Proof Theoretic Harmony in the
Substructural Era∗
Ole Thomassen Hjortland†
ole.hjortland@lmu.de
February 7, 2014
Draft—do not cite without permission
Abstract
According to the logical inferentialist, the meaning of a logical connective
is determined by the inference rules that govern its use. Proof theoretic
semantics attempts to make this idea precise in a proof theoretic framework, using for example natural deduction or sequent calculus rules. Since
Prior’s infamous connective tonk much of proof theoretic semantics have
been occupied with formal anti-tonk conditions which rule out ill-behaved
connectives (e.g. conservativeness, harmony). Common between them is that
inference rules only succeed in determining the meaning of a connective if the
proof theoretic conditions are fulfilled. On the traditional account, however,
such conditions are insensitive to substructural dimensions of proof theory,
e.g. the distinction between additive and multiplicative connectives. We
argue that proof theoretic semantics ought to have the resources to attribute
different meanings to substructurally distinct connectives. Subsequently we
show how to develop a notion of proof theoretic harmony that preserves
substructural distinctions from introduction to elimination rules. The substructural account of harmony can rule out cases of nonconservativeness that
previous accounts have not dealt with.
Keywords: Logical Inferentialism · Proof Theory · Proof Theoretic Semantics · Harmony · Substructural Logic · Natural Deduction · Sequent
Calculus
∗ This
research is supported by the Humboldt Foundation. I am very grateful for insightful
comments on earlier drafts from Roy Dyckhoff, Bruno Jacinto, Jon Litland, Julien Murzi, Stephen
Read, Greg Restall, Peter Schroeder-Heister, Florian Steinberger, and Luca Tranchini.
† Munich Center for Mathematical Philosophy (MCMP), Ludwig-Maximilians-Universität (LMU),
80539 München, Germany.
1
Introduction
A popular idea about the semantics of logical connectives is that their meanings
are fixed by inferential roles. In other words, the meaning of a connective is
determined exhaustively by the inference rules that govern its correct use. The
position is sometimes called logical inferentialism or, in its formal guise, proof
theoretic semantics (PTS). Both Carnap (1934) and Popper (1946) defended it
in an early form, but their versions were met by a well-known objection. Prior
(1961) rejected the thought that a connective’s meaning can be fixed by inference
rules alone, e.g. introduction and elimination rules in a natural deduction system.
He introduced the artificial connective ‘tonk’ (H) to show that it cannot be the
case that inferential rules are sufficient to determine meanings:
A
AHB
AHB
B
(HI)
(HE)
The problem is that in the presence of a transitive deducibility relation, the
rules for H lead to triviality.1 On the assumption that an inconsistent connective
cannot be meaningful, Prior invites us to conclude that basic inference rules are
not meaning-determining.
Since then a number of authors have spoken in defence of PTS. The problem,
it is argued, is not with logical inferentialism in itself, but with the idea that there
are no conditions on successful stipulation of new logical expressions. Inference
rules can successfully determine the meaning of a logical connective, but not any
old set of inference rules will do. Following Belnap’s (1962) suggestion that new
inference rules should be conservative and unique extensions of the antecedent
system, a host of proof theoretic conditions on inference rules have been proposed.
The overarching idea is not simply to rule out tonk. More generally, with the
proof theoretic condition in place, inference rules should succeed in determining
the meaning of the associated connective. Dummett (1973, 1991) labelled such
a condition ‘harmony’, alluding to the apparent lack of deductive equilibrium
between the introduction and elimination rule for H. However, formulating the
details of such a condition has proved challenging.2
In what follows we will assume that the inferentialist has got the big picture
right. When the appropriate proof theoretic condition is satisfied, the meaning
of a logical connective is fixed by its inference rules. The challenge will be to
give the exact nature of the proof theoretic condition. We follow a promising
proposal known as general elimination harmony (cf. Read 2000, 2010, 2013,
Francez & Dyckhoff 2012), but argue that the proposal does not give the
correct proof theoretic conditions for substructural connectives. In particular,
substructural connectives can lead to nonconservativeness in certain cases. We
1 Specifically, with H alone we get that anything follows from a non-empty set of premises. In the
presence of, say, a negation, we get full triviality.
2 The debate has to a large extent been part of a polemic between classical and intuitionistic
logic. For the traditional revisionist arguments, see for example Prawitz (1977), Dummett (1991),
Tennant (1987). For some recent approaches to harmony in classical logic, see Milne (1994, 2002),
Read (2000), and Rumfitt (2000)
2
then introduce a new version of general elimination harmony that can handle
a variety of substructural connectives. It can be shown that the new harmony
conditions (in conjunction with some further restrictions on introduction rules)
entail normalization and conservativeness.
Section 1 gives a brief introduction to substructural connectives in natural
deduction and sequent calculus. Sections 2-4 introduces the notion of general
elimination harmony, before we in Section 5-6 show the problems that arise when
substructural connectives are considered. Finally, in Sections 7-9 we present a
new proof theoretic harmony condition, and show how it manages substructural
connectives.
1
Substructural Logic
The traditional inferentialist has been chiefly occupied with proof theoretic
conditions for intuitionistic and classical logic. More recent work in PTS,
however, has had a different focus. PTS has now expanded to include work
in a range of different systems, both classical extensions such as modal logics,
and various non-classical logics. The systems that will be the main concern in
what follows are so-called substructural logics. Roughly put, these are logics
that result from omitting structural rules from sequent calculus systems, or,
alternatively, restricting discharge policies for assumptions in natural deduction
systems.3 In philosophical logic, such systems are already familiar in the forms of
relevant logics, paraconsistent logics, and non-monotonic logics, while others such
as linear logics and affine logics are becoming more popular. Applications range
from linguistics (e.g. Barker 2010) to truth-theoretic paradoxes (e.g. Shapiro
2010, Beall & Murzi 2013), from reasoning with vague predicates (e.g. Zardini
2008) to formal epistemology (e.g. Sequoiah-Grayson 2009).
What makes a logic substructural? In sequent calculus systems there are
two types of rules: operational rules and structural rules. The operational rules
introduce a connective on the right or the left, and correspond to introduction
and elimination rules respectively. The structural rules, on the other hand,
do not govern a particular connective. Instead they reflect inferences that are
permissible because of the properties of the deducibility relation (e.g. transitivity,
reflexivity). For example, in Gentzen’s original (intuitionistic) sequent calculus
we have the following structural rules:
(Id)
A⇒A
Γ⇒C
Γ, A ⇒ C
(K)
Γ, A, A ⇒ C
Γ, A ⇒ C
Γ ⇒ A Σ, A ⇒ C
Γ, Σ ⇒ C
(W )
Γ, A, B, Γ′ ⇒ C
Γ, B, A, Γ′ ⇒ C
(Ex)
(Cut)
In addition to the identity axiom (reflexivity), we have the structural rule of
weakening (K). Weakening reflects the fact that the deducibility relation in
3 Linear logic is the most well-known of the substructural logics. It is first introduced in Girard
(1987). Standard introductions include Restall (2000) and Paoli (2002).
3
question is monotonic. Since the antecedents and succedents are ordered sets, we
also have the structural rule of contraction (W ) (allowing us to identify copies
of the same formula) and the structural rule of exchange (Ex) (allowing us to
permute the order of premises in the antecedent).
In what follows, however, we will be interested in formulating a harmony
condition in natural deduction systems. In natural deduction there are no explicit
structural rules, but the structural properties can nevertheless be identified.4 For
example, an open assumption A at a leaf node of a derivation tree corresponds
to an instance A ⇒ A of the identity axiom. The structural rules of weakening
and contraction are a bit more tricky. They correspond to global discharge
policies for closing assumptions. Consider the standard introduction rule for the
conditional:
Γ, [A]u
..
..
B
A→B
(→I)(u)
An application of → I introduces the formula A → B from a subderivation of B
from A (together with open assumptions Γ), while simultaneously discharging the
assumption A. The structural rules manifest themselves in the policies governing
the discharge of assumptions. In standard natural deduction systems, e.g.
classical and intuitionistic logic, there are two special discharge policies: multiple
discharge and vacuous discharge. The former—corresponding to structural
contraction—says that multiple copies of an assumption A can be discharged with
a single application of a hypothetical rule (e.g. → I). The latter—corresponding
to structural weakening—says that discharging an assumption at all is optional
in an application of a hypothetical rule.
Whether or not the special discharge policies are permissible in a system
make a difference to what is derivable. In both the classical and intuitionistic
→-fragment the policies are essential to deriving for example the positive paradox
for the conditional and the law of absorption:
[A → (A → B)]2
A→B
1
[A]
B→A
A → (B → A)
(1)
[A]1
[A]1
B
(1)
A→B
(A → (A → B)) → (A → B)
(2)
In the leftmost derivation of the positive paradox we first apply → I without
discharging an assumption. In other words, it does not matter how A was
derived, we can always conclude B → A, for any B.5 In the rightmost derivation
of the law of contraction, the penultimate step is an application of → I where
4 See Troelstra and Schwichtenberg (2000) for a full translation between natural deduction and
sequent calculus.
5 Obviously, relevantists who object to the positive paradox of classical logic will therefore typically
reject vacuous discharge.
4
two copies of the assumption A are discharged, both indicated by the index 1.
Again, if we drop multiple discharge, the law of absorption is not derivable.6
Structural rules and discharge policies also make a difference for the project
of proof theoretic semantics. First, there is a general question of inferential
meaning: Does the presence or absence of structural rules affect the meaning
of a connective, or is the meaning determined by the operational rules (or
introduction and elimination rules) alone? Second, there is a question about
harmony in a substructural setting: How can we make proof theoretic harmony
sensitive to substructural distinctions? The first question will not be answered
in what follows, but there is an ongoing debate. Restall (2002) and Paoli (2003)
have argued that structural rules can change the logic (i.e. what is derivable)
without changing the meaning of the connectives, and a number of objections
are discussed in Hjortland (2013) and Paoli (2013). Instead, the focus in the
present paper will be on giving an adequate answer to the second question.
The observation that proof theoretic ‘anti-tonk’ conditions are sensitive to
structural rules is not new. In fact, when Belnap suggests conservativeness as
the right measure to prevent tonk and tonk-like connectives, he remarks that the
condition will depend on what he calls ’the antecedent context of deducibility’:
It seems to me that the key to a solution lies in observing that even
on the synthetic view, we are not defining our connectives ab initio,
but rather in terms of an antecendently given context of deducibility,
concerning which we have some definite notions. By that I mean that
before arriving at the problem of characterizing connectives, we have
already made some assumptions about the nature of deducibility.
That this is so can be seen immediately by observing Prior’s use of
the transitivity of deducibility in order to secure his ingenious result.
But if we note that we already have some assumptions about the
context of deducibility within which we are operating, it becomes
apparent that by too careless use of definitions, it is possible to create
a situation in which we are forced to say things inconsistent with
those assumptions. (Belnap 1962, 131)
There is an important lesson in Belnap’s remark. Conservativeness and other
proof theoretic conditions give different results depending on the structural
properties of the deducibility relation.7 As we move into a substructural settings,
therefore, connectives which previously lead to inconsistency might become
consistent.8 What is more, we will see examples where being insensitive to
structural rules lead to the nonconservative introduction of new connectives.
Worse, the nonconservativeness can result in inconsistency.
6 This
is true for the →-fragment but not true in general. Note that in the presence classical ¬, ∨
(with the law of excluded middle, A ∨ ¬A as an axiom), the law of contraction is derivable without
multiple discharge.
7 Some examples are explored in Restall (2007).
8 Cook (2005), for example, has explored tonk in a non-transitive context.
5
2
General Elimination Rules
There are several reasonable formulations of proof theoretic harmony in the
literature, e.g. definitional reflection in Schroeder-Heister (2003, 2007, forthcoming), and harmony-as-deductive-equilibrium in Tennant (1997, forthcoming).
In what follows, the focus will be on a proposal known as general elimination
harmony (GE harmony) as developed in Read (2000, 2010, 2013) and Francez
& Dyckhoff (2012).9 GE harmony is based on a unification of elimination rules
known as general elimination rules.10 The general elimination rules themselves
are formulated to capture Prawitz’s inversion principle. Prawitz sums it up as
follows: ‘By an application of an elimination rule one essentially restores what
had already been established if the major premiss of the application was inferred
by an application of an introduction rule’ (Prawitz 1965, 33). For example,
consider the standard rules for disjunction:
A
A∨B
(∨I,i)
B
A∨B
(∨E,ii)
[A]u
..
..
A∨B
C
C
[B]u
..
..
C (∨E)(u)
The inversion principle tells us that whatever is derivable from each of the grounds
for introducing a complex formula, is derivable directly from the complex formula.
Thus, since disjunction has two introduction rules (and therefore two possible
sufficient grounds, A and B) the elimination rule says that a formula C can be
derived from A ∨ B if there are already two independent derivations from A to
C and B to C respectively.
Formally, the inversion principle is encapsulated in the reduction steps of
normalization proofs. For the disjunction rules we have the following well-known
reduction conversions (we call the derivation after the conversion the redux ):
Π0
A
A∨B
[A]
Π1
C
C
[B]
Π2
C
Π0
B
A∨B
Π0
A
Π1
C
[A]
Π1
C
C
[B]
Π2
C
Π0
B
Π2
C
The first conversion shows that we can find a derivation Π1 of C directly from a
derivation Π0 of A given that the derivation of A was used to introduce A ∨ B,
and that C can be derived from both possible grounds A and B as assumptions.
The importance of these reduction conversions is not only that they lead
to normalization for intuitionistic logic, but that the normalization theorem in
turn entails conservativeness. The inversion principle is therefore a candidate
9 Note that Read’s theory of harmony in Read (2000) is different from the one given in Read
(2013). The earlier theory gave a formulation of harmony that produced a single elimination rule
from a set of introduction rules. We follow the latter formulation where a set of introduction rules
induces a set of elimination rules.
10 The main development of general elimination rules is in Schroeder-Heister (1984). Other attempts
at formulating unified elimination rules can be found in Kutschera (1968), Prawitz (1978), Martin-Löf
(1984).
6
for a local condition on inference rules that leads to a global condition on the
proof system—conservativeness, and hence relative consistency. Ultimately,
the aim is a local harmony condition based on the inversion principle that
entails normalization and conservativeness. The advantage of a local condition
is that individual connectives can be proved to yield conservative extensions of
an antecedent theory, merely because they satisfy a harmony condition. The
challenge is to provide a formal constraint on inference rules which encapsulates
the inversion principle and guarantees the existence of reduction conversions.
That is precisely the motivation for general elimination rules.
The inversion principle can be used as a guideline to formulate general
elimination rules for other connectives than disjunction. For example, the
standard introduction rule for conjunction gives two general elimination rules:
A B
A∧B
(∧I)
A∧B
C
[A]u
..
..
C (∧GE,i)
A∧B
C
[B]u
..
..
C (∧GE,ii)
The duality between disjunction and conjunction now becomes salient. The
introduction rule has two grounds A and B that are jointly sufficient, so a
formula C can be derived from A ∧ B only if it can be derived either from A or
from C. Accordingly, there will be two reduction conversions:
Π0 Π1
A B
A∧B
C
[A]
Π2
C
Π0 Π1
A B
A∧B
C
Π0
A
Π2
C
[B]
Π2
C
Π1
B
Π2
C
Unlike for disjunction, the general elimination rules for conjunction are not the
same as the standard elimination rules. However, it is easy to see that the
general elimination rules are equivalent to the standard rules.
3
Higher-Order Rules
If the grounds for introducing a complex formula is hypothetical, i.e. discharges
an assumption, then the inversion principle becomes slightly more complicated.
Consider the standard introduction rule for the intuitionistic conditional:
[A]u
..
..
B
A→B
(→I)(u)
The inversion principle tells us that the general elimination rule should allow us
to derive a formula C from A → B only if there is a derivation of C from the
grounds for introducing A → B. But the ground specified by → I is a derivation,
not a formula. The result is a general elimination rule that includes a rule as
7
assumption. We therefore say that the general elimination rule is a higher-order
rule:11
u
A
..
..
B
..
..
A→B
C (→GE)(u)
C
The rule → GE discharges an assumed rule from A to B and concludes C directly
from A → B, if C is derivable from the higher-order assumption. It is in other
words an application of the inversion principle in which the ground is a rule
rather than a formula. There is now a corresponding reduction conversion:
[A]u
Π0
B
A→B
(u)
C
Π2 v
A
.
...
B
Π1
C (v)
Π2
A
Π0
B
Π1
C
The subderivation Π2 yields A as conclusion (a special case is where A is an
assumption), hence giving a derivation of B on the (higher-order) assumption
that there is a rule that gives B directly from A. From B in turn there is a
derivation Π1 to C, and we subsequently conclude C directly from A → B and
discharge the higher-order assumption.12 Note that A is not discharged by the
application of → GE. But that is as we should expect, as A is a minor premise
in the standard modus ponens rule:
A→B
B
A
(→E)
The standard rule → E is equivalent to the higher-order rule → GE.
11 The
higher-order rules are first developed in Schroeder-Heister (1984).
(1987) suggested an alternative general elimination rule for the intuitionistic conditional,
but without the higher-order assumption:
12 Dyckhoff
A→B
A
C
Γ, [A]u
.
.
.
.
C
The rule was subsequently studied by Tennant (2002) and von Plato (2001). Olkhovikov and
Schroeder-Heister (forthcoming) have later shown that the higher-order assumptions are necessary
in order to proof-theoretically represent all intuitionistic connectives. We therefore stick with
higher-order rules throughout the presentation.
8
4
General Elimination Harmony
The inversion principle and the general elimination rules have been used to
formulate a harmony condition—so-called GE harmony. The idea is to let an
arbitrary (finite) set of introduction rules, I, induce a set of general elimination
rules, E. This approach has at least two advantages: First, it is a local condition
on inference rules, rather than a global condition on a proof system. Global
conditions such as conservativeness and consistency should in turn follow from
the local conditions when systematically applied to the proof rules. Second, GE
harmony provides a method for constructing the set of elimination rules directly
from the introduction rules. In what follows, we use the formulation in Read
(2013).
Suppose the n-place connective ⊙(β1 , ..., βn ) has m ≥ 0 introduction rules,
⊙I1 , ..., ⊙Im , each 1 ≤ i ≤ m of which has ki ≥ 0 premises, ρi1 , ..., ρiki . That is,
I⊙ =
ρ , ..., ρ
ρm1 , ..., ρmkm
11
1k1
(⊙I1 )
(⊙Im )
⊙β~
...
⊙β~
We write ⊙β~ for ⊙(β1 , ..., βn ). Each premise ρij may be either categorical,
i.e. a formula, or hypothetical, i.e. a subderivation discharging (first-order)
assumptions.13
Qm
The set of introduction rules I ⊙ induces a set with i=1 ki general elimination
⊙
⊙
⊙
rules E (I
GE E ) via the following schema:
~
⊙β
[ρ11 ]u
..
..
γ
...
γ
[ρm1 ]u
..
..
γ
(⊙GE)
⊙β~
...
[ρ1k1 ]u
..
..
γ
...
γ
[ρmkm ]u
..
..
γ
(⊙GE)
There is one general elimination rule, ⊙GE for every m-tuple of premises ρij
(1 ≤ i ≤ m, 1 ≤ j ≤ ki ). Every elimination rule will therefore have, for each
introduction rule 1 ≤ i ≤ m, exactly one minor premise with an assumption
corresponding to a premise in the ith introduction rule. The only other premise
~
is the major premise ⊙β.
We can then define GE harmony as follows:
Definition 4.1 (GE Harmony). A set of elimination rules E ⊙ are in GE harmony
with a set of introduction rules I ⊙ if I ⊙ GE E ⊙ . If a connective ⊙ has rule
sets that are in GE harmony we will say that the connective is GE harmonious.
That Def 4.1 is in accordance with the inversion principle can then be seen
by producing a schematic reduction conversion. Consider a derivation with an
application of the ith introduction rule, and then an immediately succeeding
application of an elimination rule:
13 There are, in other words, no higher-order assumptions in introduction rules. If that was allowed
the general elimination rules would take more complicated assumptions. We will bracket that option
for now.
9
(Πi1 )
ρi1
(Πiki )
ρiki
...
⊙β~
[ρ1j1 ]u
Π∗1j1
γ
γ
...
[ρmjm ]u
Π∗mjm
γ
(u)
(Πiji )
ρiji
Π∗iji
γ
(Π)
The notation ρ indicates that a premise in an introduction rule can either be
a formula A which labels the end node of the subderivation Π (e.g. in ∧I), or
a subderivation Π from some set of assumptions Γ to a conclusion B (e.g. in
→ I). We will follow Pfenning and Davies (2001) and call the existence of such a
reduction conversion local soundness. It is not in general the case, however, that
local soundness entails normalization. Whether or not the reduction conversions
lead to a normal form proof (in the presence of other permutation conversions)
depends crucially on the relationship between the formula β~ and the formula
occurring in the premises ρij . Note that the introduction rules for standard
intuitionistic connectives are such that every formula occurring in a ρij is a strict
~ We will return to this issue in Section 9 below.
subformula of β.
One desirable consequence of the definition of GE-harmony is that H is
disharmonious:
Fact 4.1. The introduction and elimination rules for tonk H are not in GE
harmony.
The introduction rule H harmoniously induces the following general elimination
rule:
A
AHB
AHB
C
GE
[A]u
..
..
C (u)
The GE harmonious elimination rule for H can be seen to normalize, and
is therefore not equivalent to the standard elimination rule (which leads to
trivialization).
5
Substructural Connectives
Let us now turn to substructural connectives. The question we want to address is
whether GE harmonious connectives in the presence of substructural deducibility
relations produce the appropriate reduction conversions. It turns out that they
do not, unless we make critical modifications to the definition.
Consider again the above GE rules for conjunction ∧:
A∧B
C
[A]u
..
..
C (∧GE,i)
A∧B
C
10
[B]u
..
..
C (∧GE,ii)
These rules are closely related to rules for conjunction in intuitionistic sequent
calculus:
Γ, A ⇒ C
Γ, A ∧ B ⇒ C
Γ, B ⇒ C
Γ, A ∧ B ⇒ C
(L∧,i)
(L∧,ii)
Informally, we can read both sets of rules as saying that if something can be
derived from the assumption of A (B), then it can be derived directly from
A ∧ B.
The above sequent rules for conjunction are known as additive rules. The
corresponding additive right-rule is the context-sharing version of conjunction
introduction:
Γ⇒A Γ⇒B
Γ⇒A∧B
(R∧)
In contrast, the multiplicative or context-free conjunction, ⊗, has the following
sequent rules:
Γ, A, B ⇒ C
Γ, A ⊗ B ⇒ C
Γ⇒A Σ⇒B
Γ, Σ ⇒ A ⊗ B
(L⊗)
(R⊗)
In classical and intuitionistic sequent calculus, the additive and multiplicative
conjunction rules are equivalent. However, ∧ and ⊗ come apart in logics where
either of the structural rules of weakening and contraction are absent, e.g.
relevant logics, affine logics, linear logics. In substructural logics, therefore,
there are two conjunctions: one additive and one multiplicative. In fact, other
connectives can also be bifurcated into two non-equivalent connectives.
We should expect there to be an analogous distinction between additive and
multiplicative conjunctions in natural deduction. It is not difficult to find a
candidate for the multiplicative elimination rule:14
A∧B
C
[A, B]u
..
..
C (⊗GE)(u)
Strictly speaking, of course, the multiplicative elimination is not in GE harmony
with the introduction rule. Def. 4.1 above requires the introduction rule to have
two distinct elimination rules. Nevertheless, the multiplicative rule in natural
deduction can also be used in a reduction conversions, but one which differs
from that of ∧ in a crucial respect:
[A, B]u
| {z }
Π0 Π1
A B
Π2
A⊗B
C (u)
C
Π0
A
|
Π1
B
{z
Π2
C
}
14 The rule can be found both in Martin-Löf (1984) and Read (2000), but not in a substructural
context.
11
Note that the redux now preserves both a subderivation ending with A and a
subderivation ending with B. That will be important in what follows.
So far only the elimination rules in the natural deduction framework are
different for ∧ and ⊗. But following Negri (2002) we can give natural deduction
introduction rules corresponding to both the additive and multiplicative sequent
calculus rules. In order to do that we need to introduce auxiliary formulae
(contexts) explicitly in the rule schemata. We will write the multiplicative ⊗I
introduction rule as follows:
Γ. Σ.
..
..
.
.
A B
A⊗B
(⊗I)
The contexts Γ, Σ are sets of (open) assumptions labelling the leaf nodes of
the subderivations terminating with A and B respectively. It should then be
clear that the standard introduction rule in both classical and intuitionistic
natural deduction corresponds to the multiplicative right rule for ⊗ in sequent
calculus. There is no restriction on the open assumptions used to derive A and
B respectively.
In the additive introduction rule, however, we will see precisely that the open
assumptions come with a decisive restriction. The additive counterpart of the
introduction rule for ∧ is as follows:
Γ.α Γ.α
..
..
.
.
A B
A∧B
A∧B
C
Γ, [A]u
..
..
C
A∧B
C
Γ, [B]u
..
..
C
In the additive rule the subderivations of A, B is adorned by the same contexts Γ
in both subderivations. That indicates that the members of the assumption class
Γ must be the same in each subderivation. Furthermore, the index α requires that
each of the assumption classes must have members with the same assumption
index. Only then can the additive conjunction rule be applied. (Recall that in
sequent calculus an application of ∧R contracts the auxiliary formulae of the
two premise sequents.) As a result, whenever a formula A in one copy of Γ is
discharged in the derivation, a copy of A in the other occurrence of Γ is also
discharged.
The result is that in the presence of vacuous and multiple discharge policies,
e.g. in intuitionistic logic, the natural deduction rules for the two conjunctions ∧
and ⊗ are equivalent. Assume the rules for ⊗, and suppose there is a derivation
both from Γ to A and from Γ to B. Then we can apply ⊗I (in the special case
where Γ = Σ):
Γ. Γ.
..
..
.
.
A B
A⊗B
12
Whenever a succeeding rule application discharges an assumption in either
the left-side Γ, multiple discharge permits the closing of the corresponding
assumption in the right-side Γ. For the other direction: Suppose there is a
derivation of A from Γ and of B from Σ. Then we substitute these derivations
with derivations of A from Γ ∪ Σ and B from Γ ∪ Σ, and apply ∧I (indexing
both copies of Γ ∪ Σ with a label α).15
For the elimination rules the role of the discharge policies are more prominent.
First assume the ∧ rules, and suppose that there is a derivation of A ∧ B and
from A, B to C. Two successive applications of ∧GE then discharges both
assumptions in the minor premise:
A∧B
A∧B
C
C
[A]1 , [B]2
..
..
C (1)
(2)
Multiple discharge now permits us to discharge both copies of A ∧ B simultaneously later in the derivation. In the other direction an application of ⊗E with
vacuous discharge is sufficient to derive both ∧GE rules.
6
Substructural Insensitivity in GE Harmony
Def. 4.1 of GE harmony does not include any explicit information about contexts. However, the standard introduction rule for conjunction in classical and
intuitionistic logic is the multiplicative one. That should be clear as there is no
restriction on the open assumptions required to derive the two premises. Thus
the introduction rule is the multiplicative ⊗I rule. If we take this rule as our
input, however, the definition of GE harmony induces the additive elimination
rules for ∧. That is an unintended effect, and although innocuous in standard
natural deduction, it leads to trouble in substructural settings. That is not
to say that matching up additive and multiplicative rules necessarily have an
adverse effect on inferential meanings. Unlike with the rules of H, for example,
there is no reason to claim that mixing additive and multiplicative rules fail to
determine the inferential meaning of the connective in question. But, if one is
interested in substructural logics, it is important to notice that mixing the rules
can lead to structural nonconservativeness. If conservativeness is a requirement
on successful implicit definition of a logical connective, then GE harmony cannot
be correct as it stands. We consider the cases of weakening and contraction in
turn.
15 The derivations Γ ∪ Σ can be constructed by weakening in additional premises via ∧GE with
vacuous discharge.
13
6.1
Weakening
To see why nonconservativeness can follow in substructural settings, consider
again the multiplicative sequent rules for ⊗.16 When an application of the cut
rule is principal for the ⊗ conjunction we have the following reduction conversion
which produces a derivation with two applications of cut on A and B:
Γ ⇒ A Σ ⇒ B Θ, A, B ⇒ C
Γ, Σ ⇒ A ⊗ B Θ, A ⊗ B ⇒ C
Γ, Σ, Θ ⇒ C
Σ⇒B
Γ ⇒ A Θ, A, B ⇒ C
Γ, Θ, B ⇒ C
Γ, Σ, Θ ⇒ C
In the conversion step for the additive ∧, on the other hand, only one cut
application is required:
Θ, A ⇒ C
Γ⇒A Γ⇒B
Γ⇒A∧B
Θ, A ∧ B ⇒ C
Γ, Θ ⇒ C
Γ ⇒ A Θ, A ⇒ C
Γ, Θ ⇒ C
Notice that both the above reductions in sequent calculus produce the exact same
sequent as the original derivation. In contrast, consider the case for a connective
⊓ which has a multiplicative introduction rule and two additive elimination rules:
Γ, A ⇒ C
Γ, A ⊓ B ⇒ C
(L⊓)(i)
Γ, B ⇒ C
Γ, A ⊓ B ⇒ C
(L⊓)(ii)
Γ⇒A Σ⇒B
Γ, Σ ⇒ A ⊓ B
(R⊓)
The appropriate conversion in the cut elimination strategy will now be as follows
(the case for B is similar):
Θ, A ⇒ C
Γ⇒A Σ⇒B
Γ, Σ ⇒ A ⊓ B
Θ, A ⊓ B ⇒ C
Γ, Σ, Θ ⇒ C
Γ ⇒ A Θ, A ⇒ C
Γ, Θ .⇒ C
.. (K)
.
Γ, Σ, Θ ⇒ C
Since the sequent Σ ⇒ B (alternatively, Γ ⇒ A) does not occur at all in the
redux, the derivation requires a number of applications of weakening to add back
the auxiliary formulae in Σ (Γ). The result, therefore, is that the conversion only
works in the presence of weakening. Put differently, the redux proves something
that could not be obtained with the original derivation. Ordinarily this is not
a problem, but in a logic without weakening, say, a relevant logic, it leads to
nonconservativeness. To see this, simply start with the redux derivation, and
subsequently use the added ⊓ rules to expand to the left-most derivation. We
can assume that the redux derivation does not include the connective ⊓ at all,
and that it is therefore a derivation in the antecedent language. The expansion
to the right-most derivation now allows us to weaken in additional premises in
the antecedent, but without explicit use of a weakening rule. In other words,
adding the mixed conjunction together with its left- and right-rule has allowed
16 See
also Troelstra (1992) for a discussion of nonconservativeness.
14
us to derive a weakened sequent. In a system without weakening this gives us
structural nonconservativeness.
It is instructive to compare the situation in natural deduction. The connective
⊓ has the following (multiplicative) introduction and (additive) elimination rules:
Γ. Σ.
..
..
.
.
A B
A⊓B
(⊓I)
A⊓B
C
Γ, [A]u
..
..
C (⊓E)(i)
A⊓B
C
Γ, [B]u
..
..
C (⊓E)(ii)
As pointed out above these are in fact the GE harmonious rules for ∧ in
both intuitionistic and classical logic. Nevertheless, the problem of structural
nonconservativeness is the same as in sequent calculus. The reduction conversion
for normalization has the same feature as the sequent calculus counterpart:
Γ
Σ Θ, [A]u
Π0 Π1 | {z }
A B
Π2
A⊓B
C (u)
C
Γ
Π0
A Θ
| {z }
Π2
C
If we compare with the reduction conversion of the multiplicative ⊗ (in Section
5) we see that the subderivation Π1 from Σ to B now becomes redundant in the
redux. Instead the derivation of C proceeds from the open assumptions in Γ ∪ Θ
alone. In order to produce a derivation of C from Γ ∪ Θ ∪ Σ, we have to weaken
the open premises in Σ back in.
More directly, we can now show that weakening has become admissible.
If we assume that C is already derivable from A, then we can use the above
rules of ⊓ to produce a derivation of C from A,B. As was the case in sequent
calculus, since the formulae can be from the ⊓-free fragment of the language,
nonconservativeness follows.
A B
A ⊓ B [A]1
A
..
..
C
(⊓E)(1)
That is, C is derivable from A, B, and there is no need for vacuous discharge. Note
the contrast to the multiplicative conjunction elimination rule, an application of
which would have required vacuous discharge.17
17 Note that the derivation is not possible if the introduction rule is additive. In that case open
assumptions in the derivations of A and B must be the same. In the derivation, however, the
application of the introduction rule has the premises A and B themselves as open assumptions,
violating the context-sharing constraint of the additive rule. Correspondingly, in sequent calculus,
the additive ∧R rule cannot be applied when the premise-sequents are the axioms A ⇒ A and
B ⇒ B respectively.
15
6.2
Contraction
A similar structural phenomenon arises when we consider a connective f which
combines an additive introduction rule with a multiplicative elimination rule.
First, in sequent calculus we have the following left- and right-rules:
Γ, A, B ⇒ C
Γ, A f B ⇒ C
(Lf)(i)
Γ⇒A Γ⇒B
Γ⇒AfB
(Rf)
The conversion step for an application of cut turns out as follows:
Γ ⇒ A Σ, A, B ⇒ C
Σ, Γ, B ⇒ C
Γ⇒B
Γ, Γ, Σ. ⇒ C
.. (W )
.
Γ, Σ ⇒ C
Γ ⇒ A Γ ⇒ B Σ, A, B ⇒ C
Γ⇒AfB
Σ, A f B ⇒ C
Γ, Σ ⇒ C
Now it is the structural rule of contraction that is required to produce a derivation
of the same sequent in the redux. In other words, without structural contraction
the derived sequent is too weak. If the connective f is added to an antecedent
theory without contraction, therefore, its introduction and elimination rules can
be applied to contract on the left.
As in the case of weakening, there is an analogous result in natural deduction.
The connective f has the additive introduction rule and the multiplicative
elimination rule:
Γ.α Γ.α
..
..
.
.
A B
AfB
(fI)
AfB
C
Γ, [A, B]u
..
..
C (fE)(u)
The reduction conversion then has the following shape:
Γα Γα Θ, [A, B]u
Π0 Π1 | {z }
A B
Π2
AfB
C (u)
C
Θ
|
Γ
Γ
Π0 Π1
A B
{z
}
Π2
C
The redux derives C from Θ and two copies Γ. But unlike the original derivation
it does not identify the two copies of Γ through the application of fI (and
index α). Instead the redux derivation requires multiple discharge in order to
derive what is already derivable without it in the original derivation. In other
words, the detour through the additive conjunction rule introduces an element
of contraction, even in a context where multiple discharge is not permissible.
What we have, therefore, is a failure of local soundness: In a contraction-free
context there is no reduction conversion of the original derivation (or even of
anything stronger, as in the case of weakening).
16
As opposed to ⊓, which made weakening admissible, the mixed connective f
makes structural contraction admissible. Consider the following derivation:18
2
2
A A
AfA
C
Γ, [A, A]1
..
..
C (1)
The derivation takes advantage of the fact that the additive introduction rule
for f allows us to treat two copies of A as identical, and that the elimination
rule can discharge two copies of A in the special case where the major premise
is A f A. As a consequence, any formula C derivable from two copies of A is
now derivable from one copy. The structural nonconservativness has given us
back multiple discharge.
The reintroduction of contraction through nonconservativeness can be particularly problematic. There are philosophically interesting formal theories that
cannot include contraction on pain of contradiction. Unrestricted rules for a
formal truth predicate and unrestricted set theoretic comprehension are examples of principles that are famously susceptible to paradox in classical logic.19
Both truth theoretic and set theoretic systems have therefore been developed
in contraction-free logics (e.g. Zardini 2011, Petersen 2000). In such systems
structural nonconservativeness leads to triviality, and so control over extensions
in order to include new connectives (and new inference rules) is particularly
critical.
7
Additive GE Harmony
What do the examples tell us about GE harmony? GE harmony, as previously
defined, is insensitive to substructural distinctions. As it stands the proposal
induces additive elimination rules for conjunction from multiplicative introduction
rules. The problem generalizes to other connectives as well: nonconservativeness
can result from adding mixed—but GE harmonious—connectives to systems
without weakening or contraction. In a substructural context, therefore, local
soundness is not entailed by GE harmony. Local soundness only follows in the
presence of special (structural) discharge policies, i.e. vacuous and multiple
discharge.
18 The premises Aα in the application of fI are assumptions. Since an assumption of A is a
special case of a derivation of A, we have derived A from A. This permits us to index the two copies
of A with α.
19 Natural deduction rules for a unrestricted truth predicate and unrestricted comprehension:
A
T (pAq)
Φ(t)
t ∈ λxΦ(x)
T (pAq)
A
(T I)
(T E)
t ∈ λxΦ(x)
(λE)
Φ(t)
17
(λE)
Even if it does not follow that there is anything semantically wrong with mixed
connectives, we have a motivation to give a modified definition of GE harmony
that also delivers local soundness for substructural systems. The strategy will be
a bifurcation of GE harmony according to the additive-multiplicative divide. Let
us first consider the case for additive rules. The set of additive introduction rules,
⊙
, is specified in a manner similar to the previous definition of GE harmony:
IA
α
G.α
G.
G.α
G.α
..
..
..
..
.
.
.
.
ρ
.
.
.
ρ
.
.
.
ρ
ρ
m1
1k1
mkm
11
(⊙I1 )
(⊙Im )
~
~
⊙β
...
⊙β
Γ.
G.
..
..
.
.
The notation ρ is slightly abusive, as it can be instantiated both as A, in the
Γ, [A]
..
..
case where the premise is categorical, or as B , in the case where the premise
is hypothetical. In both cases the members of G are the open assumptions
labelling leaf nodes of the subderivation. (In the categorical case where A is an
assumption, Γ = {A}.)
⊙
The GE harmoniously induced set of additive general elimination rules, EA
,
will now have minor premises adorned with (shared) contexts G:
G α , [ρ1j1 ]u
G α , [ρmjm ]u
..
..
..
..
~
γ
...
γ
⊙β
(⊙GE)(u)
γ
⊙
⊙
⊙
We write IA
are induced
GE EA to indicate that the elimination rules E
⊙
from the introduction rules I via the above templates. We can then give a
modified version of GE harmony:
⊙
are in
Definition 7.1 (Additive GE Harmony). A set of elimination rules EA
⊙
⊙
⊙
additive GE harmony with a set of introduction rules IA if IA GE EA
. If a
connective ⊙ has rule sets that are in additive GE harmony we will say that the
connective is additively harmonious.
Since there is a reduction conversion for rules satisfying the condition in Def.
7.1, additive GE harmony entails local soundness.
Gα
(Πi1 )
ρi1
...
⊙β~
Gα
(Πiki )
ρiki
β
H , [ρ1j1 ]
Π∗1j1
γ
γ
u
β
...
H , [ρmjm ]
Π∗mjm
γ
(u)
u
G
Πiji
H ρiji
| {z }
Π∗iji
γ
The above conversion is different from the conversions of the mixed connectives,
and therefore also different from those given by the pre-substructural Def. 4.1 of
18
GE harmony. The difference is that the redux is a derivation of γ from the very
same open assumptions as the original derivation. Hence, there is no need for
structural rules (special discharge policies) in order to either weaken or contract.
What we have, therefore, is a definition of GE harmony for additive connectives
that is sensitive to the absence of weakening and contraction.
8
Multiplicative GE Harmony
There is also a multiplicative definition of GE harmony, but one that is restricted
in the natural deduction framework. Because natural deduction, unlike sequent
calculus, typically is a single conclusion framework, we cannot have the full
range of multiplicative connectives. What we can have, however, is a definition
of GE harmony for all multiplicative connectives, ⊙, with only one introduction
⊙
rule in IM
. For example, the conjunction and the biconditional can be given
multiplicative natural deduction rules. In contrast, the multiplicative disjunction
rules would require multiple conclusion (or at least a more general deducibility
relation). Multiple conclusion in natural deduction is explored in a number
of places, e.g. Boričić (1985), Ungar (1992) and Read (2000). Precisely how
multiple conclusion natural deduction can be used to include the full range of
multiplicative connectives will not be explored here. Instead we will give only a
restricted form of multiplicative GE harmony.
Thus, let the set of multiplicative (context-free) introduction rules be simply
⊙
= {⊙I}. Then we have the following schema:
IM
G.1
..
.
ρ1
...
⊙β~
G.k
..
.
ρk
(⊙I1 )
The first index now becomes redundant since there is only one introduction rule,
but as opposed to the additive rules, the auxiliary formula in each premise can
now have distinct sets of open assumptions.
⊙
The single introduction rule in IM
induces a single general elimination rule
⊙
⊙
⊙
in EM (IM GE EM ):
⊙β~
G, [ρm1 , ..., ρmkm ]u
..
..
γ
(⊙GE)
γ
The distinct feature of the multiplicative elimination rule is that its single minor
premise takes every premise of the introduction rule as an assumption. These
are jointly discharged by the application of the elimination rule, as seen, for
example, in the multiplicative elimination rule for ⊗ above.
Definition 8.1 (Restricted Multiplicative GE Harmony). A set of elimination
⊙
rules EM
are in multiplicative GE harmony with a singleton set of introduction
19
⊙
⊙
⊙
rules IM
if IM
GE EM . If a connective ⊙ has rule sets that are in additive
GE harmony we will say that the connective is multiplicatively harmonious.
⊙
It follows immediately from the schemata that the set EM
will also be a singleton
set.
We can then proceed to give a reduction conversion for the multiplicatively
harmonious connectives:
G1
Π1
ρ1
...
⊙β~
Gk
Πk
ρk
G
γ
|
[ρ1 , ...ρk ]u
{z
}
Π
γ
G
|
(u)
Gk
G1
Π1
Πk
ρ 1 . . . ρk
{z
}
Π
γ
Again, we can now see that local soundness has been restored. The redux
derivation requires exactly the same open assumptions as the original derivation.
Let us consider an example slightly more complicated than the multiplicative
conjunction. The following is a context-free introduction rule for a multiplicative
4-ary connective, ։, together with its single general elimination rule:
u
A
.. C..
..
..
Θ
B D
|
{z
}
Γ, [A]u Σ, [C]u
..
..
..
..
..
..
E
։ (A, B, C, D)
D (u)
B
(u)
։ (A, B, C, D)
E
We can then find the following reduction conversion for the new connective:
Γ, [A]u Σ, [C]u
Π1
Π0
D
B
։ (A, B, C, D)
E
Π3
A
Θ ...
.
B
|
{z
Π2
E
Π4
C.
..
.
D
Π3
A
| {z }
Π0
B
{z
Π2
E
Γ
}
Θ
|
Π4
C
| {z }
Π1
D
Σ
}
The restriction on multiplicative harmony raises the following question: Why
develop notions of harmony in a natural deduction setting when it leads to
artificial constraints on the connectives definable? The answer, we believe, is
simply that sequent calculus is the preferable framework for proof theoretic
harmony and PTS in general. However, there is a lot of resistance towards
sequent calculus and, in particular, multiple conclusion relations in the PTS
literature (cf. Dummett 1991, Rumfitt 2008, Steinberger 2011). Although we do
not share these views, the guiding idea in the present paper has been to explore
harmony for substructural connectives in a natural deduction framework. That
20
has the advantage that it is a more direct continuation of the debates about
harmony in the literature, but the disadvantage that the expressive power is
somewhat curtailed.
9
Normalization and Consistency
It was pointed out in section §4 that the original Def. 4.1 of GE harmony does not
entail normalization. The existence of a reduction conversion, i.e. local soundness,
is guaranteed, but that alone is not sufficient for normalization. Read (2000)
gives a counterexample, a harmonious connective that leads to inconsistency
(and therefore nonconservativeness). Consider the following connective, •, read
‘bullet’:20
" • #u
..
..
u
⊥
[•]
..
..
..
..
•
C
⊥ (•I)(u)
(•GE)(u)
•
C
The •I rule harmoniously induces the •GE rule. Moreover, there exists a
reduction conversion for the • rules as prescribed by the above schema, but it
does not lead to normalization:
[•]u
Π0
⊥ (u)
•
C
Π
" • 2 #v
..
..
⊥
Π1
C (v)
Π2
•
Π0
⊥
Π1
C
The redux potentially produces a new maximum segment of the same complexity
as the original derivation. In the right-most derivation the copy of • could have
been introduced by an application of •I immediately followed by an application
of •GE. Worse, the rules for • lead to contradiction (assuming anything is
derivable from ⊥). Thus • is harmonious, yet its rules cannot be normalized
and it is nonconservative over any consistent proof system.21 Moreover, the
substructural definition of additive GE harmony Def. 7.1 does not prevent this
problem, so another condition is required for normalization and conservativeness
to follow in general.
A flat-footed condition that would do the trick is the following: any formula
occurring in the premises ρij of any introduction rule ⊙I must be a strict
20 For an earlier proof-theoretic study of normalization in the context of paradoxical expressions,
see Tennant (1982).
21 See Read (2000) for details on how triviality follows from the rules for •. The derivation that
leads to triviality is precisely an example of a derivation that cannot be put into normal form through
reduction conversions.
21
~ It follows immediately that every ρij occurring
subformula of the conclusion β.
in an elimination rule ⊙E (i.e. the assumptions in the minor premises) will also
be strict subformulae of the major premise. Obviously, purity is not satisfied
by •I, but it is satisfied by the other GE harmonious rules we have considered.
The advantage is simple. GE harmony and purity will jointly be sufficient for
normalization, the subformula property, and conservativeness. That is already
an achievement.
Nevertheless, it is not necessarily a desirable condition. First, note that
although it is sufficient for normalization of GE harmonious rules, the purity
condition is not a necessary condition for normalization. There are, in other
words, connectives whose rules normalize despite failing to satisfy the purity
condition. One example is the following rules for intuitionistic negation:
[A]u
..
..
B
[A]u
..
..
¬B
¬A
(u)
A ¬A
B
The left-most rule includes a formula B in both subderivations that can be of
any complexity, and is therefore not pure. Nevertheless, the rules can be shown
to normalize. Hence, imposing purity as an additional constraint on introduction
rules excludes some perfectly well-behaved connectives.
A more ambitious approach is to aim for an exact (necessary and sufficient)
condition on introduction and elimination rules that lead to normalization.
That is still an open problem. A conjecture is that there is at least a weaker
purity condition that entails normalization. Let us call it negative purity: every
introduction rule in I⊙ should include only premises ρij in which every formula
~ A
occurring in negative position is a strict subformula of the conclusion β.
formula A is in negative position in a premise ρij just in case it occurs as an
assumption. It should be immediately clear that •I fails the negative purity
condition, whereas the intuitionistic negation introduction rule satisfies it.
As mentioned above, the move to substructural (additive and multiplicative)
modifications of GE harmony will not change the need for a further condition to
ensure normalization. But it is interesting to note that substructural systems
do allow for normalization of rules that are not normalizable in proof systems
with weakening and contraction. The introduction and elimination rules for
• is one example, albeit somewhat artificial. Schroeder-Heister (1992) shows
that cut-elimination holds for contraction-free systems even in the presence
of a connective like •. In the absence of multiple discharge of assumptions,
normalization can be proved by an induction on the size of the derivation tree,
rather than on the complexity of maximum formulae. That this is not possible
when multiple discharge is allowed as can be seen from the following example:
22
u
u
[•] , ..., [•]
{z
}
|
Π0
⊥ (u)
•
C
Π
" • 2 #v
..
..
⊥
Π1
C (v)
Π2
•
|
...
{z
Π0
⊥
Π1
C
Π2
•
}
Since the left-most derivation requires a number of copies • to derive ⊥ via
Π0 , the same number of subderivations Π2 to • are required for the redux
derivation. The subderivation Π2 can be of any length, and therefore rules
out an induction on the size of the derivation. In general, the proof theoretic
conditions for normalization and consistency are significantly weaker in the
absence of contraction. Proofs of cut elimination already exists for contractionfree theories of unrestricted set comprehension and unrestricted truth predicates
(cf. Petersen 2000, Zardini 2011). These and similar results makes the demand
for structural conservativeness all the more pressing. If we were to introduce
a mixed additive and multiplicative connective into such a system, the result
might be nonconservativeness. In particular, if a connective makes contraction
admissible, the entire theory will become inconsistent. Substructural sensitivity
is therefore critical. The modified account of GE harmony is at least a partial
solution.
Conclusion
Proof theoretic harmony was conceived as a formal condition that excludes
trivializing connectives such as tonk. More generally, harmony conditions were
meant to ensure conservativeness once a new connective is introduced into an
antecedent theory. GE harmony is a good candidate for such a formal condition,
with the caveat that some form of purity condition is also required. But we
have seen that GE harmony in its original form can yield nonconservativeness in
substructural systems. The problem has been remedied by introducing a bifurcation between additive and multiplicative harmony. The modified substructural
account of harmony guarantees reduction conversions that leave the set of open
assumptions unaltered.
23
References
Barker, C.: 2010, Free choice permission as resource-sensitive reasoning, Semantics & Pragmatics
3(10), 1–38.
Beall, J. and Murzi, J.: 2013, Two flavors of curry paradox, Journal of Philosophy CX(3), 143–165.
Belnap, N.: 1962, Tonk, plonk and plink, Analysis 22, 30–34.
Boričić, B.: 1985, On sequence-conclusion natural deduction systems, Journal of Philosophical Logic
14(4), 359–377.
Carnap, R.: 1934, Logische Syntax der Sprache, J. Springer, Vienna.
Cook, R.: 2005, What’s wrong with tonk(?), Journal of Philosophical Logic 34, 217–226.
Dummett, M.: 1973, Frege: Philosophy of Language, Duckworth, London.
Dummett, M.: 1991, The logical basis of metaphysics, Harvard University Press, Cambridge, MA.
Dyckhoff, R.: 1987, Implementing a simple proof assistant, Workshop on Programming for Logic
Teaching, Leeds, July.
Francez, N. and Dyckhoff, R.: 2012, A note on harmony, Journal of philosophical logic 41(3), 613–628.
Girard, J.-Y.: 1987, Linear logic, Theoretical Computer Science 50, 1–102.
Hjortland, O. T.: 2013, Verbal disputes in logic: Against minimalism for logical connectives, Logique
et Analyse (forthcoming).
Kutschera, F.: 1968, Die vollständigkeit des operatorensystems {¬,ˆ, v,} für die intuitionistische
aussagenlogik im rahmen der gentzensemantik, Archiv für mathematische Logik und Grundlagenforschung pp. 3–16.
Martin-Löf, P.: 1984, Intuitionistic Type Theory, Bibliopolis, Napoli.
Milne, P.: 1994, Classical harmony: Rules of inference and the meaning of the logical constants,
Synthese 100, 49–94.
Milne, P.: 2002, Harmony, purity, simplicity and a “seemingly magical fact”, The Monist 85(4), 498–
535.
Negri, S.: 2002, A normalizing system of natural deduction for intuitionistic linear logic, Archive for
Mathematical Logic 41(8), 789–810.
Olkhovikov, G. K. and Schroeder-Heister, P.: forthcoming, On flattening elimination rules, Review
of Symbolic Logic .
Paoli, F.: 2002, Substructural Logics: A Primer, Kluwer Academic Publishers.
Paoli, F.: 2003, Quine and Slater on paraconsistency and deviance, Journal of Philosophical Logic
32, 531–548.
Paoli, F.: 2013, Semantic minimalism for logical constants, Logique et Analyse (forthcoming).
Petersen, U.: 2000, Logic without contraction as based on inclusion and unrestricted abstraction,
Studia Logica 64(3), 365–403.
Pfenning, F. and Davies, R.: 2001, A judgmental reconstruction of modal logic, Mathematical
Structures in Computer Science 11(04), 511–540.
Popper, K.: 1946, Logic without assumptions, Proceedings of the Aristotelian Society, Harrison &
Sons, Ltd., pp. 251–292.
Prawitz, D.: 1965, Natural Deduction: A proof-theoretical study, Almqvist and Wiksell, Stockholm.
Prawitz, D.: 1977, Meaning and proofs: on the conflict between classical and intuitionistic logic,
Theoria 43, 2–40.
Prawitz, D.: 1978, Proofs and the meanings and completeness of the logical constants, in J. Hintikka
(ed.), Essays on Mathematical and Philosophical Logic, Reidel, Dordrecht, pp. 25–40.
Prior, A. N.: 1961, The runabout inference ticket, Analysis 21, 38–39.
Read, S.: 2000, Harmony and autonomy in classical logic, Journal of Philosophical Logic 29, 123–154.
Read, S.: 2010, General-elimination harmony and the meaning of the logical constants, Journal of
philosophical logic 39(5), 557–576.
Read, S.: 2013, General-elimination harmony and higher-level rules, in H. Wansing (ed.), Dag
Prawitz on Proofs and Meaning, Studia Logica, Trends in Logic, Springer.
Restall, G.: 2000, An introduction to substructural logics, Routledge.
Restall, G.: 2002, Carnap’s tolerance, language change, and logical pluralism, Journal of Philosophy
99, 426–443.
Restall, G.: 2007, Proof theory and meaning: On the context of deducibility, Proceedings of Logic
Colloquium 2007, CUP.
Rumfitt, I.: 2000, “Yes” and “No”, Mind 109, 781–820.
Rumfitt, I.: 2008, Knowledge by Deduction, Grazer Philosophische Studien 77(1), 61–84.
24
Schroeder-Heister, P.: 1984, A natural extension of natural deduction, Journal of Symbolic Logic
49, 1284–1300.
Schroeder-Heister, P.: 1992, Cut-elimination in logics with definitional reflection, in D. Pearce and
H. Wansing (eds), Nonclassical Logics and Information Processing, Vol. 619 of Lecture Notes in
Computer Science, Springer, pp. 146–171.
Schroeder-Heister, P.: 2003, On the notion of assumption in logical systems, Selected papers
contributed to the sections of GAP5, Fifth International Congress of the Society for Analytical
Philosophy, Bielefeld, http://www.gap5.de/proceedings, pp. 27–48.
Schroeder-Heister, P.: 2007, Generalized definitional reflection and the inversion principle, Logica
Universalis 1(2). Forthcoming.
Schroeder-Heister, P.: forthcoming, Definitional reflection and basic logic, in M. E. Maietti, E. Palmgren and M. Rathjen (eds), Advances in Constructive Topology and Logical Foundations, Special
Issue of APAL in honour of Giovanni Sambin.
Sequoiah-Grayson, S.: 2009, A positive information logic for inferential information, Synthese
167(2), 409–431.
Shapiro, L.: 2010, Deflating logical consequence, The Philosophical Quarterly .
Steinberger, F.: 2011, Why conclusions should remain single, Journal of Philosophical Logic
40(3), 333–355.
Tennant, N.: 1982, Proof and paradox, Dialectica 36(2-3), 265–296.
Tennant, N.: 1987, Anti-realism and logic, Oxford University Press, Oxford.
Tennant, N.: 1997, The Taming of the True, Oxford University Press, Oxford.
Tennant, N.: 2002, Ultimate normal forms for parallelized natural deductions, Logic Journal of
IGPL 10(3), 299–337.
Tennant, N.: forthcoming, Inferentialism, logicism, harmony and a counterpoint, in A. Miller (ed.),
Essays for Crispin Wright: Logic, Language and Mathematics. 2 volumes, Oxford University
Press, Oxford.
Troelstra, A. S.: 1992, Lectures on Linear Logic, CSLI Lecture Notes 29, Center for the Study of
Language and Information, Stanford, California.
Troelstra, A. S. and Schwichtenberg, H.: 2000, Basic Proof Theory, Cambridge University Press,
Cambridge.
Ungar, A. M.: 1992, Normalization, Cut-Elimination and the Theory of Proofs, CSLI Publications.
von Plato, J.: 2001, Natural deduction with general elimination rules, Archive for Mathematical
Logic 40, 541–567.
Zardini, E.: 2008, Living on the slippery slope: the nature, sources and logic of vagueness, PhD
thesis, University of St. Andrews.
Zardini, E.: 2011, Truth without contra(di)ction, Review of Symbolic Logic 4(4), 498–535.
25