Philosophical Studies (2023) 180:1513–1537

https://doi.org/10.1007/s11098-023-01960-4

Curry and context: truth and validity

Keith Simmons1

Accepted: 18 April 2023 / Published online: 7 June 2023

© The Author(s), under exclusive licence to Springer Nature B.V. 2023

Abstract
A Curry paradox about truth is generated by the following sentence, written on the
board in room 101:

If the sentence on the board in room 101 is true then 1 ≠ 1.

A Curry paradox about validity is generated by the following argument, written on

the board in room 102:

The argument on the board in room 102 is valid. Therefore, 1 ≠ 1.

Though the sentence and the argument generate Curry paradoxes, they also gener-
ate more basic paradoxes, in a sense to be made clear. I argue that if we solve these
more basic paradoxes, we have solutions to both kinds of Curry paradox. The posi-
tive proposal is in part inspired by a brief remark of Gödel’s, that the paradoxes
might appear “as something analogous to dividing by zero”—so that the concepts of
truth and validity, for example, are everywhere applicable except for certain singular
points or singularities. A second central claim is that ’true’ and ’valid’ are context-
sensitive predicates. This contextual-singularity approach to the Curry paradoxes
applies also to other paradoxes of truth, validity, denotation, and predicate-exten-
sion. So a more general aim of the paper is to provide a unified response to semantic
paradox.

Keywords Truth · Validity · Paradox · Curry · Context · Singularity

A Curry paradox about truth is generated by the following sentence, written on the
board in room 101:

If the sentence on the board in room 101 is true then 1 ≠ 1.

* Keith Simmons
keith.simmons@uconn.edu
1
Philosophy Department, University of Connecticut at Storrs, 344 Mansfield Road Unit 1054,
Storrs, CT 06269‑1054, USA

13
Vol.:(0123456789)
1514 K. Simmons

A Curry paradox about validity is generated by the following argument, writ-

ten on the board in room 102:

The argument on the board in room 102 is valid.

Therefore,

1 ≠ 1.

But though these sentences generate Curry paradoxes, they also generate more
basic paradoxes, in a sense to be made clear. Solve these more basic paradoxes, I
shall argue, and we have solutions to both kinds of Curry.

1 A truth paradox

Let’s start with truth and the sentence T on the board in room 101:

(T) If the sentence on the board in room 101 is true then 1 ≠ 1.

T generates a case of contingent or empirical paradox—it matters where the sen-

tence T is written. We’re concerned, then, with a token sentence, or a sentence-
type in a context. Without losing sight of this, for convenience we can present T
more simply:

(T) If T is true then 1 ≠ 1.

I’m leaving it vague at the moment as to the language in which T is expressed—

we can loosely say that it’s English. But we can assume that the occurrence of
‘true’ in T has a certain fixed extension. It will be helpful to mark this extension
with a subscript—say, ‘trueT’. The use of the subscript here is modest: it will
serve only to track coextensive uses of ‘true’. An occurrence of ‘true’, whether
in T or not, is to be represented as ‘trueT’ if and only if it is coextensive with the
occurrence of ‘true’ in T. The subscript does not indicate what that extension is—
unlike, for example, an index indicating the level of an expression in a hierarchy
of languages. The subscript serves only to mark an occurrence of ‘true’ that has
the same extension as the occurrence in T. So we represent the sentence T as:

(T) If T is ­trueT then 1 ≠ 1

In the course of our reasoning about T, we employ the ­truthT-schema, ­TRT:

(TRT) s is ­trueT iff p,

13
Curry and context: truth and validity 1515

where ‘s’ is a name of the sentence that replaces ‘p’. And we also assume through-
out classical sentential logic (SL).

(1) T is ­trueT. [Assumption for reductio]

(2) T is ­trueT iff (if T is ­trueT then 1 ≠ 1). [­ TRT instantiated to T]

(3) If T is ­trueT then 1 ≠ 1. [1, 2 SL]

(4) 1 ≠ 1. [1, 3 SL]

We’ve reached absurdity, given that 1 = 1. So we can complete the reductio:

(5) T is not ­trueT. [1–4, SL]

(6) T is ­trueT iff (if T is ­trueT then 1 ≠ 1). [­ TRT instantiated to T]

(7) It’s not the case that (if T is ­trueT then 1 ≠ 1). [5, 6 SL]

(8) T is ­trueT. [7, SL].1

We reach a contradiction at 5 and 8, and we’re landed in paradox. We have

assumed SL, we have assumed the absurdity of ‘1 ≠ 1’, and, when we move from (5)
to (6), we assume that T can be assessed by the ­truthT-schema ­TRT. (At line 1, we
assume that T is t­rueT, from which it follows that T is assessable by T­ RT. But this
assumption is discharged when we complete the reductio.)
We have independent reason to reject the third of these assumptions: the assump-
tion that T can be assessed by ­TRT. Consider the attempt to determine whether T is
­trueT. Since T is a conditional, and its consequent is false, the evaluation of T will
depend on the evaluation of its antecedent. But the antecedent leads us back to T
itself, so whether or not the antecedent is ­trueT will depend on whether or not T is
­trueT. We are caught in an evaluative regress. The attempt to evaluate T never bot-
toms out. T is ungrounded. T is not assessable by ­TRT.
So we conclude that T is pathological, in the precise sense that T cannot be
evaluated by the t­ruthT-schema ­TRT—this is p­ athologicalityT, for short. Lines 1–8
are to be regarded as a reductio of the assumption that T is assessable by T ­ RT,
yielding:

(9) T is ­pathologicalT

But our reasoning need not end here. We can reason past pathology. T is excluded
from the extension of ‘trueT’—for if it were ­trueT, it would be assessable by ­TRT,
and not p­ athologicalT. So,

1
Assuming SL, we may deduce p from the negation of if p then q.

13
1516 K. Simmons

(10) T is not t­rueT

And from (10) and SL we obtain:

(T*) If T is ­trueT then 1 ≠ 1.2

Now T* and T are composed of the same words with the same extensions—T* is
a repetition of T. Yet T* is a true conclusion reached by valid reasoning, while T is
pathological. T and T* are tokens of the same type, yet they differ in semantic status.
Other paradoxes generate repetitions: semantically unproblematic expressions
that are tokens of the same type as pathological expressions. Suppose I write on the
board:

(L) The sentence on the board is not ­trueL,

where we represent the occurrence of ‘true’ in L modestly as ‘trueL’. We reach a

contradiction in the familiar way, and so L is ­pathologicalL: the sentence on the
board cannot be assessed by the ­truthL-schema. So L is excluded from the extension
of ‘trueL’, for if L was true it would not be ­pathologicalL. So we can truly conclude:

(L*) The sentence on the board is not ­trueL.

L* does not lead back to paradox—for that we’d need the ­truthL-schema, but we’ve
established that the sentence on the board is not assessable by the ­truthL-schema. L*
is semantically unproblematic—it is a repetition of L, but differs in semantic status.
So L* is true, but L is pathological, just as T* is true, but T is pathological. Again
we have tokens of the same type, yet with different semantic status.3 It is natural,
then, to look for a contextual explanation.
Elsewhere I have argued for a contextual treatment of the Liar, paradoxes of
denotation (König’s, Richard’s, Berry’s), and a version of Russell’s paradox in

2
Assuming SL, we may deduce if p then q from the negation of p.
3
Paradoxes of denotation and of predicate-extension also generate repetitions. Suppose I write on the
board just these three expressions: ‘pi’, ‘six’, ‘the sum of the numbers denoted by expressions on the
board’. If we assume that the third expression denotes a number, say k, we reach absurdity: k = π + 6 + k.
So the third expression is pathological, and fails to denote. But then, since the third expression fails to
denote, we may conclude: the sum of the numbers denoted by expressions on the board is π + 6. And
here we have a repetition of the third expression which is not pathological—it denotes a number, namely
π + 6.
Or suppose I write on the board just these two predicates: ‘moon of the Earth’, ‘one-membered
extension of a predicate on the board’. Suppose the second predicate has a determinate extension. Then
we reach absurdity: the second predicate has a one-membered extension if and only if it has a two-
membered extension. So the second predicate is pathological, and fails to have an extension. So we
may conclude: the extension of the first predicate is the only one-membered extension of a predicate
on the board. Here, in our conclusion, is a repetition of the first predicate, and it isn’t pathological—its
­
extension is the extension whose only member is the Moon.

13
Curry and context: truth and validity 1517

terms of the extension of predicates.4 From standard accounts of context, we can

extract the idea of tracking context-change by keeping a running record of shifts in
the information presumed to be available.5 In the course of the Liar reasoning, for
example, we establish that L is not t­ rueL, because pathological. This is new semantic
information, and it changes the context. When we finally evaluate L* as true, we do
so in a new context which is reflective with respect to the sentence on the board—a
context in which it is now established that the sentence on the board is pathological
and not t­rueL. In terms of Lewis’s notion of conversational score, the score at the
final stage of the Liar discourse is this: L is not ­trueL, L* is a repetition of L, and L*
is true. We have to accommodate our evaluation of L* as true, when, as a repetition
of L, L* cannot be t­rueL. We can do this only be identifying a shift in the extension
of ‘true’ when we evaluate L* as true. The predicate ‘true’ is context-sensitive.
There is no intrinsic difference between L and L*. The difference lies in the sche-
mas by which they’re assessed. In the course of the Liar reasoning, L is assessed by
the ­trueL-schema, and L* by the truth-schema reflective with respect to the sentence
on the board.
Similarly with T and T*. In our reasoning about T, T is assessed by the
­trueT-schema, and we reach the conclusion that T is ­pathologicalT and not ­trueT.
With this new semantic information now available, we infer T*, and declare T* true.
At this point, the conversational score is this: T is not t­rueT, T* is a repetition of T,
and T* is true. To accommodate this final evaluation of T*, we have to identify a
shift in the extension of ‘true’. Again, the predicate ‘true’ is context-sensitive. T*
is not t­rueT but it is true—trueR, let us say. And the only difference between T and
T* is the schema by which each is assessed: T by the ­truthT-schema, and T* by the
­truthR-schema ­TRR, a schema reflective with respect to T.
T* can no more be assessed as ­trueT than can T. Indeed, it is easy to check
that we can replace ‘T’ by the name of any repetition of T in (1)–(9) and estab-
lish that the repetition is ­pathologicalT. No repetition of T can be assessed by the
­truthT-schema. And just as T* can be assessed (as t­rueR) by the t­ruthR-schema ­TRR,
so can T. T can be rehabilitated: T can be reflectively assessed as t­rueR, just because
it is p­ athologicalT and not ­trueT. But in the course of our reasoning, T is assessed by
­TRT, and T* by T ­ RR. And this is where the difference between T and T* lies.
We can make these observations a little more precise by introducing the notion of
a primary representation. How should we represent the sentence T? It is of a certain
type—let’s denote it by ‘type(T)’. And the occurrence of ‘true’ in T is represented
by ‘trueT’. So a first thought is to represent T by an ordered pair <type(T),trueT>.
But this will fail to distinguish T from the repetition T*, since the occurrences of
‘true’ in T and T* are coextensive. We need also to consider the schemas by which
each is assessed in the course of our reasoning. So T is better represented by an
ordered triple <type(T),trueT,TRT>, where the third member of the triple indicates
that in our reasoning we attempt to assess T by the ­truthT-schema ­TRT. This is what
leads to the conclusion that T is ­pathologicalT. Call this the primary representation

4
See Simmons (2018).
5
See for example Stalnaker (1975), Lewis (1979), Heim (1988), Reinhart (1981), Grosz and Sidner
(1986), Halliday (1967), Clark and Haviland (1977), Chafe (1976), Brown and Yule (1983).

13
1518 K. Simmons

of T. The primary representation of T* is <type(T),trueT,TRR>, indicating that in

our reasoning we assess T* by the reflective ­trueR-schema ­TRR. This is what leads
to the conclusion that T* is true (that is, t­rueR). The difference between T and T* is
a matter of the schemas by which they’re assessed in our reasoning, indicated by the
third members of their primary representations.
If we want to capture what is going on in our reasoning, the primary representa-
tions of T and T* are required. As we saw, T can be rehabilitated—if assessed by
­TRR, T is assessed as ­trueR, just like T*. This reflective evaluation of T is repre-
sented by a secondary representation of T: <type(T),trueT,TRR>. But this secondary
representation does not capture T’s pathological status, and so it does not figure in
our contextual analysis of our reasoning. And similarly, the secondary representa-
tion <type(T),trueT,TRT> of T* does not capture T*’s unproblematic status.

2 Singularities

According to the contextual account of our reasoning about T, there is a shift in the
extension of ‘true’, represented by the shift from ‘trueT’ to ‘trueR’, where the sub-
scripts are modest, tracking only coextensiveness and lack of coextensiveness. How
might we try to pin down the extension of an occurrence of ‘true’?
I’ve argued elsewhere against a hierarchical approach, where the move from
‘trueT’ to ‘trueR’ is a move from one level of language to a higher level.6 One seri-
ous worry for the hierarchical account is that it places massive restrictions on ordi-
nary uses of ‘true’. If a use represented by ‘trueT’ is assigned a level, then it fails to
apply to all those truths at higher levels. Gödel made a parallel complaint about Rus-
sell’s simple theory of types:
“objects are divided into mutually exclusive ranges of significance, each range
consisting of those objects that can replace each other; and that therefore each
concept is significant only for arguments belonging to one of those ranges, i.e.,
for an infinitely small portion of all objects.”7
Gödel continues:
“It is not impossible that the idea of limited ranges of significance could be
carried out without the above restrictive principle. It might even turn out that
it is possible to assume every concept to be significant everywhere except
for certain ’singular points’ or ’limiting points’, so that the paradoxes would
appear as something analogous to dividing by zero.”8
In the case of truth, the idea is that the predicate ‘true’ applies almost everywhere,
except for certain singular points or singularities where their application breaks
down. In the case of T, T is a singularity of ‘trueT’.

6
See, for example, Simmons 1993. Kripke argues against a hierarchical approach in Kripke (1975). Tar-
ski had doubts about such a treatment of ‘true’ in natural language—see Tarski, (1933/1986, p. 267).
7
Kurt Gödel 1944, in Schilpp (1944, p. 149).
8
Op. cit., p. 150.

13
Curry and context: truth and validity 1519

The job of a singularity theory will be to identify pathology and singularities.

Consider T. We’ve noted already that T is ungrounded. When we attempt to assess
T by the ­truthT-schema ­TRT, we’re led into an infinite regress. This suggests that we
can capture the pathology of T via a certain kind of tree, call it the primary tree of T.
Here is the primary tree for T:

We can think of this tree as a map of the attempt to evaluate T by the schema ­TRT.
The top node of the tree is the primary representation of T. Branches lead to T’s
antecedent and consequent, suitably represented. The consequent of T contains no
context-sensitive expressions, and so it is suitably represented by its type. And any
token of its type is false, whatever the context. So we pair its type together with
a fixed value of falsity, indicated by ‘f’. The left branch leads to the primary rep-
resentation of the antecedent. The first member of the triple specifies the anteced-
ent’s type, ‘T is true’. The second member indicates that the occurrence of ‘true’
in the antecedent is to be represented by ‘trueT’. And since the conditional T is to
be assessed by ­TRT, so are its components, in particular its antecedent—and this is
indicated by the third member. Since the antecedent says that T is t­ rueT, to assess the
antecedent we will need to assess T by the schema ­TRT. So the node at the third tier,
directly below the primary representation of the antecedent, is <type(T),trueT,TRT>.
So now we have a node identical to the top node, and the tree continues as before.
So the primary tree for T contains an infinite branch. And there is no escap-
ing this infinite regress. Since T is a conditional, and the consequent is false, we
have to evaluate the antecedent if we are to evaluate T. And so on. So the infinite
branch indicates pathology—T is ungrounded. The repeated occurrence of T’s
primary representation on the infinite branch indicates that T cannot be assessed
by the t­ rueT-schema ­TRT: T is a singularity of ‘trueT’.
What about the primary tree for T*? It looks like this:

13
1520 K. Simmons

The top node is the primary representation of T*. The second member of the
triple indicates that the occurrence of ‘true’ in T* is represented by ‘trueT’, and
the third member indicates that T* is evaluated in our discourse by the reflective
­truthR-schema ­TRR. Branches lead to the antecedent and consequent of T*, suit-
ably represented. Since T* is to be evaluated by the t­ruthR-schema, so is its ante-
cedent, ‘T is ­trueT’. But now, to evaluate this antecedent, we have to determine
whether T is ­trueT. So at the third tier, T is represented by <type(T),trueT,TRT>,
where the third element indicates that T is to be evaluated by T ­ RT. This is the
primary representation of T, and the tree continues as the primary tree for T. The
leftmost branch is infinite.
T*’s primary representation does not repeat on the infinite branch—so the tree
indicates that T* is not pathological. T* is not a singularity of ‘trueR’. But T’s
primary representation does repeat, indicating that T is pathological, and a singu-
larity of ‘trueT’. T* stands above the evaluative regress in which T is caught. At
the second tier, we reflectively evaluate T’s antecedent by T ­ RR. Since T is a sin-
gularity of ‘trueT’, the antecedent is not ­trueR. Instantiating ­TRR to T*, we have

T* is ­trueR iff if T is t­rueT then 1 ≠ 1.

So T* is indeed ­trueR, since the antecedent of the conditional on the right is not
­trueR. And, just like T*, T is assessed by ­TRR as ­trueR.
So T’s status is assessment-sensitive: its semantic value depends on the schema
by which it is assessed. But this does not motivate relativism about truth. It is an
absolute matter that T is p­ athologicalT. The original attempt to assess T by ­TRT is

13
Curry and context: truth and validity 1521

a failed attempt. T* tells the whole story about T, that T is ­pathologicalT, not ­trueT,
and can be reflectively assessed as ­trueR.9
A full formal singularity theory of truth is beyond the scope of this paper. A sin-
gularity theory of truth, in which primary representations and primary trees are basic
technical notions, is provided in Simmons (1993). The singularity theory is extended
to the notions of denotation and predicate-extension in Simmons (2018). But I hope
I’ve done enough here to indicate the bare bones of such a theory.
The identification of singularities does not by itself provide a positive account of
truth—it tells us only what to leave out of the extension of a given occurrence of ‘true’.
But once we’ve identified singularities, we can take the positive step we need: we can
provide a schema for truth. Given an occurrence of ‘true’ represented by ‘trueα’,

(i) if s is not a singularity of ‘trueα’, then s is ­trueα iff p, and

(ii) if s is a singularity of ‘trueα’, then s is not t­rueα,where ‘p’ is replaced by a
sentence, and ‘s’ is replaced by a name of that sentence. This is a minimally
restricted principle for truth.

3 A truth Curry

Now consider an alternative way of reasoning about T, the sentence on the board in
room 101.

(1/) T is ­trueT. [Assumption for conditional proof CP]

(2/) T is ­trueT iff (if T is t­rueT then 1 ≠ 1). [­ TRT instantiated to T]

(3/) If T is ­trueT then 1 ≠ 1. ­[1/, ­2/ SL]

(4/) 1 ≠ 1. [1, 3 SL]

So far, we have reasoned about T just as before. But now suppose we continue as
follows:

Observe that we’ve just argued from (1) to (4). So, by conditional proof:

(5/) If T is ­trueT then 1 ≠ 1. [1–4, CP]

(6/) T is ­trueT ­[5/, ­TRT, SL]

(7/) 1 ≠ 1. ­[5/, ­6/ SL]

9
John MacFarlane characterizes relativism about truth this way: “To be a relativist about truth is to hold
that languages with assessment-sensitive expressions are at least conceptually possible”. And “To be a
relativist about truth in English (or some other natural language) is to hold that some expressions of Eng-
lish are assessment-sensitive” (MacFarlane 2014, p. 65). The case of T—and other cases of pathology—
may be seen as counterexamples to MacFarlane’s characterization of relativism about truth.

13
1522 K. Simmons

As we consider this reasoning, we should keep in mind that we are dealing with
sentences considered as tokens, or sentence-types in a context. At line 1, we assume
that T is t­rueT, for a conditional proof. It follows from this assumption that we can
apply the schema T ­ RT to T, and we semantically descend, to ­3/. Then a step of MP
yields 1 ≠ 1. So we complete the conditional proof at ­5/. We reach a repetition of T
at ­5/, a new sentence distinct from T: 5­ / is neither on the board nor self-referential.
We’ve established 5­ / by valid reasoning. And as we’ve seen, ­5/, like any repetition
of T, can be reflectively evaluated as true (­ 5/ is ­trueR). So far so good.
The Curry reasoning leads on to absurdity only because three implicit assump-
tions are now made:

(i) 5/ may be assessed by the schema ­TRT.

(ii) Since we deduced ­5/, ­5/ is ­trueT.
(iii) If ­5/ is ­trueT then T is ­trueT.

Given (i), we can apply T­ RT to 5­ /. Given (i) and (ii), we can move from 5­ / to ‘5/ is
­trueT’. And given (iii), we can move on to 6­ /. Once these further assumptions are
made explicit, we can see that the Curry reasoning slurs over the difference between
the sentences T and ­5/. We can accept assumption (iii), since 5­ / is a repetition of T.
But assumptions (i) and (ii) are unacceptable. These assumptions in turn depend on
more general assumptions:

(I) Any sentence is assessable by ­TRT.

(II) Any sentence arrived at by sound reasoning is ­trueT.

In particular, if we deduce a token of the type ‘If T is ­trueT then 1 ≠ 1’, then we may
semantically ascend via ­TRT, introduce the ­truthT-predicate, and declare the token to
be ­trueT.
But we’ve already shown that I and II are false. Consider our deduction of T*,
another repetition of T. At the reflective stage of the reasoning in which we estab-
lished T*, the schema T ­ RT has been abandoned, and T* is to be assessed by ­TRR—
as is indicated by T*’s primary representation <type(T),trueT,TRR>. T* is ­trueR, but
it isn’t t­rueT. And the same goes for T, and for any repetition of T, ­5/ included. This
is where the Curry reasoning goes wrong—we should reject assumptions I and II.
Neither ­5/ nor T can be assessed by ­TRT, and neither are ­trueT.
It’s important to see the differences between our original paradox and the Curry
paradox. Our original paradox applied T ­ RT to T only, and at no point did we apply
­TRT to a sentence other than T. Our original paradox, where we reached a contradic-
tion at lines 5 and 8, was generated from SL, the absurdity of 1 ≠ 1, and the assump-
tion that T ­ RT applies to T. The Curry paradox is generated from SL, the absurdity
of 1 ≠ 1, and the three further assumptions I, II, iii. Our original paradox is more
basic—less is needed to reach absurdity.
Our original reasoning showed that T cannot be assessed by ­TRT, that no repetition
of T can be assessed by T ­ RT, and that neither T nor any repetition of T is t­rueT. These
results required only these assumptions: SL and the absurdity of 1 ≠ 1. And these
assumptions are shared by the Curry reasoning. Given just these assumptions, our

13
Curry and context: truth and validity 1523

original reasoning shows that the Curry’s additional assumptions I and II are false. So
we have an immediate resolution of the Curry. We should reject the step from ­5/ to ­6/,
since the repetition 5­ / cannot be assessed by T
­ RT. And we should reject 6­ / itself: T is not
­trueT.

4 The Curry generalized

The sentence T has an absurd consequent, and the paradoxical Curry reasoning seems
to succeed in proving this absurdity. The Curry paradox is often presented this way.
But we can take the Curry in a more general or schematic way. The consequent can
be any sentence we like—absurd, just plain false, or true, or necessarily true. Then
what’s paradoxical isn’t that we prove an absurdity. What’s paradoxical is that we can
prove anything by, as it were, ‘logic alone’. If we start, for example, with the sentence

(U) If U is true then snow is white,

then the Curry reasoning apparently leads us to the conclusion that snow is white,
with no need for any empirical enquiry.
This suggests that we should look at a more schematic, more general version of
the Curry, generated by the sentence

(G) If G is true then φ

where φ is any sentence, with any truth value. The pathology here is that we can
prove φ, whatever φ’s truth status. So the apparent paradox is generated indepen-
dently of φ’s truth value, and any resolution must respect that. Let ‘trueG’ modestly
represent the occurrence of ‘true’ in G. The Curry reasoning goes like this:

(1) G is ­trueG. [Assumption for CP]

(2) G is ­trueG iff (if G is ­trueG then φ). [­ TRG instantiated to G]
(3) If G is ­trueG then φ [1, 2 SL]
(4) φ [1, 3 SL]
(5) If G is ­trueG then φ. [1–4, CP]
(6) G is ­trueG [5, ­TRG instantiated to 5, SL]
(7) φ [5, 6 SL]

The absurdity here is that φ is entirely arbitrary: we can prove anything from the
sentence G.
To respect this sense of absurdity, we have to treat G’s consequent as having a
fully unspecified value. If we assume from the outset some definite value for the
consequent, we may miss the way in which G leads to absurdity. For example, con-
sider the sentence U again, with a modest subscript in place:
(U) If U is t­rueU then snow is white. Suppose we assume immediately that the
consequent is t­rueU. Then by the semantics of the conditional, it follows that U is
­trueU. Then we may infer the conditional itself (by T
­ RU), and then, by MP, that snow

13
1524 K. Simmons

is white. But there’s nothing absurd here—we haven’t proved that snow is white ‘by
logic alone’. Rather, we’ve taken a circuitous route from the initial assumption that
the consequent is ­trueU to the conclusion that snow is white. So it’s crucial to make
no assumptions about the truth status of the consequent. Accordingly, we will take
the consequent’s truth status to be fully unspecified—let this status be represented
by u.
G’s pathology, like T’s, is a matter of its ungroundedness. Consider an attempt to
evaluate G. Since G is a conditional, and the truth status of its consequent is u, any evalu-
ation of G will depend on its antecedent. This is in line with Kleene’s strong three-valued
logic. Take a conditional whose consequent has value u. The conditional is true if its
antecedent is false, and false if its antecedent is true. So to evaluate G, we must evaluate
its antecedent. But the antecedent sends us back to G, and we are caught in an evaluative
loop.
These intuitive ideas can again be made more precise via the notions of primary
representation and primary tree. In parallel with T, the primary representation of G
is <type(G),trueG,TRG>. The second member of the triple indicates that the occur-
rence of ‘true’ in G is (modestly) represented by ‘trueG’. And the third member indi-
cates that in the course of the Curry reasoning, G is evaluated by the ­truthG-schema
­TRG.
The primary tree for G is this:

G’s primary tree is just like T’s, except for the representation of the consequent.
The consequent’s truth value is to be set aside, left as unspecified. G’s unground-
edness is captured by the infinite branch. G’s primary representation repeats on
this infinite branch, indicating that G is p­ athologicalG: it cannot be assessed by the
­truthG-schema. And G is a singularity of ‘trueG’.
Once we recognize that G is pathological, and excluded from the extension of
‘trueG’, we can reflectively evaluate G’s antecedent as false, and so G as true. So
G is reflectively evaluable as true, whatever the truth value of the consequent. We
can rehabilitate G, just as we can rehabilitate T: G is true. But here, this reflective

13
Curry and context: truth and validity 1525

use of ‘true’ is not coextensive with ‘trueG’—there is a shift in the extension of

‘true’.
So the reasoning of the generalized Curry goes wrong in just the same way
as the original Curry reasoning 1­ /–7/. Things go wrong with the step from 5 to
6, which requires the analogues of assumptions I and II. Neither 5 nor G are
assessable by T­ RG, and neither 5 nor G are t­rueG. The pathology of T and G is
located in the same place, in the self-referential character of these sentences, and
the attempt to evaluate them by their associated truth schemas. This brings both
kinds of truth Curry paradoxes into line with the other semantic paradoxes of
truth, denotation, and predicate-extension, as I believe they should be.10 The pre-
sent approach, in terms of context and singularities, treats all the members of this
family in the same way.

5 A validity paradox

Suppose the following argument is written on the board in room 102:

(P) The argument on the board in room 102 is valid.

Therefore

(C) 1 ≠ 1.11

P is self-referential in the sense that it refers to an argument of which it itself is a

premise. Let V be the argument to which P makes reference, and let ‘validP’ mod-
estly represent the occurrence of ‘valid’ in C. In what follows, I shall assume that
validity is defined in terms of truth: for any argument Γ, with premises π1, …, πn and
conclusion κ

(VAL) Γ is valid iff necessarily, if each of π1, …, πn is true then κ is true.

For short, a validity is necessary truth-preservation. Here, necessary truth-preserva-

tion is irrespective of logical form. For example, an argument with a necessarily true
conclusion will be valid, whatever the logical form of the argument.12

10
This then is in opposition to those who take the Curry to be a different kind of paradox from the liar.
Graham Priest claims that the Curry is different, because it has nothing to do with negation—see Priest
(1995, p. 186). Weber et al. take the Curry to be different from the liar because the Curry does not fit the
inclosure schema—see Weber et al. (2014, p. 823).
11
Stephen Read points out that an argument like this was considered by fifteenth and sixteenth century
authors (Read 2001, p. 189). Read cites Ashworth (1974, p. 125) and Roure (1962, p. 262).
12
So the notion of validity that I am working with is to be contrasted with what might be called ‘logical
validity’, the Tarskian notion where, roughly, validity is truth-preservation in virtue of logical form.

13
1526 K. Simmons

Let the set of arguments that are ­validP be ­VP. Associated with ­VP will be a set
of truths, call it ‘TP’. The ­validP arguments will be those that necessarily preserve
­truthP, where the extension of ‘trueP’ is the set T
­ P. The ­validityP-schema is:

(VALP) Γ is ­validP iff necessarily, if each of π1, …, πn is ­trueP then κ is ­trueP.

The ­truthP-schema is:

(TRP) s is ­trueP iff p,

where ‘s’ is a name of the sentence that replaces ‘p’.

We start by assuming that argument V is ­validP:

(1) V is ­validP. [Assumption]

(2) P is ­trueP iff V is ­validP ­[TRP instantiated to P]

(3) P is ­trueP [1, 2, SL]

(4) ‘1 ≠ 1’ is ­trueP. [1, 3, V

­ ALP instantiated to V]

(5) 1 ≠ 1 [4, ­TRP, SL]

We have reached absurdity, given that 1 = 1. So, we have a reductio of the

assumption at 1:

(6) V is not ­validP. [1–5, RAA]

(7) P is not ­trueP. [6, ­TRP instantiated to P, SL]

We reach 7 by just sentential logic and the necessary ­truthP and ­validityP sche-
mas. So it follows that:

(8) Necessarily, P is not ­trueP.

But then, given V

­ ALP and 8, V cannot be i­nvalidatedP—V’s premise P cannot be
­trueP. So,

(9) V is ­validP. [8, ­VALP instantiated to V]

We have a contradiction at lines 6 and 9. We can regard the argument 1–9 as a

reductio of the background assumptions that V can be assessed by ­VALP and P can
be assessed by T­ RP. These assumptions go together: we can assess A for v­ alidityP
if and only if we can assess P for ­truthP. For suppose we can assess V for v­ alidityP.

13
Curry and context: truth and validity 1527

Then we can ascertain whether V is necessarily ­truthP-preserving—and since the

conclusion C is (necessarily) not t­rueP, it must be that we can ascertain whether P
is (necessarily) ­trueP or not. Now suppose we can assess P for ­truthP. Then, given
what P says, we will be able to assign a definite validity status to V (either ­validP or
not ­validP). We have shown that these assumptions—that V and P are assessable by
­VALP and ­TRP respectively—lead to a contradiction. So we may conclude that V is
not assessable by V ­ ALP and P is not assessable by T ­ RP.
As with the sentence T, there is independent motivation for these conclusions.
V and P are ungrounded. When we try to evaluate V, we note that the conclusion
C is necessarily not t­rueP. So we have to examine the truth status of the premise
P, to determine whether it is possible for P to be ­trueP. (If it is possible, V is not
­validP; if it isn’t possible, V is v­ alidP.) But the attempt to evaluate P sends us back
to the argument V, and we are caught in an evaluative regress.
So we conclude that argument V is pathological in the precise sense that V
cannot be assessed by V ­ ALP. That is,

(10) V is ­pathologicalP.

But now it follows that:

(11) V is not ­validP.

For if V was v­ alidP, it would not be p­ athologicalP. And given that it’s been estab-
lished that V is not assessable by V ­ ALP, 11 doesn’t lead back to contradiction—
for that, we would need to assess V by ­VALP.
So at 11 we’ve established that V is not v­alidP. Indeed, given that we’ve
established 11 given only SL and the absurdity of 1 ≠ 1, we’ve established that V
is necessarily not v­ alidP.
Now suppose we assume:

(P*) V is ­validP. [Assumption]

Then, given (11), we have a contradiction, so can validly infer

(C*) 1 ≠ 1 [11, P*, SL]

And now we have a repetition of V—call it V*—produced in a context reflective with

respect to V, a context in which we have established V’s ­pathologicalityP. V*’s premise P*
and conclusion C* are composed of the same words as the premise and conclusion of V,
with the same extensions. And yet we have reached 11 by valid reasoning, and we move
validly from P* to C*, via 11. So V* is valid, while V is p­ athologicalP and not ­validP.
But though V* is valid, it is not ­validP. Indeed, the argument 1–11 can be readily
adjusted to show that V* cannot be assessed by V ­ ALP, and is not v­ alidP—simply replace
‘V’ by V* and ‘P’ by ‘P*’ and the argument goes through. Similarly for any repetition
of V: replace ‘V’ by a name of the repetition, and ‘P’ by a name of its premise. No

13
1528 K. Simmons

repetition of V is v­ alidP, V* included. So here the conversational score is this: V is not

­validP (because p­ athologicalP), V* is a repetition of V and so not v­ alidP, and V* is valid.
To accommodate our evaluation of V* as valid, we have to identify a shift of exten-
sion—V* is ­validR, let us say, but not v­ alidP. The validity predicate is context-sensitive.
Let ‘trueR’ represent the truth predicate whose extension is the set of truths whose
truth is necessarily preserved in the arguments that are ­validR: the ­validR arguments
necessarily preserve t­ruthR. V* is v­ alidR, so if P* is t­rueR then C* is ­trueR. But C* is not
­trueR, so P* is not ­trueR. Here’s the ­trueR-schema instantiated to P*:

P* is ­trueR iff V is ­validP.

Since we’ve reflectively established that V is not ­validP (because ­pathologicalP), it fol-
lows straightforwardly that P* is not t­rueR. Like P and any repetition of P, P* is not
assessable by the t­ruthP-schema—but, in a context that is reflective with respect to V,
P* is assessable by the t­ruthR-schema, as not ­trueR.
The primary representation of argument V is <type(V),validP,VALP>. Here
‘type(V)’ denotes the type of any argument with premise of type ‘The argument
on the board in room 102 is not valid’, and conclusion of type ‘1 ≠ 1’. The sec-
ond member of the triple indicates that any occurrence of ‘valid’ is to be rep-
resented by ‘validP’. In our reasoning, we attempt to assess V by the schema
­VALP, and this is indicated by the third member of the triple.
The primary tree for argument V looks like this:

The tree maps the attempt to evaluate V by the schema ­VALP. The top node is the
primary representation of V. Branches lead to the premise and the conclusion of V,
suitably represented. The conclusion C is represented by its type (since it contains
no context-sensitive expressions), paired with the value f (since any token of its type

13
Curry and context: truth and validity 1529

is false, whatever the context). The premise P is represented by its primary repre-
sentation <type(P),validP,TRP>. The second member of the triple is ‘validP’, since
the occurrence of ‘valid’ in P is represented by ‘validP’. The third member is ‘TRP’:
since we are evaluating V by ­VALP, and ­validityP is ­truthP-preservation, P is to be
evaluated by ­TRP. P makes reference to the argument V, and assesses it in terms
of ­validityP (as the second element of P’s primary representation indicates). So to
assess P, we need to assess V via the ­validityP-schema. So a vertical branch leads
from the node for P to the triple <type(V),validP,VALP>, where the third element
indicates that V is to be assessed by the ­validityP-schema. But now we have a node
identical to the top node, and the tree continues as before.
The primary tree for V contains an infinite branch. And there is no escaping this
infinite regress. Since the conclusion C is false, we have to evaluate the premise P if
we are to evaluate the argument V. So we are led down the leftmost infinite branch.
This branch indicates V’s ungroundedness. The repeated occurrence of V’s primary
representation on the infinite branch indicates that V cannot be assessed by V ­ ALP:
V is ­pathologicalP, and a singularity of ‘validP’. And the repeated occurrences of P’s
primary representation indicates that P cannot be assessed by ­truthP-schema: P is
­pathologicalP, and a singularity of ‘trueP’.
The primary tree for V* is this:

The top node is the primary representation for V*. The second member of the
triple indicates the representation of ‘valid’ in V*’s premise. The third member
indicates that, at the final stage of our reasoning, V* is assessed by the reflec-
tive ­validityR-schema ­VALR. Branches lead to the premise and the conclusion of
V*. At the second tier, the node for the premise P* is its primary representa-
tion <type(P),validP,TRR>. The second element indicates the representation of
the occurrence of ‘valid’ in P*. The third element indicates the truth-schema

13
1530 K. Simmons

by which P* is evaluated at the final stage of our reasoning. We assess P* by

the ­truthR-schema: we need to see whether P* is t­rueR to see whether ­truthR
is preserved. Now P* makes reference to the argument V, and assesses V in
terms of v­ alidityP, as the second member indicates. So to evaluate P*, we need
to evaluate the argument V by ­VALP. So the next branch leads to the triple
<type(V),validP,VALP>, which is the primary representation of V. The tree now
continues as the primary tree for argument V, and the leftmost branch is infinite.
V*’s primary representation does not repeat on the infinite branch—V* is not
pathological. Similarly, P*’s primary representation does not repeat—P* is not
pathological. Both V* and P* are reflectively evaluated, in a context where the
­pathologicalityP of V and P has been established. Consider the reflective eval-
uation of V*. We saw that our reasoning establishes (at line 11) that V is not
­validP—in fact, necessarily not v­ alidP. V is necessarily a singularity of ‘validP’.
So P* is reflectively evaluated as necessarily not t­rueR. Since V*’s premise is
necessarily not ­trueR, V* cannot be ­invalidatedR—V* is ­validR.
Both V’s primary representation and P’s primary representation do repeat on the infi-
nite branch. Both V and P are p­ athologicalP: V is a singularity of ‘validP’ and C is a singu-
larity of ‘trueP’. V* and P* stand above the evaluative loop in which V and P are caught.
Still, V and P can be rehabilitated. P can be reflectively assessed as not ­trueR, just
like P*. And V can be reflectively assessed as ­validR, just like V*.
As with truth, a full-blown singularity theory of validity is beyond the scope of
this paper. But suppose such a theory is in place. Then we can provide a schema for
validity. Given an argument Γ, with premises π1, …, πn and conclusion κ, and an
occurrence of ‘valid’ represented by ‘validα’,

(i) if Γ is not a singularity of ‘validα’, then

Γ is v­ alidα iff necessarily, if each of π1, …, πn is ­trueα then κ is ­trueα
and
(ii) if Γ is a singularity of ‘validα’, then Γ is not ­validα.
And we have a parallel schema for ‘trueα’. Restrictions are kept to a minimum.

6 A validity Curry

We can construct further validity paradoxes. We obtain a ‘contraposed’ variant of V

if we write on the board:

1 = 1. So, the argument on this board is not valid.

We can reason about this argument much as we did about V. We reach the conclu-
sion that the argument is pathological. And we can reason past pathology, moving
to a reflective context. We arrive at a valid repetition of the argument on the board.13

13
For further analysis of this validity paradox, see Simmons (2021).

13
Curry and context: truth and validity 1531

There are further validity paradoxes, in analogy with variations on the liar. We can
have looped arguments, each saying of the other that they are not valid. And we can
have extended loops, and chains of arguments. And we can have a validity-teller—
for example, if we write on the board:

1 = 1. So, the argument on the board is valid.

Just as with the liar and the truth-teller (‘This sentence is true’), we can distinguish
between cases that generate contradiction, such as V, and those that generate inde-
terminateness, such as the validity-teller. The contextual-singularity approach char-
acterizes the pathology common to both as ungroundedness, as witnessed by the
repetition of the arguments’ primary representation on an infinite branch of their
primary trees.14
But here I want to focus on one validity paradox in particular: a validity Curry. Con-
sider again the first few steps of our original reasoning about V. We began like this:

(1) V is ­validP. [Assumption]

(2) P is ­trueP iff V is ­validP ­[TRP instantiated to P]

(3) P is ­trueP [1, 2 SL]

(4) ‘1 ≠ 1’ is ­trueP. [1, 3, V

­ ALP]

(5) 1 ≠ 1 [4, ­TRP-schema]

Consider an alternative way of proceeding, where we take a distinctive Curry step.

Suppose we continue as follows:

Observe that we’ve just argued from the assumption at (1) to the conclusion at (5).
So:

(6/) V is ­validP.

(7/) P is ­trueP iff V is ­validP ­[TRP instantiated to P]

(8/) P is ­trueP [6, 7 SL]

(9/) ‘1 ≠ 1’ is ­trueP [6, 8, ­VALP]

(10/) 1 ≠ 1 [9, ­truthP-schema, SL]

14
For more discussion of these validity paradoxes, see Simmons (2021). Woodbridge and Armour-Garb
(2008) call for ‘a single treatment of validity’s pathological features” (op. cit., p71). The suggestion of
the present approach is that the underlying pathological feature is ungroundedness, and the treatment is
to identify the singularities of the context-sensitive validity predicate.

13
1532 K. Simmons

We have derived the absurdity 1 ≠ 1. This is a validity Curry paradox.

Our account of this reasoning runs parallel to the reasoning of the truth Curry.
Again we should keep in mind that we are dealing with arguments considered
as tokens, or types in a context. At line 1 we assume that V is v­ alidP. Given this
assumption, it follows that V can be assessed by ­VALP, so that V can be assessed
for ­truthP-preservation. So V’s premise and conclusion are assessable by ­TRP, so
we can instantiate T­ RP to P at line 2. By SL and V ­ ALP, we reach 1 ≠ 1 at line 5.
The argument with premise 1 and conclusion 5, call it V ­ +, is a repetition of V. V
­ +
is to be distinguished from V: it is neither on the board nor self-referential. And
­V+ can be reflectively assessed as v­ alidR, since it is a subargument of V*, which
we’ve shown to be v­ alidR. So far so good.
But now the Curry reasoning goes wrong, as it moves from 5 to 6­ /. This move
depends on three further assumptions:

(i/) The argument ­V+ may be assessed by the schema ­VALP.

(ii/) Since we’ve deduced ­5/ from 1­ /, ­V+ is ­validP.

(iii/) If ­V+ is ­validP then V is ­validP.

Given ­(i/) and ­(ii/), we can introduce the v­alidityP predicate, and declare ­V+
­validP. Given (­ iii/), we can then move on to ­6/.
But we must reject assumptions ­(i/) and ­(ii/). These in turn depend on these
more general assumptions:

(I/) Any argument is assessable by ­VALP.

(II/) Any deduction is ­validP.

And ­I/ and I­ I/ are both false. The deduction of V*’s conclusion from its premise is
valid (that is, v­ alidR), but it is not v­ alidP. Our original reasoning about V and V*
established that V is neither assessable by ­VALP or ­validP, and the same goes for
any repetition of V, V* and ­V+ included.
Again, in parallel with the case of the truth Curry, we should compare and
contrast the original reasoning about V and the Curry reasoning. Our original
reasoning applied V ­ ALP to V only—at no point did we apply ­VALP to an argu-
ment other than V. Our original validity paradox, where we generated a contra-
diction at lines 6 and 9, was generated from SL, the absurdity of 1 ≠ 1, and the
assumption that V can be assessed by ­VALP. The validity Curry paradox is gener-
ated from SL, the absurdity of 1 ≠ 1, and the further assumptions I­/, ­II/, ­iii/. Our
original paradox is more basic—we need less to reach absurdity.
Our original reasoning shows that V cannot be assessed by ­VALP, that no rep-
etition of V is assessable by ­VALP, and that V is not ­validP. These results are

13
Curry and context: truth and validity 1533

established from assuming just SL together with the absurdity of 1 ≠ 1. But

these assumptions are shared by the validity Curry reasoning. And these shared
assumptions suffice to show that the validity Curry’s further assumptions I and II
are false. So we have an immediate resolution of the validity Curry. We should
reject the step from 5 to 6­ /, since the repetition V
­ + cannot be assessed by V
­ ALP.
/ 15
And we should reject ­6 itself: V is not ­validP.

7 The validity Curry generalized

The argument V has an absurd conclusion, 1 ≠ 1. The validity Curry purported

to show that we could deduce this absurdity. But just as with the truth Curry, we
can think of the validity Curry in another way: not in terms of proving absurdi-
ties, but in terms of proving anything, by logic alone. We can prove that 1 ≠ 1,
but we can also prove that snow is black, or snow is white, or 1 = 1, by logic
alone.
So consider an argument with a conclusion whose truth value is irrelevant to the
derivation of paradox. Consider this argument, on the board in room 104:

(Q) The argument on the board in room 104 is valid.

Therefore,

(D) φ

where φ is a place-holder for any sentence, which may be absurd, plain false, plain
true, or necessarily true. Let the argument referred to by Q be W, and let ‘validQ’
modestly represent the occurrence of ‘valid’ in Q. We now reason in exact parallel
to the validity Curry reasoning about V:

15
Here’s a variant of the argument V. Suppose the following sentence is written on the board in room
103:
(P/) The argument from the sentence on the board in room 103 to ‘1 ≠ 1’ is valid.
Let the argument to which ­P/ refers be V ­ /. Then we can reason about V ­ / exactly as we originally rea-
soned about V. Take the original reasoning 1–11 of Sect. 5, replace ‘V’ by ‘V/’ and ‘P’ by ‘P/’, and we
obtain an argument that establishes that V ­ / cannot be assessed by the validityP//-schema (with the modest
subscript in place). We establish that ­V/ and ­P/ are ­pathologicalP/, ­V/ is not ­validP/, and ­P/ is not trueP//.
Curry reasoning parallel to 1–10/ of Sect. 6 goes wrong at the corresponding point. A repetition of V ­ / is
/ /
obtained, and then it is falsely assumed that this repetition is assessable by VALP/ , as validP/ .
A version of ­P/ is the main focus of Beall and Murzi (2013). Beall and Murzi explore a substructural
approach which rejects Structural Contraction. According to the present singularity proposal, classical
logic is fully preserved.

13
1534 K. Simmons

(1//) W is ­validQ. [Assumption]

(2//) Q is ­trueQ iff W is ­validQ ­[TRQ instantiated to Q]

(3//) Q is ­trueQ. ­[1//, ­2//, SL]

(4//) φ is ­trueQ. [1, 3, ­VALQ instantiated to W]

(5//) φ [4, ­TRP instantiated to φ]

And now we make the distinctive Curry move, given that we’ve just deduced 5­ //
from ­1//:

(6//) W is ­validQ.

And from here, we take steps exactly analogous to 7–10 above, using ­VALQ, ­TRQ,
and SL to reach

(10//) φ.

The absurdity here is that we can prove anything, given the argument W.
As we saw with the generalized Curry sentence G, we must not assume anything
about the value of φ if we’re to retain this sense of absurdity. For example, suppose we
assume in advance that φ is necessarily true (perhaps φ is the sentence ‘1 = 1’). Then
we might suppose that absurdity doesn’t arise. Since the W’s conclusion D is necessar-
ily ­trueQ, W is ­validQ. It follows that Q is ­trueP. Given that W is ­validQ and Q is ­trueQ,
it follows that D is ­trueQ. But this was already assumed in advance—so there is no
absurdity.
So we make no assumptions about the truth status of φ—we let its truth status be
u. The pathology of the argument W is a matter of its ungroundedness, in parallel
with the generalized Curry sentence G. Consider the attempt to evaluate W. Since
the truth value of D can play no part in the evaluation of W, W can be evaluated only
if we evaluate its antecedent. In general, if an argument has a conclusion with an
unspecified truth value, the validity of the argument may still be a determinate mat-
ter: if the antecedent is necessarily false, then the argument will be valid. So we turn
to the premise Q of W. But that sends us back to W, and we’re launched on an infinite
regress.
The primary tree for W maps out this attempt to evaluate W via the
­validityQ-schema ­VALQ:

13
Curry and context: truth and validity 1535

W’s primary tree is just like V’s, except for the representation D’s truth value as
u. The primary representations of W and Q repeat on the infinite branch. W is
­pathologicalQ (cannot be assessed by ­VALQ), and Q is p­ athologicalQ (cannot be
assessed by ­TRQ). So W is not ­validQ, and Q is not ­trueQ. W is a singularity of
‘validQ’ and Q is a singularity of ‘trueQ’.
We can observe that Q cannot be t­rueQ—Q is necessarily a singularity of
‘trueQ’. For if Q were t­rueQ, then W would have a t­rueQ premise and a f­alseQ
conclusion, and so W would not be v­ alidQ—and then Q would not be t­rueQ.
Once we reflect on Q, and see that Q is necessarily a singularity of ‘trueQ’, we
can reflectively evaluate Q as necessarily not true—say, necessarily not t­rueR.
But then, necessarily, Q cannot have a t­rueR premise and a f­alseR conclusion,
because it cannot have a t­rueR premise. That is, Q is v­ alidR, where ‘validR’ and
‘validQ’ are not coextensive. We can reflectively establish that W is valid, what-
ever the truth value of its conclusion. We can rehabilitate W.
Returning to our Curry reasoning, it’s the move to 6­ // that is the culprit. Just
as V was conflated with V ­ +, so W is conflated with the argument with prem-
ise ­1 and conclusion ­5 —let this argument be ­W+. But where things really go
// //

wrong is with the assumption that the deduction from ­1// to ­5// is ­validQ. This is
to assume the analogues of assumptions ­I/ and ­II/ in Sect. 6—to assume that any
argument is assessable by ­VALQ, and that any deduction is ­validQ. But these
assumptions are false: W­ + cannot be assessed by ­VALQ, and neither ­W+ nor W
are ­validQ. We should reject the move from ­5// to ­6//, and reject ­6// itself.

13
1536 K. Simmons

8 Concluding remarks

The truth paradox of Sect. 1 turned into a proof that the sentence T, and any repeti-
tion of T, are ­pathologicalT. This shows where the Curry reasoning in Sect. 3 breaks
down. And this resolution carries over to the generalized Curry reasoning of Sect. 4.
In parallel, the validity paradox of Sect. 5 turned into a proof that V, and any repeti-
tion of V, are ­pathologicalV. This resolves the validity Curry of Sect. 6, and, in turn,
the generalized validity Curry of Sect. 7. Both kinds of truth Curry and both kinds
of validity Curry are resolved in the same way: pathology is a matter of ungrounded-
ness, witnessed by primary trees. Occurrences of ‘true’ and ‘valid’ have singulari-
ties, sentences or arguments to which they fail to apply. But these singularities are
evaluable, as true or as valid, in suitably reflective contexts. The predicates ‘true’
and ‘valid’ are context-sensitive.
One aim of the singularity approach is to resolve semantic paradox while preserv-
ing classical logic and semantics. A second aim is to respect Tarski’s claim that a
characteristic feature of natural language is its universality:
“The common language is universal and is intended to be so. It is supposed
to provide adequate facilities for expressing everything that can be expressed
at all, in any language whatsoever; it is continually expanding to satisfy this
requirement”.16
Any restriction on the expressive power of natural language should be kept to a min-
imum. According to the singularity proposal, any use of ‘true’ or ‘valid’ is as close
to universal as possible—their close-to-global reach reflects the expressive power of
natural language. And the singularities of a given occurrence of ‘true’ or ‘valid’ are
within the reach of another.
There is much more to be said that’s beyond the scope of this paper. I haven’t
presented a full-blown singularity theory here.17 And revenge paradoxes must be
addressed. For a contextual approach, revenge comes in the form of sentences such
as ‘This sentence is not true in any context’, and arguments such as ‘This sentence is
valid in any context. So, 1 ≠ 1.’ I discuss revenge in detail elsewhere.18 In this paper,
my focus has been the Curry paradox, in all its forms. I’ve located pathology where
I think it should be: Curry sentences and arguments are ungrounded. And I’ve sug-
gested that the Curry paradoxes should be treated in the way that all paradoxes of
truth, denotation, extension, and validity should be treated—in terms of singularities
and context-sensitivity.

Funding Not applicable.

16
Tarski (1969, p. 89).
17
Again, see Simmons (2018) for a fully developed singularity theory of truth, denotation, and predi-
cate-extension.
18
For an extended discussion of revenge paradoxes of truth, denotation, and predicate-extension as they
arise for the singularity theory, see Simmons (2018). For revenge paradoxes of validity, see Simmons
(2021).

13
Curry and context: truth and validity 1537

Availability of data and materials Not applicable.

Code availability Not applicable.

Declarations
Conflict of interest Not applicable.

References
Ashworth, E. J. (1974). Language and logic in the post-medieval period. Reidel.
Beall, J., & Murzi, J. (2013). Two flavors of curry paradox. Journal of Philosophy, 110, 143–165.
Brown, G., & Yule, G. (1983). Discourse analysis. Cambridge University Press.
Chafe, W. (1976). Givenness, contrastiveness, definiteness, subjects, topics, and point of view. In C. Li (Ed.),
Subject and topic (pp. 25–55). Academic Press.
Clark, H., & Haviland, S. (1977). Comprehension and the given-new contract. In R. Freedle (Ed.), Discourse
production and comprehension (pp. 1–40). Lawrence Erlbaum Associates.
Gödel, K. (1944). Russell’s mathematical logic. In P. A. Schilpp (Ed.), The philosophy of Bertrand Russell
(pp. 123–153). Tudor.
Grosz, B., & Sidner, C. (1986). Attention, intention, and the structure of discourse. Computational Linguis-
tics, 12, 175–204.
Halliday, M. A. K. (1967). Notes on transitivity and theme in English: Part 2. Journal of Linguistics, 3,
199–244.
Heim, I. (1988). The semantics of definite and indefinite noun phrases. Garland Publishing.
Kripke, S. (1975). Outline of a theory of truth. Journal of Philosophy, 72, 690–716.
Lewis, D. (1979). Scorekeeping in a language game. Journal of Philosophical Logic, 8, 339–359.
MacFarlane, J. (2014). Assessment sensitivity: Relative truth and its applications. Oxford University Press.
Priest, G. (1995). Beyond the limits of thought. Cambridge University Press.
Read, S. (2001). Self-reference and validity revisited. In M. Yrjönsuuri (Ed.), Medieval formal logic (pp.
183–196). Kluwer.
Reinhart, T. (1981). Pragmatics and linguistics: An analysis of sentence topics. Philosophica, 27(1), 53–94.
Roure, M. L. (1962). Le traité ‘Des Propositions Insolubles’ de Jean de Celaya. Archives D’histoire Doctri-
nale Et Littéraire Du Moyen Age, 29, 235–336.
Schilpp, P. A. (Ed.). (1944). The philosophy of Bertrand Russell. Open Court.
Simmons, K. (1993). Universality and the liar: An essay on truth and the diagonal argument. Cambridge
University Press.
Simmons, K. (2018). Semantic singularities: Paradoxes of reference, predication, and truth. Oxford Univer-
sity Press.
Simmons, K. (2021). Paradoxes of validity. Philosophical Studies, 179, 383–403.
Stalnaker, R. C. (1975). Indicative conditionals. Philosophia, 5, 269–286.
Tarski, A. (1933/1986). The concept of truth in formalized languages (pp. 152–278).
Tarski, A. (1969). Truth and proof. Scientific American, 220, 63–77.
Tarski, A. (1986). Logic, semantics, metamathematics (2nd ed.). Hackett.
Weber, Z., Ripley, D., Priest, G., Hyde, D., & Colyvan, M. (2014). Tolerating gluts. Mind, 123(491),
813–828.
Woodbridge, J., & Armour-Garb, B. (2008). The pathology of validity. Synthese, 160, 63–74.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and
applicable law.

13