Willard Quine's "Two Dogmas of Empiricism"

"Two Dogmas of Empiricism"
from Philosophical Review, January 1951, 60(1): 20-43.

Introduction

Quine's "Two Dogmas" was a revolutionary article that is still debated today. Quine's findings have big consequences for not only epistemology and metaphysics, but philosophy of mind and philosophy of science.

Think of "Two Dogmas" as having two parts. Quine's own way of splitting the parts is to suggest that empiricism has two dogmas (two principles that are never questioned) which in fact are false (Quine is an empiricist too, actually. He simply wants a more "robust" or honest empiricism). The "Two Dogmas" are:
  • that there is a principled distinction between analytic and synthetic propositions
  • that reductionism is true
The first dogma argues that there is a real way to distinguish between propositions that are analytic and propositions that are synthetic. The second refers to the empiricist claim that the meaning of a single proposition is its verification conditions. Think of it this way:

"The sun is at the center of the solar system"

The meaning of the proposition is clear -- I know what it would mean for it to be true (the sun is at the center) and I know what it would be like if it were false (the sun would not be at the center). Either way, there is a simple experience which confirms or confutes the proposition.

Notice that some propositions, then, are meaningless (they have no verification conditions). The proposition "2 + 2 = 4" is not verifiable. No empirical circumstance confirms or confutes it (though some circumstances may exemplify or fail to exemplify it). These statements are taken dogmatically to be the "analytic" statements, and the suggestion (combining it with the theory of reductionism) is that (1) there are analytic truths, truths "grounded in meaning independently of matters of fact" and synthetic truths, truths "grounded in fact" that are entirely separate; and (2) that synthetic propositions are reducible to combinations of contents derived from experience (and thus verifiable by those contents). Analytic statements would be statements not so verifiable. So the two dogmas, Quine thinks, are related. As such, Quine will attack both.

Think of Quine's project here as this:

Part I: (sections 1 - 4) the Negative Attack

Here Quine will attack all of the traditional ways in which "analyticity" has been explained. None of these ways, he suggests, provides a foudnation for the term.

Part II: (section 5): the Positive Attack

First, Quine will attack reductionism and show that holism -- his theory of meaning -- is actually the right one. If Quine's theory is right, however, the analytic/synthetic cannot be a sustainable dichotomy.

Let's turn to the arguments.

Part I: Negative Attacks on the Tradition

Section 1: Background for Analyticity

The origin of the distinction between analytic and synthetic Quine traces back through Kant to Hume and Leibniz, but their accounts are imprecise. The first attempt is to argue that

Statement S is analytic if and only if a denial of that statement results in a contradiction.

Quine's argument here is straightforward: what does this mean? Does this presuppose analyticity? He thinks it does, so he moves on to other problems with Kant's definition:

(a) it is restricted to statements of subject/predicate form (are all analytic truths of the form "X is P"?)
(b) Kant relies on a metaphorical notion of "conceptual containment" that is unexplained

Lastly, Quine notes, Kant says that analytic propositions are

(c) "true in terms of meanings".

It is here that Quine begins to "dig in". What is meaning?

Quine suggests that many have confused "meaning" with "naming", or "meaning" with "reference". What he means is that there is a difference between the meaning of a term and the extension of the term. Here's an example:

"Mark Twain" and "the man who wrote Huck Finn" actually refer to the same object (the same guy). So they are names for the same person. But they do not have the same meaning. Whereas Mark Twain could not possibly have been anyone but Mark Twain (the person the name refers to) the description "the man who wrote Huck Finn" did not have to refer to Mark Twain. Similarly, while it turns out that the expressions "creatures with hearts" and "creatures with kidneys" refer to the same exact entities in the world, the expressions are different in meaning.

This being the case, Quine notes, meaning turns to a discussion of the "synonymy of linguistic forms". So we say things like "X means the same as Y" if and only if (henceforth I'll say iff ) X and Y are synonymous.

So far, so good. Now Quine notes that there are two types of statements we call "analytic".
  • (1) "No unmarried man is married"
  • (2) "No bachelor is married"
(1), Quine says is "logically true". It is true under any reinterpretation of the non-logical particles (these being "no" and "is"). So for such statements to be true requires a presupposed stock of logical particles ("no", "or", "and", etc). But (1) is not Quine's target. (2) is.

What about (2)? How is it analytic? Quine notes that most people argue that (2) is analytic iff the terms used are synonymous -- i.e. if one can substitute one for the other and translate (2) into (1). (So one would substitute in (2) the world "bachelor" for "unmarried" and end up with (1).)

Now we're moving, says Quine. But what is synonymy?

One proposed explanation is given by Rudolph Carnap. Carnap suggests that a language can be broken up into atomic parts. He suggests that we would wind up with an exhaustive list of all of the simple atomic statements like "the car is silver" and "the table is twenty pounds" and so on. Once we list all of these atomic statements, Carnap suggests, we assign truth values to them. So it turns out that in this world "the table is twenty pounds" is true, as is "the moon is circular" and so on. Once we have done so, we have given what Carnap calls a state description of the language.

Once we've listed all the atomic statements and their truth-values, we can build up other sentences using the logical particles. So let's say that "the moon is circular" (call is P) and "the table is twenty pounds" (call it Q) are true. Then
  • P & Q is true
  • P v Q is true
  • ~ P is false
  • ~ ~ P is true
and so on. So using the rules of the logical particles we can build up more complex statements and know their truth values (since this is dictated by the truth-conditions of the particles themselves).

Now, according to Carnap,

Statment S is analytic in language L (the one under consideration)
 
iff
Statement S is assigned the truth value "T" in any possible state description of the sentences in L.

So, in other words, regardless what truth values are assigned to the atomic statements of the language, the analytic statements will always turn out to be true.

In other words Carnap is saying (using state descriptions as his terminology) analytic truths are true in all possible worlds (a possible world being a possible state description).

Quine's argument against Carnap is a good one. Carnap's proposed definition of analyticity only worlds if we've already stipulated that statement such as "X is married" and "X is not a bachelor" are independent. In other words, Carnap's definition only works if it is the case that if a state description assigns "T" to "X is married" then it assigns "F" to "X is not a bachelor".

Quine is right. If "All bachelors are unmarried" is analytic, then there can't be a possible state description where "X is married" and "X is not a bachelor" get different truth values. But then it seems like Carnap's claim that analytic truths are analytic because they are true under any state description is not really the case. It is really any state description following certain stipulations.

But Carnap offers us no reasons for making such stipulations.

Thus, Quine argues, Carnap's definition works for statements such as (1) -- which are logically true. But not (2). And statements like (2) are the ones we want to figure out. Since Carnap gives us no way to understand synonymy

Section 2: Definitions

Often synonyms are treated as definitions. If "bachelor' is defined as "unmarried" we can substitute one for the other in (2) and wind up with the logical truth (1) that we need.

But what counts as a definition?

Quine notes that we can turn to the dictionary to find definitions, which are always given in terms of synonymies. We look up "bachelor" and it is defined as "unmarried male".

But who says what is what? Quine notes that the lexicographer is really an empirical scientist. If he records "bachelor" as defined by "unmarried male", then he is recording what he believes to be a pre-existing synonymy. If this is the case, however, dictionary definitions presuppose synonymy instead of explain synonymy.

Quine says that the only non-question begging example here is definition by fiat. We can introduce a word as a definition for another simply by conventional decision. In such a case we create the synonymy by stipulation. Unfortunately, Quine notes, most definitions do not work this way, and presuppose synonymy.

So definitions can't help us to understand synonymy, our current best hope of understanding analyticity.

Section 3: Interchangability Salva Veritate

It is often argued that two words or statements are synonymous when the pairs are interchangable salva veritate. To be interchangable in this way means that you can exchange one of the pair for the other in a sentence and not affect the truth value of that sentence.

Ex. Let's say that "bachelor" and "unmarried" are interchangable salva veritate. Then we should be able to make the substitutions in any sentences in which they appear. So:

"Fred is a bachelor" (let's say it is true). So "Fred is unmarried" should also be true if they are interchangable. It is.
However, Quine notes that it is clearly the case that we would not accept any interchangability of these two terms. For instance, what about "Fred has a bachelor of arts degree". If this is true, does that mean we should be able to interchange "unmarried" to say "Fred has an unmarried of arts degree" and say something true?

The rejoinder here, Quine notes, would be that "bachelor of arts" is actually one word (although it looks like three). So interchanging "unmarried" for "bachelor" would not really be a "word for word" substitution.

Quine says "fine, no problem". He accepts this and argues that still interchangability salva veritate will not suffice to explain synonymy. Quine suggests that the typical argument says this:

(1) "X and Y are synonymous" if and only if "All and only Xs are Ys" is analytic

Of course, we need an account of synonymy not presupposing analyticity. So it is argued that:

(2) X and Y are synonymous if and only if X and Y can be interchanged salva veritate

Once we have gotten rid of the "what is a word" problem, (2) seems to be true. Next it is suggested that of course

(3) Necessarily all and only bachelors are bachelors

is true, and that if "bachelor" and "unmarried man" are interchangable salva veritate then

(4) All and only bachelors are unmarried is analytic.

If this is right, analyticity is explained via interchangability salva veritate.

Unfortunately, Quine says, the argument here has a bit of "hocus pocus" about it. The main problem here is that (3) makes use of the term "necessity", a term which only makes sense under the given argument when it is used to introduce an analytic statement (recall that for the empiricist only analytic statements are necessary!). So "necessary" seems to be a part of the concept of analytic and so cannot be used in (3) as part of the proof to get us to (4), which is the discovery and introduction of the analytic.

There is, however, a way to get rid of the term "necessary". Quine argues that the only way to do this is to  deal with a strictly extensional language (a language that uses terms only referring to the world). In an extensional language (for an empiricist) there will be no term "necessary". So let's try the argument again in such a language.

Let's stipulate an extensional language L. In this language we have laid out all of the predicates (words like "red" or "heavy") and relations (phrases like "taller than" or "heavier than"), plus all the logical notation (terms like "or", "and", "if and only if"). Since the language is extensional , the meaning of a word is the object(s) to which it refers. This being the case, two terms X and Y will be interchangable salva veritate if and only if X and Y refer to the same things.

The unfortunate point here is that here we aren't dealing with synonymy, since interchangability salva veritate will never be sufficient for synonymy (since synonymy refers to meanings and extensional languages do not deal with meanings).

Since extensional languages are the only ones that don't have "necessary" as a term, synonymy fails as a condition for analyticity.

Section 4: Semantical Rules

It is also suggested that the difficulty in separating the analytic sentences in a language from the synthetic sentences is not a difficulty pertaining to the distinction itself, but rather to the "messy nature" of natural languages (natural languages are languages used by people). It is suggested by some that it would be easy to draw the distinction in an artificial language (say, a computer language).

Let's see. Quine thinks not.

Let's call our artificial language "L". What we want to know is what "analytic for" in "S is analytic for L" stands for.

Try One:

We can immediately make it more specific by claiming to look a a specific artificial language, Lo. We can claim that the semantic rules of Lo tell us which of the sentences of Lo are analytic and which are not. The problem here is that we know which sentences are analytic, but not what analyticity means.

In other words, we could define "analytic-for-Lo" as "that set of sentences" but not "analytic" in "analytic-for-Lo".

Try Two:

We know that of the sentences in Lo, the analytic ones will be a subset of the sentences in the set of all the true sentences in Lo. So perhaps we can claim that

S is analytic if and only if S is not merely true but rather true by semantic rule

What, though, does "semantic rule" mean? Here one could argue that the semantic rules are those rules that are stipulated by the "postulates" (axioms) of the language. But which sentences are the postulates? Any set of sentences in Lo, Quine argues, could be postulates.

Quine's revolutionary point: The word postulate is significant only relative to some act of inquiry. If this is true, then Quine's radical point comes out (561, left side):

"If this is true, no subclass of the truths of L are intrinsically more a rule than any other.
So if "analytic" means "true by rule", then no truth of L is analytic to the exclusion of any other."

The moral of the story: if the significance of "postulate" (upon which "semantical rule" and then "analytic" will depend) depends on the our interests of inquiry, and not some special logical status of sentences then there is no principled way to call any set of sentences in a language "analytic" as opposed to "synthetic".

Part II: Quine's Positive Arguments

Section 5: The Verification Theory and Reductionism:

A final attempt to establish analyticity would be to understand meaning as verification- "the meaning of a statement is the method of empirically confirming or infirming it". The meanings of terms or statements are correlated with experiences (translated into terms which appeal only to sense-data). Carnap rules for correlation, but eventually abandoned the attempt. The failure of any verificationism is the realization that there are no discrete experiences which function to confirm or deny any discrete statement. "[O]ur statements about the external world face the tribunal of sense experience not individually but as a corporate body". This view is called holism. The rejection of verificationism eliminates the possibility of defining analyticity as maximally verifiable. The fundamental mistake is in thinking we can ultimately separate the factual elements of a sentence that contribute to its truth from the logical elements that contribute to its truth.

Empiricism without the Dogmas:

Our beliefs form a web, with the outer fringes connecting to experience. Revision at the edges leads to revisions elsewhere in the web, but the decisions of where the revisions will occurs are underdetermined by the logical relations among the beliefs. In the revision, "any statement can be held true, come what may, if we make drastic enough changes elsewhere in the system". The difference between analytic and synthetic reduces to the feeling of willingness to revise based on experience-those we call "analytic" are simply the more "recalcitrant" ones. We develop hypotheses and posit theoretical, non-empirical entities to explain empirical phenomena and accept their existence as long as they continue to do their explanatory work. Ultimately there isn't a split between those scientific conceptual schemes we pragmatically accept because of their continued verification and those we accept as logically certain. All beliefs are in the end "thoroughly pragmatic."