Harold I. Brown
Department of Philosophy
Northern Illinois University
DeKalb, IL 60115
U. S. A.

Epistemic Concepts: A Naturalistic Approach*

Several forms of naturalism are currently extant. Proponents of the various approaches disagree on matters of strategy and detail but one theme is common: we have not received any revelations about the nature of the world -- including our own nature. Whatever knowledge we have has been acquired through a fallible process of conjecture and revision. This common theme will bring to mind the writings of Karl Popper and, in many respects, Popper is the father of contemporary naturalism. Along with Popper, the form of naturalism that I would defend is realistic in the following sense: it considers the acquisition of knowledge of the nature of the world to be a pursuable long-term goal of our epistemic activities. (See Brown [1987, 1988, 1990].) Popper's central interest in truth has led him to object to the pervasive concern with concepts among contemporary philosophers. Truth, Popper insists, is the fundamental epistemic concern; propositions are the bearers of truth; and the evaluation of propositions should be at the center of our epistemic focus (e.g., 1965, pp. 18-21; 1972, pp. 123-24). Concern with concepts, Popper maintains, is a distraction. Yet, this leaves us in an odd position. When we study a particular subject matter, one of our main problems is to determine what kinds of entities and processes occur in that domain. But the kinds of entities and processes we attribute to a domain will be captured in the concepts we use for describing that domain and, from a naturalistic point of view, concepts are no more available through revelation than are propositions. As our knowledge develops, we must not only propose and evaluate propositions, we must also propose and evaluate concepts.

This last point is familiar from recent work in the history and philosophy of science. At certain key junctures in the development of science such concepts as natural place, phlogiston and caloric were abandoned; concepts such as energy, gene, and quark were introduced; and, more controversially, concepts such as mass, space, and time were revised. An adequate theory of knowledge must include an account of the nature of concepts and of the ways in which concepts are evaluated. Moreover, an adequate theory of knowledge must include an account of how we can introduce new concepts that capture ideas never previously considered.

Throughout his career, Wilfrid Sellars defended a version of naturalism that takes this concern with concepts quite seriously and he developed a theory of concepts that includes an account of how concepts are introduced and evaluated. I have discussed Sellars' account of concepts in some detail elsewhere and argued that Sellars has provided the best available approach for understanding conceptual change in science (Brown, [1986]). In the present paper I will apply the Sellarsian account of concepts to a domain in which Sellars himself never applied that account: our epistemic concepts. My starting point will be a thesis that is familiar from the writings of a number of contemporary proponents of a thoroughgoing naturalism: as we attempt to learn about the world, we also learn how to learn about the world. (See, for example, Hooker [1987]; Laudan [1984, 1987]; Shapere, [1984, 1987].) In other words, from a naturalistic perspective, the nature and limits of human knowledge, and the means by which knowledge is to be pursued, are no more given to us than are the nature of the planets or the functions of our heart and liver. Translated into Sellarsian terms, part of the task of coming to understand human knowledge is to develop a system of concepts that adequately captures the nature of our knowledge. On this view, it is not enough to analyze such concepts as knowledge, belief, justification, evidence, and truth. These concepts have resulted from our experience thus far in attempting to develop knowledge, along with our reflections on that experience. But there is no guarantee that our present system of epistemic concepts cuts our actual epistemic situation at its joints. Thus once we become clear on the content of our current epistemic concepts, we may find it desireable to propose revisions or even radical changes in these concepts.

In this paper I will not propose any radical transformations in our epistemic concepts. Rather, my current concern is to argue that we can gain considerable insight into the content of our epistemic concepts by treating them as a Sellarsian conceptual system. I will proceed in two main steps. First, I will offer a relatively brief summary of a Sellarsian view of concepts. Second, I will apply this account of concepts to our epistemic concepts. I will not attempt a complete account of our epistemic concepts within the scope of a single paper, but I will use Sellars' account of concepts to clarify and defend a correspondence view of truth. It is interesting to note that while Sellars defended a correspondence view of truth, he never applied his own account of concepts to this task. As a result, the account of 'correspondence' to be developed here will be rather different than Sellars' account.

Part I Conceptual Systems {1}

The central theme in Sellars' account of concepts is that there are no isolated concepts. Concepts occur only as members of systems of interrelated concept and the content of a concept is determined by the role that concept plays in a conceptual system. There are, however, several different elements that may be involved in determining a conceptual role and there are different kinds of conceptual systems in which conceptual roles are specified in different ways. I will develop my summary of Sellars' views on concepts by considering the various factors that enter into the specification of conceptual roles.

There is one preliminary point that must be noted. Sellars usually treats concepts as if they were linguistic entities and he tends to use the terms 'conceptual system' and 'language' interchangeably. But it is an open question whether only linguistically competent beings wield concepts, and Sellars occasionally expresses doubts about the identification of concepts with linguistic entities -- although his remarks on this topic can be somewhat confusing. For example, Sellars notes that he is construing 'language' rather broadly when he identifies conceptual systems with languages (e.g. MMB, p. 91) and he explicitly denies that he is identifying concepts and words (e.g. SPR, p. 162). In other places he identifies conceptual thinking with language use but maintains that thinking is a more comprehensive notion than conceptual thinking (e.g. SK, pp. 303-4); and he sometimes notes that meaningful language is richer than the set of terms that stand for concepts (e.g., SPR, p. 115). In this paper I want to leave open the relation between concepts and words. As a result, I will reserve the term 'meaning' for cases in which I am explicitly considering linguistic entities and talk of the content of concepts and of how this content is specified. However, along with Sellars, I will not hesitate to make inferences from claims about words to claims about concepts in situations in which such inferences are clearly appropriate. Such inferences are particularly appropriate when we are concerned with adult human beings who are sufficiently sophisticated to reflect on epistemic matters. In this case, linguistic practice provides a central source of information about concepts.

1. Inferential Patterns

For Sellars there is one feature that all conceptual systems share. Since concepts occur only as members of systems of concepts, every concept is (at least) partially specified by its relations to other concepts. These relations provide a set of inference-tickets that allow us to infer directly from one concept to another. As a result, subsuming an item under a concept puts us in a position to make additional claims about that item. Sellars maintains that our beliefs about a particular subject matter are embodied in these inference-tickets. In particular, to accept a universal proposition such as 'All C are D' is equivalent, he maintains, to accepting a rule that allows us to infer 'x is D' from 'x is C'. Sellars is particularly insistent that, in practice, the inference from 'x is C' to 'x is D' is an immediate inference. Neither the proposition 'All C are D' nor the rule derived from that proposition appear as premises of this inference. (See, e.g., IM; SRI, pp. 198-204.) Rather, the proposition and the rule are located in the metalanguage that governs a conceptual system. In effect, Sellars is generalizing Lewis Carroll's (1895) thesis that the rules governing an inference cannot be premises of that inference. But this point takes on wide significance because, according to Sellars, to include the rule that 'x is C' entails 'x is D' in a metalanguage, is to build the corresponding inference into our concept of C. In other words, the rules of inference governing our use of a concept play a central role in determining the content of that concept. Moreover, every proposition of the form 'All C are D' that we accept specifies an aspect of the concept C. {2}

It is particularly important to note that Sellars rejects the claim that only analytic propositions play a role in specifying the content of concepts. Sellars accepts the analytic/synthetic distinction, but he limits the former to expressions of formal logic which provide formal rules of inference (e.g., SPR pp. 292-293, 298-299). Universal synthetic propositions provide material rules of inference and, Sellars maintains, both kinds of rules may play a role in specifying the content of a concept.

One consequence of this view is that conceptual change is a rather common phenomenon. Whenever we accept a new proposition concerning some subject matter, or reject a previously accepted proposition, we change the concepts that we use with respect to that subject matter. While many will find this to be a thoroughly objectionable consequence, there are several reasons why Sellars considers this consequence to be a virtue. First of all, for Sellars, we must not think of conceptual change only as a matter of rejecting one concept and accepting a different concept. Rather, conceptual change is typically a matter of degree. Sellars provides an account of the ways in which one concept can be similar to another concept and his full theory of concepts yields a set of tools for carrying out a detailed comparison of the ways in which two conceptual systems -- such as successive conceptual systems in a scientific field -- are similar and different (see Brown [1986] for details). A Sellarsian approach also allows us to understand the sense in which, for example, the relativistic and classical concepts of time are both time-concepts and in which classical and intuitionist negation are both forms of negation (see, especially, CC). In addition, Sellars' view provides the basis for an account of how an individual can learn a set of concepts by first learning a primitive version of those concepts and then expanding and deepening that understanding. And, a Sellarsian approach provides an account of how piecemeal change in a set of concepts can yield a new system of concepts for dealing with a subject matter that we have not previously thought about. I will return to these matters below when I discuss Sellars' views on analogy. For the moment I want to stress one point: the thesis that conceptual change is rather common is especially appropriate given Sellars' radical naturalism. From this perspective, we develop concepts as we attempt to find our way about the world, and we alter our concepts when we revise our understanding of some aspect of the world.

A related problem arises from Sellars' holistic approach to concepts. Given the connections among the concepts that form a particular conceptual system, a change in any of these connections will entail changes in all of the concepts that constitute that system. However, this consequence is less troubling for Sellars than it is for some other holistic accounts of concepts. Sellars' view does not require that all of our concepts are linked into a single massive conceptual system so that any conceptual change at all entails changes in every concept we possess.{3} Rather, we should think of ourselves as wielding a number of conceptual systems that remain distinct much in the way in which a multilingual individual speaks distinct languages. I will not try to specify precisely what constitutes a single conceptual system; I suspect that this notion must remain somewhat fuzzy. But we do have clear examples of distinct conceptual systems that are available to us. Someone may lack color concepts without any consequent deficit in her logical concepts, in her mastery of abstract algebra, in her grasp of classical mechanics, or in her system of moral concepts. Moreover, an individual may make use of two or more conceptual systems that are similar, but still distinct. So, a physicist may teach classical mechanics while doing research in relativity without confusing the two conceptual systems. This is no more mysterious than the case of a scholar who does research in middle English and writes up her results in twentieth-century English.

Note also that even though a change in any concept will change every concept in the relevant system, the resulting changes will often be quite minor. It is not as if every alteration of a concept requires that we learn a whole new conceptual system ex nihilo.

2. Entry and Departure Transitions

Most conceptual systems embody information about extra-systemic items; the exact role that these items play in specifying our concepts has long been a central issue in philosophical discussions of concepts. One of the most interesting and fruitful aspects of Sellars' view is his recognition that there are different kinds of concepts that are distinguished by the nature of their relations (or lack of relations) to an extra-systemic subject matter. Sellars admits two different kinds of relations between a concept and an extra-systemic subject matter -- entry transitions and departure transitions. {4} The nature and significance of these transitions can best be approached by examples.

A. For Sellars, inferences take place only within a conceptual system; moves between a conceptual system and extra-systemic items must be non-inferential. Sellars calls cases in which we subsume an item under a concept 'entry transitions' since they are transitions from some encounter with an item into a conceptual system. Note, however, that in describing entry transitions as non-inferential Sellars is not maintaining either that they are indubitable or that they are unlearned.{5} Rather, Sellars considers entry transitions to be habitual responses to stimuli -- but maintains that these responses play a role in specifying the content of the relevant concepts. In other words, the concepts we use for describing extra-systemic items are jointly specified by entry-transitions and by inferential patterns (e.g., SPR, p. 316).

Consider, for example, the concepts we use for describing colors of physical objects. The English word 'red' and the French word 'rouge' express the same concept because they enter into the same set of intra-systemic relations and entry transitions. First, the intra-systemic relations between such terms as 'red', 'orange', 'yellow' 'blue', and 'color' parallel those between 'rouge', 'orange', 'jaune', 'bleu', and 'couleur'; second, the same perceived items that lead English speakers to move into their color vocabulary at the 'red' node will lead French speakers to move into their color vocabulary at the 'rouge' node. If I were systematically to use 'red' where a French speaker uses 'orange', 'orange' where she uses 'jaune', and so forth, my 'red' would not translate into her 'rouge' even if there were no relevant differences in the respective intra-systemic connections. Sellars' point is a familiar one: two systems of concepts can share a great deal of formal structure but can still be different conceptual systems because they attach to extra-systemic items in different ways.{6} Sellars' view looks, at this point, very much like the traditional account of a scientific theory as a formal calculus plus a set of correspondence rules, but that account can be taken only as a first step towards Sellars' position. Sellars applies this view even to such 'empirical' concepts as red. Moreover, we shall see as we proceed that the full Sellarsian account of concepts is much richer than the traditional account of theoretical concepts.

We must be particularly firm in resisting any temptation to conclude that only the entry transitions play a role in determining the content of a concept -- even when we are dealing with elementary property concepts. To see why, consider an example. In his early work on the spectrum, Newton acknowledged only five spectral colors; orange and indigo were not included. He saw these colors and he named them, but considered them to be boundaries between colors, rather than distinct members of the spectrum (cf., Topper [1990]). At this stage of his research, Newton's concepts of orange and indigo entered into the same entry transitions as his later concepts -- which we have inherited. That is, the early Newton would have said 'orange' whenever we say 'orange' and 'indigo' when we say 'indigo'. But these concepts entered into somewhat different inferential patterns than they did in his later work. On a Sellarsian account, Newton's move from a five-color spectrum to a seven-color spectrum involved a degree of conceptual change, even though there may have been no changes in the relevant entry-transitions.

The full story with respect to entry transitions is rather more complex than is indicated by the discussion thus far. In rejecting 'the given' Sellars denies that there is a particular matched set of experiences and concepts that provide the only legitimate basis for entry transitions. As the above example indicates, even basic property concepts are not fully specified by entry transitions. At the same time, Sellars holds that a perceived item can serve as the basis for a wide variety of entry transitions. Given a sufficiently rich body of conceptual systems and a sufficiently well developed set of habits, we can move non-inferentially from a perceived item into a particular conceptual system at any of a number of different points, or into any one of several different conceptual systems. For example, when I notice a fork on the table, I may conceptualize it solely in terms of culinary utensils. But I may also note that the fork is silver-plated and move directly into a system of chemical elements and their properties. A biologist may glance at a tree and jump directly into the conceptual system of evolutionary biology. An anthropologist might move habitually from noticing a particular interaction between individuals into a system of kinship relations. In this last case, the observation might involve more than one sense (just as I might note that a car has a problem with its exhaust system as a result of what I simultaneously hear and smell) and may take place over several minutes. We can even move from some perceived situation directly into a theoretical framework that is connected only quite obliquely with the items perceived. Thus a physicist might move from seeing a pattern of lines on a photograph directly into the conceptual system of bubble chambers and neutrino interactions.

Let me dwell on this last example in order to make a further point. Sometimes we originally learn of a particular connection through inference, but later this connection becomes sufficiently familiar that it provides a non-inferential entry transition. A scientist who is designing an experiment may go through a detailed argument to show that a particular phenomenon should occur under specified circumstances -- say, that a particular pattern of traces should occur on certain bubble chamber photographs. Suppose that the experiment is carried out, the traces are found, and this provides the first bit of evidence for the existence of xions. After further work, and further confirmation, xions become well established particles, and this pattern becomes their standard signature. We might now arrive at a situation in which scientists who see this pattern recognize the presence of xions without inference -- they move directly from this pattern to the concept of an xion and its associated conceptual system. Once this occurs, the concept of an xion is different than before that signature was accepted, but the new concept is quite similar to the concept it replaces.

This last example raises a touchy question. Suppose Jane and Mary fully agree on the set of inferences licensed by the concept of a chair, but when faced with a particular modernistic contrivance Jane denies that the item is a chair while Mary unhesitatingly takes the item to be a chair. Do we want to say that this is a case of two people with slightly different concepts of a chair -- as a Sellarsian approach would seem to require -- or do we want to describe this as a case of two people who share a concept but hold different beliefs about whether a particular item falls under that concept? The crucial point to be made for present purposes is that this is not a question that can be settled once and for all by a priori examination of our current concepts of a concept and a belief. Rather, the question can only be answered by developing adequate accounts of the nature of concepts and beliefs. Since I am currently working under the hypothesis that Sellars has provided the best available account of concepts, I will assume the Sellarsian answer to this question subject to revision as our understanding grows. {7}

The previous paragraph will serve to introduce an important extension of Sellars' views. Sellars' discussions of entry transitions are confined to cases where he is concerned with physical objects. But if our beliefs about any subject matter are embodied in a set of concepts, then any such conceptual system will be related to its subject matter by a set of entry transitions. For example, a system of grammatical concepts is used to describe certain aspects of language. When we are engaged in a grammatical analysis we are moving between instances of language (which we may not currently be seeing or hearing) and our grammatical conceptual system -- much as we move between perceived items and our everyday framework of material objects when we are walking down the street. Note, also, that an individual learns a system of grammatical concepts in a piecemeal fashion and that somewhat different systems of grammatical concepts are appropriate for studying different languages. For example, the concept of a split infinitive plays no role in French or German grammar, while the concept of a detachable prefix or an ablative absolute plays no role in English grammar. The study of languages that have not yet been systematically studied may well require the invention of new grammatical concepts. In Part II of this paper I will apply this general approach to the relation between our epistemic concepts and our actual epistemic situation.

B. According to Sellars, there are genuine conceptual systems that are completely constituted by inferential patterns. These are formal systems. The constants of propositional logic provide a clear and simple example of a formal system. These constants form a self-contained conceptual system that can be used along with other conceptual systems. When this occurs, propositional logic provides machinery for combining propositions into more complex propositions, and for making inferences from such combined propositions irrespective of the subject matter of those propositions. (See Sicha [1976] and Brown [1986] for further discussion.) It is important to be clear that an application of logic to a particular subject matter does not constitute an entry transition. When we make such an application, we are not encountering an instance of, say, conjunction that plays a role in specifying what conjunction is. Conjunction remains the same whatever the subject matter to which it is applied. On a Sellarsian view, the logical constants are full-blown concepts even though no entry transitions are involved in specifying them. They are concepts of a different sort than those whose specification requires entry transitions, but they are not second-class concepts.

Mathematical systems such as the arithmetic of integers or linear algebra are also formal conceptual systems that are completely constituted by inference rules. Like propositional logic, mathematical systems can be applied to many different domains, but these applications play no role in determining the content of the mathematical concepts. Differentiation remains differentiation and integration remains integration irrespective of the subject matter under study. But mathematical concepts are less general than logical concepts. Propositional logic can presumably be applied to any propositions whatsoever, but it is an empirical matter whether a particular system of mathematics can be applied at all to a given subject matter, or whether a particular application will yield only true conclusions from true premises. If the latter does not occur we might try a different mathematical system or reconsider a physical hypothesis, but in either case the mathematical concepts we are using remain unchanged. {8}

C. Now consider departure transitions. As Sellars uses this notion, a departure transition occurs when one acts in a particular way in response to a concept. If, having recognized that an item is a chair, I proceed to sit on it, I am making a system departure transition. Note, however, that identifying an item as a chair does not require that I sit, nor does the identification of an object as red require any particular kind of action with respect to that quality. As a result, departure transitions play no role in specifying these concepts. But, Sellars holds, departure transitions are constitutive of normative concepts. For example, Sellars maintains that when a series of inferences leads to the conclusion that I that I ought to do x, this brings along a tendency actually to do x (SPR, pp. 350-52; SM, pp. 176-79). This tendency to do x in response to an 'ought' is part of the concept of an obligation; someone who has no tendency to do what she ought to do does not fully grasp the concept of an obligation no matter how well she can carry out the relevant inferences. Thus system departure transitions play a role in specifying the concept of an obligation. Sellars is, again, particularly concerned to stress that normative concepts are genuine concepts even though they are not tied to the extra-conceptual world by entry transitions. However, Sellars holds that normative concepts are essentially related to an extra-systemic subject matter -- but this is a different kind of relation than we find in the case of descriptive concepts.

D. There is one further case that Sellars hints at but does not discuss in any detail. Some concepts are both descriptive and normative -- that is, they essentially involve both entry and departure transitions. The concept of danger can provide one example. Someone who does not recognize a rogue elephant or a rogue gunslinger as dangerous -- that is, who does not move immediately from the recognition of such an item to the concept of danger -- does not fully grasp the concept of danger. At the same time, identifying an item as dangerous requires that some action be taken with respect to it, although the exact nature of this action will depend on the details of the situation. In Part II I will argue that many epistemic concepts are of this sort.

Having introduced this final type of concept, two clarifications should be noted. First, we must distinguish between a mixed concept and a mixed conceptual system. A mixed concept is one whose specification includes both entry and departure transitions. Any conceptual system that includes mixed concepts will be a mixed conceptual system, although not every concept in that system need be mixed. A mixed conceptual system can also occur because the system contains both normative and descriptive concepts, although it contains no concepts that are themselves mixed.

Second, it would be inappropriate to hold that a conceptual system can be mixed because it includes both formal and non-formal concepts. Formal concepts are specified independently of any entry or departure transitions and can only occur as members of formal conceptual systems.

3. Conceptual Functions

It is important, at this point, to be clear on the difference between using a concept and reflecting on that use. On a Sellarsian view, our use of a concept is completely captured in our mastery of the inferential patterns into which the concept enters along with any appropriate entry or departure transitions. But this mastery can be completely habitual and can occur without any explicit understanding of the elements that are involved in our habitual behavior. {9} Sometimes, however, we move from using a concept to reflecting on this use -- for example, when we undertake an analysis of a concept. When this occurs, we turn our attention to explicit consideration of the elements involved in specifying that concept. A major part of a conceptual analysis will be an examination of inferential patterns and any relevant entry and departure transitions. But there is an additional item that enters into a reflective consideration of a concept. We include a concept in a conceptual system in order to achieve some cognitive end and a full examination of a concept must include an account of its function in our intellectual economy. We have already begun this phase of an analysis when we note that a concept has a descriptive function or a normative function, but these are not the only functions that we must consider. For example, Sellars notes that '

'Perceive' is both a psychological term for a certain kind of on-goings involving the organism and the environment, and also indicates an endorsement. Thus, to say that somebody perceives something implies that the process has reached its object' (ME, pp. 292).

This observation leads Sellars to introduce the term 'ostensible perceiving' whose function is to refer only to the psychological/descriptive aspect of 'perceiving' (ME, pp. 292, see also SS, p. 89). Turning to a rather different example, a full account of the concept of mass in classical physics should include a discussion of the function of mass as a measure of resistance to acceleration.

Before proceeding, it is important that we pay some attention to our terminology in order to avoid a potential ambiguity. It is natural to use the term 'conceptual role' to refer to the job that a concept does in a conceptual system. However, the Sellarsian slogan which maintains that the content of a concept is determined by its role in a conceptual system requires that a concept's role include its inferential patterns and its entry and departure transitions, along with its function in the conceptual system. In order to avoid this ambiguity I will henceforth restrict 'role' to cases in which I am concerned with a complete account of a concept, thereby maintaining the Sellarsian slogan. When I am concerned only with the reasons why a concept is included in a conceptual system, I will speak of the function of that concept. {10} Thus, a complete analysis of a concept will include an account of its function, along with an account of its inferential patterns and any entry or departure transitions.

It is worth emphasizing that Sellars takes the distinction between the actual use of a concept and reflection on that use quite seriously. We can engage in reflection for many different reasons. In some cases, we may be interested in an explicit formulation of the contents of a concept. In other cases, we may be considering the wisdom of adding a new concept to our conceptual repertoire or deleting a familiar concept from that repertoire. In still other cases, we may consider changing a concept by changing some of the items that determine its content. Sellars has provided the following account of one case in which we are considering an alteration of an empirical concept.

Suppose that 'phi' and 'psi' are empirical constructs and that their conceptual meaning is constituted . . . by their role in a network of material (and formal) moves. Suppose that these moves do not include the move from 'x is phi' to 'x is psi'. Now suppose that we begin to discover (using this frame) that many phi's are psi and that we discover no exceptions. At this stage the sentence 'All phi's are psi' looms as an 'hypothesis', by which is meant that it has a problematical status with respect to the categories of explanation. In terms of these categories we look to a resolution of the problematical situation along one of the following lines.
(a) We discover that we can derive 'All phi's are psi' from already accepted nomologicals. (Compare the development of early geometry.)
(b) We discover that we can derive 'If C, then all phi's are psi's from already accepted nomologicals, where C is a circumstance we know to obtain.
(c) We decide to adopt -- and teach ourselves -- the material move from 'x is phi' to 'x is psi'. . . . This constitutes, of course, an enrichment of the conceptual meanings of 'phi' and 'psi'. (SPR, p. 357).

In a similar vein, those who reject the analytic/synthetic distinction are proposing that we eliminate a pair of contrasting concepts from the conceptual system that philosophers have long used for considering the logical properties of propositions. Those who are led by reflection to change their position on the analytic/synthetic distinction will also have to undertake to change certain habits.

4. Analogy

The development of knowledge is, in part, a history of postulating new entities, introducing new concepts, and thinking thoughts that had not previously been thought. Still, our ability to think new thoughts and communicate new ideas must be based on currently available concepts. According to Sellars, analogy is the key to this ability. Indeed, analogy is central to Sellars' entire philosophical project. Consider, for example, the following remark from the first chapter of Science and Metaphysics:

If the notion of one family of characteristics being analogous to another family of characteristics is obscure and difficult it is nevertheless as essential to the philosophy of science as it has been to theology and, it would seem, somewhat more fruitful. That it is a powerful tool for resolving perennial problems in epistemology and metaphysics is a central theme of this book (SM, p. 18).

The notion of analogy is closely related to the notion of similarity. For example, when we construct an analogy to introduce a new kind of entity, we postulate an entity that is similar to familiar entities. The most straightforward case occurs when the new entity shares some properties of familiar entities but differs in other properties. So, molecules were once conceived as similar to billiard balls in that molecules have mass, velocity and spatio-temporal location, while lacking color, temperature and other traditional secondary qualities. Analogies of this sort are important and useful but, Sellars argues, they only provide a means of introducing new entities that are constructed out of familiar properties. If we are to avoid the myth of the given, we need a means of introducing new properties -- properties that may have not occurred in our previous experience or thought. Sellars responds by arguing that while new entities must share properties with familiar entities, these need not be first-order properties; the familiar analogy between space and time will illustrate this point (SRI, pp. 180-82). Spatial relations do not share any first-order properties with temporal relations, but the two are analogous because they share such second-order properties as transitivity and asymmetry. In addition, Sellars maintains that an analogy must be accompanied by a metalinguistic commentary in which we describe the ways in which the new entities are similar to and different from those entities that provide the basis for the analogy. The above remarks about space and time and about molecules and billiard balls would be parts of the relevant metalinguistic commentaries. (Cf., SRI, pp. 182-84; Brown [1986, pp. 292-301].)

Sellars' key idea seems to be that if a newly proposed entity shares only higher-order properties with familiar entities, we have opened up logical space for these new entities to have different first-order properties than any we have previously encountered. But while this is important as far as it goes, it leaves us with no account of how we come to conceive what these new first-order properties are. The example of space and time is not really helpful here. It does serve to illustrate Sellars' point that analogies can be based on higher-order properties, but it draws an analogy between two items that are already familiar to us, and thus does show how we introduce new first-order properties. However, Sellars has provided us with the essential clues we need to solve this problem. For, on a Sellarsian view, to propose a new property we must introduce a new concept and the nature of this property will be built into that concept. But concepts only occur as parts of conceptual systems. Thus to introduce a new property we must introduce a new conceptual system; to understand the nature of the property it is necessary and sufficient that we learn that conceptual system. Moreover, this point does not apply only to concepts that refer to properties. Whenever we introduce a new concept, we do so by introducing a new conceptual system. Mastery of the new concept is equivalent to mastery of the new conceptual system.

But how are these new conceptual systems introduced? A Sellarsian response is that new conceptual systems are introduced via analogies that are constructed by making systematic changes in an already available conceptual system. These changes can come on any, or all, of the dimensions involved in specifying the content of a concept: we can make changes in inferential patterns, entry or departure transitions, or the functions of concepts in their respective conceptual systems. The familiar system provides a bridge to the new system, and thus provides a bridge from familiar concepts to new concepts. Moreover, the metalinguistic commentary that we offer to explain an analogy will describe the ways in which the new concepts are similar to and different from familiar concepts. In other words, I am proposing that we use Sellars' own account of conceptual systems as the basis for filling out his account of analogies.

Some examples may be helpful here. The concepts of space and time in special relativity can be understood as having been introduced in the manner sketched above. It has often been noted that there are major differences in the inferential patterns in which the classical and relativistic concepts of space and time occur -- but the classical and relativistic patterns are not totally unrelated to each other. For example, the four-dimensional space-time interval is an extension of the classical three-dimensional concept of length. We also find changes when we look at entry transitions, but these changes are much less dramatic than the changes in the inferential patterns. Special relativity places restrictions on measurements of spatial and temporal gaps that are not required by classical mechanics, but all relativistically legitimate measurements of space and time are also classically legitimate. In addition, the classical and relativistic concepts of space and time all function as measures of gaps between events.

Turning to a very different context, the history of logic has included the introduction of such concepts as intuitionistic negation and material implication. On a Sellarsian approach, these new concepts are derived from more familiar concepts by systematic changes in the relevant inferential patterns while maintaining essentially the same functions. Since these are formal concepts, they involve no entry or departure transitions.

It is particularly important to note that the fact that a new conceptual system was introduced by analogy with an older system does not eliminate the novelty of the new system. Analogy is not reduction. Rather, analogy is a means of using older concepts as a basis for introducing new concepts. Furthermore, a conceptual system that is introduced by analogy can become an autonomous conceptual system that may be understood and learned independently of the older system that once provided the basis for the analogy. This is central to Sellars' attack on 'the given' and to his view that we can come to think about the world in ways that are different from and independent of the ways our ancestors thought about the world. (See, for example, SRI, pp. 178; SPR, p. 192; SM, pp. 49, 146; NO, pp. 91-2.) Note especially that the fact that a type entity was originally conceived by analogy with more familiar entities says nothing about the relative ontological status of the two types of entities. Newly conceived entities may be ontologically independent of the entities that we thought of first; it may even turn out that entities discovered earlier in our epistemic history are ontologically dependent on entities that were only discovered later. In other words, the historical development of human thought does not provide a permanent constraint on what we can think or on what exists.

Finally, analogy can do other cognitive jobs besides serving as a means of conceptual innovation. In particular, analogy with familiar concepts is often the best means we have for coming to understand less familiar concepts -- irrespective of any historical relations that may have obtained between these concepts. Thus the systematic construction of analogies can provide a key tool in teaching unfamiliar concepts or the conceptual systems of other cultures, as well as a bridge to historically earlier conceptual systems. Indeed, an analogy between two conceptual systems can work equally well in either direction. Thus the same analogies that were used to introduce a new conceptual system may serve at a later time (say, in an historical discussion) as a means of explaining an older system to those who are familiar only with its successor.

Part II Epistemic Concepts

Our epistemic concepts form a system that includes such concepts as knowledge, belief, truth, justification, meaning, confirmation, a priority, and others. {11} Moreover, this is not a formal system. Its subject matter includes all of those items that may be subject to epistemic descriptions or evaluations -- everyday beliefs, scientific theories, philosophical claims, and more. For ease of exposition, I will assume that epistemic terms are predicates that attach to propositions. A Sellarsian account of this system will contain four parts: (1) A discussion of intra-systemic inferential patterns among the concepts that make up this system; (2) A discussion of the entry and departure transitions that tie our epistemic concepts to specific propositions; (3) A discussion of the functions of the various concepts in our system of epistemic concepts; (4) Where appropriate, the use of analogies to further explicate these concepts or to introduce new concepts. While a complete account of our epistemic concepts is well beyond the scope of a single paper, a brief exploration of some familiar inferential patterns will allow us to narrow our present focus.

1. Inferential Patterns

Traditional analyses of specific epistemic concepts amount to explorations of inferential patterns that connect these concepts. Some familiar examples will illustrate this point. Knowledge is commonly analyzed as justified true belief (plus, perhaps, some additional condition). According to this analysis, the claim that I know 'p' permits us to infer that I believe 'p', that this belief is justified, and that 'p' is true. But we may not infer from, say,'I believe "p" ' to ' "p" is true' or conversely. Nor may we infer from 'I have a justified belief that p' to '"p" is true' or conversely. Note, also, that we are already in an area in which philosophers disagree over the appropriate inferential patterns. For example, in The Republic Plato held that knowledge is not a form of enhanced belief but a distinct epistemic category. In other words, Plato would alter this system so that no inference from knowledge to belief would be permitted. I will not pursue this issue any further in the present paper.

Consider, now, one indication of the limits of such purely intra-systemic analysis. Our assessment of whether anything has been accomplished depends on whether we think that the concepts occurring in the analysis are somehow clearer than the concept being analyzed. That many philosophers do not think so is underlined by the continuing attempts at further analyses of these concepts. For example, the concept of truth is not generally considered to be intrinsically clearer than the concept of knowledge. The most common analyses of truth are in terms of correspondence or coherence, but correspondence is often held to be unintelligible and, while it is generally agreed that coherence requires more than just logical consistency, it far from clear what else is required. And consistency, which is surely necessary for coherence, is best explicated as possibility of mutual truth -- which returns us to the concept we are supposedly analyzing.

Turning to justification, note that many propositions derive their justification from their relations to other propositions which are already justified. The best understood case of this form of justification occurs when the newly justified proposition is deduced from its justifiers; but now an account of deductive validity becomes desirable. The most promising approach is to analyze validity in terms of truth-preserving transformations -- which returns us to the problematic concept of truth. In addition, maximal justification of the conclusion of a valid deduction requires that the premises be true. Another common form of justification is by induction. There is no widely accepted account of inductive justification but we can note that a maximally successful inductive justification also requires true premises. We can weaken the conditions for both deductive and inductive justification by requiring only probable premises. But now we must attach probability values to propositions, and this presumably means assessing the probability that a given proposition is true. Thus the move to probabilities does not allow us to evade the task of coming to grips with truth. Nor do we do any better if we shift to coherence analyses of justification. As we have already seen, coherence at least requires logical consistency, which brings us back to truth.

A similar situation appears if we consider a second major area of epistemic analysis: meaning. Attempts to analyze meaning in terms of truth-conditions or in terms of verifiability presumably require that the concept of truth be less problematic than the concept of meaning. Moreover, whatever approach we take to the analysis of meaning, we are going to need some notion of meaning equivalence. One necessary condition for meaning equivalence of two expressions is that they be substitutable for each other, in at least some contexts, salva veritatae.

These brief examples should be sufficient to motivate two claims. First, the process of mapping out inferential relations between our epistemic concepts provides some insight into the contents of these concepts but does not leave us with a perspicuous account of any single concept. Something is still missing. This situation is to be expected on a Sellarsian view. Since concepts occur only as members of conceptual systems, part of an account of a concept consists in mapping out its relations to other concepts. But a more complete account requires that we go outside the system of epistemic concepts -- and a Sellarsian approach tells us what more is needed.

Second, our brief exploration of intra-systemic relations suggest that truth plays a central role in our familiar system of epistemic concepts. Our accounts of other concepts tend to converge on the concept of truth while the claim that 'p' is true does not appear to license inferences to many

other epistemic concepts -- although it may license, and even require, specific behaviors. The remainder of the discussion in the present paper will be devoted to this concept. In particular, I will use the remaining aspects of our Sellarsian account of concepts to clarify and defend a correspondence view of truth. As I noted in the introduction to this paper, Sellars also defends a correspondence view of truth but he does not do so in terms of his own account of the nature of concepts.{12} Space limitations do not permit a detailed discussion of Sellars' account, but I will consider someaspects of this account as we proceed. Nevertheless, I propose to apply Sellars' views on concepts in a way that he never applied them and this will lead us to an account of truth that is only in partial agreement with Sellars' own account.

2. Entry and Departure Transitions

The concept of truth is both descriptive and normative. To assert that a proposition is true is to make a descriptive claim. Moreover, recognizing that a proposition is true requires that we undertake to believe that proposition. Thus we should expect to find entry transitions into our system of epistemic concepts at the truth node, as well as departure transitions that are required by the concept of truth. It will be useful to approach this topic by considering some features of Sellars' own account of truth.

Sellars begins his essay 'Truth and "Correspondence" ' by noting that while equivalences of the form ' "p" is true if and only if p' are central to a correspondence view of truth, such equivalences cannot provide a complete account of 'correspondence' since this formula 'is viewed with the greatest equanimity by pragmatist and coherentist alike' (SPR, p. 197). Sellars treats the semantic formula as a license to infer 'p' from ' "p" is true' and conversely, yet Sellars is far from clear about the nature of this inference. He tells us that 'if the word "true" gets its sense from this type of inference . . . "true" is a sign that something is to be done -- for inferring is a doing' (SPR, p. 206, see also the note on p. 224). But if to infer is to do something, and if all concepts are partially specified by inferences, why does this point require special mention in the case of truth? Sellars also notes that the inference from ' "p" is true' to 'p' is different from an ordinary deductive inference. In the latter case, the principle that legitimates the inference (e.g., modus ponens) does not occur among the premises; in the case of the semantic formula, the expression '"p" is true' is both a premise and the principle that justifies our asserting 'p' (SM, pp. 101-2). This is an unusual form of inference indeed. In fact, it is not inference at all as Sellars usually understands this notion. Rather, I suggest that we describe 'p' as true under one of two circumstances: either as a result of inferences made in our system of epistemic concepts, or as a result of an entry transition from the conceptual system in which 'p' occurs into our epistemic system. When we drop 'true' and remove the quotation marks from " 'p' " we are making a departure transition from our epistemic system to the system in which 'p' occurs. I want to examine each of these transitions.

In general, we make an entry transition into our system of epistemic concepts when we undertake reflection on the epistemic status of a proposition. In order to be clear on exactly what this involves, we must keep three different items in mind: (1) a specific conceptual system S; (2) the metalanguage for S, S*; and (3) our system of epistemic concepts. For example, S may be classical physics or evolutionary biology or our system of everyday material-object concepts. For each habitual inference that we make when using S, the metalanguage S* contains both a proposition and a rule. If I can infer D from C in S, then S* contains the proposition

PS*: All C are D

and the rule

RS*: Given C, D may be inferred.

The relation between RS* and PS* is analogous to the relation between, say, modus ponens and the expression

MPT: [p & (p --> q)] --> q.

Modus ponens is a formal rule of inference and is valid if and only if MPT is a tautology. RS* is a material rule of inference and we accept RS* as legitimate if and only if we hold that PS* is true. The rule RS* licenses an inference in S in much the way that modus ponens licenses an inference in all conceptual systems. The proposition PS* provides the focus of our attention when we turn to epistemic reflection.

Recall that, according to Sellars, if we have mastered S, the inference from C to D occurs as a matter of habit, but this habitual inference can occur with or without our understanding S. Understanding requires that we grasp the metalinguistic rule that licenses an habitual inference. Moreover, once we understand an inference, we are in a position to reflect on its adequacy and to consider such questions as whether we wish to keep this inference in our repertoire. If not, we can undertake a conscious effort to alter our inferential habits. In addition, once we have reached a sufficient level of conceptual sophistication to be engaged in metalinguistic reflection on our inferences, we will also be able to consider the import of an inference even though it is not currently part of our habitual repertoire. This may occur, for example, when we are considering whether to accept at proposition and thus adopt the corresponding rule and undertake to make the appropriate inference habitual. {14}

Consider, then, the case in which we are explicitly considering whether to accept PS* as true. Now the concept of truth occurs only in our epistemic conceptual system; when we undertake epistemic reflection we move into this epistemic system. None of the propositions we reflect on occur in our epistemic system, but names of these propositions can be generated in our epistemic system as the need arises. Since Sellars holds that inferences take place only within a particular conceptual system, the move into our system of epistemic concepts must take place non-inferentially. In other words, when we explicitly consider the epistemic status of a proposition, we make an entry transition from one conceptual system into another.{15}

Consider, next, a proposition 'p' that we confidently consider to be true. There are a number of situations in which we might want to note explicitly that 'p' is true. For example, 'p' might be related to another proposition whose truth-value is currently in doubt, or we might just be interested in listing the propositions we take to be true in a particular domain. Our confident belief that 'p' is true translates, in Sellarsian terms, into a tendency to make an entry transition directly from 'p' to ' "p" is true' -- that is, a tendency to think that 'p' is true and to utter ' "p" is true'. The tendency to make this entry transition plays a role in specifying our concept of truth -- just as our concept of red is partially specified by the our tendency to describe red objects as red.

On the other hand, suppose that a process of epistemic reflection leads us to the conclusion that 'p' is true. In Sellarsian terms, this will bring along a tendency to utter 'p' and to behave as if 'p' is true.{16} The move from recognizing that 'p' is true to uttering 'p' and acting as if 'p' is true are examples of departure transitions from our system of epistemic concepts back to S* and S, respectively. The tendency to make these departure transitions is also part of our concept of truth -- just as the tendency actually to do what we conclude that we ought to do is part of our concept of ought. In other words, the concept of truth has a normative dimension: it requires certain kinds of behavior and, for Sellars, if we understand a normative concept, we will experience a tendency to act in the required manner. {17}

To sum up this phase of our discussion, truth is both a descriptive and a normative concept. These aspects of the concept are captured in the fact that there are both entry and departure transitions that play a role in specifying the concept of truth. In describing these entry and departure transitions, we are engaged in one of the tasks required for an adequate analysis of truth -- a task that takes us beyond the exploration of intra-systemic relations between truth and other epistemic concepts.

We are now in a position to deal with a potentially confusing feature of Sellars' own account of truth. In his later work, Sellars defined 'truth' as 'semantical assertability': 'p' is true just in case the rules of the language in which 'p' occurs allow the assertion of 'p' (e.g., SM, ch. 4). In one respect, this doctrine is in conformity with Sellars' views on the nature of conceptual systems. On this view, everything we know or believe about a particular subject matter is built into a conceptual system. Once we explicitly accept the rule governing a specific inference in that system, we have done all that can be done to license a transition to the corresponding epistemic statement ' "p" is true'. To be sure, 'p' has been declared true only given our adoption of a specific conceptual system, and we may later change our mind about the status of that system. But our grounds for adopting a conceptual system are also grounds for accepting as true all of the propositions associated with that system. Still, two problems now arise. First, we seem to be left with a rather sweeping relativism. Which propositions are true appears to have been relativized to specific conceptual systems and we are left without any vocabulary for asking if a proposition that is assertable in some conceptual system is, in fact, true. This question is particularly pressing given Sellars' view that we are continually altering our conceptual systems, and thus altering our views on which propositions are true. Second, Sellars is not a relativist. Sellars is a scientific realist, and unless he is blatantly inconsistent, this relativization of truth cannot be the entire story.

Sellars attempts to reconcile these two positions by maintaining that while truth is relative to specific conceptual systems, some conceptual systems are more adequate than others. For example, in the case of physical science there is (presumably) one maximally adequate conceptual system in each domain and science will, one hopes, arrive at that system in the long run. In effect, we have here a distinction between 'truth' and 'ideal truth'. Truth is whatever is semantically assertable in some conceptual system, and different conceptual systems yield different truths. Ideal truth is what is semantically assertable in a completely adequate conceptual system.

But this, I think, is just a cumbersome way of saying something that can be said much more clearly. We accept a conceptual system because we believe that the associated propositions are true, but it often turns out that this was a mistake, and we try again. In doing so, we reject the claim that all the propositions assertable in the abandoned system are true. This need be no more troubling and no more mysterious than cases in which we describe a physical object as red but then, after further examination, conclude that it is not red after all. I can see no advantage to defining 'true' in such a way that the contents of a conceptual system that we no longer accept are true -- although less adequate than the contents of the conceptual system that we currently accept. Rather, one long-term goal of science (and other disciplines as well) is to establish a set of conceptual systems such that all of the propositions associated with each of these systems is indeed true. This is a traditional view of truth and the only one we need. To be sure, if we think that the attempt to find fully adequate conceptual systems is a long-term goal in most domains, then we need a way of distinguishing what is acceptable given a particular well-established conceptual system from what is not acceptable in that system. But we already have sufficient means available for making these distinctions. We regularly talk about a proposition being justified, about our having sufficient grounds for believing that a proposition is true, and about working in a conceptual system that takes the truth of some proposition as established. At the same time, we can still make the intelligible -- and often highly desireable -- remark that an accepted proposition may be false. In other words, there is no reason to use 'truth' to refer to anything but 'ideal truth' and this is the only way I will use 'truth' in the remainder of this paper. We can integrate this decision into our account of truth by considering the function of truth in our epistemic system.

3. Conceptual Functions

I submit that truth functions in our system of epistemic concepts as an ideal in a sense that is close to Kant's.{18} Kant considered the idea of the complete causal sequence responsible for an event as one example of a regulative ideal: we can never establish such a sequence, but the ideal directs us to continue the search for causes preceding those causes we have already discovered, and this quest increases the scope of our empirical knowledge. Now in taking truth as a regulative ideal we need not be quite so pessimistic about our ability to achieve this ideal. There may well be domains in which we can complete the quest for truth. But whether we can achieve this end or not, truth plays the role of an ideal at which many of our cognitive projects aim. Moreover, pursuit of this ideal has served as a major impetus to the correction of previous beliefs and a powerful generator of new discoveries. {19}

Treating truth as an ideal will help us understand why our epistemic system contains some of the concepts and inferential links that we find there -- that is, it will help us understand the function that these concepts play in our epistemic considerations. Let me illustrate the point by considering some aspects of the relation between truth, knowledge and justification.

While truth functions as a central epistemic end, there is an important sense in which knowledge is a more pressing concern than truth. For it is often difficult to determine if a proposition is true and, as Plato argued in the Theaetetus, true belief without reasons for that belief is of limited value. If we have merely stumbled into a true belief, we can stumble out of it just as easily. This need for reasons is captured in the concept of justification. Moreover, while justification and truth are both necessary conditions for knowledge, we treat these conditions differently when we must assess the epistemic status of a proposition. We do not carry out independent assessments of whether a proposition is justified and of whether it is true in attempting to assess a knowledge claim. Rather, our justification procedures provide our working criteria for truth and we accept a proposition as known when we judge that we have provided sufficient justification to consider it true. Still, we maintain the distinction between knowledge and justified belief because we recognize that our justification procedures are fallible; even a well justified belief may be false. We should also note that justification admits of degrees and that knowledge requires adequate justification. As a result, knowledge is doubly ideal and thus even harder to achieve than truth. Thus, we often find ourselves acting on the basis of those beliefs that are best justified, given existing constraints, while we recognize that these beliefs are not adequately justified and thus not instances of knowledge (although they may be true). Our current system of epistemic concepts includes the resources needed for making these distinctions.

Note also that the reasons for taking truth as a fundamental epistemic end are not terribly bewildering. We seek truth because we want to find out how things are. In purely intellectual terms this is an end in itself, but it is also an end of considerable practical importance. Whether we are balancing a checkbook, determining the carrying capacity of a beam, deciding if the beer is cold, assessing the best treatment for a disease, or what have you, we want the correct answer to our question -- that is, we want the truth. This pragmatic desire for truth becomes particularly pressing when we recognize that decisions to act on the basis of accepted beliefs typically involve inferences as to what we might encounter in the future. But inferences are no more reliable than their premises and thus true beliefs about the items we are dealing with provide a necessary component of a maximally reliable prediction technique. Whether our inferences are deductive or inductive, we have failed to provide maximally reliable grounds for accepting the conclusion of an inference unless the premises are true.

In other words, a number of important features of our epistemic situation are captured in the distinctions and links that we find in our available epistemic system. From a Sellarsian point of view, these links are material connections that have been built into our epistemic concepts as a result of our attempts to understand our epistemic situation. Moreover, these material connections are subject to reconsideration as we come to learn more about our epistemic situation.

4. Analogy

Propositions purport to carry information and a true proposition is one that does carry the information it purports to carry. {20} Moreover, it is this idea of 'carrying information' that is the key point of a correspondence account of truth: a proposition corresponds to its subject matter when it carries information about that subject matter. This requires no mysterious 'third thing' that somehow ties the proposition to that subject matter. Indeed, the impression that there is something mysterious about the claim that truth is correspondence derives from a misunderstanding about the nature of this claim. To see why, note that correspondence is not itself a member of our system of epistemic concepts. Rather, the idea that truth is correspondence occurs as part of a metalinguistic gloss on the concept of truth. In this gloss, 'correspondence' is being used in an analogical sense to further clarify the concept of truth -- 'correspondence' is not being offered as a definition of 'truth'. The significance of this claim can be brought out by developing the analogy in some detail.

We will begin with a non-analogical notion of correspondence that has nothing in particular to do with propositions or conceptual systems: two distinct items correspond in certain respects when they are identical in those respects. Consider, as an example, the Eiffel Tower and a model of the Eiffel Tower. In a relatively simple case, the model will share some features of the Eiffel Tower, although it must differ from its prototype in some respects. Where the properties of the model and prototype are the same, they correspond. To first approximation, we can say that a better model corresponds to the original in more features than a poorer model. As an extreme example, consider a detailed copy of the Eiffel Tower, located in Texas, in which every dimension of the tower, along with the materials, their current degree of wear and corrosion, and so forth, have been matched. Given these correspondences, we can learn a great deal about the Eiffel Tower by studying the model. The model will provide the information we want, and will thus serve as an epistemic proxy for the original. But there will always be some properties in which the model does not correspond to the prototype. At the very least, the model's latitude and longitude will not correspond to that of the prototype and thus we could not learn the prototype's latitude and longitude solely by studying the model. Nor could we learn the distance of the Eiffel Tower from the Seine or the Ecole Militaire, nor whether the Eiffel Tower is currently wet from rain, by studying our model. In a similar way, identical twins correspond in height, facial features, genetic makeup, and so forth, and either twin will be as good as the other (and thus one twin can serve as a model for the other) if we wish to determine any of these properties. But, because the twins are distinct individuals, there will always be some features in which they do not correspond -- that is, there will be properties of one twin that we cannot discover by examining the other twin. These include current location, marital status, perhaps weight, and much more.

Now the claim that a better model shares more features of the original than a poorer model is only a first approximation because there is a pragmatic aspect involved in the notion of a model. We typically create a model for a specific purpose, and this purpose plays a role in determining the respects in which we want the model and the original to correspond. Given a particular purpose, a model may become a poorer model if it shares too many features with its original. A tourist's model of Eiffel Tower that weighed as much as the actual tower would be a very poor model indeed. Here we might be satisfied with a scale model. We might, for example, construct a plastic model that included a distinct piece for each distinct strut in the Eiffel Tower, with the length of each member reduced in the same proportion. In doing so, we might make no attempt to provide a scale model of the cross-section of each strut, and there might be no consistent relation between the weight of a strut in the model and the weight of the corresponding strut in the tower itself. In this case, the number of struts of the model would correspond to the number of struts of the tower, while there would be no correspondence between the weight of a strut in the model and the weight of the corresponding strut in the tower. There is, however, an extended sense in which the length of a strut in the model corresponds to the length of a strut in the tower. For while we cannot determine the length of a tower strut just by measuring the corresponding strut in the model, we can determine this length if we know the scale and supplement the measurement with the appropriate calculation. The relation between the length of a strut in the model and in the tower is somewhat indirect, but these correspond in the sense that -- given appropriate background knowledge -- information about the original can be derived from the model.

A tourist's model of the Eiffel Tower need not even be a scale model; for this purpose, it will often be enough if the model shares the overall gestalt of the Eiffel Tower. We might even introduce considerable distortions of scale in order to include other kinds of information that a tourist might want in a model. So, a model of the Eiffel Tower might be on a base that shows the tower's general relation to the Seine, the Champs de Mars, and the Ecole Militaire -- even though none of these relations are constructed to any specific scale. This model carries considerably less detailed information about the tower than a scale model, although our latest model carries some additional, but not very precise, geographical information. Of course, we could improve this geographical information. If the base were made to scale (even though the tower is not to scale) and included a line corresponding to the near bank of the Seine, then the distance from a point at the base of the tower to the Seine could be determined from the model. This distance on the model would correspond to the distance in Paris, even though the image of the Seine on the model has no far bank and contains no water.

By now the key point I am after should be clear. When a model corresponds to a prototype in a particular respect, the model serves as a source of information about that feature of the prototype -- we can learn that feature of the prototype from the model. In the root sense of 'correspond', model A corresponds to prototype B to the extent that A duplicates B, and this duplication implies that A can serve as a source information about B. But we can now construct an analogical sense of 'corresponds' by reversing the direction of this implication. In this analogical sense, A corresponds to B in any case in which A serves as a source of information about B. {21}

We are now ready to move directly to the case that concerns us in this paper: the case in which a proposition corresponds to some item that is distinct from that proposition. In our analogical sense, a proposition, 'p', corresponds to an item B if 'p' carries information about B. A proposition need not share any features of an item to which it corresponds, nor need there be any special hook between the proposition and that item. All that is required is that 'p' can serve as a source of information about B. To the extent that 'p' provides information about B, 'p' corresponds to B. Most propositions purport to carry information about some item other than themselves, and those propositions that carry the information they purport to carry are described as true.

Let me press this account of correspondence with another example. Consider three different ways in which I might seek to convey to you the information that there are two chairs in my study: (1) I might take you into the study and let you see this for yourself; (2) I might hand you a photograph of my study showing the two chairs; (3) I might say to you, 'There are two chairs in my study'. If I take you into the room you can see the two chairs; the notion of correspondence is not relevant. The photograph and the proposition are indirect ways of informing you about my study, and they can both carry information about my study without sharing any properties with each other or with my study. If the photograph and the proposition correctly report the number of chairs in the study, then they correspond to the study in this respect. To be sure, the photograph may carry more information about the room than a single proposition does, and the photograph might thus correspond to more features of the room than the proposition does. But this does not undermine the point that we can get some information about the room from a proposition. Note also that either the photograph or the proposition can fail to correspond to the present situation in my study. In each case we have concepts available which allow us to express the relevant assessment: we describe photographs as accurate or inaccurate; we describe propositions as true or false.

I have been dwelling on cases in which a proposition corresponds to some observable situation, but nothing new need be added to apply our account to non-observables. Propositions about electrons -- or about gods and demons, for that matter -- purport to convey information about such items. True propositions do convey such information, false propositions do not convey such information. There are, of course, questions about how we determine whether a particular proposition in one of these domains is true, but that is a question about justification and it is a different question than the one that concerns us here. Here I am only concerned to explicate what it means to say that a true proposition corresponds to its subject matter. Note also that this account extends directly to propositions whose subject matter is other propositions, e.g., propositions that describe logical features of some proposition.

Finally, note that our analogical sense of 'corresponds' will serve to distinguish a correspondence account of truth from a coherentist or a pragmatist account. With regard to coherence, we need only note that coherence accounts of truth seek to capture all there is to say about truth in relations between propositions. But propositions often serve as a source of information about items that are not themselves propositions. Moreover, there are many ways in which two propositions can 'cohere' without one of these propositions serving a source of information about the other. In the case of the pragmatist account we need only note that a proposition may carry information about some situation even though that information is not useful, and that it is sometimes useful to believe a proposition even though that proposition is not true.

5. Conclusion

My aim in this paper has been to sketch a Sellarsian account of conceptual systems and apply that account to the concept of truth as it functions in our system of epistemic concepts. This required that I map out inferential relations between truth and other epistemic concepts, that I describe the entry and departure transitions that occur between truth and propositions that we describe as true, and that I offer an account of the function of truth in our system of epistemic concepts. In addition, on a Sellarsian view, analogy is a powerful tool for generating new conceptual systems and for explaining unfamiliar concepts. Since our epistemic concepts form a reasonably familiar conceptual system, analogies should not be required; the above three steps are all that are required by way of explicating truth. Nevertheless, in attempting to distinguish the present account of truth from coherence and pragmatic accounts, we encountered the claim that true propositions correspond to their subject matter. While this is a traditional claim, many philosophers contend that there is something intrinsically mysterious about the notion of correspondence being used. I attempted to remove this air of mystery by offering an account of the relevant concept of correspondence as a notion derived by analogy from a more familiar concept. Each of the four steps of this account of truth can, and should, be developed in greater detail. But aside from questions of detail, if a Sellarsian approach to conceptual systems is correct, nothing more is required in principle to explicate the concept of truth.

[Return to Main Page]


  • Brown, H. 1986. 'Sellars, Concepts and Conceptual Change'. Synthese 68, 275-307.

  • Brown, H. 1987. Observation and Objectivity. New York: Oxford University Press.

  • Brown, H. 1988. 'Normative Epistemology and Naturalized Epistemology'. Inquiry 31, 53-78.

  • Brown, H. 1990. 'Prospective Realism'. Studies in History and Philosophy of Science 21, 211-42.

  • Carroll, L. 1895. 'What the Tortoise Said to Achilles'. Mind 4, 278-80.

  • Churchland, P. M. 1979. Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge University Press.

  • Dretske, F. 1981. Knowledge and the Flow of Information. Cambridge: MIT Press.

  • Gutting, G. 1977. 'Philosophy of Science', in C. F. Delaney, et al., The Synoptic Vision. Notre Dame: University of Notre Dame Press, 73-104.

  • Hooker, C. 1987. A Realistic Theory of Science. Albany, NY: SUNY Press.

  • Laudan, L. 1984. Science and Values. Berkeley: University of California Press.

  • Laudan, L. 1987. 'Progress or Rationality? The Prospects for Normative Naturalism'. American Philosophical Quarterly 24, 19-31.

  • Popper, K. 1965, Conjectures and Refutations. New York: Harper and Row.

  • Popper, K. 1972. Objective Knowledge. Oxford: Oxford University Press.

  • Sellars, W. 1950. 'Language, Rules and Behavior', in Sidney Hook (ed.), John Dewey Philosopher of Science and Freedom. New York: Dial Press, 289-315.

  • Sellars, W. 1952. 'Mind, Meaning and Behavior'. Philosophical Studies, 3, 83-95.

  • Sellars, W. 1953. 'Inference and Meaning'. Mind 62, 313-38.

  • Sellars, W. 1958. 'Counterfactuals, Dispositions, and the Causal Modalities', in H. Feigl, M. Scriven and G. Maxwell (eds.), Minnesota Studies in the Philosophy of Science II. Minneapolis: University of Minnesota Press, 225-308.

  • Sellars, W. 1963. Science, Perception and Reality. New York: Humanities Press.

  • Sellars, W. 1964. 'Induction as Vindication', Philosophy of Science 31, 197-231.

  • Sellars, W. 1965. 'Scientific Realism or Irenic Instrumentalism', in R. Cohen and M. Wartofsky (eds.), Boston Studies in the Philosophy of Science 2. Dordrecht: D. Reidel, 171-204.

  • Sellars, W. 1968. Science and Metaphysics. New York: Humanities Press.

  • Sellars, W. 1973, 'Conceptual Change', in G. Pearce and M. Maynard (eds.), Conceptual Change. Dordrecht: D. Reidel, 77-93.

  • Sellars, W. 1974. 'Meaning as Functional Classification', Synthese 27, 417-437.

  • Sellars, W. 1975. 'The Structure of Knowledge', in H. Castañeda (ed.), Action Knowledge and Reality. Indianapolis: Bobbs-Merrill, 295-347.

  • Sellars, W. 1979. Naturalism and Ontology. Reseda, CA: Ridgeview Publishing Company.

  • Sellars, W. 1982. 'Sensa or Sensings: Reflections on the Ontology of Perception'. Philosophical Studies 41, 83-111.

  • Sellars, W. 1989. The Metaphysics of Epistemology, P. Amaral (ed.). Reseda, CA: Ridgeview Publishing Co.

  • Shapere, D. 1984. Reason and the Search for Knowledge. Dordrecht: D. Reidel.

  • Shapere, D. 1987. 'Method in the Philosophy of Science and Epistemology', in N. Nersessian (ed.), The Process of Science. Dordrecht: Martinus Nijhoff, 1-39.

  • Sicha, J. 1976. 'Logic: The Fundamentals of a Sellarsian Theory', in J. Pitt (ed.), The Philosophy of Wilfrid Sellars. Dordrecht: D. Reidel, 257-286.

  • Topper, D. 1990. 'Newton on the Number of Colours in the Spectrum'. Studies in History and Philosophy of Science 21, 269-279.

[Return to Main Page]


{*} I wish to thank Mylan Engel and Nancy Nersessian for comments on earlier versions of this paper. [Back]

Editor's Note: This article originally appeared in Inquiry 34 (1991): 323-51; and appears here through the generous help of Professor Harold Brown. The Greek characters phi and psi are rendered by their names, i.e., as 'phi' and 'psi'. AC
{1} The following abbreviations will be used in references to Sellars' texts (see the References for full citations):

CC 'Conceptual Change'
CDCM 'Counterfactuals, Dispositions, and the Causal Modalities'
IM 'Inference and Meaning'
IV 'Induction as Vindication'
LRB 'Language, Rules and Behavior'
ME The Metaphysics of Epistemology
MFC 'Meaning as Functional Classification'
MMB 'Mind, Meaning and Behavior'
NO Naturalism and Ontology
SK 'The Structure of Knowledge'
SM Science and Metaphysics
SPR Science, Perception and Reality
SRI 'Scientific Realism or Irenic Instrumentalism'
SS 'Sensa or Sensings' [Back]

{2} Sellars holds that universal generalizations are a special case of statistical generalizations of the form, 'The proportion of C's that are D's is n/m'. Thus a conceptual system may license statistical inferences, and these inferences will also be built into our concepts of C and D. For discussion see CDCM, pp. 289-96; IV, pp. 211-12; and Gutting (1977), pp. 94-100. I will limit discussion in this paper to universal generalizations. [Back]

{3} See Brown (1986, pp. 288-92) for defense of this claim. Sellars has not developed this point in any detail although there are some textual grounds for attributing this view to him, e.g., LRB, pp. 311-12. [Back]

{4} Sellars usually uses the terms 'language entry transition' and 'language departure transition' in accordance with his practice of treating conceptual systems as languages. Since I wish to remain neutral on the exact relation between concepts and language, I will speak of 'system' entry and departure transitions. [Back]

{5} See, especially, 'Empiricism and the Philosophy of Mind' in SPR. [Back]

{6} I will have more to say about 'formal structure' in sub-section B. [Back]

{7} See Churchland (1979, ch. 3) for a detailed defense of this option. [Back]

{8} This last remark should be taken as a correction of a contrary view that was defended in Brown (1986, pp. 299-300). [Back]

{9} In contemporary jargon, mastery of a concept yields procedural knowledge. [Back]

{10} This will not eliminate all possible sources of confusion. Sellars sometimes uses the term 'function' in the way in which I am using the term 'role' -- e.g., when he writes of 'Meaning as Functional Classification', as in MFC and NO, ch. 4. [Back]

{11} Although Sellars does not offer an explicit account of epistemic concepts along these lines, he does describe Chisholm as being 'concerned with buying into a system of concepts in the neighborhood of the concept of knowledge' (ME, p. 267). [Back]

{12} See particularly 'Truth and "Correspondence"' in SPR and SM chs. 4 and 5. [Back]

{13} It would seem that all of our epistemic concepts are descriptive, but they are not all normative. For example, to say that I believe 'p' does not involve a normative dimension. [Back]

{14} See the passage from Sellars quoted on p. 18. [Back]

{15} In a similar way, aesthetic or moral reflection requires an entry transition into our systems of aesthetic or moral concepts, respectively. [Back]

{16} If the truth of 'p' is a new discovery, we may have to train ourselves to behave as if 'p' is true. {17} Recall that purely descriptive concepts may lead to system departure transitions, but do not require such transitions. [Back]

{18} See also Popper 1972, pp. 29-30. [Back]

{19} This is not to say that seeking truth is the only important generator of new discoveries. See Brown (1988, Part II) for discussion. [Back]

{20} In the remainder of this paper I will use 'information' only in the sense according to which there is no such thing as false information -- cf. Dretske (1981). A false proposition fails to provide the information that it purports to provide. [Back]

{21} See Brown (1986, pp. 298-300) for further discussion of cases in which an analogy is constructed by taking a consequence of the root sense of a term as a defining characteristic of the analogical sense of that term. [Back]

[Return to Main Page]