This Week’s Special: Quine’s “Two Dogmas of Empiricism”

By Daniel A. Kaufman

On tap this week is one of the most influential philosophy papers of the last century, Willard Van Orman Quine’s “Two Dogmas of Empiricism,” which initially appeared in the Philosophical Review in 1951 and was later reprinted in his book, From a Logical Point of View, first published in 1953.

The paper is widely believed to have dealt a severe, perhaps even fatal blow to Logical Empiricism, a 20th Century version of the classical Empiricism of the Enlightenment.  At the heart of Logical Empiricism is the idea that all the well-formed statements of a language are either trivially true or false – these are the so-called “analytic a priori” statements, like “No bachelor is unmarried” – or true or false by virtue of some empirically verifiable state of affairs – the so-called “synthetic a posteriori” statements, like “No bachelor lives in New Jersey.”

These ideas become Quine’s targets in “Two Dogmas of Empiricism.”  Specifically, he maintains that:

  1. We can give no clear, non-question-begging account of analytic statements.
  2. Individual statements do not enjoy their own, individual verification conditions.

Quine’s arguments regarding analyticity go something like this.

Given that the concept of a logical truth – “No unmarried man is married” – is well-defined, we might define analytic statements as those statements that can be turned into logical truths, by substituting synonymous expressions.  Thus, we can turn “No bachelor is unmarried” into “No unmarried man is unmarried,” by substituting the terms ‘bachelor’ and ‘unmarried man’, which are synonymous.

This, of course, requires that we have some account of synonymy.  It is tempting to simply say that two terms are synonymous if they have the same meaning, but this presupposes that we have an appropriate notion of meaning to work with.  Indeed, analyticity belongs to a tight-knit family of concepts – analyticity, synonymy, meaning – all of which need explaining, so it does us no good simply to explain one in terms of the other.

The way to define synonymy, Quine says, is by way of substitutability.  Two expressions are synonymous if they are substitutable, in and out of sentences, without changing the truth value of those sentences.  But this may not be sufficient.  After all, in ordinary semantic contexts, substitutability is merely a test for co-referentiality, not synonymy.  For example, the word ‘nine’ in the sentence ‘nine is less than ten’ can be substituted with the expression ‘the number of planets in the solar system’, without changing the truth value of the sentence.  And while they certainly are co-referential – the two expressions both refer to the number 9 – they are not synonymous.

If, however, we consider semantic contexts beyond the ordinary, such as modal contexts – say a sentence of the form “Necessarily, X is F,” then substitutability does seem to be a test for synonymy.  We can see, for example, that we cannot substitute ‘nine’ in the sentence “Necessarily, nine is less than ten,” with ‘the number of planets’, without a change in truth value.  After all, while it is true that the number of solar system in our planets is less than ten, it is not a necessary truth.  This shows, then, that ‘nine’ and ‘the number of planets in the solar system’, while co-referential, are not synonymous and that only synonymous expressions will be substitutable in and out of a “Necessarily” sentence.

So, perhaps we could define synonymy as “substitutability everywhere, including sentences stating necessary truths and falsehoods.”  Having thus defined synonymy, we can then define analyticity, in terms of logical truth.

It is at this point that Quine pulls his last and most clever trick.  The account we have just given is circular or at least, effectively circular.  For the only necessary truths are the analytic ones, which means that to define synonymy in terms of substitutability and necessary truth is to presuppose that we already have an account of analyticity, which is what we were looking for, in the first place.

In response to the second Empiricist idea – that individual statements have their own verification conditions – Quine introduces what will become the very well-known and much-discussed “Web of Belief”; the idea that the various statements we assent to are, in the immediate sense, dependent upon one another for confirmation, much like the strands of a spider’s web are all connected and mutually-supporting, and only are confirmed by experience as a group, at the web’s boundaries, where it connects to the world.   Thus, ultimately, all substantive statements – even the statements of mathematics – are confirmed empirically, although it may take disconfirmation across a wide span of the web to reveal the falsity of such statements, which lie near its center.

This paper, along with his book, Word and Object (1960), set much of the agenda for analytic philosophy, after the Second World War and is one of the most widely cited philosophy papers of the 20th century.

I look forward to hearing your thoughts on Quine’s arguments and what follows from them!

Daniel A. Kaufman is Professor of Philosophy at Missouri State University, and his main areas of interest are aesthetics, epistemology, metaphysics, and the philosophy of language.


  1. Thanks for this Dan. I have been reading and discussing this paper for about 15 years now and this is, by far, the best and clearest account of the arguments in it that I have yet read.

    I would suggest that in ordinary language, when we say “A bachelor is an unmarried man” we do not intend to state a logical truth, but instead to invoke a definition. There is, I think, a distinction between a definition and a synonym, which I don’t have the time at the moment to go into in detail.

    A definition is more of a test or a membership function.

    For the second issue, I wonder if that is not something of a straw man. When people talk of reductionism in the sense that Quine is using here, did anybody ever state or imply that statements have their own individual verification conditions? Quine’s “Web of Belief” can be traced back to Neurath’s analogy of rebuilding a ship on the open sea:

    There is no way to establish fully secured, neat protocol statements as starting points of the sciences. There is no tabula rasa. We are like sailors who have to rebuild their ship on the open sea, without ever being able to dismantle it in dry-dock and reconstruct it from its best components. Only metaphysics can disappear without a trace. Imprecise ‘verbal clusters’ [Ballungen] are somehow always part of the ship. If imprecision is diminished at one place, it may well re-appear at another place to a stronger degree.

    Neurath was surely one of the central figures in the tradition that Quine critiques in this paper.


  2. Robin, yes, Carnap had an entire book called “The Logical Structure of the World,” in which he tried to give phenomenalistic definitions for every scientific term. So, not a strawman.


  3. I would go so far as to say that Quine is conflating definition and synonymy and that this is a flaw in his argument. I would say that all talk of synonyms is not relevant for considering the concept of analytic statements.

    This is also the flaw in Plato’s Euthyphro argument – Socrates extracts a definition (horizo) from Euthyphro and then treats it as a synonym (tautos) and claims that one statement can be transformed into the other by substitution.

    Also Quine’s question “who defined it?” does not seem relevant. As long as you have a definition for which you have a practical agreement with your intended audience then it does not matter who originated that definition. When I talk of the “thing with the blue handle” with my wife, it does not matter which one of us first used that term, more that we know that it refers to a specific object and we know what that object is. Surely no one is claiming that there is an objective fact of the matter about the definition of a term.

    This also highlights a problem with negative philosophy of this kind. Quine, as many philosophers do, is sawing off a branch, while sitting on the unsupported side of the branch (something of which Carnap accused Wittgenstein).


  4. Hi Dan,

    [blockquote]Robin, yes, Carnap had an entire book called “The Logical Structure of the World,” in which he tried to give phenomenalistic definitions for every scientific term. So, not a strawman.[/blockquote]

    I don’t see how that implies that individual statements have individual verification conditions. Indeed that seems antithetical to the project which is a development on Russell’s ideas which he summarised “‘Whenever possible, substitute constructions out of known entities for inferences to unknown entities’.” Russell made it clear that these were logical fictions with no meaning on their own, only as part of an aggregation.


  5. I don’t see how one can conflate synonymy and definition. A definition for ‘S’ is a synomyous expression for ‘S’.

    As for what you deem a relevant question, Quine is speaking, specifically, to the Logical Empiricists who very much deemed it a relevant question.

    I don’t see what branch Quine is sawing off (nor Wittgenstein for that matter). In Word and Object, Quine is going to go on to say that even reference is inscrutable. He never accepts — or relies upon — anything beyond purely extensional language.

    p.s. I know you have a backlog of comments with us, pending, but we do have a policy of one comment, per article, per day, for the duration of the comment thread’s life, so I will publish the one’s you’ve sent so far, one day at a time.


  6. I’m inclined to think that this Quine paper, and his earlier “Truth by convention” were mistakes. But I cannot put all of the blame on Quine. His argument seems to have been widely embraced by analytic philosophers.

    To me, Quine’s argument looks solipsistic. By looking at analytic statements in terms of synonymy, Quine has in effect made language self-contained and disconnected from reality. My alternative view is that often analytic statements are definitions that connect language to empirical reality, and as such are part of the basis for meaning and intentionality in language. Likewise, in “Truth by convention”, Quine takes conventions to be abbreviations, and overlooks their role in the creative use of generalization within mathematics.


  7. With regards to the concept of “Analytic Sentence” Determining logical truth within a formal language is possible, but translating from English into a formal language is difficult, perhaps even impossible. At one time (when I was in grad school) that task was taken up by AI researchers. There were various attempts to translate into propositional calculus or Montague Grammar, which was defined precisely for the purpose of facilitating the translation, My understanding is that alll those approaches were eventually unsuccessful.

    Natural language is inherently ambiguous. It’s a feature, not a bug, that the meaning of every utterance is affected by its context, both linguistic and non-linguistic and there is no limit to how much that context might need to be expanded.

    So what we have is not a problem in understanding analytic statements, but a problem in translating from a natural language into a language in which the logical truth is well defined.


  8. The two are directly related. Indeed they are part of one and the same project.

    That each statement has the distinct verification conditions it has is due partly to its syntactic structure and partly to the sense-data that correspond to its constituent expressions.


  9. I can’t say I really understand this. Especially the following statement:

    “My alternative view is that often analytic statements are definitions that connect language to empirical reality, and as such are part of the basis for meaning and intentionality in language.”

    Given what an analytic statement *is*, I don’t see how it can “connect language to empirical reality.” Analytic statements, by definition, are not empirically verifiable, but rather, reports, as Hume called them, “relations of ideas.”


  10. Analytic statements, by definition, are not empirically verifiable, …

    Quite right. But definitions, by their very nature, are also not empirically verifiable.

    I didn’t think I was making a hard to understand point. But, in retrospect, I should have realized that. I think I need to address it in more detail in a post on my own blog.


  11. In the later 19th Century, Western philosophies, both the stream that would become known as ‘Analytic’ and that roughly understood as ‘Phenomenology,’ began to take what would come to be called ‘Linguistic Turns.’ Initially, the presumption was that modern mathematics needed a logical foundation. Then, as the sciences began discovery and development of hitherto unknown (and previously supposed ‘unknowable’) entities, a further impetus to the Linguistic Turn seemed to be a perceived need for precise and reliable theoretic descriptions and explanations of these entities. (This seemed to be true for the social sciences as well as the natural sciences.)

    But these Linguistic Turns, rather than discovering foundations for mathematics or means of generating precise scientific statements, began to isolate philosophy, not only from these fields of inquiry, but from more general human experience as well. Eventually, the best moments of these Linguistic Turns came at the end of them – when some influential or even scandalous philsopher began unraveling or simply rejecting, the presumptions underlying the Linguistic Turn.

    “Two Dogmas” is an interesting case. It is clearly engaged in unraveling certain Positivist theories concerning language, while leaving untouched the epistemological presumptions such language theories are intended to re-enforce. (Although he does end the text suggesting a turn toward “pragmatism,” as far as I know he never fully followed up on this.) So it seems to be putting an end to a certain Linguistic Turn, but in a manner that doesn’t threaten its central project (reconstructing language as both scientifically informed and yet also somehow foundational to science). It thus remains a philosophically insular text in some ways.

    I can’t help remarking on the tone. The text reeks of Old Boys Cub at America’s Oldest University for students from the ‘Best Families.’ It is certainly smug, occasionally borders on arrogant, and is elitist through and through. Oh, that dash of French tossed in for flavor! Quine writes as though the most important activity one could engage in would be lounging around a richly furnished parlour, arguing the proper use of gerunds.

    But there’s no denying its importance. It situates itself in the history of the Analytic tradition as a kind of poisoned pill. If there is no real distinction between analytic and synthetic sentences, then any logic developed on the presumption that there is, will eventually reveal itself as a closed system – a formal language in and of itself, not a model of language per se.


  12. I don’t get the point re: tone-criticism. Quine is a relatively plain speaking philosopher, save for the occasional “desert landscape” reference. I don’t see the arrogance either. The man is just very, very smart.

    Your account of the lingusitc turn also seems somewhat skewed.

    By this — “the best moments of these Linguistic Turns came at the end of them – when some influential or even scandalous philsopher began unraveling or simply rejecting, the presumptions underlying the Linguistic Turn” — I take it you are referring to the later Wittgenstein. (If not, please correct me.) But I would maintain that he — as well as Austin, Ryle, and others — were as much a part of the linguistic turn as the logical empiricists and logicists.

    As for your last point, I’m not geting it. Could you restate?


  13. Dan,

    I’ll respond in two comments. Today, language and context:

    In the later ‘70s, I experienced something of a nervous breakdown. In the months that followed, I became obsessed with developing a symbolic language where the emotions could be reduced to operative functions. It was part of my recovery to be able to look back at that effort and realize that I had developed a language that was able to say nothing about anything of consequence that couldn’t be better said in plain English. I was reminded of this when I read Quine’s example of a possible language with “an indefinitely large stock of one-place predicates, (…) and many-placed predicates (for example, ‘G’ where ‘Gxy’ means that x loves y,) mostly having to do with extralogical subject matter.” Gxy (= ‘x loves y’) happens to be a sentence from my invented language (I don’t remember what letter I used to indicate the function). The problem is, ‘Gxy’ is completely void of content; once we introduce content, we realize that ‘Joe loves Debbie’ has not only greater clarity, but richer meaning in a given context.

    “The function of logical analysis is to analyse all knowledge, all assertions of science and *of everyday life*, in order to make clear the sense of each such assertion and the connections between them.” — Carnap,

    The danger of holding too fast to the analytic/synthetic distinction is that analytic sentences, being internally true, become the model for the kind of certainty sought from synthetic sentences by way of logical analysis, and this may not be possible. In actual practice, context is all.

    “Necessarily, all and only bachelors are unmarried men” – would not have made any sense in the Middle Ages, when ‘bachelor’ designated a subordinate knight; including divorced men in the definition would raise concern in a culture that prohibited divorce (and some dictionaries include ‘never married’ in their definitions, thus excluding divorced or widowed men); and it might cause some confusion among ethologists studying the mating behavior of seals.

    Tomorrrow: ‘Linguistic turns’ and natural languages.


  14. Quine’s “much-discussed “Web of Belief”; the idea that the various statements we assent to are, in the immediate sense, dependent upon one another for confirmation. . .all connected and mutually-supporting, and only are confirmed by experience as a group, at the web’s boundaries, where it connects to the world”
    seems to have much in common with Latour’s actor-network theory and Harman’s relational ontology. For Quine assentable statement, for Latour social agents, and for Harman objects all, at least ‘in the immediate sense’ emerge only as members of networks.


  15. Dan,
    Second (continuing) response:

    As to the most interesting twist in a ‘Linguistic Turn’ coming at its ‘end’, I mean Wittgenstein, but also Derrida, to some extent Heidegger (whose own ‘Linguistic Turn’ led into poetry); in another way Merleau-Ponty, some writings of Dewey; and don’t forget Rorty.

    ‘Linguistic Turns’ generally exhaust themselves at the point one wonders, ‘what are we talking about if we aren’t talking about what people are talking about?” – in other words, at the point when we realize that natural languages are pretty good at addressing the interests of the people who use them, and simply don’t need philosophic correction.

    “It is often hinted that the difficulty in separating analytic statements from synthetic ones in ordinary language is due to the vagueness of ordinary language and that the distinction is clear when we have a precise artificial language with explicit “semantical rules,” Quine remarks at one point, before deflating this notion as a confusion.

    One confusion here is that ordinary language is not “vague,” it is open, as it needs to be to useful in a wide variety of contexts and contingencies. (A “precise artificial language with explicit ‘semantical rules'” will almost necessarily prove closed.)

    All natural languages are systematically open in at least three ways (that concern us in this context):
    1. The definition of expressions can change. (Its past is open to its present.)
    2. New expressions can be invented. (Its present is open to its future.)
    3. Expressions from other languages can be incorporated as needed. (It is open to other languages.)

    The logical analysis of language cannot even handle this openness, how can it accommodate ‘extralogical’ matters like expressions of love? Or politic gestures like, ‘hello,’ please,’ or ‘küssdiehand’?

    However, this isn’t to say that the project was not productive: it helped develop a number of artificial, formal logical systems, and contributed to development of computer language as well – closed systems that have to be repeatedly extended to mimic the natural openness of ordinary languages.

    As to tone, let’s chalk that up to a matter of taste.


  16. On re-reading I still go back to what I said. I can’t make any sense of Quine’s treatment of definition unless I am to accept the equation of definition and synonymy, which I do not.

    In any case I can make sense of a definition without such a linkage, indeed a synonym would seem to provide a rather ineffective definition. And, as I said, the question of who created a definition is quite beside the point, as long as anyone involved in any particular discourse are agreed upon the definition.

    I can make an account of analyticity without appealing to the concept of a synonym or the originator of a definition, so I cannot make sense of any objection to it which relies on these superfluous concepts.