Penultimate draft of the paper forthcoming in the Journal of Philosophical Logic
Some changes may be reflected in the published version
A Hierarchy of Classical and Paraconsistent Logics
Eduardo Alejandro Barrio Federico Pailos Damian Szmuc
University of Buenos Aires and IIF-SADAF (CONICET), Argentina
Abstract
In this article, we will present a number of technical results concerning Classical
Logic, ST and related systems. Our main contribution consists in offering a novel
identity criterion for logics in general and, therefore, for Classical Logic. In particular, we will firstly generalize the ST phenomenon, thereby obtaining a recursively
defined hierarchy of strict-tolerant systems. Secondly, we will prove that the logics
in this hierarchy are progressively more classical, although not entirely classical.
We will claim that a logic is to be identified with an infinite sequence of consequence relations holding between increasingly complex relata: formulae, inferences,
metainferences, and so on. As a result, the present proposal allows not only to
differentiate Classical Logic from ST, but also from other systems sharing with it
their valid metainferences. Finally, we show how these results have interesting consequences for some topics in the philosophical logic literature, among them for the
debate around Logical Pluralism. The reason being that the discussion concerning
this topic is usually carried out employing a rivalry criterion for logics that will
need to be modified in light of the present investigation, according to which two
logics can be non-identical even if they share the same valid inferences.
Keywords: Substructural Logics, Cut Rule, Metainference, Classical Logic
1
Background and aim
What identifies Classical Logic as such? As easy and straightforward as answering
this question may appear, in this article we devote ourselves to discussing a number
of overlooked subtleties that, we argue, need to be taken into account in providing a
cogent answer to it. The main contribution of our work consists, therefore, in offering a
novel identity criterion for logical systems in general and, therefore, for Classical Logic.
1
The aforementioned subtleties have their origin in recent debates in the philosophical logic literature, more particularly, in a number of observations revolving around
a project carried out by the collective formed by Pablo Cobreros, Paul Egré, David
Ripley and Robert van Rooij. With the goal of solving paradoxes coming from the semantic, set-theoretic and vagueness corners, these authors have put forward a number of
conceptual and technical arguments that support them backing a nice, innovative and
quite revolutionary stance towards these riddles: the strict-tolerant approach, which
led them to entertain the non-transitive logic ST.1 The main advantage of this choice
lies, allegedly, in keeping Classical Logic as the underlying inferential framework, even
in what pertains to the problematic phenomena referred above. Thus, these authors
claim to have shown that Classical Logic—actually, ST, but both systems are identical,
according to them—can non-trivially deal with these otherwise problematic phenomena.
In this regard, it may be instructive to discuss a number of natural reasons that we
think point towards justifying that ST is not, after all, Classical Logic. First, embracing
an essentially non-transitive logic requires understanding logical consequence between
formulae in terms of something other than truth-preservation—as acknowledged in many
works revolving around ST and other substructural logics, such as [6]. But, certainly,
the notion of logical consequence featured in Classical Logic can be presented in terms
of truth-preservation. Thus, even if there is nothing essentially wrong or mistaken in
taking logical consequence to be defined as something different than the preservation of
truth from premises to conclusion, this immediately tells ST and Classical Logic apart.
Secondly, identifying ST and CL would require accepting that transitivity is inessential to the characterization of classical reasoning. But, as argued in [4], it appears to
be a rather central feature, allowing us to conduct cumulative reasoning, transitioning from lemmas to theorems and later to corollaries—regardless of the subject-matter
under discussion. In other words, if cumulative deductive progress is not guaranteed,
we cannot be sure that lemmas are knowledge checkpoints. But this seems quite an
important property of classical reasoning, while it certainly is not a feature of reasoning
carried out in ST.
Moreover, there are some other widely shared intuitions about what Classical Logic
ought to be, and about which logical principles are distinctively valid in it. Explosion,
Excluded Middle, Modus Ponens, Disjunctive Syllogism, etc. are only a few of these,
that every proper presentation of Classical Logic should comply with. Additionally, if
these are part of the core of such a logical system, then they should be expected to hold
in the context of all kinds of inferences carried out in it. That is to say, they should
hold within regular inferences between formulae, but also in the context of what we call
metainferences—namely, inferences between inferences. However, as remarked in [4]
and [28], it is illustrative to notice that ST fails to validate what might be understood
as certain metainferential forms of Explosion, Modus Ponens, Disjunctive Syllogism,
and so on. This, if anything, seems to add more grounds to the extent that ST and
Classical Logic are not identical.
Furthermore, there is the important fact that Classical Logic is widely believed to be
1
A non-exhaustive list of some works of this collective where ST is discussed is [8], [7], and [10].
2
prone to trivialization when faced with transparent truth, vague phenomena and much
more, while ST does not fall into such troubles. Interestingly, to an argument of this
sort advocates of their identity may respond that ST is nothing but a different mode of
presenting Classical Logic.2 Just like Classical Logic can be presented by means of twovalued models or valuations, it can also be presented by means of three-valued Strong
Kleene models or valuations, as we will see below. In this regard, they could argue
that which consequences can be drawn from an arbitrary piece of content or a theory
by closing it under Classical Logic—and whether cumulative reasoning is unrestrictedly
allowed—crucially depends on which presentation of Classical Logic we implement.
Faced with a reasoning of this sort, we cannot but express our substantial disagreement. Take, for example, the case of closing arbitrary theories under numerically
different logical systems, and consider the case of theories of transparent truth. Looking
at their consequences, and whether or not the resulting systems lead to triviality, provides a way—albeit a rather indirect one—to assess their strength, to compare them.
As a limit case, if two logics render exactly the same consequences, we think it is fair to
assume that, by all intents and purposes, they are the same logic.3 This is not the case,
according to the widely accepted beliefs about Classical Logic, of Classical Logic and
ST. Thus, this is why we think it is necessary to have a more direct way of identifying
and differentiating them, by providing the identity criterion that we outlined previously.
Let us clarify, though, that our aim in this article is not to develop a knock-down
philosophical argument to establish that Classical Logic and ST are different logics.
For what is worth, it is not obvious for us that there is even a way of establishing such
a thing without running into some sort question-begging reasoning. Our work here is
not intended to convince believers in the identity of Classical Logic and ST that they
are wrong. Instead, we aim at offering technical developments and precise arguments
to sustain in a formal way the rather loose and informal feeling that some people had
so far, to the extent that Classical Logic and ST are not identical.
In order to do this, these systems should be told apart by some formal and philosophically motivated criterion. In this article we advance a way of doing so, by proposing a
novel identity criterion for logical systems, which will take into account not only those
inferences between formulae which are valid in a given logical system—but also several
other inferences, e.g. between inferences themselves, which are valid in a given logical
system. As we said previously, contrary to what it seems, this is by no means a trivial
task. In fact, a number of details that have not received much attention in the literature,
until now, make this an especially intricate business.
The first difficulty appearing in our route to finding an appropriate identity criterion
for logical systems, is that the one we often use—implicitly employed by the advocates
of ST to argue it is nothing but Classical Logic—is the nowadays standard Tarskian
conception of a logic as a consequence relation between collections of formulae. Such
an account, although subject to some generalizations in the last decades, is the ruling
2
This way to understand ST was presented to us, in conversation, by Dave Ripley.
An option along these lines has been both suggested and explored in what pertains to Tarskian
logics—particularly, Classical and Intuitionistic Logic—in [27].
3
3
standard for us to distinguish logical systems. The problem is, thus, evident. We need
a different account to individuate logics and, in particular, to identify Classical Logic.
A versed reader in the recent literature revolving around the strict-tolerant approach
might point to a tempting solution, reflecting upon the set of metainferences which are
valid in these logics. As mentioned earlier and as pointed out e.g. by Barrio, Rosenblatt
and Tajer in [4], a number of classically valid metainferences—such as meta Modus
Ponens, meta Explosion and others, on which more below—fail to be valid in ST.
Thus, one might be tempted to say that it differs from Classical Logic because, even if
both systems coincide at the inferential level, they do not agree at the metainferential
level.4 However promising this solution may appear, and however easily it seems to
address the problem at hand, it cannot provide a definitive answer.
The reason for this is tied to the second difficulty we stumble upon when trying to
find an appropriate identity criterion for logics, namely that the previously discussed
phenomenon can be replicated. By this we mean that—just like ST is a system which
coincides with Classical Logic at the inferential level, but not at the metainferential
level—it is possible to obtain a system which coincides with Classical Logic at the
inferential and the metainferential level, but which does not coincide with it at the level
of the metametainferences, i.e. of the inferences between metainferences.5 Furthermore,
if the search of an extensional identity criterion is expanded to cover this case too, it
is still possible to design another ST-like logic which also complies with it, albeit nonclassical with regard to its metametametainferences, and so on and so forth.
In other words, we claim the phenomenon incarnated by the ST approach is pervasive. In fact, it is possible to obtain a sequence of logics which can be progressively
ordered in terms of their degree of classicality. To argue for this we will present a collection, a hierarchy of systems which coincide with Classical Logic in more and more
inferential levels as we move in the sequence—although for all of them there is some
inferential level at which they start to behave non-classically. This constitutes the pars
destruens of the article. The pars construens will be devoted to presenting our own
identity criterion for logical systems, and considering possible objections to it.
To this extent, the article is structured as follows. In Section 2 we introduce some
preliminary definitions that will be of use in the remainder of the article, including
the definition of metainference, metametainference, and metainference of an arbitrary
large level. Section 3 reviews the strict-tolerant phenomenon incarnated in ST, as it
is featured in the discussions appearing in the literature so far. Section 4 includes our
main technical and philosophical contributions: the hierarchy of ST-related logics, and
our novel identity criterion for logical systems. On the one hand, this section contains
4
An anonymous reviewer wonders whether we know which metainferences are valid in Classical
Logic beforehand this investigation, i.e. whether we have some means of ruling on these matters that
we are not stating upfront. In fact, throughout this article we will adopt a semantic stance towards
metainferential validity, as seen in the various definitions appearing in Section 4. This does not mean
that these issues cannot be established and explored by proof-theoretical means, but for matters of
space these will be left for another occasion.
5
Another way to fully recover Classical Logic, starting from ST, is through the introduction of a
so-called consistency or recovery operator. This path was explored by us in detail in [1].
4
a discussion that deepens the understanding of the ST phenomenon by generalizing
it to infinitely many different logical systems, and an impossibility result concerning
the extensional characterization of Classical Logic, i.e. the identification of Classical
Logic with its set of valid inferences of some inferential level. On the other, it contains
an analysis of the conceptual importance of the results associated to this hierarchy,
by providing and defending a novel identity criterion for logical systems that is able
to overcome the difficulties of the available ones and their modifications. Section 5
considers some objections to our account and our replies to them. Finally, Section 6
wraps up our work by making some concluding remarks.
2
Preliminary definitions
To understand and carry our investigation, it will be essential to have an accurate grasp
of the received view about inferences, metainferences and consequence relations. In turn,
to analyze these matters, it will be useful to fix some terminology. In what follows, we
will be working with a propositional language L equipped with the connectives ¬, ∧, ∨,
to be interpreted as negation, conjunction and disjunction, letting F OR(L) be the set
of recursively generated well-formed formulae of L. As usual, we will let Γ, ∆, and other
Greek capital letters represent sets of formulae, or sets of inferences, and Roman capital
letters A, B, C represent formulae themselves, or inferences themselves.6
An inference Γ ⇒ ∆ on L is an ordered pair hΓ, ∆i where Γ, ∆ ⊆ F OR(L), letting
SEQ0 (L) be the set of all inferences on L, to which we might occasionally refer to as
SEQ(L).
According to a nowadays standard approach, a logic is usually identified with a set of
inferences. This is very much incarnated in the definition of logic as a relation over collections of formulae, included in SEQ0 (L), and of a Tarskian logic as a logic complying
with the following structural rules—where Γ, ∆ ⊆ F OR(L) and A ∈ F OR(L).7
A⇒A
Reflexivity
Γ⇒∆
Weakening
Γ′ , Γ ⇒ ∆, ∆′
Γ ⇒ A, ∆ Γ, A ⇒ ∆
Cut
Γ⇒∆
However, the main point we discuss in this article consists—precisely—in defining
what a logic is. For this reason, we will not embrace the aforementioned identification
here. In other words, we will not assume that a logic is a relation over collection of formulae and, therefore, we will not assume that a logic is a relation of this sort complying
with this or that structural rules. Structural rules, in fact, are schematic metainferences.
But, then, given many of the critical points we will rise in this investigation—concerning
plausible identity criteria for logical systems—rely on debates involving metainferences
of some kind or another, we cannot assume structural rules are built in the definition
of a logical system. That would mean ending the investigation before even starting it.
6
Though we hope the context will make things clear enough, we will always clearly state whether,
for example, Roman capital letters A, B, C represent formulae themselves, or inferences themselves.
7
Often, the requirement is also asked that consequence relations are substitution-invariant, i.e. that
if Γ ⇒ ∆ is valid, and σ is a substitution on F OR(L), then {σ(A) | A ∈ Γ} ⇒ {σ(B) | B ∈ ∆} is valid.
5
These considerations, however, call for a clarification concerning metainferences.
Indeed, we have not yet even defined what a metainference actually is. To solve this
we will say that, intuitively, a metainference is an inference between a collection of
inferences, and an inference. To illustrate this observe the following, where the one on
the left is an inference, whereas the one on the right is a metainference.
∅⇒p
p, ¬p ⇒ q
∅⇒p⊃q
∅⇒q
In fact, just like the one on the left above is an instance of the schematic inference
usually called Explosion, the one on the right is an instance of the schematic metainference known as meta Modus Ponens—respectively depicted below.
A, ¬A ⇒ B
Γ ⇒ A, ∆ Γ ⇒ A ⊃ B, ∆
Γ ⇒ B, ∆
Of very much interest for the upcoming discussions is the following generalization
of the notion of metainference, to a metainference of an arbitrary large (finite) level. In
this vein, we will say a metainference Γ ⇒n A of level n on L (for 1 ≤ n < ω) is an
ordered pair hΓ, Ai where Γ ⊆ SEQn−1 (L) and A ∈ SEQn−1 (L). SEQn (L) is the set
of all metainferences of level n on L.
To get a feeling of these observe the following, from which the one on the left is a
metainference, the one on the middle is a metametainference (or metainference of level
2), while the one on the right is a metametametainference (or metainference of level
3). The reader can produce further examples of metainferences of arbitrary large levels,
just by using her imagination.
p⇒r q⇒r
p∨q ⇒r
r⇒s q⇒p r⇒s q⇒p
p⇒q t⇒u p⇒q t⇒u
s⇒t
s⇒t
r⇒u
r⇒u
r⇒s q⇒p
p⇒q t⇒u
s⇒t
r⇒u
r⇒s q⇒p
p⇒q t⇒u
s⇒t
r⇒u
Observe that the above definitions allow to recast e.g. all inferences as metainferences, by considering them as metainferences with empty premise sets. Similarly
e.g. for metainferences, which can be recast as metametainferences with appropriately
empty premise sets. This can be also done for formulae in an analogous manner, i.e. by
considering them as inferences with no premises.8 This is the technical reason why two
systems characterized by the same set of inferences will be forced to have the same set of
theorems, but might not necessarily be characterized by the same set of metainferences.
This also explains why two logics with the same metainferences of level n will coincide
regarding the metainferences and inferences of any level m lower than n, but might not
necessarily coincide regarding the metainferences of level k greater than n.
Metainferences are taken to be—as Dicher and Paoli put it in [12]—bona fide objects,
i.e. rightful linguistic items themselves, constituted by concrete inferences which are in
8
This approximation follows the spirit of some remarks made in [12].
6
turn constituted by pairs of collections of formulae. This marks a difference not only
with the stance adopted by Barrio, Rosenblatt and Tajer in [4], but also reflects a view
opposed to Cobreros, Egré, Ripley and van Rooij’s conception of these entities. The
former consider metainferences to be schemata and not objects themselves, whereas
the latter take them to be supervening properties under which the valid inferences of a
certain logical system might or might not happen to be closed.9
We can, and sometimes will, refer to particular metainferences and to schematic
metainferences. Two of the main schematic metainferences that will draw our attention
in what follows are corresponding versions of Cut and Explosion of each metainferential
level. These we will call, without loss of generality, meta Cut and meta Explosion and—
more particularly—we will denote them by meta...meta
{z
} Cut and |meta...meta
{z
} Explosion,
|
n
n
depending on the inferential level at which they appear. In fact, each instance of the
former is a metainference of level n−1, and each instance of the latter is a metainference
of level n. More concretely, the following metainferences
s⇒t
s⇒t
p ⇒ q, r p, r ⇒ q
s⇒t
p⇒q
s⇒t
s⇒t
p ⇒ q, r p ⇒ q, ¬r
s⇒t
p ⇒ q, u
are instances of the schematic metainferences appearing below, called meta Cut and
metameta Explosion, respectively. The reader may use her imagination to design appropriate versions of these for greater metainferential levels.
Γ1 ⇒ ∆ 1 . . . Γk ⇒ ∆ k
Γ1 ⇒ ∆ 1 . . . Γk ⇒ ∆ k
Σ ⇒ Π, A
Σ, A ⇒ Π
Γ1 ⇒ ∆ 1 . . . Γ k ⇒ ∆ k
Σ⇒Π
Γ1 ⇒ ∆ 1 . . . Γk ⇒ ∆ k
Γ1 ⇒ ∆ 1 . . . Γk ⇒ ∆ k
Σ ⇒ Π, A
Σ ⇒ Π, ¬A
Γ1 ⇒ ∆ 1 . . . Γ k ⇒ ∆ k
Σ ⇒ Π, B
With these technicalities at hand, let us delve into the ST phenomenon and the
complications it poses in offering an appropriate identity criterion for Classical Logic.10
9
See e.g. [21, p. 354] and [10, §3.3].
We should remark we will not consider metainferences with multiple conclusions, as with regular
inferences. This is not because we think there is something wrong with them, but because it will
unnecessarily complicate the presentation of the systems that we will introduce below. In this vein, we
will also not consider mixed metainferences, i.e. metainferences with premises belonging to different
inferential levels. Although there is nothing conceptually wrong about such metainferences, yet again,
working with them will make the different proofs unnecessary complicated. An additional clarification
concerns the kind of schematic metainferences that we will be taking into account to exemplify our
investigations. These will be presented in an additive instead of a multiplicative way, i.e. they will—
actually, their instances will—have shared contexts. For more on these distinctions, see e.g. [23].
10
7
3
The ST phenomenon, as we know it
To better understand the logic ST we will review some of its properties, pointing out its
relations with Classical Logic (CL hereafter) and with a certain non-classical logic, i.e.
Graham Priest’s system LP. After this, we will show how this peculiar system can be
seen as only one instance of what we call “the ST phenomenon”, that is, the definability
of systems agreeing with CL in successively many inferential levels, although not in all
of them. It is with these tools at hand that we will argue that the ST phenomenon
poses a threat to the current identity criterion for logical systems (which identifies them
with their set of valid inferences) and to any natural extension thereof.
The logic ST is sometimes discussed by its advocates as essentially motivated by
inferentialist concerns, whence it is usually presented as a sequent calculus. However,
this system can be equally well-motivated from a semantic point of view. As such,
logical consequence is defined in ST with one of the many—sixteen, as remarked by
Wintein [26]—different definitions of logical consequence which happen to coincide in
CL. In the case of this strict-tolerant system, logical consequence is defined by claiming
that Γ ⇒ ∆ is valid if and only if there is no assignment of truth-values making the
premises true while at the same time making the conclusion false.
One difference between CL and ST lies in the scope of the candidate valuations
involved in the definition of logical consequence. While for CL valuations are only
Boolean or two-valued, in the case of ST these are essentially three-valued valuations
respecting the Strong Kleene truth-tables appearing below—where ⊃ and ↔ are definable as usual, i.e A ⊃ B =def ¬A ∨ B and A ↔ B =def (A ⊃ B) ∧ (B ⊃ A).
1
2
1
2
1
2
1
2
1
2
1
2
1
2
0
1
0
0
0
1
¬
0
∧
1
1
1
0
0
0
0
∨
1
1
2
0
1
1
1
1
1
2
1
0
1
1
2
1
2
0
1
2
Definition 3.1. A Strong Kleene valuation (SK-valuation, hereafter) is a mapping from
F OR(L) to {1, 21 , 0} that respects the Strong-Kleene truth-tables above. Similarly, a
Boolean valuation is a SK-valuation whose range is {1, 0}.
Below, we define what is for a valuation to satisfy a given inference in CL and ST.
As is common practice, we will define the validity of an inference (and, more generally
below, of an inference of an arbitrary inferential level) in a certain logic as nothing more
than satisfaction by all valuations—in our case, all SK-valuations.
Definition 3.2. A Boolean valuation v satisfies an inference Γ ⇒ ∆ in CL (v CL
Γ ⇒ ∆) if and only if it is not the case that v(A) = 1 for all A ∈ Γ and v(B) = 0 for all
B ∈ ∆. Similarly, a SK-valuation v satisfies an inference Γ ⇒ ∆ in ST (v ST Γ ⇒ ∆)
if and only if it is not the case that v(A) = 1 for all A ∈ Γ and v(B) = 0 for all B ∈ ∆.
These definitions allow to prove the much commented result stating that the set of
valid inferences of CL and ST coincide—as shown e.g. in [13] and [8].11
11
Yet another case of a substructural logic which has, nevertheless, the same valid inferences than
CL is the case of the non-contractive system discussed by Lucas Rosenblatt in [24].
8
Fact 3.3. For all Γ, ∆ ⊆ F OR(L):
ST Γ ⇒ ∆
if and only if
CL Γ ⇒ ∆
Now, it is often emphasized that the strict-tolerant division giving rise to the ST
label is motivated by appealing to assertion and denial in their strict and tolerant moods
(cf. [22]), but we will claim that this label can be equally well-motivated from a semantic
point of view. This can be done by looking at the following quote:
The logic ST sets different standards for satisfaction in premises and in conclusions.
A “good” premise (a premise good enough to produce a sound argument) is one
that takes value 1. A “good” conclusion, on the other hand (a conclusion that is
not false enough to produce a counterexample) is one that takes value greater than
0. [9, p. 79]
This amounts to saying the semantic essence of the ST phenomenon consists in
its adoption of the standards of CL for taking premises to be satisfied by a valuation
and for taking conclusions not to be dissatisfied by a valuation—albeit in the context
of three-valued valuations. Thus, the semantics for ST are the semantics of a stricttolerant logic, because although the requirement for the premises is strict (i.e. that
they do satisfy the standard to count as a premise of a sound argument in CL), the
requirement for the conclusions is tolerant (i.e. that they do not satisfy the standard
to count as the conclusion of an invalid argument in CL).
These observations portray ST as a system incarnating an essentially mixed notion
of logical consequence, by means of which standards of different strength are asked for
the premises and the conclusions of valid inferences. However, three-valued systems
adopting homogeneous requirements for premises and conclusions seem equally motivated. In fact, taking the strict and the tolerant routes tout court leads, respectively,
to the well-known three-valued logics LP and K3 .
Definition 3.4. A SK-valuation v satisfies an inference Γ ⇒ ∆ in LP (v LP Γ ⇒ ∆)
if and only if it is not the case that v(A) ∈ {1, 21 } for all A ∈ Γ and v(B) = 0 for all
B ∈ ∆. Similarly, a SK-valuation v satisfies an inference Γ ⇒ ∆ in K3 (v K3 Γ ⇒ ∆)
if and only if it is not the case that v(A) = 1 for all A ∈ Γ and v(B) ∈ {0, 12 } for all
B ∈ ∆.
Letting L′ and L′′ be two systems characterized semantically by means of SKvaluations, we can define a mixed system L as in Definition 3.5—in which case we
will describe L as L′ /L′′ .12 Thus, ST can be described as K3 /LP.
Definition 3.5. A SK-valuation v satisfies an inference Γ ⇒ ∆ in L (v L Γ ⇒ ∆) if
and only if it is not the case that v L′ A for all A ∈ Γ and v 2L′′ B for all B ∈ ∆.
Going back to our main theme, though, we must recall that even if there are some
important similarities between CL and ST—indeed, even if they coincide at the inferential level—they do not coincide at every inferential level. More precisely, there are
metainferences valid in CL but invalid in ST. To wit, the following
12
This is, very much, the definition of mixed consequence discussed by Chemla, Egré and Spector in
their recent article [6]—with changes in the definition, to adapt their notation to ours, being inessential.
9
p, q ⇒ r, s p, q, r ⇒ s
p, q ⇒ s
∅ ⇒ p ∅ ⇒ ¬p
∅⇒q
are instances of the schematic metainferences depicted below, respectively called Cut
and meta Explosion (following Zardini in [28], and Barrio, Rosenblatt and Tajer in [4]).
All of them are valid in CL but invalid in ST, according Definition 3.6.
Γ ⇒ A, ∆ Γ, A ⇒ ∆
Γ⇒∆
Γ ⇒ A, ∆ Γ ⇒ ¬A, ∆
Γ ⇒ B, ∆
To observe the disparity in this regard between CL and ST—without any appeal
to a sequent calculus—it is crucial to understand how metainferences can be taken
to be valid or invalid in a certain system. In this regard we take a semantic stance,
borrowing from [14] and [12] the distinction between the local and the global definition
of metainferential validity. Saying that a metainference is locally valid according to
a certain logic L means that if the premise inferences are satisfied by a valuation in
L, so is the conclusion inference. This is different to saying that a metainference is
globally valid according to a certain logic L, which requires that whenever the premise
inferences are valid in L, so is the conclusion. In a nutshell, the difference between local
and global metainferential validity consists in nothing more than the difference between
satisfaction-preservation and validity-preservation. In what remains, we will adopt the
local notion, postponing our justification for this choice until Section 5. Below, we detail
the definition of local metainferential validity for CL and ST.13
Definition 3.6. For a metainference Γ ⇒1 A, where Γ ⊆ SEQ(L) and A ∈ SEQ(L),
a Boolean valuation v satisfies Γ ⇒1 A in CL (v CL Γ ⇒1 A) if and only if v 2CL γ
for some γ ∈ Γ, or v CL A. Similarly, a SK-valuation v satisfies Γ ⇒1 A in ST
(v ST Γ ⇒1 A) if and only if v 2ST γ for some γ ∈ Γ, or v ST A.
We close this section by calling the attention to the fact that, while the above remarks
point towards the non-classical nature of ST—regarding its valid metainferences—they
do not say much about what is particularly distinctive about its non-classicality. For
all we know, ST could be non-classical metainferentially speaking, but there could be
no recognizable pattern in its non-classical behavior. A clearer understanding of the
non-classical nature of metainferential validity in ST is, nevertheless, delivered by a
recent number of articles by Barrio, Rosenblatt and Tajer [4] and, independently, by
Pynko [19]. In their work, it is clearly described how it is possible to characterize
the metainferential non-classicality of ST in terms of the inferential non-classicality of
Graham Priest’s LP. In other words, every metainference that is invalid in ST has a
corresponding inference that is invalid in LP, and viceversa.
This result is achieved, in one of its versions, through an indispensable translation
function from metainferences to inferences, neatly laid out by Dicher and Paoli in [12].
It is interesting and illustrative to notice that our translation function below is intended
to be inspired by theirs, while simultaneously subsuming it. Indeed, while their function
13
For what it is worth, for a metainference Γ ⇒1 A, where Γ ⊆ SEQ(L) and A ∈ SEQ(L), we say it
is globally valid in ST if and only if 2ST γ for some γ ∈ Γ, or ST A.
10
allows to translate metainferences to inferences and inferences to formulae, ours allows
to translate—by applying essentially the same ideas—metainferences of any arbitrary
large inferential level to metainferences of the immediately predecessor level, as well as
metainferences to inferences, and inferences to formulae.
S
S
Definition 3.7. lower:
SEQn (L) −→ F OR(L) ∪
SEQn (L) is defined so that:
n∈ω
• lower(Γ ⇒ ∆) =
V
Γ⊃
n∈ω
W
∆
• for 1 ≤ n, lower(Γ ⇒n A) = {lower(γ) | γ ∈ Γ} ⇒n−1 lower(A)
Notice that, given the above definition, if Γ 6= ∅ and ∆ = ∅ then lower(Γ ⇒ ∆) = ¬A1 ∧
· · · ∧ ¬An for Ai ∈ Γ. Similarly, if Γ = ∅ and ∆ 6= ∅ then lower(Γ ⇒ ∆) = B1 ∨ · · · ∨ Bn
for Bi ∈ ∆. We take, furthermore, Γ ⇒0 ∆ to be just Γ ⇒ ∆. It is, then, with the help
of these definitions that we can formally rephrase the previously mentioned collapse
result for ST and LP—interpreted by Barrio, Rosenblatt and Tajer in [4] as meaning
that LP is nothing but the logic of the metainferences of ST.14
Fact 3.8. For all Γ, ∆ ⊆ F OR(L):
ST Γ ⇒1 ∆
if and only if
LP lower(Γ ⇒1 ∆)
As we said earlier, given these results one might be tempted to say that ST differs
from CL because they do not agree at the metainferential level. Two logical systems
may be claimed to be identical only if they coincide with regard not only to their valid
inferences, but also with regard to their valid metainferences. But, as promising as
this solution might be, it constitutes an unsatisfactory answer. The reason for this is
that the case of ST can be generalized. In the next section, we do this by constructing
a sequence of logics which coincide with CL in progressively many inferential levels,
although all of them start behaving non-classically at some point—and when they do,
they will just like ST behave in a way essentially tied to LP.
4
The ST phenomenon, generalized
In this section we will, first, present a logic which coincides with CL regarding its valid
metainferences, although not regarding its metametainferences, or metainferences of
level 2. Later, we will present a number of systems coinciding with CL in progressively
many inferential levels although not in all of them. This, we claim, will allow us to
provide an impossibility result: it is impossible to identify a logic with its given set of
valid inferences of any level whatsoever, for it is possible to find logics coinciding with
it in that respect despite being intuitively different from it.
To start up this part of the investigation, though, we need to focus on a system
closely related to ST, namely the logic TS. This system incarnates a mixed notion of
logical consequence that is dual to the one characteristic of ST—if ST can be described
14
For more on the relation between ST and paraconsistency, see [3] and [1].
11
as the mixed system K3 /LP, then TS can be seen as the mixed system LP/K3 . With
its help we are now able to define yet another logic that we call TS/ST, precisely
because logical consequence for it is defined in a mixed way with the help of TS and
ST.15 The definition of logical consequence for this system, appearing below, is provided
at the level of metainferences—ruling only derivatively over inferences and formulae.16
Definition 4.1. For a metainference Γ ⇒1 A, where Γ ⊆ SEQ(L) and A ∈ SEQ(L),
we say a SK-valuation v satisfies Γ ⇒1 A in TS/ST (v TS/ST Γ ⇒1 A) if and only if
v 2TS γ for some γ ∈ Γ, or v ST A.
Furthermore, the system TS/ST can be rightfully called a STian logic, for it defines
its central notion of logical consequence as outlined above by adopting the classical
standard to be satisfied as the premise of a metainference, and the classical standard
to be regarded as the counter-exemplifying conclusion of a metainference. This can be
observed in the lemma stated below, whose proof we leave to the reader as an exercise.17
Lemma 4.2. For all Γ, Θ, ∆, Π ⊆ SEQ(L), and all SK-valuations v, there is a Boolean
valuation v ∗ such that:
if
then
v 2ST Γ ⇒ ∆
v 2CL Γ ⇒ ∆
and
and
∗
v TS Θ ⇒ Π,
v ∗ CL Θ ⇒ Π
Similarly, for all Boolean valuations v, there is a SK valuation v ∗ such that:
if
v 2CL Γ ⇒ ∆
and
v CL Θ ⇒ Π,
then
v 2ST Γ ⇒ ∆
and
v ∗ TS Θ ⇒ Π
∗
In light of these remarks, that CL and TS/ST have the same valid metainferences
becomes straightforwardly provable. We will not show the details of the proof here,
given it follows from the more general Theorem 4.12 appearing below.
Fact 4.3. For all Γ ⊆ SEQ(L) and A ∈ SEQ(L):
TS/ST Γ ⇒1 A
if and only if
CL Γ ⇒1 A
Perhaps looking at TS/ST in this way will give the reader the feeling that, if this
system coincides with CL regarding its metainferences (and, hence, its inferences and
15
TS/ST is just one member of a wide family of metainferential logics. For more about such a family,
see [16]. In addition, the introduction of TS/ST has, in itself, an effect in a variety of philosophical
phenomena. To name a few, it poses new challenges to the issue of Logical Pluralism and the so-called
Collapse Argument for it (as we discussed in [2]), it can be implemented to give a new solution to
semantic paradoxes (as argued by one of us in [17]), and to suggest that paraconsistency might be
something more than the mere invalidity of certain inferential forms of Explosion (as done by us in [3]).
16
Although, as remarked by an anonymous reviewer, Definition 4.1 is an instance of Definition 4.7,
the latter requires understanding the hierarchy of logics we define next whereas the former does not.
17
An anonymous reviewer sharply points out that instead of proving conditional claims which have
conjunctions on their antecedents and consequents, we could prove the conjunction of two conditionals
linking, respectively, their first conjunct and their second conjunct. We would like to clarify that
although this is certainly possible, in the context of the overall strategy of our proofs in this article, the
current claims are sufficient enough.
12
valid formulae) then TS/ST ought to be identified with CL. Needless to say, knowing
that TS/ST is meant to be an instance of the ST phenomenon, it is quite obvious that
this conclusion must be resisted. In fact, TS/ST can be shown not to coincide with CL,
as its metametainferences (or metainferences of level 2) do not coincide with those valid
in CL. This can be checked, by paying attention to the definition of metametainferential
validity in CL and TS/ST.
Definition 4.4. For a metainference Γ ⇒2 A, where Γ ⊆ SEQ1 (L) and A ∈ SEQ1 (L),
we say a Boolean valuation v satisfies Γ ⇒2 A in CL (v CL Γ ⇒2 A) if and only if
v 2CL γ for some γ ∈ Γ, or v CL A. Similarly, a SK-valuation v satisfies Γ ⇒2 A in
TS/ST (v TS/ST Γ ⇒2 A) if and only if v 2TS/ST γ for some γ ∈ Γ, or v TS/ST A.
With the help of this notion of validity, some examples become available of the
disparity between CL and TS/ST. In particular, the following metametainferences
(instances, themselves, of the schematic metametainferences meta Cut and metameta
Explosion referred in Section 2) are valid in the former, although they are invalid in the
latter. Below, ∅ stands for the empty set of premises.
∅
p ⇒ q, r
∅
p, r ⇒ q
∅
⇒ ¬p
∅
⇒q
∅
⇒p
∅
p⇒q
Moreover, it is interesting to notice that TS/ST follows the path initiated by ST.
It coincides with CL up to a certain inferential level at which it starts to works nonclassically, but furthermore a proper characterization of its metametainferential nonclassicality shows its intimate relation with LP. Thus, not only is TS/ST non-classical
at the level of its metametainferences, but every metametainference valid in TS/ST has
a translation in terms of an inference valid in LP, as shown in the following fact (which
we will not be stopping to prove since it follows from the more general Theorem 4.16
appearing below). It can be interpreted as saying that, inasmuch as LP is the logic of
the metainferences of ST, LP is the logic of the metametainferences of TS/ST.
Fact 4.5. For all Γ ⊆ SEQ1 (L) and A ∈ SEQ1 (L):
TS/ST Γ ⇒2 A
if and only if
ST lower(Γ ⇒2 A)
In other words:
TS/ST Γ ⇒2 A
if and only if
LP lower(lower(Γ ⇒2 A))
Thus, one might be tempted to say that TS/ST differs from CL because they
do not agree at the metametainferential level. Two logical systems may be said to
be identical only if they coincide with regard not only to their valid inferences and
metainferences, but also with regard to their valid metametainferences. Once again, as
promising this solution might be, it constitutes an unsatisfactory answer. The reason
is that the ST phenomenon is indeed endemic, precluding us from identifying CL
not only with its set of valid inferences, or valid metainferences, but also with any of
its set of valid metainferences for any arbitrary large inferential level. We face this
13
impossibility because we can define a ST-like system that coincides with CL pertaining
those metainferences, but differs regarding the next metainferential level. Showing this
requires defining a sequence of systems that are progressively more coincidental with
CL. We turn to this task next.
But before, notice that once we grasp the semantic definition of logical consequence
for a given system this settles which formulae, inferences and metainferences of any
arbitrary level are valid in it. If we give the definition of logical consequence in L aiming
at regular inferences—as we did for K3 , LP, ST and even for TS—this consequently
settles both the issues of which formulae are valid and the issue of which metainferences
are valid. It determines a definitive answer to the former, since we can understand
formulae as degenerate cases of inferences with empty premises, and it determines a
definitive answer to the latter, since we know how to evaluate metainferences for these
logics, applying the notion of local metainferential validity. If, alternatively, we give the
definition of logical consequence in L aiming at metainferences—as we did for TS/ST—
this consequently settles the issue of which formulae and which inferences are valid, along
with the issue of which metametainferences (or metainferences of level 2) are valid. All
of these for reasons perfectly analogous to the previously given. It is easy to observe
that if we give the definition of logical consequence for Lj aiming at metainferences of
level k, then this will settle the issue of which metainferences of level m lower than k
are valid, along with which metainferences of level n greater than k are valid.
Thus, to construct a sequence of ST-like logics we need to give the definition of
logical consequence of each of these systems to aim at progressively higher inferential
levels. That is, the definition of logical consequence for the first logic—namely, ST—
will be aimed at regular inferences, for the second—namely, TS/ST—will be aimed at
metainferences, for the third at metametainferences, and so on and so forth. This is
what we intend to do with the pair of definitions appearing below.
Definition 4.6. The collection ST = {Li | i ∈ N} of logical systems is recursively
defined so that L0 = LP, L1 = ST, and for 2 ≤ j, Lj = Lj−1 /Lj−1 (where Lj = Ln /Lm
if Lj = Lm /Ln ).
Definition 4.7. For 2 ≤ j and Lj ∈ ST, a metainference Γ ⇒j−1 A, where Γ ⊆
SEQj−2 (L) and A ∈ SEQj−2 (L), we say a SK-valuation v satisfies Γ ⇒j−1 A in Lj
(v Lj Γ ⇒j−1 A) if and only if v 2Lj−1 γ for some γ ∈ Γ, or v Lj−1 A.
With these technical tools at hand, we show below that the hierarchy is indeed a
collection of systems instantiating the ST phenomenon. We do this with the help of
three groups of formal results, which we explain and comment after stating them.
First, we argue that the previously defined hierarchy is a set of systems instantiating
the ST phenomenon, because each logic of the hierarchy can rightfully be called a stricttolerant logic. In fact, each system of the hierarchy mirrors ST’s adoption of the classical
standard to be satisfied as the premise of an inference, and the classical standard to be
regarded as a counter-exemplifying conclusion.18 Each logic Lj of the hierarchy is taken
to be defined by its notion of validity for j-level metainferences, and in all such cases this
18
What we mean by this is that in each logic, and for every valuation v, a premise is satisfied by v in
14
definition requires that the premises of a metainference of level j − 1 do satisfy classical
standard to be regarded as the premises of a sound inference of this sort (this being the
strict part of the definition) and at the same time that the conclusion inference does not
satisfy the classical standard to be regarded as a counter-exemplifying conclusion of a
level j − 1 metainference (this being the tolerant part of the definition). These facts are
formalized in the statement of the following lemma.
Lemma 4.8. For all n ≥ 1, for all Γ, Θ ⊆ SEQn−1 (L), A, B ∈ SEQn−1 (L), and all
SK-valuations v, there is a Boolean valuation v ∗ such that:
if
v 2Ln+1 Γ ⇒n A
and
v Ln+1 Θ ⇒n B,
then
v ∗ 2CL Γ ⇒n A
and
v ∗ CL Θ ⇒n B
Similarly, for all Boolean valuations v, there is a SK-valuation v ∗ such that:
if
then
v 2CL Γ ⇒n A
v 2Ln+1 Γ ⇒n A
and
and
∗
v CL Θ ⇒n B,
v Ln+1 Θ ⇒n B
∗
Proof. We prove the first conditional, by induction on the index of the logic.
Base case: n = 1. Assume that there is a SK valuation v such that v 2L2 Γ ⇒1 A
and v L2 Θ ⇒1 B, i.e. v 2TS/ST Γ ⇒1 A and v ST/TS Θ ⇒1 B. From the fact that
v 2TS/ST Γ ⇒1 A we may infer that v TS γ, for all γ ∈ Γ and that v 2ST A. From the
fact that v ST/TS Θ ⇒1 B we may infer that v 2ST θ, for some θ ∈ Θ, or v TS B.
Furthermore, from all these facts and Lemma 4.2 we are guaranteed that there is a
Boolean valuation v ∗ such that, on the one hand, v ∗ CL γ, for all γ ∈ Γ and v ∗ 2CL A
and, on the other, v ∗ 2CL θ, for some θ ∈ Θ, or v ∗ CL B. But, by definition, this is
the same as saying that there is a Boolean valuation v ∗ such that v ∗ 2CL Γ ⇒1 A and
v ∗ CL Θ ⇒1 B.
Inductive step: n > 1. Assume that there is a SK valuation v such that v 2Ln+1
Γ ⇒1 A and v Ln+1 Θ ⇒1 B. From the fact that v 2Ln+1 Γ ⇒1 A we may infer that
v Ln γ, for all γ ∈ Γ and that v Ln A. From the fact that v Ln+1 Θ ⇒1 B we
may infer that v 2Ln θ, for some θ ∈ Θ, or v Ln B. Furthermore, from all these facts
and the Inductive Hypothesis we are guaranteed that there is a Boolean valuation v ∗
such that, on the one hand, v ∗ CL γ, for all γ ∈ Γ and v ∗ CL A and, on the other,
v ∗ 2CL θ, for some θ ∈ Θ, or v ∗ CL B. But, by definition, this is the same as saying
that there is a Boolean valuation v ∗ such that v ∗ 2CL Γ ⇒n A and v ∗ CL Θ ⇒n B.
The proof of the remaining conditional is similar.
Second, the previously defined hierarchy is a set of systems instantiating the ST
phenomenon, because every such system mimics the behavior of ST. By this we mean,
on the one hand, that all the logics of the hierarchy, after ST, have the same valid
inferences than CL—as witnessed in Lemma 4.9 and Theorem 4.10, whose proofs we
Classical Logic if and only if it is satisfied by the standard for premises, and v is a counterexample to
the validity of the conclusion in Classical Logic if and only if v is a counterexample in the logic defined
by the standard for the conclusion.
15
leave to the reader as an exercise. On the other hand, by this we mean that just
like ST coincides with CL up to its valid inferences, for each inferential level j the
hierarchy counts with a system Lj+1 which has the same valid inferences than CL up
to that inferential level. In this regard, the cumulative nature of the progression of
logics in the hierarchy, i.e. the fact that whatever validities obtain in a given logic these
are retained in the successive systems appearing in the hierarchy, is represented by
Lemmata 4.9 and 4.11. Whereas, the fact that these validities are in fact progressively
more and more classical is formalized by Theorem 4.12.
Lemma 4.9. For all j ≥ 1, for all Γ, ∆ ⊆ F OR(L), for every SK-valuation v:
v L j Γ ⇒ ∆
if and only if
v CL Γ ⇒ ∆
Theorem 4.10. For all j ≥ 1, for all Γ, ∆ ⊆ F OR(L):
L j Γ ⇒ ∆
if and only if
CL Γ ⇒ ∆
Lemma 4.11. For all j > n ≥ 2, for all Γ ⊆ SEQn−1 (L), A ∈ SEQn−1 (L), for every
SK-valuation v:
v Ln Γ ⇒n A
if and only if
v L j Γ ⇒n A
Theorem 4.12. For all n ≥ 1, for all Γ ⊆ SEQn−1 (L), A ∈ SEQn−1 (L)
Ln+1 Γ ⇒n A
if and only if
CL Γ ⇒n A
Proof. The proof is by induction on the index of the logic.
Base case: n = 1. From left to right, let us suppose that 2CL Γ ⇒1 A, from which
we infer that there is a Boolean valuation v such that v CL γ, for all γ ∈ Γ, and yet
v 2CL A. But, since Boolean valuations are a subset of SK valuations, we know by the
definition of satisfaction in TS and ST that v is also a SK valuation such that v TS γ,
for all γ ∈ Γ, and yet v 2ST A. Thus, by the definition of validity of a metainference
of level 1 in TS/ST, we know that v 2TS/ST Γ ⇒1 A, whence 2TS/ST Γ ⇒1 A. From
right to left, let us suppose that 2TS/ST Γ ⇒1 A, from which we infer that there is a
SK valuation v such that v TS γ, for all γ ∈ Γ, and yet v 2ST A. By Lemma 4.2 we
know that there is a Boolean valuation v ∗ such that v ∗ CL γ, for all γ ∈ Γ, and yet
v ∗ 2CL A. Therefore, we know that v ∗ 2CL Γ ⇒1 A, whence 2CL Γ ⇒1 A.
Inductive step: n > 1. From left to right, let us suppose that 2CL Γ ⇒1 A, from
which we infer that there is a Boolean valuation v such that v CL γ, for all γ ∈ Γ,
and yet v 2CL A. By Lemma 4.8 we know that there is a SK valuation v ∗ such that
v ∗ Ln γ, for all γ ∈ Γ, and yet v ∗ 2Ln A. Therefore, we know that v ∗ 2Ln+1 Γ ⇒n A,
whence 2Ln+1 Γ ⇒1 A. From right to left, let us suppose that 2Ln+1 Γ ⇒1 A, from
which we infer that there is a SK valuation v such that v Ln γ, for all γ ∈ Γ, and
yet v 2Ln A. By Lemma 4.8 we know that there is a Boolean valuation v ∗ such that
v ∗ CL γ, for all γ ∈ Γ, and yet v ∗ 2CL A. Therefore, we know that v ∗ 2CL Γ ⇒n A,
whence 2CL Γ ⇒1 A.
16
Third, we claim that the previously defined hierarchy is a set of systems instantiating
the ST phenomenon, because each system of the hierarchy mimics ST in coinciding with
CL up to a certain inferential point, and then starting to behave non-classically in a
way that is essentially tied to LP. The results below witness this by representing—via
the lower translation—those inferences valid in the logics of the hierarchy, in terms of
inferences of lower inferential levels which are valid in systems appearing previously in
the hierarchy. With regard to regular inferences, this is featured in Lemma 4.13 and
Theorem 4.14, whose proofs we leave to the reader as an exercise. In this respect, just
like the metainferences valid in ST (which are different from those valid in CL) can be
characterized in terms of valid inferences of LP via the lower translation, it is possible
to characterize the metainferences of level j which are valid in the logic Lj+1 of the
hierarchy—and which are, again, different from those valid in CL—in terms of valid
inferences of LP, via successive applications of the lower translation, as in Lemma 4.15
and Theorem 4.16, which immediately follows from it.
Lemma 4.13. For all j, for all Γ, ∆ ⊆ F OR(L), for every SK-valuation v:
v Lj+1 Γ ⇒ ∆
if and only if
v Lj lower(Γ ⇒ ∆)
Theorem 4.14. For all j, for all Γ, ∆ ⊆ F OR(L):
Lj+1 Γ ⇒ ∆
if and only if
Lj lower(Γ ⇒ ∆)
Lemma 4.15. For all j, for all n ≥ 1, for all Γ ⊆ SEQn−1 (L), A ∈ SEQn−1 (L), for
every SK-valuation v:
v Lj+1 Γ ⇒n A
if and only if
v Lj lower(Γ ⇒n A)
Proof. We prove this via two nested inductions. The principal induction is on the index
of the logic, and the secondary induction is on the inferential level.
Outer base case: j = 0. Recall that L0 = LP and L1 = ST.
Inner base case: n = 1. Suppose v ST Γ ⇒1 A. By Definition 4.7, this happens if
and only if either v 2ST γ, for some γ ∈ Γ, or v ST A. In turn, letting γ be of the form
∆ ⇒ Σ, the former happens if and only if v(δ) = 1, for every δ ∈ ∆, and v(σ) = 0, for
every conclusion σ ∈ Σ. Additionally, letting A be of the form Π ⇒ Λ, the latter happens
if and only if v(π) ∈ {0, 21 }, for some π V
∈ Π, orW
v(λ) = { 21 , 1}, for some λ ∈ Λ. Moreover,
the former is theVcase if and
W only if v( ∆ ⊃ Σ) = 0. Whereas, the latter is the case
if and only if v( Π ⊃ Λ) ∈ { 12 , 1}. Moving forward, the first can be alternative
understood as v 2LP lower(γ), while the second can be understood as v LP lower(A).
Finally, either these two facts entails that v LP {lower(γ) | γ ∈ Γ} ⇒ lower(A), which
by definition amounts to v LP lower(Γ ⇒1 A). The other direction just follows by
reversing the same reasoning.
Inner inductive step: 2 ≤ n. Suppose v ST Γ ⇒n A. This obtains if and only if
either v 2ST γ, for some γ ∈ Γ, or v ST A. Notice that γ ∈ Γ and A are inferences
of level n − 1. Hence, by the Inductive Hypothesis, we can establish that either v 2LP
17
lower(γ), for some γ ∈ Γ, or v LP lower(A). In either of these cases, we have that
v LP lower(Γ ⇒n A). The remaining direction just follows by reversing the same
reasoning.
Outer inductive step: 1 ≤ j.
Inner base case: n = 1. Suppose v Lj+1 Γ ⇒1 A. We know, by Lemma 4.11, that
v Lj+1 Γ ⇒1 ∆ if and only if v TS/ST Γ ⇒1 ∆. This, in turn, happens if and only if
for some γ ∈ Γ, v 2TS γ, or v ST A. Without loss of generality, let γ be of the form
∆ ⇒ Σ. Thus, v 2TS ∆ ⇒ Σ if and only if for every δ ∈ ∆, v(δ) ∈ {1, 21 }, and for
every σ ∈ Σ, v(σ) ∈ { 21 , 0}. Similarly, letting A be of the form Π ⇒ Λ, v ST Π ⇒ Λ
1
1
implies that for some π ∈ Π, v(π)
V ∈ {0,
W 2 }, or1 for some λ ∈ Λ, v(λ) ∈ { 2 , 1}. The
former obtains
ifWand only if v( ∆ ⊃ Σ) ∈ { 2 , 0}, whereas the latter obtains if and
V
only if v( Π ⊃ Λ) ∈ { 12 , 1}. Moving forward, the first can be alternative understood
as v 2K3 lower(γ), while the second can be understood as v LP lower(A). Moreover,
either these two facts entails that v ST {lower(γ) | γ ∈ Γ} ⇒ lower(A), which by
definition amounts to v ST lower(Γ ⇒1 A). Finally, by Lemma 4.9, this implies
v Lj lower(Γ ⇒1 A). The other direction just follows by reversing the same reasoning.
Inner inductive step: 2 ≤ n. Suppose v Lj+1 Γ ⇒n A. This obtains if and only
if either v 2Lj+1 γ, for some γ ∈ Γ, or v Lj+1 A. Notice that γ ∈ Γ and A are
inferences of level n − 1. Hence, by the Inductive Hypothesis, we can establish that
either v 2Lj lower(γ), for some γ ∈ Γ, or v Lj lower(A). In either of these cases,
we have that v Lj {lower(γ) | γ ∈ Γ} ⇒n lower(A), which by definition amounts
to v Lj lower(Γ ⇒n A). The remaining direction just follows by reversing the same
reasoning.
Theorem 4.16. For all j, for all n ≥ 1, for all Γ ⊆ SEQn−1 (L), A ∈ SEQn−1 (L):
Lj+1 Γ ⇒n A
if and only if
Lj lower(Γ ⇒n A)
In other words:
Lj+1 Γ ⇒n A
LP lower...(lower(Γ ⇒n A)...)
{z
}
|
if and only if
n times
All the previously discussed results are illustrated in the next figure. In the vertical axis we have the indication of the inferential levels being evaluated, while in the
horizontal axis we have the systems of the hierarchy as they are recursively generated.
18
..
.
..
.
metametainferences
w
6=
w'
6=
w''
6=
w'''
6=
w
...
metainferences
z
6=
z'
6=
z''
6=
z
=
z
...
inferences
y
6=
y'
6=
y
=
y
=
y
...
theorems
x
=
x
=
x
=
x
=
x
...
L3
...
CL
..
.
..
.
L0
LP
L1
ST
..
.
L2
TS/ST
..
.
To sum up, as we previously remarked, the hierarchy allows to convey a sort of
philosophical impossibility result: that it is impossible to identify a logic with its given
set of valid inferences of any inferential level—for it is possible to find logics which will
coincide in that respect, despite of being intuitively different from it. Such logics are
brought to our attention through the ST phenomenon. This phenomenon, already as
presented in the literature, allows to notice that the commonly used identity criterion
for logics—according to which a logic is a consequence relation between (collections
of) formulae—cannot be entirely right. This is because, as previously argued, even if
both ST and CL have the same valid inferences they are intuitively different logics.
But the ST phenomenon is not circumscribed only to the logic ST. Indeed, the logics
in our hierarchy show that it is equally dissatisfying to identify a logic with a set of
valid metainferences—for what it is worth, systems like TS/ST and CL agree in this
respect, despite of being radically different logics. Or to identify a logic with its set of
valid metametainferences, or with its set of valid metainferences of level 3, or 4, ..., or
any arbitrarily large level. Whence, the need of an appropriate identity criterion for
logical systems that can cope with cases of this sort.
Before moving to the further difficulties that coming with such a solution brings, we
would like to comment on two related issues connected with the ST phenomenon. The
first one is the characterization of paraconsistency, as a property of logical systems in
general. While normally paraconsistency is associated with the failure of some version
of Explosion—i.e. with an inference being invalid—the results hinted at here highlight
the possibility of a different phenomenon, that of paraconsistency associated with the
failure of metainferential versions of Explosion, which we might call substructural paraconsistency. Thus, there may be logics which are paraconsistent in the usual sense and
in this new sense, just like there may be logics which are paraconsistent in only one way
but not the other, i.e. logics with homogeneous policies and logics with heterogeneous
policies. The existence of such systems is confirmed by the previous results, which generalize the discussion of (anonymized). In fact, the hierarchy exhibits logics which have
a certain degree of classicality and a certain degree of paraconsistency. This is to be
expected, given the hierarchy is constructed in the image of ST, which behaves classically up to the inferential level, but behaves “paraconsistently” from the metainferential
level onward. Something essentially analogous happens for every logic in our hierarchy.
19
Particularly, for each j, the logic Lj behaves classically in the inferential levels below j,
and then “paraconsistently” from the inferential level j.
The second issue on which these results have a significant echo is that of Logical
Pluralism, especially as pertains to the so-called Collapse Argument. In essence, the
various forms of the Collapse Argument for Logical Pluralism—as discussed e.g. in [25],
[18], [20], and [15]—aim at showing how someone embracing a plurality of logics might
find herself being a Logical Monist at the end of the day, given the normative guidance
that one is expected to have from logical systems in general. Were a Pluralist to believe
the premises of an inference that is valid according to one of the logics she embraces,
but not to the other, should she believe the conclusion or not? Since there can be no
pluralism about this, Logical Pluralism collapses into Logical Monism.
Interestingly, our results resonate in this debate because two different logics need
not differ with regard to their valid inferences, but in other respect—e.g. concerning
their valid metainferences of some level. But, if a Logical Pluralist were to embrace two
logical systems with the same set of valid inferences, then she would not be affected by
the current form of the Collapse Argument. This observation should lead those opposing
the Pluralist view to refine their argument, in order to adapt that not only to logics
with the same valid inferences and different valid metainferences, but to logics with the
same valid metainferences of level n and different valid metainferences of level n + 1. In
this respect, a refined Collapse Argument could be designed, along the following lines.
Suppose a Pluralist embraces two different logics with the same inferences of level n but
different inferences of level n + 1. Were she to believe the premises of an inference of
level n + 1 that is valid according to one of the logics she embraces, but not to the other,
should she believe the conclusion or not? Since there can be no pluralism about this,
Logical Pluralism collapses into Logical Monism once again.19 A precise formulation of
this strategy was presented by us in [2]—where in all the remaining relevant respects
no argument either for or against Logical Pluralism is put forward.
Finally, let us address the elephant in the room. How are we going to identify
logical systems if the usual option of identifying it with its set of valid inferences is
no good, and neither is any of the intuitive fixes leading to identify it with its set
of valid metainferences of some arbitrarily large level n? Our proposal is, precisely,
to embrace these facts, by identifying a logic with its set of valid inferences at every
inferential level. Thus, what may suffice to differentiate a pair logical systems could
be their set of theorems, but that might not be sufficient for another pair, for which
their set of valid inferences might be required. However, there might be yet another
pair for which that could be still insufficient, requiring to draw the attention to their
valid metainferences, as could be the case with a different pair of logics whose valid
metainferences may be identical, thereby highlighting the need to pay attention to their
valid metametainferences—i.e. their valid metainferences of level 2. And so on, and so
forth. It is easy to see that no distinguished inferential level can be the stopping point,
although all of them could be of some use in differentiating certain logical systems.
To put our proposal differently, then, we may claim that a logic is to be identified
19
We would like to thank an anonymous reviewer for suggesting us to further comment on this issue.
20
with an infinite sequence of consequence relations revealing which formulae are valid,
which inferences are valid, which metainferences of level 1 are valid, which metainferences of level 2 are valid, . . . , and which metainferences of level n are valid—for every
natural number n. What we are claiming is that however a system is defined—at least
from the point of view of its semantics—this determines which formulae, inferences,
and metainferences of any arbitrary large inferential level are valid. Putting this sets
of validities one next to each other and forming a sequence allows to differentiate this
system from other systems highly resembling it in such a clear way, which suggests
taking this sequence to constitute the identity, the DNA of a logic.
Remarkably, such a criterion allows to provide an answer to our initial question:
what identifies CL as such? It is, in fact, the inferences it validates at every inferential
level that are crucial to its identity and which allow to tell it apart not only from closely
related systems like ST, but also from systems that are even more similar to it—like
TS/ST or any of the recursively defined systems appearing in our hierarchy, for that
matter. What identifies CL are all its valid inferences of every inferential level, and
each of those systems differ with it at some point. This clearly explains why ST is not
CL, and neither is TS/ST or any of the systems of our hierarchy. Subsequently, this
criterion makes it also easy to tell the difference between the systems in the hierarchy
themselves.
A further advantage of such a criterion is that it induces an interesting scale of
classicality for logical systems and, in particular, for those systems in the hierarchy. We
can certainly say that a system is similar to CL if it agrees with it to some extent, i.e.
with regard to its theorems, or inferences, or metainferences, or something else. But,
more concretely, we can say that a certain logic is k-similar to CL if it agrees with it up
to the metainferential level k—i.e. it coincides with regard to their valid metainferences
of level i ≤ k, as well as their valid inferences and their theorems. In this respect, for
instance, ST would be 0-similar while TS/ST would be 1-similar to CL. This helps
to cash out formally the fact that, in some intuitive sense, TS/ST is classical to a
greater degree than ST, whereas ST itself is classical to a greater degree than e.g. LP.
Thus, in this vein, two logics formulated over the same language will be identical in our
proposed sense if and only if they are n-similar, for every natural number n.
After these remarks, in the next section we consider some objections that may be
directed at our account, and provide appropriate responses for each of them.
5
Replies to some objections
One of the plausible things that can be called into attention as regards our discussion is
the way in which metainferences are ruled to be either valid or invalid. In this respect,
it is worth mentioning that we adopted a semantic perspective towards this issue—not
that we are not interested in other understandings of these things, but we will not be
commenting on them or on their relation with the semantic readings of metainferential
21
validity in the context of this article.20
This being said, it is the rather different available semantic approaches to this phenomenon which should be evaluated, in particular, with the aim of telling some kind
of story that explains why we decided to go with the local and not with another understanding of metainferential validity. Remember that, as discussed in Section 3, the
difference between local and global metainferential validity consists in nothing more
than the difference between satisfaction-preservation and validity-preservation in a certain logic. To wit, taking ST as an example, from the below the leftmost is locally
and globally valid, the middle one is globally but not locally valid, and the rightmost
is neither locally nor globally valid.
p, q ⇒ p p, q ⇒ q
p, q ⇒ p ∧ q
p⇒q q⇒r
p⇒r
p∧q ⇒p
p⇒p∧q
In this regard, we think that there is a considerable conceptual advantage in ruling
metainferential validity locally. It comes from the fact that the local reading allows to
provide a unified account of validity for all kinds of inferences—that is, for inferences
between formulae, between inferences, between metainferences, and so on. While ruling
over the validity of regular inferences we require that, if the premises are satisfied
(according to the standard for premise-formulae to be satisfied in the logic in question),
then at least one of the conclusions is satisfied (according to the standard for conclusionformulae to be satisfied in the said logic). Ruling over the validity of metainferences
with a local criterion asks, precisely, the same thing. Thus, if we are going to think
about logic as a discipline concerned with the study and analysis of valid inferences,
adopting something different than the local understanding of metainferential validity
would mean that validity is understood one way as pertaining to formulae and another
way as pertaining to inferences. But this, we think, seems rather undesirable. Having
a unified stance towards validity, regardless of its relata, seems like a promising and
interesting endeavor, and the local reading allows for this unification. This is why we
think local validity is the way to go.
This being said, the global understanding could, in principle, be equally suitable
for a cogent analysis of this sort. However, global validity as applied to metainferential
validity has a considerable flaw, i.e. it vastly overgenerates the set of valid metainferences. In fact, it is enough to have a metainference whose premise-set contains an invalid
inference, for the corresponding metainference to be globally valid—albeit in a trivial
sense. This is not the case with the local criterion, even if the premise-inferences are
invalid. Nevertheless, this does not represent in itself a sufficient reason for discarding
global validity. A further ground for the rejection of global validity is its uselessness for
the study of validity as a unified phenomenon. Indeed, were we to adopt that global
reading for the study of all kinds of inferences, we would need to do this for regular
inferences too. This would require of valid inferences that, instead of preserving truth
from premises to conclusion, they preserve the property of being a tautology. But this is
20
We mention this not because there is a special connection between ruling metainferences to be either
valid or invalid and adopting a semantic approach, but because there are other available ways of doing
so both semantically (e.g. global validity) and proof-theoretically (e.g. admissibility, derivability).
22
seriously non-standard, as it would deem that the inference p ⇒ q is valid—just because
p is not a tautology—as it would deem any inference with an invalid premise-formulae,
as valid. Given this, we think enough reasons have been presented to favor the local
instead of the global understanding.
An additional objection, claiming that our criterion for the identity of logical systems
is wrong, can be extracted from Dicher and Paoli’s work, [12] and [11]. These authors
favor the identification of logics with a certain class of abstract consequence relations—
as defined by Blok and Jónsson in e.g. [5]. This, together with looking at ST as
a consequence relation holding between sets of inferences and inferences, renders the
idea—built on the results of [4]—that ST is not CL, but LP.
For Dicher and Paoli to conclude this, it is essential that a logic shall be identified
with the logical system characterizing the valid inferences between the intended relata
of the system—which in the case of ST are sequents or inferences, insofar as it is mainly
motivated as a sequent calculus. This can be illustrated by the following quote.
Notice, however, that in a sequent calculus all of the action takes place at the
level of sequent-to-sequent rules, whereby from one or more sequents (intuitively
understood as ‘inferences’) we derive more sequents (i.e., more ‘inferences’). Which
is to say, the action takes place at the level of metainferences. [12, p. 8]
Whence, the system to be highlighted is the one characterized by the valid inferences
of this sort, i.e. the valid metainferences. This system happens to be Graham Priest’s
LP and, therefore, ST is not CL but LP.
Before we start discussing their conclusion, let us notice the following. There is no
doubt that Dicher and Paoli’s utilization of the Blok-Jónsson approach has, at least, one
advantage. For what it is worth, it allows to establish something we have an intuitive
grasp of, i.e. that there can be and that there are different presentations of the same
logic. This is indeed something we informally assume, for instance, when we talk about
various proof-systems (Hilbert-style axiomatizations, natural deduction calculi, sequent
calculi, etc.) for a given logic—implicitly admitting that the consequence relation in
question can hold between different relata (formulae, equations, sequents, etc.)
However useful this may be, we think Dicher and Paoli’s point can be legitimately
resisted. The strategy we are going to apply, for this purpose, is going to be fair
and clear. Their claims can be disputed, because their identity criterion for logics
can be. This, furthermore, is possible due to the fact that the said criterion severely
undergenerates, i.e. it does not qualify as logics many systems which a great deal of
people take to be genuinely so. In rejecting their criterion we will subsequently reject
their utilization of it and, furthermore, the idea that ST is LP.
The most important kind of logics that Dicher and Paoli’s identity criterion dismisses
are substructural logics. Why so? Precisely, because the Blok-Jónsson definition of an
abstract consequence relation is required to satisfy the Tarskian axioms. Thus, if for
something to be counted as a genuine logic it must be a class of abstract consequence
relations satisfying the Tarskian axioms, then substructural logics would be rendered as
non-existent. In fact, something similar can be said about multiple-conclusion logics.
Again, if for something to be counted as a genuine logic it must be a class of single23
conclusion abstract consequence relations satisfying the Tarskian axioms, then multipleconclusion logics would be automatically rendered as non-existent.
Interestingly, these authors may argue that a generalization of the Blok-Jónsson
abstract approach to logical consequence could help reinstate the logical status to substructural (and multiple-conclusion) systems. Such modification would only require of
an abstract consequence relation over an arbitrary collection A to be a relation without
asking for any of the Tarskian axioms to be satisfied—except, presumably, for substitution invariance. This, we think, will still give unintuitive results. In particular,
considering how sequent calculi are associated to logical systems according to Dicher
and Paoli, this will classify some systems as substructural (or multiple-conclusion) although it will not classify as such the ones that we usually take to be so.
To observe this notice that, as discussed in Section 2 the failure of e.g. Reflexivity,
Monotonicity, and Transitivity for the logic associated to a sequent calculus are usually
characterized by the failure or non-derivability of the metainferences discussed in Section 2. However, if we accept Dicher and Paoli’s idea that in the context of a sequent
calculus the action happens at the level of the sequent-to-sequent derivations—i.e. the
inferences are from collections of sequents to sequents themselves—all of the above are
not faithfully represented by the metainferences depicted above.
For starters, Reflexivity is usually understood as the property by means of which any
inference having the same object as premise and as conclusion is valid. But, if inferences
are only sequent-to-sequent derivations, then according to the abstract approach to
logical consequence as applied to sequent calculi Reflexivity is better expressed by the
following metainference—which properly speaking non-reflexive logics should invalidate.
Γ⇒∆
Γ⇒∆
Furthermore, Monotonicity is normally understood as the property by means of
which any valid inference can have its premise-set or its conclusion-set legitimately
augmented without leading to an invalid inference. Nevertheless, if inferences are only
sequent-to-sequent derivations, then according to the abstract approach as applied to
sequent calculi, then Monotonicity is better expressed by the following metainference of
level 2—which properly speaking non-monotonic logics should invalidate.
Γ1 ⇒ ∆ 1
Γ1 ⇒ ∆ 1
. . . Γn ⇒ ∆ n
Γ⇒∆
. . . Γn ⇒ ∆ n Σ ⇒ Π
Γ⇒∆
Finally, when applied to single-conclusion inferences, Cut is commonly understood
as the property by means of which if an inference leads to a certain conclusion, and
augmenting the premise-set of this inference with the said conclusion leads to a valid
inference with another conclusion, then the inference with the original premise-set and
the last conclusion was already valid. However, if inferences are only sequent-to-sequent
derivations, then according to the abstract approach as applied to sequent calculi, then
Cut (as applied to single-conclusion inferences) is better expressed by the following
metainference of level 2—which genuine non-transitive logics should invalidate.
24
Γ1 ⇒ ∆ 1
. . . Γn ⇒ ∆ n
Γ ⇒ ∆ Γ1 ⇒ ∆ 1 . . .
Γ⇒∆
Σ⇒Π
Γ1 ⇒ ∆ 1 . . . Γn ⇒ ∆ n
Σ⇒Π
Γn ⇒ ∆ n
We certainly think the previously mentioned are some seriously disappointing technical and conceptual consequences of applying a Tarskian version of the abstract approach
to logical consequence, in order to individuate logical systems.21 For what it is worth,
if this is how the story goes with the identity criterion applied by Dicher and Paoli to
conclude that ST is LP, we consider that these shortcomings are enough to disregard
it and, therefore, to discredit their identification of ST with Priest’s logic.
Concerning ST, then, we accept that part of the action takes place at the level of
sequent-to-sequent inferences, but just like a part of the expressive richness of a phenomenon would be lost if we identified a logic with the consequence relation holding
between formulae, another part would be lost if we only considered the consequence
relation holding between inferences. This is something our construction gives the opportunity to reflect upon.
6
Conclusion
In this article we proved a number of positive and negative results, and discussed certain
philosophical consequences of them. Concerning the positive results, we first generalized
the ST phenomenon, thereby obtaining a recursively defined hierarchy of strict-tolerant
systems. Secondly, we proved that the systems of the hierarchy are progressively more
classical. Thirdly, we showed that when the systems of the hierarchy stopped behaving classically, they started to behave paraconsistently in a way essentially tied to LP.
Concerning the negative results, we argued that the previously mentioned facts suggest
the impossibility of identifying CL with its set of valid inferences, or its set of valid
metainferences of level 1, or its set of valid metainferences of any arbitrary large inferential level, for that matter—since there will always be a system in the hierarchy which
will coincide with CL in this regard, although not in all inferential levels.
Drawing inspiration from the above, we proposed an alternative criterion to characterize CL, identifying it with an infinite sequence of consequence relations pertaining
inferences, metainferences, metametainferences, and so on and so forth—suggesting this
criterion should be applied to all logical systems, with full generality. It is important
to notice that the discussed hierarchy constitutes only the beginning of a number of
21
Moreover, for this approach to represent multiple-conclusion logics associated with sequent calculi,
it would not be enough to allow sequents whose succedents have greater-than-one cardinality. Instead,
it would require to have metainferences whose conclusion is not a single sequent, but a set of sequents as
below—where, it might be assumed to keep the symmetry with the ongoing discussion, such a transition
should be read as allowing to infer at least one of the conclusion sequents.
Γ1 ⇒ ∆1
Σ1 ⇒ Π 1
...
...
25
Γ n ⇒ ∆n
Σn ⇒ Π n
explorations. There is an endless amount of different, equally interesting hierarchies
that can be constructed and put to good use to motivate and clarify many philosophical
debates. We have reflected upon the impossibility of identifying CL extensionally with
its set of valid inferences of any arbitrary large inferential level, but a similar move could
also be done with other logics—thereby requiring to look at other hierarchies, yet to be
defined.
Finally, we have decided to only look at the finite levels of the hierarchy. Thus,
the problem of giving a definition of logical consequence for the system located at the
first limit ordinal of the hierarchy was beyond our scope. This, and many other issues,
represent deeply interesting topics which we hope to discuss in future work.
References
[1] E. Barrio, F. Pailos, and D. Szmuc. A recovery operator for non-transitive approaches. Review of Symbolic Logic. Forthcoming.
[2] E. Barrio, F. Pailos, and D. Szmuc. Substructural Logics, Pluralism and Collapse.
Synthese. Forthcoming.
[3] E. Barrio, F. Pailos, and D. Szmuc. What is a Paraconsistent Logic? In W. Carnielli
and J. Malinowski, editors, Between Consistency and Inconsistency, Trends in
Logic, pages 89–108. Springer, Dordrecht, 2018.
[4] E. Barrio, L. Rosenblatt, and D. Tajer. The logics of strict-tolerant logic. Journal
of Philosophical Logic, 44(5):551–571, 2015.
[5] W. Blok and B. Jónsson. Equivalence of consequence operations. Studia Logica,
83(1):91–110, 2006.
[6] E. Chemla, P. Egré, and B. Spector. Characterizing logical consequence in manyvalued logics. Journal of Logic and Computation, 27(7):2193–2226, 2017.
[7] P. Cobreros, P. Egré, D. Ripley, and R. van Rooij. Tolerance and Mixed Consequence in the S’valuationist Setting. Studia logica, 100(4):855–877, 2012.
[8] P. Cobreros, P. Egré, D. Ripley, and R. van Rooij. Tolerant, classical, strict. Journal
of Philosophical Logic, 41(2):347–385, 2012.
[9] P. Cobreros, P. Egré, D. Ripley, and R. van Rooij. Priest’s Motorbike and Tolerant
Identity. In R. Ciuni, H. Wansing, and C. Willkommen, editors, Recent Trends in
Logic, pages 75–83. Springer, Cham, 2014.
[10] P. Cobreros, P. Egré, D. Ripley, and R. van Rooij. Reaching transparent truth.
Mind, 122(488):841–866, 2014.
[11] B. Dicher and F. Paoli. The original sin of proof-theoretic semantics. Synthese.
Forthcoming.
[12] B. Dicher and F. Paoli. ST, LP, and tolerant metainferences. In C. Başkent
and T. M. Ferguson, editors, Graham Priest on Dialetheism and Paraconsistency.
Springer, Dordrecht. Forthcoming.
[13] J.-Y. Girard. Proof theory and logical complexity. Bibliopolis, Napoli, 1987.
[14] L. Humberstone. Valuational Semantics of Rule Derivability. Journal of Philosophical Logic, 25(5):451––461, 1996.
26
[15] R. Keefe. What Logical Pluralism Cannot Be. Synthese, 191(7):1375–1390, 2014.
[16] F. Pailos. A Family of Metainferential Logics. Journal of Applied Non-Classical
Logics. Forthcoming.
[17] F. Pailos. A fully classical truth theory characterized by substructural means. The
Review of Symbolic Logic. Forthcoming.
[18] G. Priest. Logic: One or many? In J. Woods and B. Brown, editors, Logical
Consequence: Rival Approaches. Proceedings of the 1999 Conference of the Society
of Exact Philosophy. Hermes, Stanmore, 2006.
[19] A. Pynko. Gentzen’s cut-free calculus versus the logic of paradox. Bulletin of the
Section of Logic, 39(1/2):35–42, 2010.
[20] S. Read. Monism: The One True Logic. In D. de Vidi and T. Kenyon, editors, A
Logical Approach to Philosophy: Essays in Memory of Graham Solomon. Springer,
Dordrecht, 2006.
[21] D. Ripley. Conservatively extending classical logic with transparent truth. Review
of Symbolic Logic, 5(02):354–378, 2012.
[22] D. Ripley. Paradoxes and failures of cut. Australasian Journal of Philosophy,
91(1):139–164, 2013.
[23] D. Ripley. Comparing substructural theories of truth. Ergo, 2(13):299–328, 2015.
[24] L. Rosenblatt. Non-contractive classical logic. Notre Dame Journal of Formal
Logic. Forthcoming.
[25] T. Williamson. Equivocation and existence. Proceedings of the Aristotelian Society,
88:109–127, 1987.
[26] S. Wintein. On all Strong Kleene generalizations of Classical Logic. Studia Logica,
104(3):503–545, 2016.
[27] J. Woods. Intertranslatability, theoretical equivalence, and perversion. Thought,
7(1):58–68, 2018.
[28] E. Zardini. Naive modus ponens. Journal of Philosophical Logic, 42(4):575–593,
2013.
27