Chapters 1 and 6 of Fodor and Lepore

Presented by Baljinder Sahdra

Nov. 22, 2001

Chapter 1

Introduction: A Geography of the Issues

The goal of the book

To explore whether semantic properties are holistic. The standard argument for meaning holism requires the premise that semantic properties are typically anatomic. So, another way of looking at the goal is that Fodor and Lepore wish to explore (not refute or defend!) the arguments for the view that things about natural languages and minds, like the following are anatomic properties:

'       being a symbol

'       being a symbol belonging to language L

'       having an intentional object

'       having intentional content

'       expressing a proposition

'       having a referent

'       being semantically evaluable

In other words, the issues under consideration are:

Is the connection between being a symbol and belonging to a language an internal one?  Do symbols have their being only as parts of whole language systems?

Table 1: Types of Properties

Anatomism and the Theory of Language

Issues about semantic anatomism may connect with some questions in philosophy of language.  For e.g.:

'       Could there be a private language?

'       If answer yes: Denying that the property of having a language is anatomic

'       This separates the private and language parts of private language

Two caveats:

1.     Questions about meaning holism can be rephrased into questions about whether some semantic property is anatomic, only if the semantic property is generic.

'       Generic semantic properties are those whose specifications can be taken to involve variables ranging over propositions, contents, meanings, and the like.  E.g.: the properties of

'       expressing some proposition or other

'       having some referent or other

'       having some content or other

'       It is because of these properties that there is arguably an inference from semantic anatomism to semantic holism.

'       Two doctrines are mainly under consideration:

1.     Content holism: properties like having content are holistic in that no expression in language can have them unless many other (nonsynonymous) expressions in that language have them too.

'       Thus, there can be no punctate languages.

2.     Translational holism: properties like meaning the same as some formula or another of L are holistic in the sense that nothing can translate a formula of L unless it belongs to a language containing many (nonsynonymous) formulas that translate formulas of L. 

'       This doctrine, isolated from content holism, leaves open the metaphysical possibility of punctate languages and minds.

'       Meaning holism or semantic holism is true if either content holism or translational holism is true.  It is a broader, less precise doctrine that meaning is somehow holistic.

'       Semi-holism: Asserts translational holism and allows the possibility of punctate languages and minds.  Some meaning holists may entertain the metaphysical possibility of punctate languages and minds, but deny that English (or any other natural language) is punctate.  For in English (as in any nonpunctate language), the meanings of sentences are constituted by their relations to one another. 

'       Note: Fodor and Lepore run the various kinds of meaning holism together because they all tend to "stand or fall" on much the same considerations.

'       Almost all the arguments of meaning holism in literature are arguments for content holism.

'       The only argument for translational holism in literature: The one that assumes that meanings supervene on intersentential relations, that they are something like inferential roles.  (More on this in chapter 6).

2.     When philosophers think about meaning holism questions, they may be thinking about things other than anatomism.  There is another variety of holism, namely, anthropological holism, but F & L won't consider it in detail.

'       Anthropological holism: (Wittgenstein, Austin and others.)  There is an internal connection between being a symbol and playing a role in a system of nonlinguistic conventions, practices, rituals, and performances - Forms of life.  Language ontologically depends on Forms of life.

'       Why are F & L putting it to the side?  Anthropological holism is distinct from semantic holism only insofar as it concerns the relation between language and its intentional background.  However, when you apply it to the background itself, anthropological holism reduces to the doctrine that intentional states, institutions, practices, etc. are ontologically dependent on one another.  In short, it reduces to the claim that the things in intentional background are anatomic.  So, anthropological holism just becomes semantic holism.

'       But some people could accept that there could be arbitrary punctate/atomic Forms of life.  That is, you could say that for anything linguistic to have content, there must be some atomic nonlinguistic thing that has content.  That is ok (at least for now!) for F & L.

'       Two great traditions in philosophy of language:

'       It is widely (and implicitly) argued that if a semantic property is anatomic, it is also holistic.  Anatomism about semantic properties would have the same consequences as meaning holism itself has.

'       Meaning holism jeopardizes the standard picture of learning, and it cannot account for how language could ever be learned; and hence, it must be denied (Dummett).

'       The standard picture of how we learn languages and theories:

'       Incremental learning:  We learn about the semantic properties of theories and sentences by leaning the semantics of their constituent sentences in something like the way we learn the semantics of the sentence by leaning the meaning of its constituent terms.  (Note that this presupposes that the constituents are meaningful.)

'       Partial overlap: The linguistic and theoretical commitments of speaker and hearer can overlap partially to any degree.

'       The standard picture reconciles the idea that languages have interpersonal, social existence with the patent truth that no two speakers of the same language ever speak exactly the same dialect of that language.

'       This picture makes sense only to the extent that partial consensus in usage does not require perfect consensus in usage of a language.

'       However, if holism is true, the only way I can understand your language is by understanding all of it; for if any one 'element' in your language occurs in mine, then all 'elements' in your language must occur in mine.  So, if holism is true, the standard picture of language learning must be false.

'       But how could someone understand all of a language while learning it?

'       In conclusion, semantic holism is must be denied.

'       A possible rejoinder of holists: We ought to revise our standard understanding of how languages and theories are learned and communicated! (Quine, Dennett, Stich, the Churchlands).

Broader Implications

1.     If reference holism is true, Scientific Realism goes down the drain.

2.     If meaning holism is true, we can't have intentional explanations.

3.     But on the bright side, the intentional is autonomous!

Why would Scientific Realism go down the drain?

'       Referential holism in astronomy, for e.g.: An expression has an anatomic, and hence holistic property R* iff it refers to something or other that currently accepted astronomical theories refer to.  No theory could refer to stars, for e.g., unless it also refers to planets, nebulas, black holes, etc.

'       Referential holism implies that scientific theories are empirically incommensurable unless their ontological commitments are more or less identical. But the ontological commitments of the Greek astronomy, for e.g., were not the same as those of the present day astronomy. So, the Greek astronomy and current astronomy are incommensurable.  The Greeks couldn’t even have referred to stars!  Current astronomy does not contest Greek astronomy!

'       But as per the standard argument for scientific realism, science is progressive.  By embracing present theories of astronomy, we can make more and better predictions about stars than could Greeks.  If this is only "trivially true” in that Greeks didn’t make any predictions about anything that our astronomy talks about, then the standard argument for scientific realism becomes bogus.

'       Holism about ontological commitments, as it is often argued, leads to anti-Realism (because of incommensurability).

Why can’t there be no intentional science of human nature?

'       A belief has a property T* iff it expresses a proposition that is the content of some belief of mine.  If T* is anatomic, it is holistic; and if so, it might be that nobody has thoughts that are tokens of the same type as my thought about one thing, unless that person also has thoughts that are tokens of the same type as my thoughts about a lot of things.

'       Cognitive scientists hold that higher animals act out of their beliefs and desires.  There are counterfactual-supporting generalizations which subsume mental states in virtue of their intentional contents.  It is in virtue of what they are thoughts about that thoughts fall under any generalization.

'       If holism about thought content is true, however, it may be that none of your thoughts has the property of bearing T* to any of mine (since you and I have widely different belief systems).  Not more than one person ever has thoughts about red, for e.g.; and even that could be just in one time slice (since our beliefs change from time to time).

'       Thus, if T* is holistic, there are no robust, counterfactual-supporting intentional generalizations, none that is ever satisfied by more than an individual at an instant.

'       Common argument of this sort:

'       Mental properties are holistic

'       There couldn’t really be intentional laws

'       Intentional explanations can’t be fully factual

'       Thus, there can’t be an intentional science of human nature

'       (There can’t even be a scientific theory of rationality.)

Why is there the autonomy of the intentional?

'       Looking at the bright side, the general structure of intentional explanation is autonomous or “not negotiable”.  If intentional theories are holistic, in virtue of their being holistic, they are “ipso facto” not in competition with theories of empirical sciences.  So, the following sorts of arguments can’t be true:

'       Common-sense belief psychology is “just another empirical theory”

'       On purely empirical grounds, it could turn out to be false

'       It is turning out to be false, since it proves not to be capable of integration with the rest of our developing scientific world view.

Content Identity and Content Similarity

Why aren’t people outside philosophy worried about the issues of holism?

1.     May not have noticed the undesirable consequences of it

2.     May doubt that such consequences follow

3.     It is widely supposed that even if holism precludes a robust notion of content identity, it still allows a robust notion of content similarity.  So, we may still have the notion of intentional law.

4.     See next section.

'       As per content similarity notion (borrowed from the colloquial notion of belief similarity), we may generalize by saying something like, “If somebody asks you something sort of like what is the first color you think of, then you will think of something sort of like red.”

'       The suggestion that appealing to content similarity may mitigate the severer consequences of semantic holism is simply empty.  Why?  Because we have no idea what it would be like for the ‘sort-of-like’ type generalizations to be true or false.  How can you construe belief similarity between my views and the President’s views?

'       The President believes P, Q, R, and S; and I believe P, Q, and R?

'       The President believe P and Q very strongly, and I believe equally strongly or almost as strongly?

'       But this is not believing something-similar-to-P (or more-or-less-P); only believing-most-of-P, -Q, -R, -and -S or more-or-less-strongly-believing-P.

'       By using the colloquial sense of “similar belief”, you are assuming some way of counting beliefs, and hence, some notion of belief identity.  Thus, so can’t dispense with the notion of belief identity in favor of a notion of belief similarity.  You can’t save intentional law this way, if you accept that holism is true.

'       We might coherently imagine a different notion of “similar belief” that may be both compatible with meaning holism and with there being robust intentional generalizations.  But nobody has any idea what that might be.

'       If holism is true, then the notion of “tokens of the same belief type” is defined only for the cases in which every belief is shared.  For any other cases, there is no clear notion of belief-type identity.  If we don’t even know about belief-type identity in these other ‘imagined’ cases, we have no hope of knowing belief-type similarity consistent with meaning holism.  Basically, “similar belief” argument won’t work.

'       Can the notion of physical similarity (as opposed to content similarity) save the day?

'       Nope!  Even if you grant physicalism, i.e., grant identity of belief systems supervene on physical identity, e.g., identity of molecules in the brain, it does not follow that similarity of belief systems supervene on physical similarity.  There are indefinitely many ways in which brain states could be similar, and indefinitely many ways in which they could be dissimilar.  How do you decide which similarities and differences would determine whether the beliefs are similar?  Explicating the notion of physical similarity is just as hard as explicating content similarity.

'       Can the notion of similarity-of-belief based on similarity-of-inference save the day?

'       Nope again!  Two reasons:

'       You are presupposing identity of inference, and that is not allowed by holism.  How would you explicate similar inference?

'       Some inferences would have to count for more than others.  For e.g., making inferences about tomatoes based on my thought of red is not constitutive of my having thoughts about red.  How would we know how much the difference between the red-inferences I endorse and the one that you endorse count as differences in our concept of red (since there is a huge number of red things and things about red that I know that you don’t and vice versa)?

'       In short, similarity-arguments suck because they presuppose identity!

Meaning Holism and the Analytic/Synthetic Distinction

'       Still trying to save intentional law, theory commensurability, Scientific Realism, translation, the like!

'       Embrace a notion of belief identity that is grounded in analytic/synthetic distinction?

'       Nope again! If you draw the distinction of a/s, you can’t argue from anatomism to holism.

'       Argument A:

'       Premise 1: Generic semantic properties are anatomic

'       Comments: Noncommittal about how many generic semantic properties; & could derive this premise from “inferential role” semantics, as many philosophers do.

'       Instantiation: If Smith has the belief that P, he must have other beliefs not identical to P.

'       Premise 2: There is no principled distinction between the propositions that Smith has to believe to believe that P and the propositions that Smith doesn’t have to believe to believe that P.

'       Comments:

'       Essentially, you are rejecting a/s distinction.  If you say that the principle distinction depends on the idea that if you can’t believe P unless you believe Q, then “if P, then Q” must be analytic; but there is no a/s distinction in principle.

'       “No a/s distinction” is not the same as “no analytic sentences” (unlike Quine); it is the same as “the a/s distinction isn’t principled”.

'       Conclusion: The property of being-some-or-other-belief-of-Smith’s is holistic.

'       Comments:

'       The reference to Smith is inessential.  If the argument is right, it shows that there can’t be a punctate mind or language.

'       It is a “slippery slope” argument:

'       Some a’s are F.

'       There is no principled difference b/w that a’s that are F’s and the a’s that aren’t.

'       Therefore, all a’s are F.

It is difficult to find fully explicit instances of this argument (except for Stitch's).

'       The status of argument A:

'       Not sound because:

'       Doubtful that semantic properties are anatomic, if we deny the underlying assumption of some functional role (or verificationistic) account of meaning.

'       Even if a/s is untenable, there might to some other principled way of distinguishing the propositions that you have to believe to believe P from the ones you don’t.

'       Since it is a slippery slope argument, it can easily lead from true premises to false conclusion.

'       Not valid (on the “short scope” reading of premise 1):

'       Assume that there are indefinitely many sufficient-but-not-necessary conditions for believing P; and nobody is able to believe the proposition formed by disjoining these indefinitely many sets of propositions (because it is too complicated).

'       Premise 1 and 2 would be true.

'       Neither content holism nor translation holism would follow.

'       Content holism requires that there must be many propositions that I believe if I believe that P.

'       Translation holism requires that for two people to share any belief, they must share at least one other belief.

'       What everybody wants is that meaning should be anatomic and that translation holism should nevertheless be false.

'       The “short scope” or “weak” reading: You can’t believe P unless there are other propositions that you believe.

'       You could defend anatomism on this reading, but

'       The kind of anatomism that you get if you take this reading is too weak to be worth the effort of defending.

'       It is important to argue that semantic properties are anatomic to account for the pre-theoretic intuition that two people couldn’t agree about only one thing.

'       But weak anatomism is not any better off at this than atomism.

'       The “long scope” or “strong” reading of premise 1: There are other propositions such that you can’t believe P unless you believe them.

'       This is the only kind of anatomism worth having.

'       If there is a principles a/s distinction, then the inference from anatomism to holism is blocked because premise 2 wouldn’t be ok.  Saying that premise 2 is ok requires denying a/s principled distinction; and this rules out any possibility of a “molecularist” compromise b/w atomism and holism. 

'       A molecularist says that there are other beliefs that we must also share if we are to share the belief that P, but he denies that all our other beliefs have to be shared.  But distinguishing between those that do and those that don’t requires a/s because believing P requires accepting the analytic inferences in which P figures.

'       F & L say that in what follows, they would undermine A and arguments like it.

'       Not by claiming that premise 2 is false (unlike Devitt).

'       Not by claiming that what is wrong with it is that is a slippery slope.

'       By attacking the alleged support for premise 1.  That is, by showing that there is no good reason as to why generic semantic properties are strongly anatomic; and thus, no good arguments for semantic holism, it being the stronger thesis.

'       If you have independent grounds to believe that atomistic alternative to holism doesn’t work, then arguments like A may produce a “rational conviction” in you that holism is true.

Semantic Atomism

'       Why is almost everyone a meaning holist?

'       Because there are positive arguments like A that meaning holism is true.

'       Because meaning holism is the “last log afloat” in semantics: either semantic properties are holistic or there are no such properties. 

'       If semantic properties are punctate, then, by definition, the meaning of an expression can not depend on its role in a language.  The nonholist alternative is that it depends on some punctate symbol/world relation (this is semantic atomism). 

'       Consider two historical reconstructions of the mind/world relation on which content is supposed to depend:

'       The mentalistic empiricism:

'       The semantic properties inhere in a certain class of mental particulars, “Ideas”.  These are species of images; what they mean depends on what they resemble.

'       Where is the atomistic part?  What one of one’s Ideas looks like is presumably independent of what other Ideas one has.

'       It failed!  Meaning can’t be reduced to resemblance.

'       The behaviouristic empiricism:

'       Meaning inheres in certain behavioral gestures.

'       The atomistic part?  Whether one’s behavioral repertoire includes a sound the utterance of which is reliably conditioned to dogs is, presumably, independent of what, if anything, the other sounds in your repertoire are reliably conditioned to.

'       It failed too!  Meaning can’t be reduced to behavior conditioning.

'       These two versions of meaning atomism are hopeless; not the same as admitting that meaning atomism is false; and not the same as admitting that meaning holism is true.

'       The current situation in philosophy of language:

'       Semantic properties may be anatomic, but holism may not follow from this because a/s proves to be tenable.  End up with semantic molecularism.

'       Semantic properties may be anatomic, but holism may not follow from this because, although a/s is untenable, there is some other principled way of grounding the distinction b/w the inferential relations that are constitutive of content and the ones that aren’t.  Again, end up with semantic molecularism.

'       Holism follows from the assumption that semantic properties are anatomic, but semantic properties are not in fact anatomic.  End up with meaning atomism.

The plan for the rest of the book:

'       Look at arguments for meaning holism that

'       reject an a/s distinction

'       do not assume that meaning atomism has been shown to be false

'       These arguments are not convincing.

'       The possible conclusions that we could pick from (F & L are noncommittal):

'       There aren’t any semantic properties

'       Some kind of meaning atomism is true but no one know which kind

'       There are good arguments for meaning holism, but nobody has been able to find one yet.

Chapter 6

Ned Block: Meaning Holism and Conceptual Role Semantics

'       F & L are exploring a specific argument that proposes a theory about what content is and then infers that content must be holistic.  It is an explicitly metaphysical argument with an explicitly metaphysically conclusion.

'       Thesis of the argument: The meaning of an expression is its role in a language/theory.

'       So, expressions that belong to different languages/theories are different in meaning.

'       In other words, it invites “translation holism”.

'       Block is arguing for a class of semantic theories, and thinks that some version of CRT would meet the requirements for a semantics.  But he does not say which one.

'       Since there is no clear philosophical consensus about which version of CRT is most promising, there is no CRT theorist whose formulation can be taken as fully representative of the kind.

Block’s Defense of CRT

'       See pg. 165 for the list of 8 desiderata.

'       F & L will only look at 1, 4, and 5. (1 and 5 are pivotal; 4 can’t be met by any version of CRT that implies meaning holism).

'       They think that the rest of them are likely to be satisfied by any psychologically motivated semantic theory, they throw no particular light on the status of CRT.

'       #6 can be satisfied by any “Gricean”

'       #7 is either trivially satisfiable or it is the demand that semantics be naturalistic, in which case informational theory of content is just as good as CRT.

'       #8 is satisfied by any semantic theory that has the resources to draw a narrow contend/wide content distinction; again it does not choose b/w CRT and informational theory of content.

'       #s 2 & 3 again do not choose b/w CRT and other theories of content.

'       #3: problem for CRT to avoid making meaning so relative to representational systems that translational holism is entailed.

Meaning and Reference (Desideratum #1)

'       Frege and Putnam examples show that intension/extension relation is problematic; they show that “narrow” content does not determine reference; and Block advertises that CRT (narrow meanings are conceptual roles) can offer clear resolution of the meaning/reference problems.

'       Frege: There are cases where substitutivity fails for extensionally (i.e., referentially) equivalent expressions.

'       One diagnosis: Reference and meaning are independent in at least one direction: identity of reference does not guarantee identity of meaning.

'       Putnam: Difference of reference is compatible with identity of content, assuming that the content of mental states supervene on factors that are “in the speaker’s head” (e.g., on the speaker’s neurological structure).

'       In both Frege and Putnam cases, meaning obeys an “individualistic” principle of supervience; i.e., it depends on assuming a “narrow” notion of mental content that is “in the head”. 

'       Block argues that is the virtue of CRT that it provides a notion of intentional content that permits that the Twins must be subsumed by the same intentional psychological explanations, not just by the same neurological or biochemical explanations.  In short, CRT is offered as an account of narrow content.

'       F & L grant that semantics should provide for narrow content.

'       They think that their arguments apply to theories like Harman’s in which it is wide content that is analyzed by reference to conceptual role.

'       If you identify narrow content with conceptual role, what sort of account of the meaning/reference relation do you get?

'       CRT’s solution of the Frege problem:

'       As per CRT, meaning of an expression is its inferential role (or causal role).

'       Inferential roles of coextensive expression can differ

'       Thus, CRT can distinguish the meaning of “the morning star” from the meaning of “the evening star”.

'       But whether the inferential roles of “the morning star” and “the evening star” do differ depends on how inferential roles are themselves individuated.

'       If individuation of inferential roles is as coarse-grained as the individuation of extensions, then the roles of “the morning star” and “the evening star” are not different.

'       In short, to solve Frege’s problem by appealing to distinctions among inferential roles, you need an adequate principle of individuation for inferential roles.  But not available!  It seems that the problem of individuating meaning and individuating inferential roles are identical!

'       CRT’s solution to Putnam’s problem:

'       Whereas Frege's cases require a notion of content that is more fine-grained than extensional equivalence - we can't treat coreferential expressions synonymously - Putnam's Twin cases require a notion of content that is less fine-grained than extensional equivalence - (narrowly) synonymous expressions must somehow be treated as extensionally distinct. 

'       Block wants to simultaneously deal with both of these cases that pull in opposite directions.

'       His solution is to adopt a "two factor" version of CRT.

'       Two factor CRT: has two orthogonal semantic dimensions:

'       CRT proper: provides the "aspect" of meaning that copes with Frege's problem.

'       Some independent (perhaps, causal) theory of reference: provides the aspect of meaning that copes with Putnam's problem.

'       F & L's critique:

'       Mild criticism: Neither factor is actually available.

'       More serious criticism: What keeps the two factors stuck together?  How do you align the two factors?  What precludes radical mismatches b/w intension and extension?  E.g., what precludes examples like, an expression that has an inferential role appropriate to the content 4 is a prime number but the truth conditions appropriate to the content water is wet?

'       This is just meaning/reference problem at the metatheory level: What is the connection b/w a theory that determines the meaning of an expression and a theory that determines its reference?

'       Block's solution: the CRT factor is "primary" in that it determines the nature of referential factor, but not vice versa.

'       F & L's objection:

'       Block is appealing to Kripke's ideas of causal relations determining reference.  The conceptual role of names (kind terms, etc.) is constituted by the fact that their causal relations determine their reference. 

'       But this is no help because we still can't preclude mismatches b/w intension and extension.

'       Notice use/mention fallacy: How is the sense of an expression related to its denotation, given the assumption that intensions don't determine extensions?  Block gives us a sort of description theory for terms like names, but no explanation of how determining the sense of a name is related to determining its extension.

'       Another problem for Block: Even if we grant that Block can account for Frege's problem, it will only adjudicate issues of identity of meaning; it won't answer questions about what an expression means.  That is, we don't know how narrow contents are to be expressed.

'       The answer can't be that what one's thoughts that water is wet has in common with one's Twin's thought that water is wet is that both express the narrow proposition that water is wet.  The problem with appealing to narrow proposition is that we need truth-conditions of the proposition.

Content and Behaviour (Desideratum #5)

'       D5 does not argue persuasively for CRT, because:

'       CRT is good at explaining the connection b/w the contents of mental states and their causal consequences; in fact, such connections are essential features of the propositional attitudes, as per CRT. 

'       CRT is a "functional" theory of content.

'       Note: we don't need an essential connection b/w content and behavior, only a reliable one.

'       Block's point is that if there are psychological laws that connect behavior to mental states with regard to the intentional content of the states, then we can state generalizations over wide content but not over narrow content.

'       This entirely misses the point that there could be psychological laws that are stated with respect to intentional contents; and if so, they underwrite psychological explanations, not the noncontingent mind/behavior connections.

What is the upshot of F & L's criticisms of D1 and D5?

Arguments of holism assume CRT and assume that there is no a/s distinction (recall argument A in chapter 1).  F & L are trying to undermine reasons for accepting CRT.

The Compositionality Constraint

In this section, F & L try to show that D4 can be satisfied only by forms of CRT that are either incompatible with the a/s distinction or patently preposterous on ground independent of holism.  So, they will show that (a) CRT is neither well motivated, (b) nor any version of CRT can ground an inference to holism.

Compositionality (D4)

'       The notions of compositionality, a/s distinction, and holism are all tied in this section.

'       F & L's argument goes like this:

'       You can't have semantics without compositionality.

'       Compositionality: The meaning of a sentence is in some sense a function of the meaning of the words in it (+ syntax).

'       Block takes compositionality of thought and language as self-evident.  His account of word meaning assumes it.

'       Compositionality is important because there are properties of thought, such as productivity and systematicity, that strongly suggest compositionality.

'       Productivity: Every natural language can express, and every normal mind can entertain, an open-ended/infinite set of propositions.

'       Systematicity: Any language/mind that can express/entertain the proposition P will also be able to express/entertain many propositions that are conceptually close to certain inferential potentials based on P.

'       You can't have compositionality without analyticity.

'       If meaning is inferential role, it is compositional for all and only the analytic inferences.

'       E.g., the inference 'brown cow à brown animal' is compositional; it is warranted by the inferential roles, hence, by meanings of 'brown' and 'cows'.  According to CRT, inferential roles of 'brown' and 'cow' are their meanings.  But this is just what it is means to be analytic, that is, to be warranted by the meanings of its constituent expressions.  In other words, 'brown cow à brown animal' is analytical because it is compositional.  (You can work this argument the other way around it, i.e., compositionality entails analyticity and vice versa.)

'       This is a revised version of CRT: it identifies meaning with role in analytic inference.

'       Thus, you can't have CRT without analyticity.

'       Notice a circularity of compositionality, analyticity, and meaning in the previous point.  If we try to reconcile CRT with compositionality of meaning by saying that meaning of an expression is its role in analytic inference, we are begging the question since the validity of analytic inference is guaranteed by the meanings of their constituent expressions.

'       Notice also that we can't naturalize inferential roles anymore in the revised CRT.  It is may be attractive to identify meaning with inferential role, because we can in turn identify it with casual role.  The fundamental assumption of computation theories of mental processes is that causal relations reconstruct inferential relations; and hence, there is a hope to unify semantics and psychology.  But we need a causal theory of analyticity, because we are identifying meaning with, not just any old inferential role, but with the role of an expression in analytic inference.

'       The conclusion that you can't have CRT without analyticity is a blow to the main assumption of the argument from CRT to holism, according to which there is no a/s distinction. 

'       If there is no a/s distinction, then CRT fails to satisfy the desideratum of compositionality that a semantic theory must meet.  You can't hold both CRT and holism, if you want to preserve compositionality.  You can't derive holism from CRT.

'       The holism arguments are unsound because:

'       If Quine is right about a/s distinction, CRT is incompatible with compositionality and hence false.

'       If Quine is wrong, CRT is compatible with compositionality, but there is no sound argument from CRT to holism.  Only the version of CRT that is nonholistic (given that it also embraces an a/s distinction) can be compatible with compositionality. 

Note: CRT and holism aren't necessarily incompatible.  Holism could be true, even if the main argument for it is unsound.  You could reconcile CRT with holism, but only if you were silly enough to embrace that all inferences are analytic.  You won't get semantics this way, because an acceptable semantics must be able to make sense of contingent inferences.

What might Block think?

'       Block should not have any objection to F & L's ideas about the internal connection b/w analyticity and compositionality.  He does accept a/s distinction so he can avoid holism!

'       Block would face a problem, given Quine' arguments to do away with a/s distinction.  The problem for CRT is the lack of a principled criterion for individuating inferential roles.

'       Solutions that many philosophers hope for: Quine does not preclude a denatured - graded or contextually relativized - sort of a/s distinction.

'       F & L's objection:

'       If analyticity is graded, the basic semantic notion is therefore similarity and not identity of meaning.

'       Inferences, analyticity, and compositionality are the same things.

'       So, your semantics reconstructs meaning as analytic inference and you have graded analyticity.

'       Therefore, you would have to live with the implausible notion of graded compositionality.  You would be wading in "deep waters"!

Conclusion

The holism arguments are unsound because:

'       If Quine is right about a/s distinction, CRT is incompatible with compositionality and hence false.  Since inferential roles aren't compositional and meanings are, meanings can't be inferential roles.

'       This is a blow for the received position in cognitive science that we can combine CRT with the denial of a/s distinction.

'       E.g., consider the view that concepts are stereotypes - bundles of probable or typical traits.  A/s distinction is denied because concepts aren't taken to be definitional.  Here is F & L's argument against meanings being stereotypes:

'       If meaning is compositional (which it is according to F & L), then either meaning isn't inferential role or it is role in analytic inference. 

'       If meanings of concepts are their roles in analytic inferences, meaning are not stereotypes. 

'       If meanings of concepts aren't their roles in inferences at all, meanings are still not stereotypes.

'       Meanings are not stereotypes.

'       If Quine is wrong, CRT is compatible with compositionality, but there is no sound argument from CRT to holism.  Only the version of CRT that is nonholistic (given that it also embraces an a/s distinction) can be compatible with compositionality. 

'       Is Quine right or wrong?  Take whichever position, either CRT is false or the argument from CRT to holism does not run.