Quine sets up the problem of analyticity as being a question of coming up with an adequate account of how one problematic class of statements that have traditionally been considered analytic can be reduced to another less problematic class of statements.
The first class is typified by the statement No bachelor is married, the latter class by No unmarried man is married. The initial stages of argumentation seem to be as follows:
With this as background, Quine moves into the next phase of argumentation. His overall goal is to show that we cannot give an account of analyticity. The form of his argument is disjunctive. He presents several possible accounts of analyticity. On the assumption that his disjunctive premise is exhaustive, he concludes that no account of analyticity is possible, and for us to continue to believe there are such things as analytic truths is to engage in blind faith.
His first few disjuncts are based upon the initial argumentation above. He argues that no account of synonymy is adequate. If that is true then no account of analyticity in terms of synonymy is going to be adequate. He then argues that no account of analyticity that is carried out independently of an account of synonymy will succeed either. I will now examine his arguments concerning synonymy. After that, I will look at the argument that is concerned with synonymy independent accounts of synonymy.
The terms in need of explication in (3) above are `same in meaning' and `synonymous'. Quine argues that we cannot give an adequate account of these terms. He argues disjunctively: he offers two possible ways to account for synonymy between terms: The first possibility is that synonymy comes about by one of three sorts of definition. The second possibility is that synony my between terms is simply the f act that they are interchangeable in all sentential contexts salva veritate. He argues that neither of these approaches will give an adequate account of what synonymy is. Assuming that the disjunctive premise is exhaustive, and he is correct about each of the disjuncts, he concludes that there can be no adequate account of synonymy. Since there can be no adequate account of synonymy, there can be no adequate account of how statements of type (b) are analytic if that account must depend on an account of synonymy.
In regard to the three types of definition, Quine argues that two types in some way presuppose synonymy without explaining what it is.
The third sort of definition does indeed create genuine synonymy relations, but Quine seems to think that these cases are in some way importantly different from most of the alleged cases of synonymy.
Therefore, this sort of definition cannot serve as the basis of a general account of synonymy.
In regard to the interchangeability thesis, Quine argues that it ultimately must involve some sort of presupposition of analyticity to work out. The interchangeability thesis fails because it ultimately has to make use of the notion of analyticity in order to make sense of a language that uses the modal operator `it is necessarily the case that'. That language is apparently the minimal necessary language in terms of which the interchangeability salva veritate account of sameness in meaning can be insulated from counterexamples based upon extensionally equivalent terms.
I start by presenting the initial steps of Quine's argument. I will include all the disjuncts he considers in the article. I will however first concentrate on the portion of his argument concerning definitions, interchangeability salva veritate, and synonymy. Later, I will look at his argumentation concerning a synonymy independent account of analyticity:
Even though he allows that conventional definitions can create synonymous pairs of terms, Quine still thinks that an adequate general account of synonymy has not been produced. It seems that he want an account that can be made use of in explaining synonymies that exist in natural languages. He does not think that this can be done by conventional definition alone. Perhaps it can be carried out via some sort of combination of conventional and explicative/ampliative definition. It remains to be seen if this is true. I will spell out a way this might be done after I have presented Quine's argument concerning the impossibility of a synonymy independent account of analyticity. However, for now I want to look at the argument concerning the inadequacy of interchangeability as an account of synonymy.
Once Quine dismisses definition as an adequate account of synonymy, he moves on to interchangeability salva veritate.
Interchangeability salva veritate by itself is not sufficient for sameness of meaning for two reasons: One can substitute extensionally equivalent terms and preserve truth, while one does not preserve meaning. So, if we have a language that deals only with one place properties, two or more place relations, and contains the truth functional operators, singular terms, variables for those singular terms, and quantifiers, we will not have enough to guarantee that if we are to replace one term with another, while preserving truth in some statement, that we will have also preserved the original meaning of that statement. It may be true to say that all creatures with hearts live on earth, true that all creatures with hearts are creatures with kidneys, and thereby true that all creatures with kidneys live on earth, but the first and latter sentences do not seem to mean the same thing. So it seems relative to this sort of language, interchangeability in not going to be sufficient for synonymy.
But perhaps we can enrich our language. We can add what Quine calls `intensional operators' to our language. These would be things like cognitive operators, modal operators, and the like. Quine points out the operator most likely to help us in g iving an account of analyticity: the modal operator `it is necessarily the case that'.
Necessary truths are those that are true `come what may'. Leibniz took this to mean that they are those statements that are true in every possible world. We can think of possible worlds as possible circumstances. So necessary truths are those that are true in all possible circumstances. this would seem to indicate that their truth is only an internal matter. The circumstances simply do not matter. So it would seem that necessarily true statements are true by dent of a feature havin g to do only with themselves. T hat feature cannot be the extensions of the necessarily true st atements, should they have them. Therefore, it must be their meanings that are the key feature.
So maybe if we work with a sentential matrix like `Necessarily, if x is an A, then x is a B', as a sort of test for synonymy we can find those synonymous pairs of type (b) after all. For only if the non-logical terms are exactly the same in meaning will they drop out and become irrelevant to the truth preservation of the statements we are using as test cases. If we compare the statement Necessarily, if x is a creature with a heart, then x is a creature with a kidney, with the statement Necessarily, if x is three in number, then x is odd, we see how plausible this sounds. This sort of approach also seems to work when we plug in bachelor, and unmarried.
However, Quine thinks that doing this ultimately lands us in an account of analyticity through interchangeability that involves us in some sort of circular account. His argument is difficult to follow, but it seems to be the following:
Quine now abandons the attempt to produce an account of synonymy, and attempts to give an account of analyticity in terms that are independent of synonymy. He makes use of the notion of semantic rules of languages. To simplify matters he considers artificial formal languages.
The general idea is that the analyticity of a statement is seen as relative to the semantic rules of the language of which it is a part. We might use an example: In pure propositional logic, the semantic value of a statement is one of two truth-values : `true' or `false'. There are simple statements, and compound statements.
Compound statements are created by concatenating simple statements using truth functions. (i.e., the connectives &, v, ~, etc.). Each truth function is assigned a truth table. The truth tables give semantic rules by which one can determine the truth-value or semantic value of any truth functionally compound statement that is created by use of the connectives. Analytic statements could be specified as those that describe applications of the semantic rules (the truth tables) which come out true on every possible concatenation of semantic values possible for the propositional variables in the statement. By that rule `p' in not an analytic statement of propositional logic, but `p∨~p' is.
Here is what Quine says about this:
Quine now complains that in such a definition, one unexplained term is being substituted for another. Assuming `truth' is not a problematic term, `analytic' has been replaced by `true by virtue of a semantical rule'. Using the example, we can give a semantic rule for propositional logic as follows: All statements that are of the form `p v ~p' are true. We might be inclined to say that this is a statement of a semantic rule of propositional logic. But Quine tells us that this statement is most generally described as a statement which says of a certain set of statement types that they are true. But not every statement that says of a certain set of statement types that they are true is happily described as a statement of a semantic rule. I might say, (and it might be the case that) all the predi ctions of the oracle of Delphi are true, but this would not seem to be a statement of a semantical rule of any language. Neither would my saying this sort of thing make it true that the oracular utterances are all analytically true. If all statements which say of a certain set of statement types that they are true were semantic rules, then all truths would be analytic. Clearly, it must be the case that some subset of these sorts of statements are statements of the semantic rules of languages and some are not. But what property sets off the favored subset from its brethren? Maybe that property is the fact that the favored statements point out truths that result only from the primitive postulated semantic rules (such as the truth table of disjunction in our example). The `analytic making property' would be something like `being a sentence form that receives a semantic value assignment true each time it is evaluated using the primitive semantic rules, and receives this assignment regardless of the truth value assignments of the atomic statements. Quine is not satisfied with this sort of move, and thinks that ultimately it cannot be used to give an account of analyticity. His reason seems to be the following: When we set up a formal language like propositional logic, we start by postulating some set of semantic rules as basic. Others can be defined in terms of the postulates. In formal languages, we are interested in statements in so far as they can be derived form other statements in accord with transformation rules. But it is open to us which semantic rules we treat as the basic set. E.g., we can treat rules involving negation and disjunction as the primitive semantic rules, and can then define other semantic rules involving, for example, conditionals in terms of the primitive rules we have chosen.
We might even go the other way, treating as basic the conditional, and negation, and define disjunction in terms of these. It is also possible that we could try and start with one semantic rule, perhaps the `Sheffer stroke' and attempt to define a large set of connectives in terms of it. Conversely, we could start with a rich language, containing many basic semantic rules, and define certain long formulae using the Sheffer stroke in terms of the many postulated semantic rules.
In the former kind of case, the non-basic semantic rules are derived. They can be described as abbreviatory conventions. The language could do without them. It would just be more cumbersome without them. In those sorts of languages, it seems that the derived rules are not really in an ultimate sense basic semantic rules of the language. They are not among the postulates. But since it is up to us which semantic rules will be basic, there is no sense in saying that there is in any sort of task independent sense, a privileged set of statements that follow from the basic semantic rules of propositional logic. Relative to task A, semantic rule x can be treated as being a postulate, but relative to another task B, x may be treated as being defined in terms of and being dependent upon other semantic rules that are being used as postulates. T here are many possible purely logical tasks that can be carried out using a formal language. So if `x is analytic' means `x is invariably true by the basic semantic rules of L', means `x's invariable truth is the result only of an application of a postulate of L', then any invariable truth of propositional logic could conceivably be the result merely of an application of some postulate of some L which is family related to propositional logic. It would thereby be an analytic truth by the basic semantic rules of that L.
So it seems that the following is the case: If the analyticity of a formula of some formal language L is to be defined in terms of whether or not it is (1) a statement which is true only because of the fact that it turns out true on every possible assignment of truth values to its propositional constituents, and (2) it uses only the primitive semantic rules of that language to determine if (1) is the case, and (3) what rules are treated as primitives, or postulates is something that is relative to the tasks that the creator of that system has in mind, then (4) it is true that analyticity, so defined is relative to languages. However, this does not seem to show that there are no analytic truths, as Quine seems to maintain.
But how can these considerations be applied to ordinary language? I think they can be applied in such a way as to throw doubt on Quine's strong position. Earlier, I mentioned in connection with the discussion of definition, that there might be a way of combining conventional definition, and explicative/ampliative definition so as to give some account of analyticity. Now I will try and sketch this out.
Because the relative richness of the concepts used in a language are in some way relative to tasks undertaken via that language, what may be a deductive consequence of the meaning of a word used to designate a concept of one language may not be a deductive consequence of the meaning of that same word as used in a richer language. But if there are deductive consequences of the meanings of words, then there are analytic truths. True, there cannot be analytic truths in some sort of absolute sense, but this does not seem to count against the thesis that there are at least some analytic truths. Quine claims that there are no analytic truths. To prove, as his semantic rules argument seems to, that analytic truths, if there are any, are in some way language dependent, and task dependent does not establish the stronger point. It still seems that relative to a given task, and a given way of conceptualizing a situation, that there will be some deducible consequences of that conceptualization. If the deduction of such consequences is not an analysis of the postulates, or conceptual underpinnings of that language, then what else could it be? This theory of language relative analytic truths may allow us to deal with the alleged counterexamples that are aimed at traditional examples of analytic statements. One such traditional example is the statement All bachelors are unmarried male humans of marriageable age.
Psychologists have found that people, if asked, will say that the Pope is not a bachelor. This is true despite the fact that he is an unmarried male human of marriageable age. People will also refuse to label a man who has lived in the same house as a woman with whom his is not wed with the term. Also, extremely old single men are not counted as bachelors (the Pope is once again a good example of this).
All of this is supposed to count against reading the universal generalization above as an analytic truth concerning the word `bachelor'.
It seems that there are two distinct ways to respond to this claim.
(1) `Bachelor' just means unmarried human male of marriageable age, and the counterexample shows that within that broad category there are subspecies. People may have in mind some rather typical examples of the species when they are asked to answer the questions concerning atypical examples. Because they have these typical examples in mind, more so than examples of the atypical types, they make these judgments. The empirical results do not show that there are no analytic truths concerning the term `bachelor', they just show that people can be led into error by psychological factors. It seems that empirical results could be produced that would corroborate this view. If the psychologists had asked their subjects to think carefully and tell them whether or not `strictly speaking' the Pope etc. were bachelors, it seems to me that they would have received affirmative answers.
(2) The second type of approach to these alleged counterexamples would be more in line with the way Quine looks at formal languages.
According to that view, the word `bachelor' can be seen as a symbol that is shared by various languages, each of which is a part of a motley collection called `English', or `natural language' or something of the sort. These languages are collected together by the fact that they are used by at least some people in our society at any one time.
Some languages are proper subsets of others, some languages share terms, or conceptual underpinnings, but are otherwise independent, and others might be completely independent of one another. Some terms are shared by various members of the motley crew, but vary in meaning either through variations in relative conceptual richness, or complete difference in meaning. Different languages or sublanguages can be roughly delineated by different tasks for which they were more or less consciously designed. So terms shared by distinct languages or sublanguages will vary in meaning according to the task or tasks for which the language or sublanguage exists. Being members of the overall society that makes use of this hodgepodge, we more or less pick up and use the members of the hodgepodge. Our problems with the term `bachelor' are reflective of this situation. It may be an analytic truth of language A that an unmarried male of marriageable age is a bachelor, and an analytic truth of A that an unmarried male of marriageable age who is shacking up with a woman is also a bachelor. Yet language A may be some sort of a sublanguage of a larger language, which also has language B as a part. Language B has some task different than that of A, and according to it, only unmarried males of marriageable age who are in some sense of the word eligible are bachelors. So according to this hypothetical language B the Pope does not make the cut, and neither does our shacking up guy. We might imagine language A to be used by the legal community, or by the IRS, and language B to be used by people more or less interested in who stands a realistic chance of getting hitched.
Within each language, there are certain things that are taken for granted. In the IRS language, the universe of discourse simply consists of unmarried adult males and females, and married adult males and females, and bachelor simply means unmarried adult male, because relative to the task of determining what tax rate an adult male gets, whether he is married or not is one of the relevant characteristics he may have. Whether he is eligible is irrelevant to the purposes of the IRS. In the matchmaker's language, the universe of discourse consists of unmarried viable males and females, unmarried unviable males and females, and married males and females, and bachelor means unmarried adult male who is viable husband material, because relative to the task of determining which males are possible hitchees, not only is being unmarried a relevant property, but viability (broadly construed) is a relevant property.
In general, Quine thinks that it is troubling for those that are committed to the existence of analytic truths, that all the attempted explications of what analyticity is somehow land the believer in a closed apparently circular definition of analyticity. It is explained in terms of synonymy, and interchangeability, and these themselves ultimately depend upon the notion of analyticity themselves. Now it is not clear exactly what we should take away from these states of affairs, even if we grant that they are true. Quine seems to allow that we can create, by conventional means, some analytic, or definitional terms. Yet, he thinks that aside from this, we cannot point to analytic truths of ordinary language. Yet, he also feels that logical laws, such as the law of excluded middle, are open to empirical falsification. He has in mind the particle wave duality of particle p hysics. So it seems that these to o are synthetic, or have some empirical element. But if we try to define analyticity in terms of any of the notions canvassed above, we will find ourselves explicating this family of terms by other terms in the family. Quine thinks this is a fatal flaw. It is fatal because it is circular.
However, there are other families of terms each member of which finds its meaning explained in terms of other members of the same set. Consider the terms `father', `mother' `child'. A child is the result of genetic contributions of a male and female human (a mother and a father), who account for the child's existence. A father is a male human, who along with a female human (that would be a mother) has contributed genetically to a third human (that would be a child) accounting for that human's existence. A mother is a female human who along with a male human (father) has contributed genetically to a third human (child again) accounting for that human's existence.
Does this relation between these terms throw us into grave doubts as to the viability of familial discourse, and the very possibility of making meaningful utterances about children, fathers and mothers? Does it lead us to think their is no distinction between things familial, and things non-familial? No. Even in logic, (as Quine points out in his essay), the truth functional terms are defined in terms of each other. This fact does not lead us to abandon logical discourse, or proclaim that there can be no satisfactory account of the logical connectives. It does not lead us to claim that there is no distinction between logical truths and truths of other types.
In general, if we can countenance such families of related terms, and can establish membership in such families, then there will be analytic truths. Those truths will explicate the conceptual structure, indeed the identity of those families. Similarly, if we can countenance the family of terms that Quine presents, and their conceptual interrelations, then there will be such properties as analyticity, synonymy, and necessity.
Philosophy is replete with such families of terms. The family that Quine explores is one. Another is {knowledge, truth, justification}. Another is {good, obligatory, permissible}.
In general, Quine has this problem: If we are to take interdefinability as a fatal flaw, and as an indicator that an area of discourse is either impossible, or ultimately meaningless in some way, it seems we will have to throw out not only philosophical discourse, but much discourse that has to do with matters of fact. But we (and presumably Quine) do not want to abandon the latter sort of discourse.
Why abandon the former? Perhaps there are some practical considerations. A reason that Quine has for adopting his `web of belief' view is that it is supposed to be a tonic against dogmatism.
Philosophical discourse might degenerate into dogmatism, and people will not keep their minds open, if they are convinced that there are truths that are immune to empirical falsification. If we were convinced web of belief theorists, this would be less likely to occur.
But the possible empirical falsification of logical laws does not lead Quine to aba ndon the practice of that discipline. Why then should he abandon, as impossible, the possibility of conceptual analysis in general, which is in effect exactly what he is doing? Perhaps his overriding concern is the spec ter of a recalcitrant dogmatism.
Concerning the worry over dogmatism, I think that way lies a two-edged sword, which can with equal justice be wielded against web of belief theory. If applied consistently, web of belief theory can and should land one into a firm acceptance of the alleged fact that all statements are in some way synthetic, and that there are no conceptual truths, and that even the laws of logic are (even if only slightly) empirical, and open to falsification. This would tend to degenerate into dogmatic relativism, and a quick dismissal of views of a more traditional nature. But, this would be to take up a position that the Quinean position is itself somehow independent of the web of belief, and privileged in that it is immune from empirical falsification. To take up this sort of position is just as dogmatic as is the position that claims that the Pope type examples do not show that there are no analytic truths, only inadequately grasped conceptual structures. So if there is no virtue in the one camp, then there is none in the other. It may be that there are analytic truths.
We should not dismiss that possibility