Chapter Five

HIERARCHY AND CONTEXT

* * *

5.1 From percept to proposition

5.2 The case for hierarchy

5.3 The case for context

5.4 Developmental foundationalism

* * *

5.1 From percept to proposition

Aquinas asked "Whether the Knowledge of God is the Cause of Things?" and gave us two options. On the one hand, "[t]he knowable thing is prior to knowledge, as the Philosopher says." But since "what is posterior and measured cannot be a cause," it follows that "the knowledge of God is not the cause of things." On the contrary hand, "Augustine says, Not because they are, does God know all creatures spiritual and temporal, but because he knows them, therefore they are."[1]

As a good Christian, Aquinas sided with Augustine on this one, against the pagan Aristotle. But he put before us two fundamental options. One can hold that what is, is what it is independently of any being's knowledge of it, and that what is sets the terms for what counts as knowledge. Or one can hold, in either secular or divine form, that esse est percipi, and that therefore the knowing subject ultimately sets the terms for what counts as knowledge.

In this context Aquinas advocated only the ontological side — the view that mind has ontological priority over matter. Setting the stage for the modern period in philosophy, Descartes introduced the epistemological version in his second Meditation — the view that the subject is known prior to and better than external reality. In his famous discussion of the wax he asks,

What am I to say in regard to this I which seems to apprehend this piece of wax so distinctly? Do I not know myself much more truly and much more certainly, and also much more distinctly and evidently, than I do the wax? For if I judge that the wax is or exists because I see it, evidently it follows, with yet greater evidence that I myself am or exist, inasmuch as I am thus seeing it.[2]

Epistemological subjectivism is the view that the subject has fundamental priority in setting the terms for what is knowledge. I think epistemological subjectivism is at the root of the positions that I have been fighting throughout this essay, and my sympathies, to paint broad historical strokes, are squarely with Aristotle. On specific issues I have rejected arguments intended to necessitate the adoption of subjectivist conclusions — the conclusions that dictate that one has to start "inside" and either set out on a quest for a mysterious correspondence-to-fact relation or reject such a quest as hopeless and remain stuck "inside." The skeptic's Evil Demon argument, the arguments from illusion and perceptual relativity, the arguments for perceptual inferentialism and the theory-ladenness of perception would, if any one of them were sound, force at least a minimal version of subjectivist epistemology. In rejecting these arguments in turn, I have hopefully left myself in a position to advocate the fundamental objectivity of justification and knowledge.

The results of Chapters 3 and 4, if sound, make possible the following claims about perception.

1. Perceptual states are direct relations between subject and object, not states that put subjects in an indirect relation to objects.

2. Perceptual states are independent of conceptual states, though not isolated from them.

3. Perceptual constancies are the result of physiological, not inferential, integration.

4. Perceptual focus or attention is an autonomous capacity, though it can also be conceptually guided.

5. Perceptual states, though not conceptual or propositional, are cognitive states.

Our perceptual faculties, on this view, give us a great deal of information — good information — about reality.[3] Without representationalist or skeptical worries we can proceed, as people do, as direct realists.[4] The procedure then is, in outline, straightforward. In the course of learning a language, individuals conceptualize some of their perceptual states; in the course of learning a grammar, they integrate the resulting concepts into propositions; and, in the course of learning logic, they integrate the propositions into networks. Each of these issues involves many complexities, and to the first two — conceptualization and grammar — I will not do justice here.

As mentioned above, I accept an abstraction account of concepts, one that is neither realist in the Platonic or Thomistic sense nor nominalist nor Kantian. Concepts are not what make possible perceptual discriminations, contrary to Kantian theories; they are not merely linguistic labels attached to sets of perceptual experiences, contrary to nominalist theories; and they are not intuited abstract essences, contrary to traditional conceptual realist theories.[5] Concepts are reformulations of the data given in perceptual discriminations about entities, their properties, actions, and relationships. They require transforming a wealth of perceptual information into more economical units. They are, to speak loosely, "mental space-savers" that result from an ability to "chunk" information from many percepts into a single cognitive unit.

The above paragraph is simply a statement of conclusions, not an argument for them. So my account needs supplementing by an account of concept-formation and a general defense of the conceptual level of functioning. These are beyond my scope here, so I will simply borrow a line and claim that the soul is constituted such that it is capable of performing these operations,[6] and skip over the details to issues of propositions and relationships between them, and between them and perceptual experiences.

A proposition is a grammatically structured string of concepts. I think our concepts can be grouped into four main categories: entities (dog, flower, blanket), actions/processes (run, cry, eat), properties/states of entities (pink, soft, warm), and relationships (beside, larger than, behind). A proposition asserts a relationship between two or more concepts. Each concept is a condensation of information about some entity, action/process, property, or relationship, so justifying a proposition on the basis of perceptual experience means grasping that the relationship between concepts asserted in the proposition is given in some perceptual experience(s). At this basic level, the temporal order of the process is: one perceives a certain relationship, forms or calls upon the appropriate concepts, and grammatically integrates those concepts into a single unit — a proposition. The perceptual experience thus justifies the proposition. Justification, on realist grounds, means having a connection to fact. Perception has such a connection, since it is a direct experiential relation to fact, so it has what the proposition needs.

So far we have discussed only what I call type A propositions. These are propositions that identify what is given in single perceptual experiences — for example, "The cat jumps," "My tummy hurts," "The bird is flying," "The dog is furry," "Daddy is beside Mommy." Type A propositions are the foundationalist's "basic" propositions, and the claim is that all other propositions are derived from type A propositions. Two other, relatively straightforward, types of propositions should be mentioned in this context, in order to get us up to a level of propositional complexity appropriate for introducing criticisms.

Type B propositions are generalizations and summaries from type A propositions — for example, "Dogs are furry," "Birds fly," and "Sometimes cats jump." The processes of generalization and summary require the integration of several type A propositions; part of inductive logic deals with better and worse methods of performing such integrations. Then, once one has formed some type B propositions, one can perform categorical syllogisms, which are structures of either type B propositions alone or a combination of type A and B propositions. This, of course, is part of deductive logic.

Type C propositions are compound propositions, formed by relating propositions of type A and/or type B by means of the connectives — for example, "If the cat jumps, then the bird flies" and "Either my tummy hurts or I'm imagining things." Then various combinations of propositions of types A, B, and C can yield further propositions; the combinations involved here are those standard forms of deductive argument studied in propositional calculus.

We have reached a certain level of complexity here, although much more needs to be said for a full theory of propositions: we have not, for instance, mentioned tense, metaphor, or counterfactuals. Indeed, as Rorty and others have stated forcefully, it is not at the level of propositions such as "The cat is on the mat" that accounts such as the above run into trouble, but rather at the level of propositions such as "Love makes the world go around" and "History is the story of class struggle."[7] When we get to highly theoretical propositions and propositions embedded in complex networks of other propositions, the question is whether a hierarchy and correspondence account of justification along the lines sketched above can be maintained, or whether the complex networking ultimately destroys the hierarchical structure of justification the foundationalist needs to maintain.

5.2 The case for hierarchy

The need for justification is based on the fact that at the conceptual level errors are possible. Perceptual states are direct relations of subject and object, but in conceptually processing their perceptual states human cognizers do not automatically get things right. Truth, on realist accounts, means correspondence of belief to fact. But if a subject's cognitive processes can reach either truth or falsity, then special efforts must be made to maintain correspondence. As adults we identify those cognitive processes that maintain the correspondence. Those that maintain the tie to reality are called "truth-preserving processes," and they are distinguished from those cognitive processes that do not. The concept of justification pertains to the extent to which the subject follows truth-preserving processes. By identifying the cognitive processes used to reach a given belief and comparing them to our list of truth-preserving processes, we have an explicit measure of the degree of justification of the belief.

When we investigate propositional truth-preserving processes, they seem to have two dimensions. One dimension pertains to the fact that the justified proposition depends on the justifying proposition(s), but not vice versa. If I believe "q" because I believe "If p then q" and "p", then there is an asymmetrical dependence involved. The same point holds if I believe "All w are z" because I saw and identified one hundred w's in various situations and each was z. Let's call this the hierarchy dimension: propositions are organized in asymmetrical hierarchies of dependence.

The other dimension pertains to the fact that justified propositions are always integrated into the wider context of one's entire system of propositions. Any new proposition that contradicts the rest of the system is not justified. The new proposition or the rest of the system may have to be rejected, but the new proposition is not justified until the conflict is resolved. Let's call this the contextual dimension: propositions are not held in isolation from each other.

The hierarchical dimension is the focus of this section, though we will, in the next section, return to the more difficult contextual dimension.

Broadly speaking, cognition is a process of hierarchical integration from start to finish. Sensations are integrated into percepts, percepts are integrated into concepts, concepts are integrated into propositions, propositions are integrated into arguments, and arguments are integrated into theories. And looking backwards, from the perspective of the order of dependence, theories depend upon arguments, arguments depend upon propositions, propositions depend upon concepts, concepts depend upon percepts, and percepts depend upon sensations. The dependence is asymmetrical, which is what the foundationalist's hierarchy condition claims. In Chapters 3 and 4 we investigated sensation and perception; so it is now the hierarchical features of concepts and propositions that concern us.

The case for the hierarchical structure of justification starts with the claim that there is a necessary hierarchy of concepts. There is generally a difference in the order of acquisition of concepts and the order of scientific organization of concepts, but the same hierarchical principles apply. If it is true that there is a necessary conceptual hierarchy, then since what propositions one can form depends upon what concepts one has available, propositional structure will depend upon conceptual structure.

To see the necessity of conceptual hierarchies, let's imagine a few cases in which we violate it in the order of acquisition.

Suppose I walk into a scientist's lab, hoping to learn some new things. I point to a large metallic contraption sitting on a table and ask what it is. "It's a scientific instrument," responds my scientist. I ask about a tiny, fragile-looking glass object; "That's a scientific instrument," says the scientist. I point to a wood and plastic object leaning against a wall; "Scientific instrument," intones the scientist. And so on. Clearly, given information at this level of abstraction in the hierarchy, I'm not learning anything, and will not.

The problem can work the other way. Suppose I walk into the neighboring laboratory and see various containers of liquids sitting around. I ask the scientist, "What's this?", pointing to one container. "Diexel Thombon," says the scientist. I point to another; "That's Biddle Mun Adonee," she replies. I make one last attempt, point to a different container, and hear in response the word, "Granveerenze." Again, I have really learned nothing.

Or suppose that a child points to a chair, indicating that she wants to know what it is; the parent replies "Artifact." The child points to a plastic play hammer, and the parent says, "Artifact." The child points to an automobile, and the parent replies, "Artifact." If this procedure continues, then cognitively the child is not going to progress far.

As a point of structural necessity, concepts such as "scientific instrument" and "artifact" must come later. Why? Because one must start with what's available to the person in terms of the phenomena the person's prior learning and/or perceptual faculties can discriminate. The operative question is, What differences and similarities of the entities in the person's cognitive field can he or she attend to?

If, for example, it is the case that the child's cognitive context is such that she can more readily notice the differences between the play hammer and the chair and the automobile than she can notice their similarities, then the concept "artifact" will be both unintelligible and useless to her. The starting level of concepts must be keyed to the level of phenomena her perceptual faculties are able to discriminate, just as the scientist's introduction of technical terminology to me must be keyed to the level my previous learning has brought me to.

Yet within a given level there are options, due to the differences in individuals' environments. There exists a degree of variability from individual to individual as to what is the content of his or her perceptual world; hence there is some variability from individual to individual as to what concepts will be formed first and what perceptual judgments will basic. The difference between being a city kid and being a country kid, for example, may mean that one arrives at the concept "animal" after first acquiring "dog," squirrel," and "cat" rather than after "cow," "horse," and "groundhog."

At the level of more abstract concepts, the hierarchical order is maintained, although again within the level there will be options and variability. Take for example the moderately sophisticated concept "mammal." Forming the concept "mammal" depends on a prior grasp of certain other concepts. At a minimum, someone needs concepts for at least one of the distinguishing characteristics of mammals, i.e., "hair" and "milk-producing." In virtually all cases it requires some from the following set of species that are mammals (e.g., humans, dogs, squirrels, lions, raccoons), and one or more from the set of contrast (non-mammal) species (e.g., fish, snakes, spiders).[8]

Conceptual knowledge is expressed and retained propositionally. At the propositional level, what propositions are justifiable for the subject depends upon what concepts he or she has formed. The justification of a proposition presupposes at least a minimal understanding of the concepts it integrates. So the conceptual hierarchy will to some extent be carried over to propositions. Yet at the level of proposition-to-proposition justification the primary hierarchical factor is the logic of justification. Whatever basic propositions one forms, once one is at the propositional level there are necessary relations of asymmetrical dependence: Some propositions are premises and some are conclusions drawn from them. These conclusions regularly become, in turn, premises from which further conclusions are drawn, and so on up the hierarchy.

For example, consider the proposition "It is cheaper to ship goods from Philadelphia to New York than it is to ship goods from Boston to New York." The justification of this proposition will depend, at least in part, on the proposition "Philadelphia is closer to New York than is Boston." The justification of the proposition "Philadelphia is closer to New York than is Boston" depends necessarily on the propositions: "Philadelphia is distance X from New York," "Boston is distance Y from New York," and "X is less than Y." The justification of "Boston is distance Y from New York" will require a whole raft of previous propositions — "This is a ruler," "This is a map," "New York is shown on this map," "Boston is shown on this map," and so on. Each of these propositions depends on certain conceptual distinctions, each of which ultimately depends on perceptual phenomena. In this example, "Boston is distance Y from New York" is at least two stages away from perceptual experience, while "It is cheaper to ship goods from Philadelphia to New York than it is to ship goods from Boston to New York" is at least four stages away.

The above is a highly simplified example, but it illustrates the point that there is a hierarchy. Not all propositions are at the same level in the hierarchy, as measured by the distance from perceptual phenomena.

5.3 The case for context

While justification exhibits a hierachical dimension from the outset, it also exhibits a contextual dimension. Our perceptual faculties are limited; in any given time frame one can perceive only so much. When forming a concept of a perceptual object, such as "dog" or "ball," one has a limited number of perceptual experiences, both current and remembered, to draw upon. This sets a context: a background framework that delimits the concept's breadth and depth. As one's perceptual experiences change — one can get new angles on the phenomenon, or a broader acquaintance with phenomena of that and related sorts — one's concept is revised. The revisions can involve the concept's becoming broader or more detailed, for example. When forming increasingly abstract concepts, the point holds again: one has a limited number of previously-formed concepts to draw upon. These previously-formed concepts set the context for the new one. But the new one may become revised in the light of other concepts, either new or revised ones, as we shall see below.

At the level of propositions, propositional meaning is conditioned by context. If one walked in at the end of a political speech and heard the speaker proclaim "And therefore I proudly proclaim my commitment to the principle of the right to life!" — without any background context, one wouldn't know what the speaker means, let alone whether one agrees with the speaker. Whether the speaker favors the view that each individual's life is his or hers to make the most of, or the view that every individual is automatically entitled to sustenance from all the other members of the society, or the view that abortion is wrong under any circumstances — is contextually conditioned by what was said earlier in the speech. That proposition has the meaning it does due to its place in the structure of propositions that makes up the speech.

Propositional justification is also conditioned by context. For any proposition, its justification depends upon two contextual elements: the conceptual hierarchy it is embedded in, and the propositional hierarchy it is embedded in.

To illustrate the relationship of conceptual context to propositional justification, let us take as an example the proposition "Airplanes fly." This is a justified proposition for most of us: we each have ample perceptual evidence of having seen airplanes flying overhead or having ridden in them. But this is not the whole story. One is operating on a certain definition of "flying" — say, "self-propelled motion through air." Perhaps one's concept of flying was formed in contrast to gliding and to being thrown. It may also have been formed in contrast to swimming, which is self-propelled motion through water, and to walking, which is self-propelled motion on the ground. One's concept of flying is also formed and held in context of one's knowledge of birds: one knows that birds also fly. All of this is perfectly reasonable, and all of it forms the background context to one's justified belief that airplanes fly.

Now suppose that some scientists investigate the actions of penguins when they propel themselves through water and come to the conclusion that what the penguins are doing is flying and not swimming. The dynamics of the penguins' action, they discover, are exactly the same as those of birds flying through the air, and the actions of birds moving through the air and penguins moving through the water are both very different from the characteristic actions of fish when swimming. If one hears of this new information, one then has to reevaluate the essentials of flying — which means one has to reevaluate one's concept "flying." And since the proposition "Airplanes fly" depends upon a specific characterization of flying, a reevaluation of one of the proposition's constituent concepts certainly bears upon the justification of the overall proposition. One may decide that the medium through which one moves is inessential to flying, so one will drop the phrase "through air" from one's list of the essentials of flying. One may also decide that the distinctive type of self-propelled action that penguins and other birds both perform is fundamentally different from the type of self-propelled action that airplanes do, and that, accordingly, the concept "flying" is not technically appropriate for describing what airplanes do. One might then say that airplanes thrust themselves through the air, and that to say that airplanes fly is a metaphorical extension of the concept.

The point is that new information can force a restructuring of one's conceptual hierarchy (in this case, in the direction of greater precision), which forces a reevaluation of the justification of the propositions built from those concepts. Since no concept is an island in the hierarchy, adjusting any given concept has effects on other concepts in the hierarchy, which also has effects on the propositions that depend on those concepts. This in turn means that propositions cannot be justified in a way that is immune from any future conceptual revisions. Justification is a conceptually contextual issue.

Justification is also a propositionally contextual issue. Not even type A propositions (i.e., those that conceptually identify perceptual level phenomena) are justified completely and forever out of any context. One may see a woman's dress and judge that at it is magenta. But then one may be told that the lighting is more blue than white and so become — and properly so — less inclined to believe that her dress is magenta. Other propositions can be relevant to the justification of even so basic a perceptual judgment as that of color.

Any proposition other than a type A proposition depends upon inductive support from a number of other propositions.[9] As new information comes to light, one's cognitive context changes, and the degree of justification for relevant propositions may be changed. Consider a typical example.[10]

Suppose one gets into one's car with a friend on a winter's evening, inserts the key into the ignition and turns it — but the car fails to start. Not only does the car fail to start, the engine doesn't even turn over. Now, from one's previous experience, one inductively infers that the battery is probably dead. Depending upon the range of one's experience, believing the proposition, "The battery is dead," will be justified for one to a given degree. That degree of justification can change. Suppose one's friend points out that a new battery was installed just last week. Taking this new information into account lessens the degree of justification of "The battery is dead." But then one remembers that the battery was installed by a drunken mechanic; the degree of justification of "The battery is dead" increases. But then one's friend says that he checked the installation of the battery afterwards and noted that everything was fine; the degree of justification again goes down. And so on. To bring the story to an end, suppose one finally notices that the car's light switch was left in the "On" position; then "The battery is dead" becomes highly justified for one.

The point is that propositional justification is contextual. Contexts change: new information can come to light that necessitates either a revision of one's conceptual hierarchy or of one's assessment of the degree of inductive support. This means justification requires that one remain open to new evidence, pro or con, and that one be willing to revise or reject previously justified propositions.

The question then is: Are such facts inimical to foundationalism?

5.4 Developmental foundationalism[11]

Initially there seems to be no problem reconciling the two dimensions of justification. Acquiring knowledge is a process of discrimination and integration: one first discriminates an item and then integrates it with the rest of one's beliefs. For example, one distinguishes something perceptually; then one identifies it conceptually, thus integrating that percept with the rest of one's knowledge. There is both a hierarchical element (one's perceptual awareness of reality's validating the proposition) and a contextual element (one's previous conceptual knowledge's placing the new item within the context of one's entire body of beliefs). Or it may be that contextual elements enhance the justification of two bodies of propositions further up the hierarchy. Suppose one has two different hypotheses, supported to a certain extent by perceptual evidence. If one finds that they complement each other and integrate a wider body of evidence than either can do on its own, this lends some support to each. This is like finding that two separate parts of a jigsaw puzzle, which one has worked on independently, fit together perfectly. How could such "coherence" phenomena be a problem? One strives for an integrated system of propositions, each of which is ultimately derived from perceptual evidence.

The problem seems to be the worry that the contextual dimension collapses the hierarchy dimension. The problem comes in two forms, one from the perspective of traditional foundationalism and one from the perspective of the coherence theorist. From the standpoint of traditional foundationalism, the contextual dimension is problematic, for it opens up the possibility of revisions. Yet the whole point of traditional foundationalism's insistence upon hierarchy is to trace justificatory connections back to incorrigibly known propositions, and then to build inexorably from there. But if everything is potentially open to revision, then one has no absolute, contextless certainty for any of one's propositions and thus no haven from skepticism.[12] If, however, we reject traditional foundationalism's concern with skepticism (as in Chapter 3) and its attendant quest for contextless certainty, then the possibility of revisions need not be a problem for foundationalism on that ground. As long as justification still exhibits a hierarchical structure, and as long as that structure is grounded in perceptual data, revision and reorganization among one's concepts and propositions are compatible with foundationalism. It is exactly here that the coherence theorist sees problems.

As the coherence theorist points out, the contextual dimension of justification means that one's system of beliefs is (or should be) an integrated whole. As such, revisions in one part of the system cannot help but have "ripple" effects upon the rest of the system; the system faces revisions as a whole. Yet as adults, our belief systems are incredibly huge. The ensuing problem, from the perspective of the coherence theorist, is that the system is too huge for anyone to follow all of the "ripples" back to perceptual roots in order to determine whether the system is properly grounded. If each proposition were an acontextual island, then checking correspondence to fact would be straightforward. But since each proposition is not an island, it seems that the contextual dimension means the foundational hierarchy dimension is lost. One can only trace the "ripples" so far, and this means, as Neurath suggests, that the entire system has to float: each cognizer is like a sailor "who, unable to return to dock, must reconstruct his vessel on the open sea, and is therefore forced to make use of the best constituents that are at hand."[13] One is unable to "return to dock" because it seems that to accomplish that one would either have to be able to trace an indefinite number of justificatory connections at a moment (follow all of the "ripples") or be able to take a perspective from somewhere outside the whole system in order to compare it to reality. Neither of these can one do.

Everything is right about this argument except the conclusion. It is certainly true that working through the implications of a revision can be a lot of work, especially as one's belief system becomes both more wide-ranging and more specialized. But there's no getting around this: an inconsistent set of propositions is not justified. And one needs to keep open the possibility of revisions precisely because one's system of beliefs is so huge. From a limited context it is extremely difficult to form a hypothesis that takes into account all factors. The common case is that new evidence arrives that brings to light new factors that either necessitate qualifications or rejections of old hypotheses. Or in attempting to integrate two well-grounded hypotheses in related fields, one may discover inconsistencies that necessitate revisions to one or both of the hypotheses.

And justification requires a tie to reality. Justification is a necessary condition for knowledge, and knowledge is about reality. Direct access to reality is given only in perceptual experience. So justification requires a connection to perceptual experience. In other words: the goal is truth, truth depends upon correspondence to reality, and justification is the process of establishing correspondence. So there's no getting around the need to trace the ripples and there's no getting around the need to connect the propositions to perceptual evidence.

The question, then, is how much justificatory work revisions involve, and the answer seems to be a straightforward, "It depends." This is a huge topic, but the amount of work involved seems to depend upon (1) how radical the initial revision is and (2) where in the system of one's propositions it occurs.

On (1), changes that force revisions come in degrees. Depending upon the degree, the ensuing system-wide revisions will be more or less radical. For example the change from a Ptolemaic to a Copernican astronomy is more radical than the change from a Copernican to a Keplerian astronomy. Similarly, the change from an immutability view of biological species to a Darwinian evolutionary account is more radical than a change from a Lamarckian to a Darwinian version of evolution. Accordingly, in the former cases the revisions will necessarily be systematically more wide-ranging, while in that latter cases the revisions will be more limited in scope. In the former cases there will be radical abandonments of one set of concepts for another (as, for another example, in the case of abandoning phlogiston for oxygen), while in the latter cases there may be only slight revisions of certain concepts or the accommodation of one framework into a larger one (as, for example, may be the case in the transition from Newtonian to Einsteinian physics). There is nothing necessarily easy about this: the needed revisions may not be obvious, and the transitions may take years or generations to work through.

On (2), the location of the initial revisions within a system of propositions also determines the number of ripples (i.e., justificatory connections) that need to be traced, established, or re-established. To give a straightforward example, a revision of one's views on the trade deficit can force significant changes in one's economic opinions, but it will not have much effect on one's views in art history or evolutionary theory. Some legitimate degree of compartmentalization exists within systems of propositions due to their positions in the hierarchy, and this compartmentalization limits the system-wide impact of revisions. This means, for example, that our imagined revisions to one's concepts by the "discovery" that penguins fly rather than swim will not necessarily be some hugely indefinite project. For such a moderately sophisticated change, only a moderate number of concepts and propositions are affected, and for each its initially justifying perceptual data are easily within reach. At the other end of the spectrum, very little is left untouched by Einsteinian physics due to its attempt to be an account of the principles governing virtually all phenomena; as such, it has connections to virtually every other concept and proposition. Any revision at this level entails an enormous amount of work, but it is work that must be done.

For all such cases, volumes need to be written. From historians of science we need detailed case studies. From philosophers we need full theories of concept-formation and logical methodology. And these need to be fleshed out by philosophers of science on the technical issues of theoretical reduction and integration, principles of taxonomic organization, and so on.

These projects will not work if we start by assuming the contextual and hierarchical dimensions of justification are inimical to each other. What is needed instead is a developmental account that integrates the two. This will, as a necessary step, require a rejection of the traditional picture of foundationalism shared in common by antifoundationalists and most foundationalists. Traditional pictures of foundationalist hierarchies most often speak in terms of deriving all justified propositions from a limited number of basic propositions. The resulting picture is of an inverted pyramid, to which monolithic blocks are added in an upwardly-expanding direction. First we need to flip the pyramid: at the base is an ever-expanding number of propositions. These are the basic propositions, based directly upon perceptual experiences. Then we have to realize that the pyramid is not really a good structural metaphor after all, for as one ascends the hierarchy the structure expands both upwards and outwards: depending upon what areas of investigation are pursued, some parts of the hierarchy will be more developed than others. There is nothing static about this, as discussed above. New perceptual evidence will arrive necessitating and justifying new concepts and propositions. Connections will be made between previously separate parts of the structure. The structure will be used as a base from which new levels will be erected. And over time the hierarchical structure will undergo revisions at each level. Such is the history of science.

* * *

[4] "Realism," as Maritain put it, "is lived by the intellect before being recognized by it" (1959, p. 79).

[5]p. 335). Price thinks this temptation should be resisted.

[6] Aristotle, Posterior Analytics (II, 19, 100a 13-14).

[7] Rorty (1980, in Moser & vander Nat [1987], p. 213).

[8]from observations of life forms native to its habitat.

[9] I am setting aside the issue of whether so-called analyt ic or a priori propositions exist, and whether they are exceptions to this rule.

[10] I am indebted to J. Michael Dunn for this example.

[11] The adjective is borrowed from a suggestion in Kelley (1986, p. 210).