REPLIES TO PAPERS

Terry Horgan

University of Memphis

Jaegwon Kim argues that one should distinguish naturalism from materialism, and that both should be construed as ontological rather than epistemological. I agree, on both counts. Although I have sometimes tended to slur together materialism and naturalism in of my writings (as is done in much recent philosophy), I do think that it is important to distinguish them. It is a serious philosophical task to get clearer about how each position is best articulated, and about ways that one could embrace naturalism without embracing materialism. British emergentism, for example, seems reasonably classified as a position that is naturalist but not materialist (and evidently the British emergentists themselves construed their view this way). Here are two key tenets of British emergentism, both of which seem to disqualify the view from being a form of materialism without thereby disqualifying it as a form of naturalism:

(E.1) There are emergent properties in nature, in the following sense: although (i) these properties are supervenient on certain other properties, the relevant supervenience facts are ontologically sui generis (and hence are unexplainable).

(E.2) Emergent properties are fundamental force-generating properties, in this sense: they produce additional fundamental forces that affect the distribution of matter, above and beyond the fundamental forces posited in physics.

A position worth of the label “materialism,” it seems to me, should preclude both of these emergentist theses. My notion of superdupervenience is intended as a condition that any version of materialism should satisfy, and is supposed to be incompatible with theses (E.1) and (E.2). Although sometimes, as in Horgan and Timmons (1992) and Horgan (1994), the condition is articulated in terms of the need for supervenience to be explainable “in a naturalistically acceptable way” (thereby slurring the naturalism/materialism distinction), what I had in mind was that supervenience relations must be explainable in a materialistically acceptable way. This was how I formulated the requirement in Horgan (1993), the paper in which the term ‘superdupervenience’ was explicitly introduced.[1]

I have always thought of this explainability requirement as fundamentally ontological, rather than as epistemological. The idea is that the relevant supervenience relations are not brute modal facts—not brute inter-level laws of nature (as the emergentists thought, in the case of the properties they considered emergent), and not brute synthetic necessary truths (as Moore evidently construed the status of natural/moral supervenience relations)—but instead are facts that obtain because of some “materialistically kosher” sufficient reason. I take this to mean that such supervenience facts are in principle explainable in a materialistically acceptable way. The kind of in-principle explainability I have in mind could turn out to be forever beyond the reach of human cognition: the explanations might exist in principle, and yet humans might be “cognitively closed” to them. (Colin McGinn, as I understand him, holds this view about the relation between consciousness and its physical basis; cf. McGinn 1991, 1993.)

Kim wonders whether the purpose of the superdupervenience constraint would be better served by replacing the requirement that supervenience relations be materialistically explainable—my own proposed condition that a metaphysical position should satisfy in order to be a version of materialism—with the alternative requirement (as per the principles Kim formulates as (T*) and (T**)) that supervenience relations be “naturalistically accessible.” I would resist such a replacement, because conditions (T*) and (T**) appear to be consistent with theses (E.1) and (E.2) of emergentism, whereas the desired constraint on materialism is supposed to preclude both theses.

Kim asks, “Why should an explanation of the entailment (N ® Pain) make Pain naturalistically “kosher” (as Horgan says), whereas the entailment itself cannot?” First, I again emphasize that the key issue, insofar as my invocation of superdupervenience is concerned, is what it takes to make the N-to-Pain relation materialistically kosher. Second, Kim is here using the word ‘entailment’ in a coarse-grained way that applies to a range of necessitation relations. Third, some kinds of putative necessitation relations would not be materialistically kosher, notably (i) ontologically brute inter-level nomically necessary connections (as posited by emergentists, for instance, for the case of neural properties vis-à-vis mental properties), and (ii) ontologically brute, synthetic, metaphysically necessary connections (as posited by Moore, for instance, for the case of natural properties vis-a-vis moral properties). Fourth, the appropriate way to preclude such metaphysically sui generis inter-level necessitation relations, it appears to me, is to require that supervenience relations be explainable in a materialistically acceptable way (rather than being metaphysically sui generis); hence my contention that a materialistic metaphysical position requires superdupervenience rather than mere supervenience. One way for there to be such explainability—the only clear way I have been able to identify—is this: statements about relevant physical facts (including physical laws, perhaps), in consort with certain conceptual truths about higher-level properties, logically guarantee the truth of statements expressing supervenience connections. As is pointed out in Horgan and Timmons (1992), supervenience relations between physical properties and higher-level functional properties appear to be susceptible to this explanatory format.

Is the contention that mental properties and other special-science properties are superdupervenient on physical properties a “reductionist” position, and is it thereby at odds with my own frequent espousal of “nonreductive materialism”? Kim’s discussion suggests that that he thinks the answer to both questions is ‘Yes’. My own answer is that it all depends on how one employs the word ‘reduction’, a term of art whose metaphorical connotations allow a range of different but legitimate alternative uses. In one important sense, reductionism asserts that each natural-kind property posited in the special sciences is at least nomically coextensive with some natural-kind property posited in physics, and often also asserts that the ontological basis for such an inter-theoretic “bridge law” is that the special-science property is identical to its physical “correlate.” Reductionism in this sense is what Fodor attacked in his influential paper “Special Sciences,” and also is what various philosophers (myself included) mean to repudiate in advocating what we call “nonreductive materialism.”

In another important sense, however, reductionism is a view denying both key tenets (E.1) and (E.2) of emergentism. This kind of reductionism asserts that although there may well be special-science natural-kind properties that are distinct from (and are not nomically coextensive with) physical natural-kind properties, nevertheless such properties are always resultant rather than emergent—i.e., their supervenience on certain physical properties is explainable rather than metaphysically sui generis, and consequently these properties also are not fundamental force-generating properties. In this anti-emergentist sense of ‘reductionism’, a materialist metaphysics that embraces the superdupervenience of mental properties and other special-science natural-kind properties certainly counts as a reductionist position. Thus, the view I call “nonreductive materialism” is certainly reductive in this sense.

Jessica Wilson puts heavy metaphysical weight on the idea of a fundamental force, as important for characterizing materialism. I am very sympathetic with this idea. I came to appreciate its importance by reading McLaughlin (1992), and thereafter I began to invoke fundamental forces myself. In Horgan (1993), I characterized the causal completeness of the physical this way: all fundamental causal forces are physical forces, and the laws of physics are never violated (p. 560). And it seems to me that one of emergentism’s two key tenets is also best formulated in terms of forces, as the claim that emergent properties are fundamental force-generating properties. (See (E.1) above, and the lesson from emergentism labeled (L.1) in Horgan 1993, p. 560)

As I said in Horgan (1993) p. 560, I take it that if inter-level supervenience relations are indeed robustly explainable in a materialistically acceptable way, then the supervenient properties thus explainable are not causally basic, in the sense of being fundamental force-generating properties. Given this very plausible assumption, together with the empirically well-justified assumption that the set F of fundamental forces in nature includes only physical forces that come into play at or below the level of atomic organization, the superdupervenience requirement—“Horgan’s constraint,” as Wilson calls it—evidently entails the principle that Wilson calls “Physicalism”:

Physicalism (same-subject necessitation): For every property P and Q: If Q same-subject necessitates P, every causal power bestowed by P is identical with a causal power bestowed by Q that is grounded only in the fundamental forces in F.

But contrary to Wilson, it seems to me that satisfying this principle is not enough to qualify a metaphysical position as a version of materialism; rather, Horgan’s constraint itself must be satisfied too. To see why, return to tenets (E.1) and (E.2) of emergentism, and consider a metaphysical position that embraces (E.1), rejects (E.2), and embraces Wilson’s principle. Such a view construes supervenient properties as emergent rather than resultant, even though it denies that emergent properties are fundamental force-generating properties. I.e., the view posits ontologically brute, metaphysically basic, inter-level necessitation relations that must be accepted, along with the basic laws of physics, “with natural piety.” Although such a position can qualify as a version of naturalism, I think it does not qualify as a version of materialism. There is a heavy burden of proof on anyone who would claim that a full-fledged materialist should be willing to accept inter-level supervenience-laws as fundamental “unexplained explainers” alongside the basic laws of physics. The prima facie presumption is against this claim.

Olga Markic maintains that nonreductive materialism is an unstable position, and that stability can only be achieved via its transformation into one or another position such as eliminativism, preservative irrealism, reductive materialism, epiphenomenalism, or emergentism. She contends that the instability of nonreductive materialism comes to light given my own claim that any version of materialism that is metaphysically realist about mentality requires physical-to-mental superdupervenience. Her argument, as I understand it, goes as follows. Either the superdupervenience requirement is statisfied for mentality, or not. If it is satisfied then reductive materialism obtains; mental properties are resultant, rather than emergent. But if it is not satisfied, then the only available options are irrealist versions of materialism (e.g., eliminativism or preservative irrealism) or non-materialist views (e.g., epiphenomenalism or emergentism).

In reply, I again emphasize the two above-mentioned ways of employing the word ‘reduction’. The position that I call nonreductive materialism claims that mental properties are distinct from, and are multiply realizable by, physical natural-kind properties; it also embraces physical-to-mental superdupervenience. This view is nonreductive in the sense that it embraces mental properties as real and yet denies that mental properties are identical to (or are nomically coextensive with) physical natural-kind properties. In another sense it is reductive: it construes mental properties as resultant, rather than emergent. I see no instability in this position; but one needs to keep clearly in mind that the term ‘nonreductive’ has the first sense when the label ‘nonreductive materialism’ is employed.

Markic discusses my causal compatibilism concerning mental causation vis-à-vis physical causation, and she speculates that it may be connected to the “preservative irrealism” about intentional mental properties that I recommend as a serious philosophical option in Horgan (1993, 1994). But these two themes in my work are not directly connected. Causal compatibilism is my favored approach to the causal exclusion problem; this problem arises for any materialist position that is realist about mental properties but denies that mental properties are identical to physical properties. Preservative irrealism about mental intentionality, on the other hand, is a position that I have argued is worth taking seriously in light of the real possibility that putative mental intentional properties might turn out not to be superdupervenient on the physical.

If one does embrace preservative irrealism about mental intentionality, does this mean that one must eschew the causal efficacy of the intentional aspects of mentality? Not necessarily, because the relevant causal claims might still turn out to be true. Here is a place where my work on contextual semantics, and on truth as indirect correspondence, can come into contact with my work in philosophy of mind. These interconnections are explored in Horgan (1994).

In any case, I have never been willing to fully embrace preservative irrealism about mental intentionality, even though I do think the view deserves serious consideration. Lately I have substantially backed away from this position, partly because (i) I am a realist about the phenomenal aspects of mentality, and (ii) I now hold that mental intentional properties are, by their very nature, phenomenal; cf. Horgan and Tienson (in press).

John Tienson argues against the claim that the problem of causal exclusion in philosophy of mind is a special case of a general problem about causal exclusion for special-science properties—a claim made by various philosophers, myself included. Tienson maintains, roughly, (1) that causal relations typically “go through” certain macro-properties that are resultant supervenient properties (rather than excluding such properties), and (2) that the natural-kind properties posited in the special sciences typically are such resultant, causally potent, macro-properties.[2] Perhaps he is right about this; I currently am unsure what to think about it. My lingering worry is this: Isn’t it prima facie plausible to argue, in the spirit of causal-exclusionary reasoning, that all the “real causal work” in nature is done (often collectively) by physical microparticles instantiating fundamental force-generating properties, and that resultant properties instantiated by complex wholes thereby are epiphenomenal?

But suppose that this worry can be satisfactorily handled, and that there is no genuine exclusion problem for resultant macro-properties. As Tienson himself points out, there would still be an exclusion problem in philosophy of mind insofar as mental properties fail to be resultant properties (i.e., fail to be explainable in lower-level terms). I have reluctantly come to believe that mental properties in general may well be emergent rather than resultant, in the sense of thesis (E.1) above. This is because, like a number of philosophers (including Kim and Tienson), I see no credible prospects for explaining the supervenience of the phenomenal aspects of mentality in a materialistically acceptable way; moreover, Tienson and I have argued (Horgan and Tienson, in press) that the phenomenal and intentional aspects of mentality are so thoroughly intertwined that virtually all consciously accessible mental states (including occurrent beliefs and desires) have constitutive phenomenal character—which would mean that the infamous problem of the “explanatory gap” arises for consciously accessible mental states in general.

This would also mean that there is indeed a problem of causal exclusion for mental properties in general, even if Tienson is right that there is no exclusion problem for resultant properties. Three versions of emergentism about mental properties can now be distinguished, each of which deals differently with the exclusion issue:

Classical emergentism: Mental properties are emergent, and they are fundamental force-generating properties.

Epiphenomenal emergentism: Mental properties are emergent, and they are causally inefficacious.

Compatibilist emergentism: Mental properties are emergent and are causally efficacious, but they are not fundamental force-generating properties.

Classical emergentism seems very unlikely to be true, given the steadily mounting empirical evidence that the only fundamental force-generating properties in nature are physical properties instantiated at or below the atomic level. So, if mental properties are indeed emergent rather than resultant, then evidently they can only be causally efficacious if some version of causal compatibilism is true. I think my own recommended version, in Horgan (2001) and the other papers cited therein, can be smoothly grafted onto emergentism to yield a credible form of compatibilist emergentism.

Seven Walter discusses my two papers (one collaborative with George Graham) on the theme of Jackson’s knowledge argument. As Walter points out, Horgan (1984) was a reply to Jackson on behalf of the materialist. That paper sketched an indexical version of the “mode of presentation” response to Jackson’s argument—an approach that Walter himself is inclined to favor. Walter also discusses Graham and Horgan (2000), in which Graham and I propose a variant of Jackson’s original argument that we claim poses a very serious challenge to materialism. Walter’s main claim about our Mary-Mary argument is this: to the extent that it exceeds the force of Jackson’s original challenge, it coincides with the problem of the “explanatory gap.”

I have always worried that Jackson’s thought experiment raised serious potential worries for materialism, even though his own presentation of the knowledge argument seems to me not to get to the heart of the matter. My dissatisfaction with his own formulation prompted my reply on behalf of materialism in Horgan (1984). In effect, the idea was to treat phenomenal concepts as indexical concepts of a certain specific kind that supposedly pick out materialistically respectable properties—either physical properties or else properties that are superdupervenient on physical properties. But I have increasingly come to think that neither this materialist treatment of phenomenal concepts and phenomenal properties, nor any others that I know of (or can envision), properly respects the phenomenology of subjective experience. The Mary-Mary argument attempts to show this. The key idea, as Walter points out, is that Mary-Mary’s extreme surprise upon learning what it is like to experience colors is rationally appropriate in a way that it would not be if the relevant materialist account of phenomenal concepts and phenomenal properties were correct. The default moral seems to be that phenomenal properties are not superdupervenient on physical properties—i.e., that they are emergent properties. Much as I would like to be a materialist, I have as yet found no satisfactory way of evading this conclusion.

What is the connection between the Mary-Mary argument and the problem of the explanatory gap? I think that the argument provides one specific way of appreciating this problem—one specific route to the realization that the supervenience of phenomenal properties on physical properties does not seem to be susceptible to materialistically acceptable explanation. For those who already believe this, perhaps the argument does nothing terribly novel philosophically. But against those who think that the problem of the explanatory gap can be evaded, e.g., by some version of the “mode of presentation response,” I think the Mary-Mary argument is a dialectically effective way of bringing to light the apparent inadequacy of their approach. Such accounts evidently fail to come adequately to terms with the intrinsic phenomenal character of experience—what “what it’s like” is really like.

Alex Byrne argues that a variant of Jackson’s case of black-and-white Mary teaches us that the propositional content of perception cannot be fully expressed in language. Although Byrne’s paper does not directly engage my own views, let me comment on it briefly. I am sympathetic with the contention that when Mary sees the bead, she acquires an ineffable belief about its color. However, I am not completely persuaded that this belief cannot be expressed in language; for I find myself somewhat attracted to the contention that the belief’s ineffability consists rather in the fact that Mary, prior to seeing the blue bead, does not fully possess the concept BLUE—and hence neither fully understands the word ‘blue’ nor has a full-fledged belief that the bead is blue (a belief that requires possessing, and deploying, the concept BLUE expressed by the word “blue”).[3]

But whatever exactly the ineffability of Mary’s new belief consists in, it seems to me appropriate to describe her new epistemic situation in Russellian terminology: her newly acquired, ineffable, belief is a belief by acquaintance about the bead’s color—that is, the belief deploys a color-concept that is grounded in an experiential-acquaintance relation between Mary and the property blue. Call this a direct-acquaintance concept. It also seems plausible that when she conceives the property blue under this direct-acquaintance concept, she is conceiving it essentially, as it is in itself. And of course, when one conceives the property blue in this way, one conceives it otherwise than as physical. Furthermore, the following principle seems hard to deny: If conceiving blue under this concept is a matter of both (i) conceiving it essentially, as it is in itself, and (ii) conceiving it otherwise than as physical, then the property so conceived is otherwise than physical—i.e., it is a nonphysical property. Hence, blue is a nonphysical property; and by a parallel argument, blue is not a superdupervenient property either. (Compare Horgan and Tienson 2001). The property blue may or may not really get instantiated by material objects; but whether it does or not, the argument just rehearsed leads to the conclusion that this property—whose essence is presented to us directly in experience—is not a physical property or a superdupervenient property. I find this argument quite plausible, although its conclusion is very jarring.

I turn now to papers discussing the approach to language/world and thought/world relations that I call “contextual semantics.” As I often do in articulating and defending this approach, I will adopt Putnam’s capitalization convention to refer to denizens of the mind-independent, discourse-independent, world. Also, I will follow my recently adopted practice of using the phrase ‘correct affirmability’ rather than ‘correct assertibility’. The reason for this change in terminology is that thoughts have truth-evaluable content no less than statements do; affirming can happen either in thought or in language, whereas asserting is a speech act.

Maria Reicher maintains that contextual semantics, with its appeal to the notion of truth as indirect correspondence, ultimately amounts to a paraphrase strategy, contrary to my own intentions—and hence that contextual semantics does not really succeed in the goal of showing how ontological commitment to entities like symphonies and corporations can be avoided even if statements positing these putative entities cannot be systematically paraphrased into statements not positing them. She sets forth a four-premise argument whose conclusion is this:

If s is a meaningful assertive sentence, then it must be possible in principle to give a formulation of the assertibility norms of s in terms of direct correspondence.

Three of the four premises, she says, should not be a matter of debate. With respect to the statement “Beethoven’s fifth symphony has four movements” (statement (2) in her paper), she formulates as follows two of the premises that she considers nontendentious:

If we cannot reach...an ultimate paraphrase, then we never know what has to be the case in order for (2) to be true; and if it were [in principle] impossible to reach such an ultimate paraphrase, then (2) would not be a meaningful sentence. (p. 9)

I would deny both of these claims. When I advert to “semantic standards” (or “semantic norms”) that govern various kinds of discourse (e.g., symphony discourse) and only require indirect correspondence rather than direct correspondence, the semantic standards I have in mind do not take the form of tractably formulable, cognitively surveyable, truth conditions that are expressed in language that only quantifies over ENTITIES that are REAL. I deny that in general, there are such truth conditions—or need to be. (This denial is part of my rejection of the paraphrase strategy as the only legitimate way of avoiding ontological commitment to OBJECTS and PROPERTIES answering to the relevant singular and general terms.) So I also deny that we must know such truth conditions in order to understand the meanings of our terms and concepts, or in order to employ them competently. What we need, rather, for understanding of meaning and for semantic competence, is to be able to apply the terms and concepts in a way that reasonably reflects semantic correctness reasonably well, to the extent that available evidence allows us to do so. (We need to be reasonably good at tracking epistemically warranted affirmability in the first instance, and thereby at tracking semantically correct affirmability insofar as our evidence does not mislead us or come up short.) We are reasonably good at this, even though typically the contextually operative semantic standards are such that the set of “possible worlds” in which a given statement is true (under those standards) cannot be specified by any compact, cognitively surveyable, formula that itself is governed by direct-correspondence semantic standards. This kind of tracking competence is largely constitutive of both (i) understanding the meaning of statements like (2), and (ii) knowing what has to be the case in order for such statements to be true.

I might add that I am strongly inclined to believe that the lack of tractable, cognitively surveyable, formulability is something that holds not just with respect to putative direct-correspondence truth-conditions for specific statements, but also for semantic normativity in general. I doubt that the kinds of contextually operative semantic standards that typically govern our thought and discourse are tractably specifiable and cognitively surveyable in any format, let alone in the format of direct-correspondence truth conditions. And I think that the same is probably also true for other kinds of normative standards that philosophers tend to focus on—notably, moral standards and epistemic standards. Again, such formulability is not needed. Rather, what is needed is that human judgments and judgment-tendencies should accurately reflect those standards, modulo available evidence and information. And again, I do not believe that this semantic competence requires, by any means, that the standards themselves should be formulable in a tractable and cognitively surveyable way.

There is a connection between these claims about normativity and the general conception of cognition set forth and defended in Horgan and Tienson (1996). Tienson and I argue that human cognition is too complex and too subtle to conform to programmable, algorithmic, rules over mental representations—and thus that the computational paradigm in cognitive science is inadequate. Briefly, the connection I see is this: just as human cognitive state-transitions cannot be systematized by exceptionless algorithmic rules, likewise various kinds of normativity (semantic, epistemic, moral) cannot be systematized by exceptionless general rules. Non-algorithmic cognitive processing can conform to particularistic normative standards.

Tadeusz Szubka argues, contrary to what I have maintained, that “the only gap between truth conceived as properly idealized warranted assertibility and truth construed as correct assertibility is terminological, not substantial” (p. 15). His case for this claim depends partly on the contention that although many who advocate some kind of epistemic account of truth have repudiated THE WORLD with its OBJECTS and PROPERTIES and RELATIONS, one can consistently embrace an epistemic construal of truth without embracing this kind of global metaphysical irrealism. This may well be right. And, insofar as epistemic reductionists are willing to concede the point, my dispute with them is certainly narrower than is my dispute with those “package deal metaphysical irrealists” (as I call them) who think of epistemic reductionism as intimately intertwined with global metaphysical irrealism. I am a metaphysical realist, and I want a construal of truth that comports with metaphysical realism.

Nonetheless, I do think there are gaps between semantically correct affirmability and idealized warranted affirmability. First, there is a conceptual gap, which can be expressed via what Crispin Wright (1992) calls the “Euthyphro contrast”: even to the extent that truth (i.e., semantically correct affirmability) coincides with ideal warranted affirmability, this is not because the latter constitutes the former, but rather because an epistemically ideal situation would eliminate the usual ways of being misled, mistaken, or misinformed about what is true. Truth is the key constitutive goal behind normative standards for epistemic warrant, and as such is conceptually prior to warrant or to any idealization of warrant.

Second, there is an instantiation gap, for various actual and possible cases. For example, two competing theoretical accounts about the realm of the ultra-small might be equally well warranted by all relevant epistemic standards, and might remain so even in the ideal limit of human inquiry—and yet (as all the parties to the discussion might themselves agree), there will still be an objective fact of the matter about which account (if either) is actually true. Another example is from Putnam, who now thinks it a mistake to construe truth purely epistemically: the claim “There is a group of stars outside our light cone (i.e., in some region such that a signal from any such region would have to travel faster than the speed of light to reach us) arranged as the vertices of a regular 100-gon.” Although such a claim might be true, there is no way we could know it to be true—even in the ideal limit of human inquiry; cf. Putnam (1995).

I believe that there is something importantly right in pragmatist and verificationist approaches to truth, even though I think these approaches err in attempting to reduce semantic correctness to some form of epistemic correctness. Contextual semantics seeks to retain what is right, while eschewing epistemic reductionism. One key idea is that epistemic warrant is keyed to semantic correctness in such a way that one can tell a lot about the latter by attending to the former. Warranted affirmability of the statement “Beethoven’s fifth symphony has four movements,” for instance, has very little to do with the question whether there are such mind-independently real ENTITIES as SYMPHONIES. This fact about epistemic warrant strongly suggests that the semantic correctness of this statement, as it is employed in typical contexts, is a kind of indirect correspondence that does not involve such ENTITIES; for, otherwise the epistemic standards that we actually employ vis-à-vis such statements would be far too lax to be epistemically appropriate, given the statements’ semantic content.

Another key theme of contextual semantics, also reflecting something importantly right in pragmatism and verificationism, is that standards for epistemic warrant and for semantic correctness are intimately interconnected with one another, and very likely are learned simultaneously. To a large extent, mastery of meaning is a matter of having mastered suitable epistemic standards of confirmation and disconfirmation for various statements. But again, this only reflects the fact that in general, the contextually operative warrant standards for statements in a given mode of discourse are appropriate, relative to the contextually operative semantic standards for such statements. The interconnectedness does not, by any means, show that truth is reducible to some version of epistemic warrant.

Those who advocate epistemic reductionism about truth bear the burden of proof, when their view is compared to mine. Although epistemic theories have been around a long time whereas contextual semantics has not, this sociological fact does not, by itself, shift the burden of proof back to me.

Michael Lynch sympathizes with my general approach to truth, and like me is dubious about epistemic reductionism. He also claims to believe, as I do, in a mind-independent, discourse-independent, world. But he has doubts both about my version of metaphysical realism that posits OBJECTS with PROPERTIES and RELATIONS, and about my contention that there is a limit-case use of terms and concepts under which truth is direct correspondence (rather than indirect correspondence). The centerpiece of his paper is an argument that if contextual semantics is correct, then no statement is ever semantically correct under maximally strict, direct-correspondence, semantic standards. I will address that argument below, after first commenting on several other matters.

Because of my emphasis on implicit, contextually variable, parameters that govern semantic correctness, Lynch classifies me as a “pluralist” about truth. He mentions this potential problem for pluralist accounts:

Barring an account of how various ascriptions of semantic correctness are unified, pluralist theories of truth would seem to imply that truth is equivocal. This would be unfortunate. (p. 7)

In reply, let me stress the following feature of my view about concepts governed by contextually variable semantic parameters: when parameter-settings alter from one context to another, the result is fine-grained, inter-contextual differences in concept (and in meaning) that are identity-preserving. Adapting Derrida’s terminology, I use the word “diffèrance” for this kind of identity-preserving semantic variation. Diffèrance is a much more subtle semantic difference than ordinary equivocation, because the latter involves non-identical concepts and meanings. Trans-contextual diffèrance is a feature of the notion of truth, as well. For instance, in many metalinguistic contexts (although not necessarily all), the contextually operative semantic standards governing the truth concept work in tandem with those governing terms and concepts expressed in the object language, in such a way that instances of schema T are themselves semantically correct. Since these fine-grained, trans-contextual, semantic differences in the contextually proper use of the truth-predicate are identity-preserving differences in the meaning of ‘true’, they do not make truth equivocal.

Concerning my contention that there can be limit-case, direct-correspondence, uses of terms and concepts, Lynch raises this worry:

It is difficult to avoid the conclusion that the TRUTHS hold a privileged place in Horgan’s scheme. Generally speaking, it is a norm of any rational discourse that one should have beliefs about reality and not merely about appearance. That is, were one forced to choose, one should (all other things being equal) choose to believe in what is real rather than what merely appears to be real. (p. 10)

But although the appearance/reality distinction perhaps could be pressed into service to mark the distinction between what is true under limit-case semantic standards and what is true under non-limit-case standards but is not true under limit-case standards, I would urge that in most contexts, the semantic parameters governing the terms ‘appearance’ and ‘reality’ do not work this way at all. Statements that are true under contextually operative semantic standards fall on the reality side of the divide, not the appearance side. Limit-case truths are not privileged in terms of what is worthy of belief, although they are privileged ontologically.

Concerning the putative ontological benefits of contextual semantics, Lynch raises the following objection, phrased in terms of what Jackson (1998) calls “location problems” (i.e., problems of finding a place in the mind-independent physical world for items to which our beliefs appear to be ontologically committed):

While CS allows us to avoid location problems, it gives rise to a very similar set of problems concerning whether any of our ordinary statements express FACTS about the WORLD. Call these re-location problems.... We must consider, among other questions:

Are symphonies OBJECTS (i.e., mind-independent entities)?

Are there any nonmaterial OBJECTS?

Are symphonies nonmaterial?

And so on.... This apparently undermines the main motivation for adopting CS. If CS was meant to allow a realist to avoid the frustrations of the location project, then the fact that CS causes extremely similar problems indicates that CS has gained us less ground than we thought.” (pp. 10-11).

I certainly acknowledge that we face such “relocation problems” within my general approach to semantics and ontology. Indeed, addressing such questions is a central task for metaphysics, as I see it. But in my view, this hardly means that CS has gained us less ground than we thought—much less that the main motivation for adopting CS has been undermined. On the contrary: within my approach, there is a crucial third option concerning the ontological status of entities posited in our discourse, beyond the options of either (i) locating them as ENTITIES in THE WORLD, or (ii) repudiating them and rejecting statements that posit them as false—viz., (iii) claiming that statements positing them are true under indirect-correspondence semantic standards but false under direct-correspondence standards. One key theoretical motivation for CS is the introduction of this third option into metaphysics.

Lynch objects that a central claim of CS cannot be stated under limit-case semantic standards, viz.,

(1*) Statement (1) is true,

where statement (1) is “Mozart composed 27 piano concertos.” He is right that (1*) is not true under limit-case semantic standards; rather, it is true under metalinguistic semantic standards that are assertorically consistent with those that typically operate at the object-language level to govern statements like (1*). But from my point of view this is not a problem, for two reasons. First, I do not require or expect that serious theorizing, in science or in philosophy, need always take place within discourse that is governed by maximally strict semantic standards. This is no less true for theorizing about how our language and concepts work than for any other theorizing. (I do acknowledge that the ontological foundations of contextual semantics need to be secure, in order that the whole approach not implode on itself. See Horgan and Potrc (in press), which addresses questions posed by Tienson (in press) about the ontological foundations of contextual semantics.) Second, one contextually permissible way to employ the truth predicate, which is especially useful in articulating contextual semantics, is to use it relativistically rather than categorically, thereby disengaging one’s metalinguistic discourse from specific object-language semantic standards. For example, the following claim is correct under this relativized usage of ‘true’: Statement (1) is true under the semantic standards that normally govern concerto-talk.

Let me turn now to Lynch’s argument that if CS is correct, then no statement is ever semantically correct under maximally strict semantic standards. To begin with, I myself have not much used the capitalized words ‘TRUE’ and ‘FACT’. I have mainly used the capitalization convention with words like ‘object’, ‘property’, and ‘world’. When one ascends to the metalinguistic level, one needs to keep two distinct issues separated conceptually, and it is useful to have notational devices in play that enforce this separation (and thereby avoid confusion). The first issue is whether a given object-level statement is true (semantically correct) when it is employed under maximally strict semantic standards. One limit-case use of ‘true’ would be to claim that this is so. (This usage need not be one in which the metalanguage itself is governed by limit-case semantic standards.) A second issue is whether a putatively mind-independently real PROPERTY expressed by the predicate ‘is true’ is instantiated by a given object-level statement (which would require that the instantiating statement itself be an OBJECT). Another limit-case use of ‘true’ would be to predicate the putative property TRUTH. (This usage need not necessarily be limited to cases where the semantic standards at work in the object language are limit-case object-language semantic standards.) The capitalization convention, as applied to ‘is true’ and to ‘truth’, signals the second kind of limit-case usage. But it will be helpful to have a separate explicit notation to signal the first kind of usage too. So, let ‘true_lim’ indicate truth under limit-case object-language semantic standards—a notation that leaves open the question whether truth in general, or truth_lim in particular, is a PROPERTY.

Using this notation, Lynch’s reductio ad absurdum argument evidently goes as follows. Take a statement “S” that is supposedly true_lim, and consider this statement:

(a) “S” is true_lim

Since we ourselves determine a statement’s operative semantic standards in context, whether statement (a) is true_lim is a mind-dependent matter. Hence,

(b) ““S” is true_lim” is not true_lim

Furthermore, it is a priori that

(c) If “S” says that S, then “S” is true_lim iff S.

From (b) and (c) plus the fact that “S” says that S, it follows that

(d) “S” is not true_lim.

I have several things to say about this argument. First, the fact that it is up to us whether or not the statement “S” is governed, in context, by limit-case semantic standards does not sanction the inference from (a) to (b). For, it is not up to us whether or not “S” is semantically correct under these limit-case semantic standards; rather, this depends upon whether or not “S” directly expresses how things are with the INSTANTIATION of PROPERTIES and RELATIONS by OBJECTS.

Second, I deny that instances of (c) are invariably true (and hence I also deny that (c) is an a priori truth). Suppose, for instance, that one employs the statement “Mozart composed 27 piano concertos” under semantic standards that typically govern concerto-talk, while also employing the truth-predicate to express semantic correctness under limit-case object-language semantic standards. (The subscript ‘lim’ on ‘true’ makes this latter usage explicit.) In this “mixed” mode of discourse, the truth predicate is being employed in a manner that is not assertorically consistent with the semantic standards operative in the object language, and hence the relevant instance of principle (c) does not obtain. Although it is the case that

Mozart composed 27 piano concertos,

it is not the case that

“Mozart composed 27 piano concertos” is true_lim.

Third, although I reject Lynch’s own reasoning in support of (b), as explained above, I do allow that (b) may be correct even so. For, perhaps there are no such entities as STATEMENTS, and/or perhaps there is no such property as TRUTH in general or TRUTH_lim in particular. If not, then the metalinguistic statement (a) is not true under limit-case metalinguistic semantic standards, even though the statement “S” is indeed true under limit-case object-language semantic standards.

Fourth, although I deny that instances of (c) are always correct (and hence I deny that (c) is a priori), I do concede that (c) is correct when the contextually operative semantic standards for object-language discourse are limit-case semantic standards.

All this brings me to my fifth and main point about the argument. Even if claim (b) is correct (as it is if there are no such OBJECTS as STATEMENTS, and/or there is no such PROPERTY as TRUTH_lim), and even if claim (c) is correct (as it is if the statement “S” is being employed under limit-case object-language semantic standards), Lynch’s overall argument fails anyway. For, the inference from (b), (c), and the fact that “S” says that S to the conclusion (d) is invalid. The problem is that this inference involves replacing statement (a) by the statement “S” within statement (b). This “level-dropping” substitution within the scope of the sentential context “___is not true_lim” is invalid, because the original statement within that sentential context can be richer in the range of objects and properties it posits than is the statement that replaces it, and these additional posits of the original statement may not be OBJECTS or PROPERTIES although the statement replacing it is itself true_lim. Specifically, statement (a) posits a statement (viz., the statement “S”) and a property (viz., the property truth_lim), and there may not be any such OBJECT or any such PROPERTY even if statement “S” is true_lim.

Matjaz Potrc points out an apparent tension between certain claims I want to make concerning contextual semantics on one hand, and ontology on the other hand. He argues that these claims, rightly understood, are not really in tension at all. I concur.

As Potrc points out (and as I remarked above in my comments on Maria Reicher’s paper), I have become increasingly attracted to “particularism” about semantic normativity. That is, I believe that the contextually semantic standards that typically govern most uses of concepts and terms are probably too complex and subtle to be fully systematizable by exceptionless general principles; also, I have doubts about the extent to which they can even be partially systematized by “soft” general principles containing ceteris paribus clauses. On the other hand, I have also articulated and endorsed the “nonarbitrariness of composition” as a fundamental methodological principle that demands honoring in serious ontological theorizing. I find it at work in Van Inwagen (1990); and it is invoked, for example, in Horgan and Potrc (2000) as part of the argument for the position there called “ontological blobjectivism.”

I agree with Potrc that there is not a genuine conflict between this methodological principle and my particularism about semantic normativity. The nonarbitrariness of composition holds sway when we are doing serious ontology, asking what OBJECTS and PROPERTIES there are. At the semantical level, it governs the proper use of object-talk as governed by direct-correspondence semantic standards. But this principle need not, and generally does not, hold sway insofar as one is employing language and thought under various kinds of indirect-correspondence semantic standards—where one’s posits are only aspects of “regional ontology” rather than “ultimate ontology.”

Turning now to metaethics, Josep Corbi raises several worries about the metaethical position that Mark Timmons and I have articulated and defended, which we call “nondescriptivist cognitivism.” He also challenges one of our main anti-descriptivist arguments, the one we direct against Michael Smith’s metaethical rationalism. His remarks prompt some points of clarification, both about nondescriptivist cognitivism and about our critique of Smith.

Timmons and I characterize descriptive content as “way-the-world-might-be” content. We maintain that “base case” beliefs—roughly, those non-evaluative and evaluative beliefs whose contents have the simplest kinds of logical form—are of two types: a non-evaluative belief is an is-commitment with respect to a core descriptive content, and an evaluative belief is an ought-commitment with respect to a core descriptive content. Core descriptive contents are those descriptive contents expressible by (nonevaluative) atomic sentences. Concerning the notion of a core descriptive content, Corbi says:

Core descriptive contents are not, despite appearances to the contrary, descriptive. For descriptive contents have essentially to do with a representation of “the world as being a certain way” and thereby they already have a certain direction of fit; while core descriptive contents are merely concerned with “a way-the-world-might-be content” because they are meant to be neutral with regard to the direction of fit. (p. 3)

But when we said that a descriptive content is a “way-the-world-might-be” content, the point of the modal word ‘might’ was merely to acknowledge that the world might not in fact be the way the given descriptive content says. This is a “non-success” use of the word ‘descriptive’. Under this usage, for example, the statement “Al Gore is U.S President in 2002” has descriptive content, even though it is not true—and hence does not describe how things actually are. Descriptive content certainly has objective truth conditions.

Corbi questions our remark that in the case of base-case is-commitment states but not in the case of base-case ought-commitment states, the belief’s declarative content “coincides with its core descriptive content.” But we think our remark should be noncontroversial when understood as we intended it. Consider these two belief-attributions:

(a) Carme believes that she is studying hard.

(b) Carme believes that she ought to be studying hardl.

The logical form of (b), as we construe it, is made more perspicuous this way:

(b*) Carme believes that it ought to be that she is studying hard.

Both of the beliefs attributed by these statements have the same core content, viz., the content expressed by the ‘that’-clause ‘that Carme is studying hard’. Statement (a) reveals what we meant by saying that in the case of a base-case descriptive belief, the belief’s declarative content coincides with its core descriptive content; the point is that both contents are expressed by the statement’s ‘that’-clause. Likewise, statement (b*) reveals what we mean by saying that in the case of a base-case evaluative belief, the belief’s declarative content does not coincide with its core descriptive content; here the point is that the belief’s declarative content is expressed by the ‘that’-clause appended to the word ‘believes’, whereas the belief’s core descriptive content is expressed by the distinct ‘that’-clause appended to the words ‘it ought to be’

Although Timmons and I are nondescriptivists about moral beliefs, we contend that descriptivists have reason to embrace our overall framework for belief. A descriptivist seeking to accommodate metaethical internalism could maintain, for example, that the belief expressed by statement (b*) is both an ought-commitment with respect to the core content that Carme is studying hard, and an is-commitment with respect to the putatively descriptive declarative content that it ought to be that Carme is studying hard. But Corbi objects that our framework rules out this kind of descriptivism, by virtue of a statement of ours that he quotes, viz., “Evaluative beliefs differ essentially from descriptive beliefs in the following respect: the core descriptive content of an evaluative belief does not coincide with its core declarative content.” Corbi takes us to be presupposing here that metaethical descriptivism is false. But the quoted statement was meant to be neutral about metaethical descriptivism; it can be rephrased by replacing the word ‘descriptive’ by the word ‘non-evaluative’, prior to the second occurrence of ‘beliefs’. (In the quoted sentence, we briefly adopted a common alternative use of ‘descriptive’ that unfortunately can cause confusion in the context of discussions of descriptivism vs. nondescriptivism—a use in which ‘descriptive’ essentially just means ‘non-evaluative’.)

Corbi takes us to hold that a benchmark difference between descriptive and non-descriptive content is that in the case of the former, rational agents can achieve convergence—an idea that Crispin Wright (1992) calls the “cognitive command” constraint. But this is not our view. Rather, for us the key distinguishing feature of descriptive content (as opposed to nondescriptive content) is this: for the former (but not the latter), the operative semantic standards are what we call tight. By this we mean that the semantic standards, together with how things are with respect to the PROPERTIES and RELATIONS instantiated by OBJECTS, jointly conspire to determinately fix correct-affirmability and correct-deniability status (i.e., truth or falsity) for a given declarative content. We actually agree with Corbi that in general, a requirement of rational convergence cannot even be satisfied by those discourses that are paradigmatically presented as descriptive.

This last point leads directly from Corbi’s discussion of our nondescriptivist cognitivism into his critical remarks about our argument against Michael Smith’s version of descriptivism. Corbi objects that our argument presupposes that rational convergence is a demarcating constraint on descriptive content. But this is a misunderstanding; as just explained, we do not accept such a constraint.

Our argument against Smith does indeed appeal to a certain kind of hypothetically possible non-convergence, between those on earth and those on “Moral Twin Earth.” But our use of this divergence-scenario against Smith appeals not to a putative convergence-constraint on descriptive content, but rather to the specific kinds of truth conditions that Smith himself proposes for moral claims. Smith holds, roughly, that something is good iff any rational person would desire it. The trouble is that far too little will count as good, according to this account, because of the ever-present possibility of the relevant kind of desire-divergence between different rational agents (as illustrated by our case of Earthlings and Twin Earthlings, and also by the example of Putnam and Nozick).

Smith holds that statement (T1), “Normative reasons require convergence in the desires of fully rational agents,” is a conceptual truth. Timmons and I granted this claim (at least for argument’s sake), but argued that the statement’s being a conceptual truth (if indeed it is) does not help Smith; for, the relevant sense of ‘rational’ would be a morally thick one incorporating certain specific moral standards, whereas what Smith needs in order for his rationalist account of morals to work is that there be rational convergence of desires under a morally thin notion of rationality. So our principal argument against Smith is not harmed if one calls (T1) into question (as Corbi does). Also, when Timmons and I granted for the sake of argument that (T1) is a conceptual truth (when ‘rational’ is construed in a morally thick way), we were not buying into anything like Wright’s cognitive-command constraint. Rather, the source of the alleged conceptual truth here, according to Smith, is much more local: it is the alleged constitutive content of the specific notion of a normative reason.

Stephen Barker maintains that nondescriptivist cognitivism, as Timmons and I articulate it, is incoherent because it conflates two senses of ‘content’ which, when clearly articulated, reveal inconsistencies in our position. He misconstues our position, but in an instructive way.

We reject the Semantic Assumption, which asserts that all genuinely cognitive content is descriptive content (i.e., way-the-world-might-be content). Consider the sentences that Barker labels (1)-(4):

(1) Bertie will mail the parcel.

(2) Bertie ought to mail the parcel.

(3) Jane believes that Bertie will mail the parcel.

(4) Jane believes that Bertie ought to mail the parcel.

We claim that the overall declarative content of (2), and of the belief attributed by (4), is not descriptive. So we also reject the assumption that Barker calls Orthodoxy, which asserts that a belief state is always a relation between a believer (speaker) and a proposition (or sentence, or whatever) such that what is believed is something having overall descriptive content.

Barker, focusing on our rejection of Orthodoxy, takes us to be embracing a thesis he calls Content Identity, which asserts two things: first, that the overall declarative content of sentence (1) and of the belief attributed by sentence (3) is an is-commitment with respect to the core content {Bertie mails the parcel}; and second, that the overall declarative content of sentence (2) and of the belief attributed by sentence (4) is an ought-commitment with respect to this same core content. Here is a key passage from Barker, revealing his rationale for this reading us this way:

H/T claim that they are overthrowing the orthodox conception of belief: Orthodoxy. To explain the belief attributions (3) and (4)...H/T cannot merely affirm that there are different kinds of belief states, which can take the same kinds of objects, they need rather to show that there are belief states that take different kinds of objects. So H/T must explain what these two sorts of belief-objects are meant to be and to provide evidence that there are two such types of object. H/T purport to be explaining what they mean by different kinds of belief-objects through their content identity claims, Content Identity.... This seems to be the assurance H/T supply themselves with that objects of belief can come in two forms, one form somehow representational and the other somehow nonrepresentational. And this they take to furnish themselves with the means of denying Orthodoxy. But Content Identity is an instance of flagrant content/state conflation. Declarative contents are meant to be contents. But now they are being equated with states. (pp. 8-9)

I have several things to say about this construal of our position, and about the rationale motivating the interpretation. First, it is certainly a mistaken interpretation. Timmons and I never explicitly equated declarative contents with is-commitment states or ought-commitement states, and we never intended to do so. We entirely agree with Barker that this would be a flagrant conflation between what he calls “content qua semantic object of a contentful psychological state,” and “content qua the state itself that possesses such content” (p. 6). We also agree that adopting Content Identity would make the resulting position untenable, and perhaps outright incoherent. One reason (among others) why we should reject Content Identity, as Barker rightly points out, is that we ourselves say that ought-commitments are suspended in the case of logically complex beliefs with embedded normative contents, which strongly suggests that ought-commitments “are features of force and should not then be equated with declarative content” (p. 11).

But second, the underlying philosophical demand that motivates Barker’s interpretation is an important and legitimate one (even though it causes him to misconstrue our position). In order for the H/T framework to be viable, it needs to be incorporatable within a more complete theoretical treatment of belief-states that credibly explains their ontological structure. This will have to be a theoretical treatment that repudiates Orthodoxy and offers some alternative positive account of the ontological structure of at least some belief-states—viz., beliefs whose overall declarative content is not descriptive.

Third, I acknowledge that no such positive account of the ontological structure of belief-states was provided, in Horgan and Timmons (2000). To that extent, the framework for belief we described was incomplete in an important respect. And I acknowledge that unless a credible positive account of the requisite kind can be given, our framework will not be tenable.

Fourth, I do think that the requisite treatment of the ontology of belief-states can be provided. One approach, for instance, would be to adapt the proposal in Fodor (1968) that the objects of the propositional attitudes are sentences in the “language of thought”; the class of eligible mentalese sentences could then include ones with non-descriptive content. (I do not much like Fodor’s proposal, however; cf. Horgan 1992). I myself am inclined instead to favor an approach to the ontological structure of belief-states, and to the logical form of belief-attributions, proposed by Quine in Word and Object. After canvassing various alternative construals of belief-objects, Quine wrote:

A final alternative that I find as appealing as any is simply to dispense with the objects of the propositional attitudes. We can...formulate the propositional attitudes with the help of the notations of intensional abstraction...but just cease to view these notations as singular terms referring to objects. This means viewing ‘Tom believes [Cicero denounced Catiline]’ no longer of the form ‘Fab’ with a = Tom and b = [Cicero denounced Catiline], but rather of the form ‘Fa’ with a = Tom and complex ‘F’. The verb ‘believes’ here ceases to be a term and becomes part of an operator ‘believes that’, or ‘believes [ ]’, which applied to a sentence, produces a composite absolute general term whereof the sentence is counted an immediate constituent. (Quine 1960, p. 216)

He also proposed generalizing this treatment to cover “de re” belief-constructions. He later called the envisioned propositional attitude operators “attitudinatives” (Quine 1970, pp. 32-33). In Horgan (1989) I elaborated and defended Quine’s proposal, with particular attention to explaining how the meaning of a Quineian attitudinative construction is dependent upon the meanings of its components.

Barker observes that Timmons and I adopt a “dispositionalist” treatment of beliefs whose overall declarative content is logically complex. As we said in our paper:

The essential feature of any given logically complex commitment-type is its distinctive constitutive inferential role in an agent’s cognitive economy (insofar as the agent is rational), a role involving the relevant core descriptive contents.... So, for example, consider the belief that either Jeeves mailed the parcel or Bertie ought to mail the parcel.... The embedded moral constituent, Bertie ought to mail the parcel, is in the offing in the sense that the complex commitment-state in question, together with the belief that Jeeves did not mail the parcel, inferentially yields (at least for the minimally rational agent) an ought-commitment with respect to Bertie’s mailing the parcel. (Horgan and Timmons 2000, pp. 136-7).

Barker raises a legitimate concern about how such an inferentialist approach is supposed to handle beliefs like the belief that George ought not to be President. This prompts a revision in our position. Timmons and I now would propose a non-inferentialist construal of minimimally complex negative beliefs—that is, beliefs of the logical type ~F and the logical type Ought(~F), where F is a core descriptive content (and hence is expressible as an atomic sentence). Briefly, the idea is that such negative beliefs, like their affirmative counterparts, should be construed as commitment states with a direct constitutive psychological role in cognitive economy (rather than a constitutive inferential role). There are two direct, noninferential, kinds of is-commitment with respect to a core descriptive content and likewise two direct kinds of ought-commitment—affirmative and negative.[4] With this improvement incorporated into our position, I think the treatment of embedding in section VII.2 of our paper otherwise goes through as before, mutatis mutandis.

Danilo Suster raises worries both about the general metaphilosophical approach I advocate with respect to philosophical inquiry about philosophically important concepts, and about the application of this approach to the issue of freedom and determinism. Let me begin with some clarificatory remarks, as a prelude to addressing some of his specific objections.

First, the key principles of the general metaphilosophical position that George Graham and I call “Southern Fundamentalism” are not really intended primarily as methodological strictures, but rather as ideological hypotheses of a very general and abstract sort. Whenever some specific concept is being investigated, a specific case needs to be made for the claim that the given concept falls toward the “austere” end of the spectrum of competing hypotheses about its ideological workings; likewise, a specific case needs to be made that any ideological “ideological-opulence tendencies” associated with the concept can be respectfully explained away as involving some kind of subtle, natural, mistake rather than some egregious blunder.

Second, Post-Analytic Metaphilosophy treats general claims involving philosophically interesting concepts as ideological hypotheses. (Examples are the hypothesis that knowledge is analysable as justified true belief, and the hypothesis that built into the concept of freedom is the incompatibility of freedom and determinism.) Among the most important kinds of data, vis-a-vis these hypotheses, are people’s intuitive judgments and judgment-tendencies concerning the proper application of a given concept with respect to specific concrete scenarios. Although people also sometimes have intuitive judgments or judgment-tendencies about the truth or falsity of the ideological hypotheses themselves, and although these latter intuitions are data too, they are especially defeasible because often they are good candidates to be respectfully explained away.

Third, the principle of austerity—the claim that philosophically interesting concepts generally are austere rather than opulent (one key tenet of Southern Fundamentalism)—needs to be properly understood. The idea is this: among the competing ideological hypotheses each of which does reasonably well at accommodating relevant data about the workings of the relevant concept(s), the correct ideological hypothesis generally will be comparatively more austere than the other competing hypothesis (or hypotheses). Ideological hypotheses that do badly at accommodating the relevant data—including data about people’s intuitive judgments about how to apply the concept(s) in specific concrete scenarios—usually are non-starters, even if these hypotheses construe the concept(s) as highly austere.

Let me turn now to Suster’s objections. He questions the principle of austerity, by focusing on the concept of knowledge and Gettier cases:

Take now the philosophically interesting case of knowledge. Suppose it is discovered that some justified true beliefs do not pass the Gettier test. Should we then conclude that these beliefs do not count as knowledge? According to the principle of austerity the proper conclusion is that meeting the extra condition is not a genuine prerequisite for knowledge (p. 2).

Not so, because any credible ideological hypothesis about the concept of knowledge needs to do well at accommodating people’s intuitive judgments about how to apply the concept of knowledge in concrete cases—specifically, in Gettier cases. The principle of austerity only says this: among competing ideological hypotheses that accommodate such judgments, the correct one normally will be more austere than the competing one(s). (For example, the correct one will be a hypothesis according to which I really do know that I have hands, even though my evidence for this is consistent with the possibility that I am a brain in a vat.)

We come now to the issue of freedom and determinism. I contend that people’s intuitive judgments about how to apply the notion of freedom in various concrete scenarios constitute one important kind of data in support of compatibilism. Suster counters that the incompatibilists have their own favorite stock of concrete examples, and that people’s intuitions about how to describe these examples constitute evidence for incompatibilism. He cites van Inwagen’s Martian scenario (‘M’) as an example.

I grant that an agent in scenario (‘M’) would not really act or choose freely; at any rate, this is my own intuitive judgment about the case too. But it seems to me that this fact does not count against my argument for compatibilism at all. For, I take it that our ordinary attributions of freedom to ourselves and others implicitly presuppose that such a scenario does not obtain vis-a-vis the persons of whom the attribution is made. And this presupposition is certainly justified, and indeed is surely true—even though it is defeasible.[5]

By contrast, the putative presupposition of radical causal indeterminism in the etiology of human actions and choices—which the opulent conception of freedom claims is built into our ordinary attributions of freedom—is not well justified at all. So if this presupposition is really in effect, then our ordinary evidential standards for freedom attributions are sorely lax, and our intuitive judgments about concrete cases become highly suspect because they are so sorely lacking in adequate justification. This result counts heavily against incompatibilism.

Suster questions my favored contextualist approach to freedom attributions, according to which (i) these attributions are compatible with determinism under the semantic standards that normally govern the notion of freedom, but (ii) the attributions become incompatible with determinism under maximally strict semantic standards that tend to come into play when the problem of freedom and determinism is posed explicitly. His objection, as I understand it, is that my contextualist account is not relevantly analogous to familiar contextualist treatments of knowledge attributions. He points out that these latter accounts often say that the way the standards get raised contextually in the case of knowledge—for instance, by explicitly posing the possibility that one is a brain in vat—is a matter of enlarging the range of relevant possibilities to be considered: under the raised standards, I do not know that I have hands unless my evidence precludes the possibility that I am a brain in a vat. He objects that my own contextualist treatment of freedom attributions does not work this way: it does not treat the raising of semantic standards as a matter of enlarging the range of relevant possibilities.

But although Suster is surely right about the disanalogy, I do not see it as a problem for my treatment. On the contrary: in the case of freedom attributions, it appears that raising the contextually operative semantic standards is a matter of shrinking the range of relevant possibilities—specifically, the range of circumstantially similar possible worlds that are relevant to the question whether the agent could have done otherwise. Under the semantic standards that govern freedom attributions in ordinary contexts, the compatibilist claims, this “accessibility” relation includes possible worlds in which either (a) the agent’s past is somewhat different from the actual-world past, or (b) the prevailing laws of nature are violated in certain ways prior to the agent’s choice or action (e.g., violated in a way that David Lewis, in his treatment of counterfactual conditionals, calls a “minor divergence miracle”; cf. Lewis 1979). Under maximally strict semantic standards, on the other hand, the range of relevant possibilities is diminished: the accessibility relation now holds fixed both (i) the agent’s entire past right up until the moment of choice or action, and (ii) the prevailing laws of nature.

Timothy Williamson sets forth a critique of transvaluationism, my position on vagueness. His objections persuade me that my prior writings on vagueness do not yet provide a sufficiently clear, sufficiently coherent, formulation of the view. Lately I have been working on a new articulation, specificially with an eye to addressing Williamson’s worries. Here I will provide a very brief synopsis of how I now propose to package the key ideas.

Before doing so, let me acknowledge a mistake I made in discussing iterated supervaluationism. Williamson quotes an argument I gave that was intended to show that iterated supervaluationism cannot avoid the logical incoherence that I maintain is endemic to vagueness. In Horgan (1998) I claimed that logical incoherence is manifested by the fact that an iterated supervaluationist is committed to affirming the statement

[B(0) & ~B(1)] v [[B(1) & ~B(2)] v ... v [B(10¹⁷) v ~B(10¹⁷)]

and yet under forced-march querying would be forced to say, of each respective disjunct, that it is not true. (‘Bn’ abbreviates ‘A man with n hairs on his head is bald’.) But as Williamson rightly points out, there is nothing immediately incoherent about this at all, given the semantics of supervaluationism; for, as he says, it is a familiar supervaluationist phenomenon to have a supertrue disjunction with no supertrue disjunct. So my reasoning was in error. (I do still believe that iterated supervaluationism is best construed as an implementation of transvaluationism rather than an alternative to it, but on different grounds: briefly, this construal best explains why the iterated supervaluationist is compelled to avoid pronouncing on the semantic status of each successive statement in a sorites sequence.)

Let me now briefly summarize how I propose to articulate the key claims of transvaluationism. First, vagueness essentially involves boundarylessness. Second, to say that vagueness exhibits boundarylessness is to say that for a suitable sorites sequence for a given vague term, each of the following conditions is satisfied: (1) initially there are true statements (with each predecessor of any true statement being true), (2) later there are false statements (with each successor of a false statement being false), and (3) there is no determinate fact of the matter about the transition from true statements to false statements. Third, satisfaction of condition (3) is a matter of the semantic governance of vague discourse by these two mutually unsatisfiable semantic-status principles:

The Individualistic Same-Status Principle: Each item in the sorites sequence has the same status (e.g., truth, falsity, neither true nor false) as its immediate neighbors.

The Collectivistic Status-Indeterminacy Principle: No determinate overall assignment of statuses to the items in the sequence is correct.

Fourth, such semantic governance requires that these two status-principles must be adequately respected in assertoric practice (even though the status-principles are not mutually satisfiable). Fifth, semantically appropriate assertoric practice is itself a matter of conformity with certain mutually obeyable practice standards, in particular the following two prohibitory ones:

The Individualistic Status-Attribution Prohibition: Never attribute a specific status to an item in a sorites sequence and also attribute a different, incompatible, status to its immediate neighbor.

The Collectivistic Status-Attribution Prohibition: Never affirm a determinate overall assignment of statuses to the items in a sorites sequence.

In short, although condition (3) of boundarylessness must actually be satisfied by vague discourse, its satisfaction consists in semantic governance by two mutually un-satisfiable semantic-status principles. This governance is a matter of the way the semantic-status principles undergird mutually obeyable assertoric-practice standards, notably the two prohibitory ones just stated. Vague discourse that obeys these practice-standards thereby respects the semantic-status principles as well as possible—and well enough.

Obeying the practice-standards requires one, on occasion, simply to refuse to answer certain sets of perfectly meaningful questions—such as a set of questions about the semantic statuses of the successive statements in a sorites sequence. One must adopt this non-responsive “zen attitude” (a phrase suggested by Matjz Potrč) even though no unknown facts are involved, such as sharp but unknowable boundaries. Herein is manifested, in practice, the underlying logical incoherence of vagueness.

The ideas just set forth concerning vague discourse also apply, mutatis mutandis, to vague thought-content. (Now the relevant practice-standards concern affirmatory practice in thought, rather than in language.) However, when we ask whether there could be such a phenomenon as ontological vagueness—that is, vague OBJECTS and/or vague PROPERTIES and RELATIONS—the situation is very different. Vague ENTITIES would have to exhibit an ontological form of boundarylessness—and in particular, would have to satisfy the condition that there is “no determinate fact of the matter” about transitions in a sorites sequence. But what could constitute the satisfaction of this condition, in the case of ontological vagueness? Here I see no intelligible analogue of the key idea I appeal to in the case of vague language and thought—viz., the idea of affirmatory practice being semantically governed by mutually unsatisfiable semantic status-standards. Rather, evidently the putative vague ENTITIES would have to directly satisfy both the individualistic same-status principle and the collectivistic status-indeterminacy principle. Since this is impossible, so is ontological vagueness.

Winfried Loeffler focuses on some epistemological work of David Henderson and myself. He expresses sympathy with our approach in Henderson and Horgan (2000), but urges us to further clarify the notions of “robust reliability” and “epistemically relevant possible world” that figure in Henderson and Horgan (2001). We are working on further elaboration and clarification in a book now in progress. I will offer a few tentative clarificatory remarks here, with an eye toward addressing Loeffler’s specific concerns about how to understand our examples of sample-biased inductive belief-forming processes that are de facto reliable but are not robustly reliable.[6]

Henderson and I maintain that a belief’s being objectively justified is not merely a matter of its having been produced by a belief-forming process that in fact happens to reliably produce true beliefs (even though such reliability is indeed one important dimension of objective justification). Another important dimension is robustness of reliability, which we gloss as “reliability across a very wide range of epistemically relevant possible worlds.” I am currently inclined to relativize the notion of an epistemically relevant possible world, to a cognitive agent at a time. These worlds are the ones that are consistent with the occurrence of the agent’s actual prior and present experiences—that is, the agent’s experiences prior to, and concurrent with, the occurrence of the belief itself.[7] (Such a world can be consistent with the occurrence of these experiences without being consistent with their intentional content—for example, a world in which the agent is radically experientially deceived by a Cartesian demon.)

We claim that a belief can sometimes have substantial objective justification even though the belief-forming process that produced it is not actually reliable—i.e., is not reliable in the possible world that happens to be actual. An example would be a brain-in-vat duplicate of yourself, whose transducer-internal belief-forming processes match your own. We also claim that a belief can sometimes be significantly deficient in objective justification even though it is produced by a belief-forming process that in fact happens to be reliable. This can happen, for instance, for a cognitive agent with these two attributes: (i) the agent forms inductive beliefs via inductive belief-forming mechanisms that are grossly inattentive to matters of sample bias, but (ii) the agent happens to occupy a possible world that is exceptionally homogenous in the very respects about which the agent is disposed to reason via hasty generalization. (Not only have the few cats the agent has encountered been stupid, but virtually all cats are stupid; not only have the few Frenchmen the agent has encountered been surly, but almost all Frenchmen are surly; etc.) The reason why the agent’s inductive beliefs are seriously deficient in their objective-justification status, even though they are produced by inductive belief-forming processes that fortuitously happen to be reliable, is that these processes are not robustly reliable—that is, there are epistemically relevant possible worlds that are not homogenous in the way the agent’s actual world happens to be, and in these worlds the same belief-forming processes are not reliable.

Loeffler’s main concern, as I understand it, is that the homogeneity or non-homogeneity of a cognitive agent’s actual world ought to be something that can manifest itself in potential differences in the agent’s experiences—in the appearances. He writes:

This example is not convincing. It is incoherent to suppose that two possible worlds could differ in the way that it presupposes. The homogeneity or otherwise of the population cannot be thought of simply as making a difference to the world an sich; it must surely in the end affect appearance as well. As samples become larger, the differences between a homogenous and heterogeneous population become apparent: otherwise we are unable to distinguish between homogeneous and heterogeneous populations at all. As soon as we check a sufficient number of objects, the appearance of the world will differ as well. (p. 7)

It is true enough that in possible worlds in which the agent is not radically experientially deceived (as opposed to worlds in which, for instance, the agent is deceived by a Cartesian demon or is a brain in a vat), normally there are potential experiences available to the agent that would reflect the homogeneity or nonhomogeneity of the agent’s world. But under my current proposal, these merely potential experiences are not directly germane to delimiting the class of epistemically relevant possible worlds, relative to a specific cognitive agent at the specific time when a given belief arises in the agent as the result of a sample-biased belief-forming process. For, again: the epistemically relevant possible worlds, relative to the agent at the time, are the ones consistent with the occurrence of the agent’s actual experiences prior to, and concurrent with, the occurrence of belief itself.

REFERENCES

Fodor, J. 1968. Propositional Attitudes, The Monist 61, 501-23. Reprinted in his Representations (MIT Press, 1981).

Graham, G. and Horgan, T. 2000. Mary Mary, Quite Contrary, Philosophical Studies 99, 59-87.

Henderson, D. and Horgan, T. 2000. Iceberg Epistemology, Philosophy and Phenomenological Research 61, 497-535.

Henderson, D. and Horgan, T. 2001. Practicing Safe Epistemology, Philosophical Studies 102, 227-58.

Horgan, T. 1984. Jackson on Physical Information and Qualia, Philosophical Quarterly 34, 147-52.

Horgan, T. 1989. Attitudinatives, Linguistics and Philosophy 12, 133-65.

Horgan, T. 1992. From Cognitive Science to Folk Psychology: Computation, Mental Representation, and Belief, Philosophy and Phenomenological Research 55, 449-84.

Horgan, T. 1993. From Supervenience to Superdupervenience, Mind 102, 555-86.

Horgan, T. 1994. Naturalism and Intentionality, Philosophical Studies 76, 301-26.

Horgan, T. 1998. The Transvaluationist Conception of Vagueness, The Monist 81, 316-33.

Horgan, T. 2001. Causal Compatibilism and the Exclusion Problem, Theoria 16, 95-116.

Horgan, T. and Potrč. 2000. M. Blobjectivism and Indirect Correspondence, Facta Philosophica, 2: 249-70.

Horgan, T. and Potrč, M. In Press. Addressing Questions for Blobjectivism, Facta Philosophica.

Horgan, T. and Tienson, J. 1996. Connectionism and the Philosophy of Psychology. MIT Press.

Horgan, T. and Tienson, T. 2001. Deconstructing New Wave Materialism. In C. Gillett and B. Loewer, ed., Physicalism and Its Discontents (Cambridge), 307-18.

Horgan, T. and Tienson, J. In press. The Intentionality of Phenomenology and the Phenomenology of Intentionality. In D. Chalmers, ed., Philosophy of Mind: Classical and Contemporary Readings (Oxford).

Horgan, T. and Timmons, M. 1992. Troubles on Moral Twin Earth: Moral Queerness Revived, Synthese 92, 221-60.

Horgan, T. and Timmons, M. 2000. Nondescriptivist Cognitivism: Framework for a New Metaethic, Philosophical Papers 29, 121-53.

Jackson, F. 1998. From Metaphysics to Ethics: A Defense of Conceptual Analysis. Oxford University Press.

Kim, J. 1998. Mind in the Physical World: An Essay on the Mind-Body Problem and Mental Causation. MIT Press.

Lewis, D. 1979. Counterfactual Dependence and Time’s Arrow, Nous 13, 455-76.

McGinn, C. 1991. The Hidden Structure of Consciousness. In his The Problem of Consciousness (Blackwell).

McGinn, C. 1993. Problems in Philosophy: The Limits of Inquiry. Blackwell.

McLaughlin, B. 1992. The Rise and Fall of British Emergentism. In A. Beckermann, H. Flohr, and J. Kim, eds., Emergence or Reduction? Essays on the Prospects of Nonreductive Physicalism (Walter de Gruyter).

Putnam, H. 1995. Pragmatism, Proceedings of the Aristotelian Society 95, 291-306.

Quine, W. V. O. Word and Object. MIT Press.

Quine, W. V. O. 1970. Philosophy of Logic. Prentice-Hall.

Tienson, J. In press. Questions for Blobjectivism, Facta Philosophica.

Van Inwagen, P. 1990. Material Beings. Cornell University Press.

Wright, C. 1992. Truth and Objectivity. Harvard University Press.

---

[1] In Horgan (1993) I also explicitly distinguished the philosophical question of what constitutes a broadly materialistic worldview from the philosophical question of what constitutes a broadly naturalistic worldview, and I remarked that it is also an important philosophical question how (if at all) metaphysical naturalism might differ from materialism. See note 2 on p. 556, and the statement in the text to which this note is appended.

[2] Tienson’s position on this matter is somewhat similar in spirit to Kim’s recent argument (Kim 1998, pp. 80-87) that the exclusion problem does not generalize to special sciences across the board. Kim maintains that micro-based macro-properties typically have their own causal powers. He claims that the exclusion problem really arises not for properties of wholes vis-a-vis properties of their parts, but rather for higher-order properties instantiated by an individual vis-a-vis lower-order properties instantiated by that same individual.

[3] Here I am glossing over the fact that the new experience underlying Mary’s new belief is an experience of a specific shade of blue—just as Byrne glosses over this fact in his discussion. For present purposes, then, I am ignoring the question of what it takes to acquire the generic concept BLUE, over and above experiencing this single specific shade of blue. Also, the ineffability I am talking about—and that Byrne too is talking about, I take it—is supposed to attach to a concept that involves the generic color blue, rather than to a concept that involves just a single specific shade of blue.

[4] Of course, all four kinds of belief can combine with other beliefs to generate new beliefs (for a rational agent). In this sense, even these four minimally complex kinds of belief have a constitutive inferential aspect. But the point is that this aspect does not exhaust their constitutive role in psychological economy.

[5] It should be noted that van Inwagen himself deploys scenario (‘M’) against “paradigm case arguments,” which contend that freedom-attributions cannot be mistaken for the kinds of concrete cases that serve as paradigms for the teaching of the meaning of the word ‘free’. I grant that (‘M’) effectively refutes such arguments. But although there is some similarity between paradigm-case reasoning and my own argument for incompatibilism, there is this crucial difference: my argument does not assume that satisfying the paradigmatic evidential conditions for freedom-attribution is automatically sufficient for genuine freedom. Scenario (‘M’) shows that this assumption is mistaken.

[6] My remarks will be tentative because they may not survive ongoing give-and-take between Henderson and myself. As I write, the two of us have not yet settled into joint reflective equilibrium on this matter.

[7] Does this mean that the notion of robustness of reliability also becomes relativized to a cognitive agent at a time? Not necessarily. Perhaps this notion should be characterized in something like the following way: A belief-forming process P is robustly reliable, for creatures of kind K (e.g., humans), just in case: for any temporally extended course of experience E that could be instantiated by creatures of kind K, P is reliable in a very wide range of the possible worlds that are epistemically relevant relative to E. (One might want to put further restrictions on the kinds of temporally extended courses of experience that are cited in the definition—e.g., that they have coherent intentional content, that they include experience as of being an embodied agent within a perceived ambient environment, and the like.)