April 15, 1997
The purpose of this paper is to reconsider the status of Verifiability Principle (VP) in epistemology. For that purpose, I examine the most satisfactory version of VP in the first section. In the next section I briefly look at the use of VP in Carnap and Feigl's view on the existence of physical objects. Through this example, I consider the role VP can play in epistemology in the final section. My conclusion is that maybe we can accept VP as an explication of our basic epistemological principle, but whether this is a better explication than other alternatives is not clear.
1. The most satisfactory version of Verifiability Principle
Hempel (1965) summarized and analyzed many different attempts to formulate VP as a satisfactory criterion for cognitive significance. Using his analyses, I would like to formulate what seems to me the most satisfactory version of VP:
(VP) A proposition P is cognitively significant (i.e. can be true or false) only if either
(1) P is analytic or contradictory; or
(2) P is capable of test by experiential evidence at least in principle, namely either
(a) P's all non-logical words are analyzable at least by partial interpretations, or
(b) removal of P from the theoretical system it occurs have some effect on the explanatory and predictive power of the system.
What does this principle mean? First of all, this is supposed to be a necessary condition of cognitive significance. For a reason I discuss later, a proposition which meets one of these criteria (especially 2b) can be meaningless, therefore this cannot be a sufficient condition.
Then what do each condition mean? Let us start with the condition (1). A proposition is analytic (or contradictory) if and only if it is true (or false) by rules of logic alone or by rules of logic and meanings of the words. For example, "a bachelor is unmarried" is analytic, and "a bachelor is married" is contradictory. This condition is relatively unproblematic.
The condition (2) says that the proposition should be testable "at least in principle". This qualification is added because a meaningful proposition may be untestable because of mere technological difficulties (Hempel 1965, 104, n.). Since what matters for VP is conceivability of an empirical test of it, we need the qualification.
(a) and (b) depict two alternative ways of obtaining empirically testable consequences of a proposition. According to (a), all non-logical words in the proposition should be "partially interpreted". To illustrate what "partial interpretation" means, take the concept of length as example (Hempel 1965, 110-112). The concept of length itself does not have exact empirical meaning, because the length of distance between two points can take any positive real number, while observational reports cannot distinguish, say, [[radical]]2 cm from [[radical]]2+ 10-100 cm. But combined with appropriate subsidiary hypotheses, the concept of length can be interpreted by a standard rod: if the length of interval i is r cm, then if a standard rod is applied to i, the reading of the scale is r +/- e cm. This is called a partial interpretation. Sometimes theoretical terms are not even able to be partially interpreted, however, and this is why condition (b) is added as an alternative. When we accept an axiomatized theoretical system (like a physical theory), often primitive terms in the theory appearing in axioms have no empirical interpretation (whether partial or not). Apparently we do not want to say that these axioms are meaningless. Since these axioms have indirect empirical consequences through sentences derived from them and subsidiary hypotheses, we can use (b) as a criterion for the meaningfulness of them.
A couple of comments to (a) and (b) are in order. First, a proposition which meets (a) may not meet (b) (this is why we need both). For example, Suppose that in a theoretical system there is a proposition (x) [P1x -> (Qx <-> P2x)], where P1 and P2 are observable predicates and Q is a theoretical predicate which does not appear anywhere else in the system (Hempel 1965, 114). Even though Q is interpreted by this proposition (thus this proposition meets (a)), omission of this sentence from the system does not have any empirical consequence (thus it does not meet (b)). Second, (a) and (b) does not necessary rule out all meaningless propositions. (a) does not rule out hardly meaningful propositions like "the length of the moon smells green". As for (b), suppose that S1, S2, and S3 are axioms which meet (b), and N is a meaningless proposition. Then another axiomatic system N&S1, S2, and S3 is equivalent to the first one, and N&S1 meets (b) even though it involves a meaningless proposition N (Hempel 1965, 115-116). This is why this version of VP does not provide a sufficient condition for cognitive significance.
To summarize, I think that this version of VP is successful in admitting most scientific propositions as cognitively significant. This is important for VP because they are paradigm cases of meaningfulness. On the other hand, as we saw above, it does not rule out propositions which are, intuitively speaking, not meaningful. Whether this is a serious problem depends on how we use VP in our epistemology. So let us turn to the use of VP in Carnap's account of physical objects.
2. Carnap's account of physical objects
Carnap (1950) distinguishes two kinds of questions. Internal questions are ones "of the existence of certain entities of the new kind within the framework" (206; emphasis in original). External questions are ones "concerning the existence or reality of the system of entities as a whole" (ibid.; emphasis in original). For example, questions about existence of King Arthur or unicorns are internal questions, because the concept of 'existence' in these questions is determined by the system of observable entities, and hence the way to evaluate these questions is also determined by the system. On the other hand, when we ask a question about the existence of thing world itself, this question is equivalent to the question whether we should use the linguistic framework of thing language. This is an external question because the concept of 'existence' applies to the system of thing language itself, and hence criteria internal to the system cannot evaluate the question.
According to Carnap, the answer to the external question about the thing world itself is "a matter of decision" (Carnap 1950, 207). Of course we do not choose our system of thing language when we learn, but we can choose to continue to use it or not. As the phrase 'a matter of decision' implies, the choice is somewhat arbitrary. Carnap says: "an alleged statement of the reality of the system of entities is a pseudo-statement without cognitive content" (Carnap 1950, 214). Here Carnap clearly appeals to VP. As the above version of VP shows, a proposition has an empirical consequence (e.g. partial interpretation) only relative to other propositions (e.g. subsidiary hypothesis) in the framework. On the other hand, almost by the definition of a 'framework', a theoretical framework itself does not have such outside propositions to be used for the evaluation of it. Therefore the question about the framework itself is not cognitively significant. Then, how can we choose between frameworks? According to Carnap, in this case the criteria for judgement are expediency, fruitfulness and conduciveness to the aim for which the language is intended (214).
As Feigl (1950) points out, this last point makes the choice circular. When we appeal to expediency, fruitfulness and so on to justify a linguistic framework (Feigl call this "vindication"; 116), how do we judge the fruitfulness of the framework? Obviously we have to rely on our knowledge about the world for such a judgement, and such knowledge is obtained by answering internal questions. But, in turn, the answers to internal questions are relative to the linguistic framework. Therefore the choice is circular. Feigl maintains that the circularity is not vicious, however. According to him, a circular argument may not be vicious if its conclusions does not appear literally among the premises (144).
Let us briefly compare the Carnap-Feigl argument with Austin's argument in "Other minds" (Austin 1946). Austin argues that we are entitled to a knowledge claim when we eliminate all specific doubts about the claim (the so-called special reasons requirement; Cherniak 1986, 100). For example, we are entitled to say 'it is a gold finch' when we show that the object is not a gold crest, it is not stuffed, and so on. Cartesian skepticism about the external world itself is irrelevant to our ordinary discourses, since we do not have specific doubts about it.
Does Austin's argument avoid circularity? Austin seems to deny the idea that we have choice about linguistic framework. I agree with Carnap that we have a sort of choice about this, even though I also think that the range of choice is not very wide (this point is related to the argument in the next section). Anyway, if we have the choice of framework at all, Austin's argument also gets into the same circularity as Carnap's one. In this sense, Austin's argument is no better than Carnap-Feigl's argument.
Another point of comparison is to look at VP and the special reasons requirement as two incompatible basic principles. From Carnap-Feigl's point of view, the special reasons requirement may not be specific enough to rule out the question about the external world itself. To see this, imagine that we admit that the doubt that "maybe there is a Cartesian Demon" as a legitimate specific doubt, and that we improvise a criterion to settle the doubt (maybe we agree that a coin toss is a legitimate way to decide if such a demon exists). This makes the question a legitimate one from Austin's point of view. From VP's point of view, however, the doubt about the Cartesian Demon cannot be a meaningful one, because there is no empirical procedure to settle the matter. Which is the better strategy? The answer depends on our evaluation of the status of VP (and the special reasons requirement) in epistemology.
3. Status of Verifiability Principle in epistemology
According to Feigl (1950), VP itself is a part of the circular argument, along with other deductive and inductive logic. The reason we accept VP is a practical one, namely to avoid "endless perplexity and vexation with pseudo problems" (134). And the judgement of usefulness of VP for such a purpose depends on our answers to 'meaningful' questions, classified as such by VP itself. This is circular. But Feigl maintains that fundamental rules like rules of logic or VP have some "primacy" to other parts of the circle (147). When we vindicate these rules, we are not just picking up rules for purely practical considerations, but we are explicating "what in some fashion we have known all along -- implicitly" (145). Since 'what we have known' is somewhat fixed, we can use the fixed point to start the circle. In this sense, "there can be no alternative logics in the sense in which there are alternative systems of geometry or ethics" (147; emphasis in original).
What does Feigl mean by 'explication'? Carnap's (1962) argument might help. Carnap defines explication as "transforming a given more or less inexact concept into an exact one" (Carnap 1962, 3), and summarizes four characteristics of explication: 1. the explicatum (what explicates) is to be similar to the explicandum (what is explicated), even though a considerable difference is permitted; 2. the explicatum is to be characterized in an exact form; 3. the explicatum is to be a fruitful concept; 4. the explicatum should be as simple as possible (7). Thus, the fixedness of explication is assured by characteristic 1 and practical considerations come in by 2-4.
I think that the idea, that there is some fixed point we cannot freely change in our cognitive system, is intuitively appealing. Then VP and the special reasons requirement can be seen as two different explications of our basic notion of 'meaningfulness'. Does VP explicate the fixed point better than the special reasons requirement? Among above four criteria, the version of VP seems more exact than Austin's argument for the special reasons requirement, but nothing prohibits such an articulation of the special reasons requirement. As for simplicity, VP seems more complicated than the special reasons requirement, but if the special reasons requirement needs more articulation, this does not matter. Fruitfulness is hard to judge, and heavily dependent on what we require of epistemology. If we want concrete normative directions in epistemology, VP may be more concrete than the special reasons requirement in its normative content. But other kind of requirement in epistemology may make the special reasons requirement more fruitful. Thus these practical considerations do not provide a decisive preference between these two alternatives. Then what about the criterion 1, the similarity to 'what we have known'? Here we inevitably face the problem of circularity again. This is partly a psychological question of what we believe, but a psychological statement perfectly legitimate under the special reasons requirement can be a meaningless statement under VP, and vice versa. Of course this does not necessarily mean that those who support VP and those who support the special reasons requirement never agree with each other on this matter. But we cannot expect that people will reach an agreement eventually, either. Anyway, this is enough to show that Feigl's argument that VP is an explication does not help very much in defending VP against the special reasons requirement. Therefore the priviledged status of VP which the verificationists like Carnap and Feigl give to it is not warranted.
4. Concluding remarks
In this paper, first I set up a version of VP which seems to me the most satisfactory. The major weakness of the version of VP is that it provides only necessary condition for meaningfulness. Actually this weakness does not matter when we use VP to analyze what Carnap calls external questions, including the question about the existence of physical objects, because these questions apparently do not meet VP. But when we compare the argument from VP with other arguments like Austin's one, the status of VP in epistemology becomes problematic. Feigl tries to defend VP by saying that it is an explication of our basic principle which we can not freely change, and this argument has some appeal. But again whether VP is a better explication than other alternatives like the special reasons requirement is an open question. Therefore, the status of VP in epistemology is at best a promising candidate for the explication of the basic epistemological principle (if any) we actually use in our cognitive system.
 In Hempel's paper, the condition (b) is used not as a criteria of cognitive significance of a proposition, but as a part of the criteria of cognitive significance of a theoretical system. See Hempel 1965, 113-117.
 Hempel tries to rule out this possibility with more elaborate criteria, but his strategy is not available to my version because Hempel assigns cognitive significance to each theoretical system, while my version assigns it to each proposition. See Hempel 1965,116.
 As I have mentioned in above footnotes, Hempel tries to attribute cognitive significance to each theoretical system, not to each proposition. But this attempt is not really relevant to Carnap's worry about external questions. Hempel's argument is devoid of the way to compare two 'cognitively significant' theoretical systems, and this comparison is what Carnap is talking about.
 Let us ignore the reflexive question of whether this is a good explication of the concept of 'explication' itself, for the sake of argument. See Suppe 1977, 57-60 for a critical analysis.
 This appeal comes partly from the worry that if we deny this our cognitive system collapse into a chaos. Those who use Neurath's boat metaphor would refuse such a worry, and this makes Feigl's argument about explication pointless. However, this problem is out of the scope of this paper.
 A meaningful statement under VP can be illegitimate under the special reasons requirement if the statement is testable only in principle, and we can hardly imagine how to test the claim.
Austin, J. L. (1946) "Other minds", reprinted in J. Urmson and G. Warnock eds., Philosophical Papers (1979). New York: Oxford University Press.
Carnap, R. (1950) "Empiricism, semantics, and ontology", reprinted in Meaning and Necessity enlarged edition (1956). University of Chicago Press.
--. (1962) Logical Foundations of Probability second edition. Routledge & Kegan Paul: The University of Chicago Press.
Cherniak, C. (1986) Minimal Rationality. Cambridge: The MIT Press.
Feigl, H. (1950) "De Principiis non disputandum ...? On the maning and the limits of Justification", reprinted in M. Black ed. Philosophical Analysis (1963).
Hempel (1965) "Empiricist criteria of cognitive significance" in Aspects of Scientific Explanation. Free Press.
Suppe, F. ed. (1977) The Structure of Scientific Theories, 2nd ed. University of Illinois Press.