Andrew Chrucky, Critique of Wilfrid Sellars' Materialism, 1990

CHAPTER 5

LANGUAGE AND THOUGHT

      The problem of how thought and language are related is one of the major problems in contemporary philosophy, and it is particularly troublesome for Materialism which is committed to the denial of dualism. The main reason for the difficulty is that we do not have a clear command of the concepts of thought and language. Consequently, different claims about their relation are possible -- depending on how "thought" and "language" are understood. In this chapter I propose to discuss their relation based on Sellars' latest writings.

I. DESIDERATA

      Any proposed theory of the relation of thought to language must be able to meet the following conditions of adequacy.

  1. Give an explanation for the pre-linguistic intelligent behavior of children. That children behave intelligently is commonly taken for granted. This is exemplified by various behaviors: the ability to discriminate and classify objects, the strategies and 'insights' used by children to achieve given ends. The use of tools is an obvious example. Even such seeming obvious actions as avoiding bumping into things, not tripping over things, reaching to get something, not falling from heights, must be explained. A specific recommendation would be that any proposed theory should come to grips with the findings and speculations of such child psychologists as Jean Piaget and Lev Vygotsky.{1}

  2. Give an explanation for the claims of the deaf and blind, or the deaf and mute that they had thoughts prior to learning a language; e.g., the case of Helen Keller and I. Ballard.{2} Also take into account the data available about children growing up in isolation from language, such as the case of Victor, the "wild boy of Aveyron," who was found in Southern France in 1799; the Indian girls Kamala and Amala, who were found in Midnapur, India, in 1920; Kaspar Hauser, who was found near Hanover in 1823; and the girl Genie, who was found in 1970.{3}

  3. Give an explanation for the seemingly intelligent behavior of animals as reported by such ethologists as Konrad Lorenz.{4} Animals manage to move about avoiding obstacles and dangerous situations. They stalk and kill effectively. Sometimes they use tools for certain ends. And some of these tool discoveries seem to be passed from generation to generation. For example, the learned behavior of Japanese macaques of Koshima Island to wash sand from sweet potatoes and grains of wheat before eating them. Explain the communication between animals: e.g., the dance of the bees, chimps and gorillas using a sign "language,"{5} dolphin communication.{6}

  4. Account for the possibility of humans learning a language. How is it that humans are able to learn a conventional language such as English, while no other animals can -- not even the chimps and gorillas (Washoe, Koko). What resources are needed to learn a language?

  5. Account for the existence of a distinction in our language between thoughts and language. We do make a distinction between these and describe them as if they were independent. And how do we account for such phenomena as our ability to perceive much more than we have words to express. For example, we can distinguish between various hues of colors for which we have no names. Then there is the phenomenon of not being able to express what we are thinking, or not finding the adequate words which express our thoughts. How is this possible?

  6. Distinguish, if possible, a human use of language from an ideal computer's. Can an ideal computer be said to be intelligent? What criteria must be satisfied? Is the Turing test adequate?{7} Some claim that the existence of Gödel's incompleteness theorem makes it theoretically impossible for a computer to think.{8}

  7. Relate the theory of thought and language to the findings of brain research. For example, is the theory consistent with our knowledge of various forms of brain dysfunctions, such as, aphasia, agnosia, etc.?{9}

      I am not proposing to consider all these -- that would be a staggering undertaking. I am only suggesting that any adequate theory of thought and language must do justice to these questions and explain the phenomena. What I will try to do is at least not write in such a way as if I were totally unaware of these areas of research, and I certainly don't want to claim anything that goes counter to what is well established by science.

II. THOUGHT AND CONVENTIONAL LANGUAGE

      Philosophers are divided on whether thought can exist prior to the acquisition of a conventional language. As a first approximation, it appears that behaviorists favor the total dependence of thought on language. I have in mind Watson, Ryle, Wittgenstein, Malcolm, Skinner, Rorty, Quine, and Rosenberg. The "mentalists" or "cognitivists" appear to favor the partial independence of thought from language. I have in mind Descartes, Locke, Kant, Husserl, Russell, Moore, Lewis, Piaget, Hall, Marras, Chisholm, Castañeda, Chomsky, and Fodor.

      On first impression, I would place Sellars with the behaviorists and ascribe to him the view that thought is dependent on language. And looking over the critical literature on Sellars, this view is sometimes ascribed to him.{10} In general, the shared claim of Sellars' critics is that language genetically presupposes, at least, some kind of sign understanding. These critics would all probably agree with the following observation made by Nelson Goodman:

The linguist may be forgiven for a vocational myopia that blinds him to all symbol systems other than languages. Anyone else recognizes that gestures, nods of approval and disapproval, pointings, facial expressions, bodily demonstrations, sketches, diagrams, models, play an important role in the acquisition and inculcation of skills of all sorts; and that mastery of symbols of many of these kinds occurs before, and aids enormously, in the acquisition of language.{11}

      The implication of this kind of criticism, if it is directed towards Sellars, is to attribute to him the belief that the necessary condition of having awareness, thoughts, and symbolic or conceptual abilities is the possession of a conventional language. All these critics, in rejecting Sellars' alleged account of thought and language, accept the existence of thoughts as genetically prior to language. A seeming corollary of this is the claim that we have concepts prior to language, and that we use symbols prior to language, that there are "natural signs" prior and independently of "conventional signs,"{12} and, to borrow the title of Fodor's book,{13} there is a "language of thought" prior to the language of speech.

      I had similar misgivings about Sellars' position, but after much reflection I would not make such interpretations of Sellars, at least not in these ambiguous, vague, and somewhat elliptical ways. I wouldn't even use the same formulations which Sellars himself actually uses about his views, and not without disambiguation, and setting the theses in a context. How could Thurston, for example, make her claim be consistent with a passage such as:

The 'linguistic model' begins to look far too narrow and specialized to capture the nature of thinking, even at the strictly human level -- let alone in the sense in which animals think.{14}

      I grant that a limited reading of Sellars would suggest the interpretation that thought is dependent on language, especially if it is based of his celebrated "Empiricism and the Philosophy of Mind", where we find the notorious passage cited by Thurston:

all awareness of sorts, resemblances, facts, etc., in short, all awareness of abstract entities -- indeed, all awareness even of particulars -- is a linguistic affair.{15}

But a more extensive reading of Sellars would unearth such passages as: "there is a legitimate sense in which animals can be said to think."{16}

      Sellars comes to grips with these misinterpretations of his position only after 1980 in his two essays "Behaviorism, Language and Meaning" and "Mental Events."{17} His opening statement in "Mental Events" is:

I find that I am often construed as holding that mental events in the sense of thoughts, as contrasted with aches and pains, are linguistic events. This is a misunderstanding. What I have held is that the members of a certain class of linguistic events are thoughts. The misunderstanding is simply a case of illicit conversion, the move from 'All A is B' to 'All B is A'.{18}

Sellars is clearly expressing the view that there are thoughts which are independent of language. In "Behaviorism, Language and Meaning," he tells us that:

Thus viewed from the standpoint of methodology, verbal behaviorism is perfectly compatible with the idea that there are pre-linguistic representational activities.

Indeed, the ability to have primitive representational episodes might be not only pre-linguistic, but innate .{19}

      It is clear that Sellars does not deny pre-linguistic representational abilities, as Marras, for example, seems to contend quite explicitly.

      How is this misunderstanding to be explained? It has something to do with the way Sellars characterizes a language, and how he determines which characteristics are essential to it and which ones are not. In "Behaviorism, Language and Meaning" and "Mental Events," Sellars introduces the notion of a Representational System (RS), of which a Language (L) is a species, by which "the organism constructs maps of itself in its environment, and locates itself and its behavior on the map."{20} He adds,

Such representational systems (RS) or cognitive map-makers, can be brought about by natural selection and transmitted genetically, as in the case of bees. Undoubtedly a primitive RS is also an innate endowment of human beings. The concept of innate abilities to be aware of something as something, and hence of pre-linguistic awareness is perfectly intelligible.{21}

But isn't this last sentence in contradiction with, "all awareness of sorts, resemblances, facts, etc., in short, all awareness of abstract entities -- indeed, all awareness even of particulars -- is a linguistic affair" ?{22} Before accusing Sellars of a contradiction, the more sympathetic approach is to accuse Sellars of a carelessness of formulation and of an ambiguous use of words. It is the latter approach that I will take.

      The source of the muddle lies in a failure by Sellars to remind the readers in the appropriate contexts of the distinction between a Representational System and a Language. And because he rarely writes about thoughts with an RS context presupposed, but with a clear L context presupposed, the result is the appearance that Sellars ascribes thoughts to language users only.{23}

      I will examine both RS and L, and their relation, starting with an RS. But before I examine the details of an RS, let me just point out some of the claims made about an RS. One claim is that the necessary condition for thought (of any type) is the possession of an RS. Another claim is that RSs are possessed by some animals and pre-linguistic children, as well as by language users. Still another claim is that an RS may be a genetic endowment.

      The RS-L distinction, though formulated explicitly in "Mental Events," has precursors in Sellars' writing. The first time that Sellars made a similar distinction was in "Language, Rules and Behavior" , published in 1949 -- his fourth publication. Here he made the distinction between 'tied symbol behavior' (corresponding to RS) and 'rule-regulated symbol behavior' (corresponding to L).{24} In Science and Metaphysics,{25} Sellars stresses the distinction between 'non- conceptual representations' and 'conceptual representations'. So the distinction seems to be something that may have been presupposed by Sellars in all his writings.

      How does this distinction between RS and L shed light on the misleading passages in Sellars? Let me explain it this way. RS can be viewed as a truncated analogue of L. As such, many concepts appropriate to L will also be appropriate in an analogous manner to RS.

      To make a distinction between concepts appropriate to L and RS, I will use Sellars' device of placing a prefix to a word to represent the RS analogue of an L concept. In doing this, I want to point out that Sellars' own practice here is ambiguous. He uses the 'ur-' prefix convention in some places to designate a pre-conceptual ability;{26} but at other places he uses it to designate a primitive form of a concept.{27} In view of this ambiguity, I will use the prefix 'ur-', to designate a primitive concept of L, and the prefix 'rs-' to designate an analogue in RS. One source of the misunderstanding of Sellars can, then, be explained by a failure to note, or appreciate the 'tied symbol behavior' (RS) versus 'rule-regulated symbol behavior' (L) distinction.

      But another source of misunderstanding may be due to Sellars' terminology. Sellars uses psychological verbs and their nominalizations, for the most part, to characterize only the behavior of humans with a language because he believes that the necessary condition for ascribing appropriate psychological states to a mature human subject is that the subject possesses a language. Thus, Sellars ascribes awareness, thoughts, knowledge, beliefs, etc. (in a primary sense) only to language users. And he denies that these ascriptions -- in the primary sense -- can be made to pre-linguistic animals and children. Of course Sellars' manner of speaking goes counter to a widespread linguistic practice of using psychological verbs and their nominalizations to characterize animal behavior. So the impression is left in the reader's mind that Sellars denies psychological states to pre-linguistic children and animals.

      But such a conclusion is only a half-truth. For, in fact, Sellars has always claimed that psychological ascriptions can be made to pre-linguistic children and animals in a derivative, analogous manner; as is clearly presented in the following passage:

Not all 'organized behavior' is built on linguistic structures. The most that can be claimed is that what might be called 'conceptual thinking' is essentially tied to language, and that, for obvious reasons, the central or core concept of what thinking is pertains to conceptual thinking. Thus, our common-sense understanding of what sub-conceptual thinking -- e.g., that of babies and animals -- consists in, involves viewing them as engaged in 'rudimentary' forms of conceptual thinking. We interpret their behavior using conceptual thinking as a model but qualify this model in ad hoc and unsystematic ways which really amounts to the introduction of a new notion which is nevertheless labeled 'thinking'. Such analogical extensions of concepts, when supported by experience, are by no means illegitimate. Indeed, it is essential to science. It is only when the negative analogies are overlooked that the danger of serious confusion and misunderstanding arises.{28}

      It is clear, then, that Sellars ascribes thinking, or, more precisely, in my new terminology, 'rs-thinking' to pre-linguistic children and animals. What is not clear, as the above passage expresses, is the difference between a Language and a Representational System. Let's turn to this task.

III. REPRESENTATIONAL SYSTEMS AND LANGUAGE

      In this part I want to demarcate an animal representational system, RS, which suffices for the occurrence of a 'mental event', from a conventional human language, L. Such a demarcation is implicit in Sellars' writings, but it is explicitly presented only in his recent article "Mental Events" . A great deal of obscurity has been removed for me from Sellars' writings by this essay. Understanding this demarcation has helped me to understand much of Sellars' previous writings, and many things have fallen into place. This demarcation enables us to appreciate Sellars' insistence over the years that the presence of concepts is totally dependent on the presence of a conventional language.

      A demarcation between an RS and L will enable us to understand the following claims, which, I believe, Sellars makes.

  1. An L is a species of RS.
  2. Some animals have an RS.
  3. No non-human animal (in fact) has L.
  4. No non-human animal (in fact) has concepts.
  5. 'x has concepts' is equivalent to 'x has L'

      In the following discussion, I am assuming that we are dealing with animals having the capacities of at least rats. I am also assuming that a teleological description of, for example, rat behavior is appropriate; specifically, that it is appropriate to describe a rat, for example, as wanting something, as searching for something, as avoiding something, and so on.{29} It is with this kind of background assumption that the following discussion proceeds. In one sense, it explains why teleological descriptions are appropriate.

      Minimally, to have an RS something must represent something else. Call the representing state a "symbol" and the represented thing an "object." But immediately Sellars rejects any claim that such a correlation, even a multitude of such correlations, is sufficient for an RS. Note that this objection is enough to indicate for us that Sellars would reject a "private language" as depicted by Wittgenstein. Such a purported "language" was to consist of the sign 'E' standing for some private state or object. The privacy here is not relevant. Sellars would also reject a so-called "public language" as a language, which consisted of the sign 'E' standing for some public state or object. But worse, neither of these so called 'languages' would even qualify to be RSs. What more is needed is a variety of kinds of correlations -- a complexity. He suggests that in order to have an RS "this correlation may essentially involve other correlations -- thus between it and other representational states and between representational states and actions."{30} He presents this more explicitly elsewhere:

To be sure, the child must also acquire the ability to respond to cases with the right words, but until these response patterns have been integrated with intra-linguistic moves and language-departure transitions, they have no conceptual character and do not even count as labels.{31}

      So far, then, Sellars is claiming that an organism in order to have an RS must have, as necessary conditions, the following three characteristics:

  1. An organism correlates symbols with objects which are in the environment (or in itself).
  2. An organism correlates symbols with other symbols.
  3. An organism correlates symbols with physical states, normally movements.

      All these correlations are to be understood as causal relations. In addition, (ii) is to be understood as an association of symbols, and can be called a 'Humean inference'.{32} He adds that these symbols in order to be functioning in an RS must have propositional form, i.e., these symbols must "represent an object and represent it as of a certain character."{33} The fact that our language is of the subject- predicate form, he implies, is an accidental feature, and does not by its presence demarcate an L from an RS. The function of predication could be accomplished by means other than using a separate predicate word: for example, by the fact that a symbol itself has a certain character. Thus it is possible to represent the fact that a car is white by writing the word 'car' in a special way -- let's say by using a particular font, or by writing it in a particular color, or by writing it in a particular size, or whatever. Sellars constructs such an alternative language -- calling it 'Jumblese' -- which dispenses with predicate words in "Naming and Saying."{34} He cautions that the idea that "the presence or absence of a subject-predicate structure marks a radical difference between linguistic and non-linguistic RSs" constitutes one of "two fundamental flaws in traditional conceptions of the mental."{35} The other flaw, which I will just mention but not discuss, is "The notion that essential to mental events is the involvement of abstract entities."{36} But in any case, the possession of sentence analogues can be ascribed to some animals: "a certain representational state of a trained rat could be said to be a 'This is a triangular' state, and hence to express the proposition that this is triangular."{37} Having described an RS so far, we should compare it with Sellars' description of a conventional language. The fullest exposition of the nature of language is in "Some Reflections on Language Games" ;{38} but more succinct presentations are given in other writings. The following is a good description:

Essential to any language are three types of pattern governed linguistic behavior.
(1) Language Entry Transitions:
The speaker responds, ceteris paribus, to objects in perceptual situations, to certain states of himself, with appropriate linguistic activity.
(2) Intra-linguistic Transitions:
The speaker's linguistic conceptual episodes tend to occur in patterns of valid inference (theoretical and practical), and tend not to occur in patterns which violate logical principles.
(3) Language Departure Transitions:
The speaker responds, ceteris paribus, to such linguistic conceptual episodes as 'I will now raise my hand' with an upward motion of the hand, etc.{39}

      Sellars is constructing here a model of a conventional language. It is clear to me that he lists the necessary conditions for a language, and it is clear that they are generically almost identical to the necessary conditions for an RS. So it is clear that these cannot be sufficient for L. But before I state the demarcating characteristics between an RS and an L, the above characterizations of L need some amplification and qualification.

      The first crucial point is that each type of transition in L is a causal relation, as it is in RS; and in the linguistic context the intra-linguistic transitions are described as being 'inferential' -- again, as they are in RS. These causal transitions are as follows. In the case of a mature language user, the language entry transition consists of being caused to respond with an observation (perception) sentence to sensorial stimuli; intra-language transitions consist of being caused to respond to the stimulus of a sentence with another sentence; while language departure transitions consist of causally responding to a sentence expressing an intention by an appropriate movement.

      The second crucial point is that in each type of transition a proposition is involved, as it is in RS. Propositions are sentences or their analogues (in some medium) functioning in the perception-inference-action roles. The concept of a proposition (which some may identify with a thought) is a concept of something that is translinguistically analogical to other languages or language-like structures (see below the chapter on Behaviorism). To talk of propositions, according to Sellars, is to talk abstractly about the role a sentence plays in a language or a language analogue. What is abstracted from is the peculiar linguistic medium or vehicle of expression. As an example, "Das ist rot" , "This is red" , and "RED" (in Jumblese) express the same proposition by playing a similar role in the observation- inference-action transitions in their respective linguistic mediums.

      The third crucial point is that in the cited passage Sellars uses the term 'transition' in contrast to 'action.' He argues that not all linguistic events are actions. Indeed such central kinds of events as the three kinds of transitions, i.e., perceptual takings, inferences, and volitions, are not actions for Sellars (and never become actions) for the simple reason that they are not the sort of thing that can be done on purpose. For example, "One can decide to look in the next room, but not to take there to be a burglar in the next room."{40} For perceptual responses are, as such, involuntary -- passions rather than actions. In the case of inferences, one can infer from seeing smoke that there is a fire, but it sounds odd to say that one decides to infer that there is fire. As to volitions,

I have emphasized that volitions are not actions but acts in the Aristotelian sense of actualities. It does not make sense to speak of willing to will to do A, anymore than it makes sense to speak of willing to feel sympathy for someone.{41}

      Sellars' view of language as composed of non-actions appears at first sight to be in basic disagreement with the later Wittgenstein's, J. Austin's, and the earlier J. Searle's views that language is composed of speech actions . Sellars is rejecting the claim that the performance of speech actions is basic. Of course he agrees that humans perform speech actions, but these, according to him, are explainable by speech non-actions. And if we take stock of G. Harman's essay "Three Levels of Meaning,"{42} as does Sellars, there may not be disagreement. There may simply be different types of interests. Sellars summarizes these three levels of interest as follows:

One group takes as its central theme the idea that language is, so to speak, the very medium in which we think, at least at the distinctively human level. Another finds its clue in the fact of communication. Still another focuses its attention on the kinship between such linguistic acts as stating and promising and a broad spectrum of social practices.{43}

      These three levels are hierarchically related: the third presupposes the second, which in turn presupposes the first. Sellars' interest is with the first; whereas Wittgenstein, Austin, and Searle, from this perspective, were interested in the third level.

      An RS is, as we see, analogous to an L. Both have entry, inference, and exit transitions. And the elements of both are propositions. Included in the intra-RS transitions are transitions in conformity to inference patterns, such as, seeing smoke causing an expectation of or looking for fire; or seeing lightning causing an expectation of thunder. RS- inferences, in short, are to be characterized by causal transitions, which in the context of RS are called 'Humean inferences'.

      Who has an RS? Some animals and pre-linguistic children. And, as Sellars implies, language users may at times operate in a distinct RS, i.e., distinct from their conventional language. This seems to be true of musicians and painters: "One is tempted to say that the musician not only thinks about sound, but also 'in sound',"{44} and

visual perception itself is not just a conceptualizing of colored objects within visual range -- a 'thinking about' colored objects in a certain context -- but, in a sense most difficult to analyze, a thinking in color about colored objects.{45}

      So far we have described a truncated form of L which is equivalent to an RS. And it is obvious from his remarks on animal behavior that Sellars does attribute to animals an ability to have an RS. And if the having of an RS is sufficient to have thought, then animals possessing an RS have thoughts. This sign use by animals is to be understood by analogy to a conventional language. But if the term 'language' is restricted to a conventional language, then, of course, nothing else will be a 'language' unless it can be sufficiently analogous to our language. But to say that requires asking the question: What else is necessary for a conventional language which is missing from an RS?

A. META-REPRESENTATION

      What amendments are needed, then, to move from RS to L? The obvious choice is to say that reflection has something to do with it, and to say that the minimal addition needed for this is the ability to have meta-representations. The position would then be that human thought is distinct from animal thought in including meta-thoughts.

      But this won't do for Sellars. We encounter passages such as the following:

Just as animal representational systems are, so to speak, minimal, so the animal's representation of its mapping activities, its meta-representations are equally primitive or minimal.{46}

and,

A child can have an ur-awareness of an ur-awareness which is adequate to play a role of self-awareness in primitive mapping activity, without being aware of it as a 'mental' activity; indeed without being aware of it as an 'inner' as contrasted with 'overt' representational activity. This contrast is a highly sophisticated one. Thus an ur-representation of ur- representation as representations should not be supposed to represent them as judgments, or inferences, as involving singular terms, predicates, or modalities.{47}

      In view of this complication, Sellars' position is not that animals and young children do not have thoughts or meta-thoughts, but that they do not have sophisticated types of thoughts or meta-thoughts, as is expressed in the following passage:

ability to represent these representational episodes with any degree of adequacy might presuppose a degree of sophistication which comes only with the mastery of language.{48}

B. MEDIATING PREMISES AND LOGICAL TERMS

      The suggestion as to what else is needed for a language is given in this passage:

the notion of a language which enables one to state matters of fact but does not permit argument, explanation, in short reason-giving, in accordance with the principles of formal logic, is a chimera.{49}

      The key difference between RS and L lies, then, for Sellars, in the kinds of inferences that are possible within these systems. We could call them respectively RS-inferences and L-inferences; instead, Sellars calls them 'Humean inferences' and 'Aristotelian inferences', respectively.{50} The difference between them seems to be that Humean inferences are unmediated by, while Aristotelian inferences are mediated by premises. An example of the former is:

'Smoke here', 'fire nearby'

An example of the latter is:

'Smoke here', 'If smoke anywhere, then fire nearby', 'fire nearby'

      The key difference, then, between RS and L seems to lie in the availability of mediating premises for L, but not for RS.

      Immediately the objection can be raised: Hasn't Sellars granted to animals the ability to entertain propositions; hence, premises? He has, indeed. So the presence or absence of mediating premises is not the key difference. The crucial difference lies in the nature of these mediating premises, and in the nature of the available propositions generally. The key is that the sentences of L contain logical and modal terms, such as: 'not', 'and', 'or', 'implies', 'equivalent', 'all', 'some', 'necessarily'; and such meta-words (i.e., a syntactical and semantical vocabulary) as: 'sentence', 'judgment', 'subject', 'predicate', 'operator', 'means', 'valid', etc. So Sellars' claim becomes minimally that a necessary condition for L is the availability of a logical vocabulary.

      I should point out that the staunchest defender of the private language thesis, Henri-Neri Castañeda, agrees with this understanding of a language:

Certainly Wittgenstein is entitled to define 'private language' as he pleases. And if on his definition a private language has no logical words, or only a sign 'E,' I agree that private languages are impossible.{51}

C. AUXILIARY POSITIONS

      In "Some Reflections on Language Games," Sellars called mediating premises 'auxiliary positions'{52}, and in "Mental Events" mentions their similarity to Quine's 'standing sentences'.{53} Let us take as an example of an auxiliary position the sentence 'All A are B'. Now from a formal point of view, from this sentence in conjunction with 'This is an A' we can deduce 'This is a B'. But the status of the sentence of the form 'All A are B' may vary from a logical or conceptual relation between the A and B, to a causal connection, to an accidental relation. To mark the distinction, Sellars introduces correlative principles of inference which are on a meta-level. These are divided by Sellars into formal and material principles of inference. Elsewhere,{54} Sellars uses the Carnapian terminology of logical rules (or L-rules), and physical rules (or P-rules). These are meant to capture, respectively, the notions of analytic and synthetic propositions. But he refrains from using the word 'analytic' because of its ambiguity. In "Is There a Synthetic A Priori?"{55} and in "Empiricism and Abstract Entities,"{56} he distinguishes a narrow and a broad definition of 'analytic' -- 'analytic-2' and 'analytic-1', respectively. A proposition is analytic-2 "if it is either a truth of logic or is logically true."{57} A proposition is analytic-1 if it is "true by virtue of the meanings of the terms involved."{58} And the meaning of descriptive terms, for Sellars, includes the specification of causal laws. Thus, Sellars' analytic-1 includes what traditionally are taken to be synthetic sentences. And for Sellars "'synthetic' will be used to mean neither logically true nor logically false."{59} Thus 'synthetic' is to contrast with 'analytic-2'; not with 'analytic-1'. Sellars has thus made possible an overlap between the synthetic and the analytic-1. And his formal-material distinction corresponds, then, to the analytic-2 -- (analytic-1 cum synthetic) distinction. Sellars himself is troubled by whether he has embraced a version of the synthetic a priori, and whether this is an offensive type of a priori. He concludes that the acceptance or rejection of a 'synthetic a priori' depends, then, on how the phrase is interpreted. And on the above interpretation, he does accept a version of it.{60}

      I have been talking indiscriminately about the mediating premises as 'positions', 'propositions', 'principles', 'rules', and I should add 'norms' to this list. The language of positions is used in the context of causal responses; the language of propositions is used in talking about linguistic function; while talk of rules, principles, and norms presupposes a metalanguage and the performance of actions, which are subject to evaluation (criticism). At the moment, I don't want to enter into these distinctions. All that is important for our present purposes is that L is distinguished from an RS by the availability in L of 'auxiliary positions'.

1. ADDITIONAL AUXILIARY POSITIONS: VALUES

      Sellars' discussion of 'auxiliary positions' in "Some Reflections on Language Games" is misleading by being restricted to a discussion of forms such as 'All A are B' and the correlative formal and material principles of inference. For as Sellars himself would acknowledge, the discussion of L should be seen from the perspective of values in general, and practical reasoning specifically.{61} Thus, talk of rules of inference presupposes the existence of such epistemic values as desire for consistency, coherence, truth, predictability, and explanation.

      Sellars talks about value statements in Science and Metaphysics{62} as occurring in "first level discourse", implying that auxiliary positions should be expanded. He regiments value statements to the form: 'Would that p'. Let us use 'would' as our valuation operator.{63} All discourse, in fact, for Sellars, implies values. And if values are seen from a hierarchical perspective, then, for Sellars, all "ground-floor valuings . . . are dominated by an over-arching ego-directed valuing expressed, in our terminology, by: Would that I led a satisfying life."{64} He calls this the principle of Rational Egoism.{65} This is a controversial thesis, and I will not discuss it. I am not concerned here with the correctness of this specific principle, or how it relates to the moral 'Would that we led a satisfying life'. All that I am concerned with is the logic of the situation, namely that 'Would that p' sentences are basic to understanding Language.

      Incidentally, I am puzzled by the fact that commentators on Sellars' theory of practical reasoning often omit even mentioning the 'Would that p' form. This is true of H-N. Castañeda, B. Aune, and W. Solomon.{66} Solomon, for example, writes: "Sellars' account of practical reasoning encompasses three main topics: (1) intentions; (2) volitions; and (3) the process of practical inference".{67} Solomon is truncating Sellars account. It actually begins with expressions of value: Would that p. And these are in the nature of wishes. If the wish is believed to be (practically) possible of realization, it becomes characterized as a desire. An intention is the desire to do the action A which it is believed will help bring about the consequence p; while a volition is the thought (the intention, so to say, whose time has come) which precedes and causes the action A.{68} But, to repeat, my main point is that Sellars' discussion of L in "Some Reflections on Language Games" should have been expanded to include value sentences as auxiliary positions.

2. ARE AUXILIARY POSITIONS OR RULES OF INFERENCE DISPENSABLE?

      The next question to ask is: Are 'auxiliary positions' dispensable? Sellars replied: "it is conceivable that a language game might dispense with auxiliary positions altogether, though at the expense of multiplying moves."{69} This means that the function of auxiliary positions will be taken over by the correlative inference rules.

      Now, are inference rules dispensable? Obviously not the formal rules -- without them there would be no logic, and hence no language. But what about the material rules of inference? Most of Sellars' earliest writings are taken up by this question. And his position is that they are indispensable to any empirical language. Since Sellars' metalinguistic concerns are based on Carnap's work, his polemics are directed at Carnap's position, as expressed in the following passage. The implication of Carnap's passage seems to be that P-rules are dispensable.

For the sake of brevity, we shall call all the logico-mathematical transformation rules of S logical or L-rules; and all the remainder, physical or P-rules. Whether in the construction of a language S we formulate only L-rules or include also P-rules, and, if so, to what extent, is not a logico-philosophical problem, but a matter of convention and hence, at most, a question of expedience. If P-rules are stated, we may frequently be placed in the position of having to alter the language; and if we go so far as to adopt all acknowledged sentences as valid, then we must be continuously expanding it. But there is no fundamental objections to this.{70}

      Carnap makes two observations. The first is that P-rules are dispensable from an object language. But Carnap is not clear whether their function is to be formulated in a metalanguage, or whether it is to remain unexpressed, or whether the function itself is translatable into an extensional language. If Carnap means to assert the last position, then it is tenable only if the thesis of extensionality itself is tenable. According to Carnap, this is the thesis that:

a universal language of science may be extensional; or, more exactly: for every given intensional language S1, an extensional language S2 may be constructed such that S1 can be translated into S2.{71}

An 'extension sentence' is a truth-functional sentence, and an 'extensional language' is one composed entirely of truth-functional sentences. An 'intensional sentence', by contrast, is not truth-functional, and an 'intensional language' is one which contains some intensional sentences. To deny the extensionality thesis all one need do is hold that there is at least one kind of intensional sentence which cannot be translated into an extensional one.

      Now, one of Sellars' characteristics as an analytic philosopher is the rejection of the extensionality thesis for various kinds of meta-linguistic sentences:

Now I do not know of any successful attempt to define the logical modalities, the causal modalities or normative expressions in purely extensional terms, and (as already indicated) I do not think it can be done.{72}

      Intensional sentences which cannot be translated include deontic sentences, semantical sentences, L-rules, and, as relevant to the present problem, P-rules. A P-rule, for Sellars, is a kind of conditional sentence which cannot be expressed either by a material implication (truth-functional) nor by Lewis's strict implication.{73} Sellars expedient is to introduce an new (arrow) symbol for an untranslatable intensional causal connector which has its own peculiar logic.{74}

      Once the extensionality thesis is denied for P-rules, there is the following ambiguity to be cleared up concerning Carnap's position. If Carnap's remarks apply only to an object language, then his position can be granted; indeed, Sellars' program calls for an extensional object language -- but provided that P-rules are in the metalanguage, which, in fact, is where Sellars locates them. Sellars position is, then, this:

Material rules are as essential to meaning (and hence to language and thought) as formal rules, contributing the architectural detail of its structure within the flying buttresses of logical form.{75}

      The other alternative for Carnap, if the dispensability of P-rules in an object language is not to entail the extensionality thesis, or the move to a metalanguage, is to opt for leaving P-rules in a language as unformalized and implicit. But such an alternative in order to become a live option must be backed-up by arguments to the effect that ordinary language defies formalization. At best, this alternative would suggest that ordinary language is too unsystematic to warrant formalization. Sellars is sympathetic to the complexity, but not to the skepticism. The philosophical endeavor must be to regiment ordinary language through the construction of models -- despite their shortcomings.

      The second observation that Carnap made is that if the convention is adopted to include P-rules, the result will be a perpetual change in the meaning of descriptive words in a language. This is a correct observation, and Sellars is quite conscious of it. In fact his theory of induction, from the linguistic perspective, is developed with the end in mind of discovering and amending P-rules; which amounts to the claim that a theory of induction is a theory of how to change language. Sellars, in fact, put it almost identically: "I regard rules of inductive inference as rules for the reasoned change of a language."{76}

      The conclusion to be drawn from all this is that though auxiliary positions are dispensable, their correlative L-rules and P-rules are not. And this marks the distinction between an L and an RS.

D. APING REASON

      But the inclusion or exclusion of auxiliary positions or rules of inference in L locates the difference but does not express it. The difference is more fundamental. The essential demarcation between RS and L lies, as has been noted, in the availability of logical and modal words in L, and their unavailability in RS.

      The obvious objection to this -- an objection which I have entertained for a long time, and which I thought was the Achilles' heel of Sellars' philosophy -- was to think that animal behavior displays an analogy to logical inference. And therefore, I thought, the difference between L and RS is drawn at the wrong place. But is it drawn at the wrong place?

      Sellars is quite aware of behavior in animals which resembles logical inference -- it is what has already been referred to by Sellars as 'Humean inference', and which B. Russell described as 'animal inference'.{77} Sellars also notes that Leibniz spoke of this as a "consecutiveness which imitates reason."{78} An example of a Humean inference, it will be recalled, was: 'Smoke here', 'fire nearby'. And it resembles the Aristotelian inference: 'Smoke here,' 'If smoke anywhere, then fire nearby', 'fire nearby'. The resemblance lies in the transition from similar initial states to similar end states, i.e., from an analogous observations of 'Smoke here' to analogous mental states of 'Fire nearby'.

      The objection may be pushed further. It can be conceded that the difference between RS and L is in the availability of the auxiliary positions. But this is an accidental distinction. For, after all, Sellars himself had admitted that such positions are dispensable. Their role can be taken over by the rules of inference.

      Are we to attribute an awareness of rules to animals? How would that be possible? Rules are, after all, formulated in a language. And if an animal has rules, they must be contained in a meta-RS.

      Yes, the objector may continue, Sellars has conceded that animals may be said to have meta-representations. So why can't we attribute to them rules expressed in a meta-representation?

      To answer this objection, we have to start by pointing out that Sellars distinguishes between rule-conforming (or rule-patterned) and rule-following (or rule-regulated) behavior. The difference between them is that rule conforming behavior does not involve the formulation of a rule as an element in conscious thought, while rule following behavior does. For example, a rat, after being trained to run a maze, has a representation of the maze in the sense that given the proper context, being in one position of the maze will trigger off a series of appropriate responses. Now contrast this with a human car driver who consults a map to guide him in his journey. The 'map' in the rat's case is not something consulted; whereas in the driver's case it is. To keep to the metaphor, the rat acts non-consciously in conformity to the path on the map; the driver acts by consciously following the path on the map.

      Inference rules are, then, rules which are available for consultation. While no rules are available for consultation to an RS organism. The reason is this. A rule is, as Sellars points out, always a rule of action of the form: 'If you want to bring about the result R, then if the circumstances are C, then you must (ought, may, may not, cannot, etc.) do action A'.{79} It is because an RS organism cannot infer through the process of obeying a rule that it cannot be said to 'infer' in the primary sense of 'inference', which is 'Aristotelian inference'. As Sellars puts it, "an intra-linguistic move is not in the full sense an inference unless the subject not only conforms to but obeys syntactic rules."{80}

      Why is a rule (or its analogue) not available to an animal? Sellars' answer is straightforward. An animal does not have the required logical vocabulary that is explicitly expressed by the rule formula, nor the logical connections (meanings) that are implied by the use of the rule formula.

      Suppose the objector responds by claiming that logical connector analogues are available to an RS organism. For example, a rat can be trained to respond to, let's say, a pink cube. And it refuses to respond to other colors and volumes. Isn't this refusal to respond an analogue of negation?

      Sellars asks this question explicitly: "How can a pre-logical RS ape the use of negation?"{81} His answer is that indeed a Humean RS organism has the resources of accepting or rejecting a proposition, but notes that we, as Aristotelian RSs, not only have the resources of accepting or rejecting a proposition, but we also have the resources of affirming or negating a proposition as well. And (our) concept of negation is not equivalent to (our) concept of rejection. The latter concept, he tells us, is more basic. Negation entails rejection, but not vice versa.{82}

      Sellars argues that from our perspective, if an RS organism rejects p, this entails (for us) that it does not represent p. But from our perspective, these are equivalent for it. Now our notion of not representing p includes representing not-p. And (for us) representing not-p entails rejecting p (which is equivalent for the RS to not representing p). But not representing p does not entail the representing of not-p. Therefore, argues Sellars, "The concept of rejection is more basic than the concept of negation."{83} This is to say that our concept of negation is a richer concept than that of rejection.

      This conclusion seems to be correct for yet another reason. Consider the fact that we can make a distinction between the negation of a sentence and the negation of the predicate of a sentence, i.e., 'It is not the case that p' and 'This a in not F'. The latter entails the former, but the former does not entail the latter. A Humean RS would fail to make the distinction by representing both cases through the rejection of the propositions.

      If it is granted to Sellars that negation is not available to an RS organism, then it also follows that none of the other truth-functional logical constants are available to an RS organism as well. This follows from the fact that truth-functions require one primitive logical constant with negation, and all other logical constants are interdefinable with the inclusion of the negation operator. We can make even a broader generalization by saying that no conceptual propositions at all are available to an RS organism since their meanings are dependent on their logical connections with other propositions. And as a corollary, no proposition which entails a truth-functional proposition is available to an RS organism. Now normally a truth-functional logic or a truncated form (which includes negation) is part of most other logics, such as modal and intuitive logics. If that is the case, then none of these dependent logics are available to an RS organism either.

      A specific case concerns Sellars' material rules of inference which are correlative to 'causal laws'. He writes, "For a lucid presentation of the fundamentals of a logic of causal modalities, see Burks . . . "{84} In Arthur Burks we are given the following axiom: "a causal implication [materially] implies a material implication."{85} The significance of this is that an RS organism, by this reasoning, lacks the concept of a causal law because it lacks the correlative material rule of inference.

      There is one last point which must be cleared up. All that has been said so far seems to be compatible with the view that human beings, in possessing a language, are able to perceive negative facts. This is not Sellars' view. All logical operators, as well as various modal and semantical words, are ultimately to be located in the metalanguage. Their role is to enable us to move from propositions to propositions. At the object level there are only positive sentences which picture reality. Thus, a form such as 'not-p' is to be understood as the denial of the truth of p.{86}

      With these observations in place, I can make the following conclusions on the difference between RS and L (using organisms). Although both use the entry to RS, intra-RS, and RS departure transitions, and although both use rs-propositions, and both have meta-representations; because primitive RSs lack a logical vocabulary, RS organisms can, at best, conform to inference rules. This has the appearance of aping reason. But only an L has a logical vocabulary which enables an L organism to follow rules -- both formal and material. And here is a crucial point for Sellars. To have words in one's vocabulary that are essential in formal and material inferences is to have concepts:

the conceptual meaning of a descriptive term is constituted by what can be inferred from it in accordance with the logical and extra-logical rules of inference of the language (conceptual frame) to which it belongs.{87}

      But if we expand on this by including such auxiliary positions as expressions of value and obligation, then the broader formulation becomes: "to express a concept is to be relevant to inferences which can be drawn from statements in which the expression occurs."{88}

IV. LANGUAGE AND CONCEPTS

      A major contention of Sellars' philosophy is that the possession of concepts requires the possession of a language. And anything that does not possess a language, does not possess concepts. Which is to say that the possession of concepts is materially equivalent to the possession of language. Hence, since animals and infants do not possess a language, they do not possess concepts.

      This seems to go counter to a widespread usage, especially by psychologists, which grants to animals and pre-linguistic children the possession of concepts -- albeit rudimentary concepts which are truncations or analogues of linguistic concepts. The following passage is typical of how psychologists use the word 'concept':

Concept formation refers to the process of discovering some characteristic that is common to a series of discrete objects, and that sets off these objects from all other objects. A concept refers to the description of this common element.{89}

      Even psychologists who write books with titles like Language and Thought use the word 'concept' in this way: "any concept is the internal representation of a certain class of experiences" ,{90} and "a person has learned a concept when he can with a high degree of reliability discriminate between instances and noninstances."{91}

      These descriptions of a concept by psychologists, as far as I can tell, do not differ from talk of 'stimulus discrimination' and 'stimulus generalization'. It seems that the statement 'S has a concept of C' is materially equivalent to 'S has generalized C' or 'S can discriminate C'. Peter Geach has made a similar observation about psychologists' use of 'concept': "they would say that an animal has acquired a concept if it has learned a discriminative response to some feature of its environment."{92} Sellars grants to animals ""discriminative behavior" such as is found in white rats",{93} but he does not grant them concepts. His use of 'concept', then, is different; in fact, restricted to language users. It is obvious that the word 'concept' is used in different ways not only by different groups of people, but often by individuals within a certain group, e.g., by philosophers. To avoid talk at cross-purposes the term has to be disambiguated. And since we are concerned with Sellars' philosophy, we must try to understand how he uses this word.

      Some people would accept the formula:

'S has a concept of x' if and only if 'S has a thought about x'

or, put otherwise,

The necessary and sufficient condition of having concepts is the possession of thoughts.

      This, with appropriate commentary, is acceptable to Sellars. But if it is conjoined with R. Chisholm's view that "Thoughts would be intentional even if there were no linguistic entities,"{94} then the distinction between RS-thoughts and L-thoughts would have to be introduced. Sellars could agree with Chisholm that there could be RS-thoughts even if there were no language.

      For some people any word of a language expresses a concept. For example, Geach states that it is "a sufficient condition for James's having the concept of so-and-so that he should have mastered the intelligent use (including the use in made-up sentences) of a word for so-and-so in some language.{95} This use of 'concept' is too broad for Sellars. His synonym for this use is 'sense' (and he mentions G. Frege as his source for this usage.{96}

      Sometimes the word 'concept' is restricted to a subset of words in a language. For Frege all grammatical predicates stand for concepts; whereas subjects name objects. At one point Sellars expressed the intention to use 'concept' in this sense: " I am strongly inclined to follow his [Frege's] lead and limit the term "concept" to predicative senses."{97} But this is not his preferred usage in most writings. In "Language of Theories," for example, he writes, "If one begins by listing a variety of types of expression which can without too much discomfort be said to express concepts -- noun expressions, predicative expressions, logical connectives, quantifiers . . ."{98}

      In "Conceptual Change," Sellars distinguishes two uses of the word 'concept':

In one of its less controversial uses, a concept is something a person has, namely a certain ability, and is always the concept of something, where 'something' is used in the broadest possible sense.{99}

      There is, however, another use of 'concept' in which one speaks of certain entities as concepts in a sense which, though acknowledging that they are items which thoughts can be "of" (and, indeed, uses this fact to pick them out), takes the relation between these entities and particular minds to be an "external one.{100}

      The second use has two interpretations.

Some philosophers . . . have held that concepts in the second sense are " objective" in that their existence is independent -- not only of this mind or that mind, but of mind uberhaupt. Others, while stressing their independence of particular minds have stressed objectivity in the sense of inter-subjectivity (compare the objectivity of institutions), and have suggested that conceptual entities are mind-dependent in this broader sense.{101}

      The first use makes a concept dependent on a particular mind. The second use makes a concept independent of a particular mind. And this can mean that concepts are independent of all minds; or that they are independent of any particular mind -- though dependent on some mind or other. The last is Sellars view.

      I think that these questions about the objectivity or subjectivity of concepts are secondary questions. The primary question is to specify the nature of concepts. This is the problem of specifying the category or 'genus' of a concept. Is it an object? a relation? a quality? a function? a disposition? a capacity? or whatnot. Once this question has been answered, then the question of objectivity can be raised.

      The essential trait of a concept is that it be a term in a sentence playing a necessary part in inferences: "to express a concept is to be relevant to inferences."{102} But this formulation is misleading in view of the distinction between Humean and Aristotelian inferences. What Sellars means to say is that a concept is expressed by a term in a sentence which is essential to an Aristotelian inference. And the necessary constituents of a language are the three types of transition patterns, auxiliary positions, and the principles of inference. Thus, for example, because the word "alas" has no (relevant) "inferential force" it is not a concept.{103}

      To answer the categorial question, to have a concept (in the primary sense) is to have an ability to use words in the appropriate inference patterns. Thus W. Robinson's attempt to capture Sellars' idea of a concept, given as: "A concept is a mind dependent ability to be in a state which corresponds to a property"{104} won't do. This definition would be compatible with what an RS organism has. But the introduction of the term 'concept' by Sellars is meant to demarcate something in human language users from what animals possess. This sentiment is well expressed by Geach in his criticism of the psychologists' use of 'concept':

What is at issue here is not just the way the term "concept" is to be used, but the desirability of comparing these achievements of rats and dogs with the performance of human beings who possess a concept of triangle; the psychologists I am criticizing want to play down the difference between human and animal performances, and I want to stress them.{105}

I think this is exactly Sellars' view.

A. CONCEPTS AND INFERENCE

      In the previous section, I have explained that a language involves the possession of a logical vocabulary and inferences. In this section I want to spell out these logical structures for particular categories.

      Let's start with a proposition p. To have a concept of p, by the requirement of formal rules of inference, we should be able to know various logical equivalences and implications; as that 'p' is materially equivalent to 'not not p', and that 'p' materially implies 'p or q'; that 'p and q' materially implies 'p', etc.

      Next, let's consider 'physical kinds', 'particulars', or 'individuals' generally. First, any particular must have a non-relation property, "it doesn't make sense to speak of individuals which stand in relations, but have no qualitative character."{106} Second, these materially imply various capacities and dispositional, properties expressible by causal and counterfactual implications. For example, the concept of a match materially implies such counterfactuals as 'If x is a match, and if x were scratched, then x would light'. And such counterfactuals depend on the truth of such causal laws as 'All matches if scratched, in circumstances C, light.' Indeed the concept of a physical kind materially implies clusters of such causal and counterfactual propositions.{107}

      Next, concerning the concept of universals or properties, Sellars makes three claims. The first is that the meaning of a universal is dependent on the type of particular which exemplifies it. Second, since the concept of a particular, as we saw above, depends on the physical laws into which it enters, Sellars, after presenting his arguments, concludes, "universals and laws are correlative, same universals, same laws, different universals, different laws."{108} Third, Sellars argues that the concept of a generic property is a disjunctive concept: "the generic is, at bottom, the disjunctive."{109}

      Basic factual predicates come in families of competing predicates, one or other of which must be satisfied by every object which can satisfy a predicate of that family. If a is not f1 it must be f2 or f3.{110}

      For example, the concept of a color is that of something being white or orange or violet or blue or red or . . . green. This means that to have the concept of something being red is to have the ability to infer that this thing is not green, not blue, not yellow, etc. -- "it is obvious that being (an expanse of) red implies not being (an expanse) of green."{111}

      [Given this notion of the sortal (generic, universal) as disjunctive, we are in a position to give a partial explication of Sellars' notorious passage:

all awareness of sorts, resemblances, facts, etc., in short, all awareness of abstract entities -- indeed, all awareness even of particulars -- is a linguistic affair.

      The assumption of this passage is that Determinables (sorts, kinds) are to be analyzed in terms of a disjunction of Determinates. The general formula would be:

(Concept)
If x is aware that y is Determinable, then x is aware that y is Determinate-1 or Determinate-2 or Determinate-3 or . . . Determinate-n.
For example, in the case of being aware that something is colored, this would translate as:

If x is aware that y is colored, then x is aware that x is white or orange or violet or blue or red or . . . green.

      What would be the analogous situation for animals? My best guess is that if Sellars is right about the absence of a logical vocabulary in animals, and animals are correctly credited with the ability to recognize generic features, then their ability must not be represented by the general formula above, but by the following formula:

(Intuition)
If x is aware that y is Determinable, then x is aware that y is Determinate-1 or x is aware that y is Determinate-2 or x is aware that y is Determinate-3 or . . . x is aware that y is Determinate-n.
Notice that I have labeled one formula 'concept' and the other 'intuition'. I have purposefully used these Kantian terms because these formulas capture, I believe, the distinction which Sellars made between Kant's manifold of representations (= intuitions) and the contrastive representation of a manifold (= concepts).{111a} I have in mind the first chapter of Sellars' book Science and Metaphysics: Variations on Kantian Themes, titled "Sensibility and Understanding."{111b} A Kantian intuition, for Sellars, is interpreted to be an RS (propositional) representation. Let us call it 'rs'. We can then formulate the distinction the following way:
(manifold of representations)
If x is aware of a manifold of rs's, then x is aware that rs-1 or x is aware that rs-2 or x is aware that rs-3 or . . . x is aware that rs-n.
(representation of a manifold)
If x has a representation of a manifold, then x is aware that rs-1 or rs-2 or rs-3 or . . . rs-n.
There are two other clarifications of Sellars' position which this analysis makes possible. The first is that it explains the distinction which Sellars makes between conceptual and non-conceptual representations. The second is that it sheds some light on Sellars' claim that non-conceptual representations guide conceptual representations. (AC 1996)]

V. HOLISM

      Furthermore, since the units of inference are propositions, it follows that concepts are imbedded in various propositions. And if we try to think through the extent of conceptual interconnections which the characterization of propositions, particulars, and properties entail, we will come to the Sellarsian conclusion that to have one concept implies having many; indeed, it implies having a language game. This is Sellars holism about concepts. So, "'conceptual structure' in this sense refers to language games."{112} On the one hand this holism is antithetical to the logical atomism view espoused in Wittgenstein's Tractatus {113} that "2.061 States of affairs are independent of one another.
2.062 From the existence or non-existence of one state of affairs it is impossible to infer the existence or non-existence of another." On the other hand, Sellars does not mean to espouse a Hegelian view that the meaning of a concept lies in the Absolute.{114} The clue to Sellars' holism is provided by the Wittgensteinian phrase "language game." In this sense, each person could, first, have many languages; for example, a Manifest Image and the Scientific Image. And as H. Brown points out in his commentary on Sellars' holism, "we can view the individual who learns a second (or third, etc.) science as learning a new language."{115} Second, Sellars speaks about degrees of mastery of a language. The implication of this is that it is possible to know a language in disjoined fragments; thus, denying any implications for absolute or total holism.

      In conclusion, the point of this section has been to disambiguate the concept of a 'concept', and to delineate Sellars use of it. His use is that a concept is any word in a sentence which is relevant to inferences. And I pointed out the kinds of inferences that Sellars has in mind. The upshot is that the possession of concepts is correlative with the possession of a language. And the implication of such a requirement of inferential connections needed for the possession of concepts leads to a holism about language, but, as I suggested, it need not be as extensive as the Absolute.


[Go to Chapter 6]