Laurence BonJour, Knowledge, Justification, and Truth



8. The Problem of Empirical Truth.

To summarize briefly the position at which we have arrived: I began by setting forth the familiar classical definition of knowledge as (i) belief, which is (ii) true, and (iii) justified. I then focused for two chapters on the problem of justification. The upshot of that discussion was that any attempt to justify a claim within a conceptual system in terms of something non-conceptual, as in the doctrine of the Given, must fail, roughly because the outside justifying factor, whatever it may be, must not merely be present in the knowing situation, but must be apprehended as being present. And any apprehension of it which could serve the purpose of justification must seemingly be itself a conceptual apprehension, i.e. an apprehension involving inference relations to, at the very least, the claim which it is to justify. But then the new apprehension is itself a conceptual claim, requiring justification, and the attempt to justify a conceptual claim from outside the conceptual framework has failed. From this I concluded that the only alternative to a skeptical rejection of the very possibility of justification, and thus also of the very possibility of knowledge, is a coherence theory of justification. I then tried to sketch the broad outlines of such a theory, emphasizing the way in which even observation statements are justified by considerations of coherence.

Having attended as well as I can to the issue of justification, it is time, I think, to return to our original triadic account of knowledge to see what becomes of the remaining two components if this account of justification is accepted. There seems to be no special need here for a discussion of belief; although the problems connected with the concept of belief are legion, it does not seem that they particularly affect my present epistemological concerns. With respect to truth, on the other hand, the situation seems at least to be quite different. Indeed the apparent effect of our discussion of justification is to at once raise anew and seriously aggravate the classical problem of truth (or, perhaps better, one amongst the classical problem).

It may be well to emphasize here that I am concerned in the present essay only with empirical knowledge, and thus in the present context only with empirical truth. With this restriction in mind, the classical problem of truth can be stated, in first approximation, as follows. Consider two empirical beliefs, that-P and that-Q, both justified at some time t, but the former in fact true and the latter in fact false. What exactly are we affirming of the first belief and denying of the second by saying that the one is "true" and the other is "false"? Clearly we normally want to have true rather than false beliefs, and indeed the whole elaborate apparatus of justification seems obviously to be aimed at just this result. But what result exactly? What is this "truth" which human cognitive activity is designed to achieve? There are, of course, many philosophers nowadays who refuse to take questions of this vague and general sort seriously, and who would regard my raising this one anew as a case of the most extreme sort of naivete. And such a view may well turn out to be right. But naivete, and particularly philosophical naivete, is by no means always fruitless, and I shall persist in mine for a while.

The response of the man-in-the-street to the question just adumbrated, to the extent that he can be brought to understand it, is likely to be precisely the opposite of that of the contemporary philosophers just alluded to. For common sense, what is likely to seem wrong with our question is not that it is a mistake to ask it, but rather that the answer is just too obvious to require more than a moment's consideration. What our man-in-the-street wants to say, of course, is that a belief is true just in case the content of the belief is an accurate representation of the way the world really is; i.e. just in case what is believed "corresponds to the facts." The difference between our true and false beliefs is that the former correspond to the facts while the latter do not. Thus the "correspondence theory of truth."

As is so often the case with such pronouncements of (sophisticated) common sense, it is most difficult to see how the correspondence theory can fail to be correct. But, as is also so often the case, there are problems with such a view of truth which have proved to be extremely resistant to solution. These problems arise as soon as one tries to say anything very specific about either the relata of the supposed relation of correspondence, or about the nature of the relation itself. To take the latter first, there has always been a great temptation to regard the relation of correspondence as involving somehow a fitting together or congruence of belief and fact, where such a congruence is to be understood as involving more than a merely conventional correlation. This sort of inclination is, I think, one motive behind familiar image theories of thought, characteristic especially of the British Empiricists. But every attempt to go beyond the metaphor of congruence or fitting together and give a precise characterization of such a relation runs aground on the hard fact that, once the sorts of confusion present in the image theory have been eliminated, there simply does not seem to be any clear sense in which beliefs or the sentences which express them are non-conventionally related to the world which they describe. And thus we are apparently left with a purely conventional correlation as the only explication of what is meant by "correspondence," with the disheartening result that the correspondence theory seems to say nothing more than that a belief is true if it is conventionally correlated with the world in the right way, i.e. if it is true. And these already formidable difficulties with the correspondence theory are aggravated by problems about the nature of the relata of correspondence. E.g., is the first relatum a belief? A proposition? A statement? And what is the second relatum? A fact? But which fact? How many facts are there? I shall not pursue these issues, nor attempt here any further account of the modern tribulations of the correspondence theory, but shall simply assume that all attempts to state a traditional correspondence theory have failed. Such a view seems to represent the consensus of contemporary opinion, though there are doubtless still some who would demur.{27}

These more or less familiar problems concerning the correspondence theory are aggravated in two ways by our discussion of justification. In the first place, the exorcising of the Given from epistemology also eliminates one possible version of the correspondence theory, which would take the relation of correspondence to exist primarily at the level of Lewis' "expressive" statements and to consist in a relation which might be described as "descriptive appropriateness" holding between the intended meaning of the expressive statement and the actually Given content, which relation is at once completely familiar, ineffable, and known immediately to hold, thus removing both the need and the possibility of further characterizing it.

Secondly, our account of justification in terms of coherence seems to lend impetus to the traditional rival of the correspondence theory, viz. the coherence theory of truth characteristic of absolute idealism. Thus it might be argued that once the correspondence theory has been abandoned, in the face of difficulties of the sort described above, the coherence theory is simply the clearest and most reasonable remaining alternative, especially once the holistic and temporally relative character of coherence, suggested above, is realized. Thus the suggestion would be that the belief that-P is true relative to the existing conceptual system insofar as it is part of the most coherent account of the world which is possible in that system, and true absolutely insofar as it coheres with that ideally coherent description of the world which would be given in the ideal conceptual system which is the goal of scientific advance. Such a view also makes possible a very simple account of the relationship between truth and justification: truth is simply justification in the long run. This explains why we seek truth by seeking beliefs which are justified.

This sort or view is obviously also intimately related to, if not identical with, the Peircean view of truth as relative to the final outcome of scientific inquiry. Nor should this be surprising, since the notion of conceptual development which we have seen to be involved in the notion of coherence is one of the most central ideas of pragmatism, as is the idea that justification is intimately related to action. Thus it would not be too misleading to say that the sort of view of truth just roughly sketched represents an Hegelian synthesis of central features of both of the classical rivals of the correspondence theory.

However this may be, the sort or account of truth just suggested seems to me, despite clear strong points, to have a serious deficiency which renders it most implausible from an intuitive standpoint. It seems clear enough that the notions of "ideally coherent account" and "final result of scientific inquiry" are intimately related to the notion of empirical truth. Certainly to have reason to think that one had arrived at such a coherent account or end of inquiry would be to have excellent reason to think that one had arrived at the "final" truth about the world; indeed, once the Given is abandoned it is hard to see how one could have any better reason for thinking that one had found the truth. But these other conceptions do not seem nevertheless to fully capture the concept of empirical truth, even if they are completely co-extensive with it. One way to see this is to ask why an ideally coherent account of the world is as coherent as it is, i.e. why anomalies or conflicts do not arise, or why the final result of scientific inquiry is indeed final, i.e. why no further progress is possible. I at least am very strongly inclined to say that the only answer to these questions which does not destroy the plausibility of even the claim that the notions in question are co-extensive with truth is "because such an account is true," where such an answer is not circular in the way in which it would be if our neo-coherence theory of truth were the correct one. What one wants to say is that an account of the world is true because it stands in a certain sort of relation to the way the world "really is," and that it is its standing in this relation which accounts for the fact that it is (ultimately) the most coherent account and also for the fact that it will or would (ultimately) be accepted by science. But then we are back with the correspondence theory, with all of the problems and unclarity attendant thereto.

A different way of putting the same point is to point out that although the notions of justification and truth must obviously be intimately linked with each other, it is a serious mistake to tie the latter too closely to the former. For if truth is simply equated with justification in the long run, as our neo-coherence theory of truth seems to do, it then becomes quite impossible to say why one should employ one set of standards for justification rather than another. The reason one wants to give is that one set of standards is more likely than another to yield at least an approximation to the truth, but if truth is simply justification in the long run, this reason collapses. Any standard of justification, no matter how absurd, will if consistently employed yield justification in the long run. So why not adopt whatever set of standards for justification will make the cognitive task the least arduous? In fact, why not simply equate justification with belief, or perhaps with belief by a majority if one wants an interpersonal standard, and let it go at that? Again the dimension of correspondence seems to be required if the cognitive enterprise is to be rationally intelligible.

Someone well versed in contemporary philosophy who has followed the discussion to this point is likely to remark that the reason for the present impasse is that I have simply overlooked the clearest and most cogent answer to the problem of truth, viz. the so-called semantic conception of truth developed by Tarski, Carnap, and others. Indeed many would say that to whatever extent there was a genuine problem of truth it was solved by Tarski, so that only confusions and pseudo-problems remain. In order to evaluate this claim, it is necessary to briefly examine just what the semantic conception of truth involves.

The basic idea of Tarski's account of truth can be stated as follows. Letting O be the object-language and M the meta-language, one attempts to give, in M, a definition of 'true-in-O' which has as deductive consequences all statements of the following form (T):

(T) X is true-in-O if and only if p.

where 'X' is replaced by a name or description of a sentence of O and 'p' is replaced by a translation of that sentence into M.{28} Now the significance of the achievement of constructing such a definition becomes clear when one reflects that although the number of sentences of O whose names and translations can be substituted into (T) is presumably infinite for any natural language, the definition of 'true-in-O' must be finite in length if it is to be stated at all. Thus such a definition would provide a finite account of the truth conditions of an infinite number of sentences. And the significance and appeal of the project of constructing such a definition is reinforced by Davidson's persuasive arguments to the effect that such a definition must be possible for any language which is learnable by a finite intellect, thus for all natural languages.{29}

Despite its obvious interest and importance, however, it does not seem to me that the semantic conception of truth helps in any way to solve the problem of truth with which we are presently concerned, viz. the problem of how a true empirical belief or statement is related to the world of which it is true. The key point to be noted is that what appears on the right-hand side of an equivalence of form (T) such as is a consequence of a Tarski-type truth definition is a translation of the sentence whose truth it is intended to explicate; in fact, in the case of a meta-language which contains its object-language as a sub-component, what appears on the right is just the object-language sentence itself. Thus such an equivalence seems to tell us only (i) that an object-language sentence is true if and only if its meta-language translation can be correctly asserted, i.e. is true, and (ii) what that translation is, where it may be just the sentence itself. Now (i) seems only to represent a necessary, though clearly not a sufficient, condition of adequacy for a translation; (ii) on the other hand conveys an important relation between the two languages. But it is hard to see that either (i) or (ii) says anything about the nature of truth.

That this view of the semantic conception of truth is not quite right is shown by the case in which the object-language in question is a language other than our own which we do not understand, and the meta-language is our language. For in this case the equivalences which are yielded by a Tarski-type truth definition do give us genuine and non-trivial information which is relevant to the truth of the object-language sentences. Thus to be told that, e.g.,

'Schnee ist weiss.' is true-in-German if and only if snow is white.

is, assuming that we do not already understand German, to be told something about the conditions for the sentence 'Schnee ist weiss.' being true. And a finite definition which can yield such information for all German sentences is of obvious utility in understanding German.

But are such definitions of any help in understanding truth, and especially empirical truth? After all, the sort of question about truth which I have tried to suggest in the earlier part of this section arises, if at all, just as much for familiar as for unfamiliar languages, i.e. just as much for a sentence like 'Snow is white.', which I understand, whose truth conditions I know very well, and indeed of whose truth I am quite convinced, as it does for a sentence like 'Schnee ist weiss.', for which none of these conditions may be true. The general problem is not that I do not know what the truth-conditions are for particular sentences or that I cannot in fact usually tell when particular sentences are true. It is rather that I am unclear about the relationship between empirical truths generally and the world of which they are true, though convinced for vague intuitive reasons that there must be something to be said about that relationship beyond a mere appeal to conventional correlation. And to be given a set of semantic equivalences of form (T), or even a finite statement of an infinite number of such equivalences in the form of a Tarski-type truth definition, as an answer to this problem is to quite literally be given no answer at all.

Notice that I am not claiming that a Tarski-type truth definition is trivial, even for a language which we already understand. It is not. But the reason that it is not trivial in such cases lies not in the semantic equivalences which are its consequences, but rather in the finite recursive machinery which serves to deductively generate those infinitely many equivalences. And however much that recursive machinery may say about meaning, logical form, deep structure, etc., it is hard to see how it says anything about truth, especially truth of the empirical variety.

The view that the semantic conception of truth does not help to solve the problem of empirical truth is reinforced by one further consideration, namely that whatever the semantic approach has to say about truth applies indiscriminately to all truth and says nothing distinctive about empirical truth. Thus an adequate Tarski-type truth definition for German will yield as consequences

(i) 'Zwei und zwei ist vier.' is true-in-German if and only if two and two is four.


(ii) 'Der Mord ist unmoralisch.' is true-in-German if and only if murder is immoral.

as well as

(iii) 'Der Stuhl in der Ecke ist braun.' is true-in- German if and only if the chair in the corner is brown.

The point is that anyone who wants to say that (iii) ( if true) corresponds to the world in a different or stronger sense, at least, than do (i) and (ii) (if true), will not be satisfied by the account offered by the semantic conception of truth. I think that we all want intuitively to say something like this, and that it is just this intuition which leads to the sort of problem about empirical truth which I have been discussing. I conclude that if our problem is a real one, if there is anything right in the classical correspondence theory of truth, the answer does not lie in the semantic conception of truth.

9. Truth and Picturing: a Simplified Example.

Having concluded as I have that the semantic conception of truth, whatever its virtues may be, does not provide an answer to the problem of truth sketched at the beginning of the preceding section, and in particular says nothing about empirical truth, as opposed to logical, mathematical, ethical, religious, or aesthetic truth, I shall seem to many to have backed myself into a dialectical corner from which there is no escapes. For I seem to have considered and rejected all of the available alternatives, thus providing excellent grounds for supposing that the initial question was in some way misconceived (though how exactly is difficult to say). Fortunately, however, this is not quite so, and before adopting the more or less desperate expedient of rejecting the question, I want to consider one more attempt at an answer, which differs significantly from those discussed above. As will surprise no one who has followed the essay to this point, the attempt I have in mind is that of Sellars.{30}

The central idea upon which Sellars' account of empirical truth rests is a firm distinction between what may be called the intentional order or the order of signification, on the one hand, and the real or natural order, on the other. The point here is simply that collections of utterances and inscriptions which from one standpoint are meaningful speech and convey information about the world, may from a different standpoint be considered merely as natural events, of a complicated sort and produced in complicated ways admittedly, but still basically on a par with other natural events in the world. Thus the assumption is that meaning, signification, etc., are in some sense derivative properties of linguistic events, and thus that there is a level of description at which those events can be described in purely non- intentional and non-semantic terms. (I shall use Sellars' term 'natural-linguistic object' to refer to utterances, inscriptions, etc., viewed in this non-intentional and non-semantic way.) I shall not attempt to defend this assumption, beyond remarking that besides having a definite intuitive plausibility it seems clearly basic to a broadly nominalistic philosophy; though in a certain sense the eventual outcome of the discussion will amount to at least a partial defense.

On the basis of this distinction Sellars claims to have isolated a dimension of correspondence between language and the world which is not purely conventional in character. This relation of, correspondence, or "picturing" as he calls it, is held to exist between those linguistic items which make up true empirical statements, considered merely as natural-linguistic objects in the world (though including their purely empirical, non-intentional relations to other such objects in the world, e.g. their manner of causal production by those complex natural objects which are in fact users of the language in question), on the one hand, and the objects in the world which those natural linguistic objects, or considered as part of the intentional order, would be said to be about, on the other. Thus Sellars is claiming that at least a necessary condition for language, qua language, to truly describe the world, is for the natural-linguistic objects which in fact constitute language to stand in this non-conventional relation of picturing to the objects which, qua language, they are about.{31}

Sellars' argument for this complex thesis is tortuous at best, and I shall not attempt an exegetical reconstruction of it. Rather I shall attempt to approach his position obliquely by considering a simplified example of my own, borrowing considerations from Sellars where they seem appropriate. In some ways, also, the account which emerges will be simpler than Sellars' own, and not necessarily in ways of which he would have approved. But my goal is still what might be described as a "rational reconstruction" of Sellars' view of empirical truth.

We may begin by formulating the issue as precisely as possible at this stage in the argument. It should be clear at the outset that the issue is not whether the relationship between true linguistic or conceptual items and the world of which they are true is conventional or non-conventional. That there is a very large element of convention involved in, e.g., the choice of natural-linguistic objects which are to play conceptual "roles," is something which no one could possibly want to deny. What is at issue is (1) whether there are any constraints of a broadly structural-syntactic sort on what sort of a conceptual framework can adequately represent a given sort of world, and (2) whether those constraints, assuming that there are any such, could plausibly be taken together as representing a requirement that the natural objects which serve as the elements of true assertions or beliefs in an adequate conceptual framework must stand in something like a relationship of picturing to the world which they truly describe. I want to argue, following Sellars, that an affirmative answer should be given to both of these questions.

It will help in considering these rather complex issues to have before us a simple example of a world and a conceptual system which attempts to describe it. Consider then the following simple world, which we may call "Simplia." Simplia is a three-dimensional world, two of these dimensions being spatial and one temporal. The only objects in Simplia are circular "patches" of uniform size. Each such patch is uniformly one or another of a small number of colors. These patches are in constant motion through the two-dimensional space, which we may take to be unbounded. In the course of these motions they collide, rebound, and interact in various other ways which we may suppose to vary with their colors. Suppose too that there is a cognitive being who employs a conceptual framework to observe Simplia, formulate explanatory laws, make prediction, etc. We may call both this conceptual framework and the language which embodies it "Simplese." In terms of this example, our two questions can be reformulated as follows: (1) are there any structural-syntactic constraints upon Simplese and its components deriving from the requirement that an adequate description of Simplia be possible in Simplese? and (2) do these constraints add up to a requirement that the natural-linguistic objects which function linguistically as true statements of Simplese stand in something like a picturing relationship to those objects of Simplia which, qua linguistic items, they describe? It will be useful to confine ourselves to these restricted versions of our general questions for the time being.

Our two questions, as just stated, are however seriously vague in an important respect, and require clarification before we can hope to make any real progress in answering them. This critical vagueness is located in the phrase "an adequate description of Simplia." Before we can proceed, we must have at least some inkling of what would be required for a description of Simplia to be adequate, or alternatively some idea of the respects in which such a description might fail to be adequate. This is a very large topic about which I can make only a few rather sketchy suggestions in the present context. I would submit, however, that at least an essential part of the notion of an adequate conceptual description lies in the idea that one basic goal of any cognitive activity is to represent particular states of affairs in the world. (I am not claiming here that this is the only goal of such activity, though in some sense I think that it is.) The point is that all the knowledge we might have of laws, generalizations, etc. no matter how elaborate, would be in one sense at least useless if it didn't enable us to respond to and predict particular states of affairs. Thus it would seem that a conceptual framework which did not contain the resources for (i) responding to particular states of affairs with singular observation statements, and (ii) generating other singular statements from these observation statements via its structure of laws, theories, etc., would be obviously inadequate in a most fundamental respect.

The foregoing observations are scarcely startling, and may well seem trivial to the point of uselessness. I want nevertheless to follow them up a bit in the hope that something less trivial will result. Applied to our specimen world Simplia and specimen conceptual framework Simplese, they involve the consequence that Simplese must make possible (instantaneous) description of the patches of Simplia in terms of such characteristics as color, direction and velocity of movement, etc. This in turn means at least that Simplese must contain means of referring to the patches of Simplia, and means of describing their various characteristics. I think that this can reasonably be taken to mean that Simplese must contain singular terms or their equivalent, plus appropriate predicates. Further, the stock of singular terms must be quantitatively adequate to the ontology of Simplia, which means that it must be indefinitely expandable unless the number of patches is finite and fixed. This last requirement would seem to mean that Simplese must contain something like a description operator.

Clearly we still have not emerged from triviality. The next point to be made is, however, rather more interesting. To see what it is, let us suppose, as might well be the case, that the circular patches which we have supposed Simplia to contain, rather than moving about as individuals, are normally to be found in structured "cluster" of patches, which clusters are relatively stable and immune from dissolution. We may suppose too that only certain combinations of patches in certain arrangements will form stable clusters, and that all clusters which are composed of patches of the same colors and in the same arrangement have identical dynamic and (resultant) color properties, which depend in some regular and discoverable way on the properties of the patches which compose them. To complete the picture, suppose finally that although these clusters are relatively stable, they do in some circumstances break up into their constituent patches, e.g. when they "collide" in the right way with the right sort of patch or patches.

My next suggestion should be obvious. Suppose that Simplese contains singular terms and predicates which refer to and describe, respectively, not the patches of Simplia, but only the clusters. It seems quite clear that such a language and conceptual framework could be highly successful, and to that extent adequate, in most circumstances, i.e. in circumstances which did not involve the dissolution of clusters. It is equally clear, however, that such a framework would be inadequate in at least two respects: (i) it would describe all states of affairs in Simplia in an oversimplified way, representing as simple what was actually complex; and (ii) it would be unable to describe some states of affairs at all, viz. those involving either single, unattached patches, or dissolutions (and formations) of clusters. And these inadequacies would be reflected in the fact that an account of Simplia in such a Simplese would be in two ways less coherent than an otherwise similar account in a richer Simplese which contained also terms for referring to and describing patches: (i) to the extent to which the properties of the various sorts of clusters are functions of the properties of their component patches, the account of Simplia in the weaker Simplese would contain unexplained "brute facts" about the various sorts of clusters which could be explained and systematized in the richer Simplese (in terms, of course, of other, presumably fewer, "brute facts" about patches); (ii) the account in the weaker Simplese would contain completely unexplained anomalies in its descriptions of those states of affairs involving formation or dissolution of clusters, which anomalies would be explicable in the richer Simplese.

Thus we seem to have uncovered a non-trivial constraint on an adequate conceptual framework. To be fully adequate Simplese must contain singular terms which refer to, and predicates which describe, those entities which we have supposed to be the basic particulars of Simplia, viz. the patches. Of course, Simplese could still contain singular terms and predicates which applied to clusters, but both extra-linguistic adequacy and intra-linguistic coherence would demand that such clusters be viewed as composite or complex particulars, and that it be recognized that anything which could be said about clusters could, "in principle" at least, be said in terms which applied only to the patches composing them. It is worth noting too that it might well be the case that the terms appearing in observation statements continued to be those applying to clusters. This would presumably be the case if the sensory capacities of our cognitive being who uses Simplese were incapable of discriminating individual patches, either singly or as components of clusters. But these observation statements, though phrased in terms or clusters, would still, again perhaps only "in principle," be translatable into patch terms; indeed it would be possible, though very likely not practical, to retrain our cognitive being to respond directly to the world with complicated statements about patches, instead or employing cluster-talk. And whether or not this latter step was taken, the most basic statements of the framework from an ontological, though not from a purely methodological, point or view would be singular statements about patches, including those about free patches, which would be derivable from the observation statements, together with an adequate theory of patches.{32}

The foregoing discussion represents my sketchy attempt to recreate in terms of Simplia and Simplese what I take to be an absolutely central aspect of the scientific dialectic which led from a framework of ordinary physical objects to one of atoms and now to one of whatever sorts of particles are currently regarded as most basic. It attempts to show in part in what obvious but extremely important sense the later frameworks are more adequate than the ones which preceded them, and how this greater adequacy might result from the quest for greater coherence.

Retreating to our example, we may sum up the discussion so far as follows. Simplese, if it is to make possible an adequate representation of Simplia, must contain, potentially at least, a singular term for every patch in Simplia, together with appropriate predicates to characterize such patches. (I shall have more to say later on about the issue of which and how many predicates Simplese must contain.) Such a patch-vocabulary need not be the observational vocabulary of Simplese, but it must be linked to the observational vocabulary in a way which allows the cognitive import of observations to be completely restated, in principle at least, in patch terms. It is singular statements about patches which represent the most basic description of Simplia in Simplese. These considerations, taken together, represent the first general constraint upon Simples.

The second general constraint upon Simplese is intimately related to the first and grows out of the same basic idea, viz. the idea that the basic description of a world exists at the level of singular statements describing particular states of affairs. It seems clear that if Simplese is to adequately describe Simplia at this level, it must contain resources for describing the dimensional structure of Simplia. That is, given a number of (true) singular statements about Simplia in Simplese, it is not enough for them to constitute an adequate description or partial description of Simplia that they merely assert that certain patches have certain properties somewhere and somewhen. Rather one wants to know how the patches thus characterized are spatially and temporally related to each other. An adequate framework must contain means of specifying such relations.

There are many superficially different ways in which this could be done. One way, clearly the simplest, would be to adopt a co-ordinate system, involving in the case of Simplia one temporal and two spatial co-ordinates, and then to simply attach a co-ordinate qualifier to each singular statement. Thus if 'George' is a singular term for one of the patches, a typical singular statement in Simplese might be:

Red (George) at (6, -1; 10)

which would say that a particular patch was red at spatial location (6, -1) at time 10.

Clearly the employment of such a co-ordinate system is not the only way in which the spatio-temporal relations of the patches of Simplia could be specified in Simplese. One could instead employ a system of "landmarks" of some sort, replacing the co-ordinates by specifications of the form 'near X' and 'at the time of event Y'. (This obviously assumes that some of the patches or clusters of Simplia are relatively fixed in their spatial relations to each other, so that they can serve as landmarks, but clearly something of this sort will also have to be assumed by any co-ordinate system.) Such landmark systems could be of greater or lesser complexity and specificity, depending on how many landmarks were employed. It should be fairly obvious, however, that such systems could adequately achieve their purpose only if and to the extent that they were replaceable by a co-ordinate system of the sort suggested above. That is, it is not sufficient to merely attach to the singular statements of Simplese spatial qualifiers like 'near X', 'near Y', 'near Z', etc. The user of the framework must also know the relative positions of X, Y, and Z, or else he would not know the relative positions of the patches described in the singular statements thus qualified. And similarly for temporal "landmarks". But to the extent that the relative positions of the landmarks are known, it will obviously be possible to replace the system of landmarks with a co-ordinate system of the same degree of precision which takes one of the landmarks as the origin. And it could reasonably be said that such a co-ordinate system would represent the epistemic "cash-value" to our cognitive being of the landmark system. Thus we will, I think, beg no questions if we simply assume that an adequate conceptual framework must employ a co-ordinate system of the sort indicated above to spatio-temporally qualify its singular statements. (Note too that the existence of such a co-ordinate system will do our cognitive being no good unless he is continuously aware of his own position in it; unless, that is, he is able to translate statements in terms of 'here-now' into statements in terms of the co-ordinates.) The necessity for such a co-ordinate framework will then represent the second general constraint upon Simplese.

The third general constraint upon Simplese derives from the reflection that our cognitive being, if he is like us, can observe only a very limited spatial portion of Simplia at any particular time. He can of course change his position of observation so as to observe other spatial portions of Simplia, but the patches are in constant motion so that the observation of a particular spatial portion at one time does not by itself provide him with any knowledge of that spatial portion at different times. Thus either his knowledge of Simplia will remain extremely fragmentary, extending to only a small spatio-temporal chunk of Simplia, or Simplese must contain the necessary resources to extend his knowledge beyond that portion of Simplia which he is able to observe. This means of course that Simplese must contain laws or something like them, together with all of the conceptual apparatus required for the formulation and confirmation of such laws. I shall not attempt here to spell out exactly what is involved in this requirement, a task which I could not accomplish in any detail anyway, but shall simply rely on our intuitive grasp of the matter, and content myself with making a few small points.

It should be noted, in the first place, that such laws need not be explicitly stated by our cognitive being. Rather many or most of them will be represented by the inference patterns which govern and constitute the concepts of Simplese. (Thus the requirement that Simplese contain laws was already implicit in the minimal requirement that it be a conceptual framework at all.) To take a very simple example, the concept of a "patch," as we have used it, already involves at least the idea that patches are either eternal and imperishable, or at least are very long-lived. If the former, then the concept of a "patch" would already implicitly involve the following law (where 'p' is a variable ranging over patches, the other variables range over spatio-temporal co-ordinates, and '=>' represents nomological implication):

(p).(Ex)(Ey)(Et)(p is at (x,y;t)) => (t)(Ex)(Ey)(p is at (x,y;t))

If the latter, the law would be similar but more complicated, presumably involving some sort or probability qualifications. There would be various other laws about the shape, size, color, motion characteristics, etc., of the various sorts of patches, which might or might not be explicitly stated.

A second point to be made is that such laws, whether explicitly stated or not, can usefully be viewed as principles of inference which allow our cognitive being, when he has arrived at one non-lawlike statement of an appropriate sort, to infer a different non-lawlike statement as a consequence (where such "non-lawlike" statements would normally be either singular statements, quantificational generalizations or singular statements, or truth-functional combinations of these two kinds of statements). I think, in fact, following Sellars,{33} that such an analysis of law-like statements is in fact correct, and captures the ideas (a) that the epistemic point of lawlike statements is to enable us to expand our knowledge of particular fact beyond what we actually observe, and (b) that lawlike statements do not describe the world in the same direct way that singular statements do. But I shall not insist on these points for the moment, noting only that whatever else law-like statements may be or do, they at least authorize inferences of the sort described, so that to regard them as principles of inference, though it may not be a complete account, is at least not seriously inaccurate as far as it goes.

So far I have been discussing. only the general requirement that the framework contain laws. Now it is time to see just what sort of laws it should contain. I submit, on the basis of our leading idea that the purpose of a conceptual framework, and of the cognitive use made of such a framework, is ultimately to represent the particular states of affairs which in fact obtain in the world, that Simplese will be adequate just insofar as it contains laws enabling our cognitive being to infer from what he can observe about Simplia descriptions of those spatio-temporal parts of Simplia which are outside his observation, thus gradually building up an increasingly more complete description of Simplia as his sum of individual observations increases. Ideally, of course, he would eventually be able to arrive at a complete description of Simplia at all places and times, but of course this is an ideal which is most unlikely to be achieved in practice, even in Simplia.

It must be emphasized here that the sort or structure of laws which can be arrived at by our cognitive being is largely dependent on what concepts are employed in Simplese, as should be in any case evident once the intimate relationship between concepts, inferences, and laws, discussed above, is realized. We have already seen one way in which this is so in the requirement that the singular terms and descriptive predicates of Simplese must make it possible to refer to and describe the basic ontology of Simplia. But apart from the question of basic ontology, the degree of coherence possible in a framework is also heavily determined by what descriptive predicates exist in the framework. Thus, to take a very simple example, if the dynamic properties of our patches are, as we have supposed, correlated with their colors, a conceptual framework which lumps together, say, red, orange, and yellow as one color will not be, able to contain dynamic laws of the same degree of detail and exactitude. Thus the third general constraint upon Simplese is that it contain a system of descriptive concepts and laws formulated in terms of those concepts which makes it possible for a cognitive being using Simplese and with limited powers of observation to infer from his observations as complete a description of Simplia as possible.

I believe that with the establishment of these three constraints on Simplese we are in a position to show that something like a relation of picturing or correspondence will obtain between natural-linguistic objects which are elements of true statements (or beliefs) in Simplese, or rather a subset thereof, and the objects of Simplia. But before attempting to show this, it will help greatly if we make a few simplifying assumptions which will not, I believe, affect the substance of the argument. These assumptions are as follows. (1) I shall neglect the complications involved in the possibility that our cognitive being might be physiologically able to observe only clusters of patches, rather than individual patches, and shall assume that he can directly observe individual patches, both free and in clusters. (2) I shall also neglect the (irrelevant) problems involved in the reidentification of particulars, by assuming that whenever our cognitive being encounters a particular patch for the first time he indelibly writes a unique name on it, and also that this is the only way in which patches come to have names written upon them, thus making the reidentification of Simplian particulars simple and virtually infallible. (3) It is obvious upon reflection that cognitive activity normally takes place at several different "levels," namely those of overt discourse (both vocal and inscriptional), of explicit thought, and of dispositions to thought (i.e. beliefs), and that a complete description of the world, to the extent that it exists at all, exists only at the last of these levels. I.e., we do not normally express overtly everything that we think, nor do we think explicitly everything that we would be disposed to think if confronted with appropriate questions or other circumstances. I shall assume, however, that our Simplian cognitive being writes down all those singular descriptive statements which he is even disposed to think, and that those statements are all that he writes down, though not of course all that he thinks. (In the present context, given assumption (2), a singular statement is any statement which contains the names of one or more patches, and contains no truth-functional connectives, quantifiers, nomological implication signs, etc.) I shall further assume that the time consumed in this rather large amount of writing is negligible. (4) I shall assume in accordance with our earlier discussion of laws, that the acceptance of a lawlike statement, as of any principle of inference, involves a disposition to draw all conclusions which follow by that principle from premisses which one accepts. This means that our cognitive being will write down not only singular observation statements, but also all other singular statements which follow from those observation statements by laws which he accepts.

Let us now see what results as our cognitive being moves through Simplia employing Simplese, subject to the indicated constraints and assumptions. He inscribes many observations sentences which have the general form of the description of a property (or relation) to a named patch (or patches), with an attached spatio-temporal qualifier. He also, by employing the system of laws which he gradually builds up, inscribes many other sentences of the same general form representing nomological inferences made from his stock of observation sentences. He thus gradually accumulates an extremely large stock of singular sentences which, to the extent that they are true, constitute an increasingly more detailed and accurate description of his world Simplia.

Consider this stock of sentences purely as natural-linguistic objects in the world, abstracting from the fact that they are significant components of a conceptual system. Consider first the attached spatio-temporal qualifiers, which consist, as we have seen, of ordered triads of numerals. It is possible to order these qualifiers, and thus also the sentences to which they are attached, in three dimensions according to the usual ordering of the numerals. Such an ordering can be viewed as purely factual, representing the way in which our cognitive being does in fact order numerical inscriptions in counting and other activities (which factual ordering of course reflects his normative understanding of the way in which convention dictates that such inscriptions ought to be ordered). I submit that the spatio-temporal qualifiers of Simplese, considered in light of this purely factual three-dimensional ordering, determine what might be called a "co-ordinate space," which mirrors and is a projection of the actual space-time of Simplia.{34} For every actual space-time point in Simplia, there is potentially a correlated "point" in the three-dimensional ordering of the co-ordinate qualifiers. Particular sets of co-ordinates (considered now purely as inscriptions) are linked to the particular space-time points which they project by the lawlike correlation between a sentence being inscribed as an observational response to sensory stimuli at that space-time point and the appearance in that sentence of that set of co-ordinates. This lawlike correlation is of course a consequence of the fact that an observation which is accurately and correctly made will have a spatio-temporal qualifier which represents the cognitive being's knowledge of his own position in space-time. Thus the ordering of the sentences according to this three-dimensional ordering of their co-ordinate qualifiers will reflect in a purely factual way the three-dimensional order in which the observation sentences actually were inscribed, and in which the other sentences would have been inscribed had they been observational.

It thus seems quite reasonable, in light of this factual one:one correlation between spatio-temporal points in Simplia and co-ordinate "points" in Simplese, and of the isomorphism between the spatio-temporal order of the former and the three-dimensional factual order implicit in the latter, to view the co-ordinate space of Simplese as a projection or, in an extended sense, a picture of the actual space-time of Simplia. (We may also note, with Sellars, that this picturing could be made even clearer by writing the numerical co-ordinates in the notation which employs only '0' and ' ' ', instead of ordinary numerals, so that the ordering of the co-ordinates and the relative distances between co-ordinate points would be even more clear;{35} I do not think, however, that this point matters very much.) One helpful way of visualizing this picturing relationship is to imagine the co-ordinate qualifiers of Simplese being ordered in our three-dimensional space in the obvious way, i.e. in a way which parallels an ordinary three-dimensional spatial co-ordinate system, with one of the dimensions representing Simplian time.

Having established that the co-ordinates attached to the singular sentences which are inscribed by our cognitive being constitute a non-conventional picture of the space-time of Simplia, we may now consider the singular terms appearing in such sentences, which we have supposed to be all names. These names will be correlated in two non-conventional ways with the patches to which from a semantic point of view they refer. In the first place, once a name has been applied to a given patch, an inscription which is from a linguistic point of view a token of the name in question will appear in any inscribed sentence which is linguistically a correct and accurate observation of that patch. I.e., to the extent that our cognitive being uses his framework correctly and accurately, he will be causally disposed to respond to patch George with an inscription which contains a token of 'George'. Such an observation will also, if correct, contain those co-ordinate inscriptions which are the projection into co-ordinate space of the point at which George is observed. Secondly, to the extent that the system of laws contained in Simplese is adequate, further tokens of 'George', will appear at just those places in the co-ordinate space which are projections of those places in the actual space-time of Simplia which the real George actually occupies. Thus it seems reasonable to regard the inscription 'George' as picturing or corresponding to the patch George in a way which is in an important way non-conventional. And similarly for the other names of Simplese, to the extent that Simplese is adequate and is correctly employed.

Consider finally the descriptive predicates of Simplese, first a one-place predicate like 'red'. Again to the extent that Simplese is adequate and is employed correctly, inscriptions which are tokens of 'red' will be non-conventionally correlated with instances of red in ways which parallel and are partially based on those just discussed for singular terms. First, if 'red' is an observation predicate for our cognitive being, i.e. if he is able to discriminate instances of red and has been conditioned to report them, then tokens of 'red' will appear in those inscribed sentences which are from a linguistic point of view observations of red patches, and will be attached in the (factual) manner which (linguistically) signifies predication, to singular terms which (linguistically) refer to those patches and are (factually) correlated with them. Secondly, whether or not 'red' is an observation predicate for our cognitive being, tokens of 'red' will, again to the extent that the nomological apparatus of Simplese is adequate, appear at those and only those places in the co-ordinate space which are projections of places in the actual space-time of Simplia which contain instances of red, and will be attached in the natural-linguistic manner which linguistically signifies predication to the natural-linguistic objects which are singular terms referring to those instances. Again, it seems reasonable on the basis of these two non-conventional correlations to regard the natural-linguistic objects which are inscriptions of 'red' as corresponding to or picturing the property red.

The case of relational predicates is similar, though slightly more complicated. Natural-linguistic objects which are tokens of an n-place relational predicate will, as a result of either observation or nomological inference, be attached in the natural-linguistic manner which corresponds to predication to ordered n-tuplets of natural linguistic objects which linguistically refer to ordered n-tuplets of patches which are instances of the relation in question. Again it seems reasonable to regard such natural-linguistic objects as picturing or non-conventionally corresponding to the relational property in question.

As Sellars point out,{36} it is possible to make the picturing relation even more exact by thinking of the attaching of predicates to singular terms as a way of writing the singular terms in question, and analogously for the natural-linguistic counterparts. On such a view, an inscription like "Red (George)" would be thought of as a way of writing the inscription 'George', viz. writing it in such a manner that it has the property of being-enclosed-in-parentheses- and-preceded-by-a-token-of-'red'. Given such a point of view, the picturing relation for Simplese and Simplia would consist in the fact that singular term inscriptions were correlated with patches and properties of those singular term inscriptions (of the above indicated sort) were correlated with properties of the patches. It is this sort of idea which leads Sellars to say that a perspicuous language would consist only of singular terms, which could be written in different ways in a more obvious sense, thus e.g. in different styles of type. (It should be carefully noticed that although this way of looking at the matter, and indeed most of our discussion above, assumes a certain view of the ontological status of those features of the world which are represented linguistically by predicate terms, viz. roughly that they are at least not in the world in the same sense as are the objects represented by singular terms, this assumption is in no way essential to the basic notion of picturing. Thus one who, like Bergmann, had a different view of such issues would be free to modify the account of picturing so as to treat predicates like 'red' as disguised singular terms which referred to entities in a manner exactly parallel to the reference of ordinary singular terms.)

Thus I conclude that the stock of singular descriptive sentences which our cognitive being gradually accumulates as he observes Simplia, to the extent that they are true in Simplese and Simplese is adequate for describing Simplia, constitutes a picture or projection of Simplia in a stronger sense than that of mere conventional correlation. The production of such a picture results solely from the fact that an adequate conceptual framework is correctly employed, subject to our simplifying assumptions.

It is very important to emphasize that the picturing relation which I have attempted to characterize obtains, not between a particular true sentence in Simplese and the state of affairs in Simplia which such a sentence describes, considered in isolation, but only between a relatively complete description of Simplia and Simplia as a whole. Between a particular sentence of Simplese such as:

Red (George) at (6, -1; 10)

and the state of affairs which is George's actually being red at place (6, -1) and time 10, there is clearly no non-conventional correspondence, picturing, or isomorphism of any sort. 'George' looks nothing like George, 'red' bears no discernible resemblance to red, and the difference between the numerical co-ordinates and the space-time point which they designate is, if anything, even more pronounced. Nor is this very surprising, since clearly any of a very large class of alternative inscriptions could just as well have served the purpose which 'George', 'red', and '(6, -1; 10)' serve in Simplese, and clearly it would be absurd to expect that all such inscriptions would somehow resemble the things which they represent. Any attempt to locate a picturing relation purely at the level of individual sentences is thus clearly hopeless, and it is such misguided attempts which have caused the notions of correspondence and picturing to fall into disrepute. The picturing relation which actually does obtain, if the above argument is correct, obtains in contrast only between systems of sentences and the world, and is of a broadly structural sort, involving no crude notions of simple resemblance between word and object. I.e., the co-ordinate structure of the language mirrors the spatio-temporal structure of the world, and the positions occupied within each of these structures by particulars and properties and the words which are structurally isomorphic to one another. The sense in which a single sentence pictures a single state of affairs is thus a derivative one, based on the fact that the sentence is or could be a component of a set of sentences which picture the world which contains the state of affairs.

10. Truth and Picturing: Our Conceptual System.

The thesis of the preceding section was that the true sentences of a conceptual framework adequate to describe a certain, very simple hypothetical world would, considered as natural-linguistic objects, stand in a non-conventional relation of picturing or correspondence to the objects of that world which those sentences, considered now as assertions in the framework, would be said to describe. This thesis alone, even if true, is of course somewhat less than exciting. Our world is not Simplia nor even remotely as simple as Simplia, and most at least of the simplifying assumptions made in the course of the discussion seem clearly false for our world and our conceptual system. What I want to suggest in the present section is that the same thesis which was shown in the preceding section to be true for Simplia and Simplese is in fact also true for our world and our conceptual system to the extent that our conceptual system is adequate to describe our world, the natural-linguistic objects which from a linguistic point of view comprise true singular assertions in that system will stand in a relationship of picturing or correspondence to the objects of our world.

I do not propose, however, to discuss this claim in anything like the detail in which I have already discussed the analogous claim for Simplia and Simplese. To avoid such a discussion was after all the point of the simplified example, and it would in any case be impossible for the present case for either of two reasons. First, there can be little doubt that our conceptual system is far from being completely adequate, so that one would have to discuss complicated issues about degrees of adequacy and degrees of picturing; such a discussion would be at once sufficiently intricate and sufficiently uncertain as to blur the central issues and contribute little or nothing to the advancement of the thesis. Secondly, even apart from questions of adequacy, our conceptual system and our world are sufficiently complex as to make it quite impossible to consider the issue of picturing in anything like adequate detail without far exceeding the length appropriate for the present essay. So I have little choice but to rely mainly on my simplified example to make the point, supplementing it with an attempt to show that none of the simplifications and assumptions which were made in the earlier discussion has any effect on the central point in question; so that if the thesis is valid for Simplia and Simplese, there is no good reason to doubt that it is also valid for our world and our conceptual system.

Consider to begin with the three constraints which were found to apply to Simplese and which were the basis of the picturing relation. The first of these, to the effect that an adequate conceptual system must possess singular terms and descriptive predicates which refer to and describe, respectively, the basic particulars of its world, seems also to apply to our conceptual system. As I have suggested, this constraint seems to explicate the sense in which the elaborate conceptual frameworks of contemporary micro-physics are more adequate than the common-sense framework of physical objects. Of course there are two significant differences in this regard between Simplese and our conceptual framework. (1) There is no prospect, for the moment at least, of observing the world in micro-physical terms, or of translating very many ordinary language observations into such terms. Micro-physical observations do seem to be possible in a few isolated cases, and at least partial translation in some others, but in general the complexities involved are too great. (2) For the same general sort of reason, there seems to be little or no possibility of ever producing an adequate description of the world at the level of singular micro-physical statements. But these points, though obviously of enormous practical significance, do not affect the theoretical point at issue. They merely show that from a theoretical standpoint our conceptual system is, as I have already remarked, seriously inadequate and likely to remain so. If however the scientific conclusion that the ultimate constituents of our world are sub-atomic particles is correct, it still remains true that a completely adequate description of that world would consist of singular statements at the sub-atomic level, which would then picture the particles which they described.

Someone might want to object at this point that the whole notion of "ultimately basic particulars" of a world, which is involved in the first constraint, is illegitimate. Why should we assume that any particulars are thus basic? I have some sympathy for this sort or objection, but the issue is in any case not critical. All that the argument really requires is that some particulars be more basic than others, and thus that some frameworks be more adequate than others. Since the basic claim is stated only in terms of degrees of adequacy ("to the extent that the framework is adequate"), it is perfectly all right for absolute adequacy to remain an unrealizable regulative ideal. This relativized view also enables me to make the further point that to the extent that the ordinary language framework of physical objects is adequate, the same claims will hold of it. I.e., true sentences of that framework will, as natural-linguistic objects, stand in a picturing relationship to the objects, viz. complexes of particles, which they describe. Of course the picture thus constituted will be obviously less adequate than would be an ideal picture made up of statements in micro-physical terms.

The second constraint, that the framework adequately represent the dimensional structure of its world, seems also to apply to our conceptual framework. Of course we do not employ anything like the precise co-ordinate system which I have supposed to be embodied in Simplese. Our co-ordinate system is for most purposes essentially a landmark system of a fairly crude sort, in which we generally have only the roughest sort of idea of the relative distances and directions of the various objects. And the spatial extent of the co-ordinate system varies radically with our purpose. For most purposes the surface of the earth, or some sub-region thereof, is all that matters, but for some purposes we may be concerned with the entire solar system, or with our entire galaxy, etc. Our time measurements are markedly more accurate, but even here we are often content with only a very rough and approximate specification. Again, however, none of these points makes any real difference for the thesis at issue. Rather they all show simply that our ordinary conceptual framework, though it may serve perfectly well for ordinary purposes, is theoretically inadequate in that many of its descriptions are spatio-temporally vague or imprecise. And it remains true that to the extent that it is adequate it contains a co-ordinate structure which, if spelled out explicitly, would make possible the ordering (in four dimensions) of singular statements made in the framework in a manner which would reflect the dimensional structure of our world.

The third constraint, that the framework contain concepts and laws which make it possible for users of it to infer descriptions of unobserved entities and events and thus to construct a description of their world which extends beyond the narrow limits of their actual experience, again clearly applies to our conceptual framework as well. The discovery of such laws and concepts is after all the whole point of the scientific enterprise. This point seems obvious enough to require no further discussion. I conclude that the three general constraints which we saw to apply to Simplese also apply to our conceptual system.

What then of our simplifying assumptions? Assumptions (1) and (2), though false, are harmless enough. With regard to (1), we have already seen that the question of whether the vocabulary which refers to and describes the basic particulars is used for making observation reports and/or for ordinary description of the world affects only the adequacy of the framework. It does not affect the basic point that a description of the world at any ontological level will, to the extent that it is adequate, stand in the same general sort of relation to the world as would a description at a more basic ontological level. With respect to (2), although the problem of reidentification of particulars is a genuine and important one, it does not particularly affect the argument here, so long as we insist only that an adequate framework, ours or any other, must have some method of reidentifying particulars, and will be more or less adequate as that method is more or less reliable.

Assumption (3), on the other hand, when recognized as false, does force a significant emendation of our view of the picturing relationship. Once we return to the real world, where people do not say or write all that they are disposed to think, we shall have to admit that a relatively complete picture of the world, to the extent that it exists at all, exists not at the level of overt discourse, nor at the level of explicit thought, but only at the level of dispositions to thought. And thus the status of the picture will depend to an important degree on our account of dispositions to thought. If we believe that a disposition can exist only in some actual, non-dispositional "embodiment," we may want to say that it is the elements of that embodiment which picture the world. Alternatively, we may want to say that the picture exists only potentially, rather than actually, for the most part. These are interesting issues, but they do not affect the basic claim that the elements of an adequate description of the world would constitute a picture of it, in the sense specified.

Assumption (4) seems on the other hand to pose no problems, simply because it is true. It is hard to see what would count as acceptance of a law if this did not involve at least a disposition to make inferences from singular statements to singular statements in accord with it.

I conclude that if the argument of the preceding section is sound for the case of Simplia and Simplese, it is also sound for our conceptual system and our world. To the extent that our conceptual system is adequate, it will enable us to gradually generate a system of singular statements, beliefs, and dispositions to believe which, if made explicit at the level of overt statement, and perhaps at the other levels as well, would be constituted by a system of natural objects which would picture the world in the way characterized above.

I can easily imagine there being someone who has followed the argument to this point, and who is even more or less convinced that the picturing relationship thus characterized really exists, but who is puzzled about the significance of this conclusion. What, he might ask, does this conclusion say about the problem of truth and other issues in the same dialectical vicinity? In what sense is it, as Sellars claims, the core of the old- fashioned correspondence theory of truth? What is its epistemological significance, if any? I want to devote the concluding pages of this section to an attempt to at least suggest answers to these questions.

First. There is very little point in debating about whether or not the picturing dimension of empirical truth is a correct explication of the correspondence theory. The latter theory was simply never formulated in a clear and careful enough way to provide any basis for deciding such a question. Further, there seems little reason to deny that at least part of what the correspondence theorists were trying to get at was the semantic dimension of truth formulated by Tarski. But the picturing aspect of truth, as here characterized, does represent a way in which true statements correspond to the world, and a way moreover which is limited to those statements for which the correspondence theory has always seemed most inescapably obvious, viz. singular statements of empirical fact. Thus it seems plausible at least that the picturing dimension of truth captures a significant aspect of the old correspondence theory.

Second. More importantly, the picturing dimension provides the ingredient which I suggested was lacking in the neo-coherence theory of truth sketched in section eight in this chapter. It explains why the ideally coherent account of the world reached at the end of scientific inquiry is ideally coherent and why it will not be superseded by a further account. The reason, very crudely and approximately, is that in the ideal account the natural-linguistic objects of which it is composed will picture the world exactly, so that there are no natural-linguistic objects without correlated objects in the world, nor objects in the world without correlated natural-linguistic objects, leaving no room on the linguistic level for an unfulfilled expectation or an unexpected observation, and thus yielding complete coherence and scientific satisfaction. As Sellars remarks, the picturing dimension of truth "provides that missing ingredient, the absence of which from Peirce's account of truth leaves the 'would-be' of the acceptance 'in the long run' of propositions by the scientific-community without an intelligible foundation . . ." [SM vii]. The picturing dimension also provides the underlying rationale for accepting some sets of standards for epistemic justification and not others: some sets of standards are more likely than others to lead eventually, through their continued employment, to descriptions of the world which are adequate pictures in the indicated respect (and hence to descriptions which will succeed in practice).

Third. The picturing dimension of truth represents also the cash value of the distinction between factual or empirical truth and other varieties of truth, a distinction which vanishes in the semantic conception of truth. Anyone who believes, as we all do intuitively, that statements like "The table is brown." or "The water is cold today." describe the world in a sense in which statements like "2+2=4" or "Stealing is wrong." or "The Eroica symphony is a great work of art." do not (which is not to say that these latter have nothing to do with the world), can take comfort in the above account. Of course, the picturing dimension does not, in itself, tell us precisely where to draw the line between those statements which do and those statements which do not describe the world in this strong sense. Rather it explicates what we mean by saying that some statements describe the world, while others do other things such as tell us how we should behave with respect to the world in light of such descriptions.

It is well to take note here of a possible objection to the effect that the notion of picturing cannot capture the difference between those statements which describe the world and those which do not, because only singular-statements, on our account, serve to picture the world, and no one ever wanted to say that only singular statements were descriptive. There are many other sorts of statements, so the objection goes, e.g. lawlike statements, quantified statements in descriptive terms, and truth- functional combinations of singular statements, which we all regard as factual, empirical, and descriptive, but which are not, on our account, components of pictures of the world. The answer to this objection is that statements of the indicated sorts, though descriptive in a derivative sense, are not descriptive in the same direct way that singular statements are. To see what is meant here, consider as an example a putative lawlike statement like "All ravens are black." It does not seem plausible to hold that this statement (if true) describes the world in the direct, simple sort of way in which a statement like "This raven is black." might. There is not a single state of affairs in the world which is all ravens being black in the way in which there is a single state of affairs which is this raven being black if "This raven is black." is true. Such a law describes, not a single state of affairs in the world, but rather all of the states of affairs which are its instances, and which could be individually described by singular statements. In our account this difference between laws and singular statements appears in the fact that while the latter are components of pictures, the former give, as it were, directions for constructing such pictures or standards for judging them. A similar account could be provided for quantified statements in descriptive terms, truth-functional compounds, etc. These all pick out some pictures and rule out others without themselves being components of pictures. And this is the derivative sense in which they are descriptive.

One point at which the picturing aspect of empirical truth becomes extremely critical is in connection with the familiar realist/instrumentalist dispute over the status of theoretical entities. The realist characteristically asserts that theoretical entities are real in whatever sense ordinary physical objects are real, and hence that true theoretical statements genuinely describe the "furniture of the world." The instrumentalist wants to deny this, and claims instead that talk about theoretical entities is only an "instrument" or a "calculational device" for making predictions about ordinary observable entities. There are, I think, serious confusions implicit in the instrumentalist notion of observation, some of which were explored earlier in this essay. But if no reference is made to the picturing dimension of empirical truth, it is hard to avoid the conclusion that the basic issue upon which realists and instrumentalists appear to disagree is a Scheinstreit. Both camps agree that a true theory will be part of the most coherent total scientific account. Both agree that such a theory is true in the semantic sense. Both can agree that such a theory is a useful instrument for making predictions about the sorts of phenomena which the instrumentalist wants to call observational. But in spite of these agreements, the realist still wants to say that such a theory genuinely describes the world in the strongest possible sense, while the instrumentalist wants to deny this. But does this difference really amount to anything? Can the point at issue even be clearly stated?

It is most tempting at this point to conclude that the instrumentalist/realist issue really is a pseudo-issue, and many eminent philosophers have yielded to this temptation. It seems to me, however, that this move is intuitively so implausible that it should be resisted at almost any cost. Intuitively the difference between atoms and protons really existing "out there" in the same sense in which houses and apples do, on the one hand, and it merely being useful to talk of atoms and protons in order to make predictions about houses and apples and somewhat more esoteric objects of the same sort, on the other, is simply as enormous as can be imagined. It would seem that any philosophy which is forced to declare an intuitive difference of this magnitude to be illusory is in extreme danger of having implicitly declared itself to be illusory.

Someone might want to suggest at this point that the solution to our difficulty lies in the issue of observation. That is, what the realist wants to assert and the instrumentalist wants to deny is that theoretical entities can be observed, in some circumstances at least, in the same full-blooded sense of 'observe' in which tables and apples are observed. Now this answer represents, I think, a large step in the right direction. To see, however, that it is not the whole answer, one need only ask why the instrumentalist wants to deny that theoretical entities can be observed. The answer would, I think, appeal to the principle that one cannot observe what does not genuinely exist, where this is not simply the tautology that one cannot observe what one cannot observe. The point is that the disagreement with respect to observation seems at least to be a consequence of the more basic depute, rather than a restatement of it.

The answer which I want to suggest to this problem should be obvious to anyone who has followed the discussion to this point. One way of formulating the issue between the realist and the instrumentalist is to say that for the realist the natural-linguistic objects which comprise statements of the micro-physical level genuinely picture the world in the sense characterized above, while for the instrumentalist such statements are only instruments for constructing pictures at a different level, viz. the level of ordinary observable things.{37} This is not, of course, a formulation which the parties to the dispute would be likely to immediately accept or even recognize, but I think that it preserves the essential point at issue between them. This view of the matter also enables us to give an account of how the question of observability figures in. For key amongst the factual regularities which connect natural-linguistic objects with the elements of the world which they picture are regularities pertaining to observation. This has the consequence that a term which could not, in principle, figure in an observation report also could not, in principle, picture anything in the world, since some of the correlations which constitute the picturing relationship would be lacking. This is why the issue of whether it is possible in principle to observe theoretical entities "directly" is absolutely critical for the realism/instrumentalism dispute. (And why, on the account of observation in section six, it would seem that the realist view is correct.)

This point also enables me to make a correction of my earlier statement to the effect that an understanding of the picturing dimension of empirical truth only helps us to understand the distinction between descriptive and non-descriptive statements; but does not tell us where to draw the line between them. This statement is still true, but once we see the intimate relation between picturing and observation, we see also that the decision to regard a given sort of statement as descriptive or non-descriptive will force a parallel decision about what can or cannot be observed. And once this is realized, the inclination to regard some sorts of statements as descriptive will become very slight indeed.

Fourth. The notion of picturing finally makes it possible to provide an answer to the age-old philosophical puzzle of how a cognitive being is able to know and successfully manipulate a world which is external to him and, in a sense, beyond his reach. The answer, which is as crude as the question, but not, I think, so crude as to be useless, is that such a being knows the world by coming to have within himself a picture of that world which is causally linked to both the actions of the world upon him and to his action upon the world. (Which is not, of course, to say that he knows the picture and not the world.)


To sum up briefly the results of the discussion: I began by setting forth the traditional conception of knowledge as (1) belief, which is (ii) true, and (iii) justified. I then discussed a familiar problem with regard to the element of justification: although one way of justifying a propositional claim is by means of an (inductive or deductive) inference from other (known) propositional claims, it would seem that this cannot, on pain of infinite regress, be the only way in which propositional claims are justified. This problem, together with the related problem of accounting for the role of observation in cognition, suggested the familiar idea that although many propositional claims are justified by means of inferences from further propositional claims, some such claims are justified in terms of something non-propositional and non-conceptual, viz. an immediate or Given experience.

I then proceeded to examine in detail one account of the doctrine of the Given, that of C. I. Lewis. My claim here was that the notion of Givenness represents an untenable conflation of two distinct sorts of apprehension, the conceptual and the non- conceptual or discriminative. The distinction between the two is that the former, but not the latter, involves associated inferences to further claims. My argument here was that Lewis could make the apprehension of the Given non- conceptual and non-propositional, and hence immune to error and to any need for justification, only by denying that such apprehension involves any inference relations at all, even to the propositional claim which is to be justified in terms of it. But then how the apprehension of the Given can serve to justify that claim remains an ineluctable mystery.

This rejection of the Given raised anew the problem of how knowledge is to be "'ultimately" justified. The only alternative to justification by something non-propositional and non-conceptual seems to be that all justification is propositional, and hence that a body of propositions is to be. justified only in terms of those internal inference relations among its component propositions which have traditionally gone under the heading of 'coherence', with no possible appeal to anything outside. I therefore attempted, in chapter two, to show how such a coherence conception of justification could avoid both (a) the regress problem, and (b) the charge of being totally non-empirical. The solution to (a) lay in the realization that a propositional claim can be regarded as justified if a justifying argument is available, whether or not it has been explicitly provided, The solution to (b) involves a conception of observation as a conditioned conceptual response to a sensory stimulus, involving no element of Givenness. The key point was that claims resulting (causally) from observation can be justified in terms of known empirical regularities about the behavior of the observer, e.g. that his observation reports of a certain sort are usually reliable; and also that the same sort of regularities enable one, through the mechanism of intentional action, to produce a situation where an empirical hypothesis, justified by considerations of coherence, could conflict with an observational claim, also justified by considerations of coherence, thus forcing the abandonment of one or the other, and a consequent revision of the overall structure of propositional claims. This is the closest one can come, in a coherence theory of justification, to the empirical testing of an hypothesis. My claim was that it is close enough.

Having offered this account of justification, I turned in the third chapter to the notion of truth, and tried to show that a coherence theory of justification does not require a coherence theory of truth. My argument here was a defense of Sellars' claim that empirical truth essentially involves a structural isomorphism between the linguistic objects which comprise a true account of the world, considered merely as natural objects, and the objects in the world which that account of the world describes. This structural isomorphism enables one to say that language "pictures" or is a non-conventional projection of the world which it describes, thus reinstating something like a correspondence theory of truth.

A rather different way of summing up the results of the essay would be as follows.{38} Three notions or ideas have historically played key roles in modern epistemology: (i) that of the non-conceptual or non-propositional; (ii) that of the sensory; and (iii) that of the element in cognition which directly corresponds to reality. Traditionally there has been a strong tendency to conflate these three notions into one: a non-conceptual, sensory Givenness which represents the mind's sole direct contact with the world. The burden of the present essay has been to argue, in opposition to such a view, that these three notions can and must be kept distinct. Thus chapter one argued that the non-conceptual can play no role in cognition, except insofar as it is conceptually apprehended or represented. Chapter two, on the other hand, argued that the sensory element in cognition indeed plays a central role, but is to be viewed as thoroughly conceptual in character. And the argument of chapter three attempted to show that the notion of correspondence to reality could be distinguished from both of these others, though dependent in a key way on the sensory element (i.e. observation).

One further way in which the conclusions of this essay might be viewed, one of which Sellars would certainly approve, would be as an attempted reconciliation of the central claims of the three leading philosophical schools of the last hundred years: idealism, empiricism, and realism. Its central theme is that the insights of idealism and empiricism, viz. the coherence conception of knowledge and the idea that all knowledge depends on sensory contact with the world, respectively, can be given the full credit due to them, without denying that basic realism which is so central a feature of both naive common sense and sound philosophy.

Go to Table of Contents


{27} The discussion in this paragraph is largely based on George Pitcher's helpful introduction to his anthology Truth (Englewood Cliffs, New Jersey: Prentice-Hall, 1964). [Back]

{28} Alfred Tarski, "The Concept of Truth in Formalized Languages," in his Logic, Semantics, Metamathematics (Oxford: Clarendon Press, 1956). [Back]

{29} Donald Davidson, "Truth and Meaning," Synthese 17 (1967), pp. 304-321. [Back]

{30} Sellars' main discussions of empirical truth are to be found in his paper "Truth and 'Correspondence'," chapter 6 of Science, Perception, and Reality (London: Routledge and Kegan Paul, 1963), pp. 197-224; and in chapter 5 of SM, pp. 116-150. See also his paper "Scientific Realism or Irenic Instrumentalism?" in his Philosophical Perspectives (Springfield, Illinois: Charles C. Thomas, 1967), pp. 337-369. The following discussion is largely inspired by and based on these three writings of Sellars, though the order and manner of exposition is not. [Back]

{31} "Truth and 'Correspondence'," pp. 211 ff. [Back]

{32} The point that the statements which are observationally and hence also methodologically most basic need not also be ontologically most basic is made by Sellars in SM, p. 147. [Back]

{33} Sellars account of laws as rules of inference is in the above-cited chapter of SM; also, in more detail, in "Inference and Meaning," Mind 62 (1953), pp. 313-338. [Back]

{34} Cf. "Truth and 'Correspondence'," pp. 220-21. [Back]

{35} "Truth and 'Correspondence'," p. 221. [Back]

{36} "Truth and 'Correspondence'," pp. 221-22; cf. also "Naming and Saying," in Science, Perception, and Reality. [Back]

{37} Sellars' discussion of the realism/instrumentalism issue is in "Scientific Realism or Irenic Instrumentalism," esp. pp. 368-69. [Back]

{38} I owe this way of putting the matter to Richard Rorty. [Back]

Go to Table of Contents