Chapter Four

THE MYTH OF THE MYTH OF THE GIVEN

* * *

4.1 Why appeal to a given?

4.2 Characterizing the given

4.3 Arguments against the existence of a given

4.4 Is perception an inferential process?

4.5 Is perception theory-laden?

4.6 Can the given do its job, supposing there is one?

* * *

Justification always terminates with other beliefs and not with our confronting 'raw chunks of reality', for that idea is incoherent.[1]

4.1 Why appeal to a given?

Foundationalists make two distinctive claims about justification: that it is terminal and hierarchical. Anti- foundationalists attack both of these claims. Such attacks pose a more general threat to foundationalism than do the attacks on the representationalist versions of foundationalism considered in Chapters 2 and 3, since not all versions of foundationalism are representationalist, though all necessarily make hierarchy and termination conditions for justification. Attacks focused upon the hierarchy condition will be considered in Chapter 5. This Chapter is focused upon attacks on the termination condition.

To satisfy the termination condition, foundationalists typically appeal to sensation, perception, or experience as the justificatory source of basic propositions. The "given," as any such source is generally called, is held to provide a form of non-inferential, non-propositional justification for basic propositions. The notion of a given, however, has been seen as extremely problematic; as mentioned in Chapter 1, the majority of contemporary philosophers believe the idea is incoherent. Several related questions focus the problem. Since what are to be justified are propositions, does the given — the terminal sensory or perceptual states — have to be propositional? If sensation or perception is a nonpropositional state, then can it be a cognitive state? Whether cognitive or not, can a non-propositional state justify a proposition? And can such a process — if it is a process — be non-inferential?

The attack on the given is posed as a dilemma for the foundationalist. Either there is a given or there isn't. If there isn't a given, then obviously foundationalism is dead in the water. So the first prong of the attack is a set of arguments intended to show that the whole notion of a given is unviable. The other prong argues that even if there is a given, it cannot do the justificatory task foundationalists need it to do. In either case, it follows that foundationalism is untenable, for there are no terminal points of justification. My concern in this Chapter is to lay out and respond to both prongs of the attack, thus defending the views that there is a given and that it can indeed provide a form of justification for basic propositional beliefs.

4.2 Characterizing the given

The given is supposed to be a preconceptual sensory or perceptual state of awareness.

One is aware, but one's awareness is not abstract or propositional in any form.[2] One is aware of a green patch, for example, but one is not aware that it is green — one simply sees something green or sees greenly, depending on the version. Sensations or perceptions are the stuff out of which abstractions, concepts, propositions, and theories are formed, but they are not themselves abstract, conceptual, propositional, or theoretical. The given is also supposed to be given, i.e., the subject is supposed to be essentially passive with respect to it, and not its active formative agent.

Beyond these two features, there is no uniform characterization of the given among foundationalists, for, as may be expected, they disagree about exactly what features the given has and which are to be considered essential. There are epistemological differences. In Chapter 2, for example, we considered versions of foundationalism that characterize the given as placing the subject in an epistemologically indirect relation to external objects, and in Chapter 3 we gave reasons for adopting instead a characterization of the given as an epistemologically direct relation of subject and object. There is also disagreement over to what extent the information provided by the given is integrated. Is the given the "blooming, buzzing confusion" that William James spoke of[3] — or is it sensations automatically integrated into perceptual states?

The question is important to foundationalism, for if the given is, for example, discrete point sensations, then the basic propositions will be on the order of "Here now red," "Here now bright," and "Here now round," and from propositions such as these a lot of work needs to be done before one can derive justified propositions such as "Here's a bright red ball." One is then committed to, roughly, the Carnap/Neurath project. If, by contrast, the given is automatically integrated perceptions of entities with all the normal constancies of distance, size and shape, then the foundationalist project is much more straightforward. One can proceed as do Gibson and Kelley.

These internal disagreements among foundationalists connect directly to the two main questions about the given raised most often by antifoundationalists. The first addresses the passivity claim: To what extent can the claim that the subject is passive with respect to the given be upheld in the light of the fact that some degree of processing by the subject is necessary before awareness of the given occurs? Should the processing be characterized as registering or constructing an object of awareness? The second addresses the claim for the preconceptual status of the given: To what extent can the claim that the given is nonconceptual be defended in the context of charges that perception is theory laden or the result of inferential processes?

The connection of the three issues is this. The given is either discrete sensations or integrated perception. If it is sensations, then the question is how those discrete sensations become integrated into the normal adult perceptions of entities. The almost universally suggested methods are logical and computational. If this is correct, then the question is how the logical or computational methods are acquired by the subject: at the purely sensory level subjects have no perception of entities and have yet performed no spatial or temporal integrations. It then seems that the foundationalist would be committed to at least some minimal form of nativism: the logical and computational methods are then innate and are applied subconsciously and automatically. But if this is so, then the adult's normal three-dimensional perception of entities with constant size and shape is informed by higher order logical and computational processes. So it seems, on this line of reasoning, that the subject is hardly passive; it also seems hard to see how what is given to the normal adult perceiver can be preconceptual.

On other hand, we could say the given is the perceptual level with constancies. But the perceptual process starts with discrete sensations, and therefore the normal adult three-dimensional perception of entities with the constancies must be a derived product: it must be a result of the integration of discrete sensations. If it is a derived product, then again it seems to many that it must be derived by means of logical/computational methods. And if so, then again it seems that the subject is hardly passive and that the normal adult perspective on the world is shot through conceptually. Either way it seems that the concept of the given is in trouble.

The phenomenological and experimental evidence is on the side of saying that the perceptual level of awareness is the given. When I look about me, to take the case of vision, I see entities integrated over space and time. I do not see discrete color, shape and texture units, which I then make an effort to integrate in order to see ordinary objects. The integration seems automatic from the phenomenological perspective, and indeed it takes special effort on my part to reduce my visual awareness to a flowing stream of colors and shapes. It could be responded that for an adult the perceptual level of integration seems natural only because in infancy the integrative methods were learned and automated.[4] This response, however, runs afoul of the available experimental evidence. T.G.R. Bower's famous experiments involving infants from two to twenty weeks old demonstrated that they are able to recognize and reidentify objects as the distance of the object changes and, accordingly, as the size of the retinal image changes, and that they are able to do so as quickly as an adult — thus indicating that they are not engaged in a process of consciously constructing an ordered perceptual world on the basis of discrete and fragmentary sensations.[5] The best conclusion seems to be that the integration of sensations into perceptual awareness with the usual constancies is the result of automatic processing.[6]

For these reasons I will reject the position that sensation is the given and proceed on the premise that the given is the perceptual level of awareness. In short, on this point I agree with Lewis:

we should beware of conceiving the given as a smooth undifferentiated flux; that would be wholly fictitious. Experience, when it comes, contains within it just those disjunctions which, when they are made explicit by our attention, mark the boundaries of events, "experiences" and things.[7]

In this context, the question that brings us to the heart of the issue is: Where do these "boundaries" in our experience come from? And are they compatible with the claims that the given is both passively received by the subject and preconceptual? I think they are. The conclusion reached so far is that the given is a state of direct perceptual awareness; in the rest of the Chapter I am concerned to establish that the given is a nonconceptual and nonpropositional state, and that even though the subject actively processes sensory data, the resulting perceptual awareness can still be said to be passive in the sense required by foundationalism.

4.3 Arguments against the existence of a given

The first attack on the given is to attack the radical distinction between determinate, particular, non-abstract perception and abstract, propositional thinking. The distinction, it is claimed, is either non-existent or not as sharp as the foundationalist needs it to be. Instead, it is claimed that what one perceives is dependent, as Kant argued, upon one's conceptual apparatus. One's concepts — abstract and propositional in form — determine what form one's perceptions take, which means that perception is to some extent "shot through" conceptually. Therefore, we should cash out perception as being to some extent abstract and propositional (although perception will not necessarily be as abstract as other parts of our mental life — as, for example, our scientific theories are). But, and most importantly, this means that perception becomes far more subjective. Rorty, agreeing in essence with Kant on this issue, notes that "Kant's point [was] that to change one's concepts would be to change what one experiences, to change one's 'phenomenal world.'"[8] In the 20th century, Kant's point generally is offered in one of two forms: perception is either theory-laden or formed at a deeper, less changeable level by one's linguistic habits.[9] In either case, the resulting percept is more than merely a recovery response to the available stimulus: it goes beyond the stimulus by being either shaped, added to, or distorted by one's concepts.

But if one's concepts determine one's percepts, then one's perceptual awareness — i.e., what one takes to be the given — cannot do what foundationalists require of it. Since it is already dependent for its content and form upon one's background conceptual structure, it can no longer, upon pain of circularity, provide justification for that conceptual structure. Therefore, we must account for justification solely in terms of internal coherence relations among the interconnected and interdependent conceptual parts. There is no concept-independent starting point — i.e., no given — and so there is no way for a foundationalist hierarchy to get under way.

Initially the distinction between perceptual and conceptual modes of awareness seems clear cut. Perception is of particulars that are determinate in all their dimensions. For example, to perceive a human is to perceive a particular individual of determinate height, color, shape, etc. Conception, by contrast, is abstract: one considers a class containing an indefinite number of particulars, each member of which is considered as a unit with indeterminate measurements. For example, when conceiving HUMAN, one is not thinking of any particular human, nor of any determinate height, color, shape, and so on.[10]

The antifoundationalist challenge to making this a clear cut distinction follows the following general pattern:

(a) If one has a sensory-perceptual experience of X, then one is distinguishing X.

(b) If one distinguishes X, then one distinguishes X as something.

(d) Therefore, if one has a sensory-perceptual experience of X, then one must be using concepts.[11] The key and controversial premise is (b).

In Sellars's words, with his own emphasis, the point is that "to have the ability to notice a sort of thing is already to have the concept of that sort of thing."[12] And the reason for this seems to be the belief that, to use Scheffler's words, "[w]ere it not so, indeed, we should be powerless to take hold of anything in experience: equally receptive to everything in awareness and uniformly undiscriminating, we could not properly be said to observe anything at all; at best we should confront a flat and undifferentiated given ...."[13]

Two arguments are used to justify these conclusions. The first is that perceptual states must be informed by concepts to some extent, because the needed integration of discrete sensations can only be accomplished by inferential, interpretive, or computational methods. The second is that certain informal experiments demonstrate that perception is theory-laden. In the following two sections, let us take up these arguments in turn.

4.4 Is perception an inferential process?

The claim that perception is an inferential process has a long and distinguished history. It is not hard to see why. When I look about me, I see entities — the terminal in front of me, the water glass beside it, the curtain framing the window across the way. These things appear before me as unitary items, and I am aware of no special effort required on my part in order to perceive them: I simply look and there they are. The same holds for other sensory modalities, such as hearing. The slight whir of the computer on the desk is to my ear distinct from the mildly annoying buzz of the fan behind me, and both are distinct from the sound of the dog's bark coming from somewhere outside the room. Yet a rudimentary investigation into what makes possible such seemingly straightforward awareness of distinct objects reveals that an enormous amount of information needed to be processed before my awareness of objects could occur. Take visual perception as an example. Not only am I aware that the water glass is before me, I am aware of it as being a certain shape, as being a certain color, as having a pattern on it, as being a certain distance from me, as being a certain texture, and so on. I am also perceptually aware of it as enduring through time, for each moment is not for me a completely new experience: the features of the glass are not only integrated into a unity at a given moment; they remain integrated as a unity over time. All of this was the result of the processing of the effects of energy patterns on the receptors in my eyes, and from this it seems that what appears to me as a unitary item — the water glass, with all of its specific features — can be seen as the product of the integration of discrete bits of sensory stimuli.

If we ask by what means the integration was accomplished, then a natural solution to many is perceptual inferentialism — the thesis that the data of the senses are processed along lines modeled upon conceptual processes, that is, by means of calculations, interpretations, hypothesis formation and testing, and other computations. According to perceptual inferentialism, the perceptual awareness of objects is the conclusion yielded by such processes operating upon sensory stimuli. "Conclusion" here is meant literally; von Helmholtz states the thesis clearly:

we really have here the same kind of mental operation as that involved in conclusions usually recognized as such. There appears to me to be in reality only a superficial difference between the "conclusions" of logicians and those inductive conclusions of which we recognize the result in the conceptions we gain of the outer world through our sensations.[14]

The thesis is not necessarily that the inferential processes are consciously performed, for that would fly in the face of the phenomenological evidence: I do not notice any sort of computation, inference, or hypothesis-formation when I perceive. One simply opens one's eyes (in the case of visual perception) and sees things — no conscious computation or inference needed.[15] I did not figure out, in any sense, that there are pencils and paper on the desk along with the terminal and water glass. And when my dog came into the room, I first simply heard and then saw her presence.

Empiricist inferentialists respond to this evidence with a developmental claim: the inferential processes involved are learned in infancy and then more or less automatized. As infants we performed these processes more slowly as we learned to sort out the blooming, buzzing confusion of sensory stimuli presented to us; but as we learned the processes they became effected increasingly automatically, until as mature perceivers we perform the processing entirely at the subconscious level. Nativist inferentialists, by contrast, hold that the methods are hard-wired in and performed entirely automatically and subconsciously. As a rule, whether empiricist or nativist, contemporary perceptual inferentialists do not claim that the inferential processing is consciously accessible to adult perceivers.[16]

The empiricist's developmental claim, however, runs afoul of the data from Bower's experiments, as mentioned above. Infants as young as two weeks perceive distance as quickly as adults.[17] The empiricist inferentialist could, rather hopefully, conjecture that the ability to calculate distance must be learned and automated in the first two weeks. This possibility is not, to my knowledge, ruled out by experimental evidence. Yet the empiricist inferentialist then has the difficult problem of explaining how such young infants acquire both the necessary knowledge and the facility in applying it for computing distances.

Perceptual illusions also pose problems for empiricist inferentialism. Consider the case of a perceiver subject to the illusion of the straight stick appearing bent when partially submerged in water. The stick's appearance is, on the perceptual inferentialist analysis, a conclusion based on the given sensory stimuli and the application of certain laws of perspective. On empiricist perceptual inferentialist grounds, the ability to apply those laws of perspective is learned. But if the methods are learned, then it seems they could also be unlearned or supplemented by others, so that the perceiver is no longer subject to the illusion. Suppose, for example, that the perceiver comes to be aware that he or she is viewing the stick under unusual circumstances. Further suppose that our perceiver feels the stick and comes to know that it is not really bent. Further suppose that the perceiver becomes acquainted with the laws of light refraction and what they mean for sticks viewed when partially submerged. Now, on empiricist perceptual inferentialist grounds, these new data should become integrated with all of the original data that gave rise to the illusory perception, so that if our perceiver again looks at the stick in the water, the new data would enter into his calculations, and he would no longer be subject to the illusion. This is never the case.[18] No matter how much one knows about refraction, the properties of water and light, and no matter how much one concentrates while perceiving so as to not lose awareness of the circumstances, the stick will still appear to be bent. Hence, empiricist perceptual inferentialism cannot be correct, and, rather, it must be that the perception is the result of functions performed automatically and in isolation from any higher-order conceptual input.

This leaves us only with the possibility that the integrative methods for perception are hard-wired in. But from this I do not think it follows that the nativist version of perceptual inferentialism is correct.[19] Nativist versions of perceptual inferentialism hold that what is hard-wired in should be interpreted as being modeled upon conceptual knowledge of inferential and computational methods, and, accordingly, that the automatic and subconscious processing that occurs during perception is actual inference and computation.[20] Here I think we need to be especially careful. First, we must distinguish computable and computational. A thermometer's activities are computable, but not computational: it registers changes in temperature, but it doesn't compute them. Computable is the broader concept, designating any process that can be mathematically modeled. I expect that all perceptual processes are computable, in the sense that a sufficiently sophisticated physiology of perception will eventually be able to model them. Computational, however, is a narrower concept, applicable when a person (or machine, possibly) is actually computing. And computing occurs only in cases when there is actually something to figure out — when one needs to go beyond the available information to reach the conclusion. Nativist inferentialists make the stronger claim that perception is computational, and this is because they hold that perception requires more than the registering of the information detected by the senses. What is given by the senses, on this view, does not force any particular perceptual hypothesis and so must be supplemented by calculations that determine which perceptual hypothesis best accounts for the sensory input.[21] Here we reach the key premises of perceptual inferentialism: as perceivers we are tuned only to the lower levels of information in the sensory array and, thus, by means of computational methods the subject must add to that information and construct the resulting percept. And this is where I, following the lead of Gibson, Bower, and Kelley, disagree with the nativists.

Contrary to perceptual inferentialism, I think that the information available in the sensory array is not (at least not in the huge majority of cases) in need of supplementing, and that perception is the result of physiological integrations of the information in the sensory array. Let us take up these two points in turn.

For the first point, I rely heavily on the "ecological" approach to perception, pioneered by perceptual psychologist J.J. Gibson. In his two main works, Gibson has given a detailed presentation of the incredibly rich array of structured stimulus features that are available to the human perceptual apparatus.[22] His key concept, in the case of vision, is that of the "ambient optic array": the structured light that surrounds each observer.[23] "Structured" is the opposite of "unstructured," which for Gibson means homogeneous, without differences of intensity in different parts.[24] The light that reaches us directly from the sky is unstructured: its rays are parallel. By contrast, the light that reaches us via the earth is structured by the local environment: the rays are not parallel or of equal intensity in different parts. Physical reality has structure at all levels, Gibson argues, and at the local level one's environment scatters light according to its structure. This "scatter reflection" creates structured ambient light, a set of solid angles of texture and intensity that is the ambient optic array.[25]

Gibson is also concerned that perception not be treated in terms of any static model. Vision, for example, is not a series of snapshots observed by a stationary observer; the ambient optic array should not be treated as if it were "frozen in time and as if the point of observation were motionless."[26] The point of observation is not motionless because observers are almost constantly moving in the structured array: shifting their eyes, turning or cocking their heads, walking around. As one moves through one's environment, the ambient optic array makes it possible for one to see "a continuous family of perspective transformations, an infinity of forms." Yet in the changing array, there are certain invariants due to the way the local environment has structured the ambient light. Gibson's hypothesis is that "the invariants in a family of transformations are effective stimuli for perception."[27] Accordingly, much of the work done by Gibson and those he has inspired is psychophysical experiments to determine how objects structure the optic array and how invariants in the changing structure of the optic array relate to perceived higher-order constancies of distance, shape, slant and surface orientation.

The point for us is that if Gibson's hypothesis is right, there is enough information available in the array to an observer. If what is presented to the senses is a highly structured, highly informed array, then perception will not necessarily have to be a matter of subconsciously adding to the sparse data in the sensory array or of fitting those sparse data to a fuller, hypothesized form before conscious perceptual awareness can result.[28] This will make possible the rejection of the nativist inferentialist view of perception in favor of the view that perception "is a matter of differentiating what is outside in the available stimulation, not a matter of enriching the bare sensations of classical stimulation."[29] But since the ambient array is richly structured, higher-order perceptual mechanisms are necessary to detect that structure. Here, then, is the second key claim of the ecological view: Rather than being tuned only to the lowest level of stimulus information, that of individual sense receptors, a perceptual organ is a system tuned to more sophisticated patterns in the array.[30] These more sophisticated patterns in the array are, as expected, those that make possible the awareness of the perceptual constancies of size, shape, distance, and so on. Structures among individual sensations are what perceptual systems are sensitive to, and perception, on Gibson's model, is matter of the entire organ's being activated to recover the information in the array registered by the individual receptors.[31]

Even here perceptual inferentialists could insist that the processing involved must be computational, by means of the premise that all processes of integration must be computational. Yet clearly, not all processing is computational: my stomach processes food. And not all processing involving integration is computational: consider the operations of mechanical devices such as carburetors or extrusion molds. And not all processing that involves the integration of information has to be computational: many processes in the nervous system involve integrations and disintegrations, including those in perceptual systems. To cite one example, in 1962 neurophysiologists Hubel and Wiesel reported discovering cells in the visual brain of a cat that respond to specific patterns in the stimulus array. The existence of such cells demonstrates the existence of processes of neurophysiological integration.[32] Accordingly, purely physiological accounts of the information integrating processes in the perceptual systems are entirely possible,[33] particularly if the information available in the sensory array is, as Gibson puts it, "inexhaustible." To assume in advance, as many perceptual inferentialists seem to do, that all processes involving the integration of information must be computational, is to beg the question.[34]

In this context, Kornblith's comment that Gibson has given foundationalists a big part of what they have always wanted makes a lot of sense.[35] The ecological approach allows us to say that the perceptual level of awareness, with its usual constancies of size, shape, and distance, is what is given. This squares with the phenomenological and scientific evidence. At the same time, it gives us an explanation of the integration of the information provided by discrete sensations while avoiding the problems of both inferentialist empiricism and nativism. There is no need to explain, as the empiricist version must, how two-week old infants learn how to compute distances with such ease and accuracy. And there is no need to explain, as the nativist version must, where the background interpretive knowledge came from and why we should think it has anything to do with the way the world is. By rejecting the premise common to both versions — that perceptual awareness requires interpretive, constructive processing — the foundationalist who follows the ecological approach can claim that the given is free of background interpretive processes.

We have thus disarmed one of the attacks on the given. Let us turn to the other.

4.5 Is perception theory-laden?

The second route to the conclusion that the given cannot be preconceptual relies on experiments designed to show that perceptual discrimination is theory-laden. While nothing so sophisticated as an entire theory need be involved, the experiments are intended to show that at least some background conceptual apparatus must be.

As in the case of perceptual inferentialism, it is not claimed that the process of conceptual informing is conscious. The constituting or informing is held to take place entirely subconsciously. So it does not help the advocate of the given to respond by saying that even if perception is interpreted and theory-laden, there must be something X that the subject S interprets as F, and that X is what is given. This response is not helpful because the claim is not necessarily that there is no subject-independent raw material out of which the phenomenal world is constructed. Rather, the claim is that S has no access to the raw material. S only gets the finished, conceptually informed product and so cannot distinguish what is raw material from what is interpretation. And since S has access only to the product, what is given is useless for building foundationalist justificatory structures, dependent as it is upon those very conceptual beliefs the foundationalist wants to derive from it.

First we need some criteria to determine whether perception is theory-laden. Two criteria can be stated, the satisfying of either implying that perception is theory-laden: (1) what is given is an undifferentiated field; (2) some feature of the resulting percept is not in the stimulus array. Item (1) would lead to the conclusion that perception is theory-laden by means of the premise that in order to discriminate objects in the undifferentiated field the subject would have to apply conceptual criteria. But from the results of section 4.4, it seems clear that the ambient optic array presents highly differentiated stimulus and that our perceptual systems are sophisticated enough to register this; so there is no reason to accept (1). This leaves item (2). If we determine, presumably from a scientist's third-person perspective, that what a subject claims to perceive is not matched by features in the stimulus array, then it seems clear that the subject has either added something to or distorted the actual stimulus. We can then investigate the subject's particular set of background conceptual beliefs to determine which must have been operative.

N.R. Hanson's 1958 classic discussion presents two sets of cases that are intended to meet this criterion. For each set the claim is that the subject's resulting perceptual state can only be accounted for by assuming the application of background concepts. Let's call the two sorts of cases the Sparse Data cases and the Sophisticated Identifications cases, and take them up in turn.

In Sparse Data cases the subject is presented with a line drawing and asked what he or she sees.[36] The line drawing has been carefully constructed so as to be ambiguous. When presented with a drawing of the Necker cube, for example, one can see a cube oriented in either of two different positions. In Jastrow's duck/rabbit one can see, appropriately enough, either a duck or a rabbit. In E.G. Boring's drawing one can see either the face of an old woman wearing a kerchief or a more elegantly dressed young woman with her face in profile. Let's take the old/young woman drawing as our working example.

When subjects are presented with the drawing, either the old or the young woman pops out at the subject initially, with no prompting from the person running the experiment. But since the presented data are ambiguous, some subjects initially report seeing the old woman while others initially report seeing the young woman. Typically, the person running the experiment then asks, "Can you also see the X?" — where X is the one not seen by the subject initially. And after some casting about, the subject usually reports success.

Hanson asks what accounts for the difference between perceiving one or the other. The traditional empiricist's claim is that such cases are to be accounted for in terms of separate acts of perception and interpretation: subjects perceive the same thing, but interpret it differently.[37] But this does not seem right, phenomenologically. Hanson points out that neither he nor any of the subjects is aware of separate acts of seeing and interpreting or even of a unified act of seeing-plus-interpreting. Rather, the interpreting seems built into the act of awareness — the seeing of the old or young woman comes in one movement.[38] Hanson's conclusion gathers further support from the ability of the subjects to see the other perspective only after being directed to look for it. The subject then has something in mind and tries to fit the data to a preconceived notion. In most cases the success in perceiving the other perspective would not have happened other than by being guided by concepts. Hence, Hanson concludes, perception is theory-laden.

Taking Sparse Data cases as illustrative of the way perception work usually rests upon a certain assumption about perception: that perceivers are presented only with ambiguous two-dimensional images.[39] Gregory also makes use of the Necker cube and the Boring drawing, and is explicit about this assumption. "The retinal image," he states, "gives no hint that the object is three-dimensional," yet the subject has only the two-dimensional retinal image to work with. This poses an "acute" problem for the subject, "because any two-dimensional image could represent an infinity of possible three-dimensional shapes."[40] On this premise, stylized and ambiguous two-dimensional drawings reveal the essence of perception, apparently forcing the conclusion that perception is theory-laden. The important question is whether the ambiguous-two-dimensional-image premise is accurate. And from what we have seen above, one of the central results of the ecological approach to perception is that this premise is false. The eye is not a camera, and perception is not a matter of piecing together two-dimensional photographs. Perceivers live in a three-dimensional world, and in the normal case they are presented with a structured array of light which constantly shifts as the object or subject changes position or as the light itself changes. In the Sparse Data cases, by contrast, no eye or head motions can make any difference; since the third-dimension does not exist in the object, no new information is or can be forthcoming. The whole thrust of the ecological approach is that this is the wrong way to approach an analysis of perception — that taking the Sparse Data cases as typical is the equivalent of judging out of context.

Suppose, for purposes of illustration, that we could construct a real situation that caused a momentary retinal image exactly the same as that caused by the Boring drawing. By focusing solely on that retinal image at that instant, one could guess that the object is either an old or young woman. But the next moment would remove all ambiguity: the subject or the object would shift slightly, the perceptual system would have more data, and only one perception could result. It is only by freezing a single perspective in time that any ambiguity exists. But again, perception is not a matter of inspecting two-dimensional perspectives frozen in time. The ecological conclusion, therefore, is that such cases teach us little about perception.

But they do teach us something. The phenomenon of concept-guided perception is a real phenomenon, and the relationship between perception and conception is not mechanical and not necessarily entirely modularized. Suppose, for example, that one initially perceives the young woman in the Boring drawing. Then, prompted by the experimenter, one tries to see the old woman. "Instead of looking at this line as the chin and jaw line of a young woman," the experimenter suggests, "try to see it as a big nose." Guided by this conceptually communicated information, one has in mind what one is looking for — then suddenly something "clicks" and one perceives the old woman.

Perhaps in order to see why theory-ladenness does not follow from such cases, it will help first to consider a related sort of case in which the difficulty is not ambiguity and a lack of data, but rather a glut of data — i.e., a case in which the data are overwhelming and one cannot perceptually discriminate an object one has reason to believe is there. Suppose two birders are out for a walk in a forest. One asks the other, "Do you see the catbird perched in those brambles?" The other reports that he cannot. "Look just to the right of that yellowish flower with a petal missing," the first birder suggests. Aided by this information, the other birder succeeds in spotting the catbird. In such cases, does the ensuing percept go beyond what is given in the data? Does the conceptual guidance construct a catbird or squeeze the data into the form of a perceptual catbird? Clearly not: If the second birder perceives the catbird, it is because there is a real flesh and feathers catbird there. The conceptually communicated information from the other birder merely directed the spatial focus of the second birder's perceptual system. Then once the perceptual system is focused roughly in the appropriate region, it has enough resources in the stimuli to register the catbird.

In parallel, in the Boring drawing one sees all and only the lines that are there. But because the data are structurally ambiguous — i.e., they structurally parallel the momentary patterns of information two different three-dimensional objects would present — the effect of the conceptual guidance is to suggest a pattern in a portion of the array to take as a perceptual anchor point, so to speak. But that anchor point feature is really there in the array, and since the Boring drawing does contain the minimal amount of information needed to trigger the perceptual mechanism sensitive to that sort of pattern, something "clicks" — i.e., the perceptual mechanism is automatically triggered and the other percept occurs. Yet this is not a case of seeing something that is not in the stimulus array. It is not a case of one's concepts adding to or distorting the available information, as the theory-laden conclusion requires. Suppose, by contrast, that the experimenter asks the subject, "Do you see the beauty mark on the young woman's cheek?" If there's nothing in the data that could plausibly be seen as a beauty mark, the subject will not see it. The background, conceptual expectations cannot add to the data in that sense. At most they can draw one's attention to features that are in fact there or help reorient one's perceptual focus.

There is a related phenomenon that highlights the seemingly problematic automatic nature of the perceptual discrimination involved in such cases. Surrounded by the general hubbub of noise at a party, one's ears will prick up at sound of one's name being spoken across the room. When one buys a new car, chances are one will suddenly notice cars of that make everywhere. Now, in such cases there is no conscious effort to perceptually discriminate something, in contrast to the cases of the birders seeing the catbird and the subject trying to see the old woman, in which conscious effort is very much involved. The common element is the fact that in each case the perceptual discrimination "clicks" more or less automatically in relation to one's background concepts. In raising such examples we enter the territory of the argument based on Sophisticated Identifications, so let us now turn to the issues involved there.[41] I will argue that in such cases one has a certain mental set which automatically draws one's perceptual attention to something. In some cases one's conceptual identification of the thing may also follow automatically. But the data are there, so these are not cases of adding to or distorting those data, and so such cases are not candidates for demonstrating the theory-ladenness of percep- tion.[42] One can simply program oneself to notice, to focus upon, and to identify, more or less automatically, certain perceptual stimuli.

Suppose a microbiologist looks into a microscope and reports seeing e. coli bacteria. I look into the same microscope and report seeing vaguely undulating blobs. The microbiologist, when asked, does not report anything like seeing-and-then-interpreting. To the microbiologist the perceptual discrimination and conceptual identification of the phenomenon feels automatic. His or her perceptual experience and the words "e. coli" springing to mind were simultaneous. More within my area of expertise, I look to the right and, it seems to me, simultaneously perceive books and identify them as such. Or I look out the window and, apparently simultaneously, perceive and say to myself "neighborhood kid." Concepts are constantly springing to my mind and words to my lips simultaneously with my percepts.

Now if the traditional perceiving-plus-interpretation line is right and perception is not theory-laden, then it seems there should be more independence of percepts from concepts.

To account for the phenomenological evidence, one could happily grant that perception can be theory-directed. What one notices can, in many cases, depend on past learning. One's concepts can direct one's perceptual focus. A mechanic looking at my car's engine may have his attention drawn to the oily fluid dripping from the head gasket. I may have my attention drawn to the subtle colors of the spark plug wires. If it is merely a case of having one's attention drawn, then we could agree with Hanson's transitional explanatory point: "The elements of their experiences are identical; but their conceptual organization is vastly different."[43] That is certainly true.

But Hanson and the advocates of theory-ladenness have something stronger in mind than mere theory-directedness: "theories and interpretations are 'there' in the seeing from the outset." Hanson claims that the theories and interpretations are there in two ways. First, "[o]bservation of x is shaped by prior knowledge of x." "Shaped" implies an organization imposed upon raw material that results in an organization different from that started with. So if theory shapes observations, then it must be that it organizes the raw sensory stimuli into a structure different from what was received. Second, Hanson claims that the theory and interpretation are there by incorporating into the perceptual experience more than is given by the stimulus array. For example, he remarks that if a schoolboy and a physicist are viewing a glass X-ray tube, "both see that the X-ray tube will smash if dropped."[44] Dropping and smashing are not given in the stimulus array but, according to Hanson, they are part of the perceptual experience. So the seeing is both organized and added to by one's concepts.

Is there an alternative explanation of the phenomena? Or are the only alternatives traditional empiricism's complete modularization of perception and conception/interpretation, and theory-ladenness?

While it is certainly true that background expectations and interests can influence what features of one's perceptual field will draw one's attention, what one will focus on, and what one will remember, these facts are entirely compatible with the traditional modularization thesis that the content of one's perceptual field is independent of one's concepts and theories. First, some evidence for modularity.

Earlier we noted that no amount of background theory, however firmly believed, can change the experience of a perceptual illusion. If a stick is partially submerged in water, the resulting experience is a function of the features of the ambient optical array and physiological integrations of sensory stimuli. Changing one's concepts, hypotheses, theories, or expectations changes the perceptual experience not a whit: the stick appears bent to the same degree it did before the conceptual changes. Or suppose one is told one is looking at a white object with a green light shining on it. It appears green. Yet one's theoretical knowledge is that the object in normal light is white and that a green light is shining on it. So even though one believes the object is "really" white, no matter how hard one tries this theoretical knowledge cannot influence one's perception. This points to a basic level of independence of perception from background theory. As additional confirmation, if we place a really green object beside the white one and have the green light continue to shine on both, instantly the white one will no longer appear green but white and the green one will appear green. While attending to both objects now, there is no way one can make the white object appear green, even though one knows it appeared green shortly before and that a green light is shining on it. Theory has nothing to do with the perception. Yet if perception were truly theory-laden, these are not the expected results.

A more commonplace situation is this. Suppose I am watching out the window, expecting a friend to arrive at 3:30 p.m., who I know drives a white Blazer. At 3:30 I look out my window and see a white truck turn into my driveway — "There she is," I say to myself. In the next moment, though, I look again and see that it's really a Ram Charger, a truck that looks very much like a Blazer. Such a case counts against the theory-laden view, for here we have a case of perceptual states dictating to, overturning, and reformulating conceptual states. The percept is not shaped by my expectations, wishes, or beliefs. In this situation the background conceptual states are as strong as can be: I expect to see a white Blazer, I want to see a white Blazer, and initially I believe I am seeing a white Blazer; but these background conceptual states are overridden in the face of the next moment's percept.

So, it cannot be the case that the content of perceptual states is shaped or added to by conceptual ones. Whether the perceptual state occurs depends primarily upon the information in the stimulus array; then, supposing the subject is paying attention, the most the conceptual states can do is guide the focus of the subject's perceptual states. It is also true that the conceptually guided focus may become automated with learning: my mechanic may not be able to help noticing the frayed fan belt when looking at my car's engine; a philosophy instructor may not be able not to notice the word "existence" spelled "existance" each time it appears in a student's paper; a music teacher may not be able to help noticing the sound of her student pushing her voice from her throat and not her diaphragm. Depending upon one's past learning, one's attention may be drawn automatically to specific structures in the array when they appear.[45] None of this implies shaping or adding to the data. Automating perceptual identifications is simply one form of learning.

It is further true that the words identifying these phenomena may come to mind with no conscious bidding. Phenomenologically, the conceptual identification of what is given perceptually is often not a separate act, so we now need to integrate the automation feature of the associated conceptual judgments.[46]

As adults, most of our basic concepts are applied automatically in the appropriate situations, e.g., DOG, TREE, FLOWER, FISH, BED. Supposing that one is paying attention (I believe there is a volitional element here) and has previously learned the concept, the application of it can follow automatically. Learning someone's name follows the same pattern. One may forget it a few times, then use it more or less confidently, until it may become seemingly irrevocably associated with a given face.

The question is whether the learned automation of perceptual discrimination and conceptual identification is a problem. It would be if it involved shaping, distorting or adding to the data. But hypothesizing theory-ladenness is not necessary to explain the phenomena. There are enough data in the stimulus array available to all perceivers, and the reason why some subjects notice and identify certain features in the array can be explained by pointing out that they have previously attended to those features and automated their identification. Rather than counting against objectivity, this should be one of the goals of cognition: to learn and automate the easy stuff, so as to be in a position to master the more complicated.

Automation is also possible in case involving subtler discriminations of perceptible features, e.g., learning particular makes of car (Fiero, Mustang, Saab), particular species of bird (grackle, mockingbird, brown thrasher), musical pitches, the smells and tastes of different spices (parsley, sage, rosemary, and thyme). Even reading proficiently involves a significant amount of learned automation. Consider the following string.

SOPHISTICATED PHILOSOPHERS ARE ENAMORED OF PARADOXES.

For proficient readers the automation may go up at least to the level of identifying and recognizing the individual words in the string. It would be perverse to hold that struggling to recognize particular syllables and letter combinations and laboriously grasping the meaning of each word as a whole, as beginners to reading have to do, is a purer and less subjective way to read, and that automating the recognition of individual words involves shaping, distorting or adding to the stimulus.[47] Just as automation has a valuable role in acquiring physical skills such as "finger memory" in playing a musical instrument, handling a wind surfer, or doing a somersault, automation has a valuable role in cognition.

I believe that this sort of account has either been resisted or not developed by traditionalist empiricists and foundationalists for two reasons. One is the belief that if perception and conception are not radically modularized, then theory-ladenness is the only alternative. For example, the concept of automation may at first sound suspiciously close to some form of unconscious interpretation, something that has traditionally been a part of the Kantian camp. But if an account of automation along the lines sketched above in combination with some sort of ecological account of perception is feasible, then the choice is not between complete modularization and subjectively theory-laden perception.

The second reason runs deeper, and poses a more serious threat to foundationalism. What if, for example, some mistaken conceptual identification of perceptual phenomena becomes automatic? Then one is in the position of having automated an error. And since on the foundationalist account, such a conceptual identification of a perceptual phenomenon is supposed to be part of the basic level of justificatory support, one will have infected an entire hierarchy of propositions that rests upon the mistaken one. Such automated mistakes are possible. In order to avoid them the traditional empiricist foundationalist will feel the need to entirely modularize perception and conception. Only then, is it felt, can we avoid mistaken basic propositions and be sure that our justified hierarchies are erected only upon propositions carefully considered and known to be free from error. So any sort of automatic connection between focused perceptual states and conceptual identifications will be resisted.

Here I believe we need to break with the traditional foundationalist requirement of certainty or incorrigibility of basic propositions. This requirement, prominent in the foundationalisms of Descartes, Lewis, and Chisholm, arises from a felt need to find a haven from skepticism, and in turn leads, as we saw in Chapters 2 and 3, to the adoption of representationalist methodologies. Also in Chapter 3 we saw reasons for rejecting this approach to foundationalism. It is the insistence on incorrigibility that leads foundationalists to perceive the results of the Sparse Data and Sophisticated Identification cases as threatening. But if incorrigibility is not necessary, then the possibility of a mistaken basic proposition's infecting a hierarchy of justification is not as threatening. Naturally, the fewer mistakes the better; but for two reasons mistaken basic propositions do not spell doom for a hierarchical account of justification.

First, most hierarchies of justification do not depend upon a single basic proposition. So having to reject a basic proposition later found out to be mistaken will generally not be enough to collapse an entire hierarchy. But some fragile hierarchies of justification do rest on very few basic propositions. And in some of these cases it can be a good thing to have errors in the base pointed out. We should want, for example, individuals for whom second-hand reports of ghost-sightings form the basic level of justificatory support for their belief in the paranormal to learn to be more critical of such sources of evidence. This may lead them to reject many of their cherished beliefs. The same point should hold for any justification hierarchy. Foundationalism should not be seen as a haven for epistemological conservatism; our goal is truth, not complacency. (This issue is investigated further in Chapter 5.)

Second, any mistaken automated conceptual connection is not a threat to foundationalism since automated connections are not irreversible. Conceptual automation is not, in this respect, like perception: one cannot alter the appearance of the stick-in-water illusion, but one can always go back to revise or, if need be, eliminate a previously automated concept. This means that an automated error is not embedded forever in a justificatory hierarchy, inevitably corrupting efforts to achieve higher levels of justification. For example, if one learns a foreign language solely by reading, chances are that one will automate several mistaken pronunciations. If one then hears the language spoken, one can de-automate the incorrect pronunciation and automate the correct one. When a child grows older and rebels against being called "Susie" or "Johnny," parents can unlearn the years-ingrained habit of using the diminutive and automate the approved "Susan" or "John." We can learn that some sharks are warm-blooded, and thus revise our automated thinking of them as cold-blooded killers of the deep. So the automation feature is not problematic, since it is not immutable. And since foundationalism does not require incorrigibility, automation is not an enemy to it.

Far from it: automation is extremely valuable. It speeds up routine identifications and allows one to devote mental time to making more sophisticated identifications.[48] Of course, one has to be willing to go back and check one's basic premises again, should one have reason to believe something is awry.

It is important to note, finally, that in line with our rejection of skepticism in section 3.5, one would have to have particular evidence behind any suspicions that something is awry. Skeptical questions such as, "People make mistakes, so how do you know you didn't make one somewhere?" would not be legitimate in this context, since they commit ad ignorantiam. In the absence of particular ground for doubt, the skeptic's question will carry weight for one only to the extent that one knows that one is sloppy in one's mental habits. In such cases the question of the skeptic will carry some weight because one has first hand evidence of one's generally poor mental habits — and in such cases it will follow that one cannot and should not fully trust one's hierarchy of beliefs. On the other hand, to the extent one knows one is conscientious in one's mental habits, the skeptic's general question carries no weight.

The conclusion of this section is that the Sparse Data and Sophisticated Identifications cases do not lead to the conclusion that perception is theory-laden. More needs to be said on this issue, I believe, because part of the controversy underlying the debate over how to interpret such cases stems from differences on the nature of concepts. For advocates of the theory-laden interpretation, concepts are viewed not as structures abstracted from perceptual data, but rather as active constitutive agents of the perceptual data.[49] On this latter account, what even counts as an object depends upon one's background conceptual structure; so when confronted with cases where concepts are clearly part of the situation, as in the Sparse Data and Sophisticated Identifications cases, the conclusion that perception is theory-laden seems irresistible. The issue of the nature of concepts is a huge and technical one, and so as I mentioned in Chapter 1, I will not do it full justice in this essay. I can, however, make it clear that I advocate the concepts-as-abstracted-structures account and reject entirely the broadly Kantian account of concepts as shapers of sensory stimuli. But such differences over the nature of concepts are only local skirmishes in the overall battle involving differences over the fundamental relationship between consciousness and reality. In Chapter 5, however, the issue of concepts will arise again when we discuss this overall battle and, in that context, investigate the issue of the compatibility of hierarchy and coherence.

4.6 Can the given do its job, supposing there is one?

The major problem for a foundationalist epistemology, Rorty states in the context of discussing Locke's version, is its failure to explain how nonpropositional 'knowledge of' can provide an epistemic foundation for propositional 'knowledge that.'[50] So while in the previous two sections we have set aside the idea that "[w]ithout some starting point, some initial schema, we could never get hold of the flux of experience"[51] — that we would get only a two-dimensional, undifferentiated given — the foundationalist still has the task of explaining what it means to say that a nonpropositional form of awareness about which questions of justification do not normally apply (i.e., perception) can justify a state of propositional, abstract awareness. Even if perceptions are regularly associated with or give rise to conceptual identifications, it doesn't automatically follow that the conceptual identification is justified by the perception. How can we make sense of a justificatory move from nonpropositional to propositional, from concrete to abstract, from nonjustified to justified? Anti-foundationalists hope to show that the gulf between the two is in principle too large to be bridged.

The charge is that a preconceptual given, whether sensory or perceptual, cannot be a cognitive phenomenon; a prelinguistic, preconceptual, pre-propositional phenomenon is merely a causal reaction to stimuli. And since justification is a cognitive phenomenon, it follows that the given cannot play a justificatory role. The given may cause propositions, but it cannot justify them. Regimented, the argument is as follows.

P1. Only conceptual/propositional states are cognitive.

P2. The given is not conceptual/propositional.
C1. Therefore, the given is not cognitive.
P3. Justification is a cognitive phenomenon.
C2. Therefore, the given is not part of justification.

Combining P1 and P3 yields:

C3. Justification is a conceptual/propositional issue.

My theme in this section is that C2 and C3 are both false, and that the source of the problem is P1. But since every antifoundationalist in the world accepts P1, it is not likely to be obviously false. So before presenting my reasons for rejecting P1, let us first see the context in which antifoundationalists present it.

The great-grandfather of P1 was Kant. The famous Kantian dictum, "percepts without concepts are blind," has served as a rallying point for this position. Part of the idea here is that mere perception would give us an undistinguished, formless mass. Such a mass would necessarily be "ineffable," to use Williams's word.[52] In sections 4.4 and 4.5 we rejected the notion that perceptual experience is ineffable. Yet this is not enough to completely undermine Kant's dictum. The deeper point of it is that the nonconceptual is noncognitive. One could be perceptually aware of discrete units, but by not being conceptually aware one is not cognitively aware. Kant himself argued that "appearances might, indeed, constitute intuition without thought, but not knowledge; and consequently would be for us as good as nothing."[53] So to enter the cognitive realm one must be operating propositionally, conceptually, perhaps linguistically. Otherwise, the claim runs, there is no cognition, merely stimulus-response.

In his 1981 Carus lectures, Sellars argues that an "experience itself, presumably, is not a cognitive state. It is simply a state of the perceiver which is red in the basic sense of red."[54] The experiences Sellars is referring to here are sensory experiences, viewed as mere causal reactions. Now, for an experience to become a cognitive phenomenon, the awareness would have to be "an expanse of red as an expanse of red. It is to be construed, in other words, as, in a sense to be explored, a cognitive awareness."[55] And to see something as an expanse of red requires the application of the concept RED. Therefore, on Sellars's account the difference between the noncognitive and the cognitive is that the latter involves concepts while the former doesn't.

Agreeing with Sellars, Rorty makes the claim that the ability to respond to sensory stimuli "is a causal condition for knowledge but not a ground for knowledge."[56] By a "ground" for knowledge, Rorty means something that is capable of providing justificatory support. But for that one needs a cognitive state, i.e., a propositional one, which sensation is not. If one takes the given to be nonpropositional perception, then the same point holds.

Essentially the same argument appears in Williams and Bonjour.[57] The given, notes Williams, is supposed to be independent of conceptual interpretation. If so, then it must be non-propositional. But if it is non-propositional, it can't provide a check upon anything. Since experiences just are — experiences per se cannot be true or false — it seems therefore that the idea of non-propositional knowledge is a "confused" notion. Propositions, by contrast, can be true or false. And since, as Berkeley pointed out, "[n]othing can give to another that which it hath not itself,"[58] given experiences cannot serve as a basis for propositional knowledge.[59]
Accepting this premise, we are halfway toward creating a full-fledged dilemma for the foundationalist. It seems that the foundationalist can avoid the problem only by granting that the given is itself propositional in form — and hence cognitive, and so able to confer justification. The problem then is that if the given is propositional/cognitive/justified, then it will itself require justification. And if it requires justification, then it cannot be at the base of a foundationalist justificatory structure.

In Bonjour's words, this "most fundamental and far-reaching objection" creates the following dilemma for the foundationalist.

if his intuitions or direct awarenesses or immediate apprehensions are construed as cognitive, at least quasi-judgmental (as seems clearly the more natural interpretation), then they will be both capable of providing justification for other cognitive states and in need of it themselves; but if they are construed as noncognitive, nonjudgmental, then while they will not themselves need justification, they will also be incapable of giving it. In either case, such states will be incapable of serving as an adequate foundation for knowledge. This, at bottom, is why empirical givenness is a myth.[60]

To generate one horn of the dilemma, we use the premise equating the cognitive with the "judgmental" (i.e., the propositional). To generate the other, we use the premise that any cognitive state able to confer justification itself requires justification. So we have two sub-arguments making up the dilemma. The first horn is:

If w is cognitive, then w can provide justification.

But if w is cognitive, then w needs justification.

If w needs justification, then w can't be the foundation of justification.

The second horn is:

If w is not cognitive/propositional, then w can not provide justification.

If w cannot provide justification, then w cannot be the foundation of justification.

Neither horn leads to the conclusion Bonjour needs. The second horn assumes the equation of the cognitive and the propositional. This I think is unwarranted. Cognition is a broad term: it means an awareness of reality in some form. The thrust of Chapter 3 and the previous sections of this Chapter is that perception is a form of awareness of reality, that it provides subjects with a certain amount of information about reality. The point holds generally for any sensory/perceptual mechanism: the sensory/perceptual mechanisms of bats and bloodhounds enable individuals of those species to be aware of reality in some form or other — and awareness of reality in some form is, I suggest, the root meaning of cognition. We should not overintellectualize cognition by assuming that the only form it can take is conceptual. It is true that concepts allow humans a greater cognitive range and degree of sophistication than is possible for other known species; and there are justificatory issues that arise for conceptual processes that do not apply to nonconceptual forms of awareness. It does not follow that nonconceptual forms of awareness are noncognitive and have nothing to do with justification.

Perhaps a fear of psychologism motivates suggesting that perception is noncognitive. Justification is a normative concept, while perceptual mechanisms seem to operate more or less automatically. Here we need to be wary of equating cognition and justification. Holding that perceptual states are cognitive states does not mean holding that they are justified states. A perceptual state is an awareness of reality. Justificatory issues arise consequently to perceptual states: justification relates to what conceptual identifications one makes in the context of one's perceptual states. Here errors can happen: perceptual states do not automatically force any particular conceptual identifications, except in those cases where conceptual identifications have been previously learned and automated (but these were learned and automated, and those processes are subject to justificatory concerns).

This issue leads us directly into the other horn of the dilemma — the one based on the claim that if a state can provide justification, then it needs justification. Here the problem is either that "justification" is being used equivocally or that questions are being begged. If justification means trotting out propositions from which a desired proposition follows, then perceptual states do not need justification. But if "justification" means a cognitive relation to reality, then perceptual states have all the justification they need. Perception is a state of direct awareness of reality. As reality is what all of cognition is directed toward, perceptual states are the terminal points of justification. That is where one begins, cognitively, and, working backwards, that is where tracing justification ends. But since justification is a normative issue, perhaps to avoid confusion we should specify a broader concept, such as "validation," to identify any state of being in cognitive relation to reality, and reserve "justification" for the subset of validating conceptual states. As such, questions of validation would apply to perception (requiring a philosophical defense of direct realism), but questions of justification would apply only to conceptual states (requiring accounts of concept-formation and logic). This would satisfy the foundationalist, at the same time preserving the normative content of the concept of justification. Even here a foundationalist would have to insist that the concept of justification be split into two species: those of perceptual-state-to-conceptual-state justification (involving standards for concept-formation) and conceptual-state-to-conceptual-state justification (involving standards of logic). Recognizing only the latter in formulating a dilemma for the foundationalist is begging the question. It is only by equating cognition with conception and justification with logic that the first horn of the dilemma has any force.

Rorty, however, offers an argument that is independent of any possible conceptual confusions the above dilemma involves. The passage containing the argument is worth quoting at length.

There is no reason for Sellars to object to the notion of 'knowing what pain (or redness) is like,' for this would only support the Myth of the Given, and contradict psychological nominalism, if there were some connection between knowing what pain feels like and knowing what sort of thing pain is. But the only connection is that the former is an insufficient and unnecessary causal condition for the latter [my emphasis]. It is insufficient for the obvious reason that we can know what redness is like without knowing that it is different from blue, that it is a color, and so on. It is unnecessary because we can know all that, and a great deal more, about redness while having been blind from birth, and thus not knowing what redness is like. It is just false that we cannot talk and know about what we do not have raw feels of, and equally false that if we cannot talk about them we may nevertheless have justified true beliefs about them.[61]

The central claim is that the given cannot generate justified propositions because it is neither a sufficient nor a necessary condition for justified propositions. If this is so, then the given cannot be cognitive and must be merely causal.

The insufficiency point is not objectionable to foundationalism. Certainly, being in a perceptual state does not guarantee that a conceptual or propositional state will follow. The subject also must, if the relevant concepts are not at hand, perform a process of abstraction. Even if the relevant concepts have been learned, the subject may still need to put forth effort to remain focused on the phenomenon, to try to recall the right word for the phenomenon given in the perceptual state, to connect this phenomenon with memories of related phenomena, and so on. But foundationalists need not make the sufficiency claim for the given, so this is not a threat.

The claim that the given is unnecessary for justification is the tricky one, for foundationalists most definitely claim that basic propositions must necessarily be grounded in perceptual states. Rorty claims that since any particular conceptual item can be arrived at by variety of routes, the given plays no justificatory role. Let us work through an example to see why this does not follow. Suppose a person not born deaf, say Ms. Credo, hears two different notes played on a piano. The difference between the two notes will be audible to her. She is told that one note is called "G" and the other "A-flat." She then goes on to learn about the physics and physiology of sound, including why the two sound different. Suppose Mr. Doxas is born deaf. Even so, he can come to know that there is a phenomenon called "sound," that it comes in different frequencies and wavelengths, that the speed of sound varies with elevation, that an oboe and a clarinet playing the same tone will sound different because they produce different overtones, that G is different than A-flat, and so on. So it is true that to learn about tonal differences, particular auditory experiences are not necessary. Ms. Credo has a simpler and much more direct route to the knowledge that this sound is G and that one is A-flat, though she too could have learned the difference by a route similar to that followed by Mr. Doxas. The fact that she can hear means that "That is A-flat" can be a basic proposition for her, while for Mr. Doxas the proposition "That is A-flat" necessarily must be a conclusion inferred from a vast number of other propositions. But it does not follow that no experiences at all are necessary, and that those other propositions need not be grounded in some perceptual states or others. If the gauntlet is thrown down, as it must be, the foundationalist will trace Mr. Doxas's knowledge of the difference between G and A-flat to a number of experiences — perhaps to his feeling the vibrations of a stereo speaker, and to the complex set of visual experiences that made it possible for him to infer that other people are able to communicate in a form not available to him. The propositions that capture these perceptual experiences will be Mr. Doxas's basic propositions. Energy comes in different forms, and each form of energy has effects on objects and other forms of energy. This means that phenomena that can be detected directly by one perceptual mechanism can be detected indirectly via another perceptual mechanism. It follows that what is a foundational proposition for one person need not be foundational for another. So while any given perceptual experience is not necessary for any particular proposition, it does not follow from this that no perceptual experiences are necessary to ground that proposition. Hence, Rorty's argument is invalid.

* * *

In this Chapter we have investigated all (except one) of the major attacks on the given. We have seen and set aside as inadequate the claims that the given does not exist because it is inferentially constructed or theory-laden. And we have investigated and rejected the arguments intended to show that a preconceptual given could play no justificatory role anyway. The exception is the view that while the given can play a justificatory role, its role is so far from being the whole story that it is not enough to maintain the foundationalist claim that justification is hierarchical. We will address this final attack on the given in Chapter 5, in the context of discussing the hierarchical and contextual dimensions of justification.

* * *

[2] Lewis (1929, p. 38), Moser (1989, p. 186).

[3] Whether the "blooming, buzzing confusion" comes as discrete point sensations or as an undifferentiated flux.

[4] It could also be responded that the integrative methods are innate. This response will be discussed below, in Section 4.4.

[9]Manners & Kaplan eds. [1968, p. 421 and p. 169]).

[10]lars that are determinate in all their dimensions.

[11] Moser offers a similar reconstruction (1990, p. 188).

[15] Quine & Ullian (1978, p. 22); Harman (1973, p. 19).

[18] Gregory (1970, p. 56) and Kelley (1986, p. 61) note that this holds generally for perceptual illusions.

[21]constructions to adopt (1970, pp. 25, 27).

[22] Gibson (1966 and 1979). Gibson (1982) is a posthumously published collection of his essays and articles.

[28] Gibson goes so far as the claim that the information available in ambient array is inexhaustible (1979, p. 57).

[32]logical process itself is one of computational modeling.

[34]resulting perceptual state (1970, pp. 24-25).

[36] Hanson's presentation of these is on pp. 4-8 of his (1958).

[39] Cases where this assumption is not necessarily operative are discussed below in the context of the Sophisticated

Identifications cases.

[40] Gregory (1970, pp. 56 and 25; his emphasis).

[41] Hanson's discussion appears on pp. 4-8 of (1958).

[44] Hanson (1958, pp. 10, 19, 18, respectively).

[51] Art historian E.H. Gombrich, quoted in Scheffler (1967, p. 24).

[59] Williams (1977, p. 29; also pp. 31-42, esp. p. 37).

[60] Bonjour (1985, pp. 29 and 69, respectively).