W. V. Quine, in Two Dogmas of Empiricism^[1], makes a number of claims concerning the failure of two “dogmas” of the empirical tradition. For Quine, the failure of each is, in fact, intimately attached to the same error. As he explains it, analyticity (as one of the aforementioned dogmas) begins from a position of establishing a division of sentences, or expressions, into two classes: analytic and synthetic. The division is generally intuitive one, based upon the apparently special nature of statements which are always true (called analytic). But Quine wants to say that the distinction between the two classes is in the end, incoherent. Or as he puts it, “That there is such a distinction to be drawn at all is an unempirical dogma of empiricists, a metaphysical article of faith.”^[2]

What he offers instead is a picture of language as pragmatic and naturalistic. He explains:

The totality of our so-called knowledge or beliefs,.., is a man-made fabric which impinges on experience only along the edges.... A conflict with experience at the periphery occasions readjustments in the interior of the field. Truth values have to be redistributed over some of our statements. Reevaluation of some statements entails reevaluation of others, because of their logical interconnections — the logical laws being simply certain further statements of the system,..^[3]

There is, in the final analysis, no statement of a language which is true necessarily. The contingency of truth (upon experience) is all pervasive, and infests even the most fundamental statements.

However, he continues on to suggest that we can hold some statements as more immediately contingent than others. And that this is a perfectly rational thing to do. He notes that:

A recalcitrant experience can,.., be accommodated by any of various alternative reevaluations in various alternative quarters of the total [conceptual] system; but,.., our natural tendency to disturb the total system as little as possible would lead us to focus upon these specific statements [which themselves directly conflict with experience].^[4]

And elsewhere he says directly: “Any statement can be held true come what may, if we make drastic enough adjustments elsewhere in the system.”^[5] This brings us directly to a puzzle concerning the coherency of the analytic/synthetic distinction itself.

It now seems that it would be quite easy to establish a coherent distinction between analytic and synthetic statements. The project would look something like this:^[6]

1. we have an, admittedly vague, idea of what the analytic sentences are. Let’s make this class precise arbitrarily, just by making decisions in the problematic cases.
2. we can simply decide to hold these forever
3. The analytic sentences would be just the ones that we would be disposed to hang onto come what may

At first, the proposal seems to be suggesting something rather simple. We are asked to form a class of statements, made up of unproblematic (and uninformative) members, as well as some arbitrary selection of problematic candidates. The membership set to this class is thereafter well defined. Indeed, we might understand ‘analytic’ as simply denoting membership in this class. The distinction between ‘analytic’ and non-‘analytic’ (or synthetic — thus understood) would be clear.

Quine would not have any difficulty so far. The new ‘analytic’ only shares passing resemblance to the sense of analyticity he dispels. This time ‘analyticity’ is merely the property of belonging to a particular arbitrarily well defined class of statements. In a sense, if this is all there is to the puzzle, the new term ‘analytic’ lacks any meaning at all with respect to language generally.

But there is a further stipulation. We have agreed, under our new formulation of analyticity, that any sentence which is a member of this class will be taken as true come-what-may. In this way, the member sentences share more than a passing resemblance to the notion of analyticity which Quine wants to reject as incoherent. Indeed, under every pragmatic circumstance where we appealed to analytic truth, we may do the same now, so long as the analytic statement is in fact a member of our set. So we seem to have a coherent (or a non-irrational) distinction between analytic and non-analytic statements, which also conforms to our intuitive understanding of what analyticity should mean.

A statement is understood as analytic not simply by virtue of some kind of membership in an ‘analytic class’, but also by virtue of its (stipulated) truth invariance across all conditions.

On the surface, Quine is contradicted. Here is a notion of analyticity, which seems to conform to our old understanding of the meaning, and is coherently defined. However, much of Quine’s difficulty with the concept of analyticity is not simply a doubt concerning the validity of decisions about truth invariance in borderline cases, but that any notion of truth invariance can come from no other source than arbitrary definition.

In discussing the process of definition, Quine identifies three types of relation obtaining between the definiendum and the definiens:

The definiens may be a faithful paraphrase of the definiendum into [a] narrower notation, preserving direct synonymy as of antecedent usage; or the definiens may, in the spirit of explication, improve upon the antecedent usage of the definiendum; or finally, the definiendum may be newly created notation, newly endowed with meaning here and now.^[7]

Initially, our project might have appeared to be creating a new term ‘analyticity’ (apart from the old one), conforming to the last of these three sorts of definition. But we were able to preserve at least some of the relevant aspects of the previous understanding of the term, more in the form of the second type of definition.

Quine has tremendous difficulty with the first sort of definition because of its reliance upon synonymy. There seems no way of guaranteeing the validity of synonymous translations, without a prior acceptance of analyticity. But now we appear to have analyticity.

Not quite. Analyticity (prior to revision), by virtue of its absolutely objective status, could help us in preserving truth while making translations from observation statements into non-observational synthetic statements. But if all there is to our new ‘analyticity’ is truth invariance by virtue of set membership, the rules of translation no longer guarantee truth preservation, assuming valid observations. Quine would almost certainly want to claim that it’s obvious such a class cannot serve as the foundation of a system of truth derivation from real world observations.

The problem, for Quine, is not that truth-invariant statements are (necessarily) logically contradictory to experience, but that a distinction between statements based upon truth invariance will necessarily be definitional, and be unable to help preserve consistency of belief with respect to the ‘real’ world. For Quine, the role of consistency preservation should be recognized as pragmatic limitations, not a priori linguistic constructs.

Whether Quine is accurate in his condemnation of truth-invariance outright is another issue. It seems, at least intuitively, that such animals could exist. His demonstration of the sorts of truth-invariant statements which do not exist may be well grounded and consistent. But he does not offer a demonstration of how the concept of analyticity itself is necessarily doomed to fail. (Unfortunately, this does not appear to be the appropriate place to pursue this issue further. Both for the sake of brevity, and the sanity of the reader, I shall stop here.)