Legal Knowledge and Information Systems 141

E. Schweighofer (Ed.)
© 2021 The authors and IOS Press.
This article is published online with Open Access by IOS Press and distributed under the terms
of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).
doi:10.3233/FAIA210330

A Kelsenian Deontic Logic

Agata CIABATTONI a Xavier PARENT a and Giovanni SARTOR b
a Vienna University of Technology
b Cirsfid-Alma AI, University of Bologna; EUI, Florence

Abstract. Inspired by Kelsen’s view that norms establish causal-like connections

between facts and sanctions, we develop a deontic logic in which a proposition is
obligatory iff its complement causes a violation. We provide a logic for normative
causality, define non-contextual and contextual notions of illicit and duty, and show
that the logic of such duties is well-behaved and solves the main deontic paradoxes.

Keywords. Deontic Logic, Kelsen theory, Causality, Violations

1. Introduction
We develop a framework for deontic logic that combines violation and causality. Roughly
speaking, an action is obligatory if refraining from performing it causes a violation. Our
approach is inspired by the theory of norms developed by H. Kelsen, one of the most
important legal scholars of the 20th century, in his “Pure Theory of Law,” introduced
in [1] and expanded in [2] (English translations in [3] and [4], respectively). According
to Kelsen, obligations and prohibitions are mere reflexes of sanction-norms: “the legal
order [. . . ] prohibits a certain behavior by attaching to it a sanction or [. . . ] it commands
a behavior by attaching a sanction to the opposite behavior” [4, p. 55]. This idea may
be connected to the reduction of deontic logic to alethic modal logic as proposed by
Anderson [5], though the latter does not refer to the work of Kelsen (which at that time
was only available in German, and usually only known to legal theorists).
Here we use a reduction à la Anderson, but depart from it in two respects. First,
we model the connection between a sanction (actually, a violation) and its triggering
condition as a causal relationship, rather than as a necessity relationship, as proposed
in, e.g., [6]. The necessity connection is usually understood as a strict implication which
is known to generate counter-intuitive inferences, such as Ross’s paradox [7]. It also
satisfies the reflexivity postulate (“If A then A”), which is often regarded as inappropriate
for causal reasoning. Furthermore, Anderson’s reduction is usually worked out within a
possible worlds semantics, which is not compatible with Kelsen’s view that norms do
not bear a truth-value (see [8]). Second, in Anderson’s perspective the same sanction is
the consequent of each norm. In our approach instead different unlawful facts may lead
to different violations. In this regard our approach corresponds to the working of legal
and moral systems, where distinct unlawful or immoral acts lead to distinct sanctions
or disvalues. This feature of our framework enables us to address contrary-to-duty CTD
obligations, i.e., obligations which are applicable only if other obligations are violated.
We can represent the original obligation by a norm linking the (prohibited) fact f1 to
a violation v1 , and the CTD obligation by a second norm linking the accomplishment
of f1 in combination with a further fact f2 to an additional violation v2 . The first norm
142 A. Ciabattoni et al. / A Kelsenian Deontic Logic

expresses the obligation of ¬ f1 (e.g., the obligation not to kill, in Forrester’s famous
paradox [9]), while the second norm expresses the prohibition of f2 , when f1 is the case
(the prohibition to be cruel, when killing). Hence we model CTD obligations by making
the complement of their content into an aggravating circumstance, as in legal codes.
Before presenting our formal framework, we clarify some ideas about sanction and
violation. Neither Kelsen nor Anderson claimed that every action for which a sanction is
foreseen will necessarily be followed by an action of coercive enforcement by the state
(forced execution, fine, detention, etc.), nor even by the pronouncement of a sanction
by a competent authority. Once the condition for the sanction is realised, what neces-
sarily happens is that (for Kelsen) the sanction is authorised and thus can legitimately
be applied through the appropriate legal process, or (for Anderson) that something un-
wanted has happened. Hence we will model legal norms as connecting unlawful facts to
violations, rather than to sanctions.
The paper is organised as follows. Sect. 2 presents the norms we deal with, which
causally link (unlawful) states of affairs to violations, and the logic to reason about them.
The latter is a simplified version of input-output logic [10], where part of the norma-
tive system (regulative norms) has only a propositional constant on the right hand side.
In Sect. 3 we discuss the notion of illicit, which is used to define contextual and non-
contextual duties. In accordance with Kelsen’s view that a normative system can be con-
flicting, Sect. 4 introduces the notion of (un-)obeyable system. The resulting logic of
duties is analysed in Sect. 5, using as benchmarks well known properties and paradoxes
from the deontic logic literature. Sect. 6 pinpoints a selection of topics for future research.
2. The violation logic
We introduce the base violation logic that will be used in this paper, starting with its
language.

Definition 1 (Language). Let L be an ordinary propositional language containing the

classical connectives and constants {∧, ∨, ¬, →,t, f }, and a set V of violation atoms.

Each violation atom denotes a particular violation or offence (or, following Kelsen,
the authorisation to enact a specific sanction [4, 108ff]). Norms consist in causal-like
connections, denoted by the ⇒ symbol. They link factual circumstances to violations
(regulative or violation norms) or Boolean antecedents to conclusions other than viola-
tions (constitutive or counts-as norms).

Definition 2 (Norms and Norm codes). A norm code is a finite set of:

• Violations norms: A ⇒ v, where each A is formula from L \V and v ∈ V .

• Constitutive norms (count-as norms): A ⇒ B, where A, B are formulas from L \V .

Example 1 (Auto code). A norm code capturing simple road traffic rules is:
Speed ⇒ v1 Red ∧ ¬Stop ⇒ v2
Dark ∧ ¬LightsOn ⇒ v3 BrokenLights ⇒ v4
Phone ∧ Drive ⇒ v5 BrokenFrontLights ⇒ BrokenLights
BrokenBackLights ⇒ BrokenLights Fog ∧ ¬LightsOn ⇒ v6

Note that in our example different norms establish different violations. This feature
is meant to capture the fact that unlawful or immoral situations may trigger distinct re-
 A. Ciabattoni et al. / A Kelsenian Deontic Logic 143

sponses by a legal and moral system. Such responses have to be added up to determine
how the situation is assessed by the system (expanding the generated violations makes
things worse). Our approach does not exclude that distinct facts may lead to the same
violation. However, in this case the normative system would generate a single violation,
whenever one, some, or all such norms are triggered. For instance, the violation v4 may
be triggered in two ways (via BrokenFrontLights or via BrokenBackLights). Should dis-
tinct fines be applied for broken front lights and broken back lights (to be added up when
both are the case), different violations would be triggered by each of them.
We introduce our violation logic, that specifies a causal-like entailment for norms,
in the spirit of Kelsen (who calls this entailment “imputation”, see [4, 76ff]).

Definition 3 (Violation Logic). A violation inference relation is a binary relation ⇒

between the set of propositions in L satisfying the following rules (|= is the semantical
consequence in classical propositional logic):
(Truth) t ⇒ t
(Strengthening) If A |= B and B ⇒ C, then A ⇒ C;
(Weakening) If A ⇒ B and B |= C, then A ⇒ C;
(And) If A ⇒ B and A ⇒ C then A ⇒ B ∧C;
(Cut) If A ⇒ B and A ∧ B ⇒ C, then A ⇒ C.
(Or) If A ⇒ C and B ⇒ C, then A ∨ B ⇒ C.

Although causal relations satisfy most of the rules for classical entailment, their
distinctive feature is that they are irreflexive, that is, they do not satisfy A ⇒ A. Actually,
the above relation corresponds to the “basic reusable” input-output logic (out4 ) from
[10]. It is also closely related to Bochman’s causal calculus [11], see Remark 1. The
semantics for the violation logic is essentially the one for out4 . It is “operational”, and
takes the form of a set of procedures yielding outputs for inputs. Roughly speaking, to
determine if formula A is in the output set, one considers in turn each maximal consistent
extension of the input set that is closed under the norms, and checks if it contains A. If
the answer is “yes”, then A is in the output. This semantics fits Kelsen’s idea that norms
do not bear a truth-value and that logic cannot add new norms to a code N [12, Ch. 50],
but rather identifies the input-output connections established by N, which are the object
of “rules (propositions) of law” (in German Rechtssätze)−see [4, p. 72] and [12, Ch. 49].
3. From illicits to duties
In Kelsen’s legal theory sanction norms (violation norms, in our framework) have a foun-
dational status. Other normative notions are derivative. Every behaviour that may trigger
a sanction against its author is a delict (Unrecht) and every delict is the content of the
obligation that the delict does not take place [2, p. 39]. Here we prefer to speak of an
illicit, rather than of a delict, to cover all elements (not only actions) that contribute to the
triggering of a sanction. Elementary illicits represent the minimal conditions that lead to
a violation, and elementary duties apply to their negations.

Definition 4 (Elementary Illicit and Duty). A conjunction of literals {l1 , . . . , ln } is an
elementary illicit relatively to a norm code N if and only if (iff) there is a v ∈ V such that:

1. {l1 , . . . , ln } ⇒ v; 2. no proper subset of {l1 , . . . , ln } satisfies condition 1.
A proposition A is (the content of) an elementary duty, relatively to a norm code N,
iff A ≡ ¬B and B is an elementary illicit relatively to N.
144 A. Ciabattoni et al. / A Kelsenian Deontic Logic

Example 2. In Ex. 1, {Speed}, {Red, ¬Stop}, {Dark, ¬LightsOn}, {BrokenFrontLights},

{BrokenBackLights} and {Phone, ¬ LightsOn} are elementary illicits ; their negations
(¬Speed, ¬Red ∨ Stop, etc.) are the content of elementary duties.

To develop a deontic logic we introduce the idea of generalized illicit, which covers
all possible conditions that may minimally lead to a disjunction of violations.

Definition 5 (Generalised illicit). A boolean formula A is a generalised illicit relatively

to a norm code N if there exists a set of violations S ⊆ V such that:
 
(1) A ⇒ S, and (2) there is no B such that A |= B, B |= A and B ⇒ S.

The rationale behind this notion is to ensure that the generalised illicit is not only
a sufficient, but also a necessary condition for the disjunction of the violations to take
place. Condition 2. makes A the weakest formula leading to the violations in question.
This will be the key element to the solution of the deontic paradoxes in Sect. 5.2.

Example 3. Consider again Ex. 1. The generalized illicits w.r.t. {v4 } and {v1 , v5 } are:
BrokenFrontLights ∨ BrokenBackLights and Speed ∨ (Phone ∧ Drive), respectively. Note
that BrokenFrontLights alone is not a generalized illicits as it is not the only possible
way of triggering the violation v4 (i.e., condition 2. in Def. 5 fails).

Generalised illicts have the following logical properties: they are reducible to a dis-
junction of elementary illicits, and their disjunction constitutes a new generalised illict.

Proposition 1. If A is a generalised illicit relatively to a norm code N, then there exists

a disjunction of elementary illicits {L1 , . . . , Ln } such that A ≡ L1 ∨ . . . ∨ Ln .

Proof. Let L1 , . . . Ln be all possible elementary illicits relative to N w.r.t. the violations

v ∈ S ⊆ V . L1 ∨ . . . ∨ Ln is a generalized illicit w.r.t. S. Condition 1 of Def. 5 is satisfied.
 
Indeed, if Li ⇒ v ∈ S, then by (Weakening) Li ⇒ S, and by (Or) L1 ∨ . . . ∨ Ln ⇒ S.
As for condition 2, assume there is a boolean formula B such that L1 ∨ . . . ∨ Ln |= B

and B ⇒ S. Assume w.l.o.g. that B is in disjunctive normal form, say B := N1 ∨· · ·∨Nm .
 
By (Strenghtening) and the assumption N1 ∨ · · · ∨ Nm ⇒ S it follows that Ni ⇒ S,
for all i = 1, . . . , m. Being the Li ’s all the elementary illicits triggering the violations in S
(and hence representing minimal conditions to trigger them), for each Ni there are some
literals {L1 , . . . Ln } which are included in the literals of Ni . Hence B |= L1 ∨ . . . ∨ Ln .

Corollary 1. If A1 and A2 are generalized illicit relatively to N, so is A1 ∨ A2 .

The concept of generalised illicits leads us to noncontextual duties, i.e., states of

affairs whose non-realisation lead to a violation. More formally:

Definition 6 (Noncontextual Duty). A Boolean formula A is (the content of) a noncontex-

tual duty relatively to a norm code N, denoted as ON A, iff A ≡ ¬B and B is a generalised
illicit relatively to N.

Example 4 (Ctd. from Ex. 3). ON (¬ BrokenFrontLights ∧ ¬BrokenBackLights) and

ON (¬Speed ∧ (Fog → LightsOn)) are duties.

Corollary 2. If ON A then A ≡ ¬L1 ∧ . . . ∧ ¬Ln , for L1 , . . . , Ln elementary duties.

 A. Ciabattoni et al. / A Kelsenian Deontic Logic 145

Remark 1. Out4 [10], and hence our violation logic, differ from the causal calculus
in [11] by the presence in the latter of axiom f ⇒ f . This axiom would create the fol-
lowing counter-intuitive situation when considering the corresponding obligations. From
A ⇒ B follows A∧¬B ⇒ f .1 Now, by (Weakening), for any violation v, A∧¬B ⇒ v (using
the fact that f |= v), which leads to ON (¬A ∨ B), for any constitutive norm A ⇒ B.

3.1. Putting illicits and duties in context

We consider how norms operate relatively to contexts, i.e., in circumstances considered
to be settled. We will regard contexts as kind of restrictions of the set of possible world,
which are limited to those satisfying the context. These restrictions may depend on dif-
ferent considerations such as natural necessity (laws of nature), temporal necessity (the
immutability of the past), or even the settled choices of the agent.

Definition 7 (Context). A context for a norm code is a consistent set of literals.

Here we focus on the illicits that are not settled by the context (entailed by it), so
that their happening is contingent on the choice (deliberation) of the involved agent.

Definition 8 (Contextual Illicit). A Boolean formula A is a contextual illicit (c-illicit)

relative to a norm code N and context C iff there is a set of violations S ⊆ V s.t.
 
1. (C ∪ {A}) ⇒ S
 
2. there is no B such that A |= B, B |= A and (C ∪ {B}) ⇒ S
 
3. C ⇒ S and C |= ¬A

Establishing that no weaker formula generates the considered violations, condition

2, is useful to resolve a number of deontic paradoxes, in particular those following from
the assumption of closure of the deontic operator under logical consequence. Condition
3 tells us that A is “needed” to generate the violation in question, and also that the truth
or falsity of A is not settled by the context, as shown in Remark 2.
 
Remark

2. C ⇒ S in Def 8 (3) implies that C |= A. Bycondition

1 in Def 8,
(C ∪ {A}) ⇒ S. Suppose 
C |=

A. By propositional logic C |= (C ∪ {A}). By
(Strengthening), one gets C ⇒ S. Contradiction.

Example 5. (Ctd. from Ex 1) In context C = {BrokenFrontLights}, BrokenBackLights

is not a c-illicit (due to the first half of condition 3 in Definition 8). The intuition is that
when it is settled that one the two requirements leading to the same violation v4 is met,
meeting the other becomes irrelevant.

On the basis of c-illicts we define contextual duties.

Definition 9 (Contextual Duty). A Boolean formula A is a contextual duty relatively to

a norm set N and to a context C, denoted as O(N,C) (A) iff A ≡ ¬B and B is a c-illicit
relatively to N and C (we omit N and C, when no ambiguity occurs).

We apply below our approach to two well-known deontic paradoxes pertaining to

contrary-to-duty (CTD) scenarios (see Sect. 5.2 for more paradoxes).
1 Proof: If A ⇒ B, then A ∧ ¬B ⇒ B by (Strengthening). By f ⇒ f and (Strengthening), A ∧ ¬B ∧ B ⇒ f . By

(Cut), A ∧ ¬B ⇒ f .
146 A. Ciabattoni et al. / A Kelsenian Deontic Logic

Example 6 (Forrester paradox [9]). Consider the following premises: (1) You should
not kill (2) If you kill, you should kill gently, (3) You kill. In Standard Deontic Logic
SDL [13] (1)-(3) entail that you should both kill and not kill. By considering under what
circumstance obligations (1) and (2) would be violated we have the violation norms
Kill ⇒ v1 Kill ∧ ¬KillGently ⇒ v2
where KillGently ⇒ Kill. In context {}, O¬Kill holds, in context {Kill}, OKillGently.

Example 7 (Chisholm paradox [14]). It consists of: (1) You ought to go to the assistance
of your neighbours; (2) If you go you ought to tell them that you are coming; but (3) If
you do not go then you ought not to tell them that you are coming; and (4) You do not go.
SDL [13] entails that both you ought and you ought not to tell your neighbours that you
are coming. In our framework the norms involved in this scenario are formalised as:
¬Go ⇒ v1 Go ∧ ¬TellGo ⇒ v2 ¬Go ∧ TellGo ⇒ v3
In context {}, OGo holds, in context {¬Go}, O¬TellGo.

Remark 3. Some c-illicit A relative to a norm code N and context C1 , might not be a
c-illicit relative to a superset C2 of C1 that is consistent with A and such that C2 |= ∨S,
for any (sub)set S of the violations in N. E.g., put N = {A ∧ B ⇒ v1 , D ∧ E ⇒ v2 }; A ∧ B
is a c-illicit in context C1 := {E}, but not in C2 := {A, E} (condition 2 in Def. 8 fails).

The following property, connecting violations entailed by contexts and duties, will
be useful in Sect.5.2.

Lemma 1. Let N be a norm code. If C |= ¬A1 ∨ · · · ∨ ¬An , for all duties ON A1 , . . . , ON An

 
then C ⇒ S, for every S ⊆ V .

Proof. By Def. 6 each ¬Ai is a generalized illicit relative to N and by Prop. 1 a disjunc-
tion of elementary illicits (minimal formulas that trigger violations). If there exists S s.t.
 
C ⇒ S, there are generalized illicits ¬A1 , . . . ¬Am such that C |= ¬A1 ∨ · · · ∨ ¬Am .

4. (Un)Obeyability of normative codes

Kelsen pointed out that a normative code may establish requirements (cf. [15, p.25]) that
cannot be jointly complied with: : “within [. . . ] a normative order the same behaviour
may be [. . . ] commanded and forbidden at the same time [. . . ]. This is the case if a certain
conduct is the condition of a sanction and at the same time the omission of this conduct
is also the condition of a sanction.” We formalize this intuition through two notions,
absolute and contextual unobeyability.

Definition 10 (Absolute (un)obeyability). A code N is absolutely unobeyable iff t ⇒ V ,
and it is absolutely obeyable otherwise.

Example 8. An absolutely unobeyable code is {speed ⇒ v1 ; ¬speed ⇒ v2 }. Indeed by

(Weakening) and (Or) we get t ⇒ v1 ∨ v2 .

Absolutely unobeyable codes are rare, as they involve norms that always establish
sanctions. A weaker and more common notion is that of contextual unobeyability. A
code N is unobeyable in a context C iff C entails that that a disjunction of alternative
violations will be committed, not specifying which ones will be. Thus, the agent faces a
predicament: possible violations can only be avoided by incurring in other violations.
 A. Ciabattoni et al. / A Kelsenian Deontic Logic 147

Definition 11 (Contextual unobeyability). A code N is unobeyable in a context C iff there

  
is a set of violations Vi ⊆ V such that: (1) C ⇒ Vi , and (2) for each v ∈ Vi , C ⇒ v.

Example 9. The code N := {speed ⇒ v1 ; motorway ∧ ¬speed ⇒ v2 } is unobeyable in

context {motorway} while being obeyable in context {} (through ¬Speed). Since for no

set of violations Vi ⊆ V , t ⇒N Vi , N is absolutely obeyable.

When a code N is unobeyable in a context C, in that context violations cannot be

avoided and the addressees of N are forced to deliberate on which not yet settled violation
to commit. They will have to face a “tragic dilemma”, as in the biblical story below.

Example 10 (from the Book of the Judges). Jephthah promised to God that if he was
given victory in a battle he would sacrifice (kill and dedicate to God) the first human
being that he encountered coming home. After winning the battle, he first encountered his
daughter (rather than an animal, as he may have assumed). Thus, he faced a hard choice:
either violate his promise to God, or violate the moral prohibition to kill his daughter. We
can model this situation through the code N = {Win(j) ∧ Encounter(j,d) ∧ ¬Kill(j,d) ⇒
v1 ; Kill(j,d) ⇒ v2 }, which is unobeyable in context C = {Win(j), Encounter( j, d)}.

Note that in a context of unobeyability, the agent has the duty to prevent each fact
causing a new violation. However, he has no contextual duty to prevent the tautological
disjunction of all such facts (i.e. to realise the contradictory conjunction of them). In fact
such a disjunction is not a c-illicit, being entailed by the context (Remark 2). Thus, in
the above example, in context C Jephthah has duties OKill( j, d) and O¬Kill( j, d), but
no duty O(Kill( j, d) ∧ ¬Kill( j, d)).
5. The logic of duties
In this section we analyse the logic(s) of the ON and O(N,C) modalities. We consider the
main principles discussed in the deontic logic literature, and check whether they hold.
We also use some of the best known deontic paradoxes as benchmarks.
5.1. Properties of duties
For ease of readability, we consider first the version of a given property for noncontextual
duty (i.e. unary obligation), when available.
Extensionality For non-contextual duty it takes the form (RE) “If A ≡ B, then ON (A) ≡
ON (B)”. There are two versions for conditional duty: “If ∧C ≡ ∧C , then O(N,C) (A) ≡
O(N,C ) (A)” and “If A ≡ B, then O(N,C) (A) ≡ O(N,C) (B)”. All versions trivially hold.

And introduction If ON A and ON B, then ON (A ∧ B). Assume ON A and ON B. So A ≡

¬A , and B ≡ ¬B for some A , B generalised illicits. We have (A ∧ B) ≡ (¬A ∧ ¬B ) ≡
¬(A ∨ B ). Furthermore, A ∨ B is a generalised illicit by Corollary 1. This suffices for
ON (A ∧ B). The analog principle for contextual duties does not hold unrestrictedly. E.g.,
assume that O(N,C) (A) and O(N,C) (¬A). O(N,C) (A ∧ ¬A) is not a contextual duty as the
negation of its content is equivalent to t, which is not a c-illicit (cond. 3 in Def. 8 fails)

Remark 4. The logic of ON duties contains the non-normal modal logic EC [16]. EC is
obtained by adding the C axiom (ON A ∧ ON B) → ON (A ∧ B) to the system E of so-called
classical modal logic, consisting of the sole rule of extensionality (RE).
148 A. Ciabattoni et al. / A Kelsenian Deontic Logic

Monotonicity wrt context It is the analog of the “Strengthening of the antecedent” prin-
ciple in conditional logic. It fails in its usual form “If O(N,C) (A), and C ⊆ C then
O(N,C ) (A)”, as hinted in Remark 3. The following counter-example shows that its failure
is in line with the idea that in the context of deliberation one puts aside the moral status
of the facts which are settled. Let N = {¬A ⇒ v1 }. We have O(N,0) / A, while O(N,{¬A}) A
no longer holds due to the failure of (the first half of) condition 3 in Def. 8.
Factual detachment (= detachment via modus ponens [17]) In our framework it might
be expressed as: If O(N,C) (A → B) then O(N,C∪{A}) B, for A a conjunction of liter-
als. It holds under the hypotheses: (∗) C ∪ {A} |= ¬A1 ∨ · · · ∨ ¬An , for all duties
ON A1 , . . . , ON An , and (∗∗) there is no formula D weaker than ¬B s.t. A ∧ D |= ¬B. We
show that ¬B is a c-illicit in context C ∪ {A} so that O(N,C∪{A}) B. By O(N,C) (A → B) fol-
   
lows that there is S ⊆ V s.t. (C ∪ {A ∧ ¬B}) ⇒ S, that is (C ∪ {A} ∪ {¬B}) ⇒ S.
Hence Def. 8.1 holds for ¬B relative to C ∪ {A}. Condition 2 follows from the fact that
 
if there is a D weaker than ¬B such that (C ∪ {A} ∪ {D }) ⇒ S, then A ∧ ¬B is not

a c-illicit (C ∪ {A ∧ D } ⇒ S and from ¬B |= D follows A ∧ ¬B |= A ∧ D , and from
(∗∗) that A ∧ D |= A ∧ ¬B) contradicting the hypothesis. Condition 3 also holds: the hy-
 
pothesis (∗) guarantees by Lemma 1 that (C ∪ {A}) ⇒ V , where V is the set of all
 
violations, and hence (C ∪ {A}) ⇒ S; C ∪ {A} |= B follows from C |= A → B.

No conflict ¬(ON A ∧ ON ¬A) holds for a normative system N that is absolutely

obeyable in the sense of Definition 10. Suppose N has this property. Assume by contra-
diction that both ON A and ON ¬A hold. Hence A ≡ ¬B1 , and ¬A ≡ ¬B2 for some B1 , B2
 
generalised illicits. Hence B1 ∨ B2 ⇒ (V1 ∪V2 ), and so t ⇒ (V1 ∪V2 ), since B1 ≡ ¬A

and A ≡ B2 . By (Weakening), t ⇒ V , where V is the set of all violations. Contradiction.
The contextual version of the principle –¬(O(N,C) A ∧ O(N,C) ¬A)– holds for norma-
tive systems that are contextually obeyable in C. Assume by contradiction that so is N
and that O(N,C) A and O(N,C) ¬A hold. Being A and ¬A contextual illicits w.r.t. C, there are
    
V1 and V2 s.t. (∗) (C ∪ {A}) ⇒ V1 and (C ∪ {¬A}) ⇒ V2 , and (∗∗) ∧C ⇒ V1 and
  
∧C ⇒ V2 . From (∗) it follows that C ⇒ V1 ∪V2 (i.e., condition (1) of Def. 11), and

by (∗∗) that for each v ∈ V1 ∪V2 , C ⇒ v (i.e. condition (2) of Def. 11), thus establishing
that N is unobeyable in context C. This property echoes the fact that for Kelsen normative
systems can be conflicting. Thus, for him, and in our framework, the “no conflict” axiom
(also known as D axiom in modal logic) only holds for normative systems that are either
absolutely or contextually obeyable.
5.2. Solutions to deontic paradoxes
We analyse the behaviour of our logic w.r.t. the main deontic paradoxes that have beset
SDL, which corresponds to the deontic logic obtained by Anderson using his reduction
schema. Deontic paradoxes are intended here in a broad sense as (un)derivable theorems
that are counter-intuitive in a common-sense reading.
Recall that unconstrained I/O logic [10], which is at the base of our framework, is
unable to handle many of the considered paradoxes. Two paths have been followed in
order to fix it: using constraints to filter excess outputs [18], or suitably weakening the
logic. In particular, (Weakening) and axiom (Truth) go, and at the same time a consis-
tency check restrains the application of some rules (like AND introduction). Although
based on the strongest system of unconstrained I/O logic (Def. 2), our framework elimi-
nates those paradoxes due to condition 2. in Definitions 5 and 8.
 A. Ciabattoni et al. / A Kelsenian Deontic Logic 149

Contrary-to-duty obligation (Forester’s and Chisholm’s) paradoxes As seen before, in

our framework we can overcome such paradoxes. In Forrester’s case (see Ex. 6) in con-
text {kill}, OKillGently holds, and the opposite obligation O¬KillGently does not hold.
The treatment of Chisholm’s scenario is similar (see Ex. 7).
Ross’ paradox The paradox consists in the derivation of (a) You should mail the letter
or burn it, from (b) You should mail the letter. Introduced in 1941 by Ross [19], this para-
dox has been a discussion topic ever since. It does not appear in our logics. E.g., relative
to a normative system N = {¬PostLetter ⇒ v1 }, we have the obligation OPostLetter, but
we do not have O(PostLetter ∨ BurnLetter), due to condition 2. in Def. 5.
Deontic detachment [17] It fails in its usual form: If OA and O(A → B) then OB.
This can be illustrated with Broome [20]’s counter-example: let N = {¬Exercise ⇒
v1 ; Exercise ∧ ¬EatMore ⇒ v2 } (one should exercise and if one exercises, one should
eat more). We have OExercise and O(Exercise → EatMore). But it should not be the
case that OEatMore, since, intuitively, the obligation to eat more holds only if the first
obligation is fulfilled. The correct inference is captured by the following form of deontic
detachment, called “aggregative” in [21], as one keeps track of what has been previously
detached: from OA and O(A → B), infer O(A ∧ B), which holds in our logic due to AND
introduction and Extensionality. Hence from N we can infer O(Exercise ∧ EatMore).
The contextual version of deontic detachment, with O(A → B) replaced by O(N,A) B, does
not hold either. For a counter-example, let N = {¬a ⇒ v1 , a ∧ ¬b ⇒ v2 }. We have Oa
and Oa b, but not Ob, as ¬b is not a generalized illicit (condition 1 in Def. 5 fails).
The alternative service paradox This scenario, proposed by Horty in 1994 [22], is han-
dled similarly to the previous case, by changing the disjunctive obligation in the original
formulation into a conditional obligation: from (1) You should fight in the army or per-
form alternative service (O(¬A → B)), and (2) You should not fight in the army (O¬A),
we derive that (3) You should not fight and perform alternative service (O(¬A ∧ B)).
Necessitation Ot does not hold. Take any non-empty set of norms, e.g., N = {Speed ⇒
v1 }. O(Speed ∨ ¬Speed) does not hold, because Speed ∧ ¬Speed is not a generalised
illicit, due to condition 2. in Def. 5.
Reflexivity This is the law O(N,C) ∧C. It fails, because of condition 3 in Def 8 (C |= ∧C,
see Remark 2) and so ¬ ∧C cannot be a c-illicit relative to C.
And elimination It is the principle “If O(A ∧ B) then OA”, known in the non-normal
modal logic literature as axiom M (see [16]). As suggested by a number of authors,
when A and B are not separable, such a principle is counter-intuitive, so that (to quote
Hansen) “failing a part [of the order] means that satisfying the remainder no longer makes
sense. E.g. if I am to satisfy the imperative ‘buy apples and walnuts’, and the walnuts
[...] and the apples [are meant to] land in a Waldorf salad, then it might be unwanted
and a waste of money to buy the walnuts if I cannot get the apples” [23, p. 91]. Put
N = {¬Apples ∨ ¬Walnuts ⇒ v} (or equivalently {¬Apples ⇒ v, ¬Walnuts ⇒ v}). We
have O(Apples ∧ Walnuts), but not O(Apples), due to condition (2) in Def. 5.
6. Future work
We have developed a Kelsenian deontic logic that defines obligation in terms of viola-
tion and causality. The behaviour of the resulting obligation has been analyzed using as
benchmarks well known properties and paradoxes from the deontic logic literature.
150 A. Ciabattoni et al. / A Kelsenian Deontic Logic

It has often been argued that CTDs involve different senses of “ought”, and that a
fully adequate treatment of them must be able to capture those nuances. In our analysis
of contextual duties we have focused on deliberative duties, where one puts aside the
moral or legal status of the facts which are taken as settled and one must decide what
to do. Our framework allows, however, a finer-grained analysis of duties –to be left to
future work– which could not only shed more light on the various paradoxes, but also
capture different types of “ought”-statements. Among them, we plan to investigate con-
textual (evaluative) duties that are appropriate for what is commonly referred to as the
“context of judgement” [24,25], where one assesses the moral or legal status of settled
facts through backward looking or post-eventum judgments [26, p. 157].
Furthermore, our logic is an ought-to-be deontic logic, as obligations may cover any
cause of violations. We plan to develop an ought-to-do deontic logic, reflecting Kelsen’s
notion of a delict, by carving out agentive elements within illicits and “plugging” in a
suitable logic of action in our base logic. Other topics for future research include the
question of extending the framework to support defeasible reasoning, reasoning about
exceptions, and the question of how to axiomatise and automatise the logic.2
References
[1] Kelsen H. Reine Rechtslehre Einleitung in die rechtswissenschaftliche Problematik. Mohr Siebeck;
1934.
[2] Kelsen H. Reine Rechtslehre. Franz Deuticke; 1960.
[3] Kelsen H. Introduction to the Problems of Legal Theory. Clarendon; 1992.
[4] Kelsen H. The Pure Theory of Law. University of California Press; 1967.
[5] Anderson AR. The Logic of Norms. Logique et Analyse. 1958;1:84–91.
[6] Wang PH. Anderson’s reduction and Kelsen’s Normativism. In: Pluralism and Law. Proceedings of the
20th IVR World Congress 2001. Vol. 4: Legal Reasoning. Franz Steiner Verlag; 2001. p. 104–110.
[7] Ross WD. Foundations of Ethics. Clarendon; 1939.
[8] Kelsen H. General Theory of Norms. Oxford University Press; 1990.
[9] Forrester JW. Gentle Murder, or the adverbial Samaritan. J of Philosophy. 1984;(81):193–197.
[10] Makinson D, van der Torre LWN. Input/Output Logics. J Phil Log. 2000;29(4):383–408.
[11] Bochman A. On Laws and Counterfactuals in Causal Reasoning. In: Proc. 16 Intern. Conf. on Principles
of Knowledge Representation and Reasoning (KR 2018); 2018. p. 494–503.
[12] Kelsen H. General Theory of Norms. Clarendon; 1991.
[13] von Wright GH. Deontic logic. Mind. 1951;60(237):1–15.
[14] Chisholm RM. Contrary-to-Duty Imperatives and Deontic Logic. Analysis. 1963;24:33–6.
[15] Kelsen H. On the pure theory of law. Israel Law Review. 1966:193–215.
[16] Chellas B. Modal Logic. Cambridge University Press; 1980.
[17] Greenspan PS. Conditional Oughts and Hypothetical Imperatives. Journal of Philosophy.
1975;72(10):259–276.
[18] Makinson D, van der Torre LWN. Constraints for Input/Output Logics. J Phil Log. 2001;30(2):155–185.
[19] Ross A. Imperatives and Logic. Theoria. 1941;7(1):53.
[20] Broome J. Rationality through Reasoning. West Sussex, U: Wiley-Blackwell; 2013.
[21] Parent X, van der Torre LWN. ”Sing and Dance!” - Input/Output Logics without Weakening. In: Cariani
F, et Al, editors. DEON 2014, Proceedings. vol. 8554 of LNCS. Springer; 2014. p. 149–165.
[22] Horty J. Moral Dilemmas and Nonmonotonic Logic. Journal of Philosophical Logic. 1994;23(1):35–65.
[23] Hansen J. Imperatives and Deontic logic. University of Leipzig; 2008.
[24] Thomason RH. In: Hilpinen R, editor. Deontic Logic and the Role of Freedom in Moral Deliberation.
Dordrecht: Springer Netherlands; 1981. p. 177–186.
[25] Asher N, Bonevac D. Prima Facie Obligation. Stud Logica. 1996;57(1):19–45.
[26] Hare R. The Language of Morals. Clarendon; 1991.

2 G. Sartor has been supported by the H2020 ERC Project “CompuLaw” (G.A. 833647), A. Ciabattoni and

X. Parent by the WWTF project MA16-028.