an attempt to capture the "special authority of the physical" in something like the sense that fixing the physical de— termines everything else. For discussion of whether this can be done short of token physicalism see, for example, Jaegwon Kim, "’Strong’ and ’Global’ Supervenience Revisted", Philosophy and Phenomenological Research, 48, 2, Dec. 1987, pp. 315-326. 66 A-->C B-->C A-->B Spotted by Richard Hall. 200 We are still not obliged to accept New View constraints--not by rejection of garden-variety pseudoexplanations, not by optimism about the future pro- gress of science or concern for its cognitive claims, not by minimal monism or deference to the specialness of physics. What remains? It may be argued that the notion of adequate explanation relevant here is, in fact, one more narrowly specific to the context of scientific inquiry. The New View theo- rist might argue as follows. "Look—-the whole pgipg of the New View as normative doctrine is to display necessary conditions on the adequacy of scientific explanation. Now surely we are convinced that (1) the sciences 67 are in the business of giving causal explanations, that for an enterprise to be a science, terms like ’cause’ and ’effect’ and ’event’ must be predi- cable of its theoretical entities.68 And it can hardly be denied that (2) causal explanations rely upon causal generalizations-—that is, on general— izations that subsume the things to which they apply in virtue of the 69 But, as we all know, (3) whatever has causal powers of those things. causal powers is ipso facto material (which is to say, physical).7O So, (4) an enterprise can only provide scientifically adequate-~i.e., causal-— explanations if the events referred to by its theoretical terms are identifiable with physical events--i.e., with events as individuated by the completed physics: anything to which an explanatory taxonomy applies "had 71 better be a physical thing." And to say this is simply to endorse the New 67Psysem., p. 34. 6392, p. 15. 69Psysem., p. 34. 70 Psysem., p. x. 201 View as normative doctrine." As far as I can tell, this is, indeed, Fodor’s position. In what follows, I shall argue that the appearance that the New View is entailed by a clutch of unyielding philosophical commitments with regard to the very nature of science is due to the existence of two readings of ’causal explanation’. One of these renders relatively uncontroversial the claim that the sciences necessarily deal in causal explanations, causal generalizations, and "causal powers"; the other one entails New View token physicalism. Neither, however, does both. First, then, let us consider the sense in which it is (at least reasonably) uncontroversial that science must provide causal explanations. This is the sense in which causal explanation is often contrasted with "conceptual" enterprises--analysis, explication, and so on. The distinction is a prag- matic one: achieving prediction, control, and a view of the world as (in at least some respect) orderly is a particular BEE to which language can be put, one that differs in interesting (if hard to specify) ways from the working out of logical relations among formulae, the giving of legal opinions or the naming of babies. It is to this pragmatic contrast that Fodor draws attention, for example, when he notes that in asking, ’Why is Wheaties the breakfast of champions?’, we may either be looking for "a conceptual story"--asking by virtue of what Wheaties counts as the break- fast of champions-~or we may be wondering what properties of Wheaties are 72 responsible for Wheaties-eating’s bringing about championship-winnings. Fodor calls an appropriate answer to this latter sort of question a "causal 71LOT, p. 25. 202 story". Such answers "suggest provisional values of P in the explanation schema : ’2 causes ((5 eats Wheaties) brings about (x becomes a champion)) for significantly many values of x’."73 Answers of this sort are what we expect from a science; answers of the "conceptual" variety are not. We shall want to look closely at what it takes to satisfy this schema for causal explanation. Notably, for all we are told here, there is no reason to imagine that an instance could not be made true simply by virtue of the prediction, control and understanding conferred by the theory that provides the stand-in for ’P’--by what Fodor elsewhere describes as the theory’s 74 "power and generality." The charge that an account is not 19 this sense a causal explanation can be countered by showing that the theory in quesiton does, in fact, make prediction and control possible, does, indeed, produce order among phenomena. It is in this sense that, as Fodor tells us earlier, whether what we have on our hands is a causal explanation or a "vacuous" account--a statement of conceptual or "logical" relations--is a matter of the use to which it can be put, on "the background of theory and experiment in which the explanation is intended to function, including, perhaps, the scientist’s . . . expectations about the kinds of theories in which the ex- 75 planation mgy function at some future date." If we read (1) in this light, it says, uncontroversially, that we are likely to withold 72LOT, pp. 6-8. 73LOT, p. 6. 7493, p. 119. 75PB, pp. 36-7. 203 application of ’science’ from enterprises that are not in this particular line of order—making work or that are, perhaps, not doing it very well. This reading of (1) determines, in turn, an equally uncontroversial inter— pretation of (2). If we understand giving a causal explanation as the accomplishment of a kind of local pattern-making, then our ability to do so seems clearly to depend upon the availability of more generally applicable resources. To be a causal generalization is, thus, simply to be a formula- tion apt for the creation of order across (as Fodor puts it) "significantly many" situations. Talk of generalizations’ subsuming the things to which they apply by virtue of their "causal powers" is, in turn, nothing more than a mildly provocative way of noting that scientific explanation relies upon generalizations that feature theoretical predicates or property terms. On this view, individuals come by causal powers by virtue of satisfying one or another of the predicates of a theory that "works". Causal powers trick— le downward, so to speak, from a theory’s overall success to applications of its predicates. This altogether vanilla reading of (2) commits us, how— ever, to a reading of (3) that is much less innocent. (3) now says that physics is, in fact, the only order—making theory in the game. But if, as Fodor urges, the special sciences are not only possible but actual then (3) is, on this reading, false: it is simply not the case that nothing has cau- sal powers (as we are now interpreting them) save by virtue of being an instance of a property acknowledged by the completed physics. Quite the contrary: on the view under consideration, an event can come by causal powers by virtue of instantiating a biological or a chemical or a 204 geological (or, if things turn out right, a psychological or even a sociological) property. To say that no enterprise outside of physics can make manifest an order seamless enough for such utility to be credited to its generalizations not only requires something fairly ornate by way of philosophical defense but is clearly inconsistent with Fodor’s own claims for the success of the special sciences. If the special sciences deliver counterfactual supporting generalizations, enable prediction and facilitate control, then events have causal powers by virtue of being instantiations of special properties. Physics may be first science, but it is first among equals. If we reject (3), acceptance of (1) and (2) does not carry us to (4), which is to say, to New View token physicalism. Why might it be thought that it would? The impression is due, I think, to a second way of understanding ’causal explanation’, one that involves a different sort of contrast, and that does indeed make (1) and (2) entail (4). This way of understanding ’causal explanation’, however, throws (1) and (2) into question. On this second account, the contrast between conceptual stories and causal ones has to do not with what we can accomplish with a theory, but with the sorts of entities and processes to which its theoretical terms refer. If we think of the first way of drawing the contrast as being made in the formal mode, we may think of this one as being drawn in the material mode; if we think of the first contrast as epistemic, we may think of this one as ontological. Neither of these descriptions quite fits--the important point to keep in mind is that two different distinctions are in play. Under this second in- 205 terpretation, the crucial thing about causal stories is that they are about motions and the sorts of entities that are capable of producing or modify- ing motions. They are about ”mechanisms" in the everday sense. To mistake psychological explanations for causal stories is, on this view, to miscon- strue psychologists’ talk about perceptual processes as talk about the grindings of ghostly wheels. To insist that psychology really dpgg provide causal explanations (in this sense) is to hold oneself responsible for showing that the entities to which its theoretical terms refer can play causal roles analogous to those played by real gears and cogs--i.e., by virtue of their physical shapes--and to try to gain yardage by describing mental processes along lines such as these: "[W]hen you intend to make it true that P. . . . you put into the intention box a token of a mental sym- bol that mggpg that E. And what the box does is, it churns and gurgles and computes and causes and the outcome is that you behave in a way that (ceteris paribus) makes it true that P."76 It is not surprising that Fodor should take seriously this second notion of causal explanation. We may recall that he began his career contending against (inter alia) adversaries who relied on it to make a reductive case against psychology that opened more or less as follows: "[Plsychology must be a causal science if it is a science at all. But. . . causal analysis in the strict sense can only demonstrate the contingency of motions of one kind on motions of another 77 kind," and so on. To construe ’cause’ mechanically, Fodor notes, requires, yet more stringently, that there also be "spatial and temporal 76Psysem., p. 136. 77PB, p. 40. 206 78 It is, I believe, this sense of contiguity between the movements." ’causal explanation’ that makes it important to Fodor to carry out the "Special Sciences" mission of explicating "the physical mechanisms whereby 79 and that mandates the supervenience of events conform to special laws" psychological states on brain states on which he insists in Psychosemantics. Thus: Metaphysical point: Causal powers supervene on local micro- structure. In the psychological case, they supervene on local neural structure. We abandon this principle at our peril; mind/brain supervenience (lidentity) lg our only plausible account pf hgy menégl states could have the causal powers that they d9 have. "[I]f mind/brain supervenience goes," Fodor tells us, what goes with it is "the intelligibility of mental causation,"81 which is to say, the possibility of there being psychological explanation that is, in this sense, causal. In contrast with the first interpretation, this reading yields a picture in which causal powers are relayed gpyggd, from physical events that have them (by virtue of instantiating properties of the science that treats of mo- tions and mechanical relations), to the "special" properties instantiated by these events. Understood in this way, causal explanation obviously re— quires generalizations that refer to events describable in the vocabulary 78PE, p. 41. 79LOT, p. 19. 80Psysem., p. 44. Stress added. 81Psysem.,p. 42. 207 of physics, and thus entails token physicalism. We can get to the New View, then, if we see science as essentially concerned with causal explanation 229 interpret ’causal explanation' in a way that demands reliance on gener- alizations that refer (ultimately) to local, physical interactions.82 What is required to arrive at New View token physicalism as a normative con— straint on the special sciences thus appears to be nothing other than token physicalism applied to scientific explanation generally. If this is the case, our last task is to inquire whether token physicalism deserves to be regarded as an unyielding demand on adequate scientific theory. My goal will be a relatively modest one: I shall argue that token physicalism is not nearly so well entrenched a principle. I shall work toward this conclu- sion by urging that the notion of event identity upon which token physicalism depends is not well enough understood to bear the burden of characterizing respectable science and, a fortiori, of mandating conditions for the elimination of an important theoretical idiom for which there is no replacement anywhere in sight. A number of theorists have expressed concern about token identity and, in particular, about token physicalism. I shall be leaning especially heavily on a line of criticism due to Jennifer Hornsby.83 As we have seen, 82It is natural to ask at this point what Fodor himself has to say about the phrase ’causal explanation’. What he says is that "It is simply unreasonable to require that a solution to the naturalization problem should also provide an account of causal explanation," that "semantics is respectable if it can make do with the same metaphysical apparatus that the rest of the empirical sciences require" (Psysem., p. 165n.). This seems to me unfortunate: what is in question is precisely what, in fact, is required of the rest of the empirical sciences. 208 Haugeland points out that events, as well as kinds, depend for their individuation on a theoretical vocabulary. The boundaries of "robust" events of everyday discourse--avalanches, storms, waves, sunsets, and so on-—clearly do not correspond with those of the microphysical events in which basic physics trades. This consideration applies equally to events as individuated by special sciences: if a monetary transaction-—a sale, let us say--is to be identified with anything physical, it must surely be with some aggregate or collection of microphysical events. One problem, clearly, is deciding with yhgg aggregate, precisely? Finding a principled way to bound such collections of microphysical events is, as Hornsby and others have noted, a nontrivial problem.84 The most intuitively appealing suggestion is to identify a special event contingently with the aggregate of microphysical events with which it happens to be temporally and 85 This solution cannot, however, be applied spatially coextensive. straightforwardly and its qualification bids fair to subvert the liberalizing intent of the New View proposal. Let us take a moment to see why. 83Hornsby, Jennifer, "Physicalism, Events and Part-Whole Relations" in A&E. Henceforth, Physicalism. 84Physicalism, p. 453n. For more extended discussion of this issue see Terence Morgan and Michael Tye, "Against the Token Identity Theory", in A&B. 85We can, of course, make such identity logically necessary. It has been argued, however, that doing so undermines one of two major arguments for accepting an ontology of events in the first place. See Jonathan Bennett, "Adverb—Dropping Inferences and the Lemmon Criterion" in A&B. It does not seem likely that Fodor would wish to go this way, in any case. 209 Hornsby argues that we can by no means assume that an aggregate of physical "microevents" picked out by their temporal and spatial coextensiveness with a special event can automatically be identified with a physical "macroevent". A collection of physical microevents may fail to constitute a physical macroevent because part of what we peep by being ’an event’ (of any magnitude) is being the sort of entity to which terms like ’cause’ and ’effect’ can be applied. Part of what we mean by ’a physical event’ is thus something to which causal powers can be attributed within physical theory, and not every collection of spatially and temporally proximate microphysi— cal events need satisfy this demand. Hornsby uses a parallel between continuants and events to make the point. Not every collection of proximate time slices, she notes, constitutes a continuant. [Ilnasmuch as it is in the nature of continuants to persist, we expect individual continuants to be members of kinds whose instance have intelligible, individuation sustaining persistence conditions; inasmuch as it is in the nature of events £9 cause pad 23 caused, we expect individual events to be members of kinds that pull their weight in illuminating accounts of why one thing followed 22 another. The items which are events. . .[thus] need to be singled out not merely as occupiers of space and time, but by reference to [what Ouine calls] a suitable ideology. . . [one] conditagned by the need to construct an explanatory causal nexus. Fodor nowhere explicitly characterizes events as entities that necessarily have a place in a causal nexus, but the quotation from Psychological Explanation that groups ’cause’, ’effect’ and ’event’ as necessarily pre- dicable of the referents of terms used in scientific explanations strongly suggests that this would be consistent with his views as does his more recent allusion to "the principle that events are individuated by their 86A&E, p. 454. Stress added. 210 causes and effects."87 It is certainly with events whose causal powers are owed to their instantiation of physical properties that, on the New View, the events of any respectable science must ultimately be identified. If there is no reason to assume a priori that a conjunction of spatially and temporally proximate physical microevents coextensive with a special event will itself constitute a physical macroevent-—a genuine event with causal powers due to physical law—-then special sciences can fail to meet New View constraints (and thus, fail to be respectable) even if they deliver coun— terfactual supporting generalizations. Nor will it do to say of a special science that its success with regard to prediction and control shows that the aggregates of microphysical events coextensive with its events do, in fact, constitute physical macroevents with the requisite causal powers: to do so is simply to assume the New View, and that is precisely what is in question.88 Here is Hornsby’s commentary on the case of economics. I quote at length because this is a paradigm case for the New View. Consider. . . some arbitrary macroevent e. . . . of interest to economics-—say the Public Sector Borrowing Requirement’s dropping by six million pounds in a certain period. Consider also the (putative) fusion composed from those microphysical events which will be said to ’occupy the very spatio-temporal region that e occupies.’ Is there any reason to suppose that this fusion has any special claim to the status of event? . . . We have no doubt that e itself. . . is a genuine 87Psysem., p. 38. 88To say that gpme aggregates of microphysical events can fail to con- stitute physical macroevents is not, of course, to say that ell such aggregates must so fail. It is simply to suggest that token physicalism is a more complicated and perhaps more stringent demand than might at first appear. 211 event, because 3 is the sort of thing we might be able to learn the cause of: an economist might be able to cite fea- tures of e which bring it in the scope of intelligible, more or less projectible generalizations. But. . . the fact that g has an interesting description in virtue of which we can ap- preciate its causal status, cannot in itself help to secure any description of the putative fusion (a description using microphysical vocabulary plus the word ’part’) which wougg figure in some counterfactual supporting generalization. If Hornsby is correct in arguing that microevents do not necessarily com- pose into macroevents--that we cannot capture all of the events in the world with the vocabulary of physics plus ’part of’ any more than we could capture all the kinds in the world with the vocabulary of physics plus the logical connectives--then the New View theorist who persuades us to endorse token physicalism as a demand on adequacy of causal explanation wins a py- hrric victory: this constraint may well disqualify the very sciences whose success was taken as evidence for the New View and in whose service the New View was initially advanced. Even if Hornsby is not correct—-even if, for example, we can resist the parallel she draws between events and continu— ants-~it still looks as though considerable work needs to be done before we are in a position to evaluate token physicalism as normative doctrine. I began this section by asking how bad it would be for intentional idiom were we not to have a causal semantics for the language of thought. The most conservative answer is, I think, that we cannot as yet tell, that Fodor’s eliminativist premise rests on a view of respectable scientific ex- planation that is, at the moment, seriously unclear in its demands. A less conservative answer, but one to which I am inclined, is that a theorist 89Physicalism, p. 453. Stress added. 212 whose views of scientific explanation are pragmatic, and who takes the judgments of scientific communities as authoritative with regard to theory choice, is already in a position to resist New View doctrine and thus any eliminative conditional that relies upon its acceptance. Insofar as Fodor himself appears to be committed to such naturalism, he need not rely on de- velopment of a causal semantics to legitimate the intentional idiom to which he is attached. Nor, for that matter, does he have to rely on the need to protect such idiom from elimination to justify his attempts to de- velop a causal semantics. The hope of understanding how it is that, unlike the rocks and worms and spiral nebulae, we can harbor states that are con- tentful and semantically evaluable is surely a stake adequate to justify the game. We have no need to believe, in addition, that we are keeping the sky from falling. LIST OF REFERENCES List of References Bach, Kurt, "De re Belief and Methodological Solipsism," in, A. Woodfield, ed., Thought and Object: Essay§ pp Intentionality, Oxford: Clarendon Press, 1982). Bennett, Jonathan, "Adverb-Dropping Inferences and the Lemmon Criterion", in A&E. Bourbaki, Nicholas, Elements of Mathematics: Theory Lf Sets (Reading, MA: Addison- Wesley Publishing Company, 1968). Boyd, Richard, "Metaphor and Theory Change: What is ’Metaphor' a Metaphor For?" in, Andrew Ortony, ed., Metaphor and Thought, (Cambridge: Cambridge University Press, 1979). Brown, Harold I., Perception, Theory and Commitment: the New Philosophy 9; Science (Chicago: University of Chicago Press, 1977). Camfield, Thomas M. , "The Professionalization of American Psychology, 1870-1917", Journal 9; the History Lf the Behavior Sciences, 9 (1973), pp 66- 75. Churchland, Paul M., Matter and Consciousness (Cambridge, MA: A Bradford Book/MIT Press, 1984). Danziger, Karl, "The Social Origins of Modern Psychology", in, Alan R. Buss, ed. Psychology 12 Social Context (New York, NY: Halsted Press, 1979). Martin Davies, "Individuation and the Semantics of Demonstratives", Journal pg Philosophical Logic 11 (1982), pp. 291-92. Devitt, Michael, "Against Direct Reference," Midwest Studies in Philosophy, Volume XIV, Contemporary Perspectives in the Philosophy 9; Language ll. , forthcoming. Devitt, Michael, "Meanings Just Ain’ t in the Head," in, Method, Reason, and Language: Essays in Honour of Hilagy Putnam, ed. George Boolos, Cambridge University Press, forthcoming. Devitt, Michael and Sterelny, Kim, Languag_ and Reality: An Introduction to the Philosophy of Language (Cambridge, MA: The MIT Press/A Bradford Book, 1987. Drake, Stillman, trans., Discoveries and Opinions pg Galileo (Garden City, N.Y.: Doubleday Anchor Books, 1957). 213 214 Dretske, Fred I., "Putting Information to Work", mss. Dretske, Fred I., "Reasons and Causes", mss. Dretske, Fred I., "Why Thinking Helps", mss. (Cambridge, MA:A Bradford Book/The MIT Press, 1988). Dretske, Fred I., "Misrepresentation", in Belief, Form, Content and Function, ed. Radu J. Bogdan (Oxford: Clarendon Press, 1986). Dretske, Fred I., "Precis of Knowledge and the Flow of Information" in H. Kornblith, ed., Naturalizing Epistemology (Cambridge, Mass.: The MIT Press/A Bradford Book, 1985). Dretske, Fred I., "Machines and the Mental", Proceedings and Addresses pf the American Philosophical Association 59 (1985) pp. 23-33. Dretske, Fred I., "Author's Response", The Behavioral and Brain Sciences 6 (1983) pp. 82-90. Dretske, Fred I., Knowledge and the Flow of Information (Cambridge, MA: A Bradford Book/The MIT Press, 1981). Dretske, Fred I., "The Intentionality of Cognitive States", Midwest Studies 13 Philosophy, vol. 5, ed. Peter A. French, Theodore E. Uehling, Howard K. Wettstein (Minneapolis: University of Minnesota Press, 1980. Edward, Paul, ed. The Encyclopedia 9f Philosophy, (New York: Macmillan Publishing Co., Inc. & The Free Press, 1967). Fodor, Jerry A. "A Theory of Content. Part I.: The Problem" mss. Fodor, Jerry A. "Information and Representation", mss. Fodor, Jerry A. "Making Mind Matter More", Philosophical Topics 17 (1989), pp. 59-79. Fodor, Jerry A., Psychosemantics: The Problem of Meaning 12 Ehe Philosophy pf Mind (Cambridge, MA: A Bradford Book/The MIT Press, 1987). Fodor, Jerry A.,"Why Paramecia Don't Have Mental Representations", in Peter A. French, Theodore E. Uehling and Howard K. Wettstein, eds. Midwest Studies 12 Philosophy Vol.X (Minneapolis: University of Minnesota Press, 1986). 215 Fodor, Jerry A. "Information and Association", Notre Dame Journal 9f Formal Logic 27 (1986), pp. 307-323. Fodor, Jerry A. "Semantics Wisconsin Style", Synthese 59 (1984), pp. 231-250. Fodor, Jerry A., The Modularity of Mind: An Essay_ on Faculty Psychology (Cambridge, MA: A Bradford Book/MIT Press, 1983). Fodor, Jerry A. "A Reply to Brian Loar's 'Must Beliefs Be Sentences?’ in PSA 1982, Vol. 2, pp. 644-653. Fodor, Jerry A., "Cognitive Science and the Twin-Earth Problem", Notre Dame Journal 9f Formal Logic 23 (1982), pp. 98-118. Fodor, Jerry A. ,Representations: Philosophical Essays on the Foundations of §_gnitive Science (Cambridge, MA: A Bradford Book/The MIT Press, 1981). (Rep. ) Fodor, Jerry A. "The Mind-Body Problem", Scientific American January 1981, pp. 114-123. Fodor, Jerry A., The Language cf Thought (Cambridge: Harvard University Press, 1975). Fodor, Jerry A. , Psychological Explanation: An Introduction to :33 Philosophy of Psychology (New York: Random House, 1968). Fodor, Jerry A., "Introduction; Something on the State of the Art", in Rep. Fodor, Jerry A.,"Three Cheers for Propositional Attitudes", in Rep. Fodor, Jerry A., "Three Cheers for Propositional Attitudes", in Rep. Fodor, Jerry A. and Pylyshyn, Z.W., "Connectionism and Cognitive Architecture: A Critical Analysis" Cognition (1988),pp. 3-71. Grice, H.P., "Logic and Conversation" in Th3 Logic 9f Grammar, ed. D. Davidson and G. Harman (Encino, CA: Dickenson Publishing Co., Inc., 1975). Grice, P., "Meaning", Philosophical Review, 66 (1957), 377-88. Haack, Susan, Philosophy 9f Logics, (Cambridge: Cambridge University Press, 1978. Hacking, Ian, "The Disunity of Science", Lecture, Nobel Conference XXV: The End of Science?, Gustavus Adolphus College, St. Peter, MN, October, 1989. Hacking, Ian, Representing and Intervening: Introductory Topics in the Philosophy of Natural Science (Cambridge: Cambridge University Press, 1983). 216 Hacking, Ian, The Emergence of Probability: A Philosophical Study of Early Ideas About Probability (Cambridge: Cambridge University— Press, 1975). Hall, Richard J. "Fodor's Theory of Error is Erroneous", mss. Hall, Richard J., "A View About Visual Perception", mss. Hall, Richard J., "Seeing Perfectly Dark Things and the Causal Conditions of Seeing", Theoria, 45 (1979). Haugeland, John, "The Intentionality All—Stars", Philosophical Perspectives, IV: Philosophy pf Mind and Action Theory, James E. Tomberlin, ed., forthcoming. Haugeland, John, "Weak Supervenience", Am. Phil. Quarterly, 19 (1982), pp. 93-103. Haugeland, John, "The Nature and Plausibility of Cognitivism" in Mind Design: Philosophy, Psychologyy Artificial Intelligence (Montgomery, VT: Bradford Books, Publishers, Inc., 1981). Heil, John, Review, Representations, Philosophical Books, 23 (Oct. 1982), pp. 231-233. Hills, David, "Introduction: Mental Representations and Languages of Thought" in N. Block, ed. Readings in Philosophy pf Psychology (Cambridge, MA: Harvard U. Press, 1981), vol. 2, pp. 17-18. Horgan, Terence and Tye, Michael, "Against the Token Identity Theory", in A&E. Hornsby, Jennifer, "Physicalism, Events and Part-Whole Relations", in A&E. Kaplan, D., "Dthat", mss., p. 22. Kim, Jaegwon, "'Strong' and 'Global' Supervenience Revisted", Philosophy and Phenomenological Research, 48 (1987), pp. 315—326. Lachman, Roy, Lachman, Janet L. and Butterfield, Earl C. , Cognitive Psychology and Information Processing_ An Introduction (Hillsdale, NJ: Lawrence Erlbaum Associates, Publishers, 1979). Leibniz, G.W., Monadology (1714) 17. In Leibniz: Philosophical Writings, ed. G.H.R. Parkinson, tr. Mary Morris and G.H.R. Parkinson (London: J.M. Dent & Sons, Everyman's Library, 1973). Leibniz, G. W. New Essays on Human Understandipg, trans. and ed. by Peter Remnant and Jonathan Bennett (Cambridge: Cambridge University Press, 1985). Lewis, David, Op the Plurality pf Worlds, (Oxford: Basil Blackwell, 1986). 217 Lloyd, Dan, "Mental Representation From the Bottom Up", ynthese 70 (1987), pp. 23-78. Loewer, Barry, "From Information to Intentionality", Synthese 70 (1987), 287-317. LePore, Ernest and McLaughlin, Brian P., eds., Actions and Events: Perspectives pp the Philosophy pf Donald Davidson (Oxford: Basil Blackwell, 1985), p. 55). (A&E) McGuigan, F.J., Experimental Psychology: Methods pf Research, 4th edition (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1983). Nisbett, Richard E. and Wilson, Timothy DeCamp, "Telling More Than We Can Know: Verbal Reports on Mental Processes", Psychological Review, 84 (1977), pp. 231-259. Perry, John, "Cognitive Significance and the New Theory of Reference", Nous, 22 (1988), pp. 1-18. Putnam, Hilary, "Philosophy and Our Mental Life", in Mind, Langpage app Reality: Philosophical Papers, Vol. 2 (Cambridge: Cambridge University Press, 1975). Rachlin, Howard, Behavior and Learning (San Francisco: W.H. Freeman and Company, 1976. Rescorla, Robert A., "Pavlovian conditioned inhibition", Psychological Bulletin (1969) Vol. 72, no. 2, pp. 77-94. Rorty, Richard, Philosophy and the Mirror pf Nature (Princeton, NJ: Princeton University Press, 1979). Staddon, J.E.R., Adappive Behavior and Learning (Cambridge: Cambridge University Press, 1983). Stich, Stephen,"Narrow Content Meets Fat Syntax", mss. Stich, Stephen P., From Folk Psychology £2 Cpgnitive Science: The Case Agginst Belief (Cambridge, MA: A Bradford Book/MIT Press, 1986). Stoutland, Frederick,"Davidson on Intentional Behavior", in A&E. A&E C&CA Cog. Sci. CP&IP EB I&R I&SD ICS II Intro. KFI L&R Language Learning LOT M&M Abbreviations Ernest LePore and Brian P. McLaughlin, eds., Actions and Events: Perspectives pp the Philosophy pf Donald Davidson. J.A. Fodor and Z.W. Pylyshyn, "Connectionism and Cognitive Architecture: A Critical Analysis." Jerry Fodor, "Cognitive Science and the Twin—Earth Problem." Roy Lachman, Janet L. Lachman and Earl C. Butterfield, Cognitive Psychology and Information Processing: Ag Introduction. Fred Dretske, Explaining Behavior: Reasons 12 a World pf Causes. Jerry Fodor, "Information and Representation." Martin Davies, "Individuation and the Semantics of Demonstratives." Fred Dretske, "The Intentionality of Cognitive States." Barry Loewer, "From Information to Intentionality." Jerry Fodor, "Introduction: Something on the State of the Art," in Rep. Fred I. Dretske, Knowledge and the Flow pf Information. Michael Devitt and Kim Sterelny, Language and Reality: Ag Introduction £9 the Philosophy pf Language. Jerry Fodor, "Why There Still Has to Be a Language of Thought," Appendix, Psysem. J.E.R. Staddon, Adaptive Behavior and Learning. Jerry A. Fodor, The Language pf Thought. Fred Dretske, "Machines and the Mental." 218 Misrep. MM PAs Physicalism PE PIW Precis Psysem. R&C R&I Rep. SWS Three Cheers WS WTH 219 Fred Dretske, "Misrepresentation." Jerry Fodor, The Modularity gf Mind: Ag Essay gg Faculty Psychology. Jerry Fodor, "Propositional Attitudes," in Rep. (p. 15) Jennifer Hornsby, "Physicalism, Events and Part-Whole Relations." Jerry Fodor, Psychological Explanation: Ag Introduction pg the Philosophy pf Psychology. Fred Dretske, "Putting Information to Work." Fred Dretske, "Precis of Knowledge and the Flow pf Information." Jerry A. Fodor, Psychosemantics: The Problem gf Meaning ig the Philosophy pf Mind. Fred Dretske, "Reasons and Causes." Ian Hacking, Representing and Intervening: Introductory Topics 1g the Philosophy gf Natural Science. Jerry Fodor, Representations: Philosophical Essays pg Egg Foundations gf Cognitive Science. Jerry Fodor, "Semantics Wisconsin Style." Jerry Fodor, "Three Cheers for Propositional Attitudes," in Rep. John Haugeland, "Weak Supervenience." Fred Dretske, "Why Thinking Helps." "IC "MflfllfiflWflflflfllflT/fiflflflES