I think there is something interestingly right about some version of the verificationist theory of meaning. The most crippling attack on the claim that such a theory is even possible, I take it, begins with Quine’s Two Dogmas of Empiricism. I will attempt to address what I take to be the primary strengths of this attack. I address these concerns, and attempt to show why they need not plague verificationism. This will allow me to conclude that verificationism does not fail (if it fails) for Quinean reasons.

The verificationist thesis I will defend has many serious problems upon which Quine himself never touches, and which seem (to this author) to be unsurmountable; I do not have the pagination resources to address these concerns here. The plan is to make explicit the Quinean arguments, and suggest a route of escape for the verificationist. The motivation is not to revive verificationism, but to pave the way for understanding why it fails by looking at unsuccessful attempts at its refutation.

In 1951, Quine puts an end to the consensus that sense can be made of the analytic/synthetic distinction. Now there is a new consensus consisting of something very much like the following six claims, the first four of which are basic corrolaries of meaning holism:

1. all of our perceptual beliefs (or their corresponding statements) are revisable in the face of theoretical beliefs (or statements);
2. all of our theoretical beliefs (or their corresponding statements) are revisable in the face of observational beliefs (or statements);
3. no statement is immune to revision;
4. no statement, taken on its own, is meaningful;
5. there can be no interesting, principled analytic/synthetic distinction;
6. there can be no interesting, principled theory/observation distinction.

I will examine the logical relations between these claims; the ultimate goal of this section is to sketch a strategy for rejecting claim 1, which represents a direct attack on what is sometimes called ‘the empirical basis’ of empirical statements. In section 2, I make explicit the argument against 1; in the final section, I extrapolate the consequences that my thesis has for a verificationist theory of meaning.

If the set of all perceptual and theoretical beliefs exhaust the kinds of beliefs that there are, forming a partition consisting of two partitioning subsets (the set of empirical/perceptual beliefs and the set of theoretical beliefs), then it is clear that claims 1 and 2 jointly entail 3. Although the Quinean thesis has statements as its subject matter, it seems right to assume that for every statement, there is a corresponding intent to express something meaningful. When a belief is expressed, it is often in the form of an uttered statement, and all statements that are meaningfully uttered have a corresponding underlying belief (explicit or not). This is not to say that a meaningful statement must be believed (to be true) by the utterer. But clearly the agent (at least dispositionally) believes that the statement is either true, false, or just possibly true or false. I assume nothing over and above this fact when I speak of beliefs corresponding to statements. So a meaningful observation statement, "there is a red thing before me", has as its corresponding belief something meaningful that is properly describable in much the same way as the statement suggests.

We can assume for the moment that 1 and 2 are meaningful in that there is something right about the theory/observation distinction as it pertains to either beliefs or statements. If we can show that some observation statement (which seems intuitively purely observational) is revisable in the face of theory, then perhaps they all are. If this is true, the only candidates for immunity to revision would be things like logical laws. But even these fail to be immutable (I gladly grant this point [that 2 is true] and set the issue aside permanently).

Because the revisability claim in 3 does not, as it stands, entail anything about how beliefs are revised, and by what, 3 does not entail 1 (at least in the absence of 5 or 6, as I shall explain shortly). Nor do any of 4 through 6 singly entail 1. I consider each of these in turn after examining the relation between 3 and 4.

Getting from 3 to 4 is misleadingly easy. Consider a case in which an observation statement is revised in the face of theory; it may seem natural to conclude that theory was therefore playing a significant role right from the start. One might argue that if there were no theoretical content in the statement to begin with, then there would be nothing to change in light of the relevant theoretical postulate. But this is a mistake. Changing a perceptual belief in light of some other, non-perceptual belief, for instance, does not entail that there was any theoretical content to the initial belief. The same applies to statements. Clearly, we can imagine a case in which an observer learns that he is the victim of an illusion, and thereby revises his corresponding (purely observational) belief. The agent then formulates the revised belief, which is tainted with theory somehow. But if the initial observational belief was not so tainted, then if it had any meaning prior to revision, it had meaning independently of any theoretical content. Of course, it must have had meaning in order to be revised at all (otherwise the allegedly revised belief would not be a revision but the initial formulation, properly speaking). If it can be shown that there is at least one such belief (and a corresponding statement) which is also capable of meaningfulness in the absence of any other empirical belief, then we have the negation of 4 despite the truth of 3 (since we have already granted 2). This point goes only to show that there is no entailment from 3 to 4.

4 does not specify that the holistic nature of meaning is of such a nature as to cross the theory/observation line (if there is a line). So while all observation statements may get part of their meanings from other observation statements, some might not get their meanings from theoretical statements. If this is even conceivable, then it is dubitable that 4 entails 1. Since an argument for 1 cannot take for granted that observation statements are infected with theoretical statements without begging the question, we may conclude that 4 does not singly entail 1.

A weak version of 4 plays a vital role in the formulation of claim 5, and Quine’s attack on the analytic/synthetic distinction clearly indicates this. The problem: understanding synonymy is necessary for the formulation of our most attractive attempts to draw the distinction, yet it is as vague and opaque as analyticity itself. Without a clear notion of synonymy, the analytic/synthetic distinction cannot be formulated in a principled manner (so claim 5 is true). Even if we define two terms as synonymous, and proceed to build an allegedly analytic proposition containing these terms, the definitions invoked must depend on pre-existing synonymies (Quine [1951], p. 25-7).

The pre-existing synonymy point indicates that no analytic statement is independently meaningful—the fact expressed here is a particular instance of 4 (and a point I have granted by accepting 2, since analytic statements are supposed to be purely theoretical, and true in virtue not of some empirical accident, but only logical truth via synonymy of terms). This is the first part of the Quinean argument to the effect that the formulation of an analytic/synthetic distinction is highly problematic and a metaphysical dream. Attempts to cash out analyticity in terms of Kantian concept containment also fail, since concept containment itself is at best only understood metaphorically (p. 21). This is the second part of Quine’s argument.

To simplify, we can read 5 as stating that (i) allegedly synthetic statements carry with them theoretical and linguistic conceptual presuppositions, and (ii) the notion of concept containment will not help to provide an argument against (i). Put this way, 5 entails the failure of the theory/observation distinction (claim 6), since it implies that there won’t be any purely theoretical (analytic) or observational (synthetic) statements (Quine [1951], pp. 38-9). Of course, this idea of purity is vague; if there is no distinction between empirical statements and their counterparts, this explains the vagueness, for the alleged distinction is a myth.

When we ostensibly stipulate the extensions of allegedly purely empirical or theoretical terms and statements, we rely on pre-existing (and often tacit) assumptions about the meanings of other terms and statements. While presupposing the meanings of additional terms sheds doubt on the analytic/synthetic distinction, presupposing the truth or falsity of additional statements suggests meaning holism, or claim 4. But 4 is not entailed because it is too general, claiming that no statement acquires its meaning independently. If 4 is true, then pre-existing beliefs or assumptions sneak into our allegedly independent empirical claims. This would entail 6 (or something very close to it).

Finally, assume that 3, 5 and 6 are true. So no statement is immune to revision. If this is true, then even empirical statements are revisable. 5 and 6 each entail that empirical statements are revisable in the face of what we might naively take to be purely theoretical statements. So the conjunction of 3 and 6, or 3 and 5, each entail 1.[1] Claim 1 is sufficient in itself to shed serious doubt on the verificationist theory of meaning that we will be evaluating in what follows, for the theory requires that some statements have as their foundations something like a pure empirical basis which serves as the arbiter of meaningfulness; unless we invoke only concepts that "answer to…empirical criteria which the subject can apply within his own experience" when making claims about the physical world, the resulting string of words fail to be meaningful (Lewis [1934], p. 130). If, however, allegedly empirical concepts or criteria are revisable in light of their counterparts, the thesis that an empirical basis confers meaning upon statements becomes trivial.

A theory of meaning which takes as the foundation of meaningfulness the kind of empirical experience which is supposed to be different in kind from theoretical data is only as successful as the theory/observation distinction is real and applicable; moreover, the verificationist thesis is only as interesting as our observation statements are isolable from their counterparts. But if claims 5 and 6 are true, verificationism fails; if claim 1 is true, our beliefs or reports about sense data can be altered by theory. So verificationism is uninteresting at best. To defend verificationism, we can show that each of 1, 5 and 6 are at least dubitable. To explain the strategy I will employ for this defense, it will help to summarize the considerations of this section. We came to the following conclusions:

There are several ways to go about shedding reasonable doubt on 1, 5 and 6—one of which is the following: if we can show that meaning holism (claim 4) is dubitable, then we may in good conscience resist the derivation of 5 and 6. Moreover, without 5 or 6, neither (xi) nor (xii) serve as a basis for sound arguments for 1, and we can conclude that 1 may be false. This is the route I will take in what follows.

These are the claims we will be evaluating here:

1. all of our perceptual beliefs (or their corresponding statements) are revisable in the

4. no statement, taken on its own, is meaningful;

To indicate why 4 ought to strike anyone as at least reasonably dubitable, it will help to begin by examining the conception of meaning underlying a species of verificationism that allegedly falls victim to the Quinean attack. The verificationist theory of meaning in question assumes that it is possible to attribute meaning to individual statements about physical facts (I will henceforth refer to the meaning-in-individual-statements thesis as ‘MIST’; MIST is the negation of 4). For the object of inquiry just is the unitary statement or utterance. But no thesis should play the role of an assumption if there is good reason to think the assumption is false. 4 supposedly provides the good reason. Let us see, beginning with Schlick’s (1936) formulation of the verificationist theory of meaning.

According to Schlick, it is by way of the act of definition that we construct rules for bestowing meaning upon a sentence (p. 349). The conditions dictate the following biconditional: "sentence S expresses a (synthetic) proposition just in case it is empirically verifiable in principle". Since propositions have to have meaning in order to be propositions, the biconditional represents a prescription for figuring out whether a statement is meaningful. Any instance {proposition in the extension of S?} of S must be verifiable in principle in order to meaningfully express anything empirical.

Statements that are, in virtue of empirical circumstances, unverifiable, can still be meaningful. The verifiability possibility in question is of a logical kind. Hence stating that there is life on Mars would be meaningful even if it were empirically impossible to verify the fact in one’s lifetime. Even statements that express alleged facts that violate the present laws of nature can be meaningful just in case the utterer has some kind of access to verification or falsification. Empirical possibility is not what’s at issue. The physical world reveals the truth and falsity of statements, while the logical possibility of verification endows statements with meaningfulness.

Verification of empirical claims, when carried out, is conducted by way of observation.[2] In order for this to even be possible, we must be able to stipulate what the relation is between the percepts (or sense data) and the terms involved. We agree on a logical grammar (or, more appropriately, a language) which specifies allowable relations, the extensions of terms, and conventional rules for acceptable uses of linguistic expressions. The meanings of particular statements are given, in large part, by the immediate data of sensory experience, which we interpret in terms of the stipulated syntactic and extension-fixing rules. But meaningfulness itself is given by the very fact that we have some kind of access to a method of verification or falsification (p. 351).

These are lenient senses of ‘verification’ (or ‘confirmation’) and ‘falsification’ (or ‘infirmation’) at work. We could never verify a statement in such a way as to preclude the possibility of post-verification doubt regarding the truth of the proposition. Statements do not really need to be verifiable in such a way as to yield certainty. For no empirical "hypothesis could [ever] be established as absolutely true" (p. 357). But no matter how weak the sense of ‘verification’ is, if all empirical statements and beliefs can be ‘infected’ with theoretical content, then the empirical basis of verification is either trivial or a myth.

If theoretical presuppositions sneak into our observation reports, then we will never have a purely objective method of verification—even in cases as simple as reports like, "there is a spot on this lens"; if our theories determine—even in part—what we claim to see, hear etc., then verification cannot amount to the non-biased method of figuring out whether our statement is confirmable (or infirmable); if all of our observation statements are revisable in the face of theoretical beliefs (or statements)—despite the fact that the empirical (sense-) data remain the same, Schlick’s verificationism becomes trivial. And of course, if the distinction in claim 6 is imaginary (and not just difficult to apply), then verificationism is false. Finally, if the Quinean is right about claim 4, we can bring the verificationist program to a halt right from the start by attacking the MIST assumption.

What appears to be a major challenge to MIST can be summed up in two words: name one. It seems that if MIST is true, then it ought to at least be possible to formulate one example of an independently meaningful statement. But this is false. MIST is perfectly compatible with the claim that the meanings of single statements is introspectively opaque. Perhaps these meanings only become transparent if we dissect the cognizer’s brain (or, more realistically, after we have formulated a sophisticated psychology). It is far from ad hoc to insist that the multitude of (cognitive) concepts allegedly contained in our statements are not explicit, or easily available to consciousness.[3] A brief look at a classical empiricist theory of ideas, along with some insight from contemporary psychology, will help elucidate.

Up until this point, I have been speaking very loosely about an ‘empirical basis’, and the ‘shattering’ thereof. I now want to clarify what all this jargon might amount to.

Assume that Locke and Hume were approximately right about the foundations of knowledge.[4] So even the most complex ideas and propositions we entertain are in some sense reducible to basic, atomic (sense-) impressions of sorts. Take Locke’s view, for instance. Complex ideas are concatenations of the simples. Even on such a view, many complex ideas are going to be too complex to be decomposable (in fact) into their atomic elements. But this by no means entails that decomposition is impossible in principle; the possibility in principle of decomposition can be seen if we adopt the following interpretation of the Lockean scheme.

This simple scheme presents a response to the Quinean charge that our understanding of concept containment is hopelessly metaphorical (Quine [1951], p. 21). Whether it is actually metaphorical is one question; whether it needs to be this way is another.

I do not mean to imply that the relation of concatenation is the only, or even the most common, relation tying basic concepts together to form the more complex. But there seems to be no reason to deny outright that as relations increase in complexity and basic ideas in numbers, the resulting complex ideas are the product of the same kind of stuff as BLACK DOG, differing only in degree of complexity.

If something very much like the empiricist thesis depicted here is right, then it seems that the sense in which 4 is true (and MIST false) is uninteresting. Statements get their meanings in terms of their component empirical concepts (or basic ideas); in that sense, they rely on other concepts for their meanings, and perhaps on other statements too. But I don’t think that this is precisely the kind of meaning the verificationist is concerned with. In fact, I think that if the empiricist picture is right, then a statement’s dependence on meanings of other concepts and statements is a victory, not a challenge, for the verificationist. I will explain why this is so in section 3.

Locke’s naïve empirical theory of complex idea or concept formation is not very impressive by 21^st century standards. So it is perhaps surprising that a compelling reason has yet to be devised for thinking that the theory won’t work—that is true both for armchair and laboratory reasons. Given this fact, we must entertain the possibility that the theory is true when putting forth psychological hypotheses like, "Bonzo can’t make an observation report that is free of theoretical content". This is an instance of 4, and it cannot be assumed to be true from the armchair. If Bonzo can make theory-free claims, then 4 is false, for 4 entails that Bonzo cannot make such claims.

Having granted that we ought to be careful about making psychological claims in the absence of the proper psychological arguments, I want to end this section in making explicit my thesis regarding MIST and postulates 1, 4, 5 and 6.

If it turns out that Bonzo can formulate at least some concepts and statements in the Lockean way, then there might very well be an empirical/logical idea distinction corresponding to the theory/observation distinction. Given that the ability to count is not always accompanied by the ability to perceive color, and vice versa, there’s at least one thing we know. Those faculties are different, at least in normal brains. If the faculties are different, there may very well be a real difference between the outputs of those faculties: namely, empirical and theoretical beliefs and ideas ???. So it may be possible to form ideas about the color of one’s dog without forming ideas of a theoretical nature (of which there are very likely several natural kinds [6]). If this is true in the case of Bonzo’s BLACK DOG idea, then it might also me true that the empirical content of his statement, "there is a black dog before me", really is—or can be, if he is very careful—purely empirical. The fact that Bonzo may revise his corresponding belief when he learns that he is under the influence of a hallucinogenic drug does not entail that the meaning of this particular statement (under normal conditions) is somehow inseparable from his belief in the reliability of drug-free percepts. (The belief in the reliability of perception may be a example of a theoretical postulate that ‘infects’ observation statements.)

These considerations indicate that MIST is not a ludicrous postulate, and that 4 is far from obviously right, since there is good reason to doubt meaning across-the-board-holism. Not only does this make it possible to resist the first, fifth, and sixth postulates, we are left with independent reasons for thinking that MIST is not ludicrous. Each point in turn.

First, we must admit, even in light of all this, that claims 1, 5 and 6 may still be true—just not obviously so. Interestingly, what seemed to be consequences of claim 1 no longer seem very consequential. For although it may be extremely difficult to pin-point the empirical basis of complex ideas (and therefore statements), there may very well be one. If there is one, then the theory/observation distinction isn’t a ludicrous myth, it’s just hard to get. But in this case, claim 6 is nevertheless false.

Claim 4 (meaning holism) seemed to provide a conclusive argument for the claim that all perceptual beliefs are revisable in the face of theoretical postulates. This seemed to imply that we ought to take with a grain of salt the empirical basis of observation statements. But because observation statements are supposedly grounded in sensory or perceptual beliefs, an argument for 1 requires an argument for the claim that there is no such Lockean foundation. For if there are theory-free complex ideas, and if statements are simply sets of such ideas bearing certain relations to one another (like concatenation), then the empirical basis required for a theory/observation distinction may exist. So MIST might be true, and claims 1, 4, 5 and 6 false. There is no conclusive evidence to the contrary that warrants consensus regarding 4. If such evidence were to arise, we ought to reconsider our stance; but for now, those who resist radical holism can still do so in good conscience.

I will end in pointing out the consequences that my thesis has for a verificationist theory of meaning à la Schlick.

The pressing questions were these: (1) why make the assumption that statements are the units of meaning? (2) How does one embark on the task of formulating a verificationist theory of meaning without presupposing anything about meaning (Quine’s point)? (3) How do we deal with the seemingly disturbing verificationist requirement that there must be a neat theory/observation distinction? I will save (2) for last.

In light of the fact that there is no reason (as of yet) to suppose that the verificationist assumption (MIST) is more controversial than holism, I think that the right answer to (1) is, why not? For recall that the holistic claim was a strong one: it states that no statement can be the unit of meaning. Having shown that the arguments for this claim lack validity, there is room for doubt. But there are positive for accepting verificationism too. The answer to (2) will clarify what I take to be the most important one.

(3) was answered in section 2, where we saw that the theory/observation distinction may be real, given that the empirical/theoretical distinction might be, and that statements may at bottom really be composed of simple concepts and their relations {in virtue of the fact that the underlyinig ideas/concepts are}. The question we ought to ask is that of whether verificationism can ever be useful; for it would be extremely difficult to isolate the set of theoretical concepts from their counterparts. I don’t think the theory would be useful if we attempted to apply it. For now, anyway, even the idea of a black dog is too complex for our psychology. The real objection to verificationism on the basis of the theory/observation distinction, as I see it, is just that verificationism is fruitless because we wouldn’t know how to begin categorizing simple and complex ideas—let alone statements. But this does not make verificationism false, just cumbersome.

(2) is a rhetorical expression of distaste for any attempt to define or characterize meaning in a non-circular fashion. How do we avoid this approach? We don’t. What we can do is assume that the modern empiricists’ general intuition about the foundations of knowledge was on the right track, and test the assumption by building from it a theory of meaning that does the work we want it to do. I cannot here enter into an analysis of what we want this theory to do for us, but I do want to end in stating why I think it is wise to examine our empirical claims for empirical content based on the origins of the constituent empirical concepts.

Whether we can explain away all empirical concepts in terms of sensory concepts (plus the relations they bear to one another)—this is an issue for both philosophy and psychology, I think. And the relations concepts bear to one another may forever elude us. But one thing is clear. The possibility that this is how we acquire meaningful concepts cannot be ruled out yet.

Conclusion

If it turns out that our allegedly ‘abstract ideas’ (like substance or general physical extension—two alleged ideas attacked by Hume [and sometimes Berkeley]) are grounded in nothing but myth or the overactive imagination, then whenever we set out to make any claim about the objects of our concepts or ideas and inquire into their meanings,[7] it is incumbent on us to also inquire into their origins if we can. If their origins are Humean or Lockean sensory impressions, then it is plausible to think that the meaningfulness of the statements we form using those concepts depends on whether they are reducible to real empirical concepts (i.e., their constituent parts—simple sensory ideas): if I lack the concepts DEMONIC POSSESSION or BLAT CACK, then how can I, in good conscience, insist that I am not speaking nonsense when I state that blat cacks are demonically possessed? The lack of an empirical foundation ought to deprive the statement of meaning if such a foundation is necessary for the formation of the constituent concepts are formed. This is the most compelling reason to be a verificationist.

We know that sensory experience is necessary for knowledge of physical objects, but we do not know whether it is sufficient. This is an empirical question that must itself be addressed empirically. However, it is noteworthy that even if a cognizer’s theoretical and empirical concepts will seem impossible to untangle when all the results are in, a revised and interesting verificationist theory of meaning might still be possible. The theory could preserve the intuition that if we make statements using the names of concepts that we do not in fact possess, then whatever the nature of the concepts, we utter nonsense. For in what can intended meaning consist if not what we intend to mean? And in what can an agent’s intent be grounded if not in the things he is capable of thinking?