Wilfrid Sellars and Jay F. Rosenberg

Edited in hypertext by Andrew Chrucky

Editor's Note: Due to the limitation of current hypertext, the following expressions have been substituted for the original ones. In general, if a letter has some mark over it, that mark is placed as a prefix to the expression: thus, a 'y' with a bar over it is rendered as '-y' (not to be confused with a negation of 'y' which is rendered with a bold tilde before the expression, e.g., '~y'); a 'y' with a circumflex over it is rendered as '^y'. All Greek characters (except phi) are rendered by their names: thus, 'beta', 'gamma', 'psi'; the Greek epsilon which expresses membership in a class is likewise rendered by its name as 'epsilon'. Subscripts are placed in parentheses as concatenated suffixes: thus, e.g., H(2)O is the chemical formula for water. Sellars' dot quotes are expressed by bold periods. The sign for concatenation is a bold circumflex '^'.

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

July 25, 1972: Sellars to Rosenberg
August 29, 1972: Rosenberg to Sellars
September 5, 1972: Sellars to Rosenberg
September 28, 1972: Rosenberg to Sellars
January 16, 1973: Sellars to Rosenberg



July 55, 1972

Professor Jay Rosenberg
Philosophy Department
The University of North Carolina
Chapel Hill, North Carolina 27514

Dear Jay:

The idea of a symposium on the foundations approach to knowledge is an attractive one. The issue is lurking in many contemporary controversies, even when its presence is not immediately evident. I am tempted to volunteer, but would have to reassess my commitments before I give a definite response to your exploratory questions. Roderick Firth and a good Bergmannian would make an interesting "inconsistent triad," but there are many others whose names will occur to you.

Now for the seriousness of philosophy. Your question,

How does one say perspicuously what "Facts are not objects (i.e., not particulars)" says unperspicuously?

calls for a commentary as well as an answer. And the answer itself needs to be given in stages.

The first step in the commentary is to counter with another question: Why should I want to say (perspicuously) that facts are not objects, when I clearly think that propositions are objects{1} and that facts are true propositions?{2}

Thus, from my point of view your question should be:

How does one say perspicuously what "Facts are objects (i.e. particulars)" says unperspicuously?

or, to lay aside the dimension of truth (semantic assertibility),

How does one say perspicuously what "States of affairs are objects (i.e. particulars) says unperspicuously?

(where states of affairs are entities which may or may not 'obtain').

Consider, for example,

(The state of affairs) that fa is an object (i.e. a particular).

Surely my view is that the expression 'that fa' has the form

The [.f.].a.

and is a DST which corresponds to the metalinguistic{3} sortal


which applies to any item in any language which does the job which is done in our language [in a PMese reconstruction] by an 'a' which has the property of being concatenated with a preceding 'f.'

Two comments: (1) To be an [.f.].a. is to be governed by semantical rules. In this respect, to be an [.f.].a. is like being a pawn. (2) Granted that something would not be an [.f.].a. unless it were true (a fact) that it had empirical characteristics by virtue of which it was (in its language) an .a. and was [.f.],{4} this does not mean that its being an [.f.].a. is the same thing as this truth or fact. For clearly, unless it was true of a certain object, 0, that it was of a certain empirical character, it would not be a pawn in our familiar game of chess; but this does not mean that pawns are facts and not objects.

The next step is to note that the phrase which follows the DST,

The [.f.].a.

must be understood as a transformation of 'are particulars' which adopts it to the DST. Thus,

The pawn is a particular


Pawns are particulars


The [.f.].a. is a particular

becomes [.f.].a.s are particulars.{5}

The strategy should now be clear, but it should be rounded off with a discussion of categorizing statements. It is important to note that the latter are ambiguous in a way which calls for two "rational reconstructions ."

(A) We can take (and improve) the way I took in "Empiricism and Abstract Entities." Here categorizing statements are construed as object language showings of how expressions belonging to different categories function. According to this "rational reconstruction"

1. a is a particular =df -y/y = y/a {6}

2. f is a quality =df -g/(x) gx v ~gx/f

3. p is a proposition =df -q/q v ~q/p

where the forms which are illustrated on the right are introduced as follows:

-y/fy/x =df (Eg):(y)gy <--> fy & gx{7}

where 'fy' represents any open (extensional) sentence with the variable 'y.'

-g/F(g)/f =df (EG):(g) G(g) <--> F(g) & G(f)

where 'F(g)' represents any open (extensional) sentence with the predicate variables 'g.' Again,

-q/ø(q)/p =df (Epsi) psi(q) <--> ø(q) & psi(p)

where 'ø(q)' represents any open (extensional) sentence with the sentential variable 'q.'

Notice that as thus defined, expressions beginning with '-y,' '-g' and '-q' are neither singular terms nor sortal predicates. They need not be predicative at all, save in the broad sense in which both

Tall (Tom)


Tom = Tom

are predicative with respect to Tom (roughly, say something about him), and in which

(Ex) tall x

is predicative with respect to tall, for it says something about tall mainly that tall is how something is.

Thus, strictly, the above definientia should read, respectively,

a is particular (not is a particular).
f is qualitative (not f is a quality).
p is propositional (not p is a proposition).

But once this is appreciated, one can allow the surface grammar of such categorizing statements to be that of ordinary classificatory statements.

In the above framework,

The pawn is a particular


(x) x epsilon pawn --> -y/y = y/x

(B) We can take the line I took in "Towards a Theory of the Categories." According to that analysis

--- is a particular

was construed as the material mode for (in first approximation)

. . . is a singular term

(where the subject of the former has been replaced by the corresponding explicitly metalinguistic expression). Thus

Socrates is a particular


The .Socrates. is a singular term

and hence

.Socrates.s are singular terms.

But how, on this account, are such general statements as

Men are particulars

to be construed? On the previous account there was available

(x) x epsilon Man --> -y/y = y/x

This time we need something like

For all expressions, e, if the result of concatenating e with .is a man. is a wff, then e is a singular term.

I won't attempt to tidy this up so lets see where it gets us.

Applied to

The pawn is a particular

as reducible to

Pawns are particulars

we get

For all expressions, e, if the result of concatenating e with .is a pawn. is a wff, then e is a singular term.

Applied to

(The state of affairs) that fa is a particular

we get

For all expressions, e, if the result of concatenating e with .is an [.f.].a.. is a wff, then e is a singular term.

For example, I shall write a token of [.f.].a. and call it Tom



Tom is an [.f.].a. is not only a wff, but true, and, indeed, 'Tom' is a singular term (as is the token on line 19 of this page). (i.e. fa]

I have already indicated some of the serious problems which must be disposed of before this strategy is home free. Let me add some more.

(1) What of

(The state of affairs) that aRb is an object?

The first step is obvious

The [.R.](.a., .b.) is an object

The sortal corresponding to the DST applies to any inscription in any language which does the job done in our [PMese] language by 'a's and 'b's which (in that order) have an 'R' between them. This time, however, to be an [.R.](.a., .b.) an item must be a complex object, roughly a pair of objects which satisfy a certain relation; which relation depends of course on the language to which it belongs. Now it is not customary to call complex objects or, for that matter, pairs, particulars. Indeed, many ideal language philosophers so use the term 'particular' that to speak of complex particulars is widersinnig. I do not concur, yet it is useful to distinguish between within the broad domain of objects -- which include distributive objects (e.g. the lion), average entities (the average man) and other interesting specimens -- the domain of basic individuals and composita (of which, on one interpretation, pairs, triples, etc. are examples).

Now a cat-on-a-mat is an object, though not a simple object. It is a compositum which satisfies the on-relation in the direction cat to mat. (A cat-on-a-mat is to be distinguished from -- though it is obviously related to -- a cat which is on a mat.) Granted, the object would not exist unless it was true that a cat is on a mat. But this does not entail, as noted before, that a cat on a mat is the fact that a cat is on a mat.

Thus, given a satisfactory account of pairs and their identity

(The state of affairs) that aRb is a particular (albeit a complex one)


(x,y):(x,y) epsilon [.R.](.a., .b.) --> (-u, -v)/(u,v) = (u,v)/(x,y)

It is time, now, to drop the other shoe. You must remember that in "Abstract Entities" I explained how both of the following statements could be true:

(1) Triangularity is not an individual but an attribute.
(2) Triangularity is an individual.

The argument was that 'triangularity' is ambiguous. It can mean

The .triangular.


The .the .triangular..

Thus (1) becomes

(1-1) The .triangular. is not a ST, but a predicate.

which reduces to

(1-2) .Triangular.s are not STs, but predicates,

while (2) becomes

(2-1) The .The .triangular.. is a ST (i.e. a DST)

which reduces to

(2-2) .The .triangular..s are STs (i.e. DSTs).

Correspondingly, the following are both true

(3) That fa is not an object but a state of affairs
(4) That fa is an object.

The first of these, (3), becomes

(3-1) The [.f.].a. is not a ST, but a proposition

which reduces to

(3-2) [.f.].a.s are not STs but proposition tokens,

while (4) becomes

(4-1) The .the [.f.].a.. is a ST (i.e. a DST)

which reduces to

(4-2) .The [.f.].a..s are STs (i.e. DSTs).

How, you may ask, can I reconcile

(3) That fa is not an object but a state of affairs

with my opening claim that states of affairs are objects? The answer is that although tokens of [.f.].a. are objects, they are not objects which, considered as linguistic role players, are singular terms.

Thus we have

(5) (x) x epsilon [.f.].a. --> - y/y = y/x


(6) For all expressions, e, if the result of concatenating e with .is an [.f.].a.. is a wff, then e is a singular term

both of which tell us that [.f.].a.s are particulars. On the other hand, we also have

(x) x epsilon [.f.].a. --> ~(x epsilon ST)

which is both true and consistent with each of the above.

It only remains to be noted that the sense of

That fa is an object

which is reconstructed by (4-2) above is to be carefully distinguished from both (5) and (6) above. It has as its chess parallel

The pawn is an object.
The .The pawn. is a ST (i.e. a DST).
.The pawn.s are STs (i.e. DSTs).

In this context we should not say that the pawn is a particular, but rather that it is a funny kind of object, i.e. a distributive object.

I hope that you find the above remarks sufficiently intelligible to carry on the dialogue.



Wilfrid Sellars


The University of North Carolina at Chapel Hill
Department of Philosophy

29 August 72

Professor Wilfrid Sellars
Department of Philosophy
University of Pittsburgh
Pittsburgh, Pennsylvania, 15213

Dear Wilfrid:

I see that you will be visiting with us at colloquium time. Great! My seminar on your work should be far enough along by then to have my students crying for a chance to ask you some questions. Me too, probably.

Thanks for the papers, which I am devouring, and the letter. I don't think that I've ever received a 12-page letter with footnotes before! I hope that my response (which follows immediately) will be intelligent enough to justify your obvious effort. To work:

I think that I understand now what you're trying to do, and I'd like to engage your views fairly far along, having no quarrel with most of your opening moves as such. Let me first, however, say a few things to let you decide whether I've, in fact, understood you.

1. I will need a token of [.f.].a. to talk about. Let me produce one:


Since you've used 'Tom' for one of its relatives, we'll call this one 'Jim'.

2. Now, as I understand you, the first sentence of the last paragraph is, strictly speaking, ungrammatical. In particular, the expression "a token of [.f.].a." (which you also use, on page 7 of your letter) seems to have incoherent syntax. On your view,

(1) Jim is an [.f.].a.

is a true wff, paralleling in grammar

(2) Fido is an orange dog.

Since it seems correct to say that

[.f.].a.s are tokens,

employing the analogy with

Orange dogs are animals,

the expression "a token of [.f.].a." turns out to have the syntax of "an animal of orange dog", which won't do.

3. Of course there's no problem understanding what's intended:

a token which is an [.f.].a.

(cf. "an animal which is an orange dog"). I go into the point only to let my misunderstandings about your grammar, if any, surface early.

4. Assuming that I'm OK so far, however, then, since the following transformations are permitted in the case of (2):

Fido is a dog which is orange,
Fido is a dog & Fido is orange,

I take (1) to entail:

Jim is an .a. which is [.f.]

and, therefore

(3) Jim is an .a. & Jim is [.f.].

5. This seems to mesh well with the schema in "Naming and Saying" that I was puzzling over in my long essay--your replacement for Wittgenstein's "Fact pictures fact":

Natural linguistic objects O'(1), O'(2), . . . O'(n) make up a picture of objects O(1), O(2), . . . O(n) by virtue of such and such facts about O'(1), O'(2), . . . O'(n).

6. I take it that an instantiation of this schema in the present instance would be something like

The .a. on line 3 of paragraph 1 (viz, Jim) is a picture of a by virtue of the fact that it (Jim) is [.f.].

Now we shall need to do something here about "by virtue of the fact that . . .", but your strategy seems clear enough. It means, I would suppose, something like "since it is true that. . .", yielding

Jim is a picture of a since it is true that Jim is [.f.].

And, moving down out of the metalanguage, we arrive at something like:

(4) Jim is a picture of a because Jim is [.f.].

7. Thus my reading of your late syntax seems to fit the pattern of your early schema. But (4) really puzzles me. I would expect to find, not (4), but something like

(5) Jim is a picture of a because Jim is an .a. and since, in our linguistic community, .a.s stand in certain C-H-L relations to a.

Let me drop this line of thought for the moment, however, and return to spelling out the implications of (3).

8. The first observation to make is that you are quite right to hold that Jim is a particular. For if we ask what kind of thing Jim is, the answer, by (3), has to be that Jim is an .a. and, since .a.s are, uncontroversially, particulars (natural linguistic objects), it follows that Jim is a particular.

9. But it also seems to follow that, since Jim is an .a. and since .a.s are singular terms, Jim is a singular term, and this seems wrong.

10. I have, of course, no quarrel with the conclusion that 'Jim' (note the quotes) is a singular term (cf. your conclusion at the bottom of page 7), but my conclusion that Jim (note the absence of quotes) is also a singular term is a different one, and more bothersome. The crucial point is that it seems to run directly counter to your claim on page 10 that

(3-2) [.f.].a.s are not STs but proposition tokens,

for Jim is an [.f.].a. and, as I've just argued, Jim is a ST.

11. Now there seem to be only two ways of meeting the argument in 9 above. The first is to deny that .a.s are singular terms. This, I think, you would be reluctant to do. The second is to deny that Jim is an .a.. He is rather, an [.f.], and, thus, that .a.s are Sts does not entail that Jim is a ST. But, unless I've radically misunderstood your syntax, this won't do either, for, as I read you, [.f.].a.s are .a.s, as orange dogs are dogs.

12. Now one thing I was looking for in my long essay was precisely a way of being entitled to your (3-2). I would be entitled to it, if, ala Wittgenstein, STs were natural linguistic objects and proposition "tokens" were natural linguistic facts, and there was a categorial ontological distinction between objects and facts, so that facts are not objects.

13. Like you, I am now convinced that Wittgenstein's strategy for defending (3-2) or its like won't work. But my reason is, I guess, not the same as yours. I presuppose one of Wittgenstein's moves (that facts are not objects) and use it to attack the other (that propositions are facts). I read "Facts are not objects", crudely, as "What can be stated cannot be named" and conclude that, if propositions were facts, a metalanguage would be impossible, since we must name (i.e., refer to ) propositions in any metalanguage. It isn't completely clear in my mind why you reject Wittgenstein's view that propositions are facts, but it is clear that it can't be for my reason, for you reject also what I use as a premiss, the claim that facts are not objects.

14. In any case, we are agreed that Wittgenstein's line of defense for your (3-2) won't work. Failing that, however, we need another line of defense, and what I've been suggesting is that if one adopts your view that facts are objects, there isn't going to be one. (Not, of course, that I have one to haul out of my pocket on my premisses yet.)

15. I guess, recurring to your letter, what I really need to ask you for on this point is

(a) an elucidation of the last sentence on your page 10: ". . . Although tokens of [.f.].a. are objects, they are not objects which, considered as linguistic role players, are singular terms."


(b) an explanation of 'singular term' which shows how (as you claim on page 11)

(x) x epsilon [.f.].a. --> -y/y = y/x


(x) x epsilon [.f.].a. --> ~(x epsilon ST)

can be compatible with each other, for I would have thought that

(x) x epsilon ST <--> -y/y = y/x.

(Hence it's Frege who haunts me in paragraph 33 of my long essay, since he finds 'p = q' well-formed.)

16. I suspect that your reply to (a) and (b) is going to turn on precisely the matters I'm muddled about in 5 through 7 above, but, to return there for a minute, if one holds, as you do in N&S, that what pictures picture is objects and not, as Wittgenstein would have it, facts (implicitly enforcing some sort of object/fact distinction there, by the way), then that Jim is [.f.] really seems to have nothing much to do with what it is that Jim pictures. I should like to say that that Jim is [.f.] is relevant, not to what Jim pictures but to what Jim pictures it as. Thus Jim pictures a as f by being [.f.]. But this looks suspiciously like the "Fact pictures fact" model which you reject:

that Jim is [.f.] says (shows? pictures?) that fa.

I shall wait to be straightened out on all this.

Well, I think I've probably said more than enough to allow you to pinpoint major areas of confusion and conflict. Getting all this down on paper is certainly doing wonders for my thinking on these difficult matters, so I do hope that you'll find something here worth replying to. I'd like to carry on.

Hope your new academic year is a pleasant one. See you in October.

All my best,


Jay Rosenberg


September 5, 1972

Dear Jay:

Many thanks for the long letter, the promise of which motivated my own effort. I reply at once in order to strike while the iron is hot, or, as I see it, the main question you now press can be answered by drawing a distinction which has always been implicit in my analysis, but to which I have failed until recently to give formal recognition. I first became aware of the shortcoming in question in the course of a seminar I was giving at UMass this spring. A student asked me: "How can you say that an [.f.].a. is an INDCON, when it is a PROP? Surely propositions are radically different from names.!" To which I answered, of course, that [.f.].a.s are INDCONs, but not mere INDCONs, such as occur in a list, for they are INDCONs which, in addition to playing an INDCON role, and, in so doing, refer to a, are also, by virtue of being concatenated an .f., playing the (atomic) propositional role of picturing a as f. And, indeed, the distinction which is needed to answer your question is exactly that between being a mere INDCON (as in a list) and being an INDCON which is also a PROP, e.g. an [.f.].a.

But let me get down to cases. I shall comment on your letter, paragraph by paragraph as you have numbered them:

Ad 1. Hello Jim.

Ad 2. As you yourself go on to point out, the context in which I am interested is not

Jim is a token of [.f.].a.


Jim is an [.f.].a.

I do speak informally of tokens of [.f.].a., however, and have not exclusively commented on the grammar of 'token.' Now that you press me, l suggest that

. . . is a token of - - -

is a special case of

. . . is a member of - - -


Jim is a token of [.f.].a.

stands to

Jim is an [.f.].a.


Socrates is a member of mankind

stands to

Socrates is a man

In other words, token-talk, like member-talk is one level up the semantical hierarchy from ordinary subject-predicate talk. Using it causes no trouble until we look at it instead of through it. From my point of view,

Jim is a token of [.f.].a.

has the form

Jim is a token of [.f.].a.-kind

and, made fully explicit, has the form

The INDCON^.is an [.f.].a.. is true of .Jim.

which entails

The .Jim is an [.f.].a.. is true

and carries us, via the truth move, to

Jim is an [.f.].a.

which is the way to talk about Jim in the context of our problem.

Ad 4. No comment.

Ad 5. No comment.

Ad 6. No comment.

Ad 7. We must distinguish between the C-H-L relations by virtue of which Jim refers to a, and the C-H-L relations between [.f.] INDCONs and f objects by virtue of which Jim, as an [.f.] INDCON, pictures a as f. Your (5) runs these together, i.e. runs together that which makes .a.s the linguistic representatives of a, and that which makes relevantly configured .a.s pictures of a. [See my Russell paper, pp. 51 ff.]

Ad 8. No comment.

Ad 9. Here comes the crunch. I am indeed committed to the following: .a.s are STs and Jim, being an [.f.].a., and hence a .a., is a ST.

Ad 10. My introductory paragraph should have prepared you for what is coming. I abandon my claim on p. 10 in favor of the revised claim that

(3-2)-R [.f.].a.s are not mere STs but PROPs

(I drop the word 'tokens' for reasons indicated in Ad 2. above.)

Ad 11. I have, I believe, escaped between the horns of your dilemma.

Ad 12. No comment.

Ad 13. No comment.

Ad 14. No comment.

Ad 15(a). The last sentence on p. 10 of my previous letter, which you quote, should be modified to accord with (3-2)-R as follows:

Although tokens of [.f.].a. are objects, they are not objects which, considered as linguistic role players, are mere singular terms.

Ad 15(b). Correspondingly,

(x) x epsilon [.f.].a. --> ~(x epsilon ST)

should read

(x) x epsilon [.f.].a. --> ~(x epsilon mere ST)

I am puzzled by your

(x) x epsilon ST <--> -y/y = y/x

for obviously it is false that

(x) -y/y = y/x --> x epsilon ST

since not all particulars are even linguistic, let alone singular terms. I suspect that you are confusing

(x) x epsilon ST <--> -y/y = y/x


(x) x is particular <--> -y/y = y/x

which is, on my "object language" analysis of the categorizing expression 'particular,' a definitional truth. A comment on 15(b) insofar as it concerns Frege's ghost will come at the end, because at this point it would constitute a lengthy digression.

Ad 16. It should now be clear from what I wrote in Ad 7., that I quite agree with you that

. . . that Jim is [.f.] is relevant not to what Jim pictures [i.e. the object of which Jim, as being an .a. is the linguistic representative], but to what Jim pictures it as.

You sum this up by saying that

Jim pictures a as f by being [.f.]

and express the suspicion that this takes us back to the "Fact pictures fact" model, which I have rejected. Once again I refer you to the passage in the Russell paper in which I address myself to this issue. But I really believe that it is all in N&S and T&C -- though, perhaps, somewhat in the matter in which the oak is in the acorn.

Ad Frege's ghost. According to my PMese lights

p = q

where 'p' and 'q' represent object language sentences, says no more, and no less, than

p <--> q

[in PM it is provable that p Eq q .<-->. p <-- > q, where 'Eq', to speak informally, is reflective, symmetrical and transitive.] Thus I would not object to introducing the categorial predicate 'is propositional' as follows

p is propositional =df -q/q <--> q/p

or, to use variables which don't seem to beg the question,

alpha is propositional =df -beta/beta = beta/alpha{8}

But, of course, the context

-beta/beta <--> beta/Socrates

would, be ill-formed, as would

-beta/beta = beta/snow is white

given the Leibnitz-Russell definition of '=.' For the latter would entail

Snow is white = snow is white

and this, unless it is construed as

Snow is white <--> snow is white

(in which case, caedit quaestio), would expand into

(F) F( snow is white) <--> F(snow is white)

If we construe 'F' as a variable for genuine predicates, which is implied by extending the L-R definition of identity to this context, the expression is ill-formed, since 'Snow is white' is not a singular term. If, on the other hand, we construe 'F' in terms of contexts (open sentences) in which sentences (as opposed to singular terms) can occur, then

Snow is white = snow is white

would turn out, as before, to be a rewriting of

Snow is white <--> snow is white

and not to be an identity statement at all.

But all this, of course, simply reflects my commitment to a Tractarian interpretation of PM. It belongs here simply as a way of exorcising some appearances of Frege's ghost, and, less metaphorically, as a warning against a piecemeal introduction of Fregean considerations into contexts governed by PM-Tractarian assumptions.

Well, this was written in haste and there will certainly be some repenting at leisure. But I hope it achieves its immediate purpose, which is to carry our dialogue a step further before I am engulfed by the term (and so much else). See you next month.



Wilfrid Sellars



28 September 72

Dear Wilfrid:

1. Thanks for the super-quick response to my last letter. Would that I could be as prompt, but I haven't yet achieved Distinguished Professorial leisure, and continued upward mobility is time consuming. Thanks, too, for "Reply to Quine" which I will digest at my earliest opportunity ((probably July of 1978). For all that, I do have a few thoughts which can't wait, so here they are:

You're quite right to be puzzled by my

(x) x epsilon ST <--> -y/y = y/x

2. I suspect that my reasoning (what there was of it) involved something like:

(x) [x epsilon ST <--> (z)(x represents z <- -> -y/y=y/z)]

(i.e., All and only singular terms represent particulars) and that the 'x represents z' part got lost. But that's just a guess. What I wrote was certainly false.

3. One thing I'd like to chew on a bit in your latest remarks occurs on pp. 4-5 in "Ad Frege's ghost". There occurs the following argument (I put 'fa' for 'Snow is white'):

The context -beta/beta = beta/fa would be ill-formed, given the Leibnitz- Russell definition of '='. For the latter would entail

fa = fa

and this, unless it is construed as

fa <--> fa

would expand into

(F) F(fa) <--> F(fa)

If we construe 'F' as a variable for genuine predicates, which is implied by extending the L-R definition of identity to this context, the expression is ill-formed, since 'fa' is not a singular term.

Well, what am I to make of "since 'fa' is not a singular term", given all that has gone before? My natural inclination is to read it as:

[.f.].a.s are not singular terms.

But that can't be right, [.f.].a.s are singular terms (though not mere singular terms). Still, I can't think of any other plausible reading of " 'fa' is not a singular term" and, that being so, your argument is not complete and Frege's ghost refuses to lie down and be quiet.

4. But still bigger games are afoot. Let me explain. In section VII of "Abstract Entities" (AE) you pose the question

Are there any abstract entities which are not objects?

and answer that of course there are abstract entities which are not objects. Again, on page 68 of "Toward a Theory of the Categories", you argue that it would be correct and non-paradoxical to say that there are entities which are not individuals.{9} I used to be convinced that you'd successfully established these claims. In terms of recent developments, I now think that you haven't-- and that they're false.

5. The example you give in AE are, roughly these:

(a) Lionkind is a kind and not an individual

(i.e. (a1) The .lion. is a common noun and not a ST) and

(b) Triangularity is a quality and not an individual

(i.e. (b1) The .triangular. is a predicate and not a ST).{10}

The problem is created by the fact that you address these questions before you make good the "still more basic oversimplification" which comes from not considering languages of the Jumblese sort. When the smoke from that fire has cleared, the reconstruction of 'triangularity' is no longer 'the .triangular.' but rather 'the [.triangular.] INDCON'.{11}

6. Similarly, drawing on "Classes as Abstract Entities . . .", the purified reconstruction of 'lionkind' will not be 'the .lion.' but rather

the [.epsilon(1) lion.] INDCON

If this is right, then (a1) and (b1) above become:

(a2) The [.epsilon(1) lion.] INDCON is a common noun and not a STs

(b2) The [.triangular.] INDCON is a predicate and not a ST.

And these unpack, as a first step, into

(a3) [.epsilon(1) lion.] INDCONs are common nouns and not STs

[b3) [.triangular.] INDCONs are predicates and not STs.

7. Well, you can see what's coming:

[.epsilon(1) lion.] INDCONs are INDCONS
Ergo, [.epsilon(1) lion.] INDCONs are STs

and, similarly,

[.triangular.] INDCONs are INDCONs
lNDCONs are STs
Ergo, [.triangular.] INDCONs are STs

The conclusions transcribe the material mode claims: "Lionkind is an object" and "Triangularity is an object", and are inconsistent with (a3) and (b3).

This I shall interpret as the cash value of the ancient maxim that everything which is, is particular.

8. Of course, [.epsilon(1) lion.] INDCONs and [.triangular.] INDCONs are not mere INDCONs and, hence, not mere STs. But I don't think that we can save 'kind' and 'quality' here. For if we ask what else, besides STs, [.epsilon(1) lion.] INDCONs and [.triangular.] INDCONs are, the answer it seems to me, has to be:

[.epsilon(1) lion.] INDCONs are not mere STs but PROPs


[.triangular.] INDCONs are not mere STs but PROPs

These are the formal mode counterparts of "Lionkind is a state of affairs" and "Triangularity is a state of affairs". Both, of these look wrong.

9. Well, time for some morals. What I take these animadversions to show, among other things, is that the correct transcription of

Triangularity is a quality

is not

The [.triangular.] INDCON is a predicate

(which is false), but rather, the original

The .triangular. is a predicate

which is true. To speak of triangularity, in other words, is--as I see it now--to make reference to a linguistic role which can only be filled by auxiliary signs. (While Jumblese contains [.triangular.] INDCONs, it contains no .triangular.s.) It is this essentially auxiliary character of .triangular.s, and not the non-illustratingness of "[.triangular.] INDCON" which becomes the truth behind Frege's insight that functions are essentially ungesättigt

10. Notice that, in our account of the semantics of language in general,-- the account which covers both PMese type and Jumblese type languages--both 'common noun' and 'predicate' as linguistic role designators have disappeared. They turn out to be parochial linguistic roles rather than essential linguistic roles. And this, surely, is as it should be, for Jumblese contains neither predicates nor common nouns. We wind up with two and only two basic semantic categories: INDCON and PROP. This is quite consistent with the Tractatus. In fact, as I read it, it is the Tractatus. Wittgenstein's semantics includes only 'name' and 'proposition' as basic categories.

11. This way of looking at things entitles us to say

1. There are abstract entities which are not particulars

(because, although [.triangular.] INDCONs are STs, .triangular.s are not); gives us an interpretation of

2. There really are no abstract entities

(because the only basic semantic categories are INDCON and PROP); and commits us to

3. Everything which really is, is particular

(because PROPs are STs, though not mere STs)

12. Now I'm not yet sure that I'm entirely happy with

PROPs are STs (though not mere STs)

and the resulting commitment to

Everything which really is, is particular

especially with Frege's ghost still lurking in the shadows, but I'm going to let these go until I have your reactions to what I've done in this letter. I guess it's heretical enough to warrant some comment.

13. See you in a couple of weeks. By the way, is there any chance that you could arrive here a day early or leave here a day late for Colloquium, so that my seminar could have a few hours discussion with you? Your honorarium would be adjusted accordingly.

Anxiously awaiting your next letter.



Jay Rosenberg

P.S. I'm planning on applying for a NEH Summer Fellowship and they require three references. May I send you a form?


January 16, 1973

Dear Jay:

I have been pondering long and hard about the questions you raise in your letter of 28 September. While they have not led me to change my views on any point of substance, they have convinced me that the way in which I have formulated them in print is in certain key respects inadequate and misleading. Above all they have forced me to clarify my views on the status of qualities, kinds and, last but not least, states of affairs. I will not attempt to summarize the outcome, but will let it emerge, such as it is, in my remarks on the specifics of your letter. I will avail myself of the numbered breakdown on which we agreed during my stay in Chapel Hill.

Ad 1. I hope I can get some reaction to this letter in the not too distant future, preferably before "July of 1978". I begin to hear "Time's winged chariot" loud and clear, and would like to get these things straight before it catches me. Some more "few thoughts which can't wait" would keep the dialogue rolling. This letter has turned out to be monstrously long, but not quite as long as it looks, since it is quite repetitious.

Ad 2. No comment

Ad 3. Things are beginning to warm up; but why doesn't the distinction between STs and mere STs provide the answer? Surely what combines with predicates to form statements are mere STs. This would rule out, as intended,


Ad 4. This section sets a general theme which recurs throughout the remainder of your paper. Since it contains no argument, I will not discuss it here. I will, however, return to it after commenting on your criticism of my treatment of qualities and kinds and its compatibility with the idea that they are examples of entities which are not objects.

Ad 5. We are now in the kitchen when, in the latter part of AE, I changed from equating Triangularity with the .triangular. to equating it with the [.triangular.] INDCON, my aim was to interpret Triangularity as something that could be expressed in Jumblese as well as PMese languages. Since Jumblese contains no predicates,{12} it obviously contains no .triangular.s. Thus, the metalinguistic sortal,

[.triangular.] INDCON

as used in this connection, could no longer imply that the items to which it applies consist of a .triangular. and an INDCON. It was now (informally) characterized as a common noun which applies to expressions in any language, whether Jumblese or PMese, which do the job done in our PMese language by INDCONs which are concatenated with a 'triangular,' and in other PMese languages by INDCONs which are concatenated with auxiliary expressions which function as does our 'triangular.' My reason for shifting to

.[*triangular*]. INDCON

in my "Reply to Quine" was to stress the auxiliary role of the design *triangular* in our PMese language. However, in using this new mode of representation, I should have remembered that when I first introduced the bracket notation I placed them around adjectives to indicate that they were being used to form a common noun out of a common noun; thus the brackets in

[white] dog

represented that the whole expression is a common noun formed from the common noun 'dog' and the adjective 'white,' and has the sense expressed by the phrase

dog which is white.

However, when I used this device in metalinguistic contexts I soon began, without explicit awareness of what I was doing, to construe the brackets as implying concatenation. According to this line of thought

[*triangular*] 'a'

would apply, by virtue of the brackets , to 'a's which are concatenated with a *triangular*. In this respect, the brackets played the role of the standard symbol for concatenation, '^'. The result was a sad muddle.

The above explains (without justifying) my placing the dot quotes outside the brackets in

.[*triangular*]. INDCON

for I wanted this to apply to Jumblese INDCONs which, though not concatenated with an auxiliary expression corresponding to *triangular*, nevertheless were the Jumblese counterparts of expressions in our language which consist of INDCONs concatenated with a *triangular.*

Clearly I must straighten out my notation. I have ideas about how to do this, largely based on my study of recent theories of grammar, but they will have to wait until another occasion. In the meantime, I will return to my initial use of brackets and informally characterize

[.triangular.] INDCON

as a metalinguistic sortal which applies to

INDCONs which have a character by virtue of which they function as do INDCONs which are concatenated with a .triangular. in PMese languages, and, in first instance, as do INDCONs concatenated with an *triangular* in our PMese language.

Obviously this puts a tremendous informal burden on the expressions inside the brackets. Far too much metalinguistic information is left without explicit representation; or, to put it somewhat differently, the dot quotes are playing a much richer role than that of forming a common noun which applies to inscriptions in any language which do the job done in our language by the inscription which appears between the dots.{13}

But enough of this brooding about a perspicuous notation. Without further ado I shall follow your example and use

[.triangular.] INDCON

in the intended generic sense in which it applies to PMese and Jumblese items alike.

Ad 6 and 7. You are quite right to insist that if, after reinterpreting Triangularity as the [.triangular.] INDCON, I had continued to assert both of the following

(a) Triangularity is a quality
(b) The context '--- is a quality' is the material mode for (roughly) the context '.... is a one place predicate'{14}

I would have been committed to the nonsense

(c) The [.triangular.] INDCON is a predicate.

Also, since, to use your example, Jumblese *a*s are [.triangular.] INDCONs, and hence, on this new account, express Triangularity, by accepting (c) I would have been committed to the absurdity that in the Jumblese language in question

The 'a' is a predicate.

Now, it is quite true that after offering the above reinterpretation of Triangularity in the latter part of AE, I did not discuss its consequences for the earlier analysis of the categorizing statement, 'Triangularity is a quality.' Yet I would certainly have refused to pair this statement with

The [.triangular.] INDCON is a predicate

and I clearly equated Triangularity as reinterpreted, with

that something is triangular

which places it somewhere in the category of states of affairs. Exactly where, I left rather up in the air, though I did give the essential clues, which I shall spell out shortly. The fact remains, however, that in subsequent papers, however, I continued to "oversimplify" (as I put it in AE) and to equate Triangularity with the .triangular,. and to pair

Triangularity is a quality


The .triangular. is a (certain kind of) predicate.

Obviously, therefore, I must face up to the questions,

(1) Is it correct to construe Triangularity as the [.triangular.] INDCON?

(2) If so, what sense can be made of 'Triangularity is a quality'?

As I see it, I am faced with the following alternatives:

(a) I can continue to construe 'is a quality' as material mode for 'is a (certain kind of) predicate,' in which case qualities would be parochial, and it would be incorrect to construe Triangularity, for example, as the [.triangular.] INDCON.

(b) I can reconstrue 'is a quality' to fit the reinterpretation of Triangularity as the [.triangular.] INDCON. I can do this by taking 'quality' to be the material mode for something like 'is a characterizing PROP.' In this case, both PMese and Jumblese would contain expressions which stand for qualities.

Of these two courses, (a) is the one you recommend. I have no strong objections to it, since the point l wanted to make, mainly that we can define a sense in which a Jumblese expression can pertain to Triangularity, or, more generically, a quality, can be made in a way which is compatible with (a). On the other hand, the simplest way of explaining how a Jumblese expression can pertain to a quality, e.g. Triangularity, would be to adopt alternative (b).

The important thing to see is that on alternative (a) there is, in addition to 'Triangularity is a quality' another, but non-parochial categorizing statement, which is intimately related to it, and which alternative (b) construes as synonymous with it. This non-parochial statement can be unearthed by taking seriously the idea that [.triangular.] INDCONs, being PROPs -- which of course they are -- have as their material mode of speech counterparts, states of affairs. Thus the material mode of

--- is a characterizing PROP

would have the form

.... is a ø state of affairs.

What might we substitute for 'ø'? I see no reason for not using 'qualitative,' provided that it is clearly understood that although the root of this term is (on alternative (a)) parochial, it would now be used non- parochially to apply to states of affairs which, though they can be expressed in Jumblese as well as PMese, are expressed in PMese by a concatenation of INDCONs with a predicative expression which stands for a quality -- as contrasted, for example, with one which stands for a kind.

In first approximation, to make this move is to pair the categorizing statement

That something is triangular is a qualitative state of affairs


The [.triangular.] INDCON is a characterizing PROP

In section 7 you correctly point out that

[.triangular.] INDCONs are not PREDs but INDCONs

so that if Triangularly is construed as the [.triangular.] INDCON, it would be an object, whereas on my original interpretation, which construed it as the .triangular., it would not be an object. After pointing out that on my second interpretation, Triangularity and Lionkind turn out to be objects, you characterize this as

... the cash value of the ancient maxim that everything which is, is particular. (Ital. JR.)

Though you do not explicitly make the connection, I take it that you are referring back to section 4, and arguing that when my analysis is spelled out it is inconsistent with the claim that there are entities which are not objects.

Now I assume that by italicizing the 'is,' you mean to isolate a philosophical sense of 'is' in which not everything which is, is.{15} Perhaps, then, you are prepared to admit that there is such an entity as Triangularity, which on your construction (as the .triangular.) is not an object, while denying that there is such an entity. But what makes you think that when I claimed that there are entities which are not objects, I meant that there are entities which are not objects?{16} I shall return to this topic shortly. For the moment I simply note that although the context makes it look as though it is because Triangularity (as the .triangular.) is parochial that you deny that it is, it later becomes clear that this is not your reason. Being parochial turns out to be, as it should, a sufficient but not a necessary condition for correctly denying that something really is. Thus, presumably, Disjunction (as the .or.) and Two (as the .(E2x) x epsilon(1) .KIND) would be entities which are, but which really are not. Furthermore, they are entities which are not objects.

Ad 8. You point out that if Triangularity is construed as the [.triangular.] INDCON and Lionkind as the [.epsilon(1) lion.] INDCON then, since

[.epsilon(1) lion.] INDCONs are not mere STs but PROPs


[.triangular.] INDCONs are not mere STs but PROPs

we would have, in the material mode,

Lionkind is ... a state of affairs
Triangularity is ... a state of affairs.

"Both of these," you write, "look wrong." You then go on in section 9 to draw the "moral" that "the correct transcription" of 'Triangularity is a quality' is ... the original 'the .triangular. is a predicate'."

Now, while I share your intuition about the oddness of

Triangularity is a state of affairs

I think I can account for it in a way which reconciles it with the interpretation of Triangularity as the [.triangular.] INDCON. The basic point is that what belongs in the context

--- is a state of affairs

is a that clause which results from applying 'that' to a (descriptive) statement, whether atomic, molecular or quantified. Thus we have

That a is triangular is a state of affairs
That a is triangular or b is triangular is a state of affairs
That something is triangular (i.e. that (Ex) x is triangular) is a state of affairs.


That something is triangular

as it occurs in the context

That something is triangular implies that it is trilateral{17}

is not the result of applying 'that' to a statement. It is the material mode for (in first approximation){18}

The [.triangular.] INDCON.

Thus, while

That something is triangular (i.e. the .(Ex) x is triangular.) is a state of affairs

is unproblematic,

That something is triangular (in the second sense) is a state of affairs

is incorrect, and must be replaced by

That something is triangular is a kind of state of affairs

which (again in first approximation){19} has as its formal made counterpart

The [.wise.] INDCON is a kind of PROP,

There are a number of points to be noted here. In the first place there are at least two ways in which there are kinds of PROP. (1) There are sub- categories of PROP, e.g. atomic, molecular, general, etc. Thus we have

The [.triangular.] .a. is an ATPROP
The .a is triangular or b is triangular. is a MOLPROP
The .(Ex) x is triangular. is a GENPROP

(2) Corresponding to the many states of affairs which have in common the fact that each consists in some particular thing's being triangular, thus

That a is triangular, that b is triangular, ...

there are many PROPs which have in common the fact that each is a [.triangular.] INDCON(i), for a particular value of 'i.' It is the latter sense in which there are kinds of PROP which concerns us here.

Consider, now, the following:

(a) That a is triangular is something's being triangular
(b) Something's being triangular is a kind of state of affairs.

The first of these can be quite adequately construed as the material mode for

The [.triangular.].a. is a [.triangular.] INDCON

But what of the second? The previous paragraph suggests that in some contexts the phrase 'something's being triangular' is to be construed as containing a variable (a metalinguistic variable), as contrasted with mentioning one (an object language variable) as does 'that (Ex) is triangular.' This suggests that (b) has as its formal mode counterpart

(For all i,) the [.triangular.] INDCON(i) is a kind of PROP

which reduces to

(For all i,) the [.triangular.] INDCON(i)s are a kind of PROP


[.triangular.].a.s are a kind of PROP{20}

Notice that, as usual, the plural which occurs in the formal mode of speech, when we make the move from statements about (for example)

the .or.

to statements about


does not occur in the material mode of speech. Material mode singular terms (e.g. Triangularity, that Tom is tall, etc.) yield plurals only when translated into the formal mode. Thus, to get

[.triangular.].a.s are a kind of PROP

into the material mode, we must first go to

The [.triangular.].a. is a kind of PROP

which yields

That a is triangular is a kind of state of affairs.

Perhaps we can now understand why

Triangularity is a state of affairs

"looks wrong." For while, according to the above analysis, Triangularity belongs to the category of states of affairs, the correct formulation would have to be

Triangularly is a kind of state of affairs.

Equating, as I proposed in the latter part of AE, Triangularity with that something is triangular, our categorizing statement translates into the formal mode as

(i) The [.triangular.] INDCON(i) is a kind of PROP.

Notice that to make this move we have had to reconstruct triangularity not as

The [.triangular.] INDCON

but as

the [.triangular.] INDCON(i).

The former contains the common noun 'INDCON,' the latter contains the common noun variable 'INDCON(i)' which takes as substituends specific common nouns, e.g. '.a.' where

The .a. is an INDCON


.a.s are INDCONs.

Ad 9. The "essentially auxiliary character of .triangular.s" is, of course, the key to the account of predication I have been stressing since "Naming and Saying." As auxiliary symbols, predicates have meaning only by virtue of the fact that by giving INDCONs the character of being concatenated with them, they turn them into PROPs. It is, indeed, as you say, the essentially auxiliary character of predicates which is "the truth behind Frege's insight that functions are essentially ungesaettigt" -- but remember what a miscellany Frege includes among functions (e.g. logical connectives). What baffles me is what led you to think that on my view it is "the non-illustratingness of '[.triangular.] INDCON' " which corresponds to this insight.

Ad 10. Again I can only say that of course both .epsilon(1) lion. and .triangular.s play "parochial linguistic roles rather than essential linguistic roles." But what you must not overlook is that while Jumblese contains neither adjectival predicates nor common nouns, it does contain statements which say how an individual is as contrasted with what it is. It enables the expression of both qualitative and sortal states of affairs. It contains no count nouns, but it does contain enumerative statements.

The more important claim you make in 10 is that "we wind up with two and only two basic semantic categories: INDCON and PROP." This claim which you elaborate in 11 is clearly intended to tie in with the theme you announced in 4, and a discussion of it will give some measure of unity to this sprawling letter.

Ad 10 and 11. What are we going to count as semantic categories? And what is a basic semantic category? Are 'connective ' and 'quantifiers' semantic categories? Clearly they are categories, and clearly connectives and quantifiers are subject to translation, and have senses, i.e. are proper subject matter for the context

--- means .....
--- expresses the sense .....

yet they do not, as such, correspond to anything in the world, though statements involving them do. I can understand why you would want to say that 'connective' and 'quantifier' are syntactical rather than semantical categories, whereas 'proposition' and 'singular term' are semantical (as well as) syntactical categories. Provided that one recognizes (pace Quine) that the theory of reference is a branch of the general theory of meaning, the terminology can do no harm. You are quite right to point out that Wittgenstein held that only INDCONs and PROPs (in the first instance ATPROPs) correspond to anything in the world. Thus we can agree that "Wittgenstein's semantics includes only 'name' and 'proposition' as basic categories."

The strategy of my treatment of abstract entities has always been to pour nominalistic wine into platonistic bottles. In this sense I have argued for several decades that although there are abstract entities, there really are no abstract entities. My initial insights were, at best, fragmentary, but they all finally crystallized into the analysis developed in AE. Since I hold that while there are abstract entities, there really are no abstract entities (though you seemed to have missed this), I cannot object to the core of your section 11. The reason you give for saying that "there really are no abstract entities," namely

... because the only basic semantical categories are INDCON and PROP.

strikes me as not so much wrong as misleading. One would almost expect you to conclude that there really are both particulars and (atomic) states of affairs, both INDCON and ATPROP being "basic." One reason why you don't might have been that you don't think of atomic states or affairs as abstract entities. Your actual reason, of course, is that although atomic states of affairs are abstract entities, they are, after all, INDCONs -- though not mere INDCONs.{21}

Among the abstract entities I was considering were states of affairs, thus

That a is triangular


That a is triangular is a state of affairs

is the material mode for

the [.triangular.].a. is a PROP


[.triangular.].a.s are PROPs

And when I argued that while there are states of affairs, there really are no states of affairs, it was on the ground that statements about states of affairs are paraphraseable by statements about conceptual tokens. And in general, my reason for saying that everything which really is, is particular, was not that all conceptual items which directly represent something in the world are basic singular terms, even atomic propositions being basic singular terms' (though not mere basic singular terms), but rather that all abstract entities, including states of affairs, turn out to be distributive objects, the putative names of which are DSTs. Thus the state of affairs

that a is triangular

turns out to be

the [.triangular.].a.

and is reducible to


In other words abstract singular terms turn out to be metalinguistic predicates.

Let me spell this out more precisely, so that its implications will be clear. The material mode existence statement

There are states of affairs, e.g. that a is triangular

translates into

There are PROPs, e.g. the [.triangular.].a.


The [.triangular.].a. is a PROP

which is the formal mode counterpart of a statement which has the surface grammar of singular subject predicate statement, namely

That a is triangular is a state of affairs

reduces to

[.triangular.].a.s are PROPs.

From this point of view, the statement

There are states of affairs

which appears to have the form

(Ex) x is a state of affairs

turns out, when translated into the formal mode, to have the form, not as one might expect

(Ex) x is a PROP

but rather

(EK) Ks are PROPs

which involves E-quantification of a metalinguistic predicate variable and doesn't assert the existence of PROPs. To do the latter we must make the additional step of asserting

(Ex) x is a PROP

which tells us that there are propositional tokens.

This can be summed up by saying that the idea that there "really are" no abstract entities amounts to the idea that abstract singular terms dissolve into metalinguistic predicates which are true of concreta.

But although in this sense, there "really are" no states of affairs, we can draw another distinction between abstract entities which "really are," and others which "really are not." It is this distinction which, as I see it, lies behind your remarks. The theme to begin with is indispensability. For example, Negation would be indispensable whereas Triangularity (interpreted as the .triangular.) would not. Again, definable entities would be dispensable in favor of their definientia. But, to come to the heart of the matter, and focussing our attention on entities which are "in the world," we can zero in on the question with which the Tractatus opens: Can we dispense with atomic states of affairs in favor of entities which are not states of affairs. To this question the answer is clearly no. A linguistic representation of the world cannot consist of mere names, and the relevant non-(mere names) are ATPROPs. Thus, although a monadic ATPRO is a name, it is not a mere name, but a name which is ø, where 'ø' stands for a linguistically relevant characteristic which may or may not involve the use of an auxiliary expression. Thus we have

ATPROPs are BSTs (basic singular terms), but not mere BSTs; they are BSTs which are thus-and-so (e.g. concatenated with a .triangular.)

which corresponds to

Atomic states of affairs are particulars, but not mere particulars; they are particulars which are such-and-such (e.g. triangular)

Thus whereas our former line of thought led to the conclusion that there "really are" no atomic states of affairs, this second line of thought leads to the conclusion that there "really are" atomic states of affairs. But the two conclusions are quite compatible, for the latter simply amounts to the indispensability of ATPROPs in representing the world.

Ad 12. Notice that it is not just the fact that PROP is a basic semantical category on which you are relying in section 11, for it is only ATPROPs which are STs though not mere STs. Thus, to take a previous example,

.(E2x) x epsilon(1).KINDs are PROPs, but they are not STs

Two is an entity which is not an object at all, let alone a mere object.

Only in the context of the lines of thought developed above concerning entities, objects and semantic basicness, does

ATROPs are BSTs but not mere BSTs

illuminate the claim that everything which really is, is particular. Yet there is a sense in which, representing, as it does, the correct interpretation of predication, it is the keystone of the system. For it embodies the basic truth that we say how a particular, say a, is by inscribing its name in a certain 'style,' thus by inscribing a [.triangular.].a. and not by concatenating its name with the 'name' of an abstract entity. This focuses our attention on the two dimensions of matter-of-factual connection between atomic propositions and the world: (1) the connection by virtue of which the name is hooked up with a certain object, thus .a.s with a; (2) the connection by virtue of which names inscribed in a certain style, thus [.triangular.].a. are hooked up with certain objects, thus triangular objects.

One final point remains to be elaborated. I argued in Ad 3 that what combines with a PRED to form a PROP is a mere ST. It is an .a. simpliciter, and not, for example, a [.triangular.].a. which combines with a PRED token, e.g. a .red. to form a PROP. Thus

(a is triangular) is red

is ill-formed, even though it is true that

'a is triangular's are .a.s (but not mere .a.s).

In other words, a theory of material mode categories must be careful to distinguish a sense in which atomic states of affairs are objects (though not mere objects) from a sense in which they are not objects, but states of affairs.{22}

What has been stressed up until now is the by no means unimportant sense which atomic states of affairs, unlike numerical states of affairs are objects. For, as noted above, whereas

[.triangular.].a.s are PROPs by virtue of being STs (i.e. .a.s) though not mere STs

has as its material mode counterpart

That a is triangular is a state of affairs by virtue of being an object (i.e. a), though not a mere object,

the following,

.(E1x) x epsilon(1) pope.s are not STs, but PROPs

becomes, in the material mode,

That there is one pope is not an object but a state of affairs.{23}

I think that the above is relevant to the fact that you are "not quite sure" that you are "entirely happy with

PROPs are STs (though not mere STs)

and the resulting commitment to

Everything which really is, is particular

especially with the Frege's ghost still lurking in the shadows ..." To the extent that your uneasiness is due to Frege's ghost, I hope that I have laid them to rest with my remarks in Ad 3 and the above elaboration. They suggest that when, in the theory of predication, we get down to the nitty gritty of ontology, the objects of which predicates are true are objects as such. Platonists like Bergmann construe this to mean that predicates are true of bare particulars. But on a correct theory of predication, it implies nothing of the sort. All 'object as such' rules out, as pointed out above, is such strings as

(a is triangular) is red{24}

i.e. strings of the form


where 'F' represents a genuine predicate and not any old 'open sentence,' e.g. 'not ---,' for, of course,

not (fa)

is as well formed as can be. Thus, the theory of predication in question enables us to understand why

a is triangular

doesn't assert that a nexus of exemplification obtains between a bare particular and a character. There really are triangular objects (as contrasted with objects which are merely tied to another object named Triangularity). Thus, when we say that "everything which really is, is particular," this must be construed as compatible with really is, is particular," this must be construed as compatible with

There really are states of affairs.

And this can be done along the lines I sketched on pp. 14 ff. above where it was emphasized that the representation of the world by ATPROPs cannot be reduced to its representation by mere INDCONs, and that the non- (mere INDCONs) which are necessary are the ATPROPs themselves.

To sum up:

(1) There really are atomic states of affairs

is true, as the material mode formulation of the indispensability of ATROPs in representing the world.

(2) Atomic states of affairs are particulars, but not mere particulars

reminds us that ATPROPs are BSTs which are of a linguistically relevant character over and above that by virtue of which they are the BSTs they are. And, finally,

(3) There really are no states of affairs

is the material mode formulation of the fact that the singular terms which ostensibly name states of affairs turn out, in the formal mode, to be metalinguistic predicates.

Ad 13. I had a most enjoyable stay in Chapel Hill and, in particular, found the experience of meeting with your seminar both challenging and rewarding.



Wilfrid Sellars


[Contents][Return to Main Page]
{*} I wish to thank Professor Jay Rosenberg for providing me with a copy of his correspondence with Sellars, and for his permission to place it on the Internet (AC). [Back]

{1} SPR, pp. 211 ff. [Back]

{2} However, see my concluding paragraphs for a sense in which propositions are not objects.[Back]

{3} For the purposes of your question I do not need to go into subtleties about 'speaking-out-loud' and 'inner thought episodes.' [Back]

{4} I neglect Jumblese languages which do not make use of auxiliary symbols to make the .a. of a certain character by virtue of which it translates into our language by 'a's which have the character of being concatenated with a preceding 'f.' [Back]

{5} Notice that I am dealing with linguistic tokens on the inscriptional model. Things become far more complicated when we deal with linguistic tokens as pieces of verbal behavior or inner episodes. In what sense are events objects? That is a long story which calls for a long letter in its own right. [Back]

{6} Needless to say I use the Leibnitz-Russell definition of identity. [Back]

{7} Compare the Principia definition of 'x epsilon ^y (fy),' but remember that, as thus defined, 'epsilon ^y (fy)' should not be read 'is a y such that fy,' i.e. '^y (fy)' should not be construed as a sortal. [Back]

{8} The corresponding definitions of 'is particular' and 'is predicative' with neutral variables -- after all, the burden of the definition is carried by the logical forms -- would be

alpha is particular =df -beta/beta = beta/alpha alpha is predicative =df -beta/(gamma) beta gamma v ~beta gamma/alpha [Back]

{9} 'Particular', 'object', and 'individual' seem to be interchangeable in our discussions to date and in many of your writings. There are, of course, differences--though I'm not sure we'd agree completely about what they are-- but they don't seem crucial for present purposes. I'll continue to use them interchangeably as material mode counterparts of 'singular term' until it becomes important. [Back]

{10} I abstract from the distinction between distributive individuals and others, since it is not essential to what I'm up to. I can grant that triangularity and lionkind aren't distributive objects, but they may turn out to be objects for all that. [Back]

{11) I'm sticking with the bracketing conventions you use on pages 2 and 3 of your first letter (25 July). I take it that both English *triangular(a)*s and, say, Jumblese *a*s would be [.triangular.] INDCONs. In AE, it's

the .triangular. INDCON

in "Classes as Abstract Entities . . ." it's

the .triangular. [INDCON]

and, if I recall my quick and cursory reading of "Reply to Quine" it there becomes

the .[*triangular*]. INDCON

You really ought to pick one, don't you think? [Back]

{12} Predicates, as I have emphasized since "Naming and Saying," are auxiliary expressions, and are linguistic conveniences rather than necessities. [Back]

{13} For PMese languages

[^.triangular.] INDCON

might do, construed as applying to INDCONs which have the character of being concatenated with a .triangular.. For something which applies to PMese and non-PMese expressions alike, we might try

[.^*triangular*.] INDCON

Here the only directly illustrating component would be introduced by asterisks quotes. The dot quotes would, as always, serve to form a linguistic-functional expression; in this case an adjectival functional expression which, as indicated by the brackets, combines with 'INDCON' to form a common noun which applies to INDCONs in any language which have a character which is functionally equivalent to the character of being concatenated with a *triangular* in our language. The illustrating role of the dot quoted expression as a whole would consist in the fact that it selects for functional scrutiny any sentence in our language which consists of an INDCON concatenated with a *triangular,* e.g. 'triangular a.' [Back]

{14} I say 'roughly,' because, obviously, to capture in the formal mode the specific sense of 'quality' we must pick out a far more restricted class of predicates than simply those which are one-place. I shall not attempt to botanize predicates on this occasion, beyond contrasting those which characterize (roughly adjectival predicates) and those which classify (roughly common nouns). In other words, in what follows 'quality' has, roughly, the sense of 'attribute.' [Back]

{15} In one sense the statement that there are minds is non-controversial. But are there minds? The Cartesian says yes, the Strawsonian, no. [Back]

{16} See "Towards a Theory of the Categories," p. 68, the paragraph which immediately follows the claim in question. [Back]

{17} See the discussion of this statement in AE, Philosophical Perspectives, pp. 262 ff. [Back]

{18} The reason for the qualification will emerge shortly. [Back]

{19} See previous footnote. [Back]

{20} Compare 'Dogs are a kind of animal.' [Back]

{21} I suspect you also have in mind the categorial grammar of Ajdukiewicz, according to which the basic grammatical categories are 'noun' and 'sentence,' predicates being expressions which concatenated with a noun yield a sentence, connectives being expressions which turn sentences into sentences, etc. But the relation between this distinction between basic and derivative categories to the semantic problems we have been discussing remains to be clarified. [Back]

{22} Jeff Sicha, in a personal communication, has emphasized the care with which a system of categorial classification, with its contrastive pigeon holes, must be worked out to avoid paradox. [Back]

{23} It must be remembered that there is another sense (one more step up the semantic ladder) in which it is true that

That there is one pope is an object.

In this sense it is the material mode for

The .the .(E1x) x epsilon(1) pope.. is an ST (i.e. a DST).

See the discussion of levels of abstract entities, AE, pp. 250 ff. in Philosophical Perspectives. [Back]

{24} This is not, of course, to be confused with

a, which is triangular, is red

[Return to Main Page]