Coherence, Certainty, and Epistemic Priority*

Roderick Firth
Harvard University

The Journal of Philosophy, Vol. LXVI, No. 19 (October, 1964): pp. 545-557.

Near the end of' his annual lectures on epistemology at Harvard, Lewis used to tell his students that they must ultimately choose between the theory of justification that he had been defending or something very similar to it -- and a coherence theory like that of Bosanquet. These two alternatives may not seem to confront each other quite so directly in Lewis's books on epistemology, but in his paper entitled "The Given Element in Empirical Knowledge"1 he again offers us this same choice. He explicitly defends his own theory of the given as one of "two alternatives for a plausible account of knowledge" (168), the other alternative being an "unabridged probabilism" like that of Reichenbach -- "a modernized coherence theory" (171). Although "logical and systemic relationships are important for assuring credibility." such a "probabilistic conception" of knowledge is incompatible with the fact that "no logical relationship, by itself, can ever be sufficient to establish the truth, or the credibility even, of any synthetic judgment" ( 169). "Crudely put," Lewis asserts, it
. . . strikes me as supposing that if enough probabilities can be got to lean against one another they can all be made to stand up. . . . . I think the whole system of such could provide no better assurance of anything in it than that which attaches to the contents of a well-written novel (173).

The issue outlined here is apparently one which Lewis continued to take very seriously, for in a letter written to me as recently as three years ago he expressed the fear that contemporary philosophers are, in his words, "headed back toward Bosanquet." It seems especially appropriate. therefore, in this memorial symposium, to consider some of the problems that arise when we attempt to formulate a precise definition of the issue Lewis had in mind. What kind of coherence theory did Lewis want to avoid. and what are the alternatives to it? I am convinced that we can ask no questions more important than these if we wish to understand Lewis's philosophical motivation and the full implications of his theory of knowledge.

It is possible to distinguish at least three theories (or perhaps I should say three types of theory) which can appropriately be labeled "coherence theories," and which can be defined, I believe, in such a way that none of them logically entails either of the others. These are (1) the coherence theory of truth, (2) the coherence theory of concepts, and (3) the coherence theory of justification. I shall say nothing about the coherence theory of truth except that the arguments offered in its support all seem to me to presuppose the coherence theory of justification. And I shall comment on the coherence theory of concepts only to suggest that it might be quite acceptable to Lewis even though he rejects the coherence theory of justification. It is clearly this last theory, the coherence theory of justification, which Lewis is primarily concerned to refute in his epistemological writings.


The coherence theory of concepts is the doctrine that all our concepts are related to one another in such a way that we cannot be said fully to have grasped any one of them unless we have grasped all the others: they form an organic conceptual scheme, it is said, system of meanings which cohere in such a way that introducing a new concept at any one point in the system has repercussions which are felt throughout the system. It is easy to illustrate this doctrine by restricting it to the technical concepts of some particular science and tracing the changes produced by the introduction of a new concept of space. matter or energy. But the broader implications of the doctrine can better be suggested by appealing to some commonplace concept like that of "mirror image." It can plausibly be argued that the young child who has not yet acquired the concept of "mirror image" cannot yet use the words 'see', 'touch', 'same', 'real', 'thing', 'space', 'colored', 'myself', 'left' and right' -- or perhaps any words in his vocabulary -- to mean quite what they mean to his older brother. And once the pattern of this argument has been accepted it can easily be extended to any other concept we may select.

Now it might seem at first thought that this coherence theory of concepts is incompatible with Lewis's analysis of the "sense meaning" of statements about physical objects -- and incompatible, indeed. even with the more moderate view of Locke and many other philosophers that some material- object predicates (e.g., "red") can be analyzed by means of supposedly simpler predicates (e.g. "looks reds") which we use to describe sense experience. For if a philosopher maintains that "The apple is red" can be analyzed as meaning "The apple would look red under such and such physical conditions," he is assuming that "looks red" is logically prior to "is red," i.e., that it is at least logically possible to have the concept "looks red" before we acquire the concept "is red." But if the coherence theory of concepts is correct and we cannot fully understand "looks red" unless we possess the contrasting concept "is red," then it would seem that it is not logically possible to have the concept "looks red" before we have the concept "is red." This paradox might even lead us to wonder indeed whether the conceptual interdependence of "looks" and "is" is enough to undermine Lewis's basic assumption that we can make "expressive judgments" (e.g.. "I seem to see a doorknob," "It looks as if I am seeing something red") without at the same time asserting something about the nature of "objective reality." It is these expressive judgments according to Lewis that enable us to escape the coherence theory of justification; and if it should turn out that these judgments all make some covert reference to physical objects, then -- depending, of course, on the kind of "covert reference" -- it might no longer be possible to make the epistemological distinction that Lewis requires.

There are many subtle facets of this question which cannot he explored here, but for our present purpose it is sufficient to point out that the underlying paradox is easily dissolved if we do not confuse concepts with the words used to express them. It is a genetic fact, but a fact with philosophical implications that when a child first begins to use the word 'red' with any consistency he applies it to things that look red to him whether these things are as we should say, "really red," or whether they are merely made to appear red by abnormal conditions of observation. Thus the child calls white things "red" when he sees them through red glass. In fact at this stage the child says 'red' just in those circumstances in which we, as adults, could truthfully says "looks red to me now," so that it would not be unreasonable to assert that the child is using 'red' to express a primitive form of the concept "looks red." To call this a "primitive form" of the concept "looks red" is to acknowledge that in some sense the child cannot fully understand adult usage until he is able to distinguish things that merely look red from things that really are red; but we must not suppose that the child somehow loses his primitive concept when he acquires a more sophisticated one. As Lewis points out in Chapter III of Mind and the World Order,2 the scientist and the nonscientist are able to share what Lewis calls "our common world" precisely because the scientist does not necessarily forget how to use words in their nontechnical senses; and for the same reason there is no inconsistency in maintaining that even as adults we continue to have a concept "looks red" which is logically prior to our concept "is red."

To grant Lewis this crucial point is not to deny that the vocabulary of "looks" and "seems" expressions that we use to describe sense experience is in some respect derivative from the vocabulary that we use to describe the physical world. Thus when Lewis describes his sense experience by saying "I seem to see a doorknob" his choice of words appears to reflect a linguistic rule to the effect that a sense experience should be "named after" its normal condition (in this case the condition of actually seeing a real doorknob). But such a rule like the rule in some societies that sons should he named after their fathers is merely a baptismal rule. The fact that Young Rufus is named after Old Rufus does not prevent us from learning to recognize Young Rufus before we have met Old Rufus. Analogously, the fact that key words in our "looks" and "seems" expressions are inherited from our "is" expressions does not prevent the child from consistently identifying things that look red to him (or situations in which he seems to see a doorknob) before he can consistently identify things that are red (or situations in which he really sees an "objective" doorknob). If we do not confuse baptismal rules with semantical rules (e.g., the semantical rule followed by the child who says "red" when something looks red to him) the coherence theory of concepts does not seem to be incompatible with Lewis's theories of meaning and knowledge. Let us turn, therefore, to the coherence theory of justification.


Philosophers have sometimes construed the problems of justification as though they were problems concerning the knowledge possessed by a social group; and it does of course make perfectly good sense to ask what statements we (e.g., you and I, our "culture circle," etc.) are justified in believing, and why we are justified in believing them. But Lewis seems clearly to be right in maintaining that such a question cannot be answered without first answering a more fundamental, egocentric question: Why am I, at the present moment justified in believing some statements and not justified in believing other statements? This is to be interpreted as an epistemological question not as an ethical question to which someone might in principle reply: "Because you will be happier (or more loyal to your friends) if you believe these statements rather than those"; and the ambiguous expression 'justified in believing' is to be interpreted so that we may assert without self-contradiction that someone is justified in believing a statement he does not in fact believe. It is helpful, therefore, to reformulate the question as a question about the "epistemic warrant" (or, for short, "warrant") that statements have "for me" at a particular time; and in these terms I think that the heart of the coherence theory of justification, as Lewis probably construes it, is the thesis that ultimately every statement that has some degree of' warrant for me has that particular degree of warrant because, and only because, it is related by valid principles of inference to (that is to say "coheres with") certain other statements.

To explain why I have used and italicized the word 'ultimately' in formulating this central thesis of the coherence theory, and to facilitate comparison with alternative theories of justification, it is helpful to refine the issue still further and construe the coherence theory as an answer to the question: What properties or characteristics of a statement may serve to increase its warrant? This question may in turn be formulated in a slightly different way by employing the term 'warrant-increasing property', interpreted so that, in saying that a statement S has a warrant-increasing property P for a particular person at a particular time, we imply that S would be less warranted, and that not-S would be more warranted, for that person at that time, if other things remaining the same, S did not have property P. In this terminology the question becomes: What properties of statements are warrant- increasing properties?

It is clear that advocates of the coherence theory would want to reply that if P is a warrant-increasing property of statement S, P might consist simply in S's being validly inferable from certain other statements of a specified kind. In such a case since the warrant of' S is increased, so to speak by the fact that S is validly inferable from certain other statements, P might appropriately be called an "inferential" property. Advocates of the coherence theory would surely be willing to grant, however, that there are noninferential properties (e.g.. the property of being believed by scholars with such and such characteristics) which might also increase the warrant of a statement; but to preserve coherence as the ultimate court of appeal they would insist that such a noninferential property (P') can be a warrant-increasing property of a statement S only if a particular statement about S -- the statement, namely, "If S has property P then S is true" -- is validly inferable from (coheres with) certain other specified statements.3 (This requirement might be met, for example, if P' were the property of being believed by certain scholars and if there were evidence that these scholars have usually had correct beliefs about statements similar to S in certain respects.) Thus we may say that the coherence theory of justification maintains that, if P is a warrant-increasing property of S, then either (1) P is an inferential property, or (2) P is a warrant-increasing property only because the statement "If S has the property P then S is true" has an inferential warrant-increasing property. It is convenient to summarize this by saying that all warrant-increasing properties, according to the central thesis of the coherence theory, must be "ultimately inferential."

To convert this central thesis into a fully determinate coherence theory it would have to be elaborated in two ways. (1) We should have to specify the "valid" principles of inference, deductive and inductive, that determine whether one statement coheres with, and thus confers warrant on another. And (2) we should have to specify the nature of the "certain other statements" with which a warranted statement must cohere -- the class of statements that are, we might say, "warrant- conferring." Although the problems involved in (1) are very important -- especially those which arise when we ask whether there is a set of principles of inference, and only one set, that can be selected and justified by reapplying the same standard of coherence -- these problems are neutral with respect to the central issues at stake between Lewis and the coherence theory. The problems involved in (2), however, are more directly relevant to these central issues, and we shall return to them after considering Lewis's position.


In clear opposition to the coherence theory of justification Lewis flatly denies that all warrant-increasing properties are ultimately inferential. It is a matter of some importance, which we shall consider later, that Lewis often discusses the problems of epistemic justification as problems concerning judgments, and may thus be restricting his attention to the epistemic status of statements that are actually believed (judged to he true) by a particular person at a particular time. But in any case he maintains that those statements which do reflect my present judgments about my own present experience -- including statements about sense experience, memory experience, occurrent feelings, etc. -- are certain (and hence warranted) for me at the present time, and that their certainty is not derived directly or indirectly from their coherence with other statements. There is room for debate, however, about the meaning of the word 'certain' in this context, and I think that Lewis's writings actually suggest several different alternatives to the coherence theory of justification.

There are a number of passages in Mind and the World Order, in An Analysis of Knowledge and Valuation,4 and elsewhere, in which Lewis says that "expressive judgments" (e.g., "I seem to see a doorknob") cannot be mistaken. "One cannot be mistaken," he asserts, "about the content of an immediate awareness" (MWO, 131). This is perhaps the most extreme alternative to a coherence theory of justification, and is often taken to be the only alternative that Lewis offers us. In another place in Mind and the World Order, however Lewis says interestingly enough: "All those difficulties which the psychologist encounters in dealing with reports of introspection may be sources of error in any report of the given. It may require careful self- questioning, or questioning of another, to elicit the full and correct account of the given" (62). This of course seems to imply that, in some important sense of 'can', our expressive judgments can be mistaken, and it suggests the need to distinguish what we might call "truth- evaluative" senses of 'certain' from "warrant-evaluative" senses of 'certain'. To say that a judgment is certain in a truth-evaluative sense of the word entails that the judgment is true, but to say that a judgment is certain in a warrant-evaluative sense is merely to say that the judgment (whether it be in fact true or false) is completely warranted in some specifiable sense of 'completely'. Although Lewis does sometimes assert that expressive judgments cannot be false, I believe that all the arguments he gives for the certainty of expressive judgments are arguments to show that these judgments are certain in a warrant-evaluative sense. Indeed Lewis sometimes uses the words 'indubitable' and 'incorrigible' as synonyms of 'certain', and these two words are more naturally interpreted as warrant-evaluative than as truth-evaluative. There is no logical inconsistency in asserting that someone has a false belief which he cannot rationally doubt and which he is not in a position to correct; consequently there is no inconsistency in asserting that expressive judgments are indubitable and incorrigible, while at the same time granting that some of them may be false.

In defending the doctrine that expressive judgments are certain in a warrant-evaluative sense, Lewis sometimes tries to prove much more than is necessary to refute the coherence theory of justification. Some of his arguments are apparently intended to show that, if I now judge, for example, that it looks as if I am seeing something red, I shall never at any time in the future be justified in revoking this judgment. But problems concerning the future revocation of an expressive judgment, at a time when my decision must depend in part on my memory of my present experience, are not directly relevant to the question: Are my present expressive judgments certain (and hence warranted) for me now? And, if they are, is their warrant derived entirely from coherence? I believe that Lewis's answers to these two questions are (1) that my present expressive judgments, being certain, are not only warranted for me but warranted to so high a degree that no other judgments are more warranted for me, and (2) that their, warrant is not derived to the slightest degree from coherence nor defeasible through failure to cohere with other judgments. "There is no requirement of consistency," Lewis asserts, "which is relevant to protocols."5

Again, however, it is important to observe that there are at least three weaker, and therefore perhaps more plausible positions that are also incompatible with the coherence theory of justification as we have been construing it. It might be maintained (1) that the warrant of an expressive judgment may be increased by its coherence with other judgments, and to some extent decreased by failure to cohere, but that failure to cohere can never decrease its warrant to a point at which a contradictory judgment would be more (or even equally) warranted. This would allow us to say that my present expressive judgments, although they may be false, are not now falsifiable for me. Or it might be maintained (2) that present expressive judgments, although by falsifiable by failure to cohere, always have some degree of warrant which is not derived from coherence and which is not defeasible through failure to cohere. Or, even more moderately, it might be maintained (3) that expressive judgments have some degree of "initial" non-inferential warrant which is defeasible through failure to cohere -- perhaps even allowing, in principle, for the possibility that the contradictory of an expressive judgment may be as fully warranted as any other empirical judgment.6 Although Lewis's strong position and each of these weaker positions differ markedly from one another, each of them entails a proposition which we may call "the central thesis of epistemic priority" -- the thesis that some statements have some degree of warrant which is independent of (and in this sense "prior to") the warrant (if any) that they derive from their coherence with other statements. If we decide that there are statements of this kind, our next task is to determine what warrant-increasing property these statements have in addition to properties that are ultimately inferential.

In considering this problem there is a strong temptation for those who accept the thesis of epistemic priority to say that the statement (for example) "It looks as if I am seeing something red" is warranted (or given some warrant) for me simply by the fact that it does look as if I am seeing something red; but to say this seems to imply that the statement is warranted because it is true -- because it asserts what is in fact the case. To preserve the important distinction between truth and warrant, so that in principle any empirical statement may be true but unwarranted, or false but warranted, it would be preferable to maintain that the statement has a certain degree of warrant for me because it is a statement (whether true or false) that purports to characterize (and only to characterize the content of my present experience. (This could of course be made more precise by the use of examples and other devices.) But this condition is clearly insufficient, for we should not want to hold that all statements, including all possible pairs of contradictory statements, have some degree of warrant if they satisfy this requirement.

The obvious way to meet this difficulty is to add the further condition that a statement about my present experience can have some degree of ultimate noninferential warrant for me only if I believe it to be true. This condition is suggested as we have already observed, by Lewis's use of the word 'judgment' and I am inclined to think that Lewis would consider these two conditions, taken together, to constitute a sufficient condition of epistemic priority. If he were also to maintain that these two conditions are necessary, it would not be inappropriate to say that for Lewis expressive judgments are "self-warranted" (perhaps even "self-evident"), implying by this, so to speak, that for a statement about my present sense experience, its being now judged by me to be true is an ultimate warrant-increasing property. Because of the ambiguity of the word 'judgment', however, it is unclear to me whether Lewis would actually consider this second condition (viz. that statements with noninferential warrant must be believed) to be necessary. There are of course many statements which are warranted for me, which I am justified in believing, but which I do not in fact believe; and a philosopher who accepts the thesis of epistemic priority might maintain that among these statements are some that are ultimately warranted, at least in part, noninferentially. Presumably, however, these noninferentially warranted statements would all be statements that I would now believe if I had just decided whether they were true or false; and thus it would probably be close to the spirit of Lewis's position to maintain that in the last analysis a statement can now have for me only one warrant-increasing property that is not ultimately inferential -- that compound property which consists in (1) purporting to characterize (and only to characterize) the content of my present experience, and (2) being a statement that I either now believe to be true or should now believe to be true if I had just decided whether it were true or false.

This formulation of a possible theory of epistemic priority raises a number of important and puzzling questions. We might wonder, for example, whether (2) should include some restriction on the method by which I arrive at my belief, and whether it is possible to formulate such a restriction without circularity. And we might wonder whether the strength of my belief (the confidence with which I hold it) does not have some role in determining at least the degree to which a statement has a warrant that is ultimately noninferential. Within the limits of this paper, however, I can make only a few concluding remarks about Lewis's criticism of the coherence theory, in particular about his statement already quoted, that "no logical relationship, by itself, can ever be sufficient to establish the truth, or the credibility even, of any synthetic judgment."


This statement, which I think represents the crux of many familiar arguments against the coherence theory, seems to me to reflect a conception of the coherence theory which is unnecessarily narrow and much too narrow to make the theory at all plausible. It is sometimes said (cf. AKV, 340) that the coherence theory provides us with no way of distinguishing the actual world from other "possible worlds," since statements describing any of these worlds will form equally coherent systems; and this seems also to be Lewis's point when he says that a system of statements that stand only because they "lean against one another" gives us "no better assurance of anything in it than that which attaches to the contents of a well-written novel." As we have observed, however, a philosopher who accepts what we have called the "central thesis" of the coherence theory is not thereby committed to any particular way of identifying the class of "warrant-conferring" statements with which any warranted statement must ultimately cohere. If he insists that the power to confer warrant resides only in warranted statements, and that warranted statements constitute a perfectly democratic society in which each member receives its warrant from coherence with all the others, then indeed he will not be able to explain why one system of coherent statements is warranted and another is not. But this difficulty can be avoided if he adopts a less democratic position and recognizes an elite class of' "basic" warrant-conferring statements which, although it may include some statements that are not warranted, excludes a great many statements that are warranted. If he can identify this class by reference to something other than mere coherence, he may be able, so to speak, to tie the entire set of warranted statements to the possible world in which we actually live. In fact he would then be in a position to agree with Lewis, without giving up the coherence theory, that no logical (inferential) relationship by itself, "can ever be sufficient to establish the truth or even the credibility, of any synthetic judgment."

There are many interesting ways in which we might delimit such a class of basic warrant-conferring statements for a particular person at a particular time, but perhaps the traditional and most plausible way is to restrict this class to statements, whatever their logical form or subject matter, that are actually believed by that person at that time.7 If 'believed' is interpreted liberally enough so that this class includes a large number of very general theoretical statements, there is some grounds for holding that the inferential relationships among them - - and at some points the lack of any inferential relationship -- are sufficient to determine which statements are warranted, which statements are not, and the relative degrees of warrant among them.8 Within this elite class, so to speak, each statement, whether itself warranted or not, would have a voice in determining the epistemic status of every other statement in the class. And statements outside the class -- statements which have not yet been thought about, or which, for some other reason are neither believed nor disbelieved -- could be said to be "derivatively" warranted if in fact -- whether anybody knows it or not -- they cohere with the warranted statements in this warrant-conferring class. This would mean, in effect, that these derivatively warranted statements are second-class citizens: they receive warrant from members of the class of basic warrant-conferring statements (and are thus tied down to the actual world), but they have no independent authority in determining whether any other statement is warranted.

A position of this kind seems to me to avoid Lewis's logical objection to the coherence theory of justification and thus to demonstrate that the issue between this theory and the thesis of epistemic priority must ultimately be decided on purely empirical grounds. It is of course difficult to formulate precise criteria for settling such an issue, but advocates of the coherence theory have commonly tried to defend their position by appealing to the actual practices of scientists and other rational men, and presumably these practices are relevant to the issue even if not absolutely decisive. On this basis I think it would be very difficult to defend Lewis's strong position that some statements are certain in a sense that makes coherence completely irrelevant to their warrant. But I think on the other hand that rational men often believe statements about their own sense experience with much greater confidence than they could justify by inference from other beliefs; and this suggests that we may accept the thesis of epistemic priority and try to choose among the three weaker positions that entail this thesis. It has been my intention in this paper to formulate an issue and not to defend this particular conclusion. But if the thesis of epistemic priority is, as I think, correct, the methodological consequences are momentous whether or not we accept Lewis's doctrine of certainty. For at least we can say in that case that Lewis has always been right in maintaining that the major task of a theory of empirical knowledge is to show how it is possible -- by means of a theory of meaning and suitable principles of inference - - for statements that have independent noninferential, warrant to serve as the ground of all the rest of our empirical knowledge.


* To be presented m a commemorative symposium on C. I. Lewis, at the sixty-first annual meeting of the American Philosophical Association Eastern Division, December 28, 1964. A revised and extended version of a paper presented in the Lewis Memorial Symposium, Harvard University, April 23, 1964.

1 In a symposium with Hans Reichenbach and Nelson Goodman, in The Philosophical Review, 61, 2 (April 1952): 168-175.

2 New York: Scribner's, 1929; hereafter referred to as MWO.

3 For simplicity I assume that the noninferential warrant-increasing properly P' is only one step removed from the ultimate appeal to coherence; but in principle there might he a long intervening chain of noninferential warrant increasing properties.

4 La Salle, Ill.: Open Court, 1964; hereafter referred to as AKV.

5 "The Given Element in Empirical Knowledge," MWO. Italics mine.

6 Lewis himself defend a position analogous to (3) with respect to present memory judgments about the past -- as opposed to present judgments about experiences (including memory judgments) occurring in the present (AKV, 354 ff.) H. H. Price's "Principle of Confirmability" represents an analogous position with respect to judgments about presently perceived material things; see Perception (New York: McBride, 1933). p. 185. But "initial" warrant is derived by Lewis and Price from an "assumption" or "principle,". and can thus be construed as "inferential."

7 Cf. Brand Blanshard, The Nature of Thought (New York: Macmillan, 1940), vol. II, p. 272: "What the ultimate standard means in practice is the system of present knowledge as apprehended by a particular mind."

8 The rules that would determine these things might be similar to those proposed by R. B. Brandt for the selection of warranted memory beliefs (recollections) in his "Memory Beliefs," The Philosophical Review, 64, 1 (January, 1955): 88. Brandt's rule, however, "advises accepting recollections when there is no positive support from the system" and is thus compatible with the thesis of epistemic priority.

Transcribed into hypertext by Andrew Chrucky, Sept. 14, 1997.