Forthcoming in Unifying the Philosophy of Truth, ed. T. Achourioti, F. Fujimoto, H. Galinon, and J.
Martinez-Fernandez (Springer).
Validity and truth-preservation
Julien Murzi & Lionel Shapiro∗
December 6, 2012
Abstract
The revisionary approach to semantic paradox is commonly thought to have a
somewhat uncomfortable corollary, viz. that, on pain of triviality, we cannot
affirm that all valid arguments preserve truth (Beall, 2007, 2009; Field, 2008,
2009b). We show that the standard arguments for this conclusion all break
down once (i) the structural rule of contraction is restricted and (ii) how the
premises can be aggregated—so that they can be said to jointly entail a given
conclusion—is appropriately understood. In addition, we briefly rehearse
some reasons for restricting structural contraction.
Keywords: Truth-preservation · Validity · Naïve view of truth · Curry’s
Paradox · Contraction · Modus Ponens · Substructural logics · Incompleteness Theorems
Logical orthodoxy has it that valid arguments preserve truth (see e.g. Etchemendy,
1990; Harman, 1986, 2009):
(VTP) If an argument is valid, then, if all its premises are true, then its
conclusion is also true.
∗ University
of Kent and Munich Center for Mathematical Philosophy, Ludwig-Maximilians
Universität [j.murzi@kent.ac.uk] & University of Connecticut [lionel.shapiro@uconn.edu].
Thanks to Jc Beall, Colin Caret, Roy Cook, Charlie Donahue, Ole T. Hjortland, Jeff Ketland, Hannes
Leitgeb, Francesco Paoli, Stephen Read, Greg Restall for helpful discussion on some of the topics
discussed herein, and to Dave Ripley and a referee for detailed comments on a previous draft. Julien
Murzi warmly thanks the Alexander von Humboldt Foundation, the University of Padua, and the
School of European Culture and Languages at the University of Kent for generous financial support.
Lionel Shapiro is grateful to the Arché Research Centre at the University of St Andrews for making
possible a productive visit.
1
Intuitive as it may seem, this claim, on natural enough interpretations of ‘if’ and
‘true’, turns out to be highly problematic. Hartry Field has argued that its most
immediate justification requires all the logical and semantic resources that yield the
standard semantic version of Curry’s Paradox. Worse yet, both Field and Jc Beall
have observed that the claim that valid arguments preserve truth almost immediately yields absurdity via Curry-like reasoning in most logics (Field, 2008; Beall,
2007, 2009). Moreover, Field has argued that, by Gödel’s Second Incompleteness
Theorem, any semantic theory that declares all valid arguments truth-preserving
must be inconsistent (Field, 2006, 2008, 2009b,a). We can’t coherently require that
valid arguments preserve truth, or so the thought goes.1
Two main ingredients are required for this conclusion: that the conditional
occurring in VTP detaches, i.e. satisfies Modus Ponens, and the naïve view of truth,
viz. that (at the very least) the truth predicate must satisfy the (unrestricted) TScheme
(T-Scheme) Tr (pαq) ↔ α,
where Tr (...) expresses truth, and pαq is a name of α. Both assumptions lie at the
heart of the leading contemporary revisionary approaches to semantic paradox. These
include recent implementations (e.g. Brady, 2006; Field, 2003, 2007, 2008; Horsten,
2009) of the paracomplete approach inspired by Martin and Woodruff (1975) and
Kripke (1975), as well as paraconsistent approaches (see e.g. Asenjo, 1966; Asenjo
and Tamburino, 1975; Priest, 1979, 2006a,b; Beall, 2009). Paracomplete approaches
solve paradoxes such as the Liar by assigning the Liar sentence a value in between
truth and falsity, thus invalidating the Law of Excluded Middle. Paraconsistent
approaches solve the Liar by taking the Liar sentence to be both true and false,
avoiding absurdity by invalidating the classically and intuitionistically valid principle of Ex Contradictione Quodlibet. Both approaches have sought to preserve room
for a detaching conditional that underwrites the T-Scheme. And when such a conditional threatens to reintroduce absurdity through Curry’s Paradox, both approaches
have offered a common diagnosis: they take it to show that this conditional cannot
satisfy the law of contraction:
(Contraction) (α → (α → β)) → (α → β).
1 Shapiro
(2011) refers to the the claim that VTP and the naïve view of truth we introduce in the
next paragraph yield triviality as the ‘Field-Beall thesis’.
2
More generally, they require that a theory of truth be robustly contraction free (‘rcf’,
for short); free, essentially, of a a conditional satisfying Contraction and other
natural principles such as Modus Ponens (Restall, 1993).
In this paper, we assume for argument’s sake the naïve view of truth, and argue
that this view doesn’t in fact require rejecting VTP. However, maintaining VTP
requires more than revising logic so as to ensure that Contraction is no longer a
theorem. Rather, it involves adopting a logic that lacks one or more of the rules
usually thought to correspond to basic features of reasoning in the context of
assumptions. We will focus on the structural rule of contraction
(SContr)
Γ, α, α ⊢ β
Γ, α ⊢ β
Once SContr is rejected, we will see, the standard objections against VTP all break
down. The standard arguments against VTP at best support the weaker conclusion
that, given the naïve view of truth, either VTP or SContr (or perhaps some other
structural feature of the consequence relation) should be rejected.
To be sure, rcf theorists, especially Field, are aware of the existence of substructural revisionary approaches. Field dismisses them, though, as “radical,” (Field,
2008, p. 10) and as “very desperate measures” that are, ultimately, not needed
(Field, 2009a, p. 350). He writes:
I haven’t seen sufficient reason to explore this kind of approach (which
I find very hard to get my head around), since I believe we can do quite
well without it. ... [Hence] I will take the standard structural rules for
granted. (Field, 2008, pp. 10-11; also 283n)
However, while we agree with Field that more work needs to be done to make
sense of a failure of SContr, we’d like to stress that giving up VTP is also a radical
move. What is more, revisionary theorists have at least one powerful reason to
reject SContr. Let us assume, as is often done, that the “valid” arguments include
those whose goodness depends on rules governing the truth and validity predicates
(McGee, 1991; Whittle, 2004; Priest, 2006a,b; Field, 2007, 2008; Zardini, 2011). Then
there exist validity-involving versions of Curry’s Paradox which cannot be solved
by revising the logic’s operational rules (those governing the behavior of logical
vocabulary) to ensure that the theory is robustly contraction free. This is because
the only operational rules these versions of Curry’s Paradox employ are a pair
3
of rules governing a validity predicate, rules that are arguably essential to that
predicate’s expressing validity (Shapiro, 2011; Beall and Murzi, 2011).
It has long been known that Curry-paradoxical reasoning can be blocked by
adopting a “substructural” logic lacking SContr.2 Yet we’re not aware of any detailed examinations of how the various challenges to VTP are affected by adopting
such logics.3 What makes matters delicate is that all the challenges to VTP involve
arguments with multiple premises. Hence how we may respond to the challenges
depends crucially on how we understand what it means for a conclusion to follow
validly from all of the premises taken jointly. Even stating what truth-preservation
amounts to requires us to represent such joint consequence using some logical
connective in place of the above informal ‘all’ or (in the case of arguments with
finitely many premises) in place of the corresponding ‘and.’ Once SContr is rejected,
various possibilities open up for the logical behavior of such an ‘and’, with different choices having different implication for the challenges to VTP. Moreover, the
possibility arises that there are two suitable connectives, corresponding to different
modes in which premises may be understood as taken jointly. Our chief aim is
to clarify this poorly understood complex of issues and challenge the received
wisdom that VTP is incompatible with revisionary approaches to paradox.
Two final qualifications. The structural feature of validity encapsulated in SContr
isn’t the only standardly accepted structural feature whose rejection would block
the validity-involving versions of Curry’s Paradox and allow a defense of VTP
against the standard objections. An alternative “substructural” strategy, proposed
by Ripley (2011), involves restricting the transitivity of validity as reflected in the
structural rule of Cut.4 While we will occasionally remark on this approach, we
do not have space to compare it with the strategy of giving up SContr.5 In what
follows, we will assume (as rcf theorists typically do) that validity is transitive.
2 See
Slaney (1990), Restall (1994) and Field (2008, p. 283n).
is some relevant discussion in Shapiro (2011) and Zardini (2011).
4 Weir (2005) also addresses semantic paradox by restricting the transitivity of validity, though
this shows up in his natural deduction system as a structure-based restriction on the use of operational
rules.
5 Both of these “substructural” approaches to semantic paradox have an advantage worth mentioning: they allow for a unified approach to the paradoxes of self-reference (Weir, 2005; Zardini,
2011; Ripley, 2011), as opposed to the piecemeal approach proposed by current rcf theories, where
similar paradoxes, e.g. the Liar and Curry, are treated in radically different ways. In recent unpublished work, Beall uses the desideratum of uniformity as one motivation for a new approach
to paradox—one that retains the standardly accepted structural rules but gives up on a detaching
conditional altogether. For a sketch of that approach, see Beall (2011).
3 There
4
Likewise, we won’t here be able to discuss the various ways in which one might try
to make sense of and motivate the failure of SContr.6
The remainder of this paper is structured thus. §1 introduces the standard arguments in favor of rejecting VTP. §2 observes that VTP follows from what we call the
naïve view of validity, viz. that the validity predicate satisfies (generalisations of) the
Rule of Necessitation and the T axiom. It then rehearses some reasons for thinking
that the naïve view of validity is in tension with SContr, and considers a couple of
possible objections to this claim. §3 examines various possible interpretations of
VTP, interpretations that become available once SContr is rejected. Specifically, it
considers different ways of understanding the claim that an argument’s premises
are all true, as one finds in linear logic and what we call dual-bunching logics. It
then argues that, once SContr is rejected, the standard objections to VTP are all
blocked. §4 offers some concluding remarks.
1 Three challenges to VTP
We focus on three challenges to VTP: that the most obvious argument in defense of
this principle rests on inconsistent premises, that VTP yields triviality via Currylike reasoning, and that Gödel-like reasoning shows that no consistent recursively
axiomatizable semantic theory can endorse VTP.
1.1 The Validity Argument and Curry’s Paradox
Field (2008, §2.1, §19.2) considers an argument, which he calls the Validity Argument, to the effect that “an inference is valid if and only if it is logically necessary
that it preserves truth” (Field, 2008, p. 284). If sound, the argument for this biconditional’s ’only if’ direction would seem to establish VTP. However, Field argues,
it can’t be sound. Let’s use α1 , ..., αn ⊢ β to mean that ”the argument from the
premises α1 , ..., αn to the conclusion β is logically valid” (Field, 2008, p. 42). And
let Tr-I and Tr-E, respectively, be the rules that one may infer Tr (pαq) from α in any
context of assumptions, and vice versa. Then Field reasons thus (we have adapted
his terminology):
6 For
discussion of this important topic, see Shapiro (2011); Zardini (2011); Beall and Murzi (2011);
Mares and Paoli (2012).
5
‘Only if’ direction: Suppose α1 , ..., αn ⊢ β.
Then by Tr-E,
Tr (pα1 q), ..., Tr (pαn q) ⊢ β; and by Tr-I, Tr (pα1 q), ..., Tr (pαn q) ⊢ Tr (pβq).
By ∧-E, Tr (pα1 q) ∧ ... ∧ Tr (pαn q) ⊢ Tr (pβq). So by →-I, ⊢ Tr (pα1 q) ∧
... ∧ Tr (pαn q) → Tr (pβq). That is, the claim that if the premises α1 , ..., αn
are true, so is the conclusion, is valid, i.e. holds of logical necessity.
‘If’ direction: Suppose ⊢ Tr (pα1 q) ∧ ... ∧ Tr (pαn q) → Tr (pβq). By
Modus Ponens, Tr (pα1 q) ∧ ... ∧ Tr (pαn q) ⊢ Tr (pβq). So by ∧-I,
Tr (pα1 q), ..., Tr (pαn q) ⊢ Tr (pβq). So by Tr-I, α1 , ..., αn ⊢ Tr (pβq); and
by Tr-E, α1 , ..., αn ⊢ β. (Field, 2008, p. 284).7
Two features of this Validity Argument call for comment. First, notice that it is
conducted in a metalanguage containing a validity predicate (the turnstile), but no
truth predicate. In taking the argument to establish VTP, then, Field is assuming
that the object-language sentence Tr (pα1 q) ∧ ... ∧ Tr (pαn q) → Tr (pβq) expresses
the claim that if α1 , ..., αn are all true, so is β. In §3, we will see that once giving
up structural contraction is an option, it becomes controversial whether the claim
that “all premises are true” should be expressed using a connective for which the
inferences Field justifies using ∧-I and ∧-E are valid. Second, one might worry that
the Validity Argument presupposes its own conclusion. The argument establishes
that if an argument is valid, then the conditional claiming that the argument
preserves truth will likewise be valid. But we couldn’t take this as establishing VTP
itself unless we took for granted that valid sentences are true—a claim that is a special
case of VTP. Still, even if the Validity Argument doesn’t suffice to establish VTP, it
does undermine the objections that have been offered against VTP. That is because
these objections (which all involve multi-premise arguments) don’t purport to
challenge the claim that valid sentences are true. Thus the Validity Argument should
count as a defense of VTP.8
7 It
may help to make Field’s reasoning for the ’only if’ direction explicit in natural deduction
format, for the special case where we are considering an argument from the single premise α to the
conclusion β. Complications raised by the multiple-premise case will be discussed in §3.
Tr (pαq) ⊢ Tr (pαq)
Tr-E
α⊢β
Tr (pαq) ⊢ α
Cut
Tr (pαq) ⊢ β
Tr-I
Tr (pαq) ⊢ Tr (p βq)
→-I
⊢ Tr (pαq) → Tr (p βq)
8 In
§2.1, we will see that if our object-language contains a validity predicate, it is also possible
to derive VTP using an intuitively compelling elimination rule for that predicate. While we will
6
Field suggests that the Validity Argument, though it “looks thoroughly convincing at first sight,” can’t be accepted, since it relies on Tr-I, Tr-E, →-I, and →-E,
“which the Curry Paradox shows to be jointly inconsistent” (Field, 2008, pp. 43, 284).
Let us unpack this a little. The Diagonal Lemma allows us to construct a sentence
κ which, up to equivalence, intuitively says that, if it’s true, then (say) you will
win the lottery. Assuming that our theory of truth T is strong enough to prove the
Diagonal Lemma, this means that
⊢T κ ↔ ( Tr (pκq) → ⊥).
Let Π now be the following derivation of the further theorem Tr (pκq) → ⊥:
Tr (pκq) ⊢T Tr (pκq)
Tr-E
⊢T κ ↔ Tr (pκq) → ⊥
Tr (pκq) ⊢T κ
→-E
Tr (pκq) ⊢T Trpκq → ⊥
Tr (pκq) ⊢T Tr (pκq)
→-E
Tr (pκq), Tr (pκq) ⊢T ⊥
SContr
Tr (pκq) ⊢T ⊥
→-I
⊢T Tr (pκq) → ⊥
Using Π, we can then ‘prove’ that you will win the lottery:
Π
⊢T Tr (pκq) → ⊥
Π
⊢T Tr (pκq) → ⊥
⊢T κ ↔ ( Tr (pκq) ↔ ⊥)
⊢T κ
Tr-I
⊢T Tr (pκq)
→-E
⊢T ⊥
→-E
This is the (standard) conditional-involving version of Curry’s Paradox, or c-Curry,
as we’ll call it.9 The derivation makes use of Tr-I, Tr-E, →-I and →-E, just like
the Validity Argument. Hence, Field argues, one can’t accept the latter without
thereby validating the former. Rcf theorists invalidate c-Curry by rejecting →-I,
thus resisting Π’s final step (Priest, 2006b; Field, 2008; Beall, 2009; Beall and Murzi,
2011). Therefore, Field suggests, they must reject the ‘only if’ direction of the
Validity Argument, too.
However, as Field notes, the above derivation makes use of the rule SContr.
Hence if SContr is rejected—as proposed in this context by Brady (2006), Zardini
(2011), Shapiro (2011), and Beall and Murzi (2011)—Curry’s paradox no longer
discuss only a predicate that applies to single-premise arguments, a generalized version of that
derivation would be subject to all our conclusions about the Validity Argument.
9 This terminology was introduced in Beall and Murzi (2011).
7
stands in the way of our embracing the principles used in the Validity Argument for
VTP. One complication: we should note in advance that it isn’t clear that all types
of contraction-free logics we will be considering support theories of arithmetic that
prove a Diagonal Lemma. Where this isn’t the case, the reader should suppose that
some other means of self-reference built into our semantic theory is responsible
for the Curry paradoxes we will be considering. In what follows, we will ignore
this complication, and assume that T has the resources for at least simulating
self-reference.
Will rejecting SContr allow us to endorse the Validity Argument, then? As we
will see below, matters are not this simple. Field’s argument makes crucial use
of rules governing the conjunction symbolized by ∧. Once we no longer accept
the standard structural rules, however, the rules for conjunction can take nonequivalent forms, and the soundness of the Validity Argument now depends on
which of the available rules for ∧ we accept. In §3, we will examine which of the
contraction-free logics that have been proposed in response to semantic paradox
underwrite the Validity Argument.10
1.2 From VTP to absurdity via the Modus Ponens axiom
In addition to criticizing the most obvious defense of VTP, Field offers two arguments
according to which VTP can’t be embraced without absurdity. In the remainder of
this section, then, let us examine whether we can at least affirm that valid arguments
preserve truth. For simplicity’s sake, we focus for now on arguments with only one
premise. (Issues raised by multiple-premise arguments will be considered in detail
10 Let
us briefly consider how the Validity Argument fares on the alternative substructural approach that restricts transitivity. In the version of c-Curry given above, in natural deduction format,
SContr is the only structural rule used. By contrast, the parallel Curry derivation in sequent calculus
format will conclude with the following use of the structural rule of Cut
⊢T Tr (pκ q)
Tr (pκ q) ⊢T ⊥
⊢T ⊥
Ripley (2011) proposes a semantic theory that blocks c-Curry reasoning by invalidating Cut. His
theory adds rules for Tr to a sequent calculus with entirely classical operational rules and structural
rules except for Cut, which is no longer admissible in the presence of the truth rules. We would
like to make two observations about Ripley’s proposal. On the one hand, since it retains the rule
→-I, it allows a defense of the Validity Argument’s “only if” direction (his truth rules replace Cut
in the note above), and thus of VTP. On the other hand, though Ripley’s theory also endorses
the conclusion of every instance of the Validity Argument’s “if” direction, it won’t allow the above
intuitive argument, since it renders the rule →-E inadmissible. See note 46 below.
8
in §3 below.) We will try to affirm VTP in the object-language itself, by introducing
a predicate Val(x, y) which intuitively expresses that the argument from x to y is
valid. VTP may now be naturally represented thus (see Beall, 2009):
(V0) Val (pαq, pβq) → ( Tr (pαq) → Tr (pβq)).11
As Field and Beall point out, V0 entails absurdity, based on principles accepted by
rcf theorists (Field, 2006; Beall, 2007; Field, 2008; Beall, 2009).
Since rcf theorists do not accept the rule →-I, we will need two additional ingredients to obtain paradox from V0. First, the rules Tr-I and Tr-E no longer suffice;
our semantic theory T needs to underwrite all instances of the T-Scheme. Second,
we will use the principle that if ⊢T α ↔ β, then α and β are intersubstitutable within
conditionals.12 Given these presuppositions, V0 entails
(V1) Val (pαq, pβq) → (α → β).
Now let us assume, as rcf theorists do, that our theory T implies the validity of a
single-premise version of the Modus Ponens rule:
(VMP) Val (p(α → β) ∧ αq, pβq).
Hence V1 in turn entails the Modus Ponens axiom:
(MPA) (α → β) ∧ α → β.13
However, Meyer et al. (1979) show that MPA generates Curry’s Paradox. The only
additional ingredient we need is the claim that it is a theorem that “conjunction is
idempotent,” i.e. that ⊢ α ↔ α ∧ α.
To see why this is so, recall that we have assumed T is strong enough to ensure
⊢T κ ↔ ( Tr (pκq) → ⊥). Hence, given the T-Scheme and the above substitutivity
principle, ⊢T κ ↔ (κ → ⊥). We can now derive absurdity starting with the relevant
11 Strictly
speaking, this should be expressed a universal generalisation on codes of sentences, but,
for the sake of simplicity, we won’t bother.
12 This principle is endorsed by Field (2008, p. 253) and Beall (2009, pp. 28, 35).
13 Following Restall (1994), this is sometimes referred to as pseudo Modus Ponens. See also Priest
(1980), where it is described as the “counterfeit” Modus Ponens axiom.
9
instance of MPA:
(κ → ⊥) ∧ κ → ⊥.
Substituting κ for the equivalent κ → ⊥ gives us κ ∧ κ → ⊥. In view of our
assumption that ⊢T κ ↔ κ ∧ κ, another substitution of equivalents yields κ → ⊥.
By substituting κ for κ → ⊥ once again, we get κ. Finally, we use →-E to derive ⊥
from κ → ⊥ together with κ.
Since VTP and VMP jointly entail the paradox-generating MPA, it would thus
appear that rcf theorists can’t consistently assert that valid arguments preserve
truth.14 Field (2008, p. 377) and Beall (2009, p. 35) accept the foregoing argument,
and consequently reject the claim that valid arguments are guaranteed to preserve
truth (assuming, again, that truth-preservation is expressed using a detaching
conditional that underwrites the T-Scheme). The need to reject VTP is a perhaps
surprising, although ultimately unavoidable, corollary of the revisionary approach
to paradox, or so they argue.15
1.3 From VTP to inconsistency via the Consistency Argument
A second argument for rejecting VTP (Field, 2006, 2008, 2009b) proceeds via Gödel’s
Second Incompleteness Theorem, which states that no consistent recursively axiomatisable theory containing a modicum of arithmetic can prove its own consistency. Field first argues that if an otherwise suitable semantic theory could prove
that all its rules of inference preserve truth, it could prove its own consistency.
Hence, by Gödel’s theorem, no semantic theory that qualifies as a “remotely adequate mathematical theory” can prove that its rules of inference preserve truth. Yet
insofar as we endorse the orthodox semantic principle VTP, Field says, we should
be able to consistently add to our semantic theory an axiom stating that its rules
of inference preserve truth (see Field, 2009a, p. 351n10). Hence, he concludes, we
should reject VTP.
To establish the first step in this argument against VTP, Field considers what
he calls the Consistency Argument (Field, 2006, pp. 567-8). This is an argument
14 See
Beall (2007), Beall (2009, pp. 34-41), Shapiro (2011, p. 341) and Beall and Murzi (2011).
Beall,
by contrast, does accept the negation of VTP. Indeed, he accepts that there are valid arguments, e.g.
the argument from κ and κ → ⊥ to ⊥, that fail to preserve truth. However, as Field and Beall both
note, Beall’s position doesn’t require accepting that there are valid arguments whose premises are all
true and whose conclusion is false. See Field (2006, p. 597) and Beall (2009, p. 36).
15 For Field, who rejects excluded middle, rejecting VTP doesn’t mean accepting its negation.
10
which, one might think, one should be able to run within any theory T containing a
truth predicate satisfying the unrestricted T-Scheme. The argument proceeds by “(i)
inductively proving within T that all its theorems are true, and (ii) inferring from
the truth of all theorems of T that T is consistent.” Though intuitively sound, the
Consistency Argument must fail if T is to be consistent.
Field’s claim is that the failure of the Consistency Argument must be blamed
on an illicit appeal to VTP. He observes that (ii) can’t be problematic for those
who hold that “inconsistencies imply everything.” The target theories “certainly
imply ¬ Tr (p0 = 1q), so the soundness of T would imply that ‘0 = 1’ isn’t a theorem
of T; and this implies that T is consistent” (Field, 2008, p. 286-7). However, (ii)
will be equally unproblematic for any paraconsistent theorist who holds that an
adequate semantic theory must imply the universal generalization over instances
of the schema ¬ Tr (pα ∧ ¬αq). In this case as well, if T could prove that all its
theorems are true, it would thereby prove that no contradiction is a theorem (Field,
2006, pp. 593-5). Field therefore concludes that the problem with the Consistency
Argument must lie with (i). The argument by induction alluded to in (i) proceeds
as follows: “(1) Each axiom of T is true, (2) Each rule of inference of T preserves
truth [in the sense of VTP, whence] (3) All theorems of T are true.” Field argues
persuasively that “[t]he only place that the argument can conceivably go wrong is ...
in (2)” (Field, 2008, p. 287). This conclusion is endorsed by Beall (2009, pp. 115-6).
In sum, not only does the seemingly obvious Validity Argument in favor of
VTP fail, but there are at least two arguments against accepting VTP—or so contemporary revisionary wisdom goes. As Beall writes: “such a claim ... needs to be
rejected, and I reject it” (Beall, 2009, p. 35).
2 Naïve validity and Validity Curry
What role, then, if any, is left for the notion of validity, if we can no longer affirm
that valid arguments preserve truth? Field (2008, 2009b, 2010) suggests that validity
normatively constrains belief: very roughly, one shouldn’t fully believe the premises
of a valid argument without fully believing its conclusion. We take no position here
on whether the role of the notion of validity can be explained without recourse
to truth-preservation.16 Instead, we’ll suggest in the remainder of this paper that
16 For
the record, we think that even if VTP holds, an explanation of the role of the notion of
validity will have to involve normative considerations such as those Field advances.
11
revisionary theorists need not and should not reject VTP. Provided they accept certain
basic principles that would appear to govern the notion of validity, revisionary
theorists are required on pain of paradox to adopt the very kind of logic that allows
them to embrace VTP.
2.1 Naïve validity
Still restricting our attention to single-premise arguments, consider the following
two principles for the use of the validity predicate: that, if one can derive ψ from φ,
one can derive on no assumptions that the argument from φ to ψ is valid, and that,
from φ and the claim that the argument from φ to ψ is valid, one can infer ψ.17
Both rules are highly intuitive. If Val ( x, y) expresses validity, it seems natural to
assume that an adequate semantic theory T must include the following introduction
rule for Val ( x, y), which, by analogy with →-I or Conditional Proof, we’ll call
Validity Proof :
(VP)
α ⊢T β
⊢T Val (pαq, pβq)
If T’s rules are valid, and we can derive β from α in T, then T must be able to assert
the sentence Val (pαq, pβq), expressing that the argument from α to β is valid. But
it also seems natural to assume that T contains an elimination rule for Val ( x, y),
which we’ll call Validity Detachment:
(VD)
Γ ⊢T Val (pαq, pβq)
Γ, ∆ ⊢T β
∆ ⊢T α
If, from a given context of assumptions, we can derive in T the sentence α and from
another context we can derive that the argument from α to β is valid, then it must
be possible (from the assumptions taken together) to derive β.18
17 To
the best of our knowledge, these rules are first discussed in Priest (2010). For further
discussion, see Beall and Murzi (2011) and Murzi (2011). Shapiro (2011) proposes introducing a
validity predicate governed by the equivalences Val (pαq, p βq) ⊣⊢T α ⇒ β, where ⇒ is an entailment
connective whose introduction and elimination rules in turn render VP and VD derivable. Such a
connective is common in the tradition of relevant and paraconsistent logic: see e.g. Anderson and
Belnap (1975, p. 7) and Priest and Routley (1982).
18 We have written the rule VP without side assumptions. That is because the acceptability of a
version including side assumptions
(VP∗ )
Γ, α ⊢T β
Γ ⊢T Val (pαq, p βq)
12
The rules VP and VD can also be viewed as generalizations of natural rules for a
predicate that expresses logical truth: namely, analogues of the rule of Necessitation
and of a rule corresponding to the T axiom. To see this, it is sufficient to instantiate
VP and VD using a constant T expressing logical truth. Instantiating VP yields
a notational variant of Necessitation, rewritten using our two place predicate
Val ( x, y) in place of a necessity operator:
(NEC∗ )
T ⊢T β
⊢T Val (pTq, pβq)
Likewise, instantiating VD thus
Γ ⊢T Val (pTq, pβq)
Γ, T ⊢T β
T ⊢T T
yields a notational variant of a rule corresponding to the T axiom for a necessity
operator:
(T∗ ) Val (pTq, pβq), T ⊢T β
The intuitiveness of our rules VP and VD is thus underscored by the close connection
they underwrite between the behavior of a predicate expressing logical truth and
the behavior of an operator expressing logical necessity.
We will therefore call the view that ‘valid’ satisfies VP and VD the naïve view of
validity (Murzi, 2011). One first point that deserves emphasis is that, on the naïve
truth of truth we’ve assumed at the beginning of this paper, such a view entails V0,
our object-language statement of VTP for single-premise arguments. This can be
shown using what is essentially a version of Field’s Validity Argument, except that
the validity of the argument from α to β is now expressed using an object-language
predicate rather than using a turnstile in the metalanguage:19
depends on the properties of the structural comma. For example, if the comma obeys weakening
and we get β, α ⊢T β, then VP∗ allows us to derive β ⊢T Val (pαq, p βq). But where β is contingent, it shouldn’t follow from β that it is entailed by any sentence. A similar problem arises if the
comma obeys exchange. From VD and Cut we get Val (pαq, pαq), α ⊢T α, whence exchange yields
α, Val (pαq, pαq) ⊢T α and VP∗ allows us to derive α ⊢T Val (pVal (pαq, pαq)q, pαq). But if α is contingent, it shouldn’t follow from α that it is entailed by a logical truth. Zardini (2012), whose comma
obeys both weakening and exchange, avoids these problems by restricting the side assumptions in
VP∗ to logical compounds of validity claims. See also Priest and Routley (1982).
19 Ripley (2011) offers a similar defense of VTP, using VP and the sequent α, Val (pαq, p βq) ⊢ β.
T
Shapiro (2011) explains that on the version of the naïve view presented there (see note 17 above),
Val (pαq, p βq) implies Tr (pαq) ⇒ Tr (p βq).
13
Tr (pαq) ⊢T Tr (pαq)
Tr-E
Val(pαq, pβq) ⊢T Val(pαq, pβq)
Tr (pαq) ⊢T α
VD
Val(pαq, pβq), Tr (pαq) ⊢T β
Tr-I
Val(pαq, pβq), Tr (pαq) ⊢T Tr (pβq)
→-I
Val(pαq, pβq) ⊢T Tr (pαq) → Tr (pβq)
→-I
⊢T Val(pαq, pβq) → ( Tr (pαq) → Tr (pβq))
A second point to notice is that, natural though they may seem, VP and VD lead us
into trouble—which should of course be expected, since NEC∗ and T∗ are nothing
but the key ingredients of the Myhill-Kaplan-Montague Paradox, or Paradox of the
Knower (Myhill, 1960; Kaplan and Montague, 1960; Murzi, 2011).20
2.2 Validity Curry
The Diagonal Lemma allows us to construct a sentence π, which intuitively says of
itself, up to equivalence, that it validly entails that you will win the lottery:
⊢T π ↔ Val (pπq, p⊥q)
Let Σ now be the following derivation of the further theorem Val (pπq, p⊥q):
π ⊢T π
⊢T π ↔ Val (pπq, p⊥q)
→-E
π ⊢T Val (pπq, p⊥q)
π ⊢T π
VD
π, π ⊢T ⊥
SContr
π ⊢T ⊥
VP
⊢T Val (pπq, p⊥q)
Using Σ, we can then ‘prove’ that you will win the lottery
Σ
⊢T Val (pπq, p⊥q)
Σ
⊢T π ↔ Val (pπq, p⊥q)
⊢T Val (pπq, p⊥q)
→-E
⊢T π
VD
⊢T ⊥
Our revisionary theory of truth and validity, T, proves on no assumptions that you
will win the lottery.21 Call this the Validity Curry, or v-Curry, for short, to contrast
20 Shapiro
(2011) identifies two challenges to the naïve view: a “direct argument” that it leads
straight to paradox, and an “indirect argument” that it entails a version of the paradox-producing
VTP.
21 To the best of our knowledge, the first known occurrence of the Validity Curry is in the 16thcentury author Jean de Celaya. See Read (2001, fn. 11-12) and references therein. Albert of Saxony
discusses a contrapositive version of the paradox in his Insolubles (Read, 2010, p. 211). A more
14
it with the standard conditional-involving version of Curry’s Paradox, or c-Curry.22
As we explained above, rcf theorists invalidate c-Curry by rejecting →-I. Unlike
c-Curry, however, the v-Curry Paradox makes no use of →-I, and hence it cannot
be invalidated by rejecting such a rule. On the other hand, the above derivation
of v-Curry presupposes SContr (Beall and Murzi, 2011). Hence if VP and VD hold,
there is only one revisionary way out of the v-Curry Paradox, viz. rejecting SContr,
thus adopting a substructural logic—a logic where some of the standardly accepted
structural rules fail (Shapiro, 2011; Beall and Murzi, 2011; Murzi, 2011; Zardini,
2011).23
Before examining in §3 how rejecting SContr affects VTP and the Validity Argument, we’d first like to offer a partial defence of our claim that v-Curry Paradox is a
reason for revisionary logician to adopt a substructural logic. To this end, we’ll consider in the next section two natural responses to the claim that the Validity Curry
is a genuine paradox of validity, and offer replies on the substructural logician’s
behalf.
2.3 A genuine paradox of validity
If the v-Curry Paradox isn’t a genuine paradox of validity, one of VP and VD must
not unrestrictedly hold. As it turns out, there are prima facie compelling reasons for
restricting both.24 One argument against VP runs thus. One simply notices that the
subproof in the v-Curry derivation relies on a substitution instance of the logically
invalid biconditional proved by the Diagonal Lemma, viz. π ↔ Val (pπq, p⊥q), and
hence isn’t logically valid, contrary to what an application of VP at the end of the
subproof would require. Furthermore, in both versions of the v-Curry Paradox, VD
recent version can be found in Priest and Routley (1982), and surfaces again in Whittle (2004, fn.
3), Clark (2007, pp. 234-5) and Shapiro (2011, fn. 29). For a first comprehensive discussion of the
Validity Curry, see Beall and Murzi (2011). For a defence of the claim that Validity Curry is a genuine
paradox of validity, see §2.3 below and Murzi (2011).
22 This terminology was first introduced in Beall and Murzi (2011). Ultimately, however, the
distinction in terms of predicate versus connective may not be the essential one. Whittle (2004)
and Shapiro (2011) discuss a version of Curry’s Paradox, involving a “consequence connective” or
“entailment connective,” which poses much the same challenge to rcf theorists as does v-Curry.
23 For an early anticipation of the argument from naïve validity to the rejection of SContr (in the
form of multiple discharge of assumptions), see Priest and Routley (1982). Priest and Routley, whose
entailment connective obeys analogues of VP and VD, discuss several resulting paradoxes which
they blame on the “suppression of innocent premises.” By contrast, Ripley (2011) blocks v-Curry at
the final step using VD, which is inadmissible in his nontransitive theory for the same reason that
→-E is inadmissible. See note 45 below.
24 Thanks to Roy Cook and Jeff Ketland for raising these potential concerns.
15
gets used in the subderivation, and, it might be objected, surely such a rule isn’t
logical. More precisely, Roy Cook (2012) has argued that the T-Scheme isn’t logically
valid, if by logical validity one means truth under all uniform interpretations of the
non-logical vocabulary. Cook’s reasoning would apply equally to the status of VP
and VD.25
These objections have an important virtue: they help us understand what the
v-Curry Paradox really is a paradox of. More precisely, they show that the v-Curry
Paradox is not paradox of purely logical, or interpretational, in John Etchemendy’s
term, validity (Etchemendy, 1990).26 Indeed, a recent result by Jeff Ketland shows
that purely logical validity cannot be paradoxical. Ketland (2012) proves that Peano
Arithmetic (PA) can be conservatively extended by means of a predicate expressing
logical validity, governed by intuitive principles that are themselves derivable in
PA. It follows that purely logical validity is a consistent notion if PA is consistent,
which should be enough to warrant belief that purely logical validity simply is
consistent.
However, it seems to us that there are broader notions of validity than purely
logical validity. 27 Thus, neither of the above objections applies to versions of the vCurry Paradox in which ‘valid’ expresses representational validity, whereby (roughly)
validity is equated with preservation of truth in all possible circumstances (Read,
1988; Etchemendy, 1990; McGee, 1991). But VP, VD and the arithmetic required
to prove the Diagonal Lemma are, at least intuitively, valid in this sense.28 Nor
25 Field
(2008, §20.4) himself advances versions of this line of argument, while discussing what is
in effect a validity-involving version of the Knower Paradox resting on NEC∗ and T∗ . See especially
Field (2008, p. 304 and p. 306). On the question whether his conception of the extension of the
validity predicate consistently allows him to do so, see note 25 below.
26 Here we take the logical vocabulary to be the standard vocabulary of some first-order, perhaps
non-classical, logic.
27 Several semantic theorists, including rcf theorists such as Field and Priest, resort to non purely
logical notions of validity. For instance, Field (2007, 2008) extensionally identifies validity with,
essentially, preservation of truth in all ZFC models of a certain kind, thus taking validity to (wildly)
exceed purely logical validity. (Incidentally, it seems to us that this use of ‘valid’ is in tension with
the purely logical sense Field (2008) appeals to at p. 304 and especially p. 306.) Likewise, McGee
(1991, p. 43-9) takes logical necessity to extend to arithmetic and truth-theoretic principles.
28 It might be objected that such a notion of validity presupposes VTP, and hence cannot be
appealed to in the present context, where the question whether VTP can be consistently upheld
is the very point at issue. Our modest aim here, however, is simply to suggest that someone who
already thinks, following perhaps logical orthodoxy, that valid arguments preserve truth and that,
accordingly, consequence is to be explicated in terms of truth-preservation has a reason—the v-Curry
Paradox—to reject SContr. Once SContr is rejected, the standard challenges to VTP no longer stand,
as we’ll see in §3 below. But, it seems to us, no illicit or question-begging appeal to VTP has been
made in the course of the foregoing reasoning. We thank an anonymous referee for raising this
16
does the objection that VP cannot be legitimately applied to non-purely logical
subderivations apply to conceptions of validity which take ‘valid’ to express the
consequence relation of one’s semantic theory, provided that the naïve validity
rules and enough arithmetic are part of that relation.29 Insofar as VD is valid in one
of these broader senses, and insofar as the VP and VD govern the use a predicate
expressing validity in that sense, there is at least one—important—reading of ‘valid’
on which the use of VP in the v-Curry derivation is sound. The v-Curry Paradox is
a paradox of validity, not purely logical validity.
To be sure, one might instead either reject VP on different grounds, or perhaps
reject VD. One natural enough argument against the latter rule runs thus. Suppose
validity is recursive. Then, one might argue, T∗ , and hence VD, must fail. For, if
validity is recursively enumerable, an argument is valid if and only if its conclusion
can be derived from its premises in some recursively axiomatisable theory T. That
is, the validity predicate Val ( x, y) is just a notational variant of Prov T ( x, y), where
this expresses that there is a T-derivation of y from x. Yet, the argument continues,
we know from Löb’s Theorem that, if T contains enough arithmetic (if it proves
the so-called derivability conditions), T cannot contain, on pain of triviality, all
instances of the provability-in-T analogue of T∗ , Prov T (pTq, pαq) → α. Hence, one
might conclude, T may not contain all instances of T∗ either, and hence of VD, a
fortiori.
We find this conclusion problematic. It seems to us that rejecting VD, or VP, for
that matter, isn’t really a comfortable option for proponents of the naïve view of
truth. In a nutshell, together with the naïve view of truth, the naïve view of validity
is but an instance of the general thought underpinning the revisionary approach to
paradox—what we may call the naïve view of semantic properties. 30 This is the view
that one cannot revise naïve semantic principles without thereby also revising naïve
potential concern.
29 In recent unpublished work, Cook in fact shows how this response can be strengthened: it is
possible to formulate a modified Validity Curry paradox in such a way that the arithmetic necessary
to prove the Diagonal Lemma need not be included in the scope of the validity relation.
30 This view is implicitly assumed in the work of contemporary revisionary theorists—see e.g.
Priest (2006b), Field (2007, 2008); Beall (2009); Beall and Murzi (2011). In particular, it is implicit in
their assumption that the paradoxes of validity are (in an interesting sense) of the same kind as the
Liar and -Curry. One defence of that possibly controversial assumption would involve arguing that
the Paradox of the Knower is nothing but a weakened Liar, and that, as we’ve observed in §2.1,
v-Curry is nothing but a generalised Knower, so that whatever the nature of the first paradox, it is
inherited by the other two. See also Read (2001) and Beall and Murzi (2011). We should finally stress
that in calling validity a semantic property, we merely intend to point to these parallels, without
relying on any particular conception of what makes a property semantic.
17
semantic properties, and that, on pain of triviality, semantic properties should be
held fixed, and logic must change. Arguably, the naïve view of semantic properties
has it that validity is factive, and that we, and hence our semantic theory, must
be able to say so, on pain of not being able to consistently assert what we know
to be true. If T does indeed meet the conditions for Löb’s Theorem, we would
like to suggest, then the correct reaction to the objection is instead to concede that
Val ( x, y) can’t be replaced with Prov T ( x, y), and hence that that naïve validity is
not recursively enumerable.31
It might be objected that we could revise, or refine, our naïve conception of
validity, which is after all naïve (McGee, 1991, p. 45). But, then, a parallel argument
would show that, when faced with the Liar Paradox, the c-Curry Paradox, and
other paradoxes of truth, we should similarly revise our conception of truth, which
is precisely what proponents of the naïve view of semantic properties take to be
the wrong response to semantic paradox. For the time being, we’ll assume that
the Validity Curry is a genuine paradox of validity, and that giving up SContr, as
suggested in Shapiro (2011) and Zardini (2011), is a legitimate revisionary response
to it, and to semantic paradoxes more generally. We shall now argue that, on this
admittedly controversial assumption, of which we’ve only offered a partial defence,
all three arguments for rejecting VTP break down.
3 Validity and truth-preservation
All three challenges to VTP turn out to rest crucially on how our object-language
expresses validity and truth-preservation for arguments with multiple premises.
First, recall that Field argues that the most obvious defense of VTP, the Validity Argument, rests on principles that yield paradox. As we have pointed out,
the Validity Argument presupposes that the truth-preservingness of an inference from α1 , ..., αn to β can be expressed using the object-language sentence
Tr (pα1 q) ∧ ... ∧ Tr (pαn q) → Tr (pβq). Second, the argument from VTP to PMP
31 We
don’t have space to expand on this point here. Priest (2006b, §3.2) argues at length that
the “naïve notion of proof” is recursive, whence naïve provability, a species of naïve validity, is
recursively enumerable. Here we simply notice that his arguments are consistent with the view that
naïve validity isn’t. Finally, we’d like to point out that some SContr-free semantic theories extending
contraction-free arithmetics may not be strong enough to satisfy Löb’s Theorem’s applicability
conditions, in which case the objection from Löb’s Theorem we are considering would not apply in
the first place.
18
and absurdity used the simplifying assumption that the validity of the two-premise
Modus Ponens rule can be expressed using a single-premise validity predicate as
Val (pα ∧ (α → β)q, pβq). Finally, spelling out the Consistency Argument requires
expressing in the object-language the claim that each of our semantic theory T’s
rules of inference preserves truth, where these will include multi-premise rules
such as Modus Ponens.
3.1 Premise-aggregating connectives
We will therefore assume that truth-preservation and validity for arguments with
a finite number of premises can be expressed using some “premise-aggregating
connective” ⊙:32
(a) The claim that the argument from premises α1 , ..., αn , taken together, to
conclusion β preserves truth can be expressed in the object-language as
Tr (pα1 q) ⊙ ... ⊙ Tr (pαn q) → Tr (pβq).
(b) The claim that the argument from premises α1 , ..., αn , taken together, to conclusion β is valid can be expressed using the object-language’s binary validity
predicate as Val (pα1 ⊙ ... ⊙ αn q, pβq).
Is there an understanding of the logical behavior of ⊙ on which (a) and (b) are true,
but each of our three challenges to VTP is blocked?
Before examining the three challenges in turn, we now consider the chief options
for the rules governing ⊙ in the context of a substructural natural deduction system.
For the time being, we will work within a structural framework in which the “taking
together” of assumptions—which we have indicated with commas to the left of the
turnstile—can be represented using “multisets.” These are structures that behave
like sets except for the fact that they keep track of the number of occurrences of
each member (Meyer and McRobbie, 1982a,b). The philosophical significance of
multiset structure in natural deduction has been explained in many ways, and the
same is the case for the more complex structure we will consider later. This isn’t the
place to compare various interpretations or defend one of them.33 Our aim, rather,
32 For
arguments with an infinite number of premises, we will need universal quantification to
express truth-preservation. None of the objections to VTP we will consider, however, depend on
consideration of infinite-premise arguments.
33 We have each made different suggestions in previous work: Shapiro (2011) and Beall and Murzi
(2011). See also Read (1988), Slaney (1990), Restall (2000), and Paoli (2002).
19
is to explain how moving to a deduction system in which the structure referred to
on the left of the turnstile is finer-grained than a set affects the standard objections
to VTP.
Using multisets rather than (e.g.) sequences renders redundant Gentzen’s
structural rule of exchange:
(SExch)
Γ, α, β ⊢ γ
Γ, β, α ⊢ γ
By contrast, SContr isn’t redundant, nor is the structural rule of weakening:
(SWeak)
Γ, α ⊢ γ
Γ, β, α ⊢ γ
Indeed, once one or more of SContr and SWeak is rejected, one can formulate
operational rules for two different connectives, rules that become equivalent only in
the presence of both SContr and SWeak. These are the rules that govern, respectively,
the “multiplicative” and “additive” conjunctions of linear logic, a multiset-based
logic in which both SWeak and SContr are rejected (Girard, 1987) :34
(⊗-I)
(&-I)
Γ⊢α
∆⊢β
Γ, ∆ ⊢ α ⊗ β
Γ⊢α
Γ⊢β
Γ⊢α&β
(⊗-E)
(&-E1)
Γ, α, β ⊢ γ
∆ ⊢ α⊗β
Γ, ∆ ⊢ γ
Γ⊢α&β
Γ⊢α
(&-E2)
Γ⊢α&β
Γ⊢β
Since it will be important later, we note that the structural comma appears in
the rules for the multiplicative ⊗, whereas it does not appear in the rules for
the additive &. In the terminology of Belnap (1982, 1993), the additive rules are
“structure-free” while the multiplicative rules are “structure-dependent.” Finally, in
this structural setting, our assumption of the transitivity of validity can be codified
using the following version of the cut rule:
(Cut)
Γ⊢α
∆, α ⊢ β
∆, Γ ⊢ β
34 While
linear logic is standardly presented in sequent calculus format, the above natural deduction rules appear in Troelstra (1992, p. 57) and O’Hearn and Pym (1999).
20
3.2 The Validity Argument
The first point we would like to make is that, in the absence of SContr, the ’only if’
direction of the Validity Argument (the direction that would establish VTP) fails
when the premise-aggregating connective ⊙ is construed as the additive & in a
multiset-based logic.
To see why, note that when rewritten using &, this direction of the Validity Argument requires deriving Tr (pα1 q) & ... & Tr (pαn q) ⊢ Tr (pβq) from Tr (α1 ), ..., Tr (αn ) ⊢
Tr ( β). That in turn requires n − 1 uses of the inference pattern
(&-L)
Γ, α1 , α2 ⊢ β
Γ, α1 & α2 ⊢ β
Field himself justifies this inference by appeal to the rule &-E. Indeed, in the
presence of SContr, either of our twin elimination rules &-E1 and &-E2 yields &-L.
Here is a derivation using &-E2, SContr, Cut, and the reflexivity of validity:
α1 & α2 ⊢ α1 & α2
&-E2
α1 & α2 ⊢ α2
Γ, α1 , α2 ⊢ β
Cut
Γ, α1 , α1 & α2 ⊢ β
α1 & α2 ⊢ α1
Cut
Γ, α1 & α2 , α1 & α2 ⊢ β
SContr
Γ, α1 & α2 ⊢ β
In a logic without SContr, on the other hand, &-L fails. Moreover, this remains the
case if we accept SWeak, thus strengthening linear logic into what is known as an
“affine” logic.35
Hence, insofar as we wish to preserve the Validity Argument while rejecting
SCont (and thus avoiding c-Curry and v-Curry), we ought not interpret the premiseaggregating ⊙ as the additive conjunction & of a multiset-based logic. On the
other hand, both directions of the Validity Argument go through, even in the
absence of SContr, provided that ⊙ is construed as the multiplicative ⊗. Given
α1 ⊗ α2 ⊢ α1 ⊗ α2 , the rule ⊗-E immediately yields the inference required for the
argument’s “only if” direction:36
35 In
that case, however, the “if” direction of the Validity Argument will go through
for & as premise-aggregating connective. Deriving Tr (pα1 q), ..., Tr (pαn q) ⊢ Tr (p βq) from
Tr (pα1 q) & ... & Tr (pαn q) ⊢ Tr (p βq) requires the inverse of &-L, which obtains in the presence of
SWeak.
36 In single-conclusion sequent calculus formulations (which suffice for our purposes, as our
derivations all involve the language’s negation-free fragment), the connective ⊗ is governed by the
twin rules ⊗-I and ⊗-L.
21
(⊗-L)
Γ, α1 , α2 ⊢ β
Γ, α1 ⊗ α2 ⊢ β
Indeed, with ⊗ as premise-aggregating connective, Elia Zardini (2011) has recently
proved a generalization of the Validity Argument’s “only if” conclusion.37 And the
“if” direction is no harder to establish.
Summarizing, we can say that Field’s objection to the “only if” direction of
the Validity Argument fails when our semantic theory is based on an underlying
logic that lacks SContr, as long as this logic is multiset-based and we state the
argument’s conclusion using multiplicative conjunction. Admittedly, this method
of vindicating the Validity Argument carries a cost. Multiset-based logics can
contain no connectives that behave like the conjunction or disjunction of classical
logic (e.g. Belnap, 1993). In the case of the additive connectives, for example, we
lose Distribution: α & ( β ∨ γ) ⊢ (α & β) ∨ (α & γ). On the multiplicative side, we
lose Simplification: α ⊗ β ⊢ α. Adding the rule SWeak, as Zardini proposes, restores
the latter. But, as we will see below, we still lose Square-increasingness: α ⊢ α ⊗ α.
However, adopting a multiset-based logic isn’t the only way to vindicate the
Validity Argument by rejecting a structural contraction rule. A second way is to use
one of the many substructural logics in which assumptions are regarded as “taken
together” in two different ways. In such “dual-bunching” logics, the structures
referred to on the left of the turnstile are not multisets, but rather finer-grained
“bunches” specified using two different punctuation marks (Read, 1988; Slaney,
1990; Restall, 2000). This alternative is of interest for two reasons. First, unlike
multiset-based logics, dual-bunching logics do feature connectives whose behavior
is classical to the extent that they satisfy Distribution, Simplification, and Squareincreasingness. Secondly, as we will see in §3.3, multiset-based and dual-bunching
logics underwrite different interpretations of the way in which rejecting structural
contraction blocks the argument against VTP via the Modus Ponens axiom.
The first kind of bunching is used to formulate all the structure-dependent operational rules. For this reason, it will be convenient to indicate this kind of bunching
using the comma (though the semicolon is more standard). That way, we can retain
our rules →-I, →-E, VD and ⊗-I, as long as Γ and ∆ are now understood as bunches
rather than multisets. On the other hand, we need a generalized version of ⊗-E,
37 Field’s
own reasoning, as sketched in §1.3, amounts to a special case of Zardini’s proof: the case
in which we are considering the truth-preservingness of a single-conclusion argument and employ
no side assumptions. Zardini’s proof does not depend on his acceptance of SWeak.
22
where ∆(α, β) stands for any bunch of which α, β is a subbunch:38
(⊗-Edb )
∆(α, β) ⊢ γ
Γ ⊢ α⊗β
∆(Γ) ⊢ γ
In dual-bunching logics, one or more the standard structural rules SContr, SWeak or
SExch is rejected for the comma.39 Just as for multiset-based logics, rejecting SContr
suffices to block the above derivations of c-Curry and v-Curry.
What is distinctive about dual-bunching logics is the introduction of a second
kind of bunching of assumptions, which we will indicate using the colon. This
“extensional” bunching obeys all the standard structural rules:
Γ(∆ : ∆) ⊢ β
Γ(∆) ⊢ γ
Γ(∆ : ∆′ ) ⊢ γ
(eSWeak)
( eSExch)
(eSContr)
Γ(∆′ : ∆) ⊢ γ
Γ(∆) ⊢ β
Γ(∆′ : ∆) ⊢ γ
Unlike the “intensional” comma, the colon need not get mentioned in operational
rules for any connective.40
We are now ready to consider how the Validity Argument fares for dualbunching logics. First, the reasoning challenged by Field goes through provided
the conclusion is formulated using the structural comma together with the multiplicative ⊗ as the premise-aggregating connective. That is because we retain ⊗-L,
now generalizable to
(⊗-Ldb )
Γ ( α1 , α2 ) ⊢ β
Γ ( α1 ⊗ α2 ) ⊢ β
Construed this way, the Validity Argument’s “only if” direction establishes that
α1 , ..., αn ⊢T β only if ⊢T Tr (pα1 q) ⊗ ... ⊗ Tr (pαn q) → Tr (pβq). Moreover, a parallel
result now holds for the connective &, known in this structural context as “extensional” conjunction.41 This is because the fact that the colon obeys eSContr allows
us to replicate the above derivation of &-L, yielding
38 For
definitions, see Read (1988, §4.1) and Restall (2000, pp. 19-20). In sequent calculus formulations, ⊗-Edb is replaced by ⊗-L. Sequent calculi of this type were developed independently for
fragments of relevant logics by Minc (1976) and by Dunn, whose version appears in Anderson
and Belnap (1975, §28.5). For natural deduction formulations, see Read (1988), Slaney (1990) and
O’Hearn and Pym (1999), whose use of the comma we follow.
39 Rather than rejecting the structural rule of associativity, we are avoiding the need for such a rule
by allowing our comma to retain its variable polyadicity.
40 Since assumptions can now be embedded in bunches specified using both comma and colon,
we also need to generalize our statement of the cut rule:
(Cutdb )
Γ⊢α
∆(α) ⊢ β
∆(Γ) ⊢ β
41 But
see (Paoli, 2007, pp. 569-71) for opposition to the standard claim that the extensional
conjunction of such logics is “truth functional.”
23
(&-Ldb )
Γ ( α1 : α2 ) ⊢ β
Γ ( α1 & α2 ) ⊢ β
Accordingly, the Validity Argument also goes through when the conclusion is
formulated using the structural colon together with & as the premise-aggregating
connective. Construed this way, it establishes that α1 : ... : αn ⊢T β only if
⊢T Tr (pα1 q) & ... & Tr (pαn q) → Tr (pβq).42 According to dual-bunching logics,
then, there are different kinds of multi-premise arguments, represented using different
antecedent structure, and the validity of each kind of argument entails a different kind of truth-preservation, expressed in the object-language using different
premise-aggregating connectives.
There are thus at least two general ways to vindicate the Validity Argument by
rejecting SContr: one can use a multiset-based logic with multiplicative conjunction
as premise-aggregating connective, or a dual-bunching logic. Versions of both
approaches are known to make possible a naïve theory of truth (either a consistent
paracomplete theory or a nontrivial paraconsistent theory).43 We will return to the
difference between the two approaches in the next section. For now, we merely
note that they yield logics that conflict for the fragment of the language whose
only connectives are & and the corresponding disjunction ∨. Recall that the rules
for these connectives don’t even mention the nonstandard comma structure. It
follows that on the dual-bunching approach, the single-premise validities of this
fragment will be exactly those of the corresponding fragment of classical logic. As
explained above, this stands in contrast to the conjunctive/disjunctive fragment
of additive or multiplicative linear logic.44 The philosophical interpretation of
42 The point extends naturally to cases in which the assumptions are aggregated using both kinds of
structure. For instance, α1 : (α2 , α3 ) ⊢T β only if ⊢T Tr (pα1 q) & ( Tr (pα1 q) ⊗ Tr (pαn q)) → Tr (p βq).
43 Most work on this issue has concerned the closely parallel case of a naïve set theory featuring an
unrestricted axiom of comprehension. For proofs of the consistency or nontriviality of unrestricted
comprehension in some “weak relevant logics” that can be specified via dual-bunching natural
deduction, see Brady (1983, 1989, 2006). For applications of Brady’s techniques to naïve truth-theory,
see Priest (1991) and Beall (2009), which do not however consider natural deduction systems. As for
multiset-based logics, the consistency of unrestricted comprehension in an affine logic was shown
by V. Grishin in 1974: see Došen (1993). For the consistency of a naïve truth theory based on an
affine logic, see Zardini (2011).
44 As Dave Ripley pointed out to us, a dual-bunching logic could also retain a connective & that
A
behaves like the “additive” conjunction and disjunction of a multiset-based logic, for instance in
failing to validate Distribution over the corresponding ∨ A . To achieve this, replace &-E1 and &-E2
with
(& A -E1)
Γ, α ⊢ γ
∆ ⊢ α &A β
Γ, ∆ ⊢ γ
(& A -E2)
24
Γ, β ⊢ γ
∆ ⊢ α &A β
Γ, ∆ ⊢ γ
nonstandard antecedent structure—whether dual-bunching or multiset-based—
remains a controversial and important issue. However, it isn’t one we can address
in this paper, which has the more limited aim of exploring how such logics allow a
defense of VTP against the various challenges that have been raised against that
thesis.45
3.3 From VTP to absurdity via the Modus Ponens axiom
We now turn to the objection that VTP entails the Modus Ponens axiom, and thus
absurdity via c-Curry reasoning. Using a generic premise-aggregating connective,
we can state, respectively, the validity of Modus Ponens and the Modus Ponens axiom
as follows:
(VMP⊙ ) Val (p(α → β) ⊙ αq, pβq)
(MPA⊙ ) (α → β) ⊙ α → β.
In §1.2 we saw that VTP, when expressed in the object-language, implies
(V1) Val (pαq, pβq) → (α → β).
It follows that if our naïve semantic theory implies VMP⊙ , it also implies the
absurdity-threatening MPA⊙ . Thus, in order to evaluate the objection, we need to
answer two questions:
(1) If we reject SContr, will our semantic theory still imply VMP⊙ ? Equivalently,
in view of VP and VD, will our underlying contraction-free logic still give us
(α → β) ⊙ α ⊢ β?
(2) If we reject SContr, will MPA⊙ still yield absurdity?
A negative answer to (1) or (2) will show that the objection against VTP fails.46
By contrast, in the presence of Cutdb , our original &-E1 and &-E2 have the same “extensional” effect
as the rules
(&-E1db )
Γ(α) ⊢ γ
∆⊢α&β
Γ(∆) ⊢ γ
(&-E2db )
45 For
Γ( β) ⊢ γ
∆⊢α&β
Γ(∆) ⊢ γ
relevant work on the interpretation of dual-bunching systems, see Read (1988) and Slaney
(1990). For a recent and novel suggestion toward an interpretation of multiset-based systems, see
Zardini (2011). For a sketch of a more deflationary approach to antecedent structure, see Shapiro
(2011).
46 According to the theory proposed by Ripley (2011) based on Cobreros et al. (2011), which is
“substructural” only in rejecting Cut, the objection to VTP we are considering in this section fails
25
The answers to these questions vary depending on which connective we employ
as our ⊙. For the additive & of a contraction-free logic, the answer to (1) is negative
(Restall, 1994, pp. 35-6). It should help to display how SContr is involved in the
usual derivation:
(α → β) & α ⊢ (α → β) & α
(α → β) & α ⊢ (α → β) & α
&-E
&-E
(α → β) & α ⊢ α → β
(α → β) & α ⊢ α
→-E
(α → β) & α, (α → β) & α ⊢ β
SContr
(α → β) & α ⊢ β
But the objection to VTP fails as well when we use the the multiplicative ⊗. This
time, the answer to (1) is affirmative:
α→β⊢α→β
α⊢α
→-E
α → α, α ⊢ β
(α → β) ⊗ α ⊢ (α → β) ⊗ α
⊗-E
(α → β) ⊗ α ⊢ β
However, now the answer to (2) is negative. That is because, as already noted in
Meyer et al. (1979), the argument from MPA⊙ to absurdity depends essentially on
the left-to-right direction of the Idempotence law ⊢ α ↔ α ⊙ α. But when we use
multiplicative conjunction in a contraction-free logic, we lose this law (Zardini,
2011). Again, notice how SContr is involved in its usual derivation:
α→α
α → α ⊗-I
α, α ⊢ α ⊗ α
SContr
α ⊢ α⊗α
→-I
⊢ α → α⊗α
In summary, to derive absurdity from VTP, the objector presupposes that there is
some connective ⊙ that meets two conditions:
because MPA fails to yield absurdity. This is because the argument’s final step from ⊢T κ → ⊥ and
⊢T κ to ⊢T ⊥ fails. In Ripley’s sequent calculus, the rule →-E is inadmissible in the absence of Cut.
Indeed, Ripley holds (p.c) that →-E shouldn’t be regarded as fundamental to the logic of a detaching
conditional, as it covertly builds in extraneous transitivity in comparison with the sequent calculus
rule
(→-L)
Γ⊢α
∆, β ⊢ γ
∆, α → β, Γ ⊢ γ
To this, defenders of →-E may reply that each of →-E and →-L builds in transitivity in comparison
with the other rule, and in comparison with α → β, α ⊢ β. It is true, as Ripley shows, that the
transitivity built in by →-E (which, given →-I, yields Cut) can be blamed for paradox. But in view
of the option of blaming paradox on SContr instead, this won’t suffice to show that →-L is a more
fundamental rule than →-E.
26
(a) it serves as premise-aggregator for the valid argument α → β, α ⊢ β, so that
we have the single-premise rule (α → β) ⊙ α ⊢ β and VMP⊙ , and
(b) it satisfies the left-to-right direction of Idempotence, ⊢ α → α ⊙ α.
Yet we have now seen that one or the other of these conditions fails for each of our
candidate connectives.47
At this point, a critic of VTP might object that the response just given is at
best incomplete. We have shown that the argument from VTP to absurdity fails,
in the absence of SContr, when either & or ⊗ is used to state the premise VMP⊙ .
Still, the critic insists, our task remains that of explaining why the argument fails
when ⊙ expresses our ordinary notion of conjunction. After all, ordinary conjunction
appears to satisfy both conditions (a) and (b): both single-premise Modus Ponens
and Idempotence. If we are to avoid absurdity in the presence of a naïve theory
of truth, we have argued, at least one of these appearances must be mistaken. The
challenge is to explain which.
Zardini (2011, 2012) argues that condition (a) clearly holds for our “informal
notion of conjunction.” Accordingly, he maintains that ordinary conjunction is best
captured by the multiplicative connective ⊗ of an affine logic—where the presence
of SWeak guarantees such ordinary features as Simplification. Yet, as he recognizes,
someone else might argue that condition (b) clearly holds for ordinary conjunction.
More generally, we would add, one might maintain that the usual lattice properties
are essential to our ordinary conjunction ∧, whence from α ⊢ β and α ⊢ γ it must
follow that α ⊢ ( β ∧ γ), even in the case where α = β = γ.
We don’t propose to settle this dispute about our informal notion of conjunction,
or examine whether there is a univocal such notion.48 Instead, we will now explain
how the dispute is affected by the availability of dual-bunching logics. The chief
reason Zardini insists that ordinary conjunction meets condition (a) is that he takes
conjunction to be an all-purpose premise-aggregating connective. As he writes,
conjunction is the connective we use to make explicit “how premises are combined
in a multi-premise argument” (Zardini, 2012). In order for ⊙ to be conjunction, he
holds, it is non-negotiable that it satisfy the rule
(⊙-L)
Γ, α1 , α2 ⊢ β
Γ, α1 ⊙ α2 ⊢ β
47 It makes no difference whether these connectives are those of a multiset-based or dual-branching
logic. Nor, in the latter case, would it make a difference if we considered & A in place of &.
48 For arguments to the contrary, see Paoli (2007); Mares and Paoli (2012).
27
In a multiset-based logic without SContr, we have seen, the additive connective
& violates ⊙-L. We have a counterexample in the failure of α → β, α ⊢ β to yield
(α → β) & α ⊢ β. This is the chief reason why he concludes that ⊗ has a stronger
claim than & to represent our informal notion of conjunction.49
But once dual-bunching logics are an option, matters get more complicated. In
such logics we have both &-Ldb and ⊗-Ldb . The additive & corresponds to one mode
in which premises may be combined, marked by our colon, while the multiplicative
⊗ corresponds to another mode, marked by our comma (Read, 1988). According to
dual-bunching logics, & doesn’t serve as premise-aggregating connective for Modus
Ponens, since we don’t have α → β : α ⊢ β. Yet & serves as premise-aggregating
connective for other arguments, e.g. α : β ∨ γ ⊢ (α & β) ∨ γ. Hence it is no
longer clear that Zardini’s view, on which ordinary conjunction is multiplicative
and obeys single-premise Modus Ponens but not Idempotence, holds an advantage
over the alternative view on which ordinary conjunction is additive and satisfies
Idempotence but not single-premise Modus Ponens. Giving up single-premise Modus
Ponens, understood in terms of ordinary conjunction, needn’t amount to giving
up conjunction’s role as a premise-aggregating connective in a natural deduction
system. Of course, as we noted above, the philosophical significance of the two-fold
bunching of premises needs to be elucidated. But that is also the case for the simpler
premise structure in multiset-based deduction systems.
In this section, we have shown that the standard argument from VTP to absurdity breaks down in substructural theories which do not validate SContr, and have
explained how the details of where it breaks down depend on which connective of
the contraction-free logic we use to represent the conjunction appealed to in the
standard argument.
3.4 The Consistency Argument
Let us finally turn to the Consistency Argument, and the resulting challenge to
VTP from Gödel’s Second Incompleteness Theorem. There are two ways one might
respond: argue that Gödel’s limitative results don’t obtain for theories of arithmetic
based on contraction-free logics, or argue that the Consistency Argument fails for
49 Ole
T. Hjortland (2012) has recently proposed using an affine logic with additive conjunction
and disjunction in a revisionary approach to semantic paradox. We take no position here on whether
the consideration just rehearsed poses a serious problem for that approach.
28
such logics. Since there are contraction-free theories of arithmetic for which the
results hold, we won’t rely exclusively on the former strategy.50
The Consistency Argument requires one to prove, within one’s semantic theory
T, the following induction step: if all conclusions of derivations of length ≤ n are
true, then all conclusions of derivations of length n + 1 are true. To prove this, it
suffices to prove, for each rule R, that
(TPR ) If all the premises of an instance of R are true, then the corresponding instance of the conclusion will be true.51
Now consider a rule R such that the theory proves that R has precisely two premises.
To establish TPR we will then need to prove
(TP2R ) For all x, y, z such that x and y are the two premises of an instance
of R and z its corresponding conclusion: if x is true and y is true,
then z is true.
But how are we to understand the ‘and’ in TP2R ?
If ‘all’ in TPR is understood as the standard “lattice-theoretical” or additive
quantifier (Paoli, 2005), then TP2R will only help establish TPR provided ‘and’ is
likewise construed as additive.52 But when R is the two-premise Modus Ponens, we
won’t be able to prove TPR on this construal. That is because we have already seen
that we don’t have any instance of ⊢T (α → β) & α → β. This should mean that we
don’t have any instance of ⊢T Tr (pα → βq) & Tr (pαq) → Tr (pβq) either, whence
we can’t prove the generalization TPR . In fact, that is Field’s own explanation of
how the Consistency Argument breaks down for paracomplete and paraconsistent
theories (Field, 2008, pp. 377-8). Unlike Field, we don’t attribute this breakdown
to the argument’s illicit appeal to VTP. In our view, rather, the breakdown of the
50 Restall
(1994, ch. 11) shows that that an arithmetic based on the dual-bunching contractionfree logic RWK (which he calls CK) is classical Peano arithmetic, but it isn’t known whether RWK
supports a nontrivial naïve semantic theory in which Tr (pαq) is everywhere intersubstitutable with
α (see Hjortland, 2012).
51 Here we are no longer thinking of natural deduction rules, but rather of the rules of a Hilbert
system, rules for generating theorems.
52 Here is a rough explanation. In the course of deriving TP in our object-language, we will need
R
to establish, under the assumption that three arbitrary sentences (denoted by a1 , a2 and b) are the
respective premises and conclusion of an instance of R, the claim ∀ x ( x = a1 ∨ x = a2 → Tr ( x )) ⊢
Tr (b). Assuming ∀ is lattice-theoretical, this claim will follow from Tr ( a1 ) & Tr ( a2 ) ⊢ Tr (b), whereas
it won’t follow from Tr ( a1 ) ⊗ Tr ( a2 ) ⊢ Tr (b). For we have ∀ xφ( x ) ⊢ φ( a1 ) & φ( a2 )... & ...φ( an ), but
not ∀ xφ( x ) ⊢ φ( a1 ) ⊗ φ( a2 )... ⊗ ...φ( an ). See Běhounek et al. (2007).
29
Consistency Argument (on the standard interpretation of the quantifier) results
from the argument’s illicit use of & as premise-aggregator for the two-premise
Modus Ponens rule.53
Perhaps, then, we could rescue the Consistency Argument by interpreting the
‘all’ in TPR as some kind of multiplicative quantifier, one that stands to ⊗ the way
the standard universal quantifier stands to &. Where R is Modus Ponens, we should
indeed be able to prove TP2R with ‘and’ interpreted as ⊗, since ⊗ does serve as
premise-aggregator for Modus Ponens. If this is to help establish TPR , however, we
would need to know more about the envisioned multiplicative quantifier. Paoli
(2005) and Mares and Paoli (2012) note that there is no accepted theory of how such
a quantifier should behave. One option is presented by Zardini (2011) in the context
of a multiset-based logic. But Zardini’s multiplicative quantifier won’t serve the
purposes of anyone who wishes to use the Consistency Argument to criticize VTP.
For he characterizes the behavior of the multiplicative quantifier using an ω-rule as
(right-)introduction rule. Hence, the semantic theory based on this logic won’t be
recursively axiomatisable, and won’t satisfy the conditions for Gödel’s theorem.
4 Concluding remarks
In this paper, we’ve argued for two main claims. First, the v-Curry Paradox
shows that SContr is in tension with natural principles governing some (intuitive
enough) notions of validity. Hence, if, as we’ve assumed, the validity relation is
transitive, revisionary theorists have strong reason to give up SContr. Second, the
standard challenges to VTP presented in §1 all break down once SContr is dropped.
Rejecting SContr opens up non-classical ways of aggregating together premises—
ways which no longer underwrite the objections to VTP. To be sure, it may be
argued instead that the notion of validity that is shown to be paradoxical by the
v-Curry Paradox should be rejected as incoherent. Validity, one might think, is
interpretational, or purely logical, validity: truth on all uniform interpretations of
the non-logical vocabulary. This, however, does not seem in line with the seemingly
compelling thought, championed by rcf theorists such as Field (2007, 2008) and
53 Field
himself claims that TP2R will “obviously” fail to establish TPR when the former is
understood using what is, in effect, multiplicative conjunction. See Field (2006, p. 597) and
Field (2008, p. 379). In his discussion, Tr (pα → βq) → ( Tr (pαq) → Tr (p βq)) takes the place
of Tr (pα → βq) ⊗ Tr (pαq) → Tr (p βq), which is equivalent to the former in the logics we are
considering. See also Priest (2010).
30
Priest (2006b,a), that logical validity is a species of a more general notion of validity.
Alternatively, it may be contended that paradox-prone notions of validity must be
refined, and made less naïve (McGee, 1991). But this, too, we’ve argued, doesn’t
seem like a viable option for proponents of the revisionary approach to paradox,
who rather recommend revising our theory of logic, while preserving the naïve
semantic principles. If neither of these foregoing options is viable, then SContr
must be restricted on pain of triviality, and we can continue to maintain that valid
arguments preserve truth.
References
Anderson, A. and Belnap, N.: 1975, Entailment: the Logic of Relevance and Necessity,
Vol. 1, Princeton University Press.
Asenjo, F. G.: 1966, A calculus of antinomies, Notre Dame Journal of Formal Logic
16, 103–105.
Asenjo, F. and Tamburino, J.: 1975, Logic of antinomies, Notre Dame Journal of Formal
Logic 16, 17–44.
Beall, J.: 2007, Truth and paradox: a philosophical sketch, in D. Jacquette (ed.),
Philosophy of Logic, Elsevier, Oxford, pp. 325–410.
Beall, J.: 2009, Spandrels of Truth, Oxford University Press, Oxford.
Beall, J.: 2011, Multiple-conclusion lp and default classicality, Review of Symbolic
Logic 4(1), 326–336.
Beall, J. and Murzi, J.: 2011, Two flavors of curry’s paradox. Forthcoming in The
Journal of Philosophy.
Belnap, N.: 1982, Display logic, Journal of Philosophical Logic 11, 375–417.
Belnap, N.: 1993, Life in the undistributed middle, in P. Schroeder-Heister and
K. Došen (eds), Substructural Logics, Oxford University Press, Oxford, pp. 31–41.
Brady, R.: 1983, The simple consistency of set theory based on the logic CSQ, Notre
Dame Journal of Formal Logic 24, 431–39.
31
Brady, R.: 1989, The Non-Triviality of Dialectical Set Theory, in B. P. R. Routley
and J. Norman (eds), Paraconsistent Logic: Essays on the Inconsistent, Philosophia,
Munich, pp. 437–71.
Brady, R.: 2006, Universal Logic, CSLI Publications, Stanford.
Běhounek, L., Cintula, P. and Horčík: 2007, Multiplicative quantifiers in fuzzy and
substructural logics. Presented at Logic Colloquium 2007, Wrocław.
Clark, M.: 2007, Paradoxes from A to Z, second edn, Routledge, London and New
York.
Cobreros, P., Egré, P., Ripley, D. and van Rooij, R.: 2011, Tolerant, classical, strict.
Forthcoming in Journal of Philosophical Logic.
Cook, R.: 2012, The t-schema is not a logical truth. Forthcoming in Analysis.
Došen, K.: 1993, A historical introduction to substructural logics, in P. SchroederHeister and K. Došen (eds), Substructural Logics, Oxford University Press, Oxford,
pp. 1–30.
Etchemendy, J.: 1990, The Concept of Logical Consequence, Harvard University Press,
Harvard (Mass.).
Field, H.: 2003, A revenge-immune solution to the semantic paradoxes, Journal of
Philosophical Logic 32, 139–77.
Field, H.: 2006, Truth and the unprovability of consistency, Mind 115(459), 567–606.
Field, H.: 2007, Solving the paradoxes, escaping revenge, in J. Beall (ed.), Revenge of
the Liar, Oxford University Press, pp. 53–144.
Field, H.: 2008, Saving Truth from Paradox, Oxford University Press, Oxford.
Field, H.: 2009a, Pluralism in logic, Review of Symbolic Logic 2(2), 342–359.
Field, H.: 2009b, What is the normative role of logic?, Proceedings of the Aristotelian
Society 83, 251–68.
Field, H.: 2010, Validity: model-theoretic, proof-theoretic and genuine. To appear.
Girard, J. Y.: 1987, Linear logic, Theoretical Computer Science 50, 1–102.
32
Harman, G.: 1986, Change in View. Principles of Reasoning, MIT Press, Cambridge
(Mass.).
Harman, G.: 2009, Field on the normative role of logic, Proceedings of the Aristotelian
Society 109(1), 333–335.
Hjortland, O. T.: 2012, Truth, paracompleteness and substructural logic. Unpublished manuscript.
Horsten, L.: 2009, Levity, Mind 118(471), 555–581.
Kaplan, D. and Montague, R.: 1960, A paradox regained, Notre Dame Journal of
Formal Logic 1, 79–90.
Ketland, J.: 2012, Validity as a primitive, Analysis 72(3), 421–430. Forthcoming in
Analysis.
Kripke, S.: 1975, Outline of a theory of truth, Journal of Philosophy 72, 690–716.
Mares, E. and Paoli, F.: 2012, Logical consequence and the paradoxes. To appear.
Martin, R. L. (ed.): 1984, Recent Essays on Truth and the Liar Paradox, Oxford University Press, New York.
Martin, R. L. and Woodruff, P. W.: 1975, On representing ‘true-in-L’ in L, Philosophia
5, 217–221. Reprinted in Martin (1984).
McGee, V.: 1991, Truth, Vagueness, and Paradox, Hackett Publishing Company,
Indianapolis.
Meyer, R. K. and McRobbie, M. A.: 1982a, Multisets and relevant implication i,
Australasian Journal of Philosophy 60, 107–139.
Meyer, R. K. and McRobbie, M. A.: 1982b, Multisets and relevant implication ii,
Australasian Journal of Philosophy 60, 265–181.
Meyer, R. K., Routley, R. and Dunn, J. M.: 1979, Curry’s paradox, Analysis 39, 124–
128.
Minc, G.: 1976, Cut-elimination theorem in relevant logics, Journal of Soviet Mathematics 6, 422–428.
33
Murzi, J.: 2011, Validity curry. Under review.
Myhill, J.: 1960, Some remarks on the notion of proof, Journal of Philosophy
57(14), 461–71.
O’Hearn, P. and Pym, D.: 1999, The logic of bunched implications, Bulletin of
Symbolic Logic 5, 215–44.
Paoli, F.: 2002, Substructural logics: a primer, Kluwer Academic Publishers, Dordrecht.
Paoli, F.: 2005, The ambiguity of quantifiers, Philosophical Studies 124, 313–30.
Paoli, F.: 2007, Implicational paradoxes and the meaning of logical constants,
Australasian Journal of Philosophy 85, 533–79.
Priest, G.: 1979, The logic of paradox, Journal of Philosophical Logic 8, 219–241.
Priest, G.: 1980, Sense, entailment, and modus ponens, Journal of Philosophical Logic
9, 415–35.
Priest, G.: 1991, Intensional paradoxes, Notre Dame Journal of Formal Logic 32, 193–
211.
Priest, G.: 2006a, Doubt Truth to be a Liar, Oxford University Press, Oxford.
Priest, G.: 2006b, In Contradiction, Oxford University Press, Oxford. Expanded
edition (first published 1987 Kluwer-Dordrecht).
Priest, G.: 2010, Hopes fade for saving truth, Philosophy 85, 109–40. Critical notice
of Field (2008).
Priest, G. and Routley, R.: 1982, Lessons from Pseudo-Scotus, Philosophical Studies
42, 189–99.
Read, S.: 1988, Relevant Logic: A Philosophical Examination of Inference, Basil Blackwell,
London.
Read, S.: 2001, Self-reference and validity revisited, in M. Yrjonsuuri (ed.), Medieval
Formal Logic, Kluwer Academic Publishers, pp. 183–96.
34
Read, S.: 2010, Field’s paradox and its medieval solution, History and Philosophy of
Logic 31(2), 161–176.
Restall, G.: 1993, How to be really contraction free, Studia Logica 52, 381–91.
Restall, G.: 1994, On Logics Without Contraction, PhD thesis, The University of
Queensland.
Restall, G.: 2000, An Introduction to Substructural Logics, New York: Routledge.
Ripley, D.: 2011, Paradoxes and failures of cut. Forthcoming in Australasian Journal
of Philosophy.
Shapiro, L.: 2011, Deflating logical consequence, The Philosophical Quarterly 61, 320–
42.
Slaney, J.: 1990, A general logic, The Australasian Journal of Philosophy 68(1), 74–88.
Troelstra, A.: 1992, Lectures on Linear Logic, CSLI, Stanford.
Weir, A.: 2005, naïve truth and sophisticated logic, in J. Beall and B. Armour-Garb
(eds), Deflationism and Paradox, Oxford University Press, Oxford, pp. 218–249.
Whittle, B.: 2004, Dialetheism, logical consequence and hierarchy, Analysis
64(4), 318–26.
Zardini, E.: 2011, Truth without contra(di)ction, Review of Symbolic Logic 4, 498–535.
Zardini, E.: 2012, Naive Modus Ponens. Forthcoming in the Journal of Philosophical
Logic.
35