Foundations Without Certainty

There has been a revival of interest in Hegel of late among English-speaking philosophers. Although he is still regarded as maddeningly obscure, a number of important philosophers (including Quine, Sellars, Feyerabend and Rorty) have been attracted by a doctrine prominently associated with Hegel, the coherence theory of truth. In order to hold the coherence theory of truth, it is obvious that you must hold what might be called the coherence theory of truth-testing as well: for if this theory is wrong and we can test some statements (even if only in part) by, for example, introspection as well as in terms of coherence, truth must involve something more than just coherence. My arguments against the coherence theory of truth-testing, since it is implied by the coherence theory proper, are indirectly against the coherence theory as well. I also argue that unless some version of the Private Language Argument is successful, it is virtually impossible to defend the coherence theory of truth-testing (and hence the coherence theory proper) without denying the existence of experiences, thereby committing oneself to materialism. If I am correct in this, Hegel (as an idealist) is in a much worse position here than materialists such as Quine, Feyerabend and Rorty.

Nicholas Rescher’s recent book The Coherence Theory of Truth (Oxford: Clarendon Press, 1973), is not as the title suggests a defense of the coherence theory proper but rather a defense of the coherence theory of truth-testing.

As such, it is a far more formidable adversary than the coherence theory proper for those of us who regard that theory as out of the question but who, though we think that some beliefs can be tested by introspection rather than by the coherence test and that such beliefs are in some sense the foundations of empirical knowledge, nevertheless recognize that this is not easy to defend. Rescher holds that there are seven basic claims to which a foundationalist is committed, and he is not attacking a straw man. Most foundationalists hold the views he attributes to them, but they render themselves unnecessarily vulnerable in doing so. I argue — and this is the main point of my paper — that a theory can qualify as a kind of foundationalism even if it abandons the claim that non-inferential judgments (roughly, judgments which have a non-inferential justification) [My account does not require such judgments to be absolutely certain or even that they could not be made more certain by adding inferential considerations. Nor should it be confused with the view that they are simply judgments that lack inferential justification because such judgments might have no justification at all] and judgments about the given are absolutely certain. This enables me to construct a theory that is still a kind of foundationalism but which is much more difficult to oppose because it abandons all but one of the seven basic claims that Rescher attacks.

I also argue that the claim that there is a given is independent of the claim that there are non-inferential judgments. Roughly, while a judgment is about the given so long as its referent is a part (or the whole) of the speaker’s current experience, such a judgment may nevertheless contain an inferential component pertaining, e.g., to linguistic usage. This independence is important because it is commonly supposed that any bona fide judgment must carry in one way or another some such inferential baggage and as I argue, even if this is so, the existence of such judgments would still be enough to show that the claim that we can only test our beliefs in terms of their coherence with other beliefs is false.

Finally, though I oppose the coherence theory of truth, I do not commit myself to the correspondence theory. I only need to defend what might be called an external requirement theory of truth: for a statement to be true (provided it is about something other than beliefs) it must do more than fulfill the internal requirement of coherence with some batch of beliefs; some sort of requirement regarding features of the world that are external to the batch of beliefs must be fulfilled. By limiting myself to this less demanding theory which, though compatible with the correspondence theory, does not imply it, I avoid the awkward problem of trying to find a sense in which beliefs could correspond with the states of affairs they are about.


Underlying six of the seven claims that Rescher associates with foundationalism is the contention that (barring a slip of the tongue) first-person experience reports cannot be mistaken. This is false. Consider the following case. A reports that he has an eleven-sided after-image. B says that he must be mistaken because the object A saw before having this after-image had twelve sides. B thinks that A must have miscounted the sides of his after-image. Surely this is more plausible than the claim that A sometimes has after-images with a different number of sides from the objects he has just seen. (A similar example could be constructed with A reporting the number of sides of his visual image when actually looking at something so that the sort of error I have described is not confined to the narrow class of afterimage reports.) Some foundationalists would respond that since A’s claim applies only to his experience and has no logical implications for physical objects, information in regard to physical objects cannot be used as evidence against A's assertion. But there are good empirical reasons for supposing that when you perceive a certain sort of physical object, you will have a certain sort of after-image. And it is easy to understand how a person could make a mistake in counting the sides of his after-image. (His miscounting could be either an instantaneous intuitive judgment or something which took place over a period of time.) Another view is that even if experience reports could be mistaken, the report of the person who had the experience should always be given precedence over conflicting claims by other people. But this is also wrong because (as we have seen) there might be perfectly good reasons to give precedence to the judgment of another person. And it has been claimed (e.g., by Kurt Baier) [Kurt Baier, "Smart on Sensations," The Australasian Journal of Philosophy 40 (1962), reprinted in C.V. Borst, ed., The Mind/Brain Identity Theory, (Macmillan, 1970)] that there is a linguistic rule to the effect that we must accept first-person experience reports unless we think that the person is insincere or is misusing the language. But my after-image example shows that there is no such rule: surely it wouldn’t do for A to reject B's challenge as linguistically improper.

Where does this leave us in regard to foundationalism? (1) I have not denied the existence of non-inferential judgments, even though I have admitted that inferential judgments (such as B’s) may override non-inferential judgments (such as A's). (2) Nor have I denied that there is a given and that in making judgments about it we can apply the introspection test [Some readers may be bothered by talk about the introspective testing of reports of pain and other sensations. The view would be that if you say you are testing a claim there is a presupposition that there is a reason for doubting it; that the presupposition would be false in regard to most sensation reports; and therefore that it is meaningless to speak of testing in the case of most sensation reports. But Grice has made it clear that this sort of presupposition can be cancelled, so there is no real problem] for truth: we can examine our experiences to check our beliefs about them [Although we normally speak of examining physical objects rather than experiences, it makes perfectly good sense to speak of examining experiences. For example, a doctor is asking you to examine your experience when he asks if your pain is sharp or dull, to examine your visual sensation when he asks if your vision is still blurred after you look through your new lenses, and to examine your auditory sensations when he asks if the ringing in your ears is high-pitched or not. Strictly speaking, we don't always "examine" our experiences when we apply the test of introspection. You are said to examine something only when a certain amount of care is required - as in the case of a judgment about an after-image with a considerable number of sides. In other cases (for example, in deciding whether you have a slight pain or that you feel a bit queasy) - though you might need to focus your attention on a certain aspect of your experience no particular care would be necessary so it would be odd to speak of an examination. In other cases, such as being asked if you are in pain when you are in fact suffering acutely, you wouldn't even need to focus your attention on the pain to answer the question. At the other extreme, there are some characteristics of our experience that are so complex that we can't determine what they are even by careful introspective examination, for instance, the speckled hen case. Once one has seen that introspection is fallible and that it is easier to make mistakes in some cases than in others, it becomes obvious that the case of the speckled hen, rather than constituting a dilemma, is simply a case where the characteristic in question is too complicated to examine in a short period of time. In the after-image case, an introspective judgment can be overthrown by an examination of the physical object plus an appeal to the laws of vision. In the speckled hen case, the latter sort of test is the only one that would be at all reliable. Introspection is normally thought of not merely as having an internal rather than an external object but as requiring that you focus your attention on that object. In the interest of simplicity, I shall disregard this last requirement so that I can speak, for example, of the introspection of intense pains as well as of mild ones].

These points qualify my view as a kind of foundationalism. However, I am forced to make drastic changes in the traditional theory — to admit not merely that first-person experience reports can be mistaken but that such statements may be less certain than some statements about physical objects and some general laws. E.g., A’s statement about his after-image was less certain than B’s statement that eleven-sided after-images don’t follow the perception of twelve-sided objects.) Also some statements based on memory are more certain than some statements about present experience: if A had felt an intense pain a few seconds before he had his after-image, he could be more certain that he had had a pain than that his after-image had a given number of sides.


I am now ready to show that my modified foundationalism (henceforth 'modest foundationalism') can reject all but the first of the following seven claims which Rescher attributes to traditional foundationalists.
1. On the F/l (foundationalist/inductivist) approach there are two distinct sorts of truths, the immediate and the derivative, while for the coherentist all truth is of a piece, (p. 209).
A modest foundationalist would disagree with part but not all of the coherentist position here. In holding that there are some non- inferential judgments and/or that there are judgments about the given, a modest foundationalist would disagree; but in holding that the bare fact that a judgement is non-inferential or refers to the given does not guarantee its truth, he is in agreement.
2. On the F/l approach experience is called upon to provide truths (i.e., immediately evident truths), while for the coherentist it only provides ‘raw’ data. (p. 209).
Clearly, a modest foundationalist rejects the F/l claim since he holds that experience-reports are corrigible rather than "immediately evident truths."
3. On the F/l approach discursive-inductive or deductive processes require an input of truth if truths are to be an output (which is exactly why an immediate, non-discursive route to truth must be postulated). The coherence analysis differs fundamentally in this regard, (p. 209).
I take it that by an input of truths, Rescher means an input of statements whose truth is certain, and it has indeed been held by many foundationalists that unless we start from such reports we cannot derive statements whose truth is even probable [This view is attacked in Mark Pastin's "C. I. Lewis's Radical Foundationalism," Nous 9, (1975), 407-20]. The modest foundationalist like the coherentist holds on the other hand that we can derive probable truths from probable premises, and further, that we can derive a conclusion that has a higher degree of probability than any of the probable premises on which it is based. Consider the following case. I start out believing (i) that I have an eleven-sided after-image (non-inferentially); (ii) that the perception of eleven-sided figures is characteristically followed by eleven-sided after-images (on the basis of memory); and (iii) that I have been told by reliable persons that the figure I just perceived has eleven- sides (on the basis of memory). Using a discursive procedure, relying on all three beliefs, I can make my belief that I had an eleven-sided after-image even more probable than it was at the outset.
4. On the F/l approach the initial ‘givens’ are wholly non-discursive and fixed invariants, while on the coherence approach the data represent a mixture of experiential and discursive elements. And thus, the non-discursive 'raw' data are, for the coherentist, only one part of the total data; they are by no means fixed and sacred but subject to reappraisal and revision, (p. 209).
In effect, a coherentist is holding that experiential claims are not on a drastically different cognitive footing than discursive claims. The modest foundationalist can agree. For example, my belief that I am having an eleven-sided after-image is not on a drastically different footing from my belief (based on memory) that afterimages have the same number of sides as the objects responsible for them and should therefore not be treated as "fixed and sacred.”
5. On the F/l approach nothing whatever that happens at the epistemically later stages of the analysis can possibly affect the starting point of basic truths, while on the coherentist approach there is a feedback loop through which the data themselves can be conditioned by the outcome of a coherence analysis (in other contexts) and their status is subject to re-evaluation in the light of new insights regarding their plausibility, (p. 209).
The sort of feedback loop described by Rescher would be perfectly acceptable to a modest foundationalist for reasons given in 4.
6. Unlike the F/l approach the coherence analysis does not require a sharp disparity in the treatment of particular and general propositions ('observation statements’ and ‘laws’), (p. 210).
My response to 6 is also included in my response to 4, since I granted there that it may be justifiable to give up a particular proposition because it conflicts with a general proposition.
7. On the F/l approach the body of ‘evidence’ from which the reasoning proceeds must be self-consistent. The coherence analysis has no need for this unrealistic supposition, (p. 210).
Again the modest foundationalist can agree with the coherentist: A's belief that he had an eleven-sided after-image was not consistent with his belief that the object responsible for his after-image had twelve sides and that perceptions of twelve-sided objects are not followed by eleven-sided after-images. Nevertheless, A was in a position to reach a rational conclusion: he had probably miscounted the sides of his after-image.


The theory which has emerged is remarkably similar to Rescher’s kind of coherence view. Is it so similar to his that it no longer qualifies as a kind of foundationalism? I think not. There are three main differences between our views, any one of which would establish me as a foundationalist. I hold (1) that there are non-inferential judgments and that in making them we apply the introspection test for truth (corrigible though it may be); (2) that even if no judgments are completely non-inferential there can be judgments about the given and we can apply the introspection test for truth to them; and (3) that although sensation-reports are not incorrigible, they do have another sort of special status.

1. Non-inferential judgments and the introspection test

It is relevant to note that Rescher (unlike me) must hold that sensation-reports can’t be justified by an introspective examination of experiences. He must hold rather that here (as in the case of memories) we just find ourselves holding certain beliefs — in this case, beliefs about our current experiences. He might add that a person is led to do so because he is in a certain physical state but he could not hold (as I would) that we are sometimes in a position to determine by introspection that a belief about an experience is or is not correct. It is here that Rescher’s theory encounters its most serious obstacle. My biggest problem, on the other hand, is with the claim that there are non-inferential judgments (though I can if necessary do without this claim).

I should first say what I mean by “non-inferential judgment” since the expression is used in a variety of ways. Traditional foundationalists seem usually to have had in mind empirical judgments whose justification is not inferential. To that extent I agree, but since they think that such judgments can't possibly be mistaken, they have tended to include the requirement of infallibility in the notion. That I must reject.

Rorty offers a different account. He uses “non-inferential report” to stand for a statement about which requests for justification are normally considered inapplicable but which can nonetheless be empirically confirmed [Richard Rorty, "Mind-body Identity, Privacy and Categories," originally in The Review of Metaphysics (1965), reprinted in C.V. Borst, ed., The Mind/ Brain Identity Theory Macmillan, 1970)]. I can accept a definition which leaves open the possibility of confirmation — e.g., a person who doubted the accuracy of his own after-image report could confirm it by finding that the object responsible for his image had the same number of sides as he attributed to his image — but I wish to build more into the notion of a non-inferential judgment than that requests for justification should normally be considered misplaced.

Other philosophers have thought of non-inferential judgements either as (1) judgements that we have no procedure to overthrow or as (2) judgements that we are forbidden to challenge by linguistic convention. For example, in pain reports Baier holds that there is a linguistic convention that so long as the speaker is deemed honest he is to be treated as the ultimate authority. Obviously, I must reject these two since I hold that honest non-inferential judgements can be rejected. One more point. Rescher claims that the bare fact one holds a belief is a reason for thinking that it is true. I agree but I don't want to use ‘non-inferential judgement' to stand for any judgement worthy of some confidence simply because one believes it. By 'non-inferential judgement,” I mean rather a judgement that has non-inferential justification which provides a reason for accepting it over and above the fact that one happens to believe it.

The view that such judgments exist faces the objection that even such a limited claim as that a pain exists now involves the implicit or presupposed inferential claim that one is using a public language correctly — or if not a public language, at least a private one — and even this modest judgment reflects a degree of conceptual sophistication that could only be acquired through training. But even though I agree that I must have training, that need not be made even an implicit part of my judgment. Furthermore, even if my judgment contains logical inferences in some form or another, that is all right: I only want a judgment that is free of empirical inferences. As for the presupposition that I am speaking a public language, that could be cancelled: the Private Language Argument has, I think, been sufficiently discredited for that to be permissible. Finally, I don’t think that I need to claim that my present pain is like some past experience or even that I used the same word earlier to express the same concept. I need only hold that my feeling falls under the concept I now associate with the word 'pain'.

2. The given without non-inferential judgments

Some readers may feel that my account of non-inferential judgments throws out the baby with the bath water: that in my attempt to describe a judgment that is completely free of inferential components I have stripped away so much that we are left not with an extremely modest judgment but with no judgment at all. I am not convinced that this objection works but even if it did, foundationalism could still survive. It is generally supposed that (a) if there are judgments about the given, coherentism is false, but that (b) if there are no non-inferential judgments, there are no judgments about the given. I think that an account of the given can be found such that (b) is mistaken yet the existence of a given is still incompatible with coherentism. What then is a judgment about the given? As in the case of non-inferential judgments, it has been the accepted view that for a judgment to be about the given its truth must be absolutely certain. Some writers seem to hold this on the grounds that the given is identical with a judgment about it; others because of the extreme modesty of judgments with no logical implications beyond the given; and some hold that there is a linguistic rule against challenges to reports about the given or that there is no procedure for overriding such reports. But if after-image reports and other reports of visual experiences are to count as reports of the given  as they surely must  all of these views of the given are false. Still another view is that the given must be such that it could be the referent of a non-inferential judgment. Since I hold that there can be a given even if there are no non-inferential judgments, I need a definition of 'given' without this requirement.

I hold then that something is given for a person making a judgment about it if it is not epistemically separated from him at the time he makes his judgment, that is, if he is not separated from the referent of his judgment in a way that could cause him to be mistaken about it. This doesn’t mean that a judgment about the given cant be wrong because (as in the after-image case) you can make a mistake in examining the given  only that you can't make a mistake because you are separated from it [In a work in progress, Rorty uses "contents of an inner arena" to stand for the events from which I would say that one is not epistemically separated. He takes the theory of an inner arena to derive from the Cartesian claim that reports of certain sorts of events are incorrigible. I think that the derivation goes (or ought to go) the other way, and I take the incorrigibility thesis to be a bastard born of the truth that there is an inner arena coupled with the error that you can't conceivably make mistakes in examining its contents - an error which reflects in part the idea that its contents are all terribly simple and obvious]. I think that this requirement is met if the referent of a judgment is a part (or the whole) of the speakers experience at the time he makes the judgment. On the other hand, if the referent is either (1) separated in time from a person's judgment or (2) contemporaneous with it but not contained in his experience, the referent is, I believe, epistemically separated from him: there would be a separation between him and the referent that could cause him to be mistaken about it. (Though I think that a judgment is about the given if it meets these two requirements, they are not part of my definition: it does not automatically rule against the belief that ostensive memories sometimes place us in direct contact with the past, or against there being no epistemic separation between an observer and the physical object he observes.)

Talk of freedom from epistemic separation besides being metaphorical may be somewhat vague but I think it is fair to say that if something is in a person's stream of consciousness at the time he makes a judgment about it he is not epistemically separated from it  if anything is not epistemically separated from you, surely it is your own experiences. This account of the given allows one to admit that mistakes can be made about it but it is still such that if there is a given the view that beliefs can only be tested by seeing whether they cohere with other beliefs is wrong. If there is a given, beliefs about it can surely be tested by examining it, and that is not the coherence test.

The given can also be characterized as follows: something is given for a person if he is in a position to make a judgment about it that is non-inferential except for inferential baggage regarding linguistic usage and/or comparison items. Here again experiences would be given, one could consistently admit that mistakes can be made about the given, and its existence would be incompatible with the truth of the coherence theory of truth-testing (provided at any rate that introspection is possible).

My attack against the coherence theory of truth-testing (and indirectly against the coherence theory proper) looks too easy to work: surely theories which claim such sophisticated advocates cant imply such an absurdity as that introspection is impossible. But they do. What makes the coherence theory of truth-testing so interesting and provocative is not that coherence is claimed to be one of the ways in which we can test for truth but that it is claimed to be the only one; that no belief is ever tested even in part by anything other than coherence. And advocates of the theory have been remarkably impressive in showing that one can go much further with coherence in accounting for the ways in which we test for truth than would have been thought possible. Still the coherence theory does not account for introspection: when a person judges, for example, that he has a pain, unless his judgment is extremely atypical, he is not doing so solely on the grounds that the belief that he is in pain coheres with his other beliefs.

Furthermore, if one remembers that the alternative to coherentism has been commonly thought to be traditional foundationalism with the untenable claim that experience-reports are incorrigible and the various errors which this implies, it is easier to understand why philosophers should have overlooked the crucial weakness of coherentism in regard to experience reports. Also, though many of us regard the claim that there are no introspective judgments about experiences as absurd, a number of the most distinguished coherentists would regard it instead as an important truth, either because (like Quine or Feyerabend) they think there are no experiences, or because they have been led by some version of the Private Language Argument to suppose that even if there are experiences, we lack the language to make introspective judgments about them.

Another question arises: if, as I argue, you are a foundationalist if you acknowledge the given, even if you admit that there can be mistakes about it, how important is the difference between this kind of foundationalism and Rescher's form of coherentism? A first step towards answering this is to categorize empirical judgments in terms of their relation to the given. When a person judges that he is in pain, the referent, pain, is given but he is perhaps also implying or presupposing the existence of 'comparison elements, i.e., other pains and/or facts about linguistic usage, neither of which is given. I shall call such a judgment a given/non-given judgment: the referent (here pain) is given; the comparison elements and linguistic facts are not. A coherentist on the other hand since he supposes that there is no given would classify it as a non-given/non-given judgment. Thus Quine speaks of the claim that there are experiences (and a fortiori, painful experiences) as an hypothesis [W.V. Quine, "On Mental Entities/' from Proceedings of the American Academy of Arts and Sciences, 80 (1951), reprinted in John O'Connor, ed., Modern Materialism: Readings on Mind/Body Identity (Harcourt Brace and World Co., 1969) pp. 121-23]. The customary (and I think correct) view is that whenever you are conscious an experience is given. Granted this, it seems absurd to say that it is no more than an hypothesis that there are experiences. Quine seems, however, not unhappy to hold this and goes on to urge us on grounds of ontological simplicity to reject the 'hypothesis that there are experiences [Although it is certainly odd to speak of the claim that one is having an experience as an hypothesis, a rationale can be constructed for doing so. Usually when one speaks of an hypothesis, the referent is not claimed to be given for the person making the hypothesis. However, it can make sense to speak of an hypothesis in a case where the referent is claimed to be given. Consider the following case. A synthetic tea meant to taste like chamomile tea has been produced. When A is given some, he thinks (though he is not sure) that the taste which is part of his given when he sips some is the same taste as that which he experienced at some time in the past when he drank some real chamomile tea, but he isn't sure. With some doubt he says I think that this really is the taste of chamomile tea." His judgment is clearly hypothetical despite the fact that the referent is given. A modest foundationalist could admit that though it is odd, it is nevertheless justifiable to say that the statement that one is having an experience is that kind of hypothesis].

Consider next a claim based on ostensive memory that a pain existed in the past. This would qualify for a modest foundationalist as well as for a coherentist as a non-given/non-given judgment because the referent as well as the comparison elements and the facts about linguistic usage would not be given. Some non-given/non-given judgments are supported in part by something given. If I look at a bottle and judge it yellow, the referent of my judgment, the bottle, is not a part of my given; but my claim about the bottle can be supported by a claim about my sensory experience which is given  or so I see it. A coherentist on the other hand would be forced to disagree; he is committed to holding that judgments are never supported by statements about the given. He can hold that when confronted with a yellow bottle you are typically caused to believe that such a bottle is in front of you and that part of the causal chain is a certain sort of sensory experience, but he cannot hold that you can support your claim about the bottle by examining your experience introspectively and judging that it is the right sort of experience for you to have if you are indeed looking at a yellow bottle. He cant say this because it would imply that we can make judgments about the given.

The rejection of such judgments commits coherentists to the denial of introspection and this creates pressure on them to adopt materialism. Why? Because if there are experiences it is hard to see why introspection should be impossible. (This may not be as obvious for visual or auditory sensations as for pain, but pain is enough.) Accordingly, there is pressure on coherentists to deny the existence of experiences; but if you deny that there are experiences, you have become a materialist. And recent coherentists (unlike those in the nineteenth century) have in fact tended to do just this.

3. A special status other than incorrigibility for judgments about the given

Before arguing that judgments about the given do in fact have a special status other than incorrigibility, I shall try to clear away some mistaken reasons for thinking that they do.

(i) Judgments about the given have a special status because they can be used to criticize other empirical judgments, but criticism can never go in the opposite direction. This is the traditional view which I have rejected.

(ii) Judgments about the given enjoy a special status because without them all attempts at justifying empirical beliefs would involve either a series of inferences leading to an infinite regress or a series that eventually formed a circle.

Some philosophers have argued that one or perhaps both of these alternatives are epistemically acceptable. I disagree, but this is not the place to argue the question; the point I wish to make is that a modest coherentist need not accept either of these alternatives. A coherentist must hold that you cannot apply the introspection test to your beliefs because coherence is the only test there is; but he can nevertheless avoid the regress and the circle by holding that we can have some confidence in a belief before it is tested. Foundationalists might object to this view, but they would be unable to do without it. As C. I. Lewis points out, although it is logically impossible to justify ostensive memories by the correspondence test, we must take the bare fact that we have an ostensive memory as providing at least slight grounds for thinking that it is true. The only alternative would be to abandon all claims that go beyond solipsism of the present moment. It appears then that the difference between modest coherentism and foundationalism is not that the coherentist needs to rely on some untested beliefs while the foundationalist does not. It is rather that the coherentist must rely more extensively on untested beliefs, especially in the area of perception.

(iii) Judgments about the given enjoy a special status because all justification of empirical judgments proceeds from judgments about the given. This is simply false. We often start with ostensive memories or simply with an idea coming to mind that strikes us as plausible.

(iv) For a person to have a belief about the external world he must first have sensory experiences. He may not remember the occasions on which he had the experiences but that doesn't matter. On the other hand, it is not necessary to have any beliefs about the external world in order to have beliefs about sensory experiences. Therefore we must have beliefs about the given in the form of sensory experiences to have beliefs about the external world, but the reverse does not hold true.

At least two serious objections can be raised, (a) The argument is invalid: the first premise only claims that we must have sensory experiences in order later to have beliefs about the external world, not (as the conclusion requires) that we must have beliefs about such experiences. Furthermore if the argument were amended to claim that it is beliefs about sensory experiences that are required, it would lose much of its plausibility because the new premise would be dubious at best, (b) The theory in question is a causal hypothesis rather than a theory of how our beliefs should be justified. Can a theory of justification be based on the causal hypothesis? It could be claimed that because we must have sensory experiences first in order later to have beliefs about the world we have no alternative but to start from judgements about sensory experiences in justifying our beliefs about the world. But this is simply a repetition of (iii).

(v) It could be urged instead that in justifying beliefs we can indeed start with claims about physical objects but that we shouldn’t do so. We should rather pattern the order of justification after the order in which our beliefs originally arose  the order of sensory experiences followed by beliefs about the external world  so that an ideal defense of a claim about the external world would start from judgments about sensory experiences. But even if the account of the origin of our beliefs about the external world is correct, that would not guarantee that an ideal defense of judgments about the external world would proceed in the same order. It might, but we would need a good reason for supposing that it would. It could be claimed that this is the ideal order because, even though judgments about sensory experiences can be mistaken, they are on the whole more reliable than judgments about the external world.

There are at least two reasons why this won't work. First, even if judgments whose referents are given are usually more reliable than judgments whose referents are not given, it does not follow and it is certainly not true that all judgments about the given are more reliable than any judgments about things that are not given. Consequently if you always wanted to start with the judgment in which you were the most confident, you would sometimes start with judgments about things that were not given. Secondly, once it is granted that judgments about the given can be criticized on the basis of judgments not about the given it doesn't matter particularly where you start. It would matter if you were going to stick to the judgment you started from come what may, but as long as you are prepared to drop the original judgment when it is under sufficient pressure it is pretty much a matter of convenience whether you start from a relatively weak judgment or a relatively strong one. Lets suppose that A is a relatively strong judgement and B a relatively weak one. If you start with A and discover a conflict with B, you should continue to accept A albeit with less confidence. But you would end up in exactly the same position if you started with B and found that it conflicted with A. That is, you should end up believing A, though with less confidence than if it did not conflict with B.

So much for bad reasons for assigning a special status to judgments about the given. Now for a good one. Insofar as there are reasons for holding empirical beliefs other than those arising directly or indirectly from the bare fact that certain beliefs happen to be held, they arise from judgments about the given. Or, relating this to the metaphor of a foundation, if you want something to support an empirical belief other than further beliefs whose strongest recommendation is that they are believed and that they cohere with other beliefs, you must rely on judgments about the given. Reasons which rely on such support can appropriately be called 'non-coherentist reasons' because coherentists are committed to the view that there is no such support and therefore that there are no such reasons. This yields a third way of describing the special status of judgments about the given: there are non-coherentist reasons for accepting judgments about the given and judgments supported by them but there are no such reasons for accepting any other empirical judgments.

It might be objected to my claim that there are non-coherentist reasons for accepting judgments about the given that having admitted that such judgments can be mistaken and can be criticized on the basis of judgments about the world, I must also admit that we need support from statements about the world to establish the general reliability of judgments about the given, i.e., experience-reports, even if once this reliability is established particular experience-reports need no further support from statements about the world to warrant something approaching complete confidence.

Verificationism could underlie the view that this kind of support is necessary. The claim would be that experience-reports are not meaningful unless their truth can be established conclusively; that ex hypothesi this is impossible without an appeal to other sources of information; that a statement clearly can't be reliable unless it is meaningful; and therefore that we cant show that experience-reports are reliable unless we appeal to some other source of information. This argument is valid but it relies on verificationism, a view which I think has been adequately refuted (though not everyone would agree). Rather than reviewing previous refutations, let me offer a new one. One of the many absurd consequences of verificationism is that there can be no meaningful language. Paradoxically, this conclusion can be derived from the premises of the following version of the Private Language Argument which includes verificationism as a crucial premise. A wants to add a word to his language that will stand for a certain sort of sensation  say, a pain. He has one and decides to call it an E-experience. A minute later he has an experience which his memory tells him is like the first one. Since his memory is fallible, he would like to check it but it is logically impossible for him to do so. Therefore, according to verificationism, it is meaningless to say that the second experience is like the first. But to say that an experience is an E-experience would be to say exactly that, so it is meaningless to say that an experience is an E-experience. So A has failed to coin a term in a private language.

Is the situation essentially different for a public language? Suppose you want to determine if something is a chair. You ask someone if it is. But for this to do any good you must know that you are in fact appealing to a person. Can you be certain? One is tempted to say it is obvious, but that won't do. If you could say that, you could just as well say the same about the purported pain. So you have to find out if you are indeed addressing a person. But to justify this belief, you would have to rely on your fallible memory. You could of course ask a second person if the word "person" was correctly applied to what you took to be the first person. But you would have the same problem with this second supposed person that you had with the first, and so on in an infinite regress. Thus if we accept verificationism, we must accept the absurd conclusion that public as well as private languages are impossible.

But perhaps one doesn't need to rely on verificationism to show that once we admit that experience-reports can be attacked on the basis of statements about the world, we are forced to rely on statements about the world to show that experience-reports are at all trustworthy. It might be claimed (call it the credentials claim) that if a basic source of empirical information is fallible it is wrong to trust findings issuing from that source until the reliability of the source is checked in terms of findings from some other basic source. Sensory experiences are one such basic source of information; ostensive memories are another. Accordingly, if the credentials claim holds true it must do so for ostensive memories. But if it did, we would be plunged into skepticism regarding everything beyond the contents of the present moment: if we weren't prepared to believe that having an ostensive memory that an event occurred made it at least slightly probable that it had occurred, we would have no grounds for believing that anything has happened in the past and therefore no grounds for any of the inductive arguments necessary to justify empirical beliefs that extend beyond solipsism of the present moment. Thus it would appear that the credentials claim is unacceptable.

Furthermore there is a special reason for trusting either a genuinely non-inferential judgment or a judgment about the given that is absent for any other empirical judgment. If sensation-reports can be stripped down to a point where they are genuinely non-inferential, even though such judgments are still subject to examination errors they are not (like inferential judgments) subject to errors of inference. Alternatively, if there are no completely non-inferential judgments but there is a given, there would still be an important difference between reports of the given and other empirical judgments. In judgments about the given, the referent would be given, and inferences would be limited to matters of linguistic usage and/or comparison items; whereas in a judgment not about the given, besides these inferences (which are normally so reliable that we tend to ignore them) you would need a typically less reliable inference about the nature or the existence of the non-given referent of the judgment. Thus, if there are genuinely non-inferential judgments or if there is a given, there is a class of empirical judgments that is immune to an extremely important sort of error to which all other empirical judgments are exposed [Rorty offers a principle that might be used to block this claim insofar as it deals with statements about the given. I need to distinguish errors arising from a faulty examination of the given (as in the after-image case) from errors about what I have called inferential baggage. Rorty suggests (op. cit, p. 204) that if we do not have a way of determining whether mis-examining or misnaming obtains, we do not have a genuine contrast between them (and he would presumably extend this claim to the other contrasts). His principle can be interpreted in two ways: it can be taken as requiring verification that a given alternative obtains or as requiring no more than confirmation. If it is construed as requiring verification, besides the fact that verificationism has been refuted more than once, his principle implies that it wouldn't even make sense to claim in the after-image case that the person miscounted the sides of his image rather than misspeaking, e.g., saying 'twelve' when he meant 'thirteen' , because we couldn't make absolutely sure that he was doing one rather than the other. But surely the distinction in question makes perfectly good sense. If on the other hand confirmation is all that he requires, the various distinctions will stand because there are many times when we can confirm the claim that one sort of error rather than another has been made. For example, in the after-image case one could ask the person whether he had misspoken, whether he had really meant that his image had twelve sides. And if he said that he meant it, and you thought that he might have made the more serious error of thinking that 'twelve' means 'thirteen', you could give him time to count the sides of a twelve-sided object so that an error of inspection was extremely unlikely and see whether he thought that 'twelve" was indeed the right name for the number of its sides. In neither case would you have verified your diagnosis of his mistakes but in both cases you would have confirmed it].


  1. It is sometimes reasonable to abandon a belief about your own sensations because of your beliefs about external objects and about general laws.
  2. This admission requires abandonment of six out of seven views traditionally association with foundationalism.
  3. This admission is nonetheless compatible with a modified form of foundationalism because it leaves room for the view (a) that introspection as well as coherence is a test for truth, (b) that there are non-inferential judgments and/or judgments about the given, and (c) that although sensation-reports are not infallible, they still have a special status in the justification of our beliefs about physical objects and general laws because they offer non-coherentist support for such beliefs and because fallible though they may be, they are immune to an important kind of error to which all other empirical judgments are exposed. Each of these three claims is incompatible with any form of coherentism.
I have argued in addition that (a), (b), and (c) are correct, though I haven’t proved this because all three rest on the supposition that there are experiences, and I haven’t provided the analysis needed to make it clear that experiences are a kind of thing whose existence is as obvious as it seems. Finally, I have shown that there could be a given even if there were no completely non-inferential judgments, and that it is virtually impossible to deny that there is a given and the opportunity to apply introspection as a test for truth unless, unlike Hegel, you are willing to follow materialism in its denial of experiences.

[This paper has benefited from comments and criticism from Jonathan Bennett, Steven Savitt, Jack Stewart and EarlWinkler]

Foundations without Certainty
Author(s): R. I. Sikora
Source: Canadian Journal of Philosophy, Vol. 8, No. 2 (Jun., 1978), pp. 227-245
Published by: Canadian Journal of Philosophy
Stable URL: