Descriptive Atomism and Foundational Holism Compositionality and Semantic Properties

While holism and atomism are often treated as mutually exclusive approaches to semantic theory, the apparent tension between the two usually results from running together distinct levels of semantic explanation. In particular, there is no reason why one can’t combine an atomistic conception of what the semantic values of our words are (one’s “descriptive semantics”), with a holistic explanation of why they have those values (one’s “foundational semantics”). Most objections to holism can be shown to apply only to holistic version of descriptive semantics, and do not tell against any sorts of holistic foundational semantics. As Davidson’s work will be used to illustrate, by clearly distinguishing foundational and descriptive semantics, one can capture the most appealing features of both holism and atomism. Semantic holism and semantic atomism are typically viewed as mutually exclusive options when thinking about the nature of concepts and semantic content. Nevertheless, the apparent tension between atomistic and holistic theories often results from running together distinct levels of semantic explanation. For instance, in what follows, it will be argued that Davidson can justly be characterized as both an atomist and a holist about meaning. There is no contradiction in Davidson’s position because he combines an atomistic story about what the semantic values of our words are, with a holistic explanation of why they have those values. When discussing how our individual words hook on to the world, he presents a holistic theory, and when explaining how we understand sentences and complex expressions in terms of their parts, he gives an atomistic theory. The problems of intentionality and productivity are distinct, and different sorts of theories can be used to respond to them. Many assume that if you are a holist or atomist at one level, you must be so at both, but as Davidson’s work illustrates, one can easily combine these perspectives that might otherwise seem incompatible. Henry Jackman Descriptive Atomism and Foundational Holism Compositionality and Conceptual Role To see how one could be both a holist and an atomist, we should first consider how the contrast between the two positions is laid out by the authors who are probably the most prominent defenders of the atomistic approach, Jerry Fodor and Ernest Lepore (hereafter “F&L”). F&L frame the holism/atomism dispute in terms of a distinction between “Old Testament” and “New Testament” Semantics (hereafter “OTS” & “NTS”). OTS takes meaning to be a type of “symbol-world relation”, so, say, ‘dog’ means dog “because of some (nonsemantic) relation that holds between the symbol and the animal”. By contrast, according to NTS, “the meaning of an expression is at least partially constituted by the expression’s inferential relations.” For NTS, the meaning of, say, “dog” in my language is partially constituted by the fact that I’m disposed to infer “x is a mammal” or “x is not a cat” from “x is a dog”, and will also infer “x is a dog” from “x is a Poodle”, etc.. According to F&L, NTS slides inevitably into holism (because there is no principled way to make a distinction between those inferential connections which are meaning constitutive and those which are not) while OTS tends to be atomistic. F&L have many objections to holism and NTS, most of which relate to the purported fact that NTS makes meaning so unstable and idiosyncratic that there is no room left for any constancy of meaning over change of belief, since any time we change a belief, the inferential roles of all of our terms change. For instance, if Peter infers “x tends to like cheese” from “x is a dog”, and Roger doesn’t, then the two don’t mean the same thing by “dog”. However, this means that they are making different inferences when they conclude, say, “x is not a dog” from “x is a building” so they must mean something different by “building” as well, and so on throughout the language. Because of this seeming instability, there can be no objective disagreement over matters of fact (since any two people who disagree about a sentence will have different See especially their paper “Why meaning (probably) isn’t conceptual role” (Fodor & Lepore ). F&L p. . Popular candidates for such a relation being resemblance, association, and (Fodor’s favorite) asymmetric causal dependence. (Margolis takes atomism itself to be characterized in this way, claiming that atomism is simply the position that concepts are identified by their extension (Margolis , p. ). F&L , p. . See F&L , 00 . F&L , p. . inferential roles for it, and thus mean something different by it), or communication in the sense of grasping the content of what another says (since no two people – or even no two stages of a single person – ever have precisely the same beliefs, no two people ever mean the same thing by any of their terms). Nevertheless, unattractive as F&L (quite justly) find such apparent consequence of NTS to be, they also recognize that these are all bullets that philosophers enamored of NTS could bite (indeed, have bitten). Consequently, they take the most decisive objection to NTS to be that it is incompatible with the compositionality of meaning, and so compositionality requires that we be atomists. F&L summarize this argument against NTS as follows: ( ) meanings are compositional, ( ) but inferential roles are not compositional ( ) so meanings can’t be inferential roles. To use their example, the meaning of “brown cow” is composed from the meaning of “brown” and the meaning of “cow”, but it doesn’t seem as if the inferential role of “brow cow” is composed by the inferential role of “brown” and the inferential role of “cow”. There are inferences that we would make about brown cows that we would not make about either cows in general or brown things in general. For instance, we might endorse the inference: [brown cow -> dangerous], even if we don’t endorse [brown -> dangerous] or [cow -> dangerous]. As F&L put it: prima facie, the inferential role of ‘brown cow’ depends not only on the inferential role of ‘brown’ and the inferential role of ‘cow’, but also on what you happen to believe about brown cows. So, unlike meaning, inferential role is, in the general case, not compositional. 0 If such criticisms are on the right track, then it is simply a mistake to identify See F&L . I respond to such objections in Jackman , 00 a. F&L , p. . F&L , p. . Similar arguments apply to those attempts to identify meaning with prototypes. (The prototype for ‘pet fish’ (small, gold, lives in small glass bowl, etc.), doesn’t come from the prototypes of ‘pet’ and ‘fish’. (See Fodor & Lepore .) Fodor & Lepore , p. . 0 F&L , p. . Of course, one might argue that the inferential role of “brown cow” can be erived from the inferential roles of “brown” and “cow” in that it is part of the inferential role of “brown” that “brown -> (cow -> dangerous)” and part of the inferential role of “cow” that “cow -> (brown Henry Jackman 0 Descriptive Atomism and Foundational Holism a term’s meaning with its inferential role. However, such criticisms of inferential role semantics need not force us to give up holism. In particular, while it may be the case that the ‘compositionality argument’ shows that a term’s meaning cannot be identified with its inferential role, it does not prevent one from claiming that facts about what a term means at least partially supervene upon such inferential roles. In other words, concepts need not be individuated in terms of the (possibly non-semantic) facts that they supervene upon. F&L assume that unless we renounce holism completely, we must identify meaning with conceptual role, and this commitment can be avoided by the holist who distinguishes: ( ) the idea that meaning can supervene upon inferential relations (where claims about inferential relations helping to ‘constitute’ meaning are understood as claims about what determines meaning), and ( ) the idea that meanings be identified with inferential relations (where the claim about what helps ‘constitute’ meaning is understood as a claim about what meanings are composed of). After all, Philosophers who stress the relation between meaning and conceptual role could be viewed as having one (or more) of the following five positions in mind. CR : A term has the meaning it does because it has a certain conceptual role. CR : A term’s meaning supervenes upon its conceptual role. CR : A term’s meaning is determined by its conceptual role. CR : A term’s meaning is constituted by its conceptual role. CR : A term’s meaning is its conceptual role. While F&L often treat these positions as interchangeable, they are all diffe-> dangerous)”. However, this just shows that, for the inferential role theorist, the meaning of, say, “brown”, is partially determined by the meaning of “brown cow”, and so there is not a one-directional flow of meaning from the simple terms to the complex ones. This is enough to show that for the inferential role theorist, meaning is not compositional in any traditional sense. For a related distinction, see Brigandt 00 , p. . I have also recently discovered (much to my dismay) that this point has also been persuasively presented with respect to the compositionality problem in Pagin . For CR , see F&L , pp , ; for CR , see F&L , pp. 0, , 0, , ; for CR , see F&L , pp. , , , (p. being the definition of NTS); for CR , see F&L 00 , pp. , , , pp. , , , , , , , (p. being the title). For the claim that F&L’s argurent, and strictly speaking, the compositionality argument only tells against CR . Some sorts of holistic theories, for instance those that individuate concepts in terms of their ‘narrow’ conceptual roles, or the sort of ‘state-space’ semantics associate with Paul Churchland, may indeed be committed to something like CR , but the sort of holistic semantics associated with Davidson’s work is more plausibly tied to weaker supervenience claims like CR . Supervenience requires only that there can’t be a difference of meaning without there being a difference in conceptual role. The function between conceptual role and meaning can thus be many to one rather than the one-toone function associated with CR . Further, if we think that meaning supervenes only partially on conceptual role (which seems compatible with F&L’s characterization of NTS), then the relation between meaning and conceptual role need not be a function at all, and there could be differences of meaning that didn’t have any associated differences of conceptual role. If one distinguishes between inferential roles helping to determine meaning and their being identified with meanings, one can see how a roughly Davidsonian theory can give a holistic explanation of the semantic relations between symbols and the world without abandoning the OTS claim that meanings should be individuated in terms of such symbol/world relations rather than their holistic supervenience base. Indeed, it is noteworthy that when one reads Davidson’s writings on semantic theory, he seems like he is engaged in traditional OTS, and his theory seems to have the satisfaction of the compositionality constraint as one of its primary desiderata. On the other hand, when his writings focus on radical interpretation (how we come up with the semantic theory for a particular speaker) he ment runs against the rocks because it conflates CR and CR , see Warfield . It might also tell against CR if CR is interpreted as being equivalent to CR (an interpretive claim I take to be characteristic of NTS), but CR can also quite naturally be interpreted as standing for CR or CR , in which case the objection doesn’t hold. See, for instance, Churchland . This is partially why the holism involved with Davidsonian radical interpretation is ‘moderate’ rather than ‘radical’. (For a discussion of this distinction, see Jackman , 00 a.) (For an earlier discussion of the importance of the one-to-one/many-to-one distinction for presenting a plausible account of holism, see Pagin , pp. and the works cited therein.) F&L , p. . Something like this line would have to be taken by the Davidsonian unless ‘conceptual role’ is taken in its ‘widest’ possible sense. See, for instance, Davidson , . Henry Jackman Descriptive Atomism and Foundational Holism sounds more like someone doing NTS. 0 This may make it hard to answer the question of whether Davidson is an atomist or holist tout court, but it represents no inconsistency on Davidson’s part. There is nothing wrong with combining what is, in essence, an atomistic semantic theory with a holistic theory of interpretation. (Even when one adds that constraints on interpretation are partially ‘constitutive’ of the meaning-facts under investigation.) One can usefully characterize all of this in terms of Robert Stalnaker’s distinction between “descriptive” and “foundational” semantics. Stalnaker presents the distinction as follows: First, there are questions of what I will call “descriptive semantics”. A descriptive-semantic theory is a theory that says what the semantics for the language is, without saying what it is about the practice of using that language that explains why that semantics is the right one. A descriptive semantic theory assigns semantic values to the expressions of the language, and explains how the semantic values of complex expressions are a function of the semantic values of their parts ... Second, there are questions, which I will call questions of ‘foundational semantics’, about what the facts are that give expressions their semantic values, or more generally, about what makes it the case that the language spoken by a particular individual or community has a particular descriptive semantics. Stalnaker notes that many run these two sorts of semantic projects together, assuming that the constraints on one must also be in play for the other. For instance, Stalnaker argues that Searle and Dummett reject Millian approaches to descriptive semantics because they implicitly assume that an answer to the descriptive question must provide an answer for the foundational question as well. In much the same way, one can understand F&L as rejecting holistic answers to the foundational question because they implicitly take them to also be answers to the descriptive semantic question. Indeed, both OTS and NTS, as understood by F&L, can be understood as working within this sort of ‘methodologically monistic’ framework where one’s descriptive and foundational semantics must either be both atomistic or both holistic. However, once this sort of methodological monism is given up, the constraints on each level will 0 See, for instance, Davidson , , , , , . It thus isn’t surprising that he is one of the primary targets of F&L . See, once again, Pagin , pp. . Stalnaker , p. . Stalnaker , . He refers here to Searle and Dummett . open up considerably. Davidson provides, as mentioned above, an example of this sort of methodological pluralism. To put things very roughly, on the Davidsonian view of descriptive semantics, we can understand the meaning of a term in terms of its satisfaction conditions, with, say, “cow” being satisfied by members of the set of cows and “brown” being satisfied by members of the set of brown things. In this respect, Davidson is no different from atomists like Fodor. Since semantic values of basic concepts are individuated extensionally, compositionality can be accounted for by allowing that complex concepts are constructed classically, with “brown cow” getting the semantic value it does by picking out the set of objects that satisfy both “brown” and “cow”. Nevertheless, the story will be very different at the level of foundational semantics. The Davidsonian understands the satisfaction conditions of any word in a speaker’s language as determined by the set of assignments that would maximize the truth of all of that speaker’s commitments. Many of these commitments would be associated with inferential roles such as the transition from “cow” to “animal” or “cow” to “member of a natural kind”, but others (such as those commitments manifested in one’s past applications of a term to particular objects) may not. Whether a commitment relating to a term is made true by a candidate extension for that term obviously depends upon the semantic values of the other words in the sentence held true, so the assignment of values that maximizes the truth of the speaker’s commitments will have to be determined holistically rather than on a word-by-word basis. For instance, if we decide to individuate a biological kind term, say “cow”, in terms of the genetic make-up of its members rather than their evolutionary history, this will affect the extension of the other biological kind concepts in the language, Further, it seems clear that compositionality is only a constraint on one’s descriptive semantics, and, for instance, Fodor makes no effort to show that compositionality is satisfied at the foundational level. Asymmetric causal dependencies, for instance, don’t compose in the way that meanings do, since our causal relation to “brown cow” is not a simple combination of our causal history with “brown” and our causal history with “cow” (see Margolis & Laurence b, p. ). It is only by presupposing methodological monism that would lead one to think that adopting an atomistic foundational semantics like Fodor’s would be required to satisfy the compositionality constraint. This is usually explained in terms of the Davidsonian’s commitment to the ‘Principle of Charity’. For a more complete discussion of this (and explanation of why this account of Charity sometimes can be attributed more comfortably to the “Davidsonian” than it can to Davidson himself ), see Jackman , 00 a. Henry Jackman Descriptive Atomism and Foundational Holism which will then be individuated in the same way. The extension we settle on for “cow” will thus affect the extension of a term like “horse”. In much the same way, the extension of “brown” will be partially determined by the extension assigned to, say, “ochre” if the speaker is committed to the belief that nothing is both brown and ochre. The Davidsonian thus gives a holistic foundational semantics for how “cow” comes to be satisfied by members of the set of cows and how “brown” comes to be satisfied by the set of brown things. Consequently, it looks as if F&L’s claim that “the plausible candidates for the semantically relevant symbol to world relations all look to be atomistic” may rest on conflating descriptive and foundational semantics. The question of what the semantically relevant symbol world relations are is clearly one for one’s foundational semantics, but all of the reasons F&L give for preferring atomic theories to holistic ones apply at the level of one’s descriptive semantics. Indeed, holistic accounts of the mind/world relation, such as Davidson’s, have a number of advantages over rival atomistic foundational semantic theories such as the sort of ‘asymmetric dependence’ account that Fodor prefers. According to such asymmetric dependence theories, “cow” means cow because: ( ) The concept COW stands in a lawful relation, L, to the property of being a cow, ( ) Other lawful relations involving COW, L 1 -L n , are asymmetrically dependent on the lawful relation between COW and cow. That is, L 1 -L n wouldn’t hold unless L did, and not the other way around. I will not rehearse in any detail all of the problems commonly perceived to afflict this sort of theory, but will merely suggest that many of them can be understood as arising from its atomistic presuppositions rather than its particular details. For instance, such theories seem ill-suited for dealing with parts of the language like “big”, “of”, “quickly”, or “unless” all of which seem to relate to other concepts rather than being intelligible on their own. The suggestion that one could have say, these three concepts without any others seems much F&L , p. . See Fodor , 0, . For a version of this formulation see, Margolis & Lawrence b, p. 0. For a discussion for how such cases cause problems for asymmetric dependence theories, see Margolis & Laurence p. . less credible than the idea that there could be a creature whose only concepts were “red” “cold” and “sweet”. The atomist might seem to require a different foundational semantics for these others parts of speech, while the holist can allow that, in spite of their occasionally having different sorts of semantic values, the foundational story about how every word acquires its value is the same throughout the language. Further, atomism at the foundational level commits one to a story about what concepts supervene upon that pushes one towards endorsing either a strong form of innateness about concepts, an unusually robust type of metaphysical realism, or both. Similar problems arise with other atomistic foundational theories, such as the crudest version of the causal theory of reference that emerged in the wake of Kripke’s work and quickly ran afoul of the problem that the objects we are causally related to can be grouped in all sorts of ways. 0 Holistic foundational semantic theories, then, are much less likely to leave one with unintuitive psychological and metaphysical commitments than their atomistic rivals, and while they were typically viewed as carrying some serious baggage of their own, most of this has been shown to come only from being a holist at the descriptive level as well.