Notes by RBJ on

Two Dogmas of Empiricism

by Willard Van Orman Quine


The two dogmas are (1) the analytic/synthetic distinction (2) reductionism (to sense data). Quine claims that both are ill-founded.
1. Background for Analyticity
Mainly leading to the reduction of analyticity to synonymy. Also criticises Carnap's account of analyticity via state-descriptions.
2. Definition
Concerned with disposing of the idea that definitions provide a basis for the synonyms required for establishing analyticity.
3. Interchangeability
Discussion of the use of interchangeability salve-veritate to explain cognitive synonymy.
4. Semantical Rules
Mainly concerned with arguing that the problem with "analyticity" is not rooted in the vagueness of ordinary language but is as severe for artificial languages with semantic rules. Primarily addressing Carnap's approach to defining analyticity.
5. The Verification Theory and Reductionism
6. Empiricism Without the Dogmas



The two dogmas are:
  1. That there is some fundamental cleavage between truths which are analytic or grounded in meaning independently of matters of fact, and truths which are synthetic or grounded in fact.
  2. That each meaningful statement is equivalent to some logical construct upon terms which refer to immediate experience.
Quine mentions two effects of abandoning the dogmas:
  1. a blurring of the supposed boundary between speculative metaphysics and natural science
  2. a shift towards pragmatism

Background for Analyticity

Quine begins by mentioning distinctions drawn by Leibniz and Hume as foreshadowing the analytic/synthetic distinction, which he attributes to Kant. After mentioning Kant's two definitions of the distinction, he goes on to infer from Kant's subsequent usage that his intent can be restated: "a statement is analytic if it is true by virtue of meanings and independent of fact". Quine therefore proceeds to a discussion of meaning.

Meaning is to be distinguished from denotation or extension. A relationship between essence and meaning is mooted. The question "what sort of things are meanings?" is raised.

Here Quine, because of doubts about what meanings can be, proposed to abandon meanings "recognising as the primary business of the theory of meaning simply the synonymy of linguistic forms and the analyticity of statements".

This is in fact incorrect. A complete account of synonymy and analyticity in a language would still fail to give an adequate account of the meaning of synthetic sentences. Of course this might still be given by intralinguistic synonymy claims if all the concepts in the language were available in some other language already well defined, but there is no reason to suppose that they will be. It is remarkable furthermore that Quine at this point appears to make semantics parasitic on the notion of analyticity, thereby (if he were successful) guaranteeing that scepticism about analyticity would translate into scepticism about meaning in general (and hence about truth).

Quine now divides analytic truths into two groups.

  1. Those which he calls "logically true" and characterises as remaining true under all reinterpretations of components other than logical particles
  2. Those which can be turned into logical truths by "putting synonyms for synonyms".
Quine believes himself to have reduced the problem of defining analyticity to that of defining synonymy. In this he is mistaken for it is not in general the case that analyticity can be defined in terms of synonymy.

One special case in which it does hold is in a language in which all analytic sentences are "synonymous". It is doubtful that English falls under this case, but many formal languages are given a purely extensional semantics in which this holds. In those cases where this does hold, it is not a useful reduction, for the claim that a statement A is analytic can immediately be translated into the claim that it is synonymous with "true" (or "A and A"), but this is not a helpful step.

More importantly it is not the case that in practice it is possible either in natural languages or in formal languages to establish the analyticity of sentences by substituting for synonymous constituents.

In the case of natural languages, for example, there may be example of conceptual inclusion or exclusion which are not underwritten by any relations of synonymy. For example being red may entail being coloured, but there is may be no available expression which can be substituted for "coloured" to show this. Similarly being red may be incompatible with being blue, but there is no way of showing this by substitution which does not beg the question. (e.g. substitute "red and not blue" for "red").

In the case of formal languages, it is not uncommon to define a formal language starting with the primitives of set theory allowing additional constants to be introduced by arbitrary conservative extensions (a conservative extension is a new axiom which leaves the interpretation of previously introduced constants unchanged). Such extensions underwrite the demonstration of a variety of analytic truths but do not operate in the same way as a traditional definition. e.g. one might say, "let alpha be a natural number greater than 45". "alpha > 0" is a trivial consequence and is analytic in the language thus defined, but its derivation does not take place by substituting for alpha a defining synonym.

Finally Quine mentions and criticises Carnap's explanation of analyticity in terms of "state-descriptions", referring to [Carnap47] pp. 9ff.

Quine in effect points out that Carnap's definition does not define analyticity but only logical truth, and in this he is correct.

This is an elementary mistake on the part of Carnap not a fundamental problem in this method. Up to §2.1 Carnap is OK for it is clear that he intends the semantics of the non-logical constants to be taken into account in the determination of analyticity. However at §2.2 he comes up with a final formulation which makes this impossible and hence defines analyticity as validity (in the case that the language is first order and a state description is an interpretation, with similar consequences likely in other contexts).

Carnap subsequently makes good this error in a paper entitled "Meaning postulates" which is included as Appendix B in the second edition of [Carnap47]. He does this in a way which clearly relates to his original intention as given in §2.1.


In this section Quine disposes of the idea that the synonyms needed to establish analyticity can be obtained from definitions.

He discusses three different kinds of definition:

  1. those where the intent is to give an accurate account of the meaning of some pre-existent concept
  2. those where the intent is to "explicate" a prior concept, by giving a closely related but more precisely defined concept
  3. those where a completely new concept is introduced by definition
In the first case the definition is merely an account of a known (or alleged) synonymy relation and provides no independent basis for such.

In the second case Quine argues that for the definition to succeed in explicating it must be related to prior usage, that this relation presupposes synonymy relations, and that this kind of definition cannot therefore provide a basis for synonymy claims.

Quine does not distinguish between whether such a definition succeeds in "explicating" and whether it succeeds in defining. Even if such a definition is a poor explication, it may nevertheless be a good definition. Analyticity of claims involving use of the concept as defined do not depend in any way on the prior usage of the concept but only on the definition.

In the third case Quine accedes that "we have a really transparent case of synonymy created by definition".

Having made this concession he then proceeds to discuss the role of definition in formal work.

Quine concludes "In formal and informal work alike, thus, we find that definition -- except in the extreme case of the explicitly conventional introduction of new notations -- hinges on prior relations of synonymy.", and therefore proposes to say no more of definitions.

Here Quine has managed to make analyticity in formal languages appear to be exceptional, marginal and unimportant. This is in my opinion highly misleading.

It is misleading as to the extent to which analyticity is well-defined in formal languages, for he incorrectly suggests that this will not be the case unless completely new concepts are being introduced. His grounds for doubting that explicative definitions provide a basis for judgements of analyticity are unsatisfactory. It is misleading also as to the importance of analyticity in formal languages and its ability to underpin the semantics of concepts in natural languages. It is a feature of mathematics in the twentieth century that even though the languages used by mathematicians are rarely completely formal, their semantic precision is founded in concepts, such as that of a set, which have been made precise through formal theories.

Quine accepts as an exception to his negative conclusions only "the extreme case of the explicitly conventional introduction of new notation". However, the use of formal languages is in general such an extreme. They are used precisely to discard previous language and to reconstruct in a more precise manner a language adequate for the purposes in hand. That this language will often contain concepts which correspond, completely or loosely, with previously known concepts is for Quine a ground for scepticism, but for the authors of such languages may be completely irrelevant. The question whether the language has a well-defined semantics and whether analyticity for this language is well-defined do not depend upon such matters. (it does depend upon the language in which the semantics is defined, but this is another matter)


The central argument of this section seeks to demonstrate that "cognitive synonymy" cannot be defined in terms of "interchangeability salve-veritate" without recourse to the notion of analyticity.

The most significant point made to support this claim is that "interchangeability salve-veritate" (once you get the fine detail of what that means right) is extensional, i.e. two concepts are thus interchangeable if the have the same extension. Except in the special case of extensional languages, this does not amount to "cognitive synonymy" and does not provide a basis for a judgement about analyticity.

I agree with Quine's conclusion here, if not with all the detail which leads to it.

Quine concludes by not only abandoning the attempt to explain synonymy via interchangeability, but also the attempt to explain analyticity via synonymy.

Semantical Rules

En passant, at the beginning of this section is Quine's complaint about the difficulty in deciding whether "anything green is extended" is analytic. This is he says down to the meaning of analytic, not to any difficulty with the meanings of green or extended. I believe that Grice has observed that this is a bad example because we have difficulty even in deciding whether it is true, and this can hardly be because of a difficulty with the concept of analyticity.

The main purpose of this section is to repudiate the suggestion that difficulties with analyticity are due to opacity in the semantics of natural languages and don't apply to precise artificial languages.

Quine takes Carnap to be the primary source on semantics of artificial languages, and it is Carnap's methods he targets. He considers two methods, which he attributes to Carnap, of defining analyticity for some articifial language. In the first a set of rules are used which stipulate which sentences are analytic. In the second a set of semantic rules which do not mention analyticity is used. The analytic sentences are then defined as those whose truth follows from the semantic rules.

Quine's objection to the first approach is that the rules cannot be understood because they contain the word analytic which has not been defined. His objection to the second is that the rules, thought they do not contain the word analytic, do appeal to the notion of "semantic rule" which is meaningless.

Quine seems to think that no definition can be understood if it mentions the concept being defined. But all definitions do this and it is not merely unexceptionable but indispensable.

Next Quine back's off and says that we can define "analytic-for-L" for specific L, but that we must give a general definition for variable "L", and can't do that.

However, this is in fact what Carnap does. His method involves defining the semantics of an arbitrary language using a set of L-rules. He then defines the notion of analyticity for all these languages in terms of the L-rules for the specific language.

Quine next proceeds to compare the idea of semantical rule with that of postulate. There is nothing distinctive about a postulate other than that it has been chosen to be a member of some set of postulates for some purpose, and the same is true of semantic rules. If analytic is defined as "true by semantic rules" then no statement is any more analytic than any other, since they are all entailed by some set of rules.

But clearly anyone putting forward this account of analyticity will in define analyticity as true by "the" rules which define the semantics of the language. That the set of analytic sentences changes when the semantics changes is wholly expected and unproblematic.

Quine now reveals that his notion of a language does not involve the semantics of the language (and hence that his account of analyticity as a relation between a sentence and a language was defective) and observes that one could fix the problem by regarding the semantics as part of the language. However, he alleges that this is equivalent to including the definition of analyticity in the language and appears to think this to absurd even to countenance.

This section concludes with possible the best second level explication of analyticity I know (after a first level explication as "true in virtue of meaning").

Quine evidently thinks this fails, but he has in fact offered no real criticism of this explication, his attacks (fallacious though they are) have been directed at more detailed accounts of how the semantics of a language can be formulated.

The Verification Theory and Reductionism

Now Quine considers the verification theory of meaning.

He observes that if the verification theory does provide an adequate account of meaning then it can be used to define analyticity.

In fact he claims that analyticity can be defined as "synonymous with a logically true statement", which is not in general the case. However, more importantly he seems here to clearly connect his denial of analyticity with scepticism about semantics.

Quine therefore considers the relation between a sentence and its method of confirmation (or refutation). This leads him first to talk about radical reductionism, ultimately concluding that this is untenable. He claims that the "dogma of reductionism" is implicit in the idea that statements in isolation can admit of confirmation at all.

This is asserted rather than argued, and seems to me false. Why can I not hold the latter and deny the former? In fact I do. Here is another point at which Quine's scepticism about semantics is clear. He is close to asserting here that sentences do not have truth conditions.

Empiricism without the Dogmas

Here Quine argues holistically that the entire body of our beliefs should be evaluated without recourse to the idea that individual statements have empirical content. He advocates that we make choices about truth values the purpose of which seems to be to make the logical connections between the peripheral and other statements, and that the statements of the logical laws are subject to this revision. He does not advocate that we give any attention to semantics, but he does advocate that we decide what is logically true.

One wonders what is to prevent us from deciding that the logical truths are the analytic sentences and from determining the analytic sentences by deciding on the semantics of our languages?

UP HOME © RBJ created 2001/8/27 modified 2010/3/24