Correspondence between Wilfrid Sellars and Bruce Aune

Edited in hypertext by Andrew Chrucky: Revised and expanded Jan. 16, 1997.

The correspondence prior to 1975 was contributed by Professor Bruce Aune, and the 1975-1979 correspondence was contributed by Professor Willem deVries. It is reproduced here with the permission of Professor Aune.

Editor's Note: Due to the limitation of current hypertext, the following expressions have been substituted for the original ones. All Greek characters are rendered by their names: thus, 'alpha', 'beta', 'gamma'. The square of a number is expressed as follows: x^2.

Logical connectors and quantifiers are all expressed by bold characters as follows:

--> = material implication
<--> = material equivalence
& =conjunction
~ = negation
v =disjunction
(x) = universal quantifier
(Ex) =particular ('existential') quantifier


October 19, 1961: Sellars to Aune
October 23, 1961: Aune to Sellars
November 11, 1961: Sellars to Aune
Fall 1964: Aune to Sellars
October 3, 1969: Sellars to Aune
January 28, 1970: Sellars to Aune
September 3, 1970: Aune to Sellars
June 23, 1975: Sellars to Aune
June 9, 1979: Aune to Sellars
July 3, 1979: Sellars to Aune


October 19, 1961

Dear Bruce

I have enjoyed reading your Free Will paper. It is very well written, and I am in close sympathy with its general drift. I am a bit uneasy about some features of the argument, however, and feel that you may be misinterpreting what Taylor says -- although on the central issue you are right and he is wrong.

Classically it was customary to distinguish between 'absolute' and 'hypothetical' and 'relative' modalities. (The terminology differed, but the distinctions, by and large, did not.) Thus p v ~p is absolutely necessary, while q is necessary on the hypothesis p and (p --> q). In the case of the physical or natural modalities, if lightning then thunder is absolutely necessary, but thunder is necessary on the hypothesis of lightning. As for the relative modalities, q will be said to be necessary relative to p, if q is necessary on the hypothesis that p and it is implied that the hypothesis is true. Let us use 'NP[p]' to say that p is necessary in the absolute physical sense, 'MP[p]' to say that p is possible in the absolute physical sense, 'NP[q]/p' to say that q is necessary (in the physical sense) on the hypothesis that p, and, correspondingly, 'MP[q]/p' for the case of hypothetical physical possibility. Finally let us express relative physical necessity and possibility by adding an exclamation mark to the two last forms, thus: 'NP[q]/p!' and 'MP[q]/p!'

You are of course right that from 'p and NP[if p then q]' we cannot conclude 'NP[q].' We can, however, conclude 'NP[q]/p!' And if we represent '(Ep) NP[q]/p!' by 'NPR[q]' -- Anglice use 'q is relatively physically necessary' for 'There is a p such that p is the case and if p then q is physically necessary in the basic or absolute sense' -- we can move from 'NP[q]/p!' to 'NPR[q].'

So much for basic modal notions. To make the points about 'ability to do something' which I want to make, I must introduce some more terminology the most important of which is the notion of a 'minimal action'. Abstracting from mental actions or doings, a minimal action can be roughly characterized as a bodily change which is under our voluntary control and is not brought about by bringing about some other change which is under our voluntary control. Roughly it is an action which does not have a 'smaller' action as its initial segment.

Let 'Am(xt)' mean x does minimal action Am at t

Let 'Vam(x,t) mean x wills at t to do Am

Let ' t' ' mean the time just before t

Let 'CAN[Am(x,t)]' mean x is able to do Am at t


CAN[Am(x,t)] = NP[Vam(x,t) --> Am(x,t)] = NP[Am(x,t)]/Vam(x,t)

Notice that




but not

(K) MP[Am(x,t)]/K(x,t')
where 'K' is a variable ranging over states of affairs

though, of course it does entail


We can now introduce a more inclusive concept of action and a corresponding concept of ability to do.

Actions other than minimal, and circumstances of minimal actions, are represented by 'A' and 'C', respectively, and introduced in terms (roughly) of the schemas:

A(x,t) = R(x,t) = (EC)(EAm) C(x,t) & Am(x,t) & NP[C(x,t) & Am(x,t) --> R]

where 'R' stands for the "result" and 'R(x,t)' has the sense of 'x at t' brings about R'.

(It will be noticed that I have attempted to tie the notion of 'circumstance' to that of action in such a way that circumstances are what combine with minimal actions to bring about the existence of more inclusive actions. In the sense of circumstance I am attempting to characterize, being paralysed would not be a circumstance. But this is not essential to the points I want to make in this letter.)

We can now define 'x is able to do A' as follows:

CAN[A(x,t)] = (EAm)(EC) CAN[Am(x,t)] & C(x,t) & NP[C(x,t) & Am(x,t) --> A(x,t)]

The relation of determinism to ability can now be formulated as follows:

If determinism is true, then 'x did not do A at t' i.e. '~A(x,t)' entails

(EK) K(x,t') & NP[K(x,t') --> ~A(x,t)]

Consider the case of minimal actions. From

K(x,t') & NP[K(x,t') --> ~Am(x,t)]

it would, of course, be incorrect to infer


but quite correct to infer not only

~MP[K(x,t') & Am(x,t)]



and even


which, though it asserts the relative impossibility of a doing, turns out, when unpacked, to assert the impossibility of a complex state of affairs rather than that of a doing, as you well point out.

Now the crux of the matter is that


doesn't entail


The latter has, in accordance with our definitions, the sense of

~NP[Vam(x,t) --> Am(x,t)]

Thus, although 'It was (relatively) P-impossible that x do Am' and 'It was impossible for x to do Am', i.e.

'x was unable to do A'

look deceptively alike, the former has the form

[EF] F(x,t') & NP[F(x,t') --> ~Am(x,t)]

whereas the latter has the form

~NP[F(x,t') --> Am(x,t)]

The above applies to minimal actions. The extension of the point to non-minimal actions is straightforward. As in the case of minimal actions, '~MPR[A(x,t)' is compatible with 'CAN[A(x,t)]'. The latter is falsified by any one of the following:

~(EAm)(EC) CAN[Am(x,t) & C(x,t) & NP[Am(x,t) --> A(x,t)]

I.e. there was no combination of a minimal action which x was able to do together with a circumstance which obtained which would bring about a doing of A

(EAm)(EC) CAN[Am(x,t) & NP[Am(x,t) --> A(x,t)] & ~C(x,t)

I.e.There was a minimal action which x was able to do at t which would result in a doing of A if the circumstances had been of a certain kind which they weren't

(EAm)(EC) C(x,t) & NPtAm(x,t) --> A(x,t)] & ~CAN[Am(x,t)]

I.e. x was in a circumstance which if combined with a certain kind of minimal action would have brought about a doing of A, but x was not able to do the minimal action at the time in question.

Well, Bruce, I hope the above is not too incomprehensible, and if comprehensible a worthwhile comment on your stimulating paper. You will find a hint of these views in a footnote to the phenomenalism paper.

As ever,


[Return to Table of Contents]

October 23, 1961

Dear Wilfrid,

Your letter came this morning, and it pleased me very much to get your comments and to hear that you found my paper interesting, You were the first, in fact the only, one to comment seriously on it -- and for this, as well as for excellence of your comments, I am deeply grateful.

Feigl was here last week to give a series of lectures, and we talked about the paper a little. He found it interesting -- and my conclusions a little surprising -- but he didn't have time to think the argument through, and hence didn't have much to say by way of comment on it. He did mention, however, that he thought it was worth publishing, and he said that if I could get it down to the equivalent of 16 pages of Phil. Studies, he would recommend it for publication in your journal. I have revised it since he left, however, and now I feel that it is too long. At any rate, in order to free myself from constant revision, I sent the revised version off to Phil. Rev. I sent it there because they make their decisions very promptly, and because Mind and Phil. Quarterly have papers of mine they have yet to print. (Perhaps sending it was premature, but then, thanks to your letter, I have additional ammunition for making replies.)

But now for some comments on your comments, and then some comments on your phenomenalism paper, First, I liked your idea of minimal actions. Actually, I had something like this in the back of my mind, but I didn't know quite how to put it, That was why I said so much for trying. I know it sounds awkward to speak of trying to move one's own finger, say; but I don't think it is nonsense. A paralyzed child may try to move his legs but fail; though we don't normally speak of trying if he succeeds. But because there are cases where one can try to move a muscle, an analysis of trying -- a just analysis, that is -- must wrestle with the problems of thought and action, and this means handling the notion of minimal actions, both mental and physical. Because these problems are bound up with the now hotly debated notion of "the will," I tried to circumvent the whole tangle by fixing on the admittedly loaded concept of trying. Once this concept is unpacked, however, something like your view of thought and action (as expressed, e.g., in your Language Games paper) will have to emerge, along with, of course, your comment on minimal actions.

I was very interested in your analysis of hypothetical and relative necessity, which was quite new to me. It is possible that Taylor had something like this in mind, though I am sure that he was too confused about physical necessity to envisage anything like the complex analysis`you present. In fact I was moved to write about physical necessity as I did because I thought he, like many contemporary Oxbridgians, was led to plump for an anti-scientific view of man because, among other things, he was confused about the implication of the physical modalities.

Your own suggestions, especially about the logic of "x is able to do A," struck me as extremely thorough-going, making me realize that my analysis scarcely scratched the surface of a highly complex matter. I found one difficulty in your account, however. Your definition of "x is able to do A," namely

CAN[A(x,t)] = (EAm)(EC): CAN[Am(x,t)] & C(x,t) & NP[Am(x,t) --> A(x,t)],

seems to imply that one has a given ability only in favorable circumstances. But doesn't it make sense to say "I have the ability to read French novels, but because I am not in favorable circumstances, no such novels being in my office, I cannot presently exercise this ability.

If, in accordance with this difficulty, you delete the "C(x,t)" as a main conjunct in your definition (and put it, perhaps, in the antecedent of "NP(Am(x,t) --> A(x,t))," since it cannot be NP that A(x,t) follows from Am(x,t) in any circumstances), then, given that CAN(A(x,t)), for some x, the event of A(x,t) depends on two factors, viz. Vam(x,t) and C(x,t). This seems more in line with the ordinary notion of ability, for it would be odd if one could gain and lose a number of abilities merely by undergoing the change in circumstances involved in walking round the block!

Again, in your formulae at the bottom of page 3 illustrating possible falsification of "CAN(A(x,t))," I think you should include "C(x,t)" in the antecedent of "NP(Am(x,t) --> A(x,t))," for in connection with the second one, viz.

(EAm)(EC): CAN(Am(x,t)) & NP(Am(x,t) --> A(x,t)) & ~C(x,t),

it might happen that Vam(x,t) occurs (since it is consistent with C(x,t)), and this would mean, in view of CAN(Am(x,t)), that A(x,t) must also occur -- which would be very odd if ~CAN(A(x,t)) obtains.

Now if you are willing to make the suggested change in the position of "C(x,t)," your analysis comes a lot closer to the one expressed in my paper, Certain differences, of course, remain, two of which are slight and another of which is more basic. The slight or trivial differences are these: (1) you speak of willing and minimal actions while I speak of trying, and (2) your definition contains the operator "NP." These differences are not basic, i.e. they indicate no serious disagreement because (a) I am willing to admit that the analysis of trying must mention willing and minimal actions (though willing, too, needs careful analysis), and (b) I use a complicated subjunctive conditional in my definition, and when this is analyzed symbolically it will probably require modal operators. Our basic difference, then, lies in my use of a bound variable, ranging over states or conditions of persons. (If one's abilities are not partly determined by one's present circumstances, "state" is perhaps a bad word: "condition" is probably much better in that it lacks some of the temporal connotations of the latter.)

How can we resolve our apparent disagreement about the bound variable? Well, perhaps there really is no disagreement here; perhaps the variable will appear when the notion of willing is analyzed. If willing is to involve the framework of thoughts, i.e., "self-directed commands" then, since thinking essentially involves frames of mind which are dispositional states, then a bound variable will probably have to be supplied. On the other hand, you might maintain that "CAN(A(x,t))" does not really involve more than a conjunction of hypotheticals, that abilities are really just "conjunctive-conditional properties." (For your definition, unpacked, is

(EAm)(EC): NP(Vam (x,t) --> AM(x,t)) & NP(C(x,t) --> A(x,t)),

where the above suggestion about changing the position of "C(x,t)" is carried out,) Now, if it is your view that abilities are iffy in this way, then I am inclined to disagree, though I have no strong argument to offer. I mainly have a feeling, a strong one, that "x has the ability A" is to be unpacked along the lines of "x is such that if ...," where the "is such that" requires symbolic treatment along the lines of "(EØ)(Øx, now & (y)(Øy,t' ... ))," the universal quantifiers being required in order to pin down, though indirectly, the condition in point. As I said, however, I have no strong arguments to offer for this interpretation, which also applies, though in a slightly different way to dispositions,

A word, now, on your Phenomenalism paper. The earlier stages of your argument raise no special puzzles for me: having already pondered your views very carefully, I found it easygoing and elegantly done, But the later sections bothered me a bit. At first, when I thought about your points on sensations and logical subjects, I felt that the conclusions of my Feigl paper were neatly wrecked. But now, after another week's thought, I think that my position there is still sound -- though it is far less thorough-going than yours. Let me explain.

As I set up my solution to the problem of sensory consciousness, I am still within the framework of physical things, though I assume, to be sure, that it is still possible to speak of neurophysiological processes. I do of course look forward to a utopian stage of neurophysiological research, but the stage at which I handle the puzzle is still not as utopian as yours -- that is, the stage I consider has not yet reached the point where the physical thing framework is dispensable for the biological sciences. I assume that the physiologist still speaks of animals and their nervous systems, though he is capable of identifying highly intricate neural conditions and specifying them in terms of various functors. Having identified such conditions, however, he might discover that some of them are associated with characteristic patterns of observable behavior -- behavior, that is, that we commonsensically term the effects or manifestations of certain raw feels. Allowing that our talk of a condition or state of a person is elastic enough to permit us to speak of a person's being in a certain Ø-state (instead of a part of his brain's being in such a state -- cf. my remarks on pp. 28-29 of the Feigl paper), then, given that we can characterize raw feels in behavioral terms, i.e., by description (as the condition such that . . . ), and taking seriously the implied uniqueness condition (as discussed in my paper, p. 27), it is easy to proves given the requisite progress in neurophysiology, that R = C, where "R" applies to a given raw feel and "C" applies to a brain state characterized in neurophysiological terms, i.e., in terms of spike potentials, etc., of certain groups or nerve fibers. (Wow, what a sentence!) And once we admit that introspective reports are legitimate, we must take seriously ascription of phenomenal properties to his raw feels -- which means, given that R is a raw feel and P is a phenomenal property truly asserted to it, that a cortical state C also has that phenomenal property, Since C is specified in terms of physical-2 functors, it may be said that C has phenomenal as well as mathematical properties. Since C is a Ø-state of a person, more accurately a set of events in a person's cortex, it is C, not its phenomenal or mathematical properties themselves, that results in observable behavior and results from changes in his sense-organs or in his immediately perceptible environment, This means that though my position involves a sort of dualism (as so far developed), it is very different from either parallelism, interactionism, or the usual forms of epiphenomenalism.

Now I repeat that I do not disagree with your solutions; in fact I am certain that something very like your position must be true. Still, I think that my view is sound as at least a partial resolution of a puzzle formulated within the frame of persons, their experiences, and states of their nervous system. As far as our differences are concerned, I am inclined to think that they stem from the stage of scientific progress with respect to which we formulate a solution of the raw feel -- brain state problem.

Your reflections on the future framework of neurophysiology or physical science generally, while I as strongly inclined to agree with them (I feel that they must be sound), still take my breath away. For I wonder how, within this frame, we are to deal with the semantic matters of empirical significance, public confirmability, and the like, must we always retain the physical-thing-person frame for these purposes? That is, can we have a pragmatic metalanguage without terms for persons, and can a comprehensive semantics be utterly separate from pragmatics?

[Editors note: Professor Aune thinks that there may be another page of this letter -- but cannot find it.]

[Return to Table of Contents]


November 11, 1961

Dear Bruce:

Thanks for the careful letter. I enjoyed reading -- studying it. here are some comments and meta-comments.

1. You are, of course, quite correct in pointing out that 'Am(x,t)' is not a sufficient condition of 'A(x,t)', i.e. that 'NP[Am(x,t) --> A(x,t)]' presupposes the obtaining of a favorable conditions. I was quite aware of this in giving my definition of 'CAN[A(x,t)]'. In effect I was using 'NP[... --> ...]' in the sense of the short arrow introduced at the top of p. 252 in my Vol.II essay. This wasn't too ill-advised at that stage -- compare the definition of 'Sol(x,t)' as 'immersed(x,t) --> disintegrates(x,t)', which will do for the occasion -- but it messed things up in the subsequent discussion. It would have been better to conjoin 'C(x,t)' with 'Am(x,t)' in the antecedent of the NP clause right from the beginning, or, at least, to have used the concept of hypothetical necessity, thus,

NP[Am(x,t) --> A(x,t)]/C(x,t)

and given the following definition,

CAN[A(x,t)] =def (EAm)(EC) CAN[Am(x,t)] & C(x,t) & NP[Am(x,t) --> A(x,t)]/C(x,t)

2. I am inclined to distinguish between 'x can do A at t' and 'x is (at t) able to do A', e.g. between 'x can swim at t' and 'x is (at t) able to swim.' The concepts are obviously closely related. If we define "Able-to-do-A(x,tj' as

(EAm)(EC) CAN[Am(x,t)] & NP[Am(x,t) --> A(x,t)]/C(x,t)

then we could define 'CAN[Am(x,t)]' in terms of this predicate somewhat as follows,

CAN[A(x,t)] =def Able-to-do-A(x,t) & C(x,t)

But there is no need to ring the changes on possible analysanda. It might, however, be worth noting that in the case of minimal actions


would coincide with


both having the sense of 'NP[Vam(x,t) --> Am(x,t)]' though I grant that the concept of a minimal action needs careful exploration to determine whether it coincides with the notion of what can be done that circumstances cannot hinder.

3. As for the question whether dispositions and abilities are to be analyzed in terms of 'such that if . . . .', this might or might not hold us up, depending on whether you conceive of the quantified variable as ranging over "micro-states" or as ranging over thing kinds (and circumstances). I think it important to distinguish between the analysis of a disposition and the explanation of the disposition. A failure to draw a distinction between these two interpretations of "such that" involves a danger of mixing levels of discourse (roughly observation and theoretical language) and I suspect that your treatment of the sensory consciousness problem tacitly commits this error. But of this more later.

Notice that if we take the thing-kind cum circumstances approach, and use the special variable 'p' to express the categorial nature of person as contrasted with ordinary kinds, and recognize further the 'internal relation' between being a person and willing, we see that

CAN[Am(p,t)] NP[Vam(p,t) --> Am(p,t)]

needs no supplementation by quantified variables or such thats -- which is not, of course, to say that it is incapable of further analysis.

4. Notice also the difference between being paralysed and being hindered. This distinction is analytically related to that between minimal and non-minimal actions. Abbreviating 'x is paralysed at t with respect to Am' by 'ParAm(x,t)', we can define it as follows,

ParAm(x,t) =def NP[Vam(x,t) --> ~Am(x,t)]

Once again, the definition of a paralysis must be distinguished from its explanation, whether molar or micro-. 'ParAm(x,t)' clearly entails '~CAN[Am(x,t)]', i.e. '~NP[Vam(x,t)]'. Is the converse true? Does '~CAN[Am(x,t)]' entail 'ParAm(x,t)'? It sounds as though it should, but it does not follow from our definitions as they have so far been explicated that this entailment obtains. For, schematically, '~NP[fx --> gx]' obviously doesn't entail 'NP[fx --> gx]'

5. A closer examination of the situation is rewarding. Notice that whereas ordinarily the conjunction of`singular statements

f(a,t1) & g(a,t1')

merely authorizes the consequence

Possible[f(a,t1) & g(a,t1')]

(from actuality to possibility the argument is good); in the case of volitions we can go from

Vam(Tom,tl) & Am (Tom, t1')


CAN[Am (Tom, t1)]

which, in spite of the word 'CAN' expresses a statement of necessity, for it is equivalent to

NP[Vam(Tom,t1) --> Am(Tom,t1')]

so that in this case from actuality to necessity the argument is good. To understand why this is so is to understand why we think that "we have a non-indicative insight into the causal nexus between volitions and minimal actions." Why is it so? The answer lies in the explication of the concept of a volition. Roughly we mean by a volition (pertaining to minimal actions) a thought which culminates in a minimal action unless one is paralysed. Thus we have not only

Vam(x,t) & CAN[Am(x,t)]



which on our definitions is analytic, but also

Vam(x,t) & ~ParAm(x,t)



which, on the preceding definitions is not. To bring out the character of the second entailment, we must note that



CAN[Am(x,t)] V ParAm(x,t)

ex vi terminorum. A person may not yet have acquired the ability to will to do Am, but if he has, then it is a necessary truth that if he does so, either he is paralysed or he performs Am. Further ramifications of this point might be spelled out on a subsequent occasion.

6. A minor point. Not all doing involves trying to do: but all succeeding in doing does, 'x succeeded' entails 'x tried.' An interesting question is 'Does it make sense to speak of trying to do a minimal action? Certainly, one can (a) try to find out if one is able to do Am; (b) try to acquire the ability to do Am. But my suspicion is that 'x tried to do Am' is nonsense unless it is used to mean 'x tried to do A and for some people A is a minimal action.'

7. I want to take up the problem of sensory consciousness on a subsequent occasion, so I will limit myself to a few off the cuff remarks. As far as I get it, your solution is along naturalistic Aristotelian lines. One state of a person has both 'phenomenal' and 'physical-2' properties. The raw feel state and the neurophysiological state are identical. They are two only in the sense in which the author of Waverly and the author of Ivanhoe are two.

The state which has (phenomenal property) = the state which has (physical-2 property)

The attractiveness of this solution (which is essentially the same as the "double knowledge" solution developed by my father in his first book (Critical Realism) and subsequent publications) lies in avoiding (at least prima facie) both dualism of things and dualism of episodes, and, in particular, dualism of bodily states and epiphenomena. Instead it has a dualism (parallelism?) of attributes of single states. It is not too different from the classical double aspect theory, save that the latter tended to think of aspects not as properties or attributes, but as 'abstract particulars' (thus tending to revert to parallelism).

8. In spite of its plausibility I don't think it will work, unless it is tacitly reinterpreted along the lines of my solution. As I indicated in section 3, I think that your interpretation of dispositional statements has led you to a mixing of conceptual frameworks. The mistake is analogous to that involved in saying, for example, that the elasticity of a rubber band is identical with a certain microproperty of the system of sub-atomic particles "of which it consists," or that the wetness of water is identical with a physical-2 property of H(2)O. The crucial issue is "In what sense arc glasses of water and their states identical with collections of particles anti their states?" These are familiar points, but their application to concepts pertaining to persons may not be obvious until you ask yourself in what sense a network of neurons is "part of a person."

9. I am enclosing a copy of two lectures I gave last year at Pittsburgh. It may help resolve the issue one way or the other.

10. By the way, I am still not certain what is going to happen to the phenomenalism paper. I sent it to Kantstudien two years ago and have as yet received no definite commitment on Gottfried Martin's part to print it. The objection is the length. Is there any chance that it would be acceptable for the Oberlin volume? I did read a substantial section there. Would you be willing to explore the matter with your colleagues?

11. It would be nice if we could narrow our differences (which in any case are on matters of detail, rather than broad principle) by correspondence. But a thorough exploration of the issues will have to wait until we can talk things over. I hope it will be soon.

As ever,


[Return to Table of Contents]

[Aune's Note: This is a letter about a manuscript presented by Sellars in a symposium at the APA in Dec. 1964. The letter in undated, but possibly sent in the Fall of 1964.]


I liked your notes on intentionality very much, but I am afraid that most readers will find it excessively tough-going and obscure,

The difficulty lies in paragraphs 10-13.

At the end of #9 (# = paragraph) you announce your next move as that of exploring what it is for a token of a sentence in, say, German to express a proposition. Yet in the next paragraph, where this exploration is supposed to start, you give every appearance of changing the subject -- for without any explanation you begin to talk about concepts and meaning. To anyone unfamiliar with your work, this transition will seem bewildering.

In order to avoid this difficulty you must relate meaning to expressing and concepts to propositions in a clearer, more explicit manner. And you must also, at some stage, explicitly draw the inference that because meaning is not a relation, statements describing the proposition expressed by certain sentences are not genuinely, or really, relational either, and that to speak of the proposition expressed by a certain sentence is not to relate that sentence to an extra linguistic abstract entity. (This sort of inference is needed because without it the reader will, I am sure, wonder about the bearing your 'meaning is not a relation' slogan has on the task at hand, which is, again, that of explaining what it is for a sentence to express a proposition.

In order to clarify fully the #s 10-13 you might also do this following (this is only a suggestion which occurs to me now). At the beginning of #10 you might remark that before moving directly to the complex case of propositions, you want to begin with something a little simpler, viz. the rest of your discussion in #10 and all of #s 11, 12, and 13. This done, add a paragraph relating the previous discussion of nouns and concepts to the problem at hand, i.e. to propositions and the meaning of sentences. I believe this could be done in a very short space by paring

'Himmel' (in G) expr the concept sky with 'Es regnet' expr the prop . . .


'Himmel' (in G) means sky with 'Es regnet' means it is raining

You could then explicitly introduce your proposal about props, viz that "'Es regnet' means it is raining" is, at bottom, the PMese stt "'Es regnet' < it is raining; and that the non-relational counterpart to "'Es regnet' expresses the prop that it is raining" is "'Es regnet's (in G) are .it is raining.s" Finally, with just a few words you could construct almost a word-for-word counterpart to #13, viz. "To claim that 'Es regnet' (in G) means it is raining because .....

I might of course misunderstood your intentions in such a way that the above suggestions are unworkable. But as I see it now, they would help the ordinary reader quite a lot.

I have also made a few comments on punctuation, style, etc. They are in red pencil on the copy. I put down everything I could think of, no matter how insignificant and questionable, because you might possible decide that certain tiny things are worth changing.


[Return to Table of Contents]

University of Pittsburgh
Pittsburgh, Pennsylvania 15213

Department of Philosophy

Friday Oct 3, 1969

Dear Bruce

I really appreciate your care in following up the question you raised about the inadequacy of the principle

(p implies q) implies (shall[p] implies shall[q]).

I found your letter exactly to the point and sat right down to reflect on an answer. Perhaps the following remarks will help orient further discussion.

I have never -- not even on p. 190 of the Castaneda volume -- said that 'exportation holds for implication.' After all, I was brought up on Lewis & Langford and learned almost at my mother's knee to avoid confusing

p < (q < r)


p < (q --> r)

What I have done is argue that there is a generic notion of implication which is such that (roughly)

'p' implies 'q' <--> the consequence 'p,q' is justified

In other words, if 'p' implies 'q' in the generic sense, then if 'p' belongs to one's corpus then the addition of 'q' is justified. This generic notion of implication leaves it open whether or not the justification is purely "formal" or "ultra-conceptual" or whether it would appeal to matters of fact. (I will not distinguish "formal" from "ultra-conceptual" in this letter, since my point can be made without exploring physical implication -- though my treatment of the latter would, I believe, add grist to my mill).What logicians have tried to do is systematize certain species of implication where the justification would appeal to purely formal considerations. I would certainly agree with your point that exportation does not hold with regard to the implications they systematized.

Clearly 'p --> q' should not be read as ' 'p' implies 'q'.' The latter means that the move from 'p' to 'q' is justified and, usually, that it is logically or conceptually justified. On the other hand, given that 'p --> q' is true, the sequence 'p,q' is justified, though not formally so. The relevant formally true implication is, of course,

'p & p --> q' implies 'q'

on the other hand, knowing the latter, and knowing that 'p --> q' is true, I know that if I am entitled to add 'p' to my corpus, I am also entitled to add 'q.'

Similarly, if I know that 'p' is true, then I know that 'p --> q: --> q' is true, because

'p --> [p --> q: --> q]'

is formally true. As a result, I know that this sequence '(p --> q)

, q' is justified, though not formally so. (as you correctly found out

p < [p --> q: --> q]

does not entail

p < [p --> q: < q].)

In other words, I know -- though not on purely formal grounds -- that I am entitled to add 'q' to my corpus if I am entitled to add 'p --> q.' This means that in the generic sense of 'implies' I know that

'p --> q' implies 'q'

It would, in my opinion, be a serious mistake to treat any of the standard theories of logical implication as an explication of the pre-analytic notion of implication which we constantly employ in practical and theoretical reasoning. If I gave the impression in the Castañeda volume that the term 'implies' as it occurs in the examples of practical reasoning which I gave has the sense of 'strictly implies' or 'logically implies' (i.e. L-implies) this was carelessness on my part. I was well aware that the arguments are fallacious if 'implies' is taken in the usual sense given by logicians. Perhaps the above remarks will throw light on what I was trying to do with the phrase 'relative implication' or 'implication relative to an assumption.'

As far as the new form you raise is concerned, I simply don't see it. We must, in the case of theoretical argument distinguish not only between the validity of an argument and its goodness (i.e. its being valid and having true premises) but between these and the reasonableness of a given person offering it. Thus

All mules are barren.
Daisy (we shall suppose) is a mule.
So, Daisy is barren.

is both valid and good. But if Jones believes either of the premises to be false, it would be, in a legitimate sense, unreasonable of him to offer it.

It would be unreasonable of me to reason

Shall [I will do A, if p]
So Shall [I will do A]

if I doubt that p. But this is because it would be unreasonable of me to believe

'I will do A, if p' implies 'I will do A'

if I do not accept p. The point in question is independent of the distinction between 'theoretical' and 'practical reasoning.'

I was delighted to learn that Herbert is better. I must drop him a line.



[Return to Table of Contents]

New York, N. Y. 10021

Jan. 28, 1970

Dear Bruce --

The paragraph at the bottom of page 2 of my letter is not a model of clarity. Let me try again along somewhat different lines. My aim was to explicate the concept

alpha implies beta relative to assumption delta

in a way which brings out its relation to the idea of one proposition implying another simpliciter.

The argument should have begun by pointing out that if, for example, 'p' implies 'q' simpliciter,then, without knowing the truth values of 'p' and 'q' we can reject the sequence


(We can also reject the sequence

p (= ~~p)

by virtue of the contrapositive: '~q' implies '~p.' For present purposes I can treat such pairs of sequences, which differ only in their order, as the same.)

Consider, now, the claim that 'p --> q' implies 'q' relative to the assumption 'p' [I change my example to fit the structure of reasoning implying conditional intentions -- see the example analyzed in George and Hector's book.] To what extent can we analyze this case along similar lines? Thus, can we say that given the truth of 'p' we can reject the sequence

p --> q

without knowing the truth values of 'p --> q' and 'q'?

The following line of thought seems to support an affirmative answer: We know, to begin with, that the following implication implication is logically true

'p & p --> q' implies 'q'

Consequently we know that without knowing the truth values of 'p' 'p --> q' and 'q' we can reject the sequence

p --> q

The above implication by itself, however, is compatible with rejecting the sequence

p --> q

for it is compatible with accepting the sequence

p --> q

of which it is a part. If, however, we now add the assumption that 'p' is true, the latter sequence becomes unacceptable, and hence both the sequence

p --> q


p --> q

are unacceptable.

Let me call these sequences the 'p' elaborations of the sequence

p --> q

and suppose that since both 'p' elaborations of the latter sequence are unacceptable, the sequence itself is unacceptable. The principle here is that if a sequence is to be acceptable, then for any proposition alpha either the sequence which results by adding alpha or the sequence which results by adding ~alpha must be acceptable, i.e. one or other of the alpha-elaborations of the sequence must be acceptable. (This can be put in terms of embedding the sequence in a model set or possible world.)

Thus, if we took only the logical implication

'p & p --> q' implies 'q'

into account, the sequence

p --> q

would not merely be specified as unacceptable, for its 'p' elaboration in terms of '~p' is acceptable, though that in terms of 'p' is not.

It might be objected that I am bringing in other formulas of acceptability than implication, for in the above account the unacceptability of

p --> q

rests on the truth of 'p' and hence on the falsity of '~p' as a constituent of the sequence? So what?! The fact remains that in this case the unacceptability of the sequence

p --> q

can be known without the truth values of its constituent and this is what I am claiming to be the core notion of implication.

What of the example in my letter?

'p' implies 'q' relative to the assumption 'p --> q'

The pattern is the same. We are entitled to rule out the sequence


without knowing the truth value of 'p' and 'q.' Once again the equally true implication

'p & p --> q' implies 'q'

enables us to rule out the sequence

p --> q

But it still permits the sequence


for it permits

~(p --> q)

of which it is a proper part.

Now let us stipulate that 'p --> q' is true. Both 'p --> q' elaboration of
p --> q
are now unacceptable, and, hence, the sequence itself is unacceptable, and known to be so independently of knowing the truth values of 'p' and 'q' -- though, it does, of course, require knowledge of the truth value of 'p --> q.'

I think that some such general notion of implication throws light on a number of interesting uses of 'implies,' though to show this would require an elaboration and sophistication of the above account. Consider, for example,

'Nixon is wise' implies 'Nixon exists'

This rules out the sequence

Nixon is wise
Nixon does not exist

Notice, however, that the argument implication is now equivalent to the contraposition

'Nixon does not exist' implies 'Nixon is not wise'

This, however, is but the beginning of larger story.

I hope this makes sufficient sense (and is legible enough) to be of help.



[Return to Table of Contents]

September 3, 1970

Dear Wilfrid,

Thanks for the letter of July 20. I was on vacation when it arrived, so I didn't get to it until recently. I wasn't convinced by your last remarks (see below) and I decided to write up a little paper putting all my doubts about your views on practical reasoning together. My thought was that I might produce a long study of you on practical reasoning for the planned Castañeda volume and I could include the material from this short paper as a section. But I am not committed to the plan. The paper (which I enclose) is not intended for general circulation; I am mainly interested in your reaction. As you will see, my doubts about the foundation of your theory of practical reasoning are serious, but the further development of your theory is, I believe, unaffected by my criticisms.

Now about your letter. Your statement (5) is problematic. Surely if P implies Q, then it is false that P & ~Q. Hence, if I know that P implies Q, then I know that either P is false or Q is true. I do not, of course, know what the exact values of P and Q are (or at least I need not), but then the same may be true of my knowledge that P only materially implies Q.

Note that, strictly speaking, your (1) is false. If I know that P is true and Q is false, I know the truth-values of P and Q but I do not know that P materially implies Q since ~(P --> Q). What you meant in laying down (1) was that if I know that P is false or Q is true, then I know that P materially implies Q. This claim is, of course, correct, but it doesn't cast doubt on anything I said in my last letter.

As I understand your letter, you are distinguishing your generic implication from material implication on the ground that while knowing that P is false or Q is true is sufficient for knowing that P materially implies Q. it is not sufficient for knowing that P generically implies Q. This strategy does, of course, distinguish the two kinds of implication in a weak sense, but it does not throw any light on the generic sense of implication which I knew full well would not be truth functional. The point of my last letter was that what you claimed to be necessary for P implies Q -- namely, that one could reject the sequence P,~Q without knowing the exact truth-values of P and Q -- is also necessary for P materially implies Q. The reason that this condition is necessary in both cases is that the knowledge of an implication, whether generic or only material, requires one to know that the sequence P,~Q Is unacceptable.

Tell me what you think of the enclosed paper.

Best regards,



I saw Herbert Feigl when I was in Minneapolis. His legs are bad, but he looks surprisingly good. We had a pleasant chat.

[Return to Table of Contents]

University of Pittsburgh

Department of Philosophy

June 23, 1975

Professor Bruce Aune
Philosophy Department
University of Massachusetts
Amherst, Massachusetts 01003

Dear Bruce:

Thanks for the two letters and for the handsome manuscript which has just arrived. I had intended to wait until you had a chance to get re-settled in Amherst before I began bugging you, but I cannot refrain from sending a few disjointed comments to show how interested I am in getting a dialogue going.

I don't think I am as far away from my "earlier simpler theory" as you take me to be. I am just trying to fit it into a larger context which places less stress on inferences, and more on the place of practical reasoning in a theory of persons as agents. But you will be the best judge of that after I wrestle with a few of your probing questions.

Let me begin by noting that in your letter to Hector of May 31, 1975 you write (p. 3):

. . . let me emphasize again that, although I regard "I will do A" and "I shall do A" as semantically or formally indistinguishable, I do not think (or claim) that they are indistinguishable tout court. Their difference is logically unimportant but practically very important. Not only do utterances of "I will do A" express attitudes that are not expressed by utterances of "I shall do A," but thoughts of the form [sic] "I will do A" have important causal properties that thoughts of the form "I shall do A" do not have.

I put in the 'sic' to call attention to the conceptual tension involved in saying that formally indistinguishable utterances have different forms. You will reply, of course, that different senses of 'form' are involved in these two contexts. But would that not be to grant that there is more than a 'practical' and 'causal' difference between 'I will do A' and 'I shall do A'? I suspect that the old issue as to what is to count as a 'logical' principle is lurking in the underbrush. Is the difference between 'I will do A' and 'I shall do A' a difference in 'logical' form? When is a difference in form a 'logical' difference? Are all differences in 'conceptual' power differences in 'logical' powers? (Similar considerations relate to the term 'semantical.') To point up these tentative queries, I note that on p. l of your letter to Hector you say that 'P and Q' and 'P although Q' have the same logical form. I don't object to this. I share your taste for keeping the term 'logical' and 'logical form' on a tight leash. I would, however, object if you went on to say that the difference between these two statements is merely 'practical' and 'causal,' i.e. that there is no interesting sense of 'form' in which the difference is a formal one, or sense of 'logical' in which the difference is a logical one.

Perhaps, from a strictly logical point of view one captures the logic of 'Shall [I to do A]' by identifying it with 'Shall [I will do A]'{1}. From this point of view,

Shall [if I will do A, I not to do B]
So, shall [if I will do B, I not to do A]

coincides with (can not be distinguished from)

Shall [if I will do A, I will not do B]
So, shall [if I will do B, I will not do A]

Another way of putting this might be to say that from a strictly logical point of view 'I to do A' occurs amid logical connectives only as a corrupt version of 'I will do A.'

Yet there might be a conceptual difference between the two expressions which is to be explicated along the lines suggested in "Volitions Re-Affirmed." The difference (relating to the role of the So-be-it principle) might be called 'dialectical' rather than 'logical.'

But enough of this background painting! Let me turn to specifics. On p. 2 of your letter to me you write:

Though I have argued . . . that the S-operator is not really necessary, it is certainly OK as an explicit indicator that a formula expresses a volitional attitude.

I suspect that here is the crux. I think of a 'volitional attitude' in terms of tokening or having propensities to token expressions having a constituent with the conceptual powers of an S-operator. To be sure, these conceptual powers include the causal power of 'S [I to do A now],' but they are not exhausted by it. Thus from my point of view, to attempt to explain the S-operator in terms of an independently specifiable 'volitional attitude' is somewhat like trying to explain 'not' in terms of an independently specifiable 'attitude' of negation.

The above has a direct tie-in with your doubts about my emphasis on the distinction between intentions to do and intentions that something be the case. My views on the importance of this distinction developed from brooding about Kant's argument for the existence of God: (roughly)

Virtue ought always to be rewarded. So, it can be the case that virtue is always rewarded. Only if God exists can it be the case that virtue is always rewarded. Therefore, God exists.

The fallacy is exposed by unpacking the first premise as 'If any agent(s) is (are) in a position to bring it about that virtue is always rewarded, he (they) ought to do so.' This unpacking of 'ought to be' in terms of 'ought to do' blocks Kant's simplistic use of the 'ought implies can' principle. But this unpacking surely requires a corresponding unpacking of 'shall be' in terms of 'shall do,' (i.e. anglice 'will do'). This I have always believed, though I have never paid enough attention to spelling it out.

It is, undoubtedly, an oversimplification to say that 'It shall be the case that-p' is equivalent to 'I shall do what I can to make it the case that-p.' Yet this account is, I believe, on the right track, even though it hasn't gone very far. Is it an analysis? That would be too strong a claim; but the truth it contains must be grounded in an analysis of the relevant concepts -- which I shall attempt to sketch below. For the moment I shall simply express my conviction that there is a very strong conceptual tie between intending that something be the case and intending to something, i.e., that

Jones intends that-p but has no intention to do anything (not 'anything specific,' 'anything').

is conceptually incoherent.

Clearly the putative analysans must be construed as somehow involving a ceteris paribus clause. Yet this is the most trivial aspect of the problem, for it merely points up the fact that insofar as one is rational, one is prepared to take new information into account and even to allow for the possibility of simply changing ones mind. I quite agree that it would be irrational to intend to do all of a number of as yet unspecified actions each of which would ensure its being the case that-p.{2} Why keep a dog and bark yourself?

So we don't want

Shall be [p]

to imply (in the desired sense)

Shall [(A)(t) I do A at t ensures p --> I to do A at A at t]

Well, why not

Shall [(EA)(Et) I do A at t ensures p --> I to do A at t]?

Suppose I accept 'It shall be the case that-p'. I discover that by doing A1 (which I abhor) at t1, I can ensure that-p. Unless I am really dedicated to its being the case that-p, I will ponder

Shall be [alpha and p and I do A1 at t1]?
Shall be [alpha and not-p and I not do A1 at t1]?

and choose

Shall be [alpha and not-p and I not do A1 at t1]

Notice, however, that since it was not stipulated that my doing A1 at t1 was necessary to its being the case that-p (but only that it was sufficient), I can still intend that-p

Shall be [alpha' and p]

where the background constituent represented by 'alpha' ' now includes 'I do not do A1 at t1.' According to the putative analysis, we would have

Shall [alpha' and ((EA)(Et) I do A at t ensures p --> I do A at t)]

It is time to give a little more cash in support of the claim that we are on the trail of an analysis. Surely we need a distinction between those intention-constituents which get into a scenario through So-be-it, and those which are regarded by the agent as up to him and not yet decided. Remember that even if it has been decided that something will be the case -- a fact -- it still gets into a scenario -- into the scope of an S-operator -- by virtue of being relevant to a choice between alternative scenarios, thus

Shall [alpha and p and I bring about X]?
Shall [alpha and p and I bring about not-X]?

We would expect surface grammar to reflect these distinctions. When I stress the difference between

Shall [I will do A]


Shall [I to do A]

my concern is to capture the way in which the surface grammar of practical sentences reflect connections which are conceptual, though not in your sense logical.

It should be clear by now that from my point of view the fundamental unit of intending is the scenario. As a first (and crude) approximation I give it the form

I shall* bring lt about that [. . .]{3}

All 'shall be's and 'shall* do's are to be construed as abstractions from this matrix. And if we construe 'I shall* bring it about that [. . .]' as 'I shall* do what I can to bring it about that [. . .]', then, while

Shall be [p]

is not itself to be analyzed as

I shall* do what I can to bring it about that-p

it presupposes that 'p' is a constituent of the scenario having the form

I shall* bring it about that [. . . and p . . .]

Only if 'p' is up to the agent does the latter imply

I shall* do what I can to bring it about that-p

Even if 'p' is not up to the agent, however, it implies the weaker

Shall be [p].

Notice that it is because this cumbersome shall-operator

I shall bring it about that . . .

takes declarative sentences (atomic, molecular, quantified) in its scope, that one is motivated to introduce indicators to reflect the surface grammar distinctions which appear in ordinary language pertaining to intentions. This explains why I wrestled with 'targets' in my paper on "Contrary-to-Duty Imperatives."

If all this is not completely misguided, I can withdraw my claim that 'It shall be the case that-p' is equivalent to 'I shall* do what I can to make it the case that-p,' while preserving the intuition which led me to make it. Perhaps the best thing to do is to stress the importance of the clause 'what I can.' For if the agent believes that 'p' is not up to him a fortiori he believes that there is nothing he can do to make it the case that-p and the equivalence collapses. The general point that intentions differ from wishes ('shall' from 'would') by virtue of the fact that 'Jones intends that-p' implies 'Jones thinks it possible that-p' ('Jones does not think it impossible that-p'?) is a relevant consideration here.

But what, then, are we to say from the standpoint of the logic (in the strict, Aune, sense) of intentions about the suggested equivalence between

Shall [p]


Shall [(EA)(Et) I do A at t ensures p --> I do A at t]

I see no reason to abandon it. It preserves at the hard core level the connection between intending that-p and intending to do what one can to ensure its being the case that-p, if that-p is up to one. Of course, if it turns out that that-p is up to one, say, by doing A1, one may decide to pass up the opportunity. But, then, as you rightly stress in your letter to Hector, logic never forces us to take a volitional attitude towards any state of affairs, not even

Shall [p or not-p]


As ever,

Wilfrid Sellars


[Return to Table of Contents]

June 9, 1979

Dear Wilfrid,

In my last letter to you I claimed that a key principle you use in connection with your concept of dependent implication is, as I now see it, invalid, but I was too lazy to think up a counter-instance to that principle. Lying in bed this morning, I decided to think up a counter-instance, and the following occurred to me.

The objectionable principle is this: If p implies q rel to r and p implies q implies and is implied by (IIB) s implies t, then s implies t rel to r.

Here's a counter-example.
r = x is a male (M)
p = x is unmarried (U)
q = x is a bachelor (B)
s = x is unmarried (U)
t = x is a spinster.
Given this interpretation, the following appears to be true, though both p implies q and s implies t are false:

(K): p implies q IIB s implies t.

But while it is true that p implies q rel r (i.e. p & r implies q), it is false that s implies t rel r (i.e. s & r implies t). QED.

The kicker here is the principle (K), according to which being unmarried implies being a bachelor iff being unmarried implies being a spinster. It seems to me that this principle is clearly true, though I have no way of proving it. The key to constructing a counter-example for your principle is, in any case, this: find interpretations of p implies q and s implies t according to which both formulas are false and yet imply one another, where the conjunction of p and r nevertheless implies q. I think that (K) does the trick, but here is another possibility if you are not convinced. Let x and y be positive integers. Then the following is true:

x^2 = y and y < 2 implies x^2 = 1.

But for an arbitrary positive integer w, it seems true that

(y < 2 implies x^2 = 1) implies & is implied by (w < 2 implies x^2 = 1).

Yet it is not true that

x^2 = y & w < 2 implies x^2 = 1.

If neither of these counter-instances convinces you, I am sure that I can find others. There is something tricky about them, for we have to find two false formulas that imply one another, and the notion of implication is somewhat obscure. Nevertheless, to go back to my last example, it seems obvious that, for arbitrary x and y, we can assert and deny that y < 2 implies x^2 = 1 just when we can assert and deny that, for an arbitrary w, w < 2 implies x^2 = 1. The same for the kicker in my first example: we can assert and deny that being unmarried is sufficient for being a bachelor just when we can assert and deny that being unmarried is sufficient for being a spinster. So I think both of my counter-examples are OK.

Tell me what you think.

All the best,


[Return to Table of Contents]

July 3, 1973

Professor Bruce Aune
Philosophy Department
University of Massachusetts
Amherst, Massachusetts 01002

Dear Bruce:

Thank you for the friendly opening paragraph of your letter. The episode at the Fogelin's was, I believe, a case of misperception all around. I was exhilarated by having my first lecture -- always the most difficult -- behind me. I can easily see why my mood seemed aggressive. It was certainly expansive -- and the line between expansion and aggression is notoriously difficult to draw. Since one is at least as responsible for the appearance one presents as for the reality, my sincere apology.

You have long been one of the very few on whom I can count for understanding criticism. You are at home -- if not always comfortably -- in the dialectic, and can quickly spot questionable moves. This is why, in the few cases in which you have not convinced me, I keep on trying to convince you.

The current exchange is a case in point. I don't think I am flogging a dead horse, if I attempt at least a partial rebuttal. Thus I most certainly did not "tacitly" appeal to "a principle of Interchange of Equivalents" in order to get from

(1) p --> q implies q (relative to p)


(2) S(p --> q) implies S(q) (relative to p).

I quite agree that the following argument is invalid,

p --> q implies q (relative to p)
[p --> q implies q) implies and is implied by (IIB) [l --> m implies m]
therefore, 1 --> m implies m (relative to p)

What I actually did was set up what you refer to as the 'fundamental principle'

P implies Q IIB S(P) implies S(Q)

and interpret the two occurrences of 'implies' as a generic term for relations which authorize inference. I claimed that 'logical implication,' 'nomological implication' and 'implication on an hypothesis' are species of this relation. (A check on what I wrote in "Thought and Action" and Science and Metaphysics will confirm this.)

Thus, as I saw it,

p --> q implies-relative-to-p q IIB S(p --> q) implies-relative-to-p S(q)

is simply a special case of the fundamental principle rather than something to be established by a combination of the fundamental principle and "a principle of Interchange of Equivalents." There may be -- indeed undoubtedly are -- problems about the moves I made. But they are not open to your counter-example.

What, then, of the idea that implies-relative-to-p is an inference authorizing relation? Obviously,

(p --> q) --> q

does not hold in all possible worlds involving p and q. It would be an obvious howler to argue

If p, then, necessarily, (p --> q) --> q
Therefore, it is necessary that (p --> q) --> q

The major premise is true, the conclusion, however, false.

On the other hand,

(p --> q) --> q

does hold in all possible worlds in which p holds.{4}

Does it not make sense to say

Given that-p (or on the hypotheses that-p), p --> q implies q

where this means

Given that-p, one may infer q from p --> q?

The reasoning

If p, one may infer q from p --> q
Therefore, one may infer q from p --> q

involves, indeed, a detaching of

. . . one may infer q from p --> q

But this detaching does not involve a commitment to the unconditionality of this permissible inference. That is to say, it does not involve a commitment to

p --> q logically implies q


If p, it is permissible to interrupt
Therefore, it is permissible to interrupt

This also involves a detaching of the conclusion

. . . it is permissible to interrupt

but it does not involve a commitment to

It is unconditionally permissible to interrupt

Compare, also,

If one has promised to do A, one is under an obligation to do A
Jones has promised to do A
Therefore, Jones is under an obligation to do A

Obligations are essentially tied to grounds, but this does not mean that obligation statements have to specify their grounds.

Jones is under an obligation to do A

is, in an important sense, a complete statement. Yet it is, by virtue of the conceptual grammar of obligation talk, defeasible. It raises the question 'Why?' Appropriate replies are

Because he promised to do A.
Aune said he ought to, so there must be some reason.
On reflection, I see that I was mistaken. There is no reason why Jones is under an obligation to do A. I withdraw the assertion.


One may infer q from p --> q

is complete, but defeasible. In particular, one can reply to the question 'Why?'

Because: p and in all possible worlds in which p is the case, q is the case if p --> q is the case.

Of course, the statement

One is logically entitled to infer p from p --> q

where this means

One is, as a mere matter of logic, entitled to infer q from p --> q

is false. But only by confusion can one conclude that

One is entitled to infer q from p --> q

is false.


I should, perhaps, have made it more explicit that

alpha implies-relative-to-gamma beta

has the same sense as

If gamma then alpha would imply beta

i.e., to put it in terms of inference,

If gamma then one would be entitled to infer beta from alpha.

It is this which underlies the reasoning

p --> q implies-relative-to-p q
Therefore, p --> q implies q

Once again, the conclusion of this argument does not have the sense of

p --> q logically implies q


N [(p --> q) --> q]

and the argument does not commit the fallacy of exportation.


To turn to the bearing of the preceding on the logic of conditional intentions, consider the following sequence

1. p Hyp.
2. S(p --> q) Hyp.
3. p & p --> q implies q Logic
4. p --> q implies-relative-to-p q Rel. Imp.
5. p --> q implies q 1, 4, Rel. Imp. -> Imp.
6. S(p --> q) implies S(q) 5, Fund. Prin.
7. S(q) 2, 6, M.P.

Compare it with

1. p Hyp.
2. S(p --> q) Hyp.
3. S(p) 1, So-be-it
4. S[p & (p --> q)] 2, 3, C.I. [Shall]
5. p & (p --> q) implies q Logic
6. S[p & (p --> q)] implies S(q)Fund. Prin.
7. S(q)4, 6, M.P.

As far as I can see, these arguments are conceptually equivalent. In each case one is led to consider possible worlds in which p obtains, and, therefore, possible scenarios (intention structures) in the content of which p is a conjunct. And this is done in terms of conceptual grammar of beliefs and intentions which brings them together without admitting

p and shall (p --> q)

into the depth grammar, i.e., without doing violence to the scope-ishness of the shall operator.


The reference to the scope-ishness of the shall operator reminds me of your comment on my animadversions on tense logic in "On Reasoning About Values." You write,

. . . what you did . . . was to reformulate the general lines of your view and to add some remarks directed at my position (not my objections), on the logical character of such operators as "It was the case that . . ." Thus I really didn't have anything to say in reply. I will grant, for example, that "It was the case that . . ." is a logical operator, a logical constant, but this doesn't commit me to anything, one way or the other, about the S-operator.

But my point wasn't that tense locutions are logical operators or logical constants (they aren't!) It was rather that tense locutions have a "logic" or "conceptual grammar" which is not reducible to principles of the propositional and functional calculi. This conceptual grammar can be presented in terms of a system of implications. Thus I was comparing the implications involved in the meaning of 'shall' to the implications involved in the meaning of tense locutions, e.g.

'S is P (at t)' implies 'It will always be the case that S was P (at t)'

Thus, in the passage to which you referred I wasn't attacking your position, at least directly, but, rather, highlighting what I regard as one of the strengths of my own position, i.e., the fact that it does have something specific to say about the conceptual role of 'shall'/'will'. I have always stressed that just as the characteristic expressions of language entry transitions involve both world-->word connections and intra-linguistic word--> word connections (inference patterns), so the characteristic expression of language departure transitions involve both word-->world connections and inference patterns.

There is, I take it, no disagreement in principle between us that the role of 'shall' and its cousins can be described in causal terms. Furthermore, I would not regard it as incorrect to emphasize, as you do, that the distinctive feature of intentions is their causal role in bringing about actions. But I would immediately want to add, as you do not, that the causal role of expressions of intention is not limited to this word-->world role -- (a) they are causally involved in specific inference patterns; (b) these specific inference patterns are distinctive moments in the distinctiveness of the framework of intentions.

But perhaps I have misunderstood you. In any event, I hope that my pressing of this point will help us clear away a persistent source of polemical exchanges on what must surely be, in the last analysis, peripheral issues.

You write,

What I think I have done is to point to a number of difficulties in your theory and to offer an alternative account that lacks the difficulties and yet does justice to your claim that practical logic (to the extent that there is such a thing) is entirely derivative from ordinary indicative logic. My nonsystem [own system?] is internally consistent and has a clear-cut semantical interpretation. The obvious place to attack it is to attack the semantical interpretation . . .

I hope that the opening paragraphs of this letter have clarified my strategy for dealing with the difficulties you have raised with my theory. Furthermore, since we agree that "practical logic (to the extent that there is such a thing) is entirely derivative from ordinary indicative logic," I would expect to find no disagreement with respect to the semantical interpretation of valid argument forms relating scenarios to scenarios. If the general thrust of this letter is correct, such differences as there are between us concern not 'practical logic' in this narrow sense, with respect to which semantical interpretation is decisive, but rather the conceptual grammar of resolutive locutions. From my point of view, practical logic in the narrow sense is embedded in this conceptual grammar, and its formulations, to be philosophically perspicuous, must reflect the depth grammar of resolutive, and, in particular, the scope-ishness with respect to which you and Hector have lined up against me from the start.

Have, as they say, a good summer,


Wilfrid Sellars


P.S. I enclose a Xerox of my previous letter in which typos in the arrow formulae have been corrected. W.

[Return to Table of Contents] [Return to Main Page]


{1} I am, of course, using my idiolect in which 'will' is simply the future declarative. [Back]

{2} As usual one abstracts from considerations of probability. I tried to develop a strategy for taking probability into account in the concluding section of "Induction as Vindication." [Back]

{3} From now on, 'shall' with an asterisk will represent the English 'will.' [Back]

{4} It is worth reflecting that the fundamental intuitions involved in the concept of implication relation to an hypotheses are captured by the role of subordinate proofs in systems of natural deduction (e.g. Fitch). [Back]

[Return to Table of Contents] [Return to Main Page]