Correspondence between Wilfrid Sellars and Ausonio Marras*

Edited in hypertext by Andrew Chrucky

Editor's note: The first letter (Jan. 3, 1972) is Sellars' initial comments on the proofs of Marras' article, "On Sellars' Linguistic Theory of Conceptual Activity," Canadian Journal of Philosophy 2 (1972-1973): 471-83. This article was published together with Sellars' reply, "Reply to Marras,": 485-93, and Marras' rejoinder, "Reply to Sellars,": 495-50.

The second letter (Nov. 25, 1975) is Sellars' commentary on a draft of Marras' "The Behaviourist Foundation of Sellars' Semantics" Dialogue 16 (1977): 664-675. In a footnote to his published version, Marras wrote: "An earlier draft of this paper was read at the Pacific Division meetings of the American Philosophical Association held at Berkeley, California, on May 27-29, 1976. I am grateful to Wilfrid Sellars and R.W. Binkeley, who provided extensive and valuable comments to the earlier draft in private correspondence. While certain alterations resulted from their comments, I doubt that I have succeeded in securing their agreement on all points."

The third letter (June 15, 1976) is Marras' reply to this second letter.

For ease of reference to the hypertext version of EPM, I have inserted in brackets next to the page numbers of EPM (in SPR), the appropriate passage numbering. I have also inserted in the second letter in square brackets the corresponding page numbers of Marras' published paper.

Dot-quotes are rendered by bold periods.

Jan. 3, 1972: Sellars to Marras
Nov. 26, 1975: Sellars to Marras
June 15, 1976: Marras to Sellars

Faculty of Arts and Sciences
University of Pittsburgh
Pittsburgh, Pennsylvania 15213

Department of Philosophy

January 3, 1972

Professor Ausonio Marras
Philosophy Department
University of Western Ontario
London, Ontario

Dear Marras:

I have been reading your really excellent paper on my Jonesean myth approach to inner conceptual episodes. The CJP [Canadian Journal of Philosophy] has offered me an opportunity to reply, and I am planning to take it, in spite of the fact that it is logically impossible for me to do all the things I am committed to do during these next few months.

The charge of circularity you make has, as you might expect, been made before, but never so lucidly and, from my point of view, so usefully. It may also not surprise you that in my reply I shall draw a distinction between different levels of 'having a conceptual ability,' ranging from knowing how to make the appropriate linguistic moves and transitions with respect to sentences involving the expression which 'stands for' a given concept, to knowing how to make moves and transitions in language games containing relevant rule sentences. In your last footnote you quote me as referring to "full-fledged" members of the linguistic community. The term 'full-fledged' implies a contrast with fledglings, and I have certainly never intended to deny that at a certain stage fledglings have conceptual abilities. I have, to be sure, insisted that merely to respond to red things with 'red' is not to have a conceptual ability; but this leaves lots of scope for rudimentary conceptual abilities which do not involve obeying ought-to-do's. I shall try to spell this out in my reply, but in the meantime you might look at the key paper in the whole series: "Some Reflections on Language Games." You'll notice a general coincidence of aim in that paper with that of the more recent paper to which you direct your attention, though the earlier paper limits itself to conceptual activity of the Rylean type.

The key concept is that of pattern-governed behavior, and the key claim is that all linguistic activity is essentially pattern-governed behavior -- whether OL [object language], ML [meta-language] or MML [meta-meta-language] ... In other words, rule obeying activity is a special case of pattern-governed behavior, where the 'patterns' involve the expressions of the rule in their aspects of language entry transitions, intra-linguistic moves and language-departure transitions. (See particularly sections 15-18; 22-3; 36; 53 ff.) From this point of view, the role of linguistic oughts, which belong to the repetoire of full-fledged language users, is to maintain or improve pattern-governed linguistic activity at the level to which they apply, whether this activity is by full-fledged language users or by neophytes.

Please remember that, on my account, all concepts pertaining to action have their primary (in the order of conceptual) locus in the Rylean framework. I am afraid that in your last footnote you equate 'psychological' with 'mentalistic.' Jones, of course, has (as I see it) a full battery of psychological concepts pertaining to 'intending' and related states, as well as to 'thinking-that.' His concepts, however, are Ryle-psychological.

But I shall try to spell all this out in my reply. Meantime, let me express once more my pleasure in reading your clear and challenging paper, and my gratitude for having the opportunity to clarify what I must admit to be at least obscurities in my formulations, if not in my thought.

Sincerely yours,

Wilfrid Sellars



University of Pittsburgh

Department of Philosophy

November 26, 1975

Professor Ausonio Marras
Philosophy Department
University of Western Ontario
London, Canada

Dear Ausonio:

I am grateful to you for the opportunity to comment on your paper on the "behaviorist 'foundation' " of my semantics. I do indeed "find it of interest," for the central question you raise does point to a weakness of the argument of EPM. How serious this weakness is and whether it can be remedied without destroying the main thrust of the paper are the matters I want to take up in this letter.

Perhaps the most useful point with which to begin concerns what I had in mind when I spoke of the Rylean language, the language of the pre-Jonesean community. The passage in which it was introduced described it as "a language of which the fundamental descriptive vocabulary speaks of public properties of public objects located in space and enduring through time." (p. 178 [#48]; I have added the italics.) It also, of course, contains "the vocabulary or formal logic." (p. 179 [#49].) What else does it contain? Of course it does not contain "the fundamental resources of semantical discourse." (p. 179 [#49].) These are exactly that which, according to the argument, has to be added "in order that these talking animals may come to recognize themselves as animals that think ..." (p. 179 [#49].) By 'think', of course, in this passage, I meant 'in the inner episode sense.' As for 'methodology,' your statement that "theoretical thinking is as old as thinking itself" (p. 12 [674]) is of a piece with my statement that "science is continuous with common sense, and the ways in which the scientist seeks to explain empirical phenomena are refinements of the ways in which plain men, however crudely and schematically, have attempted to understand their environment and their fellow man since the dawn or intelligence." (p. 183 [end of #51].)

Now, you are tempted to construe my Rylean language as "the rough equivalent of a Carnapian physicalist 'thing-language.' "(p. 5 [667].) But this, though not completely wrong-headed, misses one essential point. Carnap's thing-language contains only a descriptive and a logical vocabulary. My Rylean language, however, as introduced in the passage quoted above could perfectly well contain the prescriptive vocabulary of practical reasoning (and other passages in EPM imply that it does -- see Section VIII, passim). What my Rylese and Carnap's thing-language have in common is a rejection of the vocabulary of 'inner episodes' in the relevant sense. And, indeed, when, in the passage you quote from Science and Metaphysics I wrote that "dot-quote common nouns have a purely Rylean sense" (p. 161), the previous discussions made it clear that the sense of these common nouns involves a reference to linguistic rules, just as the sense of chess common nouns involves a reference to the rules of chess. As I pointed out at the end of "Language as Thought and as Communication," the chief flaw of EPM was, as I saw (and still see) it my failure to be clear about the exact nature of 'means' statements. I was groping for a precise account. My intuitions, however, were on the right track, and I can even read my earlier formulations without too keen a sense of embarrassment. When I agreed, in the Chisholm Correspondence that "... one additional expression must be taken as primitive, specifically 'means' or 'designates,' " I emphasized that "although 'means' is in a grammatical sense a 'relation word,' it is no more to be assimilated to words for descriptive relations than is 'ought,' " and that "although it is a 'descriptive' predicate if one means by 'descriptive' that it is not a logical term, it is not in any more interesting (or usual) sense a descriptive term," (pp. 523-4). The view which I subsequently expressed by saying that 'means' is a specialized form of the copula was almost on the tip of my tongue. In the passage I have just quoted, I was using the expression 'logical term' to refer to quantifiers, connectives and the symbol for identity. I was rejecting such views as, for example, that

x means y (in G) =df
x = 'rot' and y = red or
x = 'gelb' and y = yellow or
x = 'und' and y = and or
(See SPR, p. 201; "Empiricism and Abstract Entities," p. 463). I was certainly not as clear then as, hopefully, I am now about the status of the copula, but even then I would most certainly have put it in a different box from the logical terms listed above.

On p. 3 [666] you ask "How, exactly, is talk about meaning related to its behavioral foundation? ... Is it a factual relation?" Now it is obviously a matter of fact that

'rot' (in G) means red
'rot's (in G) are .red.s

This, of course, is not at issue. The question is rather, in first approximation, Is it a matter of fact that if occurrences of 'rot' (in G) exhibit such and such relevant{1} behavioral uniformities, then 'rot' (in G) means red? You suggest that on my view "Rylean facts serve as empirical criteria for the truth of semantical statements." (p. 4 [666].) I do not object to this. I do, however, find puzzling your characterization of "the ... relation in which criteria stand to what they are criteria for" as an " 'empirical bridge.' " (p. 4 [666].) Now the criteria are, indeed empirical, but is the relation between a criterion and that for which it is a criterion empirical? The question deserves more attention than you give it, for it is central to your argument.

I have argued that, at a level which abstracts from specific linguistic materials and linguistic structures, Rylean uniformities pertaining to the expressions quoted on the left hand side of a 'means' statement must be the same as the uniformities which pertain to the expressions on the right.{2} You argue that if (which I would not grant) "the relation between semantical statements and the relevant Rylean facts is a factual, non-analytic relation," it can not be construed as "a purely contingent one." (p. 4 [666].) Rejecting my solution, which takes the concept of a criterion seriously, you propose an alternative:

The alternative ... is to suppose that the empirical generalizations (italics mine, WS) which obtain between (true) semantical statements and Rylean facts stand in need of explanation, and to provide such an explanation is at the same time to provide the required conceptual link between the Rylean framework and the framework of semantical concepts. (p. 4 [667].)

I am not yet clear as to exactly what would count as an example of the sort of 'empirical generalization' you have in mind. You turn aside to take another look at my relations with "logical behaviorism." It is here that the passage in which you misassimilate the Rylean framework to the "Carnapian physicalist 'thing-language' " occurs. You then go on to write, correctly, that

... the point of the myth is to show how the addition to the Rylean language or the resources of semantical discourse and of theoretical discourse suffices to explain the acquisition of mentalistic discourse.

The reference to theoretical discourse is pertinent, but not of the essence, for I could have made an explicit commitment to the idea that pre-Joneseans were able to give theoretical explanations of purely physical phenomena, Indeed, I would have done so had I been on my toes, for the idea is implicit in the way I set up the problem faced by Jones. (See,particularly, the top of p. 181 [#51] of SPR.) What is of the essence, however, is the claim you proceed to make that

What the myth does not explain, however, is the acquisition of semantical discourse itself. Sellars does briefly address himself to this point, but only to evade it.

Now, it is indeed true that I raise the question "... how could the addition of these resources be construed as reasonable? where the resources are "in the first place the fundamental resources of semantical discourse," (SPR, p. 179 [#49]) and, having raised it, give no answer. But, I was probably wise not to make the attempt, though the question is of great intrinsic interest and importance. For my attention remained focused on the question whether discourse about meaning was to be analyzed in mentalistic terms, and the basic flaw in my argument was not the failure to answer the above question, but in the inadequate character of my account of the nature of the discourse about meaning.

I certainly agree that the question concerning "the acquisition of semantical discourse itself" is, as you put it, "... crucial if we are concerned with clarifying the relations of [my] semantics to its Rylean foundation." (p. 6 [668].) You rightly add that

Since the acquisition of semantical discourse by Sellars' Rylean ancestors is surely not to be viewed as a gift from the gods, but must, instead, be regarded as a natural development from within the primitive Rylean language itself, we must suppose that the conceptual and methodological resources of the primitive language must have been rich enough to permit this development. (p. 6 [668].)

This is an excellent point, and I must admit that I did not do it justice in EPM, for the simple reason that I was not yet able to cash out my intuitions concerning 'means.' But since you do not limit your paper to the 1956 Sellars' slice, but are evaluating the approach to semantics which EPM was just getting underway, it is surely fair play on my part to draw, as you yourself do, on later formulations to defend my Jonesean myth.

Thus, let us take into account the fact that Jones, from the beginning, is in a framework which includes the language of ought-to-be and ought-to-do. What he needs to do is not to invent this dimension of discourse, but, at most, to apply it to linguistic.behavior. I say "at most," because, while I certainly did not explicitly state in EPM after Jones was introduced that he was in a framework which recognized ought-to-be's and ought-to-do's with respect to linguistic behavior, the ideas which found their most adequate expression in "Language as Thought and as Communication" were already present in Section VIII -- see p. 166, paragraph 2 from the bottom [#33]; also the paragraph beginning on the bottom of p. 167 [#35] -- where the stage was being set for his appearance. I would have had no more qualms about attributing such a framework to Jones than about granting that he was already familiar with theoretical explanation in non-mentalistic contexts.

But to be in the framework of linguistic ought-to-be's and ought-to-do's is not ipso facto to be in the framework of meaning. What needs to be added? Not much, given the resources at hand. Our Ryleans are presumably able to classify a wide variety of things and events. They also play chess, and use sortals, like 'pawn,' the senses of which are bound up with rules. Thus, as far as semantics is concerned, Jones' chief invention is the introduction of what I have called 'illustrating quotes' to form sortals which classify linguistic expressions in terms of their function in the language to which they belong. Add to this the concept of a distributive singular term and its use to explicate the status of such ostensible 'names' as 'triangularity,' and Jones has the resources to regiment

'dreieckig' (in G) means triangular


'dreieckig's (in G) are .triangular.s


'dreieckig' (in G) stands for triangularity


'dreieckig' is the German .triangular.


'dreieckig's are German .triangular.s

and the Jonesean semantics is in full swing. (The interpretation of truth in terms of de-quoting, and the recursive specification of classes of truth conditions is too long a story to sketch here.)

I hope you find these rather hasty comments a meaningful part of our continuing dialogue. I certainly found your paper both perceptive and helpful.


Wilfrid Sellars


Department of Philosophy
The University of Western Ontario
London, Ontario N6A 3K7

June 15, 1976

Professor Wilfrid Sellars
Department of Philosophy
University of Pittsburgh
Pittsburgh, Pennsylvania

Dear Wilfrid:

I have, at last, been able to get back to your reply to my paper on "The Behaviorist Foundations of Sellars' Semantics". Needless to say, your comments were, as usual, lucid and penetrating, and I think I now see more clearly where the argument in my paper needs buttressing. Let me, then, directly turn to this; later I shall take up a few (I hope) minor points.

The central problem I raised in my paper concerned the Ryleans' acquisition of semantical discourse: to the extent that they lack the methodological resources of theoretical (postulational) discourse, I claimed, they are committed either to a logical behaviorist or to an instrumentalist account of linguistic meaning. On the assumption that either of these alternatives would be (even for you) unacceptable, I proposed as a solution that the Ryleans be allowed to engage in theoretical reasoning and thus be allowed to introduce semantical concepts as theoretical concepts, in agreement with the requirements of a "thoroughgoing" methodological behaviorism (in the sense of EPM, 185 [#54-55]).

Your objection, as I understand it, is essentially as follows. The problem I have raised only arises because my characterization of the Rylean language is inadequate in that it mentions only its descriptive aspect and fails to take account of its equally important prescriptive aspect. Although this aspect is not explicitly mentioned in the section of EPM where the Rylean language is introduced, there are other sections in EPM as well as in other writings in which it is clear that you wish to ascribe to the Ryleans the ability to engage in practical reasoning. The Ryleans, in other words, operate all along in a framework of rules (e.g. they are able to play chess and classify the pieces in terms of the rules which govern their use). Thus all that the Ryleans had to do in order to acquire the resources of semantical discourse was simply to apply this framework of rules and practical reasoning to linguistic behavior. As soon as we allow the Ryleans the ability to classify things and events not only in terms of their descriptive characteristics but also in terms of their rule-governed functions, all that remains to be done as far as semantics is concerned is "the introduction of ... 'illustrating quotes' to form sortals [dot-quote common nouns] which classify linguistic expressions in terms of their function in the language to which they belong." The acquisition of semantical discourse by the Ryleans should not therefore appear mysterious.

In order to respond to your objection it may be helpful to distinguish two aspects which are involved in the Ryleans' acquisition of semantical discourse: first, the recognition that certain sounds and noises emitted by people in determinate circumstances are meaningful, or, as you say, "non-parroting"; second, the articulation of a (classificatory) system which specifies a meaning or function for each distinguishable set of sounds and noises. Now while to specify the meaning or function of an utterance (by classifying it with respect to the rules which govern its function) indeed requires essentially the use of prescriptive language, it does not follow that the claim that some utterances are meaningful, non-parroting, rule-governed and thus essentially classifiable with respect to prescriptive criteria, is itself a prescriptive claim. To make such a claim is instead to make a "descriptive" (or, more precisely, as I shall presently suggest, a theoretical) claim: it is to claim that certain configurations of sounds and noises, unlike other configurations, enter into patterns of (intelligent) behavior not by virtue of their overt, observable characteristics but by virtue of their resulting from the application of certain (prescriptive) criteria. (Recall your claim in "Notes on Intentionality": meaningful overt speech is to be 'understood' in terms of uniformities which result from "the causal efficacy of rule-expressions".)

Now if you protest that I am running together two distinct things, namely (to put it in a different terminology), an account of linguistic competence (the 'grammar' which specifies the rules of a language; semantics proper) and a theory of linguistic behavior, in a sense you are right. While I distinguish the two, I am urging that both are involved in the main question at issue: How is the Ryleans' acquisition of semantical discourse to be explained? Although the Ryleans were, of course, speaking meaningfully and following rules all along, prior to the acquisition of semantical discourse they did not know they were doing so, for by hypothesis they did not yet have the concept of linguisic meaning and linguistic rule. How did they come to know that their linguistic behavior was rule-governed; how did they come upon the concepts of linguistic meaning and linguistic rule? They did not just "dream them up": instead, I suggest, they introduced them (perhaps on the analogy of the functions of chess pieces and the rules of chess) in the attempt to provide (better) explanations of their behavior.

Thus, I am suggesting, to characterize a linguistic episode in semantical terms is, at least in part, to relate it to an explanatory framework. And this, as I see it, is in accordance with the whole thrust of the Jonesean myth. What was the theoretical motivation for postulating thoughts? It was to explain the remarkable fact that people "behave intelligently not only when their conduct is threaded on a string of overt verbal episodes ... but also when no detectable verbal output is present" (EPM, 186 [#56]). Clearly, mental episodes are postulated, at least in the first instance, in order to extend whatever explanatory function verbal episodes were already recognized to have; and they have that function by virtue of their semantical, not by virtue of their observable, characteristics. (For the explanatory purpose of Jones' theory it is only such semantical characteristics that are analogically extended to mental episodes.)

Thus, without wishing to minimize the prescriptive aspect of semantical discourse, we must nevertheless admit that in the context of the Jonesean myth the ascription of semantical characteristics to verbal episodes must have performed an explanatory and thus a "descriptive" (non-prescriptive) function. If so, the problem I raised in my paper still remains: given the primitive, non-theoretical resources of the Ryleans, the realization in the course of their conceptual evolution that certain sounds and noises are meaningful or non-parroting must have amounted merely to the realization that they exemplify certain behavioral regularities (i.e. rule-governed patterns), or, alternatively, to the realization that such and such inferences are licensed by their occurrence. It goes without saying that the explanatory value of these 'realizations' is as limited as is the explanatory power of logical behaviorism and instrumentalism themselves.

Seen in the context of the foregoing discussion, your recognition that "theoretical thinking is as old as thinking itself" gives me comfort; for now we can remove the last impediment to a non-logical-behaviorist and non-instrumentalist account of the Ryleans' acquisition of semantical discourse, precisely in the way I suggested: let the Ryleans be (proto) theoreticians and they will be in a position to introduce semantical concepts as theoretical concepts. These concepts, while classifiable with reference to rules and prescriptive criteria, will enter into theoretical explanations of people's behavior and will thus explain the 'Rylean behavioral uniformities' exemplified by that behavior. Those uniformities will be explained by being deducible from a theory, consisting of a complex set of statements, some of which will be statements about rules, some about contextual conditions, some about internal (cognitive) states of the speakers, and some will be 'bridge laws' relating some of those statements to statements about observable behavior.

Given this theoretical link between semantical statements and Rylean uniformities, we can also understand how the latter can serve as criteria for the former. When in my paper I construed this criterial relation as an empirical relation, I did not wish to imply that this relation is purely contingent: it is indeed a necessary relation, but one which has a theoretical, and hence factual, grounding. In other words, I only wished to deny that the relation is analytical, as the logical behaviorists have tended to construe it. (In Kant's terminology, the criterial relation might be construed as a synthetic a priori relation.)

In response to the claim that the concept of linguistic meaning is a theoretical concept one might wonder: If theoretical concepts require a model for their introduction, what might the model for linguistic meaning be? One might suggest the following two possibilities: 1) Linguistic meaning is analogous to the "meaning" or function of chess pieces, and to ascribe meaning to an utterance is to theoretically ascribe properties to it analogous to the properties that make a piece of behavior a move in a game. 2) Linguistic meaning is an inner (unobservable) mediating response to a linguistic stimulus, understood on the model of the overt response to a thing or situation 'signified' by the linguistic stimulus. I don't know which of these two views would be more promising from the standpoint of theoretical psychology; and I don't know that they are really incompatible. The latter view, which is represented by mediation-theorists such as Osgood, would be compatible with the theoretical approach you refer to as 'methodological behaviorism'.

One final point. It may seem that in alluding to the behavioral 'foundation' of your semantics I betray a misunderstanding of your whole enterprise: after all, who more than yourself could be opposed to the foundationalist approach to knowledge and concept formation? Perhaps so; but let me remind you that your introduction of the framework of thoughts in EPM was in some relevant sense foundationalist: as you put it in your Synthese reply to Putnam's and Dennett's comments on your paper "Meaning as Functional Classification", you construed the framework of thoughts as "a 'theoretical' framework related to the 'observation' framework of overt linguistic behavior (somewhat as microphysical theory is related to the macro-framework of perceptible things)". Taking a hint from this methodological approach, my suggestion has been that the very framework of 'overt linguistic behavior', i.e., the framework of semantically characterizable items, is itself a theoretical framework related to the observable framework of Rylean behavioral episodes. (Note that while sounds and noises are indeed observable, their meaning is not.)

As I re-read this letter, I fear I have not been able to make my points as clear as I had hoped. While I feel confident that my approach is in the right direction, I am still groping towards a clear statement of all the issues involved. Needless to say, any further light you might be able to shed would be most welcome. I might add that I am looking forward with great anticipation to the 'workshop on Sellars' to be held at VPI [Virginia Polytechnic Institute] in November, to which I have been invited to participate.

Let me thank you again for your helpful comments, and excuse the delay with which I have responded.



Ausonio Marras

[Return to Main Page][Contents]


{*} I wish to thank Professor Marras for supplying me with the first and third letter, and for permitting me to publish his letter to Sellars.[Back]

{1} See the next footnote.[Back]

{2} I omit considerations about degrees of meaning. I also avoid far more difficult issues concerning the filtering of 'raw' statistical linguistic data through a framework which distinguishes between those uniformities of various kinds and levels which constitute the conceptual resources of a language, and those which pertain to the employment of these resources. The possibility of drawing such a distinction with respect to the Heracleitean flux which is a natural language is certainly a $64,000 question for empirical linguistics. If, however, we are able to conceive of a 'tidy' language with respect to which these distinctions can be clearly drawn, the issue between us can at least get off the ground.[Back]

[Rreturn to Main Page]