Académique Documents
Professionnel Documents
Culture Documents
DISSERTATION
von Herrn M. A. J
org Hansen
g e b o r e n a m 09.09.1967
in Cochem
Eingereicht am 24.01.2008
Offentlich
verteidigt am 18.11.2008
Tag der Verleihung: 25.11.2008
Abstract
Deontic logic has so far been almost exclusively interpreted by a possible
worlds semantics and ideality or preference relations between the worlds. In
distinguishing itself from attempts at a logic of norms or logic of imperatives, deontic logic has portrayed its formulas as deontic propositions, true
or false statements about what is obligatory, permitted or forbidden according to some unspecified normative system. Based on this idea, the imperatival tradition of deontic logic, an unorthodox sidearm of the mainstream,
has interpreted the formulas of deontic logic not with respect to ideal or best
worlds, but to given sets of norms or imperatives. In a series of papers standing in this tradition, I have shown that by using an explicitly represented set
of imperatives and what they command as the logical semantics by which
the truth of deontic formulas is defined, all the standard systems of monadic
and dyadic deontic logic can be reconstructed, i.e. they are sound and complete with respect to such semantics. The basic concepts are motivated here
from a position of imperativological scepticism, and the main results are
summarized.
Keywords:
deontic logic, logic of imperatives, modal logic
Contents
Introduction
1 The
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Logic of Imperatives
Beginnings: Poincares Proposal . . . .
Jrgensens Dilemma . . . . . . . . . .
Dubislavs Trick and Related Theories
Explanations of Imperative Inferences .
1.4.1 Logic of Satisfaction . . . . . .
1.4.2 Logic of Existence . . . . . . . .
1.4.3 Logic of Non-Factual Existence
1.4.4 Formalistic Approaches . . . . .
Rosss Paradoxes . . . . . . . . . . . .
Ordinary Language Arguments . . . .
The Way to Go Forward . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 Deontic Logic
2.1 Beginnings: Leibnizs Discovery . . . . . . . . .
2.2 Standard Deontic Logic . . . . . . . . . . . . . .
2.3 Andersons Reduction . . . . . . . . . . . . . .
2.4 Possible Worlds Semantics . . . . . . . . . . . .
2.5 Dyadic Deontic Logic . . . . . . . . . . . . . . .
2.6 Doubts and Questions . . . . . . . . . . . . . .
2.6.1 Doubts About Possible Worlds . . . . . .
2.6.2 The Question of Conflicts and Dilemmas
2.6.3 Interpretations of Deontic Logic . . . . .
2.7 A Fundamental Problem of Deontic Logic . . .
ii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
3
5
8
15
16
17
19
21
25
35
46
.
.
.
.
.
.
.
.
.
.
50
50
53
55
57
60
66
66
69
72
77
117
Acknowledgements
120
Bibliography
121
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
142
142
164
187
215
249
Anhang
274
Zusammenfassung . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
Bibliographische Beschreibung . . . . . . . . . . . . . . . . . . . . . 281
Lebenslauf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
iii
iv
Publikationen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Vortrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Versicherung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Introduction
At the biannual EON-workshop on Deontic Logic in Computer Science,
held at the University of Bologna, 810 January 1998, David Makinson drew
attention to the distinction between norms, that cannot meaningfully be
termed true or false, and deontic propositions, true or false statements about
what is obligatory, permitted or forbidden according to the norms, and that
this distinction has tended to get lost in modern deontic logic literature.
He therefore called for a reconstruction of deontic logic as a logic concerned with norms, but in accord with the philosophical position that norms
are devoid of truth values. Motivated by this call I have, in a series of research papers since then, provided a semantics for deontic logic that defines
its operators by means of an explicitly given set of imperatives. It can be
demonstrated that all of the standard systems of monadic and dyadic deontic
logic can be reconstructed with respect to such an imperative-based semantics, but also other systems are explored that consider specific issues, like
the possibility of conflicts and dilemmas, or the accommodation of priority
relations between the imperatives.
Here, the topic is explained and the results are presented that have been
achieved so far. Chapter 1 addresses the question of whether there exists a
logic of imperatives. If it does, then it would perhaps be easier to devise a
deontic logic as a logic that reflects already given logical relations between
imperatives, but the arguments make it clear that our start must be from
a position of imperativological scepticism, and no direct logical relations
between imperatives can be assumed. Chapter 2 lays out the main systems
of deontic logic that are later to be reconstructed, and explains the background of Makinsons criticism directed at traditional deontic logic. Chapter
3 first provides an overview over an already existing imperatival tradition
of deontic logic that has, instead of using standard possible worlds semantics with ideality or preference relations, interpreted deontic operators with
1
INTRODUCTION
respect to given sets of imperatives or norms. Then the imperative semantics used here to reconstruct the systems of deontic logic is explained, and
the results of the previous research are summarized. Finally it is examined
in which way previous paradoxes of deontic logic may be treated within
the new semantical framework. The conclusion explains what the results
mean for our understanding of deontic logic, and hints at possible further
developments.
Chapter 1
The Logic of Imperatives
1.1
Beginnings: Poincar
es Proposal
LOGIC OF IMPERATIVES
LOGIC OF IMPERATIVES
1.2
Jrgensens Dilemma
For such textbook definition cf. Mates [147] p.5, Lemmon [134] p.1, Hodges [98] p.55.
LOGIC OF IMPERATIVES
LOGIC OF IMPERATIVES
also conclude that there must be another place that is also called Frankfurt
(namely Frankfurt/Oder), and that this place does not lie near the Main,
because otherwise the extra information would not have been discriminating.
But the sign post does not point into the direction of this other city. Similarly
the information that is obtained by interpreting an imperative utterance may
be used to infer some other information. But this is not inferring imperatives.
The fact that imperatives are traditionally not considered to be true or
false finds its explanation in the different intentions in which imperatives
and indicatives are used. The main use of indicatives is to convey what the
speaker believes the world to be like. If it is so, then the sentence is called
true, if not, then it is called false and the recipient might point out that I
should perhaps change my beliefs. By use of an imperative I tell the addressee
what I want to be done. If the addressee does what is demanded, the action
may be qualified as right, or satisfactory with respect to the command, and
if not, then the behavior of the addressee is in some sense wrong and I will
perhaps remind the agent of his or her obligation. So truth and falsity are
the qualities of descriptions when things are or are not as they have been
described, while right or wrong are the qualities of acts that are or are not
in accordance with what has been prescribed. Descriptions and prescriptions
have a different direction of fit, and true/false are the terms used to express
the match/mismatch on the language side in case of a descriptive use of
language, and right/wrong are the terms employed for the match/mismatch
on the world side in case of a prescriptive use.4 Therefore it is a confusion
of language, and indicates a misunderstanding of the intention in which the
sentence has been uttered, if imperatives are termed true or false.
Accordingly, the most effort regarding Jrgensens dilemma has been
spent on developing alternative definitions for imperative inferences, rather
than arguing for the application of the terms of truth and falsity to imperatives unless one is already convinced by the dilemma that such things as
inferences with imperatives are at all impossible (e.g. Keene [120]).
4
This explanation of why the terms of truth and falsity are not applicable to normative
uses of language originates with Anscombe [19] 32. Independent accounts can be found
in Kenny [122] p. 68 and Peczenik [166], [167] who speaks of the norm as a qualifying
utterance. The dual terms right/wrong are used as corresponding qualifications e.g. by
Englis [52] and Kelsen [121] p. 132.
LOGIC OF IMPERATIVES
1.3
Mally [145] p. 12 seems to have introduced the symbolism !A, which was then employed
by Hofstadter & McKinsey [99] for the imperative that demands that A be the case.
However, Mally intended !A to be interpreted theoretically, as an assertion or assumption
that A ought to be, which we now call a deontic proposition and formalize by OA.
6
Cf. Dubislavs use of the plural when stating that an inference from demand-sentences
will now be formally facilitated by the following convention, and Dubislavs summary,
in which he stresses that no demand-sentence can be derived from premisses that do not
contain at least one demand-sentence.
LOGIC OF IMPERATIVES
LOGIC OF IMPERATIVES
10
has an indicative parallel-sentence which describes the contents of the command or wish. Jrgensen suggests that an imperative consists of an imperative factor and an indicative factor, where the first indicates that something
is commanded, and the second what is commanded. The indicative factor
can then be separated from the imperative and formulated in indicative sentences describing the action, change or state of affairs which is commanded.
Applying the rules to these latter sentences we can thus indirectly apply the
rules of logic to the imperative sentences to make their entailments explicit.
Writing A for the assertion that the state of affairs A holds, this slightly
different analysis of imperative inferences is illustrated by the next figure:7
LOGIC OF IMPERATIVES
11
logical reasoning is done with descriptors only - this Hare calls the principle
of the dictive indifference of logic, there is no special need for a logic of
imperatives. Rather, all logic is recast as a logic of descriptors, where if the
descriptors of the premisses describe a state of affairs, then the descriptor of
the conclusion describes, at least partially, the same state of affairs. Hares
view of logic is pictured in the next figure:
LOGIC OF IMPERATIVES
12
This explanation and the formulation of the principle below is most clearly expressed
in Weinberger [234] p. 172. Weinberger uses the term coordination instead of correspondence, but this suggests an onto or even one-to-one mapping.
11
Besides Weinberger cf. Hamblin [74] pp. 151-152: Take the exact words of the imperative, and transform them into indicative mood (...) Now the worlds which extensionally
satisfy the imperative are just those of which the description is true., and Moutafakis
theorem T3 ([155] p. 155), which expresses the equivalence of the statements that an
imperative is satisfied and that a description of the prescribed action as performed is true.
12
These include Simon [196], who converts commands to declarative mode by removing
the imperative operators from them, obtaining a theory in which all recipients obey the
commands, and then applies the ordinary laws of logic to derive new relations that may
be converted back into commands. According to Niiniluoto [156], an imperative !p entails
imperative !q if p entails q. Very close to (DCM) is von Wrights account in [246] pp. 71,
164, where he defines the content of a prescription as the prescribed thing, and defines
that a command is entailed by a second command or by a set of commands if the content
LOGIC OF IMPERATIVES
13
LOGIC OF IMPERATIVES
14
Lemmon expresses reservations regarding his definition, but only because he thinks
that it does not sufficiently restrict imperative conclusions to statements about future
actions. A definition similar to Lemmons seems to be intended by Philipps [171] p. 364
who defines: to do p is forbidden! is true iff the indicative someone does p is incompatible
with the class of valid prescriptions, where compatible means that if the indicative is true,
then at least one prescription is violated.
LOGIC OF IMPERATIVES
15
1.4
!A
!B
is valid. Dubislavs trick can easily be applied to e.g. sentences of the form
I doubt that .... Then from I doubt that he is staying at his sisters place
in San Francisco follows I doubt that he is staying in San Francisco, which,
though we can derive he is staying in San Francisco from he is staying at
this sisters place in San Francisco, seems wrong: I might not doubt that he
is staying in San Francisco, but doubt very much that he is staying with his
sister. So why should Dubislavs trick work for imperatives if it would not
for other expressions?
LOGIC OF IMPERATIVES
1.4.1
16
Logic of Satisfaction
On one interpretation, which has been called the logic of satisfaction, the
scheme (ImpInf) is understood as stating that if the imperative sentence !A
is satisfied, then it must be that some other imperative sentence !B is also
satisfied. This interpretation is usually attributed to Hofstadter & McKinsey
[99], whose formalization of the scheme (ImpInf) would be !A >!B, which
is derivable in their axiom system whenever A B is classically derivable.
It is immediate that (ImpInf) must be valid on this interpretation whenever
the ordinary argument
A
B
is valid for the descriptive sentences A and B: If A classically implies B, then
it must be that if !A is satisfied, then A is true and also B is true and hence
!B is satisfied. Thus Dubislavs trick receives its semantic justification. But
if (ImpInf) is interpreted in this way, then it seems one should also accept
the following scheme:
!A
A
According to our (informal) convention, !A represents an imperative sentence
that is satisfied iff A is true. So it must be that if !A is satisfied, then A is
true and so the above scheme is valid. But on the look of it, this scheme
seems to state that from an imperative that demands A it can be inferred
that A is the case, which is nonsense. And this misunderstanding reveals
that when we spoke of the possibility of an inference in which the premisses
and the conclusion are imperatives, it seems that we talk about inferring an
imperative from some other imperatives, and not about reasoning whether
or not the imperatives in question are satisfied. So though the inferences
of a logic of satisfaction are valid in the interpretation in which they were
intended, thus interpreted inferences seem not to be what we want from a
logic of imperatives. For these reasons, Ross [188] p. 61 and also Hare [88]
doubted that a logic of satisfaction is what one has in mind in the case of
practical inferences.16
16
Kanger [119] p. 49 and Fllesdal & Hilpinen [53] p. 7 criticize Hofstadter & McKinsey
for making !A and A equivalent, which is somewhat unfair since the intended interpretation of their formulas (in terms of satisfaction) is not presented.
LOGIC OF IMPERATIVES
1.4.2
17
Logic of Existence
When we speak of inferring one imperative from some other imperative, this
could mean that the existence of an imperative is logically deduced from the
existence of some other imperative.
What is meant by saying that an imperative exists ? First, it could be the
existence of an utterance of some sentence in the imperative mood by some
commanding agent towards some commanded subject.17 Second, one might
demand that the utterance, as a performative use of the imperative sentence,
was effective and established an imperativum, i. e. a command, demand,
request or the like. For this it may be required that the commanding agent
had the will to command (and did not use the words for fun) as well as some
authority (power to punish or reward) over the addressee.18 Third, for an
order by legal authorities in this capacity to come into legal existence it
may be required that the authority was competent to utter it according to
the legal rules of some normative system that confers such competence, and
similar for bodies that are constituted not by law but by other rules like a
firm or Robins band.19
Yet however much the concept of existence is thus refined, it seems to
require the presence of actual facts: a (still alive?) speaker, a linguistic
entity like an utterance and circumstances of speaking, a certain attitude of
the speaker towards the act of speaking, a backing of the speaker by force or
an authority conferred by existing and/or valid rules, etc. But it is difficult
to see how logic can stipulate such an existence. This is illustrated in the
following example by Aleksander Peczenik:20
17
This existence is what Frey [58], along with an additional property of justification,
infers in imperative inferences: If the imperatives that appear in the premisses exist and
are justified, then also the imperatives derived from these exist and are justified (p. 465).
Freys justification means that what is demanded is good regarding some aim of the
commanding agent, called axiological validity in Ziembi
nski [260].
18
Cf. von Wright [246] p. 120-126. This is Ziembi
nskis [260] thetic validity. Lemmon
[133] seems to have this notion of validity or existence in mind when he demands that
the entailment of imperatives must be defined in terms of what imperatives are in force
at a given time.
19
Bulygin [39] uses the term systemic validity. According to Weinberger [228], [231] p.
259, this validity takes the place of truth as the hereditary trait (Erbeigenschaft) that is
transferred from the premisses to the normative conclusion in inferences with normative
sentences (Norms
atze).
20
Quotation from a letter by A. Peczenik to R. Walter, printed in [218] p. 395
LOGIC OF IMPERATIVES
18
Kenny [122] first pointed out that valid and invalid, interpreted as meaning commanded and not commanded, are not the analogues of true and false, but of stated
and not stated.
LOGIC OF IMPERATIVES
1.4.3
19
To get around this difficulty one may consider to interpret existence not
with respect to natural facts, but with respect to some ideal world of ought
or an assumed normative system that is closed under consequences, where
the closure operation may be understood e.g. in the sense of derivability by
use of Dubislavs convention. So if the agents use of the imperative mood
has resulted in the existence of a command in the world of ought, or, due
to the agents legal competence as e.g. a police officer, created an order
that now belongs to the normative system, then all the consequences of the
command that can be derived by an appropriate method exist in this world or
system as well.22 What thus has ideal existence is not the imperative sentence
as a spatial and temporal phenomenon or as a grammatically correct or
meaningful combination of words. Rather, it is what the use of an imperative
sentence expresses or accomplishes a command, request etc. Then it must
be that not only commands, requests etc. exist in this sense that in fact
have been expressed by a performative use of a sentence in the imperative
mood, but also some that only can be expressed. For if all that exists
in the normative system already exists as a result of a pragmatic use of
language, then there would be no need to let the normative system e.g.
be closed under a consequence operation. That what we can express by
using language (commands, requests, assertions etc.) has some existence,
22
The world of ought terminology originates with Walter [217], who is however following
Kelsen [121] p.195 in that an individual norm does not exist before the general norm
was applied by a judge, so orders that can only be deduced do not exist in Walters
world of ought. The idea to explain logical relations between norm sentences (like
imperative sentences) in terms of their existence in a system of norms that is closed
under consequences is that of Stenius [205]. In Opalek & Wolenski [165], norms are
non-linguistic entities expressed by (descriptively interpreted) deontic statements, and
normative systems consist not only of norms that have been expressed by a normative
authority, but also of the consequences of these basic obligations. In Alchourron &
Bulygins hyletic variant of a conception of norms [8], implicitly promulgated norms have
existence in a logically closed normative system, and descriptively interpreted (deontic)
norm propositions are then propositions about the existence of norms (in that system).
Holl
ander [100] promotes the idea of a deontically perfect world where norms exist that
obey logical principles, like that conflicts are excluded. Kelsen [121] pp. 187188 rejects
the idea of an ideal existence of norms because there is no ideal act of will that creates
them, and rejects the whole idea of a logic of norms.
LOGIC OF IMPERATIVES
20
x : Forest(x)
y : Tree(y)
The argument is analytical when the words forest and tree have their
usual meaning and for all that understand this meaning and thus know that
there cannot be forests without trees. By starting to talk about the (ideal)
existence of commands it seems that we silently changed (ImpInf) into
(2)
which appears confused. This is because the argument form is not used as
it is usually used, and now we do not know what to make of it. We are used
to filling in the blanks of the argument form
(3)
a
b
Cf. Stenius [205] according to whom all normative systems include a norm that
demands a tautology.
24
Note that the topic of this discussion has not suddenly become the ontological status of notoriously difficult concepts of practical philosophy and jurisprudence, like moral
obligations, natural law, human rights, laws of custom etc. Our concern are still ordinary
sentences in the imperative mood, addressed e.g. to a husband, secretary, student, child
or dog (cf. Ziemba [259] p. 386).
LOGIC OF IMPERATIVES
21
where a means a forest and b means a tree. At most, this is a mistaken way
to try to express (1). Similarly, the scheme (2) must be corrected into (5):
(5)
There exists a command given by to x by the use of the imperative sentence !A.
Therefore: There exists a command given by to x by the use of
the imperative sentence !B.
1.4.4
Formalistic Approaches
That the world of ought approach thus only provides arguments with descriptive
sentences is accepted by Walter [217], for he turns to identify imperatives with descriptions
of the world-of-ought-existence of a command that is created by the use of an imperative
consequently such sentences can be true or false and therefore part of logical inferences.
Thus imperative logic is reduced to indicative logic, where the difficult part is now the
verification of some descriptive sentences.
LOGIC OF IMPERATIVES
22
LOGIC OF IMPERATIVES
23
indicated: the pieces as propositions, the signs as not, or, and brackets, the initial positions as axiomatic basis, the game rules as the usual
rules of substitution and modus ponens, and the achievable game positions
as derivable formulas. This characterization of propositional logic is meant
by Dubislav as an exposition of Booles [34] idea that the validity of the
processes of analysis does not depend upon the interpretation of the symbols which are employed, but solely upon the laws of their combination. In
Dubislavs view, which he calls the formalistic theory, this description of
logic functions as a mould for all scientific theory: a theory is constituted
by a pure calculus (of formulas and rules), combined with a fixed interpretation. Observational sentences are captured in formulas that can be used
alongside axioms or derivable formulas of the system to derive other formulas
within the calculus. Then the assignment of these derived, or better: calculated formulas is reversed, i.e. they are translated back into observational
sentences. If these are regularly true, then the observational sentences are
explained by the theory. If a calculated observational sentence turns out
to be false, then the theory is erroneous. Thus it also becomes possible to
decide between competing, non-isomorphic theories.
The usefulness of the Dubislavs formalistic approach for the problem
of imperative logic is immediate. In fact, Dubislavs own proposal in [50]
satisfies all requirements in [49] for being a theory of imperative inference:
there are entities that may function as premisses and conclusions, namely
imperative sentences. There is an interpretation that assigns each imperative sentence a formula, namely that of the indicative parallel sentence in
the calculus of ordinary logic. There is a calculus, namely ordinary logic,
that tells us what formulas can be derived from the formulas assigned to the
imperative sentences that function as premisses. And finally, this assignment
is reversible to provide derived imperative sentences. Other authors taking
their cues from Tarskis [210] syntactical definition of consequence relations
and deductive systems,27 Tarskis [211] definition of truth,28 Gentzens [63]
27
Cf. Alchourr
on & Bulygin [8] who employ a formal consequence relation to explain
what norms are implicitly promulgated by a set of norms.
28
Both R
odig [187] and Yoshino [257] appeal to Tarski and argue that meaningful operations with prescriptions are made possible by supposing that normative attributes like
obligatory or punishable may be applied to actions. Rodig draws attention to the problem of objective verifiability and therefore truth of such statements. But he circumvents
the problem by assuming that meta-language truth conditions can be given, which is sufficient to handle normative attributes as normal predicates in the object language. Rodig
LOGIC OF IMPERATIVES
24
idea that to define a symbol is to give rules for its introduction and elimination,29 and Wittgensteins dictum that the meaning of a word is its use ([239]
43) have similarly argued that instead of searching in vain for analogues
of truth values, it suffices for an explanation of imperative inferences to give
formal rules for obtaining imperatives from other imperatives.
If this formalistic approach to the logic of imperatives is accepted, we are
still not finished yet. If the assignment of formulas, calculations and backtranslations of derived formulas are to be more than a game, there must be
some way to judge the adequateness of the theory, and be it only to decide
between competing proposals.30 In analogy to Dubislavs general approach,
where a theory is only an explanation of phenomena if its calculated observational sentences are regularly true, one should require of any proposed
logic of imperatives that the imperative it derives from other sentences are
normally not true of course, but accepted as implicit in other sentences
that are used as premisses. This resembles what is called the soundness of a
calculus: if the calculus allows false (unacceptable) conclusions to be drawn
from true (accepted) premisses, then it must be discarded as unsound.31
I now turn to the question of adequacy in this sense.
and Yoshino then use these predicates to formalize e.g. a norm that says that helping
in an emergency situation is obligatory as acts: In emergency(act) Helping(act)
Obligatory(act). The puzzling thing is that if this really is a prescription (norm), i.e. it
makes so far unregulated acts of helping in cases of emergency obligatory, then for no
such act the truth of the part Obligatory(act) can be established before the truth of the
whole is established. This at least differs from Tarskis compositional truth definition.
29
Cf. Alchourr
on & Martino [10] who provide a calculus with an introduction rule
for a prescriptively interpreted O-operator, where their rule corresponds to the modified
Dubislav convention (DCM) plus a requirement of joint satisfiability.
30
It seems consensus that there must be some test of adequacy. Weinberger [224] writes
that one must test a rule for the logical manipulation of norm sentences for its adequacy
for the area of normative thought, and Sosa [201] speaks of a control of commonsense
that is necessary because otherwise there would simply be no end to the possible logics.
31
The other possibility, that the calculus does not provide all the inferences from premisses that are acceptable (usually called completeness), is less harmful and can be dealt
with by e.g. refining it. For a similar definition of adequacy cf. Chellas [44] p. 4, where
however the terminology is vice versa.
LOGIC OF IMPERATIVES
1.5
25
Rosss Paradoxes
is a valid imperative inference. Ross himself points out that his paradox is
not paradoxical if this validity of an imperative inference is understood in
the sense of a logic of satisfaction. If the letter is posted and the imperative
!A satisfied, then the imperative !(A B) will likewise be satisfied this
is no more paradoxical than that A B can be inferred from A. But if
the meaning of imperative inference refers to anything like the validity or
existence of an imperative, then Ross claims that his inference is not only
not immediately felt to be evident, but rather evidently false.
Why does Rosss example of an imperative inference seem paradoxical?
In particular, regarding the formalistic theory of imperative inference given
in the last section, why should it be paradoxical to say that if one uses the
imperative !A for commanding, then one implicitly also commands !(AB)?
One explanation has been that that by using a disjunctive imperative, i.e.
an imperative sentence that like !(A B) is satisfied if some state of affairs or
some other state of affairs holds, the authority has left it to the subject how
to satisfy her command. Suppose Romeo hands a letter to Mercutio with
the words Post the letter or burn it, but relieve me from deciding its fate
LOGIC OF IMPERATIVES
26
and mine, would his friend not be free to do as he pleases? Analyzing this
freedom, it has been argued that giving a command entails an imperative
permission or implicitly authorizes to carry out the actions required to satisfy
the command.32 So the imperative post the letter or burn it would contain
the permission I hereby permit you to post the letter or burn it. Now
explicit disjunctive permissions are often understood in a strong sense that
grants both disjuncts: when someone says help yourself to a cup of coffee or
a cup of tea, then the guest is permitted to help herself to coffee and also
permitted to help herself to tea (though possibly not both). So one obtains
the following chain:
(2)
The reason why the inference from the first line of (3) to the second line seems
not objectionable, while the similar inference from the second line in (2) to
its third line appears somehow wrong, may lie in the fact that the imperative
to post the letter or burn it that is used in the reasoning is only implicit,
i.e. derived, while Mercutios reasoning was about an imperative that was
explicitly used by Romeo. So one could modify ones view on the second
step in (2) by saying that one is only allowed to infer a strong permission
to do what is commanded if this command is not itself derived. I return to
32
Cf. Chellas [44] p. 19 for the term imperative permission and Keene [120] for the
implicit authorization.
33
The idea to explain the counterintuitive nature of Rosss paradox using the also, or
even more, counterintuitive inference to you may post the letter or burn it was von
Wrights in [249] pp. 2122, also cf. von Wright [256] pp. 121122 and Hintikka [96].
LOGIC OF IMPERATIVES
27
According to Hare [88] it is just a conversational implicature that gets canceled. But
it seems that by saying go via Berwick or Coldstream the authority really leaves it to
the agent which route she wants to take and later retracts this choice , while someone
who says e.g. the tickets are upstairs or in the car, and later adds they are not in the
car only made it seem as if the tickets could be in either location. If the order was only
given further orders pending, as Hare also argues, then the first order was not complete,
because it left the agent unable to determine how to fulfill it. It is as if the authority had
said in the middle of a sentence: hang on, Im not finished yet, Ill be right back.
35
This was the point in Williamss [238] criticism of Hares [88] scheme.
36
Cf. Stenius [206]: Free choice permission is too strong a concept to be useful.
LOGIC OF IMPERATIVES
28
It seems the kind of deliberation that one would expect of reasonable agents.
Likewise, Mercutio, having been asked by Romeo to post the letter or burn
it, might be found reasoning in the following way:
(6)
One might dispute whether Mercutios fear is really on a par with a road
blocked e.g. by a landslide. But if we suppose it is, then Mercutios reasoning
seems as impeccable as that of the officer. Now return to Rosss paradox:
here the agent was ordered to post the letter. Implicit in this imperative, so
we are told by Dubislavs convention (DC), is the imperative post the letter
or burn it. Imagine that the agent is not able to post the letter for some
cause (the postal workers are on strike and the mail bins have been locked
up). So the agent could reason in the following way:
(7)
But this reasoning is absurd. Just because the agent cannot fulfill her obligation to post the letter, this does not mean that she is obliged to do something
that was never mentioned, and in fact could be anything: the words burn
the letter could be replaced e.g. by go to the zoo, kill a passer-by or love
your neighbor and the inference would be just as valid if it is valid.37
Now the agent, in reasoning in the above settings, used indicative statements about natural facts like that something cannot be done to reason
37
This is Weinbergers [225], [226] explanation of why Rosss paradox poses a problem.
LOGIC OF IMPERATIVES
29
about the imperatives go via Berwick or via Coldstream, post the letter
or post the letter or burn it. But inferences that mix imperatives and indicatives are notoriously troublesome and should perhaps be avoided. As
MacKay [139] points out, both of the following inferences
Go fly a kite!
You are going to drop dead.
Therefore: Drop dead!
You are going to fly a kite.
Drop dead!
Therefore: Go fly a kite!
are validated by Dubislavs extended convention (DEC), where both inferences seem plainly invalid, and so perhaps (DEC) should not be accepted.
Yet consider again the case of the officer. Imagine that it was not the original
authority that issued the command not to go via Coldstream, but someone
else, like the officers husband (who in the past had some bad experience on
this road). Since there is some discretion in the authoritys order, there is
no reason why the officer should not give in to her husbands request, and so
the following reasoning of the officer seems correct:
(8)
But then the following reasoning of the agent to whom Rosss imperative
post the letter was addressed must be likewise correct:
(10)
There is some discretion in the (implicit) order to post the letter or burn it,
LOGIC OF IMPERATIVES
30
so why should the agent not take an additional request into account? The
only difference between the (9) and (10) is that in (9) the reasoning appeals
to an explicitly used imperative, whereas in (10) it starts by considering an
order that was only implicit in the use of some imperative. So maybe what
was wrong was that derived imperatives were used, without paying enough
attention to the fact that the derived imperatives are only part of a system, that the explicitly used imperatives have not ceased to exist, that
imperatives that are only derived do not exist on quite the same level as
explicit imperatives, or that the agent is somehow expected to make use of
the logically strongest information that is available.38 So we are back at the
proposal that a difference must be made between explicitly used imperatives, and imperatives that only derive from explicit imperatives. But to
require that reasoning with imperatives starts with explicit imperatives, and
must not start with imperatives that are only inferred, reveals an unusual,
non-classical meaning of imperative inference. For classically, logical inferences may very well be conducted by proving first that some assumptions
have some desired conclusion, and then show that the assumptions follow
from an accepted set of premisses. This is facilitated by the transitivity of
classical consequence: if A Cn(B) and B Cn(C) then A Cn(C) (consequences of the consequences are also consequences), or the monotonicity
rule: if A Cn(X) then A Cn(X Y ) (what follows from some axioms
also follows from a larger set of axioms).
Rosss paradox seems to demonstrate that given the imperative inferences
provided e.g. by Dubislavs convention, it becomes necessary to distinguish
between the imperatives that are explicitly given and the imperatives that are
inferred: agents can use the former for their reasoning, but not always the latter, or not the latter by themselves, which makes reasoning with imperatives
somehow non-classical. And so there may yet be another way to get around
the difficulties: perhaps Rosss example is not really a case of an imperative
38
R
odig ([187] p. 184185) points out that by deriving the norm to post the letter or
burn it, the original order to post the letter does not cease to exist, and that it is the
conjunction of both norms that must be satisfied. That the entailed norms do not exist
in quite the same way as explicit norms is the idea of von Wright e.g. in [255] and [256] p.
114 and p. 122. According to Stenius [206], the use of post the letter or burn it carries
the tacit information that a stronger regulation like post the letter does not belong to
the codex. For the idea that using a weaker sentence post the letter or burn it violates
a conversational presupposition cf. Hintikka [96]. Also cf. Hamblin [74] p. 88: implicit
imperatives may be different from the real thing, and we should be wary of loading them
up with the full range of imperative properties.
LOGIC OF IMPERATIVES
31
R-consequence is defined by Weingartner & Schurzs [237], [236] in the following way:
a propositional formula A is a R-consequence of a set of formulas X iff (i) X classically
implies A and (ii) it is not possible to uniformly replace a proposition letter at at least
one of its occurrences in A by a random proposition letter without making the classical
inference invalid.
40
The origin of the example is unclear. The name Paradox of the Window is used e.g.
by Stranzinger [207] and Weinberger [232].
LOGIC OF IMPERATIVES
(11)
(12)
32
Suppose that wants x to practice the piano, but neighbors have already
complained about the disturbance and even called the police on a previous
occasion. So does not want x to play the piano while the window is open.
Closing the window will reduce the noise so much that the neighbors are left
with nothing to complain about. Suppose then that sends x to play the
piano, using the words close the window and play the piano. A little bit
later, the following discussion ensues between and :
:
:
:
I told her to play the piano, but I didnt hear her doing it all
afternoon.
Well, at least she closed the window.
Why should she do that?
You ordered her to close the window, thats what she did, so
she did something right, didnt she?
You ordered her to play the piano, thats what she did, so
dont try to wiggle out of your responsibilities.
LOGIC OF IMPERATIVES
33
It gives the paradox a further twist if we imagine that playing the piano with the
window open is explicitly forbidden. For by Dubislavs convention (DC), the imperative
!(A B) (dont play the piano while the window is open), is derivable from the
imperative !(A B) (close the window and play the piano). But it seems that the
additional prohibition is best formalized as a conditional imperative (in Hofstadter &
McKinseys [99] formalism: B !A). Conditional imperatives pose other problems
outside the current topic. In any case, one would still have to say that playing the piano
with the window not closed was satisfactory with regard to some (derived) imperative.
42
Cf. Hare [88], Weinberger [223] and [235]. These difficulties led Weinberger to reject
the validity of an inference from !(A B) to !A in his publications since [223].
LOGIC OF IMPERATIVES
34
The inference is clearly absurd and so Kennys logic does not help us to solve
the paradoxes.
In Rosss first paradox, the imperative to post the letter or burn it was
inferred from the imperative to post the letter, thus forcing one to acknowledge that some (though only inferred) imperative is satisfied by burning the
letter. In the window paradox we could infer the imperative play the piano from the imperative close the window and play the piano, thus forcing
us to acknowledge that an (inferred) imperative is satisfied when the piano
is played with the window wide open. In both cases, we would much rather
say that no imperative was satisfied by burning the letter that was meant to
be posted, and by playing the piano with the window open when it should
have been closed. This, I think, is the main cause why Rosss paradox and
the window paradox give rise to counterintuitive feelings, or are paradoxi43
44
LOGIC OF IMPERATIVES
35
1.6
Maybe it is not really the case that all options have run out to redefine
Dubislavs scheme in a way so that it avoids Rosss paradoxes. Maybe we
have to replace the classical logic that appears in his scheme by yet another
logic, or develop such a logic.45 But it is hard to see what kind of logic this
could be, since most logics, including other non-monotonic logics, will permit
us to either infer !(A B) from !A or !A from !(A B), and so at least one of
the two paradoxes will arise. So I think, after all these troublesome attempts
to define a logic of imperatives, it is worthwhile to take another look at
Poincares proposal that originally started the controversy.
Poincares only explicit example of an inference with an imperative conclusion has the following form:
(1)
Do this!
This cannot be done without that.
Therefore: Do that!
Cf. Keene [120]: What we wanted here is a logic of actions, in which a well-defined
concept of inclusion plays a leading role.
LOGIC OF IMPERATIVES
36
But here the two sentences that function as premisses are both descriptive.
Since Poincare explained that an imperative cannot be derived from indicative premisses alone (and there is no reason not to follow him), this cannot
be an imperative inference, and there must be something more involved than
the drawing of a logical conclusion. One such other function of the therefore appearing at the front of the last sentence of (3) is not to reason, but
to motivate, as in:
(4)
Here the speaker motivates the imperative to take the bus by explaining that
driving into town is impossible, since the car is broken. So similarly, what
seems to happen in (3) is that I motivate my (new) imperative drive me in
a northerly direction by an already given command and an assumed fact.
Consider again the proposed inference (2). Just like indicative inferences
are explained by the fact that someone who accepts (or: assents to) the
premisses must also accept the conclusion, Hare [87] has argued that an
imperative inference is one where someone who assents to all imperative
premisses must also assent to the imperative conclusion:
A sentence p entails a sentence q if and only if the fact that a
LOGIC OF IMPERATIVES
37
LOGIC OF IMPERATIVES
38
the taxi driver to drive me north? I might be absolutely sure that the airport is to the north, but I would still blame the taxi driver for not going
where I requested if, opposite to what I believed, the airport is in fact to
the south-west of my starting point and the driver still went north. So I did
not utter such a request, and would not even imply such a request, lest I be
charged by the driver for going there instead of the airport. All we can say
is that the taxi driver would also be satisfying, and so seemingly assenting
to, a purely hypothetical request to drive me in a northerly direction, if she
satisfies the request to drive me to the airport and the airport does in fact
lie in a northerly direction. But this is again not a logic that infers one imperative from some set of other imperatives and/or indicatives, but the logic
of satisfaction as explained in sec. 1.4.1.46
It would be nice to have real life examples, cases of ordinary reasoning
with imperative premisses and an imperative conclusion, i.e. instances of
(ImpInf)
!A
!B
where !A and !B are sentences in the imperative mood, and where the use
of the inference not the imperatives is either accepted in some ordinary
discourse, or opposed (and the person who uses it blamed for being unreasonable or illogical).
Use of indicative arguments in everyday discourse often occurs in singular
sentences, like
(5)
(6)
(7)
Here two descriptive sentences are linked with the adverb so (similar adverbs
would be therefore or hence). (5) seems analytical if one understands
enough to be elliptical for enough to make up job-losses elsewhere. (6)
is analytical if one knows that A is a top-grade and that English is one
of several high-school subjects. (7) is made into a logical argument by the
assumed background knowledge that Pnin is a novel by Nabokov. It is
often not easy to distinguish such indicative arguments from sentences that
present reasons, motives or are otherwise explanatory, for these also use the
46
LOGIC OF IMPERATIVES
39
(8) explains why today the speaker used the bus. Since the bus need not
have been the only means to get into town, or the speaker may have stayed
at home, the hearer cannot just conclude the second part from the first. (9)
presents the psychological motive why the speaker asked the other person to
go shopping with her. Other people might have been motivated differently
by the desire to make friends with that person. In (10), a natural event is
explained by a certain state of affairs. Again, this is not a logical argument:
the birds could also have not flown in, or flown in but not nested in the
ceiling. Now the adverbs so andtherefore can also be used to meaningfully
link imperatives. Consider the following examples:
(11)
(12)
(13)
(14)
LOGIC OF IMPERATIVES
40
names, or serve them plenty of alcohol, or maybe the guests are easygoing
and do not really require any effort on the hosts part to make themselves
at home. Similarly, in (13) the more readily accepted advice to keep vermin
out of the house is used as a rationale to make the addressee accept the
drudgery of having to patch up the roof. The most promising candidate
for an appeal to analyticity seems to be (14), i.e. that the imperative to
read all of Nabokovs novels includes the imperative to read Pnin, given the
background knowledge that Pnin is a novel by Nabokov. Note that when
making the background knowledge explicit, it becomes a case of Dubislavs
extended convention (DEC). Such a sentence may be used e.g. by a teacher
of a literature course when addressing her students. But again we cannot
rule out that this is just a case of complementing an imperative by a second,
more specific one, as we sometimes do to get things done.
Adherents of Dubislavs convention (DC) must also accept the following
argument:
(15)
But it seems dubious what reason the speaker could have for adding the so
part. Just aiming for top-marks in some subject area is clearly not what the
speaker wants the addressee to do. More meaningful would be the converse,
(15a)
Aim for top marks in at least one subject area, so aim for an A in
English!
where the advice to aim for A in English is rationalized by the wish to have
the student achieve top-marks somewhere. But since the student could not
know from the first imperative that it was the subject of English that the
speaker wanted her to achieve top marks in, this would like (12) and (13)
rather be a motivating so, and not a use of so that appeals to a logical
capability.
Matters are further complicated by the fact that expressions of the following kind can also be meaningfully employed:
(16)
(17)
(18)
LOGIC OF IMPERATIVES
41
In all three sentences, the first part is descriptive and the second is in the
imperative mood. We have already noted in the case of (3) that such arguments exist, but for anyone who agrees to Poincares thesis that imperative
conclusions do not follow from an indicative premisses it is clear that (16)
(18) cannot represent valid arguments. (16) seems again a case where the
so is used to motivate the advice that is expressed by the imperative. The
so does not express a logical relation, for sometimes it is better to use a car
that stutters than a coach that wont take one back. In (17) the indicative
gives a reason why the speaker wants her request to be followed. According to Hamblin [74], such reason-providing indicatives are often attached to
advice-expressing imperatives, yet here the imperative might also be an order
(e.g. of a parent). For the same reason the speaker might have ordered the
agent to hand over her key, and not to leave it under the mat, and so what
is expressed is again not a logical relation. (18) seems also like presenting
a motive for inviting Gill to the party (she is the addressees best friend),
but here things might be a bit more complicated the expression could be
elliptical for:
(18.a) Invite your best friends to the party, Gill is your best friend, so
invite her to your party!
This is very similar to what Dubislav considered a valid argument, namely
his inference from thou shalt not kill to Cain shall not kill Abel. But then,
(18) might also be elliptical for
(18.b) Gill is your best friend, one invites ones best friends to ones parties,
so invite her to your party!
where the second part (which is not in the imperative mood) appeals to the
existence of a rule that the speaker might consider binding, or binding for the
addressee. Then this is rather a case of reason-giving, and not of a logical
inference: the speaker motivates her imperative by asking the speaker to
conform to some preexisting rule.
To tell the uses of therefores and sos that are motivating, reasongiving or explanatory in a non-logical sense, apart from those that separate
the premisses from the conclusion in an argument that is intended to be a
logical one, we can use the following trick: instead of therefore or so, use a
clause like ... It follows logically from this that ... to separate the sentences.
LOGIC OF IMPERATIVES
42
The new phrase makes the appeal to a logical capability explicit. Where the
original adverbs so and therefore were used to indicate a (claimed) logical
inference, the new formulations
(5.a)
(6.a)
(7.a)
I couldnt get the car started. It follows logically from this that I
took the bus.
(9.a) I wanted to make friends with her. It follows logically from this
that I asked her if she would go shopping with me.
(10.a) There were holes in the roof. It follows logically from this that birds
had come in and were roosting in the rafters.
The phrase it follows logically from this makes again an appeal to some
shared understanding of used words, concepts and background. But here,
this background knowledged obviously does not allow one to conclude the
second sentence from the first. The listener could not have known from the
first sentences in these examples that the speaker took the bus, asked someone
to go out shopping or has birds nesting in the roof of her house. So claiming,
as the rephrased sentences do, that the second part can be concluded from the
first, makes the sentences seem irritating, weird and false, while the earlier
LOGIC OF IMPERATIVES
43
LOGIC OF IMPERATIVES
44
John, read all of Nabokovs novels. So you can conclude for yourself:
John, read Pnin!
It seems that in (14.f) and (14.g) the speaker has not just asked the addressee
of the first command to draw a conclusion, but in this process to give
himself the command expressed by the second sentence, i.e. to tell himself to
LOGIC OF IMPERATIVES
45
read Pnin. When the addressee is made explicit in the inferred command,
it looks as if the addressee is additionally asked to use his own first name
when telling himself to read Pnin which is a weird thing to ask of anybody.
And this points at another problem of (14.f) and (14.g): if the person who
commands read all of Nabokovs novels (the teacher) and the person who
commands read Pnin (John himself) are not identical, how can the second
imperative be inferred from the first?
By contrast, all of the above phrases can be employed for deontic sentences (non-imperative sentences that do not prescribe, but describe what
ought to be done) without difficulty:
(19)
(19.a)
(19.b)
(19.c)
(19.d)
(19.e)
(19.f)
LOGIC OF IMPERATIVES
46
of the adverb so suffices to explain why the sentence (14) seemed meaningful: the teacher, perhaps asked by John whether he also has to read Pnin,
motivates the more specific imperative to read this book by prefixing to it
the general requirement to read all of Nabokovs novels, thus making it clear
that Pnin is in fact one of the books that John has to read. (14.f) appears
comparatively less strange than the other reformulations because to ask John
to give himself the imperative to read Pnin may be a (roundabout) way
to make sure he actually reads it. To understand (18) we do not need to
determine whether the speaker refers to an explicit command to invite ones
friends, or a social custom to do so, because what is in any case implicit in
(18) is an appeal to a preexisting obligation to motivate the agent to do what
the speaker wants her to do. It also explains why (15) seemed so strangely
pointless: the reason for using the less specific imperative to achieve some
top marks is not sufficiently explained by prefixing to it a more specific imperative to achieve top marks in English.47 And so it seems that all of the
imperative arguments (11)(18) are really cases of reason-giving and motivation, and the sos and therefores used in these expressions that like
Poincares donc, or the alsos, daher s and deshalbs of German language,
may be used to connect both indicative and imperative sentences, provide
only reasons, explanations or motives in the case of imperatives, and do not
indicate claims of analyticity.
So I want to dare the hypothesis that there are no examples of imperative
inferences, i.e. logical conclusions in the imperative mood, drawn from at
least one premiss in the imperative mood, to be found in ordinary language
arguments. They only appear in the writings of some philosophers.
1.7
Other authors have noted before the conspicuous absence of imperative arguments from natural language. Wedeking [221] argued that there are no
cases in which we actually use commands in arguments, and that words like
therefore before imperative sentences are employed not to mark inferences,
but for the purpose of reason-giving, of motivating the subject. Harrison
[90] argued from a point of grammar and semantics that there is no logic
47
Note that the same strangeness does not necessarily arise for deontic logic. The dean
of one faculty may say to another: Our students are obliged to have an A in English,
so yes, ours are like yours obliged to achieve top marks in at least one subject are.
LOGIC OF IMPERATIVES
47
of imperatives; the difference between his position and mine (that there are
no imperative inferences in ordinary language, and so a logic of imperatives has no point) seems very subtle.48 Even more conspicuously, there have
been little challenges to these arguments.49 On the contrary, there is a whole
tradition of normological or imperativological skepticism, of authors who
have denied the existence of a logic of norms or imperatives.50
If there are, as a matter of fact, in ordinary language, no argument forms
that resemble imperative inferences, then there also is no place for a formal
theory for such a logic. Presenting formalizations of such a logic would be
writing about what Dubislav [49] called an Unding or chimaera: a non-thing
that exists only as a concept, but no real object falls under the concept.
So did Poincare commit a mistake? Did he confuse an important insight
by Hume [107] on the use of is and ought that facts cannot be used to
48
Harrison writes (pp. 110/111, 124/125): The expression I conclude: Shut the door
does not make sense, nor does the expression I conclude that dont. One can say: So
shut the door and Therefore shut the door and Shut the door because ... but the
function of the words so, therefore and because is not in this context to indicate that
Dont shut the door is a conclusion. They have some other function, which philosophers
have confused with that of indicating that what follows them is a conclusion. (...) The
reason, therefore, why therefore and so can precede post the letter, but I conclude can
not, is that I conclude can precede only propositions (and only they can be conclusions)
and indicate that reasons have been given for the proposition, but therefore and so can
precede either propositions or imperatives. When so and therefore precede imperatives,
however, they are not reasons for the imperative, as the unwary might suppose, but for
the action enjoined, advised, recommended or directed by it. Harrison concludes (p.
81): There is no such thing as imperative logic (...) There are indeed logical relations
between one imperative and another, but this simply supports a logic in which the premises
and conclusions are indicative statements about imperatives (p. 81, my emphasis). The
closeness of Harrisons account to mine was pointed out to me by Lou Goble.
49
Casta
neda [43], replying to Wedeking, grants that differences in the meaning of inferential words in indicative and imperative inferences may exist, but argues that they do
not prohibit a concept of imperative inferences in parallel to indicative ones. This seems to
miss the point, it echoes Dubislavs convention without explaining what such formalisms
are to formalize. I owe the reference to Wedeking and Casta
neda to Peter Vranas.
50
Such authors include G. H. von Wright who writes in [256] p. 109: And now I
too, after a long and winding itinerary have come to the same view: logical relations,
e.g. of contradiction and entailment, cannot exist between (genuine) norms.. Above
we have already noted that Hamblin [74] p.89, Sellars [193] p. 239-240, von Kutschera
[128] and Philipp [168], [169] have expressed scepticism or denied the possibility of a
logic of imperatives altogether. For imperatives also cf. Moritz [153], Williams [238],
Keene [120], Opalek & Wole
nski [164]. The term is coined from Weinbergers [230] term
normological skepticism which denies logical relations not only between imperatives, but
any prescriptive language. The main proponent of normological scepticism is Kelsen [121].
LOGIC OF IMPERATIVES
48
argue that they must be so or that other facts should be made similar to them
with a statement about grammar? Curiously, in his essay, Poincare [172]
never claimed to have discovered the logic of imperatives of which he was
heralded as the pioneer. His main argument is that findings of science can
influence moral reasoning. He just presumes that, like scientific arguments
consist of sentences in the indicative mood, moral reasoning is conducted
using sentences in the imperative mood. It is true that facts can influence
the reasoning of agents about their obligations: Hares officer, who upon
being commanded to go to Edinburgh via Coldstream or Berwick finds the
road via Coldstream blocked, acts quite reasonably by concluding that she
now ought to go via Berwick. But this is a reasoning about what obligations
she has, it is a deontic argument, and not a case of inferring imperatives. So
Poincares main argument is correct, but the assumed parallelism between
sentences in indicative and imperative mood, that they can both feature in
logical arguments, does not exist. Our language does not work that way.
There are several ways to go forward from a position of imperativological
skepticism. First, one might continue the logic of imperatives as a logic of
satisfaction. The logic of satisfaction states which imperatives must also be
satisfied if some other imperatives are satisfied, and it may also be used to
state which imperatives will be violated by satisfying other imperatives. We
can use the notion of satisfaction to distinguish imperatives that might be
seen as redundant in a set of imperatives in the sense that these will also be
satisfied if some other, different imperatives are satisfied, or identify subsets
of imperatives that cannot be all satisfied and so conflict. By providing these
concepts, the logic of satisfaction, though it may appear trivial, remains a
meaningful and correct way to talk about imperatives.51
Second, imperatives normally express the wish or desire on the part of
the person or authority using the imperative that what is commanded is
satisfied. But it seems unreasonable to wish for A to be realized, but also
for A to be realized, and in this sense two wishes may exclude another. If
imperatives express wishes of one particular person, we can then point out
to her what wishes may be unreasonable. Likewise it might be desirable to
view the norms of a particular society as if they all were the wishes of one
person, the law giver, and logic may then give advice as to which norms
must be revised so that the system is reasonable. This is the position of G.
51
Cf. C. G. Hempels [91] remark with regard to Rosss Paradox that a logic of satisfaction should not be so easily rejected.
LOGIC OF IMPERATIVES
49
H. von Wright in his late work on normative logic, cf. e.g. [255], [256].52
Finally, there is deontic logic. Deontic logic uses the modal expressions it
is obligatory that, it is permitted that, it is prohibited that etc. to describe
what ought to be, is permitted or is prohibited according to given imperatives
or norms. Deontic logic has been disparagingly called a kind of ersatz truth,
that merely mirrors logical relations that already exist between imperatives
or norms, and so we should rather look for this logic than studying a deontic
logic that only reflects it and so must result in a dull isomorphism.53 But it
has been the logic of imperatives that has kept escaping us, while sentences
that use deontic expressions can easily be used to form valid arguments. So
maybe it is not a logic of imperatives that is the proper subject of study
and makes deontic logic just an ersatz theory, but it is the other way round,54
and the idea of a logic of imperatives has been a fata morgana, leading us to
ever more futile attempts to explain inference relations between imperatives,
to find analogues of truth values, or new logics to explain Dubislavs scheme,
whereas any plausibility of this idea was just a reflection of the real, but quite
distinct possibility of a logic about imperatives, namely of deontic logic.55 If
my hypothesis holds, then it is the only logic regarding normative concepts
such as obligation that we should be concerned with. So it is to this logic
that I now turn.
52
The idea that commands can be identified with wishes, which in the above sense
relate to each other, goes back to Bentham [29] pp. 9597. Note that this does not force
one to acknowledge that there is a logic of commands. There is a difference between a
theoretical inconsistency and a practical inconsistency, or: what is inconsistent, and what
is inconsistent to say, are two entirely different things (Harrison [90] p. 95).
53
This is Hares view in [88] p. 325; also cf. Alchourron [2] pp. 264266; Kalinowski
[112] p. 134; Weinberger [230] p. 58, [232]; Wagner & Haag [216] p. 102. The idea that
deontic logic reflects the logical properties of norms is that of von Wright in [246] p. 134.
54
Cf. Alchourr
on & Bulygin [7] p. 463: This logic of norms is, so to say, a reflection
of the logic of normative propositions. It is because we regard as inconsistent a system in
which it is true that Ox p and Ox p, that we say that the norms !p and !p are incompatible.
So it is the logic of norm propositions which yields the foundations for the logic of norms..
55
Thus Hares [87] proposed inference of Take this box to the station from the imperative Take all the boxes to the station and the statement This is one of the boxes
reflects a deontic deliberation of the agent: I have been ordered to take all the boxes to
the station, this is one of the boxes, so I must take this box to the station. Casta
neda [43]
argues in favor of imperative inferences that we can certainly say: John, go home follows
from, and was inferred by Smith from, John, go home and study. While logicians use
sentences of the form: ... follows from ..., with all kinds of expressions filling the gaps,
it seems to me that the underlying natural inference is again deontic: Smith concludes
from Johns having been ordered to go home and study that John ought to go home.
Chapter 2
Deontic Logic
2.1
Deontic logic is a branch of modal logic that is concerned with the logic of natural language expressions such as it is obligatory that ..., it is permitted
that ..., it is forbidden that .... It seems that Gottfried Wilhelm Leibniz
(16461716) was the first to explain in detail that these legal modalities
appear to be related in a similar way to each other as the alethic modalities
that describe something as necessary, possible and impossible. Leibniz, in an
unpublished manuscript dating from 1671 for his book Elementa Juris Naturalis, proposes that what is obligatory (debitum) is what is necessary for a
good man to do, what is permitted (licitum) is what is possible for a good
man to do, and what is forbidden (injustum, illicitum) is what is impossible
for a good man to do. Leibniz lets a fourth legal modality, that of indifference, correspond to contingencies as contingent is what is possible but
not necessary, so what is indifferent is what is permitted but not obligatory.
Thus Leibniz seeks to transfer all the complications, transpositions and oppositions of the modalities from Aristotles De Interpretatione to the iuris
modalia ([131] pp. 480-481).1
While it would be unfair to say that this analogy between legal concepts
and their natural or alethic counterparts became forgotten among logical
1
As Knuuttila [124] explained, the idea that a good man cannot violate the law can be
traced through the middle ages back to Aristotles Nicomachean Ethics EN 1100b 33-35,
where he states that no happy man can become miserable, for he will never do the acts
that are hateful and mean.
50
DEONTIC LOGIC
51
alethic
necessary
possible
contingent
impossible
Modal Categories
epistemic
deontic
verified
obligatory
unfalsified
permitted
undecided
indifferent
falsified
forbidden
existential
universal
existing
partial
empty
contraries
implies
oppositions
implies
It is possible that A
=It is not necessary that not A
=It is not impossible that A
2
Watts [220] mentions the analogy several times, cf. part II ch. 2 sect. iv. As Hruschka
[106] points out, Gottfried Achenwall [1] 12, 44 uses language very similar to Leibniz
when distinguishing between the actio moraliter possibilis, actio moraliter impossibilis and
the actio moraliter necessaria, identifying the necessitas moralis with obligatio. Bentham
[29] p. 97 constructs a deontic square similar to the one below, and shortly before the
advent of axiomatized deontic logic, Grelling [73] suggests the construction of a formal
syntax of ought-sentences after the image of modal logic.
3
Cf. von Wright [242] p. 2. The changes to von Wrights original table are due to
Anderson and Moore [18] who added the quantificational concept of partial (for some
but not for all), and the epistemic concept of unfalsified (verified or undecided).
DEONTIC LOGIC
52
contraries
implies
oppositions
implies
It is permitted that A
=It is not obligatory that not A
=It is not forbidden that A
subcontraries
DEONTIC LOGIC
2.2
53
Not counting an earlier attempt by Ernst Mally [145],4 the first successful
axiomatization of the relations between deontic expressions was presented
by Georg Henrik von Wright in his 1951 article Deontic Logic [242]. Von
Wright employs three modal operators O, P and F , where OA means
that it is obligatory that A, P A means that it is permitted that A and F A
means that it is forbidden that A. The three operators are cross-defined by
letting O be primitive, and defining P A as OA and F A as OA, so
P A means that it is not obligatory that not A, and F A means that it is
obligatory that not A.5 His axiomatization, called the classical system of
deontic logic [251], can be characterized by the following axioms and rules:
(Ext) If A B is a theorem, then OA OB is a theorem.
(M)
O(A B) (OA OB)
(C)
(OA OB) O(A B)
(D)
OA P A
Here, the principle of extensionality (Ext) expresses that if A and B are
equivalent, then so are the sentences OA and OB. The principle of monotony
(M) expresses that if it is obligatory that both A and B, then it is also
obligatory that A and obligatory that B, and the principle of cumulativity
(C) expresses vice versa that if it is obligatory that A and obligatory that
B, then it is also obligatory that both A and B. Together, (M) and (C) are
equivalent to von Wrights principle of distribution (P A P B) P (A B),
meaning that if a disjunction of acts is permitted, then one of the disjuncts
is permitted, and vice versa. (D) is the characteristic deontic axiom that
expresses that if it is obligatory that A then it is permitted that A; it is the
equivalent of von Wrights principle of permission P A P A that expresses
that any act is either permitted or its negation permitted, and it is also due
to the definition of P A as OA equivalent to (OA OA). It is also
crucial for obtaining the modal relations expressed by the deontic square:6
4
that A is the case, and vice versa, or, in Mallys words, the Aquivalenz
von tats
achlich
Gefordertem und Tats
achlichem (the equivalence between the factually obligatory and the
facts) cf. [145] p. 269. Mallys error lies in his missing distinction between logical and
material implication, cf. Wole
nski [241], Goble & Lokhorst [70].
5
Von Wright originally used P as primitive and equivalently defined OA as P A,
and F A as P A, but the use of O as primitive became standard.
6
Since (D) is missing from Oskar Beckers independent publication [28], von Wright is
DEONTIC LOGIC
54
OA,
P A, F A
(D)
OA,
P A, F A
(D)
(D)
(D)
P A,
OA, F A
P A,
OA, F A
If one adds to von Wrights classical system the additional axiom (N)7
(N)
O(A A)
one obtains the equivalent of the so called standard deontic logic system
SDL, which is more rigorously formalized as follows:8
Definition 2.1 (The Language of Deontic Logic) The alphabet of the
language of deontic logic LDL is like that of propositional logic LPL except
that it additionally contains the deontic operator symbol O. LDL is defined
by the same rules as the set LPL plus the following: if A LDL , then also
OA LDL . P A abbreviates OA, and F A abbreviates OA.
Definition 2.2 (The System SDL) SDL is the smallest set that contains
all LDL -instances of tautologies and the axiom schemes (M), (C), (D) and
(N), and is closed under the rules of (Ext) and modus ponens.
rightly acclaimed to be the founder of modern deontic logic.
7
(N) makes all tautologies obligatory. While von Wright [242] p. 11 rejected it, its
philosophical import seems rather small: if something is obligatory, e.g. if OB is true,
then the truth of O(A A) can be derived by means of (C) and (Ext).
8
The deontic systems studied here will be based on propositional logic PL. The alphabet of its language LPL consists of a set of proposition letters Prop = {p1 , p2 , ...}, Boolean
connectives , , , and brackets (, ). LPL is then the smallest set such that
(a) any proposition letter p1 , p2 , ... is in the set, (b) if A, B are in the set, then so are
A, (A B), (A B), (A B). Semantically, valuation functions v : Prop {1, 0} define
the truth of a sentence A LPL (written v |=PL A) as follows: (a) v |=PL pi if and only
if (iff) v(pi ) = 1, (b) v |=PL A iff it is not the case that v |=PL A, (c) v |=PL A B iff
v |=PL A and v |=PL B, (d ) v |=PL A B iff v |=PL A or v |=PL B, (e) v |=PL A B iff
v |=PL A or v |=PL B. If all valuations that make true all sentences of a set of sentences
also make true A, then implies A (written A). If A is true under all valuations v,
A is called a tautology, if it is false under all valuations, it is called a contradiction. From
the set PL of all tautologies, the notions of provability, consistency and derivability (I
write PL A) are defined as usual. denotes an arbitrary tautology, and an arbitrary
contradiction.
DEONTIC LOGIC
55
2.3
Andersons Reduction
Independently from each other, Stig Kanger [119] and Alan Ross Anderson
[12], [14] showed in 1956/7 that the analogy between the alethic and deontic
modal concepts may not only be used to transfer the rules of the former to
the latter, but to reduce deontic logic to alethic modal logic altogether. To
this effect, a prohairetic constant Q (Kanger, Q denoting what morality
proscribes) or S (Anderson, S meaning a sanction or a bad state of affairs)
is added to any modal system from K upwards, and the deontic O-operator
is then defined in the following way, respectively:
(Def-OQ)
OA =def (Q A)
(Def-OS)
OA =def (A S)
So in Kangers interpretation, that A is obligatory means that what morality
prescribes necessitates A, and in Andersons interpretation it means that not
doing A necessitates the sanction (or the bad thing).10 If the alethic modal
9
DEONTIC LOGIC
56
Let the deontic fragment DF (KQ+ ) [DF (KS + )] contain all A KQ+ [KS + ]
with A = (B) for some B LDL . Then DF (KQ+ )[DF (KS + )] = (SDL).
The reduction of deontic to alethic modal logic is interesting in its own
right, it can even point to an history of its own. Kangers variant, that
something is obligatory if it is necessary for an ideal state, stands right in
prescribes). Knuuttila [124] p. 236 points out that a reverse reduction of the alethic
modalities to deontic modalities can be found in the Dialectica of Peter Abelard (1079
1142) who identified necessity with what nature demands, possibility with what nature
allows, and impossibility with what nature forbids.
11
Anderson showed that even this addition can be avoided if one chooses the modal
system KT and replaces S by the formula B B. For Kangers constant Q, the formula
B B constitutes a similar replacement.
DEONTIC LOGIC
57
2.4
Axiom systems characterize the logic in question by means of sets of theorems. E.g. we know that
OA P B P (A B),
meaning that if something is obligatory and something else is permitted,
then we may realize both conjunctively, is a theorem of standard deontic
logic SDL, because its theoremhood can be established by means of this
systems axioms and the rules, namely (K) and modus ponens. If A is a
theorem of e.g. SDL, we write SDL A and may call A provable in SDL. As
usual, from this notion a variety of others derive: a set of sentences LDL
is called SDL-inconsistent if and only if there are A1, ..., An in , n 1, such
that SDL (A1 ... An ), and is SDL-consistent otherwise. A sentence
12
Very similarly, Andersons sanction has been interpreted to represent simply failure
to be perfect (Prior [179] p. 146.)
13
Cf. Husserl ([108] p. 41/42, my translation): The following forms are equivalent:
A ought not to be B, and an A which is not B, is generally a bad A, or only an A,
which is not B, is a good A , an A may be B = an A, which is B, needs, for this
reason, not to be a bad A ; Menger [150]: The statement I command p (...) may
be interpreted as unless p, something unpleasant will happen ; Bohnert [33] reduces
commands to motivational declarative sentences like P B(x), where B is the behavior
desired of x and P a penalty, or E B(x), where E is some desired end. Both, Menger
and Bohnert, are aware that their definitions, phrased in terms of ordinary disjunction
and material implication, have unwanted consequences (like that all factual truths are
commanded), to avoid them Menger applies a three-valued logic, Bohnert demands that
B(x) should be causally necessary for the end E.
14
By singling out an undesirable state of affairs, an action may be defined as forbidden
if it necessarily results in this state, as obligatory if the action not- is forbidden, and as
permitted if is not forbidden (cf. Meyer [152]).
DEONTIC LOGIC
58
DEONTIC LOGIC
59
DEONTIC LOGIC
60
2.5
F A O(A B),
DEONTIC LOGIC
61
sentences will derive one of the others. But intuitively, the normative situation is not inconsistent and the sentences are independent from each other,
so their formalization should not make them otherwise. Thus this example
was termed Chisholms Paradox.
In reaction to Priors paradox, G. H. von Wright in 1956 [244] introduced a
stroke / in the scope of the operator O, where the dyadic formula O(A/B)
means that A is obligatory under the condition B. The old, monadic Ooperator is then defined as OA = O(A/ ), i.e. as an obligation that holds
in arbitrary, or default conditions. More formally, the new language is as
follows:18
Definition 2.5 (The Language of Dyadic Deontic Logic)
The alphabet of the language of dyadic deontic logic LDDL is like that of LDL ,
except that it additionally contains the stroke /. LDDL is then the smallest
set such that
(a) if A, B are in LP L , then O(A/B) is in the set,
(b) if A, B are in the set, then so are A, (A B), (A B) and (A B).
Again, P (A/B) abbreviates O(A/B) and F (A/B) abbreviates O(A/B).
OA, P A and F A abbreviate O(A/ ), P (A/ ) and F (A/ ) respectively.
With the introduction of the stroke, soon numerous new axiomatic systems of dyadic deontic deontic logic emerged. Systematically, the following
dyadic analogues of the axiom schemes of SDL appear to be good candidates
for an axiom system of dyadic deontic logic:
(Ext)
If A B is a tautology, then O(A/C) O(B/C) is a theorem.
(ExtC)
If C D is a tautology, then O(A/C) O(A/D) is a theorem.
(DM)
O(A B/C) (O(A/C) O(B/C))
(DC)
(O(A/C) O(B/C)) O(A B/C)
(DD)
O(A/C) P (A/C)
(DN)
O(A A/C)
These axioms allow reasoning with antecedents like C or D only in the very
weak form of (ExtC). But sometimes we want to use information from the antecedent for our reasoning with consequents. Consider the following example
by Powers [174]:
18
Note that the language LDDL is more restricted than LDL and e.g. does not allow
iterated or nested O-operators as in OOA or O(OA A), or mixed propositional and
deontic formulas as in B OA. I simplify matters, as an extended language would e.g.
require operators and axioms for universal necessity in order to make the system complete
with respect to the kind of semantics given below.
DEONTIC LOGIC
62
John must either not impregnate Suzy Mae or marry her. Suzy
Mae is pregnant by John. So John must marry Suzy Mae.
A rule that permits such inferences is achieved by strengthening (Ext) to a
principle of circumstantial extensionality :
(CExt)
If C (A B) is a tautology, then O(A/C) O(B/C) is a
theorem.
Using this rule, we obtain the theorem O(A B/A) O(B/A), i.e. in
a situation where A is true (Suzy is pregnant by John) the obligation to
realize A B (Either Suzy is not pregnant by John or John marries Suzy)
is equivalent to an obligation to realize B (John marries Suzy). If (Ext) is
replaced by (CExt), it is clear that either (D) or (N) must be weakened to:
(DD-R) Unless C is a contradiction, O(A/C) P (A/C) is a theorem.
(DN-R) Unless C is a contradiction, O( /C) is a theorem.
Otherwise for impossible conditions the theorem O(/) is derivable from
(DN) and (CExt), and also the contradictory theorem O(/) from (DD),
(DC) and (CExt), i.e. the set of axioms would be inconsistent.
Two other early dyadic deontic axioms or theorems were the principles of
conditionality and of rational monotony:19
(Cond)
O(A/C D) O(D A/C)
(RMon) P (D/C) (O(A/C) O(A/C D))
(Cond) expresses that if A is obligatory in the situation C-and-D, then it
is obligatory in the situation C that if one realizes D then one also realizes
A. (RMon) expresses that if one realizes something permitted, then ones
obligations do not change. Together, these axioms, with the choice between
unrestricted (DD) and restricted (DN-R), or restricted (DD-R) and unrestricted (DN), describe the two most successful systems in dyadic deontic
logic:20
19
(Cond) is Reschers [182] axiom A4, it also derives from the left-to-right direction of
von Wrights [244] axiom A2: P (A B/C) P (A/C) P (B/C A). (RMon) is derivable
from its right-to-left direction and is also Reschers [182] theorem T4.16. Given (CExt),
the conjunction of (Cond) and (RMon) is equivalent to A2.
20
As for the not so successful approaches, note first that these should not be theorems:
Up
O(A/C) O(A/C D)
Down
O(A/C) O(A/C D)
With Up or Down, any contrary-to-duty obligation, contrary to intuitions, creates a conflict of duties. E.g. consider a Chisholmian set {OA, O(B/A), O(B/A)}. With Up we
derive OB and OB from O(B/A) and O(B/A), so there is a conflict. From OA and
DEONTIC LOGIC
63
O(A/A)
[ only DDLH ]
O(/C) (O(A/D) O(A C/D))
P (/C) (O(A/D) O(A C/D))
O(A D/C) O(A/C D)
[ = (P -Cond) ]
O(D/C) (O(A/C D) O(A/C))
[ = (P -RMon) ]
O(A/C) O(A/D) O(A/C D)
O(A/C D) (O(A/C) O(A/D))
P (C/D) (O(A/C D) O(A/C))
P (A/A B) P (B/B C) P (A/A C)
P (A2 /A1 ) ... P (An /An1 ) P (A1 /An ) P (An /A1 )
O(A2 /A1 ) ... O(An /An1 ) O(A1 /An ) O(An /A1 )
DEONTIC LOGIC
64
DDLL were more successful than others was not so much because they included some widely recognized theorems, but because they are also sound
and (weakly) complete with respect to a preference semantics that models
contrary-to-duty situations in an intuitively plausible way:21 Let (at least
as good as) be a reflexive, transitive and connected preference relation on
B (the set of all Boolean valuations v : P rop {1, 0}) that additionally
satisfies the Limit Assumption:
(LA)
For all A LPL , if A = then best( A ) =
Here A = {v B | v |= A} is the subset of Boolean valuations or worlds
that make A true (the A-worlds), and for any X B, best(X) = {v
X | v X : v v }) is the set of best worlds in X. The Limit Assumption tells us that for all PL-sentences that are true at some world, there are
worlds at which they are true that are at least as good as any other world
at which the sentence might be true. Finally, alongside the usual truth definitions for Boolean operators the following truth definitions for the deontic
O-operator are used, for DDLH and DDLL respectively:
DDLH :
|= O(A/C)
iff
best( C ) A
L
DDL :
|= O(A/C)
iff
v C A ) : v C A : v
v
So for DDLH , O(A/C) is defined true iff the best C-worlds are A-worlds, and
for DDLH , O(A/C) is defined true iff there is a (C A)-world such that no
(C A)-world is at least as good.
Just as the introduction of the stroke / into the scopes of the deontic
operators was motivated by the search for a representation of contrary-toduty imperatives, so was the development of preference semantics for dyadic
deontic logic. This was expressed particularly clearly by Bengt Hansson:22
The problem (...) is what happens if somebody nevertheless
performs a forbidden act. Ideal worlds are excluded. But it may
be the case that among the still achievable words some are better
than others. There should then be an obligation to make the best
out of the sad circumstances.[84]
21
A proof of (weak) completeness for DDLH was first presented by Spohn [203], it can
be straightforwardly adapted to DDLL . As the semantical details vary between Hansson
and Lewis, I use a description that fits both systems.
22
Also cf. Mott [154]: The lesson to be learned from Chisholms paradox has been the
eminently convincing, indeed obvious, one: that what we ought to do is not determined
by what is the case in some perfect world, but by what is the case in the best world we can
get to from this world. What we ought to do depends upon how we are circumstanced.
DEONTIC LOGIC
65
Hanssons and Lewis semantics capture this idea in the form of a (possibly
infinitely descending) chain of sets of ideal, subideal etc. worlds: the so-called
system of spheres (cf. fig. 1). Ideal worlds are those that are best overall:
those in best(B). The second-best worlds are those that are best in the set
of all worlds without the best worlds, i.e. those in best(B\best(B)), and so
on. For any possible situation C, the best C-worlds are those that are in the
smallest C-permitting sphere, i.e. the smallest sphere that has a non-empty
intersection with the set of C-worlds C . The Limit Assumption guarantees
that for all such C there is such a smallest C-permitting sphere. As to how
Though such a restriction is not an option for the semantics as outlined, it is possible
in Lewis [137] semantics where the set of worlds may be a subset of B, e.g. one where it
is necessary that A. Then we have (O(B/A) A) OB.
DEONTIC LOGIC
66
which is how preference semantics solves Chisholms Paradox. Other socalled paradoxes receive similar solutions in dyadic deontic logic, like the
Paradox of the Good Samaritan [179] and that of the Gentle Murderer
[56]: in monadic deontic logic it follows from the fact that it is forbidden
to assault a man that it is also forbidden to help the assaulted man, and
an obligation not to commit murder, but otherwise murder gently, is not
expressible for the reasons given for Priors Paradox. In dyadic deontic logic
we can state without contradiction that assault and murder are forbidden,
but that in the bad situation that either happens secondary obligations to
help or do it gently obtain.
2.6
2.6.1
The axiomatic basis of deontic logic has always been feeble at best. It has
been claimed that deontic logic has not a single axiom that is not incompatible with some reasonable moral position and so they all violate the principle
of logical neutrality (cf. Sayre-McCord [191]). Worse still, the main semantical tool of traditional deontic logic, possible worlds semantics, is stretched
in its task to clarify our intuitions about possible theorems of deontic logic,
as it is itself the target of philosophical criticism. Consider the definition
DEONTIC LOGIC
67
DEONTIC LOGIC
68
DEONTIC LOGIC
69
But this seems like a bad defense of possible worlds semantics. If the motivation for deontic possible worlds semantics starts with given norms or duties,
and defines the ideal worlds as the ones where all is as is envisaged by these
norms, then why not appeal to these reasons directly and try, as Kalinowski
advises, to reconstruct deontic logic on the basis of norms?
2.6.2
A second, more practical problem is that existing norms may conflict, or, in
unforeseen circumstances, result in dilemmas. For moral norms the question
of the existence of true conflicts has been the subject of intense philosophical
debate since E. J. Lemmon [132] started it; in positive law they are, according to Alchourron [2] and Alchourron & Bulygin [6], frequently found and
no rarity, while computer scientists seem to have come to view conflicts
between specifications as inevitable. But standard deontic logic excludes the
possibility that OA and OA are both true by its axiom (D), which is equivalent to (OA OA). Possible worlds semantics makes it impossible that
all deontic alternatives make true A and also A unless there are no deontic
alternatives, which is excluded by the seriality restraint that corresponds to
axiom (D). But to drop seriality seems not a proper way to model conflicts:
in a conflict, there are not too little, but rather too many ideal ways to act.
Similarly, for the axiomatic approach, simply removing (D) from the set of
axioms will not be sufficient: unless one accepts that a case of conflict makes
anything obligatory, one would also have to remove or replace (M) or (C),
which make it possible to derive from OA and OA that OB for any B (so
called deontic explosion).
For dyadic deontic logic there is not only the problem of conflicts, but also
that of dilemmas: for any two apparently harmless statements of obligation
OA and OB, the situation (AB) creates a dilemma, i.e. a situation where
something that is obligatory can no longer be realized (unless there is the
rather artificial assumption that either A implies B or B implies A). Just like
standard deontic logic cannot accommodate conflicts, the traditional dyadic
deontic systems DDLH or DDLL do not accept dilemmas: given that (AB)
is consistent, O(A/(A B)) derives O(A/(A B)) with (DD) or (DDR), whereas with (CExt) we derive O(A/(A B)) from O(B/(A B)).
In dyadic deontic semantics, to model any two sentences OA and OB, the
system of spheres must specify which state of affairs A or B is to be preferred
in the section ||(A B)|| of the spheres, though nothing might be specified
DEONTIC LOGIC
70
about such a situation by the sentences that one aims to model consider
the above modeling of Chisholms dilemma: we assumed that not going and
not telling is better than not going and telling, though the original sentences
were silent about this question.
An easy way out of the problem of conflicts and dilemmas is to say that
the task of deontic logic is to model the moral or legal situation after it has
been resolved by appropriate mechanisms of conflict resolution. But we then
admit that deontic logic is neither able to describe conflicts nor to analyze
the resolution mechanism:
It is of no avail to say that logic pictures the situation that arises
after the normative conflict has been resolved. For rather that
what is being resolved, the conflict of norms, is the interesting
state of affairs that must be logically characterized. (Alchourron
& Bulygin [6] p. 25)
Another way out is to give up the idea that the expressions of deontic logic
somehow relate to real sets of norms and rules, describe what ought to be
the case according to some system of norms (which can be inconsistent and
thereby falsify axiom D), but that deontic logic only describes rules for rational norm giving: a rational law giver cannot enact norms which make both
OA and also OA true, because a rational law giver wants that all the norms
are obeyed, but a subject could not obey all norms if according to the norms
both OA and OA are true. This is the route that von Wright has taken
in his work since [252], and it seems also to be the one taken by Weinberger
when he writes that normative systems which contain conflicting norms can
exist in social reality, they are logically flawed, but do not loose their function and social meaning ([234] p. 25, my translation).26 But lawyers, when
they explain what ought to be done, may be done or is forbidden to do in
a certain situation, when e.g. a law has been violated or such a violation
is imminent, do not advise the law giver. They advise their clients, which
are usually the subjects of the law. Thus a solution that views deontic logic
only as a tool for law givers, which they may, unlikely as that seems, use in
the activity of passing a law, would miss a large part of ordinary language
discourses in which the expressions you are obliged to, you may, you are
forbidden to are used that deontic logic claims to formalize.
26
Cf. Holl
ander [100] p. 98 (my translation): By concluding that a norm-giver may be
reasonable or unreasonable (...), Weinberger reaches the same conclusion as von Wright.
DEONTIC LOGIC
71
The hard way is to accept the possibility of conflicts and make changes to
accommodate them and explain resolution mechanisms. Approaches to accommodate conflicts in traditional deontic logic reach from replacing (D) by
(P) O and dropping (C), and in the semantics use minimal models (neighborhood semantics) to pick alternative sets of deontic alternatives instead of
one (minimal deontic logic),27 or to use multi-ideality or multipreference
semantics that has not one deontic alternative or preference relation but several, and thus can make OA and OA true in a model that makes A true
in all ideal or best alternatives according to one relation, and A is true in
the ideal or best alternatives according to another,28 to sequent systems that
replace (C) by a rule of consistent cumulativity, which tests consistency
of A B before cumulating OA and OB to O(A B) and which can only
be used before a use of (M), and so avoids the derivation of OB from OA
and OA,29 to axiomatic systems that leave (C) intact but exchange (M)
for a rule of permitted monotony for the same reason.30 So there are ways
in which deontic logic can be modified to accommodate conflicts, and even
keep possible worlds semantics, though this semantics may not always be
very intuitive.
As for mechanisms to resolve conflicts, a natural way to do so is by
weighing the conflicting obligations and giving preference to the ones with
the highest priority.31 Though it seems not to have been done before, this
too might be accomplished by using preference semantics and in particular
Gobles [67] multiplex preference semantics, through introducing a preference
relation over the multiple ideality or preference relations (OA then holds all
things considered if A is true in the most ideal of the ideal worlds). However,
the severe changes that would be required and the multitude of proposals how
to go forward to solve the problem of accommodating conflicts and dilemmas
shows that there are no easy solutions along the old ways of thinking about
deontic logic. In fact, the difficulties of traditional deontic logic to handle
such problems have led to doubts about the whole enterprise:
27
Cf. Chellas [45] pp. 200203, 272277 for its monadic and dyadic variants, also cf.
Nortmann [159] who likewise employs minimal models but instead of (C) drops (M).
28
Cf. Schotch & Jennings [192], and Goble [66] [67] for such solutions.
29
Cf. van der Torre & Tan [214] for such a labelled sequent system.
30
Cf. Goble [69]. In [68] Goble also presents a possible worlds semantics that is sound
and complete with respect to such axiomatic systems.
31
Cf. Isaac Watts [220] part II, ch. V, sec. III, principle 10: Where two duties seem
to stand in opposition to each other, and we cannot practise both, the less must give way
to the greater, and the omission of the less is not sinful.
DEONTIC LOGIC
72
2.6.3
DEONTIC LOGIC
73
Cf. Ayer [24] pp. 105/106, Reichenbach [180] pp. 19/20, von Kutschera [128] pp. 11
14, Hilpinen [92] pp. 299/300. Bentham was the first to succinctly point at this ambiguity:
The only mood to which grammarians have thought fit to give the name imperative is
by no means the only one which is calculated to perform the office of imperation. (...)
But in many instances it will be found that the style of imperation is altogether dropped.
He proceeds to give a variety of forms of a law against the exportation of corn, among
these the legislator speaking as it were in the person of another man who is considered
as explaining the state which things are in, in consequence of the arrangements taken by
the legislator, It is not permitted to any man to export corn. It is unlawful for any man
to export corn. No man has a right to export corn. Why? because the legislator has
forbidden it. The exportation of corn is forbidden. ([29] pp. 104-106, 154)
33
Also cf. Kalinowski [114] p. 118 (my translation): For what does an expression of
the form it is obligatory that A mean? If this is not a norm but a proposition about
norms it means that the law giver L has enacted the valid norm N according to which A
is obligatory ; Opalek & Wole
nski [164] p. 384: Deontic logic is conditioned by norms,
because deontic statements are true or false depending on the existence of respective acts
of norming.; Sinowjew [197] p. 266 (my translation): The truth values of normative
DEONTIC LOGIC
74
The interpretation that the propositions of deontic logic are true and
false descriptions about what is obligatory, permitted or forbidden, and are
not themselves norms or imperatives, has not been shared by all authors.
Casta
neda [42] viewed deontic logic as a modal logic of imperatives and
resolutives.
Aqvist [21] argued that it should be possible to interpret deontic logic atheoretically as a logic of commands, in the sense that OA
expresses a command, and not a proposition about a command; occasional
oddities like the difficulty of interpreting formulas like OA should be accepted as a small price for a logical theory of commands. Such a prescriptive
interpretation of deontic logic was also adopted by Alchourron [2] when he
identified deontic logic with a logic of norms, and set out to develop a logic
of normative propositions in parallel. Chellas [44] replaced the O-operators
of deontic logic by the symbol !, where !A is to be understood as representing a natural language imperative let it be the case that A, and presents
a logic for such formulas that is equivalent to SDL. For Bailhache [25], deontic logic is the logic of deontic norms (norms that are created by the use
of deontic expressions), where the deontic formulas are then evaluated with
respect to accessible ideal worlds as usual. More recently, Kamp [116] p.
233 has demanded that the formulas of deontic logic must be interpretable
both descriptively and prescriptively. Sometimes it is not even quite clear in
which way standard accounts of deontic logic desire its formulas to be read:
Carmo & Jones [40] write that deontic logic is a formal tool needed to design normative systems and, just like Bailhache in [26], throughout call the
sentences of deontic logic norms or deontic norms and so seem to adopt
the prescriptive interpretation, but the authors then employ possible worlds
semantics to evaluate deontic propositions (norms?) as usual. Fllesdal &
Hilpinen [53] speak of deontic propositions constituting or being implied by
a normative system (e.g. pp. 13, 29), or of deontic logic formalizing imperatives (p. 26), but portray the descriptive interpretation of deontic logic as
one that they shall often resort to (p. 8). In the face of all this confusion,34 David Makinson [141] noted that work on deontic logic goes on as if a
propositions are determined according to the usual rules. E.g. a proposition it is forbidden
to smoke is true if it really is forbidden to smoke (if there exists such a norm), false if it
is permitted to smoke, and undetermined if there is neither a prohibition nor a permission
to smoke.
34
Holl
ander [100] pp. 9899 uses Hintikkas concept of a deontically perfect world to
motivate his deontic logic as a logic of norms in the following way: the deontically perfect
world is a place where conflicting norms cannot exist, as opposed to the real world where
DEONTIC LOGIC
75
distinction between norms and normative propositions has never been heard
of.
In part, the confusion about what the formulas of deontic logic mean can
be attributed to von Wright himself, who later wrote that while deontic logic
is a theory of descriptively interpreted expressions, its laws concern logical
properties of the norms themselves, and that therefore the basis of deontic
logic is a logical theory of prescriptively interpreted O- and P -expressions
([246] p. 134). This reflection of logical properties of norms in the logical
properties of deontic propositions have led von Wright to famously claim that
logic has a wider reach than truth ([245] p. vii). Still later, von Wright [251]
reaffirmed that deontic logic is a logic of norm-propositions and denied the
possibility of a logic of norms, except in the form of advice to a rational law
giver (cf. [252] p. 132, [256] p. 110).
However, it is quite clear that OA and P A cannot be interpreted as representing imperatives or norms. I have explained in sec. 1.2 why imperatives,
and for that reason, also deontic norms, cannot be meaningfully termed
true and false. But if norms are neither true nor false, then the Boolean
operators occurring in the formulas of deontic logic such as OA OB,
P A OA, OA P B, cannot have their usual, truth functional meanings as and, or, not, if ..., then. So a logic of norms, even if one believes
it exists, cannot resemble deontic logic,35 and the meaning of (sub-) formulas
such as OA must be interpreted descriptively.36 So the descriptive interpretation of the formulas of deontic logic is the only tenable one, and is also the
one used here.
What the confusion in deontic logic literature as to the meaning of its
formulas illustrates, is a need to better explain the relation between the
formulas of deontic logic, as logic of deontic propositions, and the norms
that these propositions are statements about. Von Wright hints at this when
writing in 1951 [243]:
they can. The logical relations studied are then those that exist between norms in the
deontically perfect world. But deontic alternatives, as described by Hintikka [97], [95], are
sets of propositions, not places where norms exist. If deontic propositions are true in a
world that is ideal to ours (it is not just one, as Hollander seems to assume), this is due
to other worlds that are ideal to the ideal world. All the previously mentioned problems
and confusions about deontic logic seem to eclipse here.
35
Cf. Makinson [141] p. 30. Keuth [123] and Swirydowicz [208] therefore restrict their
logic of norms to statements of normative entailment of the form !A !B.
36
That is, unless one is prepared to leave the common ground and call norms true and
false, as Kalinowski [111], [114] and Kamp [118], [116] have done.
DEONTIC LOGIC
76
DEONTIC LOGIC
77
unknown, but an explicitly represented sets of norms will clarify such discussions and the understanding of deontic logic.
2.7
In the face of the problems and doubts outlined above, and the multitude of
proposals to modify deontic logic, it has been concluded that
deontic logic has fallen into ruts. The older rut is the axiomatic
approach, with its succession of propositional and occasionally
quantified calculi. The newer one is possible worlds semantics,
with endless minor variations in the details. (Makinson [141])
In consequence, a fresh start has been called for. A representation of norms
in deontic logic could constitute such a fresh start. However, if norms are to
be represented in deontic logic, there is then the question of how to do so
without falling into the trap of Jrgensens Dilemma [110]: how can norms
be a subject of logical reasoning if norms are neither true nor false and so
cannot be part of any logical inference? This has been called the fundamental
problem of deontic logic:
It is thus a central problem we would say, a fundamental problem of deontic logic to reconstruct it in accord with the philosophical position that norms are devoid of truth values. In other
words: to explain how deontic logic is possible on a positivistic
philosophy of norms. (Makinson [141])
Jrgensens dilemma cannot be dealt with, and the fundamental problem
not solved, by making the object language of deontic logic one of norms.
As was explained above: we cannot negate norms, form conjuncts or adjuncts out of expressions representing norms, for what meaning would the
Boolean operators then have? We must let the statements of deontic logic
remain true or false statements about norms, deontic or normative propositions that describe what is obligatory, permitted or forbidden according to
the norms. However, the norms according to which the normative proposition may be true or false should be explicitly represented in the semantics
of deontic logic. Deontic logic would then, as it once aspired, not talk about
ideal states, preferred worlds or perhaps preferences over preferences, but
DEONTIC LOGIC
78
more directly about positive law, not to forget the more profane usages of
normative language like rules set up by mums and dads, employers or school
teachers, with all their curiosities and inconsistencies.
If the semantics by which we evaluate the truth of deontic formulas is
changed, there are then the questions of what truth definitions should be
used for the deontic operators, and in particular what a thus semantically
reinterpreted deontic logic will be like: will the traditional systems of deontic logic still be sound and complete, or in Makinsons terms: can they
be reconstructed with respect to such a new semantics? Or will we arrive
at completely new axiomatic systems? The next chapter describes how the
formulas of deontic logic can be evaluated with respect to a logical semantics
that explicitly represents them (I call this imperative semantics), and what
systems result.
Chapter 3
Imperative Semantics for
Deontic Logic
3.1
The idea to represent norms or imperatives in the logical semantics for deontic
logic is not entirely new. While the mainstream of deontic logic adhered to
possible worlds semantics, quite a number of authors have tried, in one way
or another, to link deontic logic to norms or to imperatives we might
call this the normative or imperatival tradition of deontic logic. What is
typical for such approaches is that they have, besides the truth definitions for
deontic operators, a set of objects that are meant to be prescriptions, or that
contains the propositions that relate to prescriptions in the sense explained
by Weinbergers Principle (W): the proposition that must be true when the
imperative is satisfied, and false if it is violated. The deontic operators
are then in turn defined with respect to this set, and not a possible worlds
semantics. Any proposal that has these features is counted as belonging to
this imperatival tradition of deontic logic.
3.1.1
As explained in sec. 2.3, Alan Ross Anderson was the first to propose to
reduce deontic logic to alethic modal logic, and define OA by means of a
constant S representing a sanction:
OA =def (A S)
79
IMPERATIVE SEMANTICS
80
Also cf. Goble [64] p. 199: There is an analytical connection between the statements
(i) it is obligatory that p and (ii) if not-p, then V, where the constant V is to represent
some violation or bad state of affairs or wrong in the world.
IMPERATIVE SEMANTICS
3.1.2
81
IMPERATIVE SEMANTICS
82
3.1.3
Regarding preference semantics, the proposal not to use one, but several sets of norms
corresponds to Gobles [66], [67] multipreference semantics that uses not just one, but
multiple preference relations to explain OA in terms in what is best according to all.
IMPERATIVE SEMANTICS
83
3.1.4
Note that p and q are here proposition variables, i.e. sentences from LP L , rather than
proposition letters. In this section I try to keep to the original texts as closely as possible.
IMPERATIVE SEMANTICS
84
logical relations between norms.4 In this respect his proposal and related
proposals5 differ from other proposals discussed in this section, and also the
proposal explained in the following sections: I do not assume any theory of
logical relations between norms or imperatives, for the simple fact that it is
doubtful and heavily disputed whether such relations exist, and that I agree
with a position of imperativological scepticism that, at least for imperatives,
altogether denies them. If, contrary to this view, such relations do exist, then
nothing is lost: we can close the set of norms or imperatives by any accepted
principle or rule and still use the formulas of deontic logic to describe what
is obligatory according to this set, just as Stenius proposes. However, if such
logical relations do not exist, then the interpretation of deontic logic as one of
expressions about sets of norms or imperatives will still remain meaningful.
3.1.5
Alchourr
on & Bulygins Expressive Conception
of Norms
For this reason, Stenius [206] later retracted his proposal and only called a system of
norms well-formed if it is not closed under consequences.
5
deontic operators via a normative system closed under rules, where Swirydowicz
uses
Ziembas [258], [259] dyadic formalization of imperatives which he enriches with logical
relations between the imperatives.
IMPERATIVE SEMANTICS
85
3.1.6
IMPERATIVE SEMANTICS
86
IMPERATIVE SEMANTICS
3.1.7
87
Input/Output Logic
In answering his own call [141] for a reconstruction of deontic logic in accord
with the position that norms are neither true nor false, David Makinson,
together with Leon van der Torre [142], [143], develops input/output logic
as a tool to extract the essential mathematical structure behind this reconstruction (cf. [144]). In their view, input/output logic should be seen not
as a new kind of logic, but as a way of handling norms: given a set of norms,
it must e.g. be determined which of these are still operative in a situation
that already violates (or satisfies) some of them. The set of norms G is taken
as a black box or transformation device, and the task of logic is then that
of a secretarial assistant, preparing the factual information that goes into
this device (the input A) and coordinating the production of the output,
written as out(G, A).
The simplest input/output operation that Makinson & van der Torre
suggest is that of simple-minded output. Let G be a set of conditional
norms (a, x), where a and x are sentences in the language of propositional
logic LPL , a describing some condition or situation, while x represents what
the norm tells us to be obligatory in that situation. A set of sentences
A LPL serves as explicit input. The operation first expands A to its
classical closure Cn(A). This is passed into the transformer G which delivers
as immediate output the set G(Cn(A)) = {x | (a, x) G and a Cn(A)}.
Simple-minded output out1 (G, A) is then again the classical closure of this
set, i.e. out1 (G, A) = Cn(out1 (G, A)). This operation is depicted by the
following figure from [144]:
IMPERATIVE SEMANTICS
88
conditional norms that are triggered by the input, i.e. whose conditions
are true given the input, are to be satisfied. Makinson & van der Torre
then refine this simple-minded output and provide further output operations
that e.g. provide reasoning by cases (if there are two norms (a, x) and (b, x)
in G, then x is in out(G, {a b})), normative detachment (if there are two
norms (a, x) and (x, y) in G then y is in out(G, {a})), or permit throughput
(if there is a norm (a, x y), then y is in out(G, {a, x})). Finally, the
operation may employ restraints to e.g. ensure that only such a (maximal
sub-) set of norms is used that provides only consistent output (output such
that
/ out(G, A)).
In the above outline, the output is a set of descriptive sentences that
might be seen as describing what is obligatory given the norms in G and the
situation described by the sentences in A (deontic interpretation). Hence we
may use an input/output operation out(G, A) to define a deontic logic via
G, A |= Ox iff x out(G, A)
i.e. Ox is true given the set of norms G and the factual descriptions A iff x
is in the output out(G, A) of the used output operation. But the operation
might also be viewed as providing new norms conditional on the situation
described by a singular sentence a as input, where the new set of norms
may be defined as out(G) = {(a, x) | x out(G, {a})}, i.e. the operation
provides the norm (a, x) just if x is in the output out(G, {a}) (normative
interpretation).6 In the first interpretation the head depicted in the above
figure acts as a kind of lawyer or legal advisor, telling the agent what ought
to be done according to the norms in the situation described by A. In the
second interpretation the head acts as a kind of judge, or subsumtion machine
(Subsumtionsautomat), handing out norms for specific situations as a result
of the (general) laws known to the judge, a picture that is remindful of the
Latin proverbs iura novit curia and da mihi facta, dabo tibi ius (the court
knows the law, give me the facts and I will give you the law). Makinson & van
der Torre leave it open how the output should be interpreted, in particular no
deontic operators are defined, but if the aim is to reconstruct deontic logic,
then obviously the first interpretation should be preferred. It is easy to see
that then e.g.
6
Cf. Makinson & van der Torre [142] p. 385. Swirydowicz [208] defines a normative
consequence relation very similar to Makinson & van der Torres basic output to provide
a method to derive norms from norms.
IMPERATIVE SEMANTICS
89
(Conj)
OA OB O(A B)
(N)
O(A A)
or statements such as if G, A |= Ox then G, A {a} |= Ox (strengthening
of the input) all hold for the simple-minded output operation out1 . The
main question will then be which kind of output operation provides the most
adequate results for a correspondingly defined deontic O-operator. In [80] I
have considered truth definitions of dyadic deontic operators that resemble
the operations of input/output logic, and discussed some of their properties.
I will to return to the results of this work shortly, but will now explain my
own approach to model norms in a semantics for deontic logic.
3.2
By stipulating that the imperatives in I are binding or valid, a notion that may in
turn depend upon some acceptance by the subjects and/or a legal community, I hope to
side-step all discussions on whether normative propositions are really true or false, or
rather just acceptable or non-acceptable (cf. Mazzarese [148], Niiniluoto [158]), for from
the same viewpoint that accepts these imperatives as binding, some normative propositions
that something is obligatory, permitted or forbidden according to the imperatives must
IMPERATIVE SEMANTICS
90
IMPERATIVE SEMANTICS
91
in do not eat with your fingers, buy apples and walnuts, write her a
letter, or at least a postcard, and these imperatives are then satisfied by
not eating with ones fingers, buying apples and buying walnuts, and writing
her a letter or writing her a postcard. But Weinbergers principle as well
as the analysis of Hofstadter & McKinsey [99] show that in terms of the
sentences that describe the satisfaction of an imperative, a formalization of
these connectives within imperatives would only amount to a doubling of the
usual Boolean operators: if we employ the name !(p1 p2 ) for the imperative
buy apples and walnuts, its associated descriptive sentence f (!(p1 p2 )) is
still p1 p2 , i.e. the imperative is still satisfied just in case its addressee
buys apples and buys walnuts. Also, we may want to use expressions that
are imperatives not in the grammatical, but in a functional sense in our
examples, like I must ask you to pack your gear and leave the property,
can you close the window or turn off the ventilation, please? or you ought
to go now, and then, with the aid of principle (W), we need not worry about
each expressions proper formalization too much.
I will, however, make two assumptions about the contents of the imperatives: i) that imperative sentences are not separable, and ii) that they are
independent. Sometimes even doing only part of what has been requested
or commanded is seen as something good and (partly) following the order,
while at other times failing a part 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 are needed for biscuits and the apples for dessert,
then it makes sense and may be required to get the walnuts even if apples
are out. (We just have the biscuits for dessert.) If, however, both 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. As the example illustrates, there is no
logical method to distinguish one case from the other. The following may
be used as a test for separability:11
(Sep) If by use of an imperative sentence i someone is required, among
other things, to do A and B, and is still required to do A even if
B cannot be done, then i is separable.
Instead of a separable imperative, we may just use two or more inseparable
imperatives to express what is required. However, ordinary usage of two
imperative sentences does not guarantee their independence. Suppose a flight
11
IMPERATIVE SEMANTICS
92
IMPERATIVE SEMANTICS
93
money on ice cream, I might not be able to buy both the apples and the walnuts, and so find myself in a dilemma. I might even have spent so much on
ice cream that I cant buy the walnuts with the change: I am in a contraryto-duty situation. But it might still be enough to purchase the apples. Depending on how the imperative is to be understood, I might still be under
an obligation to buy the apples or the walnuts (in the first case), or just the
apples (in the second), or nothing at all (if I was shopping to help prepare a
Waldorf salad).
The above approach to modeling normative situations is very basic. As
noted above, there are all kinds of norms for which the core concept would
have to be amended and adjusted think of explicit permissions (licenses),
or of general norms: in order to model, within the basic concept, a general
norm not to lie, one would have to add to the set, for every subject and any
occasion to lie, an individual, independent imperative that obliges the subject
not to lie on this occasion. While some adjustments of the basic concepts
are treated below, a more fundamental criticism of my approach would be
that it does not seem to account for norms that are not created by use of
speech. One may think e.g. of natural law, Kants categorical imperative, or
more down-to-earth rules of courtesy, custom or tradition. Now, as long as
these can be rephrased to be inseparable and independent, and correspond
to descriptions of situations in which they are satisfied or violated in the
sense explained by Weinbergers principle (W), nothing hinders us to make
I a set of such norms. The set I is not an ontological category, though it
is explained in terms of imperatives, uttered by some authority, that are,
by some subject, believed to be binding. This seems to me the best way
to reconstruct everyday normative reasoning, when we consider individual
demands from possibly different sources and try to satisfy them as best as
we can. However, I admit that when obligations are considered that are
(only) justified by an appeal to prohairetic notions something is obligatory
because it is the best thing to do, it brings about the greatest good for the
greatest number, in all possible situations everyone is better off with it done
than not done, etc. , then the more natural model may be one in terms of
ideality or preference relations between possible worlds after all.
IMPERATIVE SEMANTICS
3.3
94
3.3.1
The first paper in the series, written for the EON00 workshop held in
Toulouse, France, 2022 January 2000 and subsequently published in revised
form in 2001 [75], describes the basic imperative semantics and outlines how
it may be applied to solve Makinsons fundamental problem.
The first result is almost trivial: Let I = I, f be a basic imperative
structure and define
I |= OA iff {i1 , ..., in } I : f (i1 ), ..., f (in ) |=PL A
So according to this truth definition, OA is true if and only if there are imperatives in the set I such that their associated descriptive sentences classically
imply A, i.e. A is necessary to satisfy some imperatives. Letting the truth
definitions for Boolean operators be as usual, if I is assumed to be a set
of non-conflicting imperatives, i.e. f (I) PL , then von Wrights classical
axiomatic deontic system is sound and complete with respect to this semantics. If I is additionally assumed to be non-empty, i.e. there is at least one
imperative i I, then standard deontic logic SDL, restricted to a language
that does not permit mixed formulas and nested deontic operators, is sound
and complete with respect to this semantics.
IMPERATIVE SEMANTICS
95
IMPERATIVE SEMANTICS
96
to remove all those imperative-associated sentences from f (I) that are already violated in a situation described by C, where C represents a possibly
contrary-to-duty situation. This contraction is nave as, of course, there
may be imperative-associated sentences left that each can be individually but
not collectively true together with C. But I wanted to distinguish contraryto-duty situations from conflicts and dilemmas, and preserve the possibility
of the latter and the possible truth of O in such situations. So consequently
the truth of a dyadic deontic O-operator is defined with respect to a basic
imperative structure I = I, f as follows:
I |= O(A/C) iff f (I) n C {C} |=PL A
According to this definition, O(A/C) holds if A is necessary to satisfy all
the imperatives that are not violated in the circumstances described by C.
Curiously, the dyadic deontic logic that is proven to be sound and complete
with respect to such a truth definition is just like the system DDLH described
in sec. 2.5 (i.e. like Hanssons DSDL3) except that it lacks its typical deontic
axiom (DD-R). We can also make the full set DDLH sound and complete by
assuming that dilemmas are excluded for all possible circumstances, i.e. that
if PL C then also f (I) n C PL C. However, this would commit us to
assuming that for any two imperatives !i1 , i2 I, either |=PL f (i1 ) f (i2 )
or |=PL f (i2 ) f (i1 ) otherwise the situation (f (i1 ) f (i2 )) constitutes a
dilemma. To assume that the demands of all imperatives are thus chained
rather corresponds to the nested system of spheres provided by preference
based semantics, but it seems unlikely that any real-world set of commands
will ever meet this assumption.
3.3.2
IMPERATIVE SEMANTICS
97
I |= O1 A iff i I : f (i) = A
I |= O2 A iff i I : |=PL f (i) A
I |= O3 A iff i I : |=PL f (i) A
I |= O4 A iff i1 , ..., in I : |=PL (f (i1 ) ... f (in )) A
I |= O5 A iff f (I) |=PL A
So O1 takes obligatoriness very literal, describing A as obligatory only if there
really is an imperative that demands a state of affairs properly described by
A. O1 is the operator first proposed by Alchourron & Bulygin [5]. O2 is similar, but allows A to be replaced by any logically equivalent proposition. For
O3 A to be true it is sufficient that A is necessary to satisfy one imperative.
This is the first of the definitions for OA that van Fraassen [57] proposed.
The definition of O4 is similar to Smileys [198] definition of a deontic operator: O4 A is true iff there is non-empty set of imperatives such that A
is necessary to satisfy them. We have already seen in the first paper that
this definition, with an assumption of the imperatives being non-conflicting,
yields von Wrights classical deontic system (with the language restricted so
that O-operators do not occur iterated or nested). Finally O5 is similar to
the corrected definition used by Alchourron & Bulygin [5] in that O5 A is
true iff A is necessary to satisfy all imperatives (regardless if there are any).
This definition, together with the assumption that the imperatives do not
conflict, then yields SDL (with the language restricted as above). Sound and
complete axiomatic systems are given for all operators.
I then consider what happens if we mix the operators. Obviously we can
then express truths such as Oi A Oj A for 1 i < j 5. A maybe
unexpected finding is that though all possible mixtures of operators have
sound and (weakly) complete systems (the additionally required axioms are
forthcoming), no system but the one that only uses operators O4 and O5 is
also strongly complete: the semantics is not compact, i.e. there are infinite
sets of formulas such that each finite subset of these can be modeled using
basic imperative semantics, but not the whole sets. A similar experience
awaits us for the use of semantic restrictions. Four such restrictions are
considered:
[R-0] I =
(Non-Triviality)
[R-1] i I : PL f (i)
(Excluded Impossibility)
[R-2] f (I) PL
(Collective Satisfiability)
[R-3] i I : PL f (i)
(Excluded Necessity)
1
2
It turns out that for O and O , no axiomatic system is strongly complete
IMPERATIVE SEMANTICS
98
IMPERATIVE SEMANTICS
99
IMPERATIVE SEMANTICS
100
3.3.3
So OF (A/C) is true if A is required to satisfy some imperatives whose demands are collectively consistent with the current, possibly contrary-to-duty
situation described by C. Correspondingly, OS (A/C) is defined true if A
is required to satisfy any maximal subset of imperatives such that their
demands are collectively consistent with the situation C. Again, the OF operator allows conflicts, it may be that both OF (A/C) and OF (A/C)
are true, whereas OS cautiously only pronounces that as obligatory what
is necessary to satisfy any maximal set of imperatives whose demands are
consistent with the situation C, and so (OS (A/C) OS (A/C)) holds for
all non-contradictory C.
Similar definitions to the ones above have already been used in other contexts, and the paper proceeds to point out similarities to Kratzer and Lewiss
premise semantics [125], [138],16 Poole systems without constraints [173], the
X-logics of Siegel and Forget [194], [55], and the epistemic states-semantics
of Bochman [30], [31], [32] that is similar to the cumulative models in Kraus,
Lehmann and Magidor [126]. It is then proved that to the definitions above
correspond sound and (I prove: only weakly) complete axiomatic system,
where the axioms for OF are just like that for DDLL except that (DC) does
16
Lewis [138] (p. 233) states in a footnote that for deontic conditionals, the premises of
the premise semantics might be understood to be something that ought to hold, and so
seems the inventor of this kind of imperative semantics.
IMPERATIVE SEMANTICS
101
not hold, (DD) is replaced by (DP) P ( /C), and (CCMon) is used as additional axiom scheme, and the axioms for OS are exactly like those for DDLH ,
except that (RMon) is exchanged for (CCMon), and finally the following
mixed axiom schemes are added:
(RMonF SS )
P F (D/C) (OS (A/C) OS (A/C D))
(RMonSSF )
P S (D/C) (OS (A/C) OF (A/C D))
The OS -axioms are equivalent to those used by Kraus, Lehmann and Magidor
[126] for their system P except that the deontic axiom (DD-R) is added.
The OF -axioms are those of Bochman [30], except that his axiom (Pres) is
strengthened to (DP). The mixed schemes are again those of Bochman.
Besides the new and relatively complex completeness proof, the paper
points out a link to Gobles deontic multiplex preference semantics [66], [67]
which in terms of possible worlds ordered by several preference relations
similarly attempts to accommodate conflicting obligations. The dyadic
deontic formula O(A/C) is defined true by multiplex preference semantics
iff there is at least one preference relation such that the best C-worlds are
A-worlds (this corresponds to OF ) or if according to all preference relations
the best C-worlds are A worlds (this corresponds to OS ). I show that the
construction of a set of imperative-associated sentences described by the
earlier completeness proof, that makes true a formula of deontic logic A, can
also be used to define preference relations that make true A by use of these
definitions, and that except for a small change in the restrictions applying
to (DN) and (DD)/(DP) the above axiomatic system is then also weakly
complete with respect to such a multiple preference semantics.
3.3.4
My fourth paper on imperative semantics was originally written for the working group Law and Logic at the XXII. World Congress of Philosophy of Law
and Social Philosophy (IVR 2005), held 2429 May 2005 in Granada, Spain;
a revised version was published in 2006 [78].
While the previous paper considered conflicts between imperatives and
how deontic operators, in particular OF and OS , can be defined that take
them into account and localize these conflicts, this paper studies methods to
resolve conflicts. It is commonplace that a priority relation, or ranking, between the imperatives may help to resolve conflicts. W. D. Ross [189], to that
effect, distinguished prima facie duties, possibly conflicting obligations that
IMPERATIVE SEMANTICS
102
appear to hold before we look for some ordering, provided e.g. by considering which are more pressing, ranks of authorities, gross differences in utilities
etc., from what we view as our duties sans phrase after all these things have
been considered. While e.g. the O3 -operators employed previously might be
seen as describing prima facie duties only, letting it be sufficient for OA that
A is necessary to satisfy one imperative, I now consider prioritized imperative structures I = I, f, < that additionally employ a priority relation <
between imperatives, where i1 < i2 means that i1 takes priority over i2 , and
how these might be used to resolve conflicts to arrive at a definition for an
O-operator that describes all-things-considered obligations.
There have been quite different proposals as to how priorities might be
used to resolve conflicts. I adopt an algorithm first introduced by Rescher
[183] and further developed by Brewka [35], [36] for the purpose of determining preferred subtheories in theory revision, a proposal that makes very
weak demands on < (it is irreflexive, transitive and well-founded, i.e. infinite descending chains are excluded). This then is compared with other
methods suggested in the literature, such as Hortys [104] definition of binding imperatives, Alchourron & Makinsons [9] criterion of least exposure,
Prakkens [175] hierarchical rebuttal, Sartors [190] prevailing relation, and
Alchourrons [4] use of safe contraction. These methods then turn out to be
either equivalent, or to produce counterintuitive results in examples where
Brewkas method does not.
The dyadic deontic O-operator is then defined with respect to such prioritized imperative structures I = I, f, < as follows:
I |= O(A/C) iff I C: f () {C} PL A
This is similar to the definition of the dyadic OS -operator used in the previous paper (a definition similar to that for OF could have been used just
as well), but instead of defining O in terms of what is required to satisfy
the imperatives in all maximal subsets of I such that f () does not imply
C, i.e. all sets in I C, we use all sets in a set I C of maximal
preferred subsets such that f () does not imply C. Very roughly, for any
situation C, any set in I C is defined by first adding a maximal set of
the most important imperatives such that their demands do not derive C,
then adding to it a maximal set of the second most important imperatives
that can be added without the set of corresponding demands now deriving
C, and so on. The axiomatic system that corresponds to such a definition
is then not surprisingly, because it was not assumed that all conflicts can
IMPERATIVE SEMANTICS
103
3.3.5
IMPERATIVE SEMANTICS
104
the water boil. But if the water does not boil, e.g. because there is a power
failure or one decides not to continue with the recipe, one has not done
anything right or wrong: the opportunity to satisfy or violate the instruction
did not arise.17 Much of the literature on imperatives and deontic logic is
devoted to understanding the reasoning about conditional norms, which von
Wright [253] called the touchstone (Pr
ufstein) for normative logic.
In a paper written for the interdisciplinary seminar on Normative MultiAgent Systems NorMAS07 held 1823 March 2007 at Dagstuhl, Germany,
and subsequently published in revised form in [80], I have formalized conditional imperatives by introducing an additional function g : I LPL to basic
imperative semantics that associates with each imperative the description of a
condition that provides the opportunity for satisfying or violating the imperative (I write A !B for an imperative i such that g(i) = A and f (i) = B).
Given a set of factual descriptions W LPL , an imperative is called triggered if its antecedent is true given W then the opportunity for its satisfaction or violation arises.18 The set Triggered(W, I) = {i I | W PL g(i)}
is then the set of imperatives that are triggered given the facts W . A triggered imperative g(i) !f (i) is satisfied iff W |=PL g(i) f (i), i.e. if also
its consequent is true, and violated iff its consequent is false given the facts,
i.e. iff W |=PL g(i) f (i). This makes it possible that a conditional imperative g(i) !f (i) is neither satisfied nor violated. When an imperative is
neither satisfied nor violated because the antecedent is false, the imperative
is called obeyed.19 A subset I of imperatives is obeyable given the
facts W if m() W PL , where m(i) = g(i) f (i) is the materialization of i, i.e. the material conditional corresponding to the imperative. So
is obeyable given W iff it is not the case that for some {i1 , ..., in } we
have W PL (g(i1 ) f (i1 )) ... (g(in ) f (in )): otherwise we know that
whatever we do, i.e. given any maxiconsistent subset V of LPL that extends
W V , at least one imperative in must be violated.
The truth definitions of deontic operators are then defined with respect
these conditional imperatives, and in particular with respect to those impera17
This is Casta
nedas [41] distinction between those contents of an obligation statement
that work as circumstances and those that work as deontic foci.
18
Cf. Rescher [184], Sosa [200], van Fraassen [57] Also cf. Greenspan [72]: Oughts do
not arise, it seems, until it is too late to keep their conditions from being fulfilled. Horty
[104] seems to have introduced the term triggered.
19
The terminology differs here: Downing [48] uses the term complied by instead of
obeyed, while Vranas [?] calls the imperative avoided.
IMPERATIVE SEMANTICS
105
tives in the set of imperatives I that are triggered. Several options arise, and
in parallel to input/output logic as developed by Makinson & van der Torre
[142], [143], I discuss the following definitions for O-operators with regard to
thus extended conditional imperative structures I = I, f, g :
(td -cd1) I |= O(A/C) iff f (Triggered({C}, I)) PL A
(td -cd2) I |= O(A/C) iff V LP L C : f (Triggered(V, I)) PL A
(td -cd3) I |= O(A/C) iff f (Triggered ({C}, I)) PL A
(td -cd4) I |= O(A/C) iff V LP L C : f (Triggered (V, I)) PL A
Truth definition (td -cd1) corresponds to Makinson & van der Torres simpleminded output out1 . It pronounces A as obligatory if A is necessary to satisfy
all imperatives that are triggered by C. Definition (td -cd2), like basic output out2 , allows for reasoning by cases, it e.g. makes O(A/C D) true
in case there are two imperatives C !A and D !A and the situation is
described by C D (in the definition, each V LP L C is a maximal
subset of the language that is consistent with C). Definition (td -cd3), corresponding to reusable output out3 , allows normative detachment and makes
e.g. O(A/C) true if there are two imperatives C !B and B !A and the
situation is described by C (in the definition, the set Triggered (W, ) of iteratively triggered imperatives means the smallest subset of I such that
for all i , if f (Triggered (W, )) W PL g(i) then i Triggered (W, )).
Definition (td -cd4), corresponding to Makinson & van der Torres reusable
basic output out4 , then permits both, reasoning by cases and normative detachment.
It seems important to be able to use circumstantial reasoning, i.e. employ the information about the situation not only to determine if an imperative is triggered, but also for reasoning with its consequent. E.g. if the set of
imperatives is {A!(B C)}, with its imperative interpreted as expressing
if you have a cold, either stay in bed or wear a scarf, one would like to
obtain O(C/A B), expressing that given that I have a cold and dont stay
in bed, I ought to wear a scarf. To allow for such circumstantial reasoning
for all the above definitions, and corresponding to the throughput versions
out+ of input/output logic, the above definitions may be changed into
(td -cd1+ ) I |= O(A/C) iff f (Triggered({C}, I)) {C} PL A
(td -cd2+ ) I |= O(A/C) iff V LP L C: f (Triggered(V, I)) V PL A
(td -cd3+ ) I |= O(A/C) iff f (Triggered ({C}, I)) {C} PL A
(td -cd4+ ) I |= O(A/C) iff V LP L C : f (Triggered (V, I)) V PL A
Though all these modifications of the original simple-minded definition (td -
IMPERATIVE SEMANTICS
106
IMPERATIVE SEMANTICS
107
agent ought to do, the method should at least not hand out sets of imperatives of which it is already known beforehand that at least one imperative
must be violated. In order to respect priorities and use them to solve some
or possibly all of the conflicts, Brewkas original algorithm from [35], [36]
(cf. above sec. 3.3.4) is then employed to construct the preferred maximally obeyable subsets by testing at each step whether adding a next-lower
imperative leaves the constructed set obeyable. Instead of I, these preferred
maximally obeyable subsets are then used in the above definitions to describe
as obligatory what is, for all these sets, necessary to satisfy all the imperatives in the set that are triggered, triggered etc. I show that all of the above
definitions produce the intuitively right result for the order puzzle, and so
the proposed method is also neutral with respect to the question which of
the above definitions should be chosen.
3.4
For a long time deontic logic has been plagued by so-called paradoxes,
seemingly counterintuitive formulas that are nevertheless valid in the main
systems of deontic logic. It might be questioned whether these are at all
solvable, or do not rather stem from different uses, in quite different contexts
of normative reasoning, of the natural language phrases it is obligatory that,
it ought to be that, it is required that, it is morally necessary that, it
has been commanded that that seemingly all are to be encompassed by the
one O-operator of deontic logic. A reaction to this situation may be, as
was proposed above, to define not one, but several O-operators that talk
differently about the norms. The paradoxes of deontic logic then serve as
benchmarks how well one particular definition of an O-operator reflects the
intended usage of the natural phrases in these benchmark examples.
3.4.1
Rosss Paradoxes
The first paradox, Rosss paradox, was originally levelled not at deontic logic,
but at the logic of imperatives as designed by Jrgensen [110] in accord with
Dubislavs proposal in [50] (cf. above sec. 1.5). Ross [188] showed that if
such imperative inferences are accepted, then it is possible to derive from
IMPERATIVE SEMANTICS
108
the imperative post the letter the imperative to post the letter or burn it,
which seems counterintuitive.
In deontic logic, both in von Wrights classical system and in the standard
system SDL, all instances of the scheme OA O(A B) are theorems. If p1
is meant to formalize the sentence John posts the letter, and p2 the sentence
John burns the letter, then Op1 O(p1 p2 ) is an instance of this scheme
and formalizes the natural language expression if John ought to post the
letter, then John ought to post the letter or burn it, which might be seen as
similarly problematic as the original imperative inference.20 In sec. 3.3.2 we
have seen that definitions of O-operators of the type O4 and O5 may be used
to reconstruct the classical and the standard system of deontic logic, and
so the formula Op1 O(p1 p2 ) is valid for such readings of the O-operator.
However, it should be noted that using the definitions of these operators,
Rosss paradox appears much less problematic than it did for imperative
inferences. Though both O4 and O5 describe the normative situation, the
truth of Op1 O(p1 p2 ) for both operators cannot be used to claim that
by burning the letter some, if only derived, imperative is satisfied. A good
model of the normative situation in case of Rosss paradox is to let the
set of imperatives be {!p1 }, i.e. there is (only) the imperative to John to
post the letter in the relevant set of imperatives. The definitions of O4
and O5 then make true Op1 , so John ought to post the letter, and, since
both operators define OA as true if and only if A is necessary to satisfy
some (or all) imperatives, also O(p1 p2 ) is true, for any situation that is
satisfactory with regard to the only imperative !p1 will also make p1 p2
true. However, we do not have that bringing about p1 p2 is also sufficient
to satisfy some imperatives: in particular, there is no imperative in the set
{!p1 } that will get satisfied by bringing about p2 , i.e. burning the letter. This
was different for Rosss original setting, where burning the letter satisfies the
derived imperative !(p1 p2 ), and so seemingly forces us to acknowledge that
in this case, John did at least something right, did something that satisfied
an imperative etc. Since the set of imperatives is not closed by any logic
of imperatives (there is none), there is also no imperative in the set that
gets satisfied by burning the letter. So I think that in this way, imperative
semantics really takes the edge off Rosss paradox.
20
Cf. Danielsson [47] for a recent claim that such a theorem is problematic and should
be avoided. However, his axiomatic proposal to eliminate the paradox does not work, as
shown in [79].
IMPERATIVE SEMANTICS
109
IMPERATIVE SEMANTICS
110
3.4.2
Contrary-to-duty Paradoxes
IMPERATIVE SEMANTICS
111
in the set of imperatives. In this sense, it is not true that Jane is under a
conditional obligation to commit adultery. Now consider what happens if
the set does in fact contain such a conditional, contrary-to-duty imperative,
e.g. one that in this case commands Jane to return the stolen goods. Let
p1 !p3 be this contrary-to-duty imperative, i.e. the set of imperatives is
now {!p1 , p1 !p3 }. The first thing to note here is that the first imperative
does not derive the second imperative there is no such inference relation
between imperatives. For modeling obligations that arise from conditional
imperatives in possibly contrary-to-duty circumstances, a number of definitions were proposed in [80]. They also take into account possible priority
relations, but we need not worry about these here. When we are not interested in how obligations in one situation may influence obligations in some
other situation, we can also draw the situation description outside the scope
of the dyadic operator, i.e. make the situation part of the model and not
the operator. The simplest definition of OA with respect to a conditional
imperative structure I = I, f, g (cf. sec. 3.3.5) and a set of facts W is then
the following:
()
I, W |= OA
iff
W : f (Triggered(W, ))
PL
IMPERATIVE SEMANTICS
112
IMPERATIVE SEMANTICS
113
I, W |= OA
iff
and then OA holds only for those A that are equivalent to either p1 p2 , p1
or p2 , i.e. only going, calling and going-and-calling is obligatory. Now change
the set of facts into W = {p1 }: Chisholms sentences tell us that the man is
not going to go. The set of relevant imperatives shrinks to {p1 !p2 }: it
is the only imperative triggered and still possible to satisfy given our updated
knowledge. Both definitions () and () make OA true just for those A that
are necessary for the truth of p2 , i.e. for not telling the neighbors that he
is coming, and the stronger definitions () and () define OA true only for
those A that are equivalent to p2 . So not calling his neighbors is the only
obligation left. If we add the definition
I, W |= A iff W PL A,
for all non-deontic, propositional formulas A LPL , then in this situation
also p1 is described as true, which is what also might be demanded regarding
Chisholms paradox. Thus I think Chisholms paradox is adequately treated
in imperative semantics.
3.4.3
Finally, there have appeared paradoxes for deontic logic that also address
contrary-to-duty imperatives, but in these scenarios the contrary-to-duty situation gives rise to a special obligation to violate a primary rule. One of these
paradoxes, termed the paradox of the considerate assassin describes the following situation: the set of imperatives is {!k, !c, k !c}. The imperative
!k is intended to mean that you should not kill the witness, !c means
IMPERATIVE SEMANTICS
114
that you should not offer anyone a cigarette (including the witness), and
k !c means that if you kill the witness, you should offer her a cigarette
first. Prakken & Sergot [176], who first introduced this example (though in a
deontic setting, not a setting with explicitly represented imperatives), argue
that the solution should make Oc true for the situation k, as this applies
the imperative that is more specific for the given circumstances. However,
it is easy to see that e.g. truth definition (), with the set of imperatives as
described and the set of facts W = {k}, does not make Oc true: the set of
maximally obeyable subsets of I in this situation is {{!c}, {k !c}}, i.e.
the assassin has the choice to either keep to the rule of not offering cigarettes,
or offering, in this special case, a cigarette. Both commands apply to the
situation, and it has not been explicitly specified that the second should take
priority over the first. Hence only O(cc) is true, and thus only tautologies
are described as being obligatory.
A similar idea underlies the second example, also first proposed by Prakken
& Sergot [176] and slightly modified by Makinson [141]. Here the set of imperatives is {!f, f !(f w), d !(f w)}: There is a general prohibition
of fences !f except if there already is one in that case it should be white,
i.e. f !(f w) or if the owner has a dog, in which case the owner
should have a white fence, i.e. d !(f w). Again, the more specific imperative is intended to be applied in the situation where there already is a
fence or there is a dog, and so for the situation f the sentence O(f w)
should be true, and likewise for the situation d. For the situation f , definition () produces the right result: the set of maximally obeyable subsets
in this situation is {{f !(f w), d !(f w)}}, the only triggered imperative in this set is f !(f w), and so Of w) is defined as true.
However, things are different for the situation where there is (only) a dog,
i.e. W = {d}: the imperative that there is to be no fence is not yet violated
and so still considered relevant, and so the set of maximally obeyable subsets is {{!f, f !(f w)}, {f !(f w), d !(f w)}}. In each of the
two subsets there is just one imperative that is triggered in the situation d,
namely !f and d !(f w) respectively, and so only O(f (f w)), and
what else is necessary to make f (f w)) true, is described as obligatory.
In particular, O(f w) is false.
While it is true that neither definition (), nor any other definition proposed for imperative semantics, includes a specificity test, I am not convinced that this is a defect. For default rules that describe a knowledge base,
it might be argued that facts learned for more specific circumstances should
IMPERATIVE SEMANTICS
115
override possibly incomplete more general knowledge, and so the rule that
birds fly should not be applied to penguins once we have learned that these
do not fly, but it still should be applied to other birds. For imperatives there
does not appear to be such a general rule of overriding specificity. A soldier who is ordered before a mission to, under no circumstances, use lethal
force, and is then ordered by another officer to, if meeting a certain enemy,
kill that enemy, might, upon meeting this enemy, wonder what to do even
if both commands stem from equally ranking officers. In law there is the
legal principle lex specialis derogat legi generali, but this is not universally
applicable to all sets of norms, particularly not if they stem from from different parts of the legal system, and when it competes with other principles
like lex posterior, lex superiori, or standard argument forms like teleological
interpretation. But it seems that if there is a case where a more specific
imperative should take priority, we can use a priority ordering to express
such relations between the imperatives. So, for the first example, let the priority ordering be !k < k !c < !c, and let the priority ordering include
d !(f w) < !f in the case of the second example (where i < j means
that the imperative i takes priority over the imperative j). For prioritized
conditional imperative structures I = I, f, g, < the definition () may be
redefined as follows:21
()
I, W |= OA iff PIo (W, I) : f (Triggered(W, )) PL A
Here, the set P o (W, I) is the set of preferred maximally obeyable subsets as
defined in [80] (cf. above sec. 3.3.5). For total priority orderings like the ones
in the above examples (we ignore the additional imperative f !(f w) in
the second example, the paradox pertaining to this imperative was already
solved) it suffices that for this case the (only) set in PIo (W, I) is constructed
by adding the highest-ranking imperative i1 to i1 if W {g(i1 ) f (i1 )}
is not inconsistent, adding the second highest-ranking to i2 if i1 W
{g(i2 ) f (i2 )} is not inconsistent, and so on, and letting be the union
of all sets i such that i I. So in the first example, where W = {k},
I = {!k, !c, k !c}, and the prioritization is !k < k !c < !c, the
set of relevant (preferred) imperatives is {k !c}: in the first step, !k is
not added since it is already violated in the situation k, in the second step
k !c is added since it is consistent with k, and in the third step nothing
is added since {k c, c} is not consistent with k. Because k !c is also
21
This is similar to definition (td -pcd 1-8) that was defined with respect to preferred
maximally obeyable subsets in [80].
IMPERATIVE SEMANTICS
116
triggered given the facts W = {k}, definition ( ) then makes true Oc. So if
the assassin is going to kill the witness, and the order to offer the witness
a cigarette first takes priority over the general rule not to offer cigarettes,
then the assassin ought to offer the witness a cigarette first. The solution
for the second example, and the situation that there is a dog, is constructed
similarly. Thus both paradoxes receive a fair solution.
I think that these paradoxes also show the usefulness of imperative semantics for deontic logic. Here, priority orderings pertain to given imperatives,
not to statements of obligation that are true in relation to them. If we only
had these statements, as in traditional deontic logic, a reasoning with priorities would be much harder: we would have to ask e.g. whether a formula
OA that derives from a formula O(A B) of known priority also inherits
its rank, and if it is exactly the same rank or some lower one. Likewise it
would be hard to determine the preferred subsets that are consistent with
the situation: we do not know whether the truth of OA holds because there
is an obligation to realize A, or perhaps an obligation to realize A B which
might no longer be relevant since B is true, or which might already be
overridden by a higher-ranking obligation to realize B but canceling the
obligation from which OA derives would force us to cancel also OA, unless it
also derives from some other obligation, but in that case its rank must now
perhaps be changed, etc. But statements of obligation are not true by themselves, they are true because there exist certain norms or imperatives, and it
seems much more natural to say that it is these that get canceled if they can
no longer be satisfied (or are already satisfied), or are overridden by other
norms. Imperative semantics makes these relations easier to understand and
to model the deontic propositions that are true with respect to the norms.
Chapter 4
Conclusion
In 1998, David Makinson pointed out a fundamental problem of deontic
logic: if deontic logic is supposed to be concerned with norms, how can it
be reconstructed in accord with the philosophical position that norms are
devoid of truth values? In reaction to Makinsons call I argue that norms, or
simpler: imperatives, should be explicitly represented in the logical semantics
of deontic logic. The approach proposed here, and the imperatival tradition
of deontic logic it follows, thus depicts deontic logic as a logic about norms,
not of norms.
If there is a logic of imperatives or norms, then perhaps this approach
would not be much worth. If, in the face of Jrgensens Dilemma, we can
e.g. use Dubislavs convention to derive an imperative !A from a set of
imperative !B1 , ..., !Bn if the set {B1 , ..., Bn } classically implies A, then an
assumption of closure of the set of norms, perhaps together with an assumption of normative consistency (exclusion of conflicts), yields a deontic logic
(a logic of propositions about what is obligatory according to the norms) that
is isomorphic to such a logic of norms. The logic of normative systems proposed by Stenius serves as a typical example: here deontic logic is restricted
to stating what prescriptions exist in a normative system that is closed under given rules, i.e. the deontic proposition OA is true if and only if the
prescription OA exists in the system of of norms S. From such a viewpoint,
deontic logic must then appear as a dull isomorphism or ersatz theory.
However, Rosss paradox, in which from the imperative post the letter the
imperative post the letter or burn it is derived, and Weinbergers variant,
the window paradox, in which from the imperative close the window and
play the piano the imperative play the piano is derived, have drawn into
117
CONCLUSION
118
CONCLUSION
119
The reconstructions demonstrate that all the work put into the construction of existing deontic logic systems has not been in vain. Underlying intuitions have been captured in axiomatic systems that are still sound and
complete with respect to a semantics that explicitly represents norms. In
particular, SDL is sound and complete for a semantics that defines a deontic
O-operator in terms of what is necessary to satisfy a set of imperatives, provided that they do not conflict, and Hansson- or Lewis-type standard dyadic
deontic systems are sound and complete with respect to a semantics that
defines a dyadic deontic O-operator in terms of what is necessary to satisfy
all those imperatives that still can be satisfied in possibly contrary-to-duty
circumstances, provided that all conflicts and dilemmas are taken care of by
a total priority relation. But aside of these standard systems, new systems
have appeared that allow us to e.g. meaningfully talk about obligations even
if a priority relation does not exist or is not total and therefore conflicts
must be taken into account, and new definitions are possible for operators
that e.g. single out obligations pertaining to particular imperatives, be they
overridden by stronger ones or not, or define what is necessary and sufficient
to satisfy at least one imperative. This multitude of possible definitions does
not, in my view, mean that deontic logic has become a logic `a la carte. It
rather responds to the different circumstances in which deliberation is required, and to different ways we talk about obligations, e.g. when we treat
them as being prima facie or all things considered.
Inevitably, with the reconstruction of existing deontic logic systems the
work has just begun. It started with a semantics that represents unconditional imperatives, but as I demonstrated in [80], the representation of
conditional imperatives presents difficult choices for the definition of an appropriate O-operator. Permission is a neglected concept in deontic logic, we
must find a way to represent explicitly given licenses which override prohibitions that may apply otherwise (it appears senseless to explicitly permit something that has not previously been, and never will be, prohibited).
Norms about the creation of new imperatives or licenses must be taken into
account. Some of these problems and other challenges have been outlined in
Hansen & Pigozzi & van der Torre [81]. However, I think that the reconstructions have shown that modeling natural language imperatives and the
reasoning of the subjects they address is a fruitful approach to deontic logic.
So I am confident that the imperatival tradition of deontic logic, to which my
work contributes, will continue to be useful in developing intuitive solutions
to these problems.
Acknowledgements
My work on imperative-based semantics for deontic logic would not have been
possible without the fruitful discussions at the biannual EON workshops
where the small deontic logic community meets. In particular I want to
thank Lennart
Aqvist, Lou Goble, Jeff Horty, David Makinson and Leon van
der Torre for their constant encouragement, helpful suggestions and always
well-placed criticism. All errors that remain are mine.
120
Bibliography
[1] Achenwall, G., Prolegomena Iuris Naturalis, In Usum Auditorum, 3rd
ed., Gottingen: Sumtibus Victorini Bossigeli, 1767.
[2] Alchourron, C. E., Logic of Norms and Logic of Normative Propositions, Logique & Analyse, 12, 1969, 242268.
[3] Alchourron, C. E., The Intuitive Background of Normative Legal Discourse and its Formalization, Journal of Philosophicl Logic, 1, 1972,
447463.
[4] Alchourron, C. E., Conditionality and the Representation of Legal
Norms, in: Martino, A. A. and Socci Natali, F. (eds.), Automated
Analysis of Legal Texts, Amsterdam: North Holland, 1986, 173186.
[5] Alchourron, C. E. and Bulygin, E., The Expressive Conception of
Norms, in [94] 95124.
[6] Alchourron, C. E. and Bulygin, E., Unvollstandigkeit, Widerspr
uchlichkeit und Unbestimmtheit der Normenordnungen, in: Conte,
A. G., Hilpinen, R. and von Wright, G. H. (eds.), Deontische Logik und
Semantik, Wiesbaden: Athenaion, 1977, 2032.
[7] Alchourron, C. E. and Bulygin, E., Pragmatic Foundations for a Logic
of Norms, Rechtstheorie, 15, 1984, 453464.
[8] Alchourron, C. E. and Bulygin, E., On the Logic of Normative Systems, in: Stachowiak, H. (ed.), Handbuch pragmatischen Denkens,
Hamburg: Meiner, 1993, 273293.
[9] Alchourron, C. E. and Makinson, D., Hierarchies of Regulations and
Their Logic, in [94] 125148.
121
BIBLIOGRAPHY
122
BIBLIOGRAPHY
123
[23]
Aqvist, L., Some Results on Dyadic Deontic Logic and the Logic of
Preference, Synthese, 66, 1986, 95110.
[24] Ayer, A. J., Language, Truth and Logic, New York: Dover Publ., 1938.
[25] Bailhache, P., Several Possible Systems of Deontic Weak and Strong
Norms, Logique & Analyse, 21, 1980, 89100.
[26] Bailhache, P., Essai de logique deontique,
Philosophique J. Vrin, 1991.
Paris:
Librairie
[27] Barcan Marcus, R., Iterated Deontic Modalities, Mind, 75, 1966,
580582.
[28] Becker, O., Untersuchungen u
ul, Meisenheim/Glan:
ber den Modalkalk
Anton Hain, 1952.
[29] Bentham, J., Of Laws in General, London: Athlone Press, 1970, edited
by H. L. A. Hart. First appeared 1782.
[30] Bochman, A., Credulous Nonmonotonic Inference, in: Dean, T.
(ed.), Proceedings of the 16th International Joint Conference on Artificial Intelligence, IJCAI 99, Stockholm, August 1999, San Francisco:
Morgan Kaufmann, 1999, 30 35.
[31] Bochman, A., A Logical Theory of Nonmonotonic Inference and Belief
Change, Berlin: Springer, 2001.
[32] Bochman, A., Brave Nonmonotonic Inference and Its Kinds, Annals
of Mathematics and Artificial Intelligence, 39, 2003, 101121.
[33] Bohnert, H. G., The Semiotic Status of Commands, Philosophy of
Science, 12, 1945, 302315.
[34] Boole, G., The Mathematical Analysis of Logic, Cambridge: Macmillan, Barclay, and Machmillan, 1847.
[35] Brewka, G., Preferred Subtheories: An Extended Logical Framework
for Default Reasoning, in: Sridharan, N. S. (ed.), Proceedings of
the Eleventh International Joint Conference on Artificial Intelligence
IJCAI-89, Detroit, Michigan, USA, August 20-25, 1989, San Mateo,
Calif.: Kaufmann, 1989, 10431048.
BIBLIOGRAPHY
124
BIBLIOGRAPHY
125
BIBLIOGRAPHY
126
[64] Goble, L., The Iteration of Deontic Modalities, Logique & Analyse,
9, 1966, 197209.
[65] Goble, L., A Logic of Good, Should, and Would, Journal of Philosophical Logic, 19, 1990, 169199.
[66] Goble, L., Preference Semantics for Deontic Logics: Part I Simple
Models, Logique & Analyse, 2003, 383418.
[67] Goble, L., Preference Semantics for Deontic Logics: Part II Multiplex Models, Logique & Analyse, 2004, 335363.
[68] Goble, L., A Proposal for Dealing with Deontic Dilemmas, in: Lomuscio, A. and Nute, D. (eds.), Deontic Logic in Computer Science
(Proceedings of the 7th International Workshop on Deontic Logic in
Computer Science, DEON 2004, Madeira, May 2004), LNAI 3065,
Berlin: Springer, 2004, 75 113.
[69] Goble, L., A Logic for Deontic Dilemmas, Journal of Applied Logic,
3, 2005, 461483.
[70] Goble, L. and Lokhorst, G.-J. C., Mallys Deontik, Grazer
Philosophische Studien, 67, 2004, 3757.
[71] Gombay, A., What is Imperative Inference?, Analysis, 27, 1967, 145
152.
[72] Greenspan, P., Conditional Oughts and Hypothetical Imperatives,
Journal of Philosophy, 72, 1975, 259276.
[73] Grelling, K., Zur Logik der Sollsatze, Unity of Science Forum, 1939,
4447.
[74] Hamblin, C. L., Imperatives, Oxford: Blackwell, 1987.
[75] Hansen, J., Sets, Sentences, and Some Logics about Imperatives,
Fundamenta Informaticae, 48, 2001, 205226.
[76] Hansen, J., Problems and Results for Logics about Imperatives,
Journal of Applied Logic, 2, 2004, 3961.
BIBLIOGRAPHY
127
[77] Hansen, J., Conflicting Imperatives and Dyadic Deontic Logic, Journal of Applied Logic, 3, 2005, 484511.
[78] Hansen, J., Deontic Logics for Prioritized Imperatives, Artificial Intelligence and Law, 14, 2006, 134.
[79] Hansen, J., The Paradoxes of Deontic Logic: Alive and Kicking,
Theoria, 72, 2006, 221232.
[80] Hansen, J., Prioritized Conditional Imperatives: Problems and a New
Proposal, Autonomous Agents and Multi-Agent Systems, 17, 2008,
1135.
[81] Hansen, J., Pigozzi, G. and van der Torre, L., Ten Philosophical Problems in Deontic Logic, in: Boella, G., van der Torre, L. and Verhagen,
H. (eds.), Normative Multi-Agent Systems, no. 07122 in Dagstuhl Seminar Proceedings, Internationales Begegnungs- und Forschungszentrum
f
ur Informatik (IBFI), Schloss Dagstuhl, Germany, 2007.
[82] Hanson, W. H., Semantics for Deontic Logic, Logique & Analyse, 31,
1965, 177190.
[83] Hanson, W. H., A Logic of Commands, Logique & Analyse, 9, 1966,
329343.
[84] Hansson, B., An Analysis of Some Deontic Logics, Nous, 3, 1969,
373398, reprinted in [93] 121147.
[85] Hansson, S. O., Preference-based Deontic Logic (PDL), Journal of
Philosophical Logic, 19, 1990, 7593.
[86] Hare, R. M., Imperative Sentences, Mind, 58, 1949, 2139.
[87] Hare, R. M., The Language of Morals, Oxford: Oxford University
Press, 1952.
[88] Hare, R. M., Some Alleged Differences Between Imperatives and Indicatives, Mind, 76, 1967, 309326.
[89] Harris, Z. S., Transformational Theory, Language, 41, 1965, 363
401.
BIBLIOGRAPHY
128
[90] Harrison, J., Deontic Logic and Imperative Logic, in: Geach, P. T.
(ed.), Logic and Ethics, Dordrecht: Kluwer, 1991, 79129.
[91] Hempel, C. G., Review of Alf Ross: Imperatives and Logic, Journal
of Symbolic Logic, 6, 1941, 105106.
[92] Hilpinen, R., On Deontic Logic, Pragmatics and Modality, in: Stachowiak, H. (ed.), Pragmatik: Handbuch des pragmatischen Denkens,
vol. 4, Hamburg: Felix Meiner.
[93] Hilpinen, R. (ed.), Deontic Logic: Introductory and Systematic Readings, Dordrecht: Reidel, 1971.
[94] Hilpinen, R. (ed.), New Studies in Deontic Logic, Dordrecht: Reidel,
1981.
[95] Hintikka, J., Some Main Problems of Deontic Logic, in [93] 59104.
[96] Hintikka, J., The Ross Paradox as Evidence for the Reality of Semantical Games, in: Saarinen, E. (ed.), Game-Theoretical Semantics,
Dordrecht: Reidel, 1977, 329345.
[97] Hintikka, J. K., Quantifiers in Deontic Logic, Helsingfors: Societas
Scientiarum Fennica, Commentationes Humanarum Litterarum 23:4,
1957.
[98] Hodges, W., Logic, Harmondsworth: Penguin, 1977.
[99] Hofstadter, A. and McKinsey, J. C. C., On the Logic of Imperatives,
Philosophy of Science, 6, 1938, 446457.
[100] Hollander, P., Rechtsnorm, Logik und Wahrheitswerte. Versuch
einer kritischen Losung des Jorgensenschen Dilemmas, Baden-Baden:
Nomos, 1993.
[101] Horty, J. F., Moral Dilemmas and Nonmonotonic Logic, Journal of
Philosophical Logic, 23, 1994, 3565.
[102] Horty, J. F., Nonmonotonic Foundations for Deontic Logic, in: Nute,
D. (ed.), Defeasible Deontic Logic, Dordrecht: Kluwer, 1997, 1744.
BIBLIOGRAPHY
129
[103] Horty, J. F., Agency and Deontic Logic, Oxford: Oxford University
Press, 2000.
[104] Horty, J. F., Reasoning with Moral Conflicts, No
us, 37, 2003, 557
605.
[105] Horty, J. F., Defaults with Priorities, Journal of Philosophical Logic,
36, 2007, 367413.
[106] Hruschka, J., Das deontologische Sechseck bei Gottfried Achenwall im
Jahre 1767, Gottingen: Vandenhoek & Ruprecht, 1986.
[107] Hume, D., A Treatise of Human Nature, Oxford: Oxford University
Press, 1888, edited by L. A. Selby-Bigge. First appeared 1739.
[108] Husserl, E., Logische Untersuchungen, vol. 1, Halle: Max Niemeyer,
1900.
[109] Jacquette, D., Moral Dilemmas, Disjunctive Obligations, and Kants
Principle that Ought implies Can, Synthese, 88, 1991, 4355.
[110] Jrgensen, J., Imperatives and Logic, Erkenntnis, 7, 1938, 288296.
[111] Kalinowski, G., Le Probl`eme de la Verite en Morale et en Droit, Lyon:
Emmanuel Vitte, 1967.
[112] Kalinowski, G., Einf
uhrung in die Normenlogik, Frankfurt/M.:
Athenaum, 1973.
BIBLIOGRAPHY
130
BIBLIOGRAPHY
131
[129] Ledig, G., Zur Klarung einiger Grundbegriffe. Imperativ, Rat, Bitte,
Beschlu, Versprechen, Revue Internationale de la Theorie du Droit,
3, 1928/29, 260270.
[130] Ledig, G., Zur Logik des Sollens, Der Gerichtssaal, 100, 1931, 368
385.
[131] Leibniz, G. W., Elementa Juris Naturalis [2nd half 1671, holographic
concept B (Hannover)], in: Preussische Akademie der Wissenschaften
(ed.), Gottfried Wilhelm Leibniz: Samtliche Schriften und Briefe, vol. 1
of 6. Reihe: Philosophische Schriften, Darmstadt: Otto Reichl, 1930,
465485.
[132] Lemmon, E. J., Moral Dilemmas, Philosophical Review, 71, 1962,
139158.
[133] Lemmon, E. J., Deontic Logic and the Logic of Imperatives, Logique
& Analyse, 8, 1965, 3971.
[134] Lemmon, E. J., Beginning Logic, 2nd ed., London: Chapman and Hall,
1987.
[135] Leonard, H. S., Interrogatives, Imperatives, Truth, Falsity and Lies,
Philosophy of Science, 26, 1959, 172186.
[136] Lewis, D., Counterfactuals, Oxford: Basil Blackwell, 1973.
[137] Lewis, D., Semantic Analyses for Dyadic Deontic Logic, in: Stenlund,
S. (ed.), Logical Theory and Semantic Analysis, Dordrecht: Reidel,
1974, 1 14.
[138] Lewis, D., Ordering Semantics and Premise Semantics for Counterfactuals, Journal of Philosophical Logic, 10, 1981, 217234.
[139] MacKay, A. F., Inferential Validity and Imperative Inference Rules,
Analysis, 29, 1969, 145156.
[140] Makinson, D., General Patterns in Nonmonotonic Reasoning, in:
Gabbay, D. M., Hogger, C. and Robinson, J. (eds.), Handbook of Logic
in Artificial Intelligence and Logic Programming, vol. 3: Nonmonotonic
Reasoning and Uncertain Reasoning, Oxford: Clarendon Press, 1994,
35 110.
BIBLIOGRAPHY
132
[141] Makinson, D., On a Fundamental Problem of Deontic Logic, in: McNamara, P. and Prakken, H. (eds.), Norms, Logics and Information
Systems, Amsterdam: IOS, 1999, 2953.
[142] Makinson, D. and van der Torre, L., Input/Output Logics, Journal
of Philosophical Logic, 29, 2000, 383408.
[143] Makinson, D. and van der Torre, L., Constraints for Input/Output
Logics, Journal of Philosophical Logic, 30, 2001, 155185.
[144] Makinson, D. and van der Torre, L., What is Input/Output Logic,
in: Lowe, B., Malzkom, W. and Rasch, T. (eds.), Foundations of the
Formal Sciences II : Applications of Mathematical Logic in Philosophy
and Linguistics (Papers of a conference held in Bonn, November 10-13,
2000), Trends in Logic, vol. 17, Dordrecht: Kluwer, 2003, 163174.
[145] Mally, E., Grundgesetze des Sollens. Elemente der Logik des Willens,
Graz: Leuschner & Lubensky, 1926.
[146] Marek, V. W. and Truszczy
nski, M., Nonmonotonic Logic. ContextDependent Reasoning, Berlin: Springer, 1993.
[147] Mates, B., Elementary Logic, 2nd ed., Oxford: Oxford University Press,
1972.
[148] Mazzarese, T., Norm Propositions:
Queries, Rechtstheorie, 22, 1991, 3970.
[149] Meggle, G. (ed.), Actions, Norms, Values. Discussions with Georg Henrik von Wright, Berlin: de Gruyter, 1999.
[150] Menger, K., A Logic of the Doubtful. On Optative and Imperative Logic, Reports of a Mathematical Colloquium, 2, 1939, 5364,
reprinted in [151] 91102.
[151] Menger, K., Selected Papers in Logic and Foundations, Didactics, Economics, Dordrecht: Reidel, 1979.
[152] Meyer, J.-J. C., A Different Approach to Deontic Logic: Deontic Logic
Viewed as a Variant of Dynamic Logic, Notre Dame Journal of Formal
Logic, 29, 1988, 109136.
BIBLIOGRAPHY
133
[153] Moritz, M., Der praktische Syllogismus und das juridische Denken,
Theoria, 20, 1954, 78127.
[154] Mott, P. L., On Chishoms Paradox, Journal of Philosophical Logic,
2, 1973, 197211.
[155] Moutafakis, N. J., Imperatives and their Logic, New Delhi: Sterling,
1975.
[156] Niiniluoto, I., Truth and Legal Norms, Archiv f
ur Rechts- und Sozialphilosophie Beiheft, 25, 1985, 168190.
[157] Niiniluoto, I., Hypothetical Imperatives and Conditional Obligations, Synthese, 66, 1986, 111133.
[158] Niiniluoto, I., Norm Propositions Defended, Ratio Juris, 4, 1991,
367373.
[159] Nortmann, U., Deontische Logik ohne Paradoxien. Semantik und Logik
des Normativen, M
unchen: Philosophia Verlag, 1989.
[160] Nowell-Smith, P. H. and Lemmon, E. J., Escapism: the Logical Basis
of Ethics, Mind, 69, 1960, 289300.
[161] Nute, D. and Yu, X., Defeasible Deontic Logic. Introduction, in:
Nute, D. (ed.), Defeasible Deontic Logic, Dordrecht: Kluwer Academic
Publishers, 1997, 116.
[162] Opalek, K., On the Logical-Semantic Structure of Directives, Logique
& Analyse, 13, 1970, 169196.
[163] Opalek, K., Theorie der Direktiven und Normen, Wien: Springer, 1986.
[164] Opalek, K. and Wole
nski, J., Is, Ought, and Logic, Archiv f
ur Rechtsund Sozialphilosophie, 73, 1987, 373385.
[165] Opalek, K. and Wole
nski, J., Normative Systems, Permission and
Deontic Logic, Ratio Juris, 4, 1991, 334348.
BIBLIOGRAPHY
134
[167] Peczenik, A., Norms and Reality, Theoria, 34, 1968, 117133.
[168] Philipp, P., Logik deskriptiver normativer Begriffe, 1989/90, In [170]
pp. 241290. First published as reports of the Karl-Marx-Universitat
Leipzig, Untersuchungen zur Logik und zur Methodologie 6:6588, 1989,
and 7:5182, 1990.
[169] Philipp, P., Normative Logic Without Norms, 1991, In [170] pp. 291
301. Edited version of a presentation to the Meeting of the Central
Division of the American Philosophical Association, Chicago, 1991.
[170] Philipp, P., Logisch-philosophische Untersuchungen. Edited by Ingolf
Max and Richard Raatzsch., Berlin: de Gruyter, 1998.
[171] Philipps, L., Braucht die Rechtswissenschaft eine deontische Logik?,
in: G., J. and Maihofer, W. (eds.), Rechtstheorie. Beitrage zur Grundlagendiskkussion, Frankfurt: Klostermann, 1971, 352368.
[172] Poincare, H., Derni`eres Pensees, Paris: Ernest Flammarion, 1913.
[173] Poole, D., A Logical Framework for Default Reasoning, Artificial
Intelligence, 36, 1988, 2747.
[174] Powers, L., Some Deontic Logicians, Nous, 1, 1967, 380400.
[175] Prakken, H., Logical Tools for Modelling Legal Argument, Dordrecht:
Kluwer, 1997.
[176] Prakken, H. and Sergot, M., Contrary-to-duty obligations., Studia
Logica, 52, 1996, 91115.
[177] Prior, A. N., The Paradoxes of Derived Obligation, Mind, 63, 1954,
6465.
[178] Prior, A. N., Formal Logic, Oxford: Clarendon Press, 1955.
[179] Prior, A. N., Escapism: the Logical Basis of Ethics, in: Melden, A. I.
(ed.), Essays in Moral Philosophy, Seattle: University of Washington
Press, 1966, 135146.
[180] Reichenbach, H., Elements of Symbolic Logic, New York: CollierMacmillan, 1947.
BIBLIOGRAPHY
135
[181] Reiter, R., A Logic for Default Reasoning, Artificial Intelligence, 13,
1980, 81132.
[182] Rescher, N., An Axiom System for Deontic Logic, Philosophical Studies, 9, 1958, 2430.
[183] Rescher, N., Hypothetical Reasoning, Amsterdam: North-Holland,
1964.
[184] Rescher, N., The Logic of Commands, London: Routledge & Kegan
Paul, 1966.
[185] Rescher, N., Assertion Logic, in: Rescher, N. (ed.), Topics in Philosophical Logic, chap. XIV, Dordrecht: Reidel, 1968.
[186] Rodig, J., Kritik des Normlogischen Schlieens, Theory and Decision,
2, 1971, 7993.
BIBLIOGRAPHY
136
BIBLIOGRAPHY
137
[207] Stranzinger, R., Ein paradoxienfreies deontisches System, in: Tammelo, I. and Schreiner, H. (eds.), Strukturierungen und Entscheidungen
im Rechtsdenken, Wien: Springer, 1978, 183193.
[208] Swirydowicz,
K., Normative Consequence Relation and Consequence
Operations on the Language of Dyadic Deontic Logic, Theoria, 60,
1994, 2747.
[209] Swirydowicz,
K., A New Approach to Dyadic Deontic Logic and the
Normative Consequence Relation, in: Murawski, R. and Pogonowski,
J. (eds.), Euphony and Logos. Essays in Honour of Maria Steffen-Batog
and Tadeusz Batog, Amsterdam/Atlanta: Rodopi, 1997, 283317.
BIBLIOGRAPHY
138
BIBLIOGRAPHY
139
Blackwell,
[240] Wole
nski, J., Deontic Logic and Possible Worlds Semantics: A Historical Sketch, Studia Logica, 49, 1990, 273282.
[241] Wole
nski, J., Remarks on Mallys Deontik, in: Hieke, A. (ed.), Ernst
Mally Versuch einer Neubewertung, St. Augustin: Akademia, 1998,
7380.
[242] von Wright, G. H., Deontic Logic, Mind, 60, 1951, 115.
[243] von Wright, G. H., An Essay in Modal Logic, Amsterdam: North Holland, 1951.
[244] von Wright, G. H., A Note on Deontic Logic and Derived Obligation,
Mind, 65, 1956, 507509.
BIBLIOGRAPHY
140
[245] von Wright, G. H., Logical Studies, London: Routledge and Kegan
Paul, 1957.
[246] von Wright, G. H., Norm and Action, London: Routledge & Kegan
Paul, 1963.
[247] von Wright, G. H., A New System of Deontic Logic, Danish Yearbook
of Philosophy, 1, 1964, 173182, reprinted in [93] 105115.
[248] von Wright, G. H., A Correction to a New System of Deontic Logic,
Danish Yearbook of Philosophy, 2, 1965, 103107, reprinted in [93] 115
120.
[249] von Wright, G. H., An Essay in Deontic Logic and the General Theory
of Action, Amsterdam: North Holland, 1968.
[250] von Wright, G. H., Normenlogik, in: Lenk, H. (ed.), Normenlogik.
Grundprobleme der deontischen Logik, Pullach b. M
unchen: Verlag
Dokumentation, 1974, 25 38.
[251] von Wright, G. H., Problems and Prospects of Deontic Logic, in:
Agazzi, E. (ed.), Modern Logic: A Survey, Dordrecht: D. Reidel, 1980,
399423.
[252] von Wright, G. H., Norms, Truth and Logic, in: von Wright, G. H.
(ed.), Practical Reason: Philosophical Papers vol. I, Oxford: Blackwell,
1983, 130209.
[253] von Wright, G. H., Bedingungsnormen, ein Pr
ufstein f
ur die Normenlogik, in: Krawietz, W., Schelsky, H., Weinberger, O. and Winkler,
G. (eds.), Theorie der Normen, Berlin: Duncker & Humblot, 1984,
447456.
[254] von Wright, G. H., Is and Ought, in: Bulygin, E., Gardies, J.-L. and
Niiniluoto, I. (eds.), Man, Law and Modern Forms of Life, Dordrecht:
D. Reidel, 1985, 263281.
[255] von Wright, G. H., Is there a Logic of Norms?, Ratio Juris, 4, 1991,
265283.
[256] von Wright, G. H., A Pilgrims Progress, in: von Wright, G. H. (ed.),
The Tree of Knowledge and Other Essays, Leiden: Brill, 1993, 103113.
BIBLIOGRAPHY
141
142
143
205
IOS Press
Abstract. Though deontic logic is regarded as the logic of normative reasoning, norms as entities
lacking truth values are usually represented neither in its language nor its semantics. Limiting
ourselves to unconditional imperatives, we propose a concept for their semantic representation and
show that existing systems of monadic and dyadic deontic logic can be reconstructed accordingly.
Keywords: deontic logic, logic of imperatives
The ideas of this paper were motivated by discussions at the EON 98 conference. The analysis of dyadic deontic logics
considerably gained from Wlodek Rabinowicz critical remarks on an earlier version of this paper, and the paper entirely
profited from Leon van der Torres constant encouragement and well-placed criticism. I am obliged to an anonymous referee
for many helpful comments.
Address for correspondence: Institut fur Philosophie, Universitat Leipzig, Leipzig, Germany
144
truth-functional operators. Hence it seems there cannot be a logic of norms this is Jrgensens dilemma
[28].
Though norms are neither true nor false, one may state that according to the norms, something ought
to be (be done) or is permitted: the statements John ought to leave the room, Mary is permitted to enter, are then true or false descriptions of the normative situation. Such statements are sometimes called
normative statements, as distinguished from norms. To express principles such as
(Conj)
with Boolean operators having truth-functional meaning at all places, deontic logic has resorted to interpreting its formulas , , as representing such normative statements. 1 A possible logic of
normative statements may then reflect logical properties of underlying norms thus logic may have a
wider reach than truth, as von Wright [54] famously stated.
Norms, for the reasons given, are absent in the language of deontic logic. But since the truth of
normative statements depends on a normative situation, it seems that norms may be represented in a
logical semantics that models such truth or falsity. Standard deontic semantics evaluates deontic formulas
with respect to sets of worlds, in which some are ideal or better than others. Deviating from this
standard approach, a semantics that seeks to represent norms has been proposed by several authors,
namely Erik Stenius, Stig Kanger, Carlos Alchourron and Eugenio Bulygin, and Bas van Fraassen.
Stenius [47] uses a set of prescriptively interpreted norm sentences, each norm sentence having the
form , and being a Boolean formula. The following principle restricts :
, then
is then defined as
is true if and only if (iff)
(r-St)
if
and
The truth of descriptively interpreted sentences
(td-St)
Obviously, is a normal modal operator. Note, however, that Stenius closure principle yields Ross
paradox [44], where the imperative Post the letter or burn it! is derived from the imperative Post
the letter!: even the forbidden will satisfy a norm sentence in any set that conforms to (r-St). Also,
it is quite a matter of dispute whether norms contain indicatives 2 , and Stenius does in fact define as
sentence radical, so the formulation of his principle can be criticized, cf. Makinson [34]. Kanger
[29] and Alchourron/Bulygin [2] instead use a set of descriptive sentences, where Kanger thinks of these
sentences as representing a welfare program for the universe of discourse, and to Alchourron/Bulygin
the sentences in their set are the sentences commanded by an authority. Their truth definition for
then reads
is true iff
(td-AB)
is the consequential
closure of . Again, is normal. Ross paradox can be argued not to
arise: yields for any but does not necessarily succeed in satisfying a norm
since is not closed under consequence. Yet Kanger and Alchourron/Bulygin make it appear as if
norms can somehow be replaced by descriptive sentences in normative reasoning an approach termed
reductionist by some philosophers (cf. Moutafakis [37], Hamblin [17]). But sentences are true or false,
1
Cf.Fllesdal/Hilpinen [16]. This already holds for Mallys deontic logic [35], cf. p. 12, where he writes that to interpret !
as the theoretical statement it is a fact that ought to be is the standpoint of (his) theory that describes obligating norms and
their laws.
2
Some believe imperatives contain indicatives as their propositional core (Sosa [45], also cf. Alchourron/Bulygin [3]),
whereas others deny any straight-forward grammatical relation and more cautiously speak of a thematically parallel sentence
(Opaek [39] p. 35).
145
207
norms are not, so semantics that claims to be about norms should describe the gap, and explain how it
can be bridged. And certainly norms have their special properties: the implication that any feature one
may want to semantically model time of promulgation, authority, subject, hierarchical position, etc.
must relate to a sentence which is but one of these features, not the norm itself, seems hardly acceptable.
Van Fraassen [50] has a set of imperatives in force, where to each imperative there corresponds
(td-vF1)
is true iff there is an such that
Van Fraassen then introduces the notion of a score relative to an outcome and the set of imperatives
, where the score of the outcome is the subset of imperatives it satisfies. His final truth definition reads
(td-vF2)
is true iff there is a such that for all
where is the score of the outcome among the set of imperatives . Van Fraassens suggestive
paper extends this analysis to conditional imperatives, where these relate to sentences that describe the
opportunity for satisfying the imperative. The idea of a score has been further examined by Horty [26],
also cf. [25] p. 284, and related research on the subject of conditional imperatives was carried out by
Makinson [34].
According to Makinson [34], the fundamental problem of deontic logic is to reconstruct it such
that it is in accord with the position that norms are devoid of truth values. I propose a semantic platform, related to the above semantics, on which such reconstructions can be carried out. The paper is
confined to unconditional obligating norms (imperatives, commands), and will not treat explicit permissions, derogations, general norms, or other types. The basic semantics is given in section 2. In section
3 we provide reconstructions for systems of monadic deontic logic. In section 4 we consider the reduction of deontic logic to alethic modal logic by Anderson [7] and Kanger [29], reconstructing it such that
the Andersonian-Kangerian constants have definite meanings. Since the late Fifties, deontic logics have
employed dyadic operators to formalize conditional obligations. In section 5 we propose reconstructions
for well-known systems of dyadic deontic logics. The results may be viewed as justifying and explaining
the view that deontic logic is a logic of reasoning about, if not of, norms.
2. Imperative Structures
2.1. The Basic Concept
What is needed to model imperatives in a logical semantics? Imperatives have a source, and a recipient,
and some authority must be assumed on the part of the source, of which the recipient is a subject (cf.
Rescher [43] ch. 2). The decisive component of an imperative is held to be its theme of demand: a
state the authority wants to see realized, or an action the subject ought to perform. If this state is realized
or the action performed, then the imperative is satisfied. We can identify this state or action by using a
descriptive sentence which corresponds to the imperative and is true just when the imperative is satisfied.
This sentence will usually be rather complex, e.g. it must spell out whether chance fulfilment or an
attempt suffices, and special attention must be paid to seemingly ordinary connectives and quantifiers.
Hence transforming the exact words of the imperative into indicative mood is just an initial approach.
That there nevertheless is such a correspondence between imperatives and indicatives is even accepted
146
by the anti-reductionists (cf. Weinberger [60] p. 69, Moutafakis [37] T3 pp. 155-156, Hamblin [17]
pp. 112, 139/140, 152-153), so we may formulate the following postulate:
(W) To each imperative there corresponds a descriptive sentence that is true if the imperative is satisfied,
and false if it is not-satisfied (violated).
Let an imperative structure therefore be a pair , where is a (possibly empty) set of objects;
they are meant to be imperatives. We make no assumptions about the nature of : it may be the set
of imperatives from a specific source, directed towards a certain subject, belonging to a system, being
valid or in force at a certain time, or derivable by some here unspecified procedure from some other
set, e.g. as individual obligations from more general norms. The function
associates a
sentence in the language
of a basic logic BL with each imperative in . These are meant to be the
descriptive sentences that correspond to imperatives. In the reconstructions, BL will be a propositional
logic defined below. But BL may be any logic that accommodates concepts of truth, semantic validity
of a sentence (we write
), and satisfiability, so we could assume BL to be a logic of act
propositions (cf. Alchourron/Bulygin [3]) or a logic that represents time and intentionality. This is our
basic platform to develop and reconstruct specific deontic logics. However, this platform is open to, and
requires, extensions and restrictions to model further qualities of imperatives.
(r-1)
For all :
(r-2)
is -satisfiable
(r-3)
For all :
(r-4)
If
then there exists an such that
Someone taking the view that imperatives express a will, and that a will cannot be directed towards
things that are logically impossible, and so imperatives that demand the impossible cannot exist (cf.
Kelsen [31] p. 171), may want to restrict the semantics of imperative structures by (r-1). Kelsen
[30] p. 136 argued that conflicting norms cannot coexist within the same system. Then (r-2) might be
adequate, demanding that all imperatives must be satisfiable collectively. 3 It has often been assumed
that the nature of imperatives is to direct, to serve as a means to change present facts or prevent unwanted
changes taking place (cf. Jrgensen [28]). (r-3) might then serve to exclude the case that an imperative
demands what is true on logical grounds alone (cf. Weinberger [59] p. 121, Alchourron/Bulygin [4] p.
284). Alchourron/Bulygin [4] also believe that one cannot command without implicitly commanding
all logical consequences of the commands contents, and then (r-4) might be accepted. We have already
noted that Ross paradox is connected with such closure conditions.
The effect of employing semantic restrictions is that imperative structures that do not conform to
these are excluded from analysis. For a general exclusion the grounds must be very good indeed, and
147
209
though the above may appear reasonable, we hesitate to integrate them into the basic concept. Still,
restrictions might be useful if the behaviour of specific imperative structures that satisfy these is to be
examined. In this sense, restrictions are a primary tool to reconstruct specific deontic logics on the basis
of our concept.
such that
such that
such that
According to (td-1) is true iff is the descriptive sentence correlated to some imperative. An such
that is defined true by (td-1) might be considered an answer to the question What sentence (literally,
grammatically) corresponds to a given imperative?. 4 If the primary interest is not in the wording,
but the
-logical meaning we assume of the imperative-correlated sentences, we may instead employ
(td-2). Each of a sentence defined true by this definition may then be considered an answer to the
question What describes a state logically necessary and sufficient to satisfy one imperative?. (td-3) is
similar to van Fraassens truth definition in [50]. Each of a sentence defined true by this definition
(td-1)
(td-2)
(td-3)
(td-4)
answers the question What is necessary to satisfy one imperative?. A subject may be interested not
so much in single imperatives, but what must be achieved to have all imperatives satisfied. Then, (td4), similar to the truth definition of Alchourron/Bulygin in [2], defines each of a sentence
to be
an answer to the question What is logically necessary to satisfy all imperatives?. We are at liberty to
mix
the above,
and
any other
deontic operators
we choose to define, and can then state truths such as
Except for imperatives with identical corresponding sentences, the set of (td-1)-true -sentences will mirror the set of imperatives. So (td-1) comes close to an operator that allows us to directly talk about imperatives (cf. van Fraassen [50] p. 18 for the
need for such operators).
148
and let
be shorthand for
so
is the imperative structure we obtain
from adding a new imperative to and expanding such that
. Let
be the name of
logic about imperatives that defines deontic operators with respect to an imperative structure . Since
the semantics of may employ semantic restrictions, we call an imperative structure that conforms with
is a -structure, and
is -covered by
iff
is not a -structure
is -rejected by
iff
is a -structure, and
is -independent from
iff
is -covered by , then from the point of someone using as a tool for the analysis of a set of
If
imperatives it does not matter whether this imperative-candidate is added to the set or not, since the set
of -truths does not change.6 On the other hand, someone who believes that imperative structures should
conform to the semantic restrictions of could not tolerate the addition of an imperative-candidate
if
from , then whether it should be added to or not seems not disputable on -logical grounds alone.
We cannot further explore this topic here. But it seems important that the above semantics may not
only be used to construct descriptively interpreted deontic operators, but also to explain (meta-)logical
relations between prescriptions.
Our basic logic will be propositional logic given in the usual fashion: The alphabet has a set Prop of
proposition letters , logical signs , , , , , and brackets , . All proposition
letters are PL-sentences, and if and are PL-sentences, so are
(negation),
(conjunction),
(disjunction),
(conditional), and
(biconditional), and nothing else. Outer
brackets and corner quotes for mixed expressions are mostly omitted. The language
is the set of
PL-sentences. A valuation
assigns a truth value to each
. B is the set of
all such valuations. Truth of a PL-sentence at a valuation (we write
is defined recursively,
starting with
iff
iff
is not true
The idea of constructing normative logics as built on top of logics of normative statements appears in Alchourron/Bulygin
[3] p. 463, whereas in Alchourron/Bulygin [4] p. 285 the relation is vice versa. The idea of defining the normative status of
a considered norm by testing what happens to a system if this norm is added is the basis of von Wrights logic of normative
entailment (cf. [57], [58]).
6
The term coverage is borrowed from Rescher [43] ch. 6. The notion is akin to that of acceptance used in default logic, cf.
van der Torre [48].
149
211
write
). We suppose a sound and complete axiomatic system for PL.
provable in this system.
means that
is
is true iff
and
are co-satisfiable.
Both definitions refer to a satisfaction of imperatives; in the second the terms of ordinary satisfiability
and satisfaction of an imperative even appear mixed. This seems more that a semantic coincidence if
we accept postulate (W), for then we know a set of imperatives to be satisfiable iff the set of correlated
sentences is satisfiable, and can rephrase the definitions accordingly. Both definitions also refer to
the imperatives. However, the question is what to do if there are none. Instead of rejoicing as a
normal person would, most deontic logicians churlishly exclude this possibility by demanding that at
least something must be obligatory. We want to cover the case where the imperative set is empty, and
then nothing naturally appears to be necessary to satisfy the imperatives or as opposing an imperatives
false, and
true, for all sentences . 7
satisfaction. So the truth definitions then have
, except that additionally
The language of analysis will be , where the alphabet is like that of
we have and . Then is the smallest set such that
then and
then so are and
iff
and
iff
or is -satisfiable
We use the following restrictions:
is -satisfiable
(R1)
(R2)
(a)
(b)
If
If
(R1) excludes that the imperatives are, for reasons of the logic of their correlated sentences, not or not
all satisfiable. Otherwise we may have the result that everything is obligatory everything is necessary
to satisfy the imperatives though nothing can and nothing is permitted but it is not particular states
but the imperatives themselves that oppose their satisfaction. (R2) may then be used to rule out the case
that the imperative set is empty.
Consider the following axiom-schemes and rules:
7
Beyond intuition, to have such a definition may be important in a more dynamic setting where we successively remove
imperatives invariably satisfied or violated from a given set. With
we then have a formula true when an agent is done.
150
(Def)
(Conj)
(D)
(N)
(Ext)
If then
Let be any of the names in the left column of table 1:
TABLE 1
Name
System
(Def), (Conj), (Ext)
(Def), (Conj), (Ext), (D)
(Def), (Conj), (Ext), (N)
(Def), (Conj), (Ext), (D), (N)
Semantic Restrictions
None
(R1)
(R2)
(R1), (R2)
The axiomatic system is the smallest set of DL-sentences that contains all -instances of PLtheorems as well as all PL-instances of the axiom-schemes indicated to the right of the system name in
the center column and is closed under modus ponens and any indicated rule. If
, we write
and call provable and refutable in . If and satisfies the restrictions for indicated in
the right column of Table 1 and in the same row as the system name, we call
-satisfiable. If
(Mon)
If
then
Theorem 3.1. All systems DL , DL, SDL , and SDL are sound and complete.
Proof: Soundness is trivial with respect to truth definitions and restrictions. To prove (weak) complete
if for all
otherwise
is -satisfiable
and
by
151
213
so
and since
,
, so
is true. Let
be a conjunction member of
. Now suppose
is false, i.e.
and
is not -satisfiable. So
is not
-satisfiable, and there are
such that
. By (Mon), (Conj)
and (Def) we have
. So , containing the conjuncts
,
is contrary to our assumption refutable in .
will be -satisfiable (R1) if contains (D): If is not -satisfiable, then there are
such that
and therefore we have
by
(D), (Def), (Ext) and (Conj), so is refutable in again. will be non-empty (R2) if contains (N):
Suppose
. Then
and for one
,
is not -satisfiable, so
. But then
Remark 3.2. Suppose that in the truth definitions we drop the clause regarding empty :
iff
iff
is
-satisfiable
Then SDL is sound and complete with respect to semantics that employs these definitions, and SDL is
sound and complete for such semantics if restricted by (R1). (Validity of
is trivial with respect to the
new definitions. To adjust the above completeness proof, define
.)
If the imperative structure and the valuation are clear from the context, we will just write and . Since
accepts the principle of bivalence we have
and
.
Andersons [7] and Kangers [29] idea was to define deontic operators by a combination of alethic
operators and one constant. Unhappily, Anderson [7] and Anderson/Moore [9] explained their constant
S as representing a sanction or bad thing, and defined the statement
as meaning that not doing
necessitates the sanction. Lemmon and Nowell-Smith [32] objected, since in this wicked world it is not
always true that when (say) a robbery is committed someone suffers a sanction and accused Anderson
of the naturalistic fallacy. Interestingly, in answering these accusations Anderson [8] argued that if one
fails to fulfill ones obligations, that is the bad thing, no sanction being presupposed. But then we can
give Andersons constant a definite meaning: the bad thing is that
, since then an imperative has
been violated. Kanger [29] introduced a constant Q standing for what morality prescribes. Though
this seems just as dubious at first, Kanger likened morality to a welfare program for the universe of
discourse, represented by a set of propositions, and defined Q to be true iff all propositions of this
program are true.
is then defined true iff the truth of Q necessitates the truth of . Now we have a set
152
of sentences that might be understood as a welfare program: the set of sentences that correspond
to the imperatives. We therefore interpret Kangers constant as meaning that all sentences in are
true, hence by (W) every imperative is satisfied, so
.
Here we will have four constants, of which only two can be defined. The constants of Anderson
and Kanger neither alone nor combined suffice for logics properly parallel to
and , or axioms
expressing the principle of bivalence which the notions of satisfaction and violation inherited from our
basic logic. The language of analysis will be , where the alphabet is like that of
, except that
additionally we have operators and and constants , , , and . The definition of the set
of sentences differs from that of
only in the respect that now also , , , and are sentences, and
if is an -sentence, then also
and
are. Note that will represent Andersons constant, and
is our interpretation
of Kangers Q. The truth definitions are relative to a valuation and an imperative
structure
, and are as for -sentences, except for these additional clauses:
iff
for all :
(logical necessity)
iff
for some :
(logical possibility)
iff
(Df1)
(Df2)
(Biv1)
(Biv2)
(AK-D)
(AK-N)
(Nec)
If
then
Let be any of the names in the left column of table 2:
TABLE 2
Name
System
(Df1), (Df2), (Biv1), (Biv2), (Nec)
(Df1), (Df2), (Biv1), (Biv2), (Nec), (AK-D)
(Df1), (Df2), (Biv1), (Biv2), (Nec), (AK-N)
(Df1), (Df2), (Biv1), (Biv2), (Nec), (AK-D), (AK-N)
Semantic Restrictions
None
(R1)
(R2)
(R1), (R2)
153
(a)
If , then and
Semantics for dyadic deontic logics is usually based on a preference relation (at least as good as)
on the set of all Boolean valuations B (Hansson [19]) or a set of possible worlds W (Lewis [33], Aqvist
is true
iff
the
best
-worlds
are
-worlds,
[12]). According to the Hansson-type definition,
whereas the Lewis-type truth definition demands that there is a -world such that no 9
world is at least as good. Hansson motivated his semantics by the observation that while one should
always try to make the real world an ideal world, in circumstances in which somebody nevertheless
performs a forbidden act ideal worlds may be excluded, but some achievable worlds may still be better
than others. One should then make the best out of sad circumstances, i.e. realize the best worlds left. In
our semantic framework, the best state obviously is that all imperatives are satisfied. However, suppose
that in circumstances some imperatives cannot be satisfied. It then seems reasonable to demand that
all other imperatives still satisfiable in circumstances be satisfied, and to permit
only what is cosatisfiable in with the imperatives left. The truth of dyadic-deontic sentences
and
will be defined with respect
to
the
subset
of
imperatives
in
not
yet
violated
in
situation each
sentence such that is true answers the question: What is necessary to satisfy all imperatives
that are still satisfiable, given circumstances ? Note that we do not presuppose any special conditional
structure of the imperatives.10
For the semantics, let be a function that removes those sentences from a set which (individually)
imply the truth of a sentence :
Let be any imperative structure. is then the set of just those sentences in
that are individually satisfiable with . We give the following truth conditions:
9
Lewis-type truth definitions have become more standardly employed with Horty ([25], [27]) to define deontic operators with
regard to (general, dominance, maximin) orderings in stit-models.
10
154
217
is -satisfiable
Before we take a closer look at note that the truth definitions are strict parallels of those in section
3. Concerning their first clause, I believe that when all imperative correlated sentences are rejected by
, i.e. all imperatives are violated in circumstances , then there is nothing left to be done. To
iff
iff
and
or
call something obligatory or not permitted in such a situation seems as pointless as when there are no
imperatives at all: E.g. if Im to see to it that the two witnesses appear at the wedding, and one wont
come, I might just as well tell the other one its off. However, we shall also consider truth conditions
where this clause is dropped:
*
*
iff
iff
is
-satisfiable
We will then see that whereas conditions (a) and (b) produce Lewis-type dyadic logics, semantics that
makes use of conditions (a*) and (b*) will relate to Hansson-type systems.
situation. The logical evidence of there being a dilemma is the truth of both
,
. A
situation where
was demanded initially, but is now true, is a state of violation, not a dilemma.
Such demands are disregarded by
In a dilemma, we also know that invariably some demands
will be violated, but which ones is still up to us. Then saying You should do and also You should
may reflect a tenable position (this van Fraassen [50] argued for in detail, also cf. von Wright
do
[55], [56], Hamblin [17] ch. 5). So some logic should be able to state the dilemma, and we should not
beforehand smooth out the situation by removing one of the still satisfiable demands, or change them into
a demand for, say,
. Also, in a dyadic setting the existence of a dilemma for a situation is not
entirely devastating, even
if
is constructed as a regular operator (supporting agglomeration of contents)
so subsequently
is true: There might neither be a dilemma prior to the truth of , nor might
a dilemma ever arise again once one of the demands is violated, as it must be. So dilemma-permitting
logics with regular operators may still be of interest. 11
We use the following restrictions on , which parallel (R1), (R2) employed above (p. 211):
155
(DR1)
(DR2)
If
If
then
then
is -satisfiable
(no predicaments)
(no outlaws)
(DR1) expresses the assumption that all possible conflicts (dilemmas, predicaments) are resolved in
. If the circumstances can be realized at all, the imperatives that are then left must be co-satisfiable
with them:
Though in possible circumstances e.g. two imperatives
with
and
might be satisfiable with individually, they cannot both be satisfied any longer
and therefore are not both
Note that if (DR1) applies, then for all
allowed
in the set of imperatives.
either or .12
(DR2) demands that in all possibly realized circumstances there are still some unviolated imperatives left. I call this restriction No outlaws, since it might be understood as stating that we cannot get
into a situation, however despicable, that would completely rid us of all duties. (DR2) may be motivated
by the assumption that contains an infinite supply of imperatives for any occasion, though of course
one tautological imperative would suffice.
Consider the following axiom-schemes and rules:
then
if
if
then
if then
if then
(DDef)
(DConj)
(CExt)
(ExtC)
(Up)
(Down)
(DD )
(DN )
(DDR )
(DNR )
12
I owe this observation to Leon van der Torre. So the contents of all norms must be chained: The norm giver must, implicitly
or explicitly (e.g. by use of unless-clauses), indicate which demands should be given priority in case each is still satisfiable
individually, but not all collectively. This does not mean that in a finite code all but the strongest norms are redundant, since
some norms may be contrary-to-duty.
13
Lewis [33], van Fraassen [49] employ non-contextual extensionality and an axiom to the same effect.
14
(Up) and (Down) are derivable in the systems CU of Lewis [33], Spohn [46], and Aqvist
[11], [12], [13]. (Up) was first
employed by Rescher [42], and (Down) was first proposed by Fllesdal/Hilpinen [16], p. 31.
15
156
219
TABLE 3
Name
DDL
DDL
DSDL
DSDL
System
Restrictions
None
(DR1)
(DR2)
(DR1), (DR2)
DSDL3
DSDL3
None
(DR1)
We define the axiomatic system to be the smallest set of DDL-sentences that contains all DDLinstances of PL-theorems as well as all PL-instances of the axioms indicated to the right of the system
name in the center column, and is closed under modus ponens and any indicated rule. If
we write
and call provable in . For systems DDL , DDL, DSDL , DSDL, validity and satisfiability are
defined with respect to truth conditions (a) and (b) and the indicated restrictions, and similar for systems
DSDL3 , DSDL3 with respect to truth conditions (a*) and (b*) (the horizontal line indicates the change
in truth definitions).
(T1) is the down axiom employed e.g. by Rescher [42] and Aqvist
[12],18 and (T2) is an equivalent of
Lewis [33] down axiom (A8). In the presence of (Up) and (CExt), (T1) and (T2) are derivable from
(Down), while for a derivation of (Down) from (T1) the use of (DN R ), and for a derivation of (Down)
from (T2) the use of (DDR ) seems crucial. (T3) is Lewis [33] upward axiom (A6), derivable here from
(Up) and (CExt). (T4) and (T5) both derive from (Up) and will be useful in the proofs below.
are sound.
Proof: The validity of (DDef), (DConj), (CExt), and (ExtC) is trivial, and so is (DD ) since
. For the validity of (Up), first observe that
Assume the truth
of
so
and also
We have
16
DFL contains the systems of Danielsson [15], van Fraassen [49], and Lewis [33].
We add (ExtC), which von Kutschera erroneously believed to be derivable.
18
Von Wrights axiom (A2) in [53] is equivalent to the conjunction of (T1) and (Up), and so are von Kutscheras (A17) in [51],
and Spohns (P5). For a discussion of versions of (Up) and (Down) cf. [40].
17
157
and since ,
from monotony of we obtain
: Again observe first
For (Down),
consider
its
equivalent
that
Suppose and are
true. Either
Suppose
is -satisfiable, so
is satisfiable with ,
for all
but
we
assumed
the
contrary
for
all
So
for
to
must
but then
be true,
isbe empty,
so holds
trivially.
Or
-satisfiable.
If
then and again trivially
is -satisfiable. Suppose
-satisfiable, so for all
sois once
is satisfiable
with
so
more
and
is -satisfiable. So in all three possible cases in which
so
disjunct
definition holds for
. Concerning (DN R), if then due to (DR2)
of the truth
, so the first conjunct of the truth definition holds for
, and the second holds
trivially since
is always a consequence.
Theorem 5.2. The systems DDL , DDL, DSDL , and DSDL are complete.
for any
Proof: To prove (weak) completeness, we have to show that
-sentence
and any system
be
a
set
of
2
mutually
non-equivalent representatives of
, where is the number
. By writing PL of proposition letters in
(a)
Any (with CExt, ExtC from
equivalently replaced) conjunct of is in ,
(b)
for all
either
or
,
(c)
is -consistent.
It then suffices to prove that there is an imperative structure for that satisfies all
. The proof
follows Hansson [19] and Spohn [46], i.e. we first identify the deontic basis of a sentence , identify
supersets
such that the deontic basis of is implied by that of (which reflects partly ordinary,
not contrary-to-duty, conditionality of ), and then see to it that the deontic basis of
is implied by
sentences
with an empty deontic basis (in case
such that
. Observe that
and
corresponding set of sentences
iff
iff
iff
iff
is not defined
158
221
For each sentence
, the sentences such that may be ordered by extensional inclusion.
We define to be the set of all sentences maximal within that order, and prove some observations
regarding :
and for all : if then
such that
(L3) If
then there is a
and
Proof : If then there must be a such that
so
for
some
such
there
cannot
be a such that
,
and So and
(L4) For all sentences
.
Proof: Apparent from (L3) and the fact that
(L5) For any such that
for some , we have
, so we obtain
Proof : Suppose
Then
from (Down) , and so .
(L6) Let
for some Then either or
so
(L7) If
, then there is a sentence such that and
Proof : Suppose
We have
and
by (Down) also
. So by definition
and
of
We
have
by
(CMon)
, so
But then
with (Down) we obtain
(L8) Let
then
.
for some . Then and if
Proof :
, so with
(CMon), we have
so
from
Suppose
of , so byuse
(Down) and by definition
Then
of (Down) we obtain
contrary to the assumption
(L9) For all sentences such that
and :
Proof : From
and we have by use of (Up).
So
and
From
we have
by use of (T2).
So
we
obtain
(L1)
(L2)
iff
iff
159
, let
for some
and
,
be some proposition letter not in
is the set given by the following clauses, and is again an identity function on :
,
and
and
19
For all
iff
Suppose
. So
. We have from (L6) and (L8). From (L8) it
follows that logically implies for each
, and also that logically implies for each
.
Concerning the sentences from clause (ii), by the same
So no sentence from clause (i) can be in
reasoning logically implies and also the negation of the consequent. So neither are these
two
, so and both sides of the iff-clause prove false for any .
sentences in
Suppose
and
Due to (L7) there is a such that logically implies ,
and
for otherwise
from (L9) and
contrary to our assumption. It
and so so from
follows that
so
so
from (L6), and since (L9) gives us
we have
Assume
iff
Suppose
Then
We have for otherwise
and
by
sentences from clause (ii) are in
(T5), but we assumed
So both So
Assume Then for all ,
so
for otherwise
For any
due to (T4). So
so
for
each
such
, either
then due to (L3) there is a
such that
or So
and again So both sides of the iff-clause prove true for
any
.
By these clauses is constructed inconsistent: The sentences from clause (ii) jointly imply that some
is true, which is
denied by the sum of all sentences from clause (i). It is the job of the contraction function to remove sentences to make
consistent, if so required.
19
160
223
iff
so is satisfiable.
so also and, since
so
, is satisfiable.
(DR2) suppose
so We
from (DNR), so For
again
have
If
then
and since
If
then
so
both imperatives from clause (ii) are in and at least one must also be in
Proof: For each
then
if and
then
is then defined as
. The obvious fact here that for all , either
R
or
, thus reflects a property of deontic bases resulting from (DD ) that, when presented in
if
(i)
(ii)
Theorem 5.7. The systems DSDL3 , DSDL3 are sound and complete.
Proof: Regarding soundness, the validity of (DN ), (DNR ) is now trivial with respect to the changed
truth definitions, and so is the validity of (DD R ) for semantics restricted by (DR1). (Up) and (Down) are
proved as before (for Down there is now just one case to consider). To prove (weak) completeness, the
above proof may be adapted in the following way: First observe that for all
is now well defined,
since
due to (DNR ), (DN ). Now possibly
(just where in the previous construction
not define
any more. All other observations still hold. In the definition of , clause (ii) now has
so
, and
The case
is then proved as before. In the proof
of the verification lemma, for (DR1) case
is redundant and the remainder proved identically.
Note that also (DR2) holds trivially since any nonempty
has a nonempty intersection with or
6. Future Study
We motivated a semantics that models imperatives, and demonstrated that existing systems of deontic logic reconstruct nicely with respect to this semantics. In particular we could observe which truth
161
definitions and restrictions on imperative structures are necessary to obtain these reconstructions, while
standard possible worlds semantics tends to conceal that and how its restrictions on relations between
worlds are in fact restrictions on the presupposed normative system. We can only hint at how other
concepts may be accomodated:
We may define some subset of I as minimal requirements. We can then define operators expressing what is necessary to satisfy the minimal requirements, but also operators of supererogation as
meaning that while something is not minimally required, it still satisfies an imperative, or operators of offense as meaning that some not minimally required imperative is violated (cf. [36] for
a possible worlds setting of supererogation).
We may have hierarchies of imperative structures, and define operators in relation to satisfaction
sets that fulfil e.g. the following criterion: Any imperative of the highest authority is satisfied, any
imperative of the second highest authority is satisfied that can be if the imperatives of the highest
authority are, etc. (cf. [18] for such command hierarchies).
We might use a second function to associate (numerical) values with each imperative, and define
as true iff is necessary for achieving that the sum of all values of imperatives in the satisfaction set is the sum of all values of imperatives in , the highest achievable sum, or surpasses a
fixed value. It may then be obligatory to violate one imperative when it serves to satisfy others of
higher value.
Regarding conditional imperatives, we may follow van Fraassens proposal in [50] and employ
a function that associates a second -sentence with each imperative, describing the opportunity for its satisfaction or violation. For all valuations we then have the set
of actualized
imperatives
such that
. We can define a monadic operator
as true at iff
Some puzzling detail remains: In section 2, we have separated imperatives and their propositional
content, linking them by a function that associates a descriptive sentence with each imperative. But in
the reconstructions this separation was not used the restrictions and truth definitions could equally well
be given with respect to a set of sentences instead. This does not mean that imperatives are reducible to
indicatives. It merely shows that the reconstructed logics do not require the modelling of other qualities of
imperatives. The real puzzlement comes from the other side of the equation: Quite a number of deontic
logics may be characterized as being not about specifically normative notions at all, but as reasoning
about sets of sentences and their underlying classical (meta-)logic. To modify von Wrights famous
statement: It may have turned out that deontic logics have a wider reach than normative concepts.
References
[1] al-Hibri, A.: Deontic Logic, Washington: University Press, 1978.
[2] Alchourron, C. E. and Bulygin, E.: The Expressive Conception of Norms, in [24], 95123.
162
225
[3] Alchourron, C. E. and Bulygin, E.: Pragmatic Foundations for a Logic of Norms, Rechtstheorie, 15, 1984,
453464.
[4] Alchourron, C. E. and Bulygin, E.: On the Logic of Normative Systems, in Stachowiak, H. (ed.): Handbuch
pragmatischen Denkens, vol. IV, Hamburg: Meiner, 1993, 273294.
[5] Alchourron, C. E. and Martino, A. A.: Logic Without Truth, Ratio Juris, 3, 1990, 4667.
[6] Alchourron, C. E., Gardenfors, P., and Makinson, D.: On the Logic of Theory Change: Partial Meet Contraction and Revision Functions, Journal of Symbolic Logic, 50, 1985, 510530.
[7] Anderson, A. R.: The Formal Analysis of Normative Concepts, in Rescher, N. (ed.): The Logic of Decision
and Action, Pittsburgh: University Press, 1967, 147213.
[8] Anderson, A. R.: Some Nasty Problems in the Formal Logic of Ethics, No u s, 1, 1967, 345359.
[9] Anderson, A. R. and Moore, O. K.: The Formal Analysis of Normative Concepts, American Sociological
Review, 22, 1957, 917.
[10] Aqvist,
L.: Interpretations of Deontic Logic, Mind, 73, 1964, 246253.
[11] Aqvist,
L.: Deontic Logic, in Gabbay, D. and Guenthner, F. (eds.): Handbook of Philosophical Logic, vol.
II, Dordrecht: Reidel, 1984, 605714.
[12] Aqvist,
L.: Some Results on Dyadic Deontic Logic and the Logic of Preference, Synthese, 66, 1986,
95110.
[13] Aqvist,
L.: Introduction to Deontic Logic and the Theory of Normative Systems, Naples: Bibliopolis, 1987.
[14] Chisholm, R. M.: Contrary-to-Duty Imperatives and Deontic Logic, Analysis, 24, 1963, 3336.
[15] Danielsson, S.: Preference and Obligation, Uppsala: Filosofiska foreningen, 1968.
[16] Fllesdal, D. and Hilpinen, R.: Deontic Logic: An Introduction, in [23], 135.
[17] Hamblin, C. L.: Imperatives, Oxford: Blackwell, 1987.
[18] Hanson, W. H.: A Logic of Commands, Logique et Analyse, 9, 1966, 329343.
[19] Hansson, B.: An Analysis of Some Deontic Logics, No u s, 3, 1969, 373398. Reprinted in [23], 121147.
[20] Hansson, S. O.: New Operators for Theory Change, Theoria, 55, 1989, 114131.
[21] Hare, R. M.: Imperative Sentences, Mind, 58, 1949, 2139.
[22] Hare, R. M.: The Language of Morals, 2nd ed., Oxford: University Press, 1964.
[23] Hilpinen, R. (ed.): Deontic Logic: Introductary and Systematic Readings, Dordrecht: Reidel, 1971.
[24] Hilpinen, R. (ed.): New Studies in Deontic Logic, Dordrecht: Reidel, 1981.
[25] Horty, J. F.: Agency and Obligation, Synthese, 108, 1996, 269307.
[26] Horty, J. F.: Nonmonotonic Foundations for Deontic Logic, in [38], 1744.
[27] Horty, J. F.: Agency and Deontic Logic, New York: Oxford University Press, 2001.
[28] Jrgensen, J.: Imperatives and Logic, Erkenntnis, 7, 1937/38, 288296.
[29] Kanger, S.: New Foundations for Ethical Theory, in [23], 3658.
[30] Kelsen, H.: Reine Rechtslehre, Leipzig/Wien: Deuticke, 1934.
[31] Kelsen, H.: Allgemeine Theorie der Normen, Wien: Manz, 1979.
163
[32] Lemmon, E. J. and Nowell-Smith, P. H.: Escapism: The Logical Basis of Ethics, Mind, 69, 1960, 289300.
[33] Lewis, D.: Semantic Analyses for Dyadic Deontic Logic, in Stenlund, S. (ed.): Logical Theory and Semantic Analysis, Dordrecht: Reidel, 1974, 114.
[34] Makinson, D.: On a Fundamental Problem of Deontic Logic, in McNamara, P. and Prakken, H. (eds.):
Norms, Logics and Information Systems, Amsterdam: IOS, 1999, 2953.
[35] Mally, E.: Grundgesetze des Sollens, Graz: Leuschner und Lubensky, 1926.
[36] McNamara, P.: Doing Well Enough: Toward a Logic for Common Sense Morality, Studia Logica, 57,
1996, 167192.
[37] Moutafakis, N. J., Imperatives and their Logics, New Delhi: Sterling, 1975.
[38] Nute, D.: Defeasible Deontic Logic, Dordrecht: Kluwer, 1997.
[39] Opaek, K.: Theorie der Direktiven und Normen, Wien: Springer, 1986.
[40] Prakken, H. and Sergot, M.: Dyadic Deontic Logic and Contrary-to-Duty Obligations, in [38], 223262.
[41] Prior, A. N.: The Paradoxes of Derived Obligation, Mind, 63, 1954, 6465.
[42] Rescher, N.: An Axiom System for Deontic Logic, Philosophical Studies, 9, 1958, 2430.
[43] Rescher, N.: The Logic of Commands, London: Routledge & Kegan, 1966.
[44] Ross, A.: Imperatives and Logic, Theoria, 7, 5371.
[45] Sosa, E.: The Logic of Imperatives, Theoria, 32, 1966, 224235.
[46] Spohn, W.: An Analysis of Hanssons Dyadic Deontic Logic, Journal of Philosophical Logic, 3, 1975,
237252.
[47] Stenius, E.: The Principles of a Logic of Normative Systems, Acta Philosophica Fennica, 16, 1963, 247
260.
[48] van der Torre, L.: Reasoning about Obligations, Amsterdam: Thesis Publishers, 1997.
[49] van Fraassen, B.: The Logic of Conditional Obligation, Journal of Philosophical Logic, 1, 1972, 417438.
[50] van Fraassen, B.: Values and the Hearts Command, Journal of Philosophy, 70, 1973, 519.
[51] von Kutschera, F.: Normative Praferenzen und bedingte Gebote, in Lenk, H. (ed.): Normenlogik, Pullach:
UTB, 1974, 137165.
[52] von Wright, G. H.: Deontic Logic, Mind, 60, 1951, 115.
[53] von Wright, G. H.: A Note on Deontic Logic and Derived Obligation, Mind, 65, 1956, 507509.
[54] von Wright, G.H.: Logical Studies, London: Routledge and Kegan, 1957.
[55] von Wright, G.H.: A New System of Deontic Logic, Danish Yearbook of Philosophy, 1, 1964, 173182.
Reprinted in [23], 105120.
[56] von Wright, G.H.: A Correction to a New System of Deontic Logic, Danish Yearbook of Philosophy, 2,
1965, 103107. Reprinted in [23], 105120.
[57] von Wright, G. H.: Norms, Truth, and Logic, in von Wright, G. H.: Practical Reason, Oxford: Blackwell,
1983, 130209.
[58] von Wright, G. H.: Is There a Logic of Norms?, Ratio Juris, 4, 1991, 265283.
164
Abstract
Deviating from standard possible-worlds semantics, authors belonging to what might be called
the imperative tradition of deontic logic have proposed a semantics that directly represents norms
(or imperatives). The paper examines possible definitions of (monadic) deontic operators in such a
semantics and some properties of the resulting logical systems.
2004 Elsevier B.V. All rights reserved.
Keywords: Logic of imperatives; Deontic logic
A more comprehensive list should also include Smiley [19], and Niiniluoto [17]. Smiley, in a concept later endorsed by Ruth Barcan Marcus [15] and motivated by an Andersonian [2] definition of a constant representing
the satisfaction of all normative demands, uses a normative code consisting of propositions to define OA as
meaning that there is a finite number of propositions p1 , . . . , pn in this code such that (p1 pn A) is a
logical truth. Niiniluoto represents commands by a tuple containing a proposition p, where the truth of OA then
depends on whether A is logically implied by one such p.
1570-8683/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jal.2004.01.003
|= O 1 A
|= O 2 A
|= O 3 A
|= O 4 A
|= O 5 A
iff
iff
iff
iff
iff
A I,
B I :|=BL B A,
B I :|=BL B A,
B1 , . . . , Bn I :|=BL (B1 Bn ) A,
I |=BL A.
O 1 A is true iff A is one of the sentences in I , i.e., A (literally) corresponds to what an imperative demands. O 2 A is true iff the truth of A is BL-logically a necessary and sufficient
condition for the fulfillment of what one imperative demands.2 Similar to Niiniluotos [17]
operator, O 3 A is true iff A is BL-implied by what some imperative demands. Similar to
Smileys [19] operator, O 4 A is true iff a finite subset of imperative-associated sentences
BL-implies A. Finally and similar to AlchourrnBulygins [1] operator, O 5 A is true iff A
is BL-implied by the set of all such sentences. Note that while all of O 1 O 4 presuppose
the existence of some sentence in I , O 5 A can hold for (BL-valid) sentences A even when
I contains no object.
Discussing the general framework, I showed in [8] how a number of well-known systems of monadic and dyadic deontic logic can be reconstructed with respect to imperative
semantics. The present paper examines some details and problems connected with monadic
deontic logics facilitated by the above semantics. Section 2 gives logical systems for each
of the above operators. Section 3 then examines systems that include several of these operators. In Section 4 we take a look at possible semantic restrictions on the content of
imperatives, and see what changes must be applied to logical systems to correspond to
them. Section 5 examines more closely the definitions van Fraassen [21] proposed for
O-operators, and provides a monotonic system to resemble his reasoning, and Section 6
addresses a sceptical definition of ought proposed by Horty [12] to deal with normative
conflicts. Finally, in Section 7 I give systems and semantics for extended languages that
permit iterated and nested deontic operators of the types presented above.
2 In the terminology of Brown [4] O 2 is a type 2 operator, whereas operators O 35 would be type 1, their
scope being a necessary condition for norm satisfaction only.
165
166
41
2. Logics DL15
for deontic operators O15
1
Let our basic logic be propositional logic PL. The alphabet of the language L[PL]
has a set of proposition letters Prop = p1 , p2 , . . . , truth-functional operators , , ,
, and brackets (, ). The set of sentences is defined as usual. ,
in front
of sets of sentences means their conjunction and disjunction, and, e.g., ni=1 Ai further
abbreviates {Ai , . . . , An }. In the semantics, valuation functions v : Prop {1, 0} define
the truth of sentences A L[PL] as usual, A meaning the set of valuations v that make
A true. means an arbitrary tautology, and an arbitrary contradiction. We suppose a
sound and complete axiomatic system for PL.
In this section, we examine deontic logics DLi1 for operators O i , 1 i 5, defined
above. The upper index in the system name (and the names of most axiom schemes) indicates the type of operator used, and the subscript indicates that the scope of O i is L[PL],
not a deontic language (in Section 7 we give up this restriction). The alphabet of languages
L[DLi1 ], 1 i 5, is like L[PL] except for one additional operator O i . L[DLi1 ] is then
the smallest set such that
(a) if A L[PL] then A L[DLi1 ],
(b) if A L[PL] then O i A L[DLi1 ], and
(c) if A, B L[DLi1 ], so are A, (A B), (A B), (A B), and (A B).
Note that we permit mixed expressions like p1 O 1 p2 . The axiomatic systems DLi1
are then defined by the following clauses (we write DLi A for A DLi1 ):
1
DLi
O i ,
5,
i = 5.
A sentence A L[DLi1 ] is DLi1 -derivable from a set L[DLi1 ] (we write DLi A) iff
1
{A} is DLi1 -inconsistent. If A is derivable from the empty set, A is called refutable.
For the semantics, let v be as before, and I be a set of PL-sentences. The truthdefinitions of DLi1 -sentences are relative to v and I , where the truth of a proposition letter
is defined as before, and the truth of O i A is defined with respect to I as given in Section 1,
truth-definitions for Boolean operators being as usual. VerDLi (I, v) denotes the subset of
1
DLi1 -sentences defined true for I , v, and we write I , v |= A for A VerDLi (I, v). A is
1
167
DLi1 -valid (we write |=DLi A) iff A L[DLi1 ] is true for arbitrary I , v, and L[DLi1 ]
1
DLi1 -entails A (we write |=DLi A) iff I , v |= A for all I , v such that I , v |= B for all
1
Proof. Soundness is trivial with respect to the truth definitions employed. For completeness, we prove equivalently that if is DLi1 -consistent then is DLi1 -satisfiable:
Let A1 , A2 , . . . be a fixed enumeration of L[DLi1 ]. Let = n n , where 0 = ,
and
n+1 =
n {An+1 },
if this is consistent,
n {An+1 }, otherwise.
Clearly each n must be DLi1 -consistent with either An+1 or An+1 , hence (i) is DLi1 consistent, (ii) for all A L[DLi1 ] either A or A . Now we define:
v=
1, if p ,
0, otherwise,
I = {A | O i A }.
i = 3:
i = 4:
is DL31 -inconsistent.
Assume I, v |= O 4 B, so there is a non-empty finite set {C1 , . . . , Cn } I such that
|=PL (C1 Cn ) B. So {O 4 C1 , . . . , O 4 Cn } . But from (Ext41 ), (C41 ),
(M41 ), and PL-completeness we obtain DL4 (O 4 C1 O 4 Cn ) O 4 B, so
1
i = 5:
is DL41 -inconsistent.
Assume I, v |= O 5 B, so I |=PL B. If I = , then by strong completeness of PL
there is a non-empty finite set {C1 , . . . , Cn } I such that |=PL (C1 Cn )
B, so the r.a.a. is done as in case i = 4. If I = then DL5 O 5 B by (Ext51 ), (N41 ),
so is DL51 -inconsistent.
168
43
Op
Let the description of systems DL1 follow that of DLi1 , where the clause a) is now
Op
relative to the language L[DL1 ], and clauses b), c) hold for all i Op satisfying the
additional requirements. We have the following additional axiom schemes:
(Oi Oj ) DLOp O i A O j A, {i, j } Op, i < j ,
1
(O5 O3 ) if
PL
(O5 O4 ) if
PL
1
1
Op
Proof. Soundness is again trivial with respect to O i -truth definitions. For weak completeness, we have to prove that if DLOp A then DLOp A. We assume DLF A so A is not
Op
refutable in DL1 . One disjunct in a disjunctive normal form of A is then not refutable
Op
in DL1 . From the non-deontic conjuncts of we obtain a valuation v that satisfies them.
As for the deontic conjuncts, for all i Op let Oi = {B L[PL] | O i B is a conjunct of
}, and Qi = {B L[PL] | O i B is a conjunct of }. Let 1 , 2 be proposition letters that
do not occur in . We define
1 Op (a)
I = O1
2
2 Op (b)
(B (1 1 )) | B O
(B 1 ) | B O3
3 Op (c)
2 , (2 (B 1 )) | B O4
4 Op (d)
5
2 , (2 (B 1 )) | B O and PL B
5 Op (e).
Obviously, for all B Oi , I, v
O i B:
Proof. The semantics of DL1 for Op = {4, 5} is not compact, i.e., there are finitely
Op
satisfiable sets L[DL1 ] such that is not satisfiable: Suppose {i, j } Op. We give
non-satisfiable sets and an If for each finite f such that If , v satisfies f , v being
arbitrary.
i = 1, j = 2: Let = {O 1 A ||=PL A p1 } {O 2 p1 }. For any finite subset f of
there is a proposition letter that does not occur in any A f . Each set f is
satisfied by If = {(( ) p1 )}.
i = 1, 2, j = 3: Let = {O i A ||=PL A p1 } {O 3 p1 }. Let f be any finite subset,
and be as before. Each set f is satisfied by If = {( p1 )}.
i = 1, 2, 3, j = 4, 5: Let = {O i A PL A} {O j p1 }. Let f be any finite subset, and
be as before. Each set f is satisfied by If = {( p1 ), }.
{4,5}
is (strongly) complete.
Proof. We adapt the completeness proof of DL41 in Theorem 2.1: is now defined with
{4,5}
{4,5}
respect to an enumeration of DL1 -sentences and DL1 -consistency. I , v is constructed
4
4
as for DL1 , in particular I = {A | O A }. The proof that I , v |= O 4 A iff O 4 A is as
before. To prove that I, v |= O 5 A iff O 5 A , assume first O 5 A . Either O 4 A ,
169
170
45
I = (Non-Triviality),
A I : BL A (Excluded Impossibility),
I BL (Collective Satisfiability),
A I : BL A (Excluded Necessity).
Consider the logics DLi1 defined in Section 2 and the following additional axiom
schemes:
(XIi1 ) if PL A then DLi O i A,
1
Warranted by the following theorems, the table below lists, for languages L[DLi1 ],
1 i 5, axiom systems sound and complete with respect to DLi1 -semantics conforming
to any of the above restrictions individually (square brackets indicate weak completeness
only):
L[DLi1 ] i = 1
[R-0]
[R-1]
[R-2]
[R-3]
i=2
i =3 i =4 i =5
DLi1 + (Di1 )
DLi1 + (XNi1 )
[DLi1 ]
DL51
Theorem 4.1. The following systems are sound and (strongly) complete:
1.
2.
3.
4.
5.
6.
7.
Proof. Soundness is again obvious. The completeness proofs are forthcoming adaptations
of that for the DLi1 -systems. When there is a specific axiom, it obviously ensures that the
constructed I conforms to the given restriction. Consider cases 2, 4, 7 where there are no
specific axioms:
Case 2: O 5 by (N51 ), so by the construction I = .
Case 4: We construct as before, but now define I = {A | O i A and PL A}. Suppose O i A and A
/ I , so |=PL A. Then |=PL A p1 , |=PL A p1 ,
so by (Exti1 ) we have O i p, O i p1 , so {p1 , p1 } I, so I |=PL A. The
remainder is as before.
Case 7: Let be as before, but now I = {A | O 5 A and PL A}. Suppose O 5 A
and A
/ I , so |=PL A. But then trivially I |=PL A, and the remainder of the proof
is as before.
Theorem 4.2. DLi1 -semantics, i = 1, 2, is not compact if restricted by [R-0]. DLi1 semantics, i = 3, 4, is not compact if restricted by [R-3].
Proof. To show that DLi1 -semantics, i = 1, 2, is not compact if restricted by [R-0], let
= {O i A | A L[DLi1 ]}. To satisfy all of , I must be empty, which is excluded by
[R-0], so is not satisfiable by [R-0]-restricted semantics. But any finite subset f
is satisfied by If , v, where If = {}, and is a proposition letter that does not occur in
any A f , v being arbitrary.
For the proof that DLi1 -semantics, i = 3, 4, is not compact if restricted by [R-3], let
= {O i A PL A} {O i }. To satisfy all of , I must contain tautologies only, which
is excluded by [R-3], yet it also cannot be empty due to the expected truth of O i , so
is not satisfiable by [R-3]-restricted semantics. But any f is again satisfied by If , v, as
described above.
Theorem 4.3. The following systems are sound and weakly complete:
1. DLi1 for DLi1 -semantics restricted by [R-0], i = 1, 2.
2. DLi1 for DLi1 -semantics restricted by [R-3], i = 3, 4.
171
172
47
Proof. Soundness is again obvious. For weak completeness, suppose {A} is DLi1 consistent. To demonstrate that there are I, v such that I, v A, we do a construction as in
the proof of Theorem 3.1, i.e., form a disjunctive normal form of A, from the non-deontic
conjuncts of a non-refutable disjunct obtain a valuation v that satisfies these, and from the
deontic conjuncts construct sets Oi and Qi as before. Again, let be a proposition letter
not occurring in A.
Case 1: Let
I=
Oi , if Oi = ,
{}, otherwise.
, otherwise.
Obviously I contains no tautological element (R-3). Suppose B Oi . If there is a nontautological C Oi , for all B Oi , tautological or non-tautological, I, v O i B. If there
are tautological B Oi (only), there cannot be a tautological C Qi due to (Exti1 ) and
PL-consistency of {A}, so I and again I, v O i B. For each B Qi , I, v O i B:
B Q3 : If there is a tautological C Q3 , O3 = due to (Ext31 ), (M31 ) and PL-consistency
of {A}, and I = by the above construction, so I, v O 3 B. Otherwise, no
C O3 PL-implies B again due to (Ext31 ), (M31 ) and PL-consistency of {A}, and
if does then B is tautological, but this was excluded.
B Q4 : As before, if C Q4 is tautological then O4 = and I = , so I, v O 4 B.
Otherwise, B is neither implied by any {C1 , . . . , Cn } O4 due to (Ext41 ), (M41 ),
(C41 ) and PL-consistency of {A}, nor by I = {} since a tautological B was excluded.
If the semantics employs more than one of the above restrictions, then generally there is
a sound and (strongly) complete axiomatic system iff each of the semantics with just one
of these restrictions has such a system, and the axiomatic system is obtained by including
all axioms and rules of the systems for the singularly restricted semantics. However, this
rule fails in one case:
Theorem 4.4. DL51 -semantics is not compact if restricted by [R-0], [R-3].
Proof. To show incompactness, let = {O 5 A PL A}. To satisfy all of , I cannot
contain anything but tautologies, which is excluded by [R-3], yet it also cannot be empty
due to [R-0], so is not satisfiable by [R-0][R-3]-restricted semantics. But any finite
Oi {A L[PL] |
{},
PL
173
174
49
is obligatory. But since there is no single imperative that demands p1 , [Df-F1] does not
allow the derivation of Op1 . Van Fraassen therefore introduces the notion of a score of an
outcome v, i.e., the set of imperatives that v fulfills, and defines:
[Df-F3] OA is true iff there is a possible state of affairs v in A whose score is not
included in the score of any v in A .
Again, this definition may be reformulated in the terms used here:
[Df-F3 ] I |= O F A iff v A : v A : {B I | v |= B}
{B I | v |= B}.
Let the set of maximally PL-consistent subsets I of I be the set of subsets I I such
that (i) I , and (ii) there is no I such that I I I and I . It can then be
proven that [Df-F3 ] is equivalent to [Df-F3 ] (cf. Horty [12, p. 30, Theorem 2]):
[Df-F3 ] I |= O F A iff I I : I PL A.
Let the language L[DLF1 ] be like any of the L[DLi1 ], except that O F replaces O i in
all DLi1 -sentences. Let the axiomatic system DLF1 be like DL31 + (XI31 ) + (N31 ), except
that O F replaces O 3 . Let DLF1 -semantics be defined like DL31 -semantics, except that truth
definition [Df-F3 ] replaces that for O 3 . We then have the following results:
Theorem 5.1. DLF1 is sound and weakly complete.
Proof. Concerning soundness, the validity of (Ext1F1 ), (MF1 ), (XIF1 ) and (NF1 ) is immediate. Concerning (weak) completeness, we have to prove that if DLF A then DLF A. We
1
assume DLF A so A is not refutable in DLF1 . As before (cf. Theorem 3.1), let be a
1
non-refutable disjunct in a disjunctive normal form of A. From its non-deontic conjuncts
we again obtain a valuation v that satisfies these. Concerning the deontic conjuncts of ,
let OF be the set of PL-sentences B such that O F B is a conjunct of , and QF be the set
of PL-sentences B such that O F B is a conjunct of . For any set L[PL] let
minPL = {A | B : if PL B A then PL B A}.
Let n = card(minPL OF ). Let {1 , . . . , n } Prop be an arbitrary set of n proposition
letters not occurring in , and be a function that maps minPL OF onto the set {1 , 1
2 , . . . , 1 n1 n } of n mutually exclusive PL-sentences. We then define:
I = B (B) | B minPL OF .
Note that I = {{i} | i I }, since (B) is PL-inconsistent with any (C), B, C
minPL OF and B = C. No B OF is a contradiction due to (XIF1 ), and for each consistent
B OF the definition of minPL OF ensures that there is an I I such that I PL B,
so O F B is true. Suppose B QF but I, v O F B, so there is an I I : I PL B.
Then there is a C minPL OF such that I = {C (C)} and PL C (C) B. Since
I |= O F A
iff
I I : B1 , . . . , Bn I |=BL B1 Bn A.
It can then be shown that DL31 + (XI13 ) is sound and weakly complete with respect to the changed definition, and
the counterexample in Theorem 5.2 again disproves compactness.
175
176
51
system in terms of the O-operator used. But, regarding his additional question, it seems to
me that the ought language game extends to operators like O 1 or O 2 that more directly
talk about imperatives. So perhaps a more powerful language can provide the missing
inferences. It is immediate that the counterexample used in Theorem 3.2 to disprove com{2,4}
pactness of DL1
also disproves compactness of any semantics that employs operators
1
2
of type O or O in addition to van Fraassens operator O F . So for an improved characterization of van Fraassens third semantics it must suffice to give a weakly complete
system:
{2,F }
{2,3}
Let L[DL1 ] be the language that is like L[DL1 ] except that O F replaces O 3 in
{2,3}
all DL1 -sentences. For the semantics, the truth of O 2 , O F is defined with respect to a
set I as above, and for proposition letters and Boolean connectives it is defined as usual.
{2,F }
{2,F }
Let the axiomatic system DL1
contain all instances of L[DL1 ] in PL-theorems, be
2
F
closed under modus ponens and (Ext1 ), (Ext1 ), contain all L[PL]-instances in the axiom
schemes (MF1 ), (XIF1 ), (NF1 ) of DLF1 , and in the following schemes:
(F-1) If
PL
(F-2) DL{2,F } (
n
n
2
i=1 Ai then DL{2,F }
i=1 O Ai
1
n
2
F
F
i=1 O Ai O B O (B
n
i=1 Ai ,
n
F
i=1 Ai )) O (B
OF
n
i=1 Ai ).
Proof. For soundness of the new axiom schemes, validity of (F-1) is immediate from the
fact that if there are B1 , . . . , Bn I equivalent to A1 , . . . , An respectively, and A1 , . . . , An
are PL-consistent, then some I I contains {B1 , . . . , Bn }, so I PL A1 An .
For the validity of (F-2), suppose O 2 A1 , . . . , O 2 An . If O F (B ni=1 Ai ), then no I
I may derive B and all Ai , so if O F B is true and there is an I that derives B, it must
be inconsistent with ni=1 Ai and consequently derive B ni=1 Ai .
{2,F }
{2,F }
For (weak) completeness, we have to prove that if A is DL1 -valid, then A DL1 .
{2,F }
{2,F }
We assume A
/ DL1 , so A is not refutable in DL1 . As now usual, let be a nonrefutable disjunct in a disjunctive normal form of A. From its non-deontic conjuncts we
4 A difficulty in axiomatizing van Fraassens semantics was pointed out by Horty [11, p. 50]: If background
imperatives are coded into ought-statements, as they are here using operator O 2 , then the set {O 2 A | A L[BL]}
must derive O F B for any BL-consistent B, though for some basic logics like first order logic these are not recur{2,F }
{2,F }
-axioms is not decidable and DL1
not recursively
sively enumerable. But then due to (F-1) the set of DL1
enumerable, so for such basic logics no axiomatization in the usual sense is provided here.
again obtain a valuation v that satisfies them. As for the deontic conjuncts of , first let
O2 = B L[PL] | O 2 B is a conjunct of ,
Q2 = B L[PL] | O 2 B is a conjunct of .
Let L[PL] be the set of PL-sentences that contain only proposition letters occurring in
n
some deontic conjunct of . Let r(L[PL] ) be a set of 22 mutually non-equivalent representatives of L[PL] , n being the number of proposition letters occurring in the deontic
conjuncts of . By writing PL-sentences we now mean their unique representatives in
L[PL] . We construct a finite set with the following properties:
(a) If O F B, O F C are conjuncts of , then O F B, O F C are in .
(b) For all B r(L[PL] ), either O F B or O F B .
{2,F }
(c) {} is DL1 -consistent.
In (a), we may have to replace B, C by r(L[P L] )-equivalents using (ExtF1 ).
Let:
OF = B L[PL] | O F B ,
QF = B L[PL] | O F B .
Lemma 5.1. For all B O2 : (a) if PL B then there is a C minPL OF such that PL
C B, and (b) for all C minPL OF , if {B, C} is PL-consistent, then PL C B.
Proof. (a) is immediate from (F-1) and the definition of minPL OF (cf. the proof of Theorem 5.1). Concerning (b), either O F (B C) or O F (B C) (clause (b) in the
construction of ). If O F (B C) then PL C (B C) by definition of minPL OF ,
{2,F }
so PL C B. If O F (B C) then O 2 B, O F C, O F (B C) DL1 -derives
O F (B C) by use of (F-2), so O F (B C) by clause (c) in the construction
of . But then PL C (B C) by definition of minPL OF , so {B, C} is not PLconsistent.
Let : minPL OF L[PL] be the function defined in the proof of Theorem 5.1. Let
I = O2 B (B) | B minPL OF .
Lemma 5.2. For all I I , B L[PL] : if I = and I PL B then there is a
C minPL OF such that PL C B.
Proof. Immediate from Lemma 5.1, the construction of I , and the fact that (C) contains
no proposition letters occurring in B or C.
It remains to prove that I satisfies all deontic formulas in and :
177
O2 :
Q2 :
OF :
QF :
178
53
179
n
2
i=1 O Ai
n
2
i=1 O Ai
O F
O F
k
j =1
k
j =2
j )) O S (
k
j =1
) OS
j
n
A
)
OS
i=1 i
k
j =2
j );
j ;
n
i=1 Ai .
In FS-2 and FS-3, j is a non-empty subset of {A1 , . . . , An }. Note that (FS-3), (FS-4) are
special cases of (FS-2). (FS-3) derives (O 2 A O 2 B O F (A B)) O S (A B),
which is a syntactic version of the disjunctive solution to Marcuss dilemma: the disjunction (A B) of two possibly conflicting imperative demands A, B is obligatory, unless
there is a consistent set of imperatives that demands even A B. Hence one twin or
5 It might seem awkward that if I contains contradictory imperative demands, e.g., if I = {p , p , p } then
1
1 2
with [Df-S] the set of truths is the same as for I = {p2 }, i.e., the same as for an I without these demands. If
the fulfillment of one of the demands is to remain obligatory, an alternative would be to have a disjunction of
oughts instead of the ought of a disjunction, and define:
In the above example, (O 5 p1 O 5 p1 ) is now true under the new definition, O 5 (p1 p1 ) remains valid, and
O 5 remains false. Ignoring operator indices, we have VerDLS (I, v) VerDS (I, v) VerDL5 (I, v).
1
180
55
the other must be saved. From (FS-1) and (XIF1 ) we derive O F A O S A, so what is
obligatory according to van Fraassens ought is at least permitted in the sceptical sense. If
guilt is a notion that attaches to a violation of a sceptical ought only, then one may save a
twin without guilt, but walking away remains forbidden. From the results of Section 5 it is
{2,F,S}
immediate that DL1
cannot be strongly complete. However we obtain the following
result:
{2,F,S}
i=1 Ai , so there is some I I that contains each Ai . For the validity of (FS-2),
assume again O 2 Ai true for all i, 1 i n, and assume O F ( 1 kj =2 j ).
Those I I that contain a set j also derive 1 kj =2 j . Consider an I
that does not contain any j : Since I is not consistent with all Ai j , it must derive
j for each j , so it derives kj =2 j . If I is consistent with 1 , it derives 1
k
j =2
j , but then O F (
k
j =2
Q = B r L[PL]
| {} DL{2,F,S} O S B .
1
I is constructed as in the proof of Theorem 5.3 and additionally contains the following
elements: Let QS = {B1 , . . . , Bm }. For each Bj we add CI CII CIII to I , where:
CI = OS ,
CII = { | O2 and {CI } PL Bj },
CIII = 1 n n+1 n+j .
In CIII , 1 , . . . , n are the same proposition letters used to define : minPL OF L[PL]
(cf. the proofs of Theorems 5.1 and 5.3), and n+1 , . . . , n+m are some other letters not
occurring in . We have to prove that the modified I still satisfies all conjuncts of . For
the non-deontic conjuncts and conjuncts of the form O 2 B, O 2 B, O F B, the proof is as
before. Consider conjuncts of the form O F B, O S B, O S B, and the sets QF , OS , QS :
QF :
OS :
QS :
181
182
57
(p) = 0,
(ii)
(A) = (A),
(iii)
(A B) =
(iv)
(O i A) = (A) + 1.
(A),
(B),
if (A) (B),
otherwise,
(Extin ) If DLi
n1
A B then DLi O i A O i B, 2
5, n
5.
183
Proof. The proof proceeds just like for DLi1 . Note that strong completeness is proved
inductively for DLin , using strong completeness of DLin1 .
Let the axiomatic systems DLi be like DLin , except we now allow all DLi -instances in
PL-tautologies and relevant axiom schemes (Mi ), (Ci ), or (Ni ), and (Exti ) now reads:
(Exti ) If DLi A B then DLi O i A O i B, 2
5.
Proof. For the left-to-right direction we do an induction on the construction of DLi : Suppose DLi A. Induction basis: Assume A DLi by instantiation of sentences B1 , . . . , Bk
into a PL-theorem. Then {B1 , . . . , Bk } L[DLin ], so there is the same instance for DLin .
Assume A DLi by instantiation of sentences B1 , B2 into any of the DLi -axiom-schemes
(Mi ), (Ci ), (Ni ) or (Exti ). Then B1 , B2 L[DLin1 ], so there is the same instance for
DLin . The induction step for modus ponens is trivial. For the right-to-left direction do a
double induction on the construction of DLin : Suppose DLi A and suppose n = 1. Inducn
5, n
i
n DLn .
Proof. DLi = n DLin is immediate from Theorem 7.2, and DLin DLin+1 is proved by
induction on the construction of DLin just as in the right-to-left proof of Theorem 7.2.
We are now able to give proper truth definitions for the deontic modalities of L[DLi ]:
I , v |= p
I , v |= O 1 A
I , v |= O 2 A
I , v |= O 3 A
I , v |= O 4 A
I , v |= O 5 A
iff
iff
iff
iff
iff
iff
v(p) = 1;
A I;
n N: B I L[DL2n ]: |=DL2 B A;
n
n N: B I L[DL3n ]: |=DL3 B A;
n
n N: B1 , . . . , Bk I L[DL4n ]: |=DL4 (B1 Bk ) A;
n
n N: I L[DL5n ] |=DL5 A.
n
Theorem 7.3. DLi is sound and complete with respect to DLi -semantics.
184
59
Proof. For soundness, the validity of DLi -instances in PL-theorems is trivial, and so is
the validity of (Mi ). The validity of (Ci ) is easily established using Corollary 7.1 as well
as the soundness and completeness of each DLin . For the DL4 -validity of (Ext4 ), suppose
DL4 A B, so by Theorem 7.2 DL4 A B for n = (A B). So |=DL4 A B
n
So also I, v |= O 4 B. Suppose n
n+1
iff
iff
iff
iff
iff
A I,
B I : |=DL2 B A,
B I : |=DL3 B A,
B1 , . . . , Bk I : |=DL4 (B1 Bk ) A,
I |=DL5 A.
Proof. Immediate from the truth definitions for O i -operators and Corollary 7.2. So the
truth definitions of O i -operators are equivalent to taking DLi as basic logic.
Consider the following DLi -sentences:
(DT) O i (O i A A);
(D4) O i A O i O i A;
(D4c ) O i O i A O i A;
(DB) O i (A O i O i A);
(DE) O i A O i O i A.
None is a theorem of any DLi . To refute (D4c ) in DL5 , consider the set I = {O 5 p1 }. We
have I |=DL5 O 5 p1 , but not I |=DL5 p1 , so I, v |= O 5 O 5 p1 is true, and I, v |= O 5 p1 false
for any v. So I, v |= O 5 O 5 p1 O 5 p1 , so DL5 O 5 O 5 p1 O 5 p1 and DL5 O 5 O 5 p1
O 5 p1 by completeness of DL5 . This also refutes (DT) since O 5 (O 5 p1 p1 ) DL5 derives O 5 O 5 p1 O 5 p1 by the use of (C5 ), (Ext5 ), and (M5 ). The same set I refutes
theoremhood of (DB), I = {p1 } refutes (D4), and, e.g., an empty set I refutes (DE).
Now (DT is widely accepted in deontic logics modelled prohairetically. In semantics
employing a deontic ideality relation R, (DT) is valid if R is semi-reflexive, i.e., any
world ideal from some standpoint is ideal with respect to itself. In Hanssons analysis of
monadic deontic logic, where a set of deontic formulas is modelled by a non-empty basis of
ideal among possible worlds, (DT)if admitted as well-formedis immediate from the
universality of ideality (cf. [9, pp. 381382]). The same is true for Andersonian subsets of
sanction-free worlds within a set of worlds linked by an alethic accessibility relation, where
(DT) is derivable as soon as alethic necessities are made to hold at the actual world (cf.
[2, p. 187, Theorem OM45]). Fllesdal and Hilpinen [7] have called (DT) a plausiblelooking candidate for logical truth. And should not, as Prior [18, pp. 255256] has
claimed, the acceptance of (DT) be trivial, reading it as It ought to be the case that what
ought to be the case is the case?
The non-derivability of (DT) in any system DLi brings to the fore the distinction between what Barcan Marcus [15] has called the descriptive and evaluative uses of ought.
Our definitions of O i are descriptive, not evaluative: something is obligatory only if it is
(at least) derivable from what has been (factually) commanded. Thus O 3 O 3 p1 means that
there is an imperative that can be satisfied only if O 3 p1 is true. But having no analytical
content, a violation of this imperative is not excluded, which means that O 3 p1 is false, i.e.,
that there is, as a matter of fact, no imperative that demandsamong other thingsp1 . To
use Marcuss example: that parking on highways ought to be forbidden does not entail, in
the semantics employed here, that parking on highways is forbidden.
For similar reasons, factual statements about the set of imperative demands cannot be
meaningfully employed for conditional commands. Consider the sentence O 3 (O 3 A
A). Does it entail O 3 A? Let I = {O 3 p1 p1 }, so I, v |= O 3 (O 3 p1 p1 ) is true,
whatever v is. However, DL3 (O 3 p1 p1 ) O 3 p1 , as can be seen from I = ,
v (p1 ) = 1. So I, v O 3 p1 . Possibly one may want to admit commands like If it is not
(already) obligatory to do , then do (by virtue of this command)!, and the addressee
should be able to conclude that (now) she has an obligation to do . However, such conditional expressions are not expressible in the language and semantics presented so far.
Exploring how imperative semantics can be extended to cover such and other conditional
expressions, and how to sail around the maelstrom of self-referentiality lurking behind the
above example, must be left to further study.
185
186
61
8. Conclusion
When the question is what ought to be, according to some existing norms or imperatives, it seems natural to employ a semantics that models the meaning of deontic operators
with respect to these imperatives, rather than some notion of absolute or relative betterness. Existing modal systems relate rather simply to ought-operators of different strength,
defined with respect to a set of sentences that describe what the imperatives demand. To
mix such operators is, however, generally not an option if the resulting semantics is to be
represented by a strongly complete axiomatic system, and the same holds for the use of restrictions on the number and kind of imperatives to be considered. To cope with conflicting
imperatives, we may define a van-Fraassen-type operator that prohibits the agglomeration
of contradicting demands, and a sceptical ought-operator according to which something
is obligatory only if there is no conflicting demand; both are best represented in systems
that also contain operators which relate more directly to imperatives. In the light of the
fact that natural languages do not prohibit the use of deontic modalities within imperative
expressions, it is a welcome result that imperative semantics is possible, and sound and
completely axiomatizable for logical languages that permit iterated and nested deontic operators. Here the roads of ought defined evaluatively and ought defined with respect to
given norms part, as demonstrated by the fact that statements like it ought to be that what
ought to be is the case remain well formed, but are true accidentally only.
References
[1] C.E. Alchourrn, The expressive conception of norms, in: R. Hilpinen (Ed.), New Studies in Deontic Logic,
Reidel, Dordrecht, 1981, pp. 95123.
[2] A.R. Anderson, The formal analysis of normative concepts, in: N. Rescher (Ed.), The Logic of Decision and
Action, University Press, Pittsburgh, 1967, pp. 147213.
[3] D.O. Brink, Moral conflict and its structure, Philos. Rev. 103 (1994) 215247.
[4] M.A. Brown, Conditional and unconditional obligations for agents in time, in: M. Zakharyaschev, et al.
(Eds.), Advances in Modal Logic, CSLI, Stanford, 2001, pp. 121153.
[5] E. Conee, Against moral dilemmas, Philos. Rev. 91 (1982) 8797.
[6] A. Donagan, Consistency in rationalist moral systems, J. Philos. 81 (1984) 291309.
[7] D. Fllesdal, R. Hilpinen, Deontic logic: an introduction, in [10], pp. 135.
[8] J. Hansen, Sets, sentences, and some logics about imperatives, Fundamenta Informaticae 48 (2001) 205226.
[9] B. Hansson, An analysis of some deontic logics, Nos 3 (1969) 373398. Reprinted in [10], pp. 121147.
[10] R. Hilpinen (Ed.), Deontic Logic: Introductory and Systematic Readings, Reidel, Dordrecht, 1971.
[11] J.F. Horty, Moral dilemmas and nonmonotonic logic, J. Philos. Logic 23 (1994) 3565.
[12] J.F. Horty, Nonmonotonic foundations for deontic logic, in: D. Nute (Ed.), Defeasible Deontic Logic,
Kluwer, Dordrecht, 1997, pp. 1744.
[13] D. Jacquette, Moral dilemmas, disjunctive obligations, and Kants principle that ought implies can, Synthese 88 (1991) 4355.
[14] S. Kanger, New foundations for ethical theory, in [10], pp. 3658.
[15] Marcus, R. Barcan, Iterated deontic modalities, Mind 75 (1966) 580582.
[16] Marcus, R. Barcan, Moral dilemmas and consistency, J. Philos. 77 (1980) 121136.
[17] I. Niiniluoto, Hypothetical imperatives and conditional obligation, Synthese 66 (1986) 111133.
[18] A.N. Prior, Formal Logic, Clarendon Press, Oxford, 1955.
[19] T.J. Smiley, The logical basis of ethics, Acta Philos. Fennica 16 (1963) 237246.
[20] E. Stenius, The principles of a logic of normative systems, Acta Philos. Fennica 16 (1963) 247260.
[21] B. van Fraassen, Values and the hearts command, J. Philos. 70 (1973) 519.
187
Conflicting imperatives
and dyadic deontic logic
Jrg Hansen
Institut fr Philosophie, Universitt Leipzig, Beethovenstrae 15, D-04107 Leipzig, Germany
Available online 2 June 2005
Editors: A. Lomuscio and D. Nute
Abstract
Often a set of imperatives or norms seems satisfiable from the outset, but conflicts arise when
ways to fulfill all are ruled out by unfortunate circumstances. Semantic methods to handle normative
conflicts were devised by B. van Fraassen and J.F. Horty, but these are not sensitive to circumstances.
The present paper extends these resolution mechanisms to circumstantial inputs, defines dyadic deontic operators accordingly, and provides a sound and (weakly) complete axiomatic system for such
deontic semantics.
2005 Elsevier B.V. All rights reserved.
Keywords: Logic of imperatives; Deontic logic; Conflict of norms
188
485
there is just one thing one actually ought to do, then feelings of regret about having not
acted as one should have are out of place and one should not mind getting into similar
situations again. To avoid the derivation of the ought of a contradiction from two oughts
of equal weight but with contradictory contents, Williams argued that deontic logic should
give up the agglomeration principle
(C)
OA OB O(A B).
Lemmon had no such qualms: he advocated dropping the Kantian Principle ought implies
can
(KP) OA A,
thus allowing for obligations to bring about the impossible, and concluded:
I should like to see a proper discussion of the arguments that go to resolve moral
dilemmas, because I do not believe that this is an area of total irrationality, though I do
not believe that a traditional logical approach (the logic of imperatives, deontic logic,
and whatnot) will do either.
Regarding commands and legal norms, G.H. von Wright [45, Chapter 7], like H. Kelsen
[23, p. 211] at the time, excluded the coexistence of conflicting norms from the same
source: The giving of two conflicting norms is the expression of an irrational will; it is a
performative self-contradiction and as such a pure fact that fails to create a norm. E. Stenius
[40] and later C.E. Alchourrn and E. Bulygin [1] rejected this view: A system of norms
that is impossible to obey might be unreasonable and its norm-giver blameworthy, but its
existence does not constitute a logical contradictionconflicts are ubiquitous in systems of
positive law and logic cannot deny this fact. In his later theory, von Wright [49] concedes
that existing normative systems may or may not be contradiction-free, and reformulates
deontic principles as meta-norms for consistent norm-giving. Kelsen [24] later came to
view logic as inapplicable to law.
2. Van Fraassens proposal and Hortys variation
2.1. Van Fraassens operator O F
Not taking sides, pro or contra the existence of genuine normative conflicts, but arguing
that the view in favor seems at least tenable, B. van Fraassen [11] took up the burden of
finding plausible logical semantics that could accommodate conflicting obligations. The
intended semantics should accept the possible truth of two deontic sentences OA, OA
without committing the norm-subject to the absurd by making O(A A) true, for van
Fraassen wanted to keep the Kantian Principle. Given the existence of certain imperatives
in force, i.e., imperatives that are left as valid, relevant, not overridden, etc. by some unspecified deliberation process, van Fraassens idea was to make these imperatives part of
the logical model, and to describe something as obligatory if it serves to satisfy some, not
necessarily all, imperatives. Formally, let I be the set of imperatives in force, B be the set
of possible states of affairs, and i + B be the possible states of affairs where the imperative i I is considered fulfilled. Let A B be the set of possible states of affairs where
486
189
the indicative sentence A is considered true. Finally, let score(v) be the set of all imperatives that are fulfilled in the state of affairs v: score(v) = {i I | v i + }. Van Fraassen
then defines:
[Df-F]
O F A is true iff
v A : v A : score(v)
score(v ).
So A is obligatory if and only if (iff) there is some score that can be achieved when A is
true, which is not included in any score that could be achieved when A is true. In other
words, A is obligatory iff there are imperatives that can only be (collectively) satisfied
when A is true, but not when A is false.
By slightly changing the viewpoint, van Fraassens proposal might also be described
in the following way: let I be a set not of imperatives, but of indicative sentences in the
language LBL of some basic logic BL. The motivation is that I contains one sentence A
for each imperative i in force that is true in exactly those states of affairs in which the
imperative is fulfilled, i.e., A = i + . BL is assumed to be compact and the turnstile in
BL A means a classical consequence relation that characterizes BL, LBL , A
LBL . Let the remainder set A be the set of all maximal subsets that do not derive A,
i.e., of all such that (i) BL A, and (ii) there is no such that
and A. Then Df-F is equivalent to Df-F* ( means an arbitrary tautology):1
[Df-F ]
O F A is true
iff
I I : I BL A
O F (A B) (O F A O F B)
O F
OF
If BL A B then DL O F A O F B
190
487
To van Fraassens own puzzlement, the cases where agglomeration remains permissible do
not seem axiomatizable: object language does not reveal whether particular A and B of
some O F A and O F B are derived from the demands of imperatives that do not conflict and
so O F (A B) should be supported.2
2.2. The skeptical operator O S
The invalidation of the agglomeration principle by van Fraassens semantics did not
make them popular (cf. Donagan [8, p. 298]). Moreover, let a P F -operator expressing
permission be defined in the usual way as P F A =def O F A, and consider again I =
{A C, B C}: then O F (A C) is true, and so is O F C. Applying (MF ) and (ExtF ),
O F (A C) must be true, hence P F (A C) is false. So not even the obligatory is always
permitted, which seems strange (cf. Jacquette [22]).
In reaction to the dismissal of the agglomeration principle, Donagan [8] and Brink [7]
have claimed that even if there could be a normative demand for A and a conflicting demand for B, with BL A B, it need not follow that the norm-subject has an obligation
to realize A and an obligation to realize B. Rather, there should just be a disjunctive obligation to realize A or B. Given competing normative standards of equal weight, the strategy
of this reasoning is not to trust a single standard, but to consider obligatory only what all
standards demand. Let I be as before. Varying van Fraassens truth definition, Horty [20]
has formalized this skeptical ought as follows:3
[Df-S]
O S A is true iff I I : I BL A
O S A O F B O F (A B)
2 So agglomeration requires some consistency check of the underlying imperatives contents. Van der Torre and
Tan [41,42] proposed a two-phase deontic logic, where consistent aggregation must take place before weakening. For the present imperative semantics, I suggested a bimodal approach in [17] with an operator O 2 that more
directly talks about the imperatives. For comparisons and a new proposal cf. [15].
3 More in parallel to van Fraassens original definition, one may equivalently define:
[Df-S ] O S A is true
iff
v A : v A : score(v) score(v )
488
is valid, and the mixed system DL{F,S} containing the axiom schemes for DLF , DLS ,
the axiom scheme (CSF ), all instances of BL-theorems and modus ponensis sound and
(weakly) complete (cf. [17, Section 6]).
4 That predicaments only arise from an agents own faults, and not through misfortune or the wrongdoings of
others, is a view von Wright and Donagan ascribe to Thomas Aquinas, but this does not seem quite correct: in
the discussion of oaths (Summa Theologica II.II Qu. 89 art. 7 ad 2), Thomas considers the objection that it would
sometimes be contrary to virtue, or an obstacle to it, if one were to fulfill what one has sworn to doso oaths
need not always be binding. In answering, Thomas distinguishes oaths that are unlawful from the outset, where a
man sinned in swearing, and oaths that could be lawful at the outset but lead to an evil result through some new
and unforeseen emergency: fulfillment of such oaths is unlawful.
191
192
489
Sub-ideal demands are usually represented by a dyadic deontic sentence O(A/C), meaning that it ought to be that A given C is true. By accepting all instances of (DD-)
O(A/) P (A/) as a logical truth in [48], von Wright dismisses an inconsistent normative system as conceptual absurdity: if A is obligatory on tautological conditions (i.e.,
unconditionally obligatory), then there cannot be a likewise unconditional obligation to the
contrary. Although von Wright originally used the stronger (DD) O(A/C) P (A/C) for
arbitrary C (axiom A1 of the old system in [44], and axiom B1 of the new system in
[46]), he later turned against it, arguing that while morality makes no conflicting claims,
it is not a logical impossibility that conflicting promises can give rise to predicaments.5
Dyadic operators seem essential for even making this distinction.6
Turning object language oughts into a special sort of conditionals does not necessarily
imply a change in the formalization of the background imperatives: consider the set I =
{(C A), (C A)}, corresponding to background imperatives in the usual way. I is
also its single maximally consistent subset, which derives C, so O F C and O S C are
both true. But a single standard is no longer available once C becomes true: the imperatives
have not all been fulfilled (otherwise one would not be in condition C), and any maximal
set of imperatives that is consistent with the given circumstances cannot contain all. So the
proposal is to call A obligatory in case C iff A is necessary for doing the most that can
be achieved, given the truth of C. Formally:
[Df-DF]
O F (A/C)
iff
I I C: I {C} BL A
So O F (A/C) is true iff there is some set, among the maximal subsets of I consistent
with C, that together with C derives A. This is obviously a conservative extension of the
definition given for the unconditional case, so we may define O F A =def O F (A/).
If a cautious, disjunctive approach were appropriate for cases of conflict, then it would
be hard to see why predicaments should be treated differently: that conflicts must be accounted for at the outset, but analogues of Buridans ass cannot be brought about by fate
or unpredictable human nature, would hardly be plausible. Distrusting any single standard,
such an approach would accept, given the circumstances C, only what is necessary by any
standard that could still be metno crying over spilled milk. Formally:
[Df-DS]
O S (A/C)
iff
I I C: I {C} BL A
So O S (A/C) is true iff all the maximal subsets of I consistent with C derive A, given the
truth of C. This is again a conservative extension of the unconditional case, so one may
define O S A =def O S (A/).
After a comparison of the above definitions with similar approaches namely in the study
of nonmonotonic reasoning, I will give an axiomatic dyadic deontic system DDL{F,S} ,
which I prove to be sound and (only) weakly complete with respect to the above semantics.
5 Cf. [47], [48, pp. 36, 81, 89].
6 Below, I extend the treatment of conflicts to the area of predicaments, and do not follow von Wright in ruling
out conflicts. However, this can easily be done by axiomatically adding (DD-) to the system presented below.
490
4. Comparisons
Though the truth definitions introduced in the preceding section naturally extend the
proposals of van Fraassen and Horty for dealing with normative conflicts to the dyadic
context and the related problem of predicaments, and though their resolution mechanisms
are not exactly new (cf. below), there has not been much discussion of these concepts
in the deontic logic literature. Notably, Hortys own dyadic operator in [21] is defined
with respect to (simply) maximally non-conflicting sets of prima facie oughts, and it is
disregarded that their joint demands may now be inconsistent with the situation. But the
more general literature on nonmonotonic reasoning offers a range of parallel concepts.
Regarding O S , the most obvious parallel is Kratzer and Lewiss premise semantics in
[25] and [30] which has a set of formulas H (the premises) to define counterfactuals in
much the same way as the set I is used here in the definition of deontic conditionals.
Considering Kratzers definition, and setting aside the world-relativity of H , let SH,C =
{H H | H BL C} be the set of all subsets of H that are, according to some basic
logic BL, consistent with C. A counterfactual conditional is defined in the following
way:
H |= C A iff
In other words, C A is true iff each set in SH,C has a superset in SH,C that implies
C A.7 The truth definition is tailored for a basic logic that may fail compactness and so
accommodates sets H with ever-larger C-consistent subsets, but no maximal ones. Here,
BL was assumed to be compact, and we obtain:
Observation 1 (Relation to premise semantics). For any set I LBL :
I |= O S (A/C)
iff
I |= C A
193
194
491
|skept() A =def A
E( )
|cred() A =def A
E( )
(cf. Brass [5] for the analogy and notation). The following is then immediate:
Observation 2 (Relation to Poole systems). For any set I LBL :
I |= O S (A/C)
iff
{C} |skept(I ) A
iff
{C} |cred(I ) A
I |= O (A/C)
Regarding the O F -operator, it is perhaps not quite as obvious that its corresponding P F operator is closely related to the X-logics of Siegel and Forget [10,38]: The consequence
relation |X of these logics holds between a set of formulas and a formula A modulo a
set X of formulas, where the definition is
|X A iff
Cn {A} X = Cn( ) X
As Makinson [31] pointed out, X can be understood as a set of bad propositions that one
is to avoid. So |X A is true iff A can be realized together with without increasing
the set of bad proposition above those that were already true given . Here we have a
set I of desired propositions, so a statement seems bad if it asserts that some desired
proposition be false, e.g., A is true for some A I , or that at least one A1 , . . . , An I is
false, i.e., (A1 An ) is true. Let I = { {A1 , . . . , An } | {A1 , . . . , An } I, 1
n card(I )} be the bad set corresponding to I . We then obtain:
Observation 3 (Relation to X-logics). For any set I LPL , X = I :
I |= P F (A/C)
iff {C} |X A
492
iff
A |E B
iff
s min A : (s) BL A B
A = or s min A : (s) BL A B
Observation 4 (Relation to Bochmans epistemic states). Let I LPL , and let the corresponding epistemic state EI = S, , be such that (i) S = P(I ), (ii) (s) = Cn(s),
and (iii) s t iff t s. Then
I |= O S (A/C)
iff
iff
I |= O (A/C)
C |EI A
PL
C and C |EI A
Proof. I prove first (a) I min A iff I I A, (b) EI is an epistemic state, (c) A =
iff BL A: For (a), by definition A = {I I | I BL A}, so A is the set of subsets
of I consistent with A. I min A means that for any I min A , I = I : I I .
By definition for any I A , I = I : I I . This means there is no I I consistent
with A such that I I , which means I I A. For (b), if I I is in A , i.e., it is
/ I A then by definition of I A there is some I I A
consistent with A, and I
such that I I , so there is some I min A with I I . So EI is smooth, hence
it is an epistemic state. For (c), A = iff {I P(I ) | I BL A} = iff BL A
holds by monotony of BL. Putting together, we get: I |= O S (A/C) iff I I C: I
{C} BL A iff I min C : I BL C A iff C |EI A. Likewise: I |= O F (A/C) iff
I I C: I {C} BL A iff I min C : I BL C A iff C = and [ C =
or I min C : I BL C A] iff BL C and C |EI A.
A final parallel brings us back to deontic logic, namely the multiplex preference semantics of Goble in [1214], where a multitude of preference relations enables definitions like
all-best (universally preferred) and some-best (existentially preferred), which are then
used in definitions of deontic operators. That, in the finite case, such semantics corresponds
closely to the present account will be explicated in Section 6. Regarding meta-theory, for
a somewhat more general semantic setting the skeptical consequence relation was axiomatized by Kraus, Lehmann and Magidor [26], and the credulous consequence relation by
195
196
493
Bochman [2]. However, a completeness proof for a system that includes both8 seems to be
missing so far and this is what I shall now turn to.
5. The dyadic deontic logic DDL{F,S}
Let the basic logic be propositional logic PL: The alphabet has proposition letters
Prop = {p1 , p2 , . . .}, operators , , , , and parentheses (, ). The language LPL is defined as usual. , in front of a set of sentences means their conjunction
and disjunction, and, e.g., ni=1 Ai further abbreviates {Ai , . . . , An }. Semantically, valuation functions v : Prop {1, 0} define the truth of sentences A LPL as usual (written
v |= A), B is the set of all such valuations, and A is the extension {v B | v |= A} of A.
PL is a sound and complete axiomatic system, and PL A means that A is provable in PL.
The alphabet of the language LDDL{F,S} additionally has the operators O F , O S , and
the auxiliary /. DDL{F,S} is then the smallest set such that
{F,S}
iff
I I C: I {C} PL A
I |= O S (A/C)
iff
I I C: I {C} PL A
DDL{F,S} -satisfiable,
If I |= A, A is called
and DDL{F,S} -valid if I |= A for all I LPL
(we write |=DDL{F,S} A).
Consider the following axiom-schemes (* is the uniform index F or S):
(CExt )
(ExtC )
(DM )
(DCS )
(DCSF )
(DNS )
(DPF )
(Cond )
(CCMon )
(RMonF )
8 One might add a third (monadic) deontic modality O 2 that more directly talks about the imperatives to
axiomatize consistent agglomeration, but I must leave the details to future study (cf. [17, Section 6] for the
resulting monadic system DL{2,F,S} ).
494
(TransFSS ) and (TransSFS ), which is again equivalent. The names are from the study of nonmonotonic logics, namely reflexivity, right weakening, preservation, (conjunctive) cautious
monotony, conditionalization and disjunctive reasoning. (Ref ) is Hanssons [18, Theorem 2], (CCMon ) Reschers [36, Theorem 4.4], (Or ) is the right-to-left version of von
Wrights (B3) in [46], and (DR ) Hanssons Theorem 13. Fllesdal and Hilpinen [9] introduced the strong version (FH ) of (RMon ) (their Theorem 77). (FH+ ) is even stronger:
its displayed versions could replace (CCMon ) and (RMon ), =F, FSS, SSF. (Trans )
is transitivity of weak preference given by A B =def P (A/A B) (Lewis [28, p. 54]).
197
198
495
Spohn [39] introduced (P -LoopF ) to define the relevant equivalence classes in his completeness proof of Hanssons DSDL3, and its O-form was rediscovered by Kraus, Lehmann
and Magidor [26] who put it to the same use. Note that (CCMon ) is (P -Cond ), and (Cut )
is (P -RMon ), where the deontic operators are swapped in the P -versions.
Theorem 2. DDL{F,S} is sound.
Proof. The validity of (DM ), (DCS ), (DCSF ), (CExt ), and (ExtC ) is immediate. (DNS ),
(DPF ) are valid since any subset of LPL derives , and any maximally consistent subset
is consistent. If PL C then at least is in I C, hence I C = and then both
(DN-RF ) and (DP-RS ) hold likewise.
(CondF ) Assume O F (A/C D), so there is an I I (C D) such that I {C
D} PL A and I {C} PL D A. Since I PL (C D), also I PL C,
so by maximality there is an I I C such that I I , so there is an I
I C: I {C} PL D A, so O F (D A/C).
S
(Cond ) Assume O S (A/C D). So for all I I (C D) : I {C D} PL A.
If there is an I I C: I {C} PL D A then I {C} PL D and
I PL (C D). By maximality I I (C D): I I . Since I PL
(C D), I PL C, so there is an I I C: I I . Then I I and
by maximality of I I C, I = I and hence I = I . So I is in I (C
D) and I {C D} PL A, but this violates the assumption. So for all I
I C: I {C} PL D A, and O S (D A/C).
(CCMonF ) Assume O F (A D/C), so I I C: I {C} PL A D. Then I
{C} PL D, for otherwise I PL C which is excluded by the definition of
I C. So I PL (C D), by maximality I I (C D): I I and
I {C} PL A, I {C D} PL A. Hence O F (A/C D).
(CCMonS ) Assume O S (A D/C), so for all I I C: I {C} PL A D, and so
I {C} PL D, for otherwise I PL C contrary to the definition of I C,
and so for all I I C: I PL (C D). Suppose I I (C D), so
I PL (C D) and I PL C. By maximality I I C such that I I .
In turn I PL (C D) as just proved, so I I (C D) such that I I .
But then I = I by maximality of I I (C D), so I = I I C and
I {C} PL A as assumed. So I {C D} PL A for any I I (C D).
So O S (A/C D) is true.
F
(RMon ) Assume O F (A/C), so I I C: I {C} PL A. If P F (D/C) then
I I C: I {C} PL D. So I PL (C D). So by maximality I
I (C D): I I , so I {C} PL A, by monotony I {C D} PL A, so
O F (A/C D) is true.
FSS
(RMon ) Assume O S (A/C), so I I C: I {C} PL A, and P F (D/C), so
I I C: I {C} PL D, so I PL (C D). Suppose I I (C D),
so also I PL C and by maximality I I C : I I . We have I
{C} PL A, so B1 , . . . , Bn I : {B1 , . . . , Bn } {C} PL A by PL-compactness.
If I {C} PL A then {B1 , . . . , Bn } I , by maximality I {B1 , . . . , Bn } PL
(C D), but I {B1 , . . . , Bn } I , so I {C} PL (C D) contrary to the
496
P S (p2 /p1 )
O S (pi /p1 )
i=3
P S (p2 /p1 )
O S (pi /p1 )
i=3
P S (p2 /p1 p2 )
O S (pi /p1 p2 )
i=3
P S (p2 /p1 p2 )
O S (pi /p1 p2 )
i=3
is finitely DDL{F,S} -satisfiable: let n be the greatest index of any proposition letter occurring in some finite f . Then
If = pn+1 (p1 p2 ), pn+1 (p1 p2 ), p2 , p3 , . . . , pn
satisfies f . For easy verification, I list the relevant sets of maximal subsets:
If =
199
200
497
2.
Proof. From O S (p2 /), P S (p2 /p1 ) and the validity of (RMonFSS ), follows O F (p1 /), i.e., there is an I1 I : I1 PL p1 . Likewise from O S (p2 /)
and P S (p2 /p1 ) , it follows that there is an I2 I : I2 PL p1 . To
satisfy O S (p2 /) it is necessary that for all I I : I PL p2 , and from
O S (pi /p1 ), O S (pi /p1 ) , and the validity of (OrS ), it is obtained that for all
I I : I PL pi , i 3.
(ii) For each A, there is an IA I A : IA {A} PL p2 .
Proof. Let A {p1 , p1 p2 }. Then by observation (i) I1 I A. Since I1 PL
p2 , to satisfy P S (p2 /A) there is an IA I A such that IA I1 PL A.
So IA {A} PL (p1 p2 pn ) for some n. If n 3, then IA {A} PL
(p1 p2 pn1 ), since IA {A} PL pn is necessary for O S (pn /A) .
So IA {A} PL (p1 p2 ), so IA {A} PL p2 . Likewise, the proof for A
{p1 , p1 p2 } is obtained from I2 I A.
(iii) If A {p1 , p1 p2 } then IA {p1 , p2 , p3 , p4 , . . .} PL .
If A {p1 , p1 p2 } then IA {p1 , p2 , p3 , p4 , . . .} PL .
Proof. Suppose A {p1 , p1 p2 } and IA {p1 , p2 , p3 , p4 , . . .} PL . Then
IA {A, p2 , p3 , p4 , . . .} PL . So IA {A} PL (p2 p3 p4 pn )
for some n. But also IA {A} PL p2 p3 p4 pn by observation (ii)
and from the fact that I satisfies O S (pi /A) , 3 i n. So IA PL A, but
this contradicts IA I A. The proof for A {p1 , p1 p2 } and the set IA
{p1 , p2 , p3 , p4 , . . .} is done likewise.
It follows that Ip1 I(p1 p2 ) PL and Ip1 I(p1 p2 ) PL . This is most easily
seen by appealing to PL-semantics: some v B satisfies {p1 , p2 , p3 , p4 , . . .} and by (iii)
all elements of Ip1 as well as all of I(p1 p2 ) , so their union is satisfiable and therefore
consistent (likewise for {p1 , p2 , p3 , p4 , . . .} and Ip1 I(p1 p2 ) ). From (ii) it follows
that
Ip1 I(p1 p2 ) PL (p1 p2 ) (p1 p2 ) p2
Ip1 I(p1 p2 ) PL (p1 p2 ) (p1 p2 ) p2
But the conclusions are tautologically equivalent to p2 , so there are consistent subsets of
I that derive p2 , and I O S (p2 /), although O S (p2 /) .
Theorem 4. DDL{F,S} is weakly complete.
Proof. The proof follows the completeness proof of Spohn [39] for B. Hanssons [18]
preference-based dyadic deontic logic DSDL3. Since parts of this proof will be reused in
the next section for logics that might not include unrestricted (DN ) or (DP ), I will avoid
their use up to the last step of this proof.
498
A. Preliminaries. We must prove that if |=DDL{F,S} A then DDL{F,S} A for any A LPL .
We assume DDL{F,S} A so A is DDL{F,S} -consistent. We build a disjunctive normal
form of A and obtain a disjunction of conjunctions, where each conjunct is O (B/C)
or O (B/C). One disjunct must be DDL{F,S} -consistent. Let be that disjunct. Let the
be the PL-sentences that contain only proposition letters occur-restricted language LPL
n
, where n is
ring in . Let r(LPL ) be 22 mutually non-equivalent representatives of LPL
the number of proposition letters in . By writing PL-sentences (including and ), we
). We construct a set as follows:
now mean their unique representatives in r(LPL
(a) Any conjunct of is in .
):
(b) For all B, C r(LPL
F
either P (B/C) or O F (B/C) , and
either P S (B/C) or O S (B/C) .
(c) is DDL{F,S} -consistent.
It then suffices to find a set I LPL that makes true all B .
B. Identifying the deontic bases. We identify syntactically what Hansson called the deontic basis in an extension C (Spohn [39] writes C). Monadic deontic logic has just one
basis, dyadic deontic logic usually has one basis for any C, and here there may be several
bases, which expresses some conflict or predicament in case C.
), let
Definition 1. For any C = , C r(LPL
) | O S (A/C) },
OCS = {A r(LPL
F
) | O F (A/C) }
OC = min{A r(LPL
CA = max C r(LPL
) | P F (A/C)
CA
C=
)
Ar(LPL
201
202
499
500
203
, then O F (C OC /D) .
Proof. {C OC } {D} . If O F (C/D) , then the conclusion is trivial. Otherwise P F (C/D) , so PL D C by (C3). For r.a.a. suppose P F (OC /D) .
With O F (OC /C) we obtain O F (OC D/C) by (FH+F ), and PL OC
D by minimality of OC . But then PL D (C OC ), which refutes the assumption. Hence O F (OC /D) and O F (C OC /D) by use of (CExtF ).
D. Identifying the multiple system of spheres. If this were ordinary dyadic deontic logic
with agglomeration and so just one basis OC for any C, we would be almost done: like
Spohn [39] orders his equivalence classes [C] by a relation before, C could be ordered
into C1 , . . . , Cn with C1 = , and Ci+1 = Ci OCi until this equals . S1 , . . . , Sn with
Si = (Ci Ci+1 ), 1 i < n, is then the system of spheres. Here this method fails since
no C C is guaranteed to have a single basis. But as it turns out, C has the structure of a
multiple system of spheres that is similarly identified.
(D1) {} C.
Proof. P F (/) by (DP-RF ), and PL C for any P F (/C) , so
C , C.
(D2) For all C C, O OFC : If C O = , then C O C.
Proof. If C O = then P F (/C O) by (DP-RF ), CC O CF . We
prove CC O = C O: PL (C O) CC O is immediate from (CExtF )
and (C1). If PL CC O (C O) then {C O} PL CC O , so O F (C
O/CC O ) follows from (C7). P F (C O/CC O ) by definition, so is
DDL{F,S} -inconsistent, but we assumed otherwise.
(D3) For all C C: If PL C D, C = D, then O OFD : PL C (D O).
Proof. Either P F (C/D) , so PL D C, C = D (C3). Or O F (C/D) , so
PL O C for some O OFD and PL C (D O).
), O OF {O S }: If D = , then D = (D O).
(D4) For all D r(LPL
D
D
204
501
(D O). (a) is guaranteed by D, and (b) follows from (D3), (D4) and finiteness
). So C = D by (D3).
of r(LPL
E. Canonical construction and coincidence lemma.
)
Definition 3 (Canonical construction). For all C C {}, D r(LPL
F-Succ(D) = {D O | O OFD },
F-Chain(C) be the set of sequences D1 , . . . , Dn , 1 n, where D1 = , Di+1
F-Succ(Di ), Di = Di+1 for any 1 i < n, and Dn = C,
S-Chain(C, D) be the set of sequences D1 , . . . , Dk , Dk+1 , . . . , Dn , 1 k < n, where
S , D =D
i < n, and
D1 , . . . , Dk F-Chain(C), Di+1 = Di OD
i
i+1 for any k
i
Dn = D.
For any C C\{}, C F-Succ(C), let
] be a function that associates a unique proposition letter not
: C {} [Prop\LPL
occurring in with each element of C {},
(C, C ) = (C ) {(C ) | C F-Succ(C), C = C },
(C) = {(C ) | C F-Succ(C)}.
n1
i=1 (Ci , Ci+1 ).
i S [ D1 , . . . , Dk , Dk+1 , . . . , Dn ] = D (C)
(C)
if C = ,
otherwise.
Let I F be the set of all such i F [ C1 , . . . , Cn ], and likewise I S be the set of all such
i S [ D1 , . . . , Dk , Dk+1 , . . . , Dn ]. Then finally I = I F I S .
The definition provides the construction of the canonical set I to make all of true.
F-Succ(C) is the set of immediate contrary-to-duty successors C of C, i.e., O OFC with
C = C O. (D2) showed each C C to be such a successor of (a successor of . . .) ,
and F-Chain(C) is the set of all such chains beginning with and ending with C. is used
to make any two i F [ch(C )], i F [ch(C )], C = C being successors of (successors of. . .)
C, inconsistent with each other and with any i S [ch(C, D)] via . Since C is finite, so is
the number of proposition letters introduced by , I F , I S and I . Regarding the sequences
used to construct I , I use ch(C) for D1 , . . . , Dn F-Chain(C) with C C {} \ {},
), and ch, ch , etc. for
ch(C, D) for D1 , . . . , Dn S-Chain(C, D) with C C, D r(LPL
any sequence for which either holds. We obtain:
(E1) For all ch = D1 , . . . , Dn , PL Di+1 Di and
PL
Di Di+1 , 1
i < n.
502
(E2) If {i F [ch(C)], i F [ch(C )]} PL , then ch(C) is a segment of ch(C ) or vice versa.
(E3) If {i S [ch(C, D)], i S [ch(C , D )]} PL , then C = C and ch(C, D) is a segment of
ch(C , D ) or vice versa.
(E4) If {i F [ch(C)], i S [ch(C , D)]} PL , then ch(C) = C1 , . . . , Ci is a segment of
ch(C , D) = D1 , . . . , Dk , Dk+1 , . . . , Dn , where Dk = C and 1 i k < n.
(E5) No i F [ch(C)] or i S [ch(C, D)] I is a contradiction.
Proof. (E1) is immediate from (D4). (E24) are immediate from the definitions of and
)-conjunct and a [L \L ] . For (E5), first note that each i I consists of a r(LPL
PL
PL
conjunct. Since no proposition letter occurring in one occurs in the other, if i is a con )-conjunct, for any
tradiction, then so must be one of its conjuncts. Regarding the r(LPL
F
i [ch(C)] it is C which must be consistent since C = is excluded. For any i S [ch(C, D)]
)-conjunct is D, and ch(C, D) = D , . . . , D
the r(LPL
1
n with D1 = , Dn = D and
]-conjunct of any i F [ch(C)],
n = 1, so by (E1) D = is excluded. For the [LPL \LPL
n1
ch(C) = C1 , . . . , Cn F-Chain(C), it is i=1 (Ci , Ci+1 ), a conjunction of conjunctions
of non-negated and negated proposition letters, which cannot be a contradiction:
No conjunct (Ci , Ci+1 ), 1 i < n, is a contradiction: For any C = C C, (C ) =
(C ), and no (C ) occurs negated and non-negated in (Ci , Ci+1 ).
If (C ) occurs non-negated in (Ci , Ci+1 ) and negated in (Cj , Cj +1 ), i < j ,
then C = Ci+1 and C F-Succ(Cj ). So there is a ch(C ) F-Chain(C ), ch(C ) =
C1 , . . . , Ci , C , . . . , Cj , C , which violates (E1).
If (C ) occurs negated in (Ci , Ci+1 ) and non-negated in (Cj , Cj +1 ), i < j , then
C F-Succ(Ci ) and C = Cj +1 . PL Cj +1 Ci+1 , so PL C Ci+1 . So there are
O1 , O2 OFCi with C = Ci O1 , Ci+1 = Ci O2 , and PL (Ci O1 ) (Ci
O2 ). Then PL O2 (Ci O1 ), and with (CExtF ) PL O2 O1 . By minimality
O2 = O1 and C = Ci+1 , but (Ci , Ci+1 ) left (Ci+1 ) non-negated.
]-conjunct of i S [ch(C, D)], the case that (C ) occurs non-negated in
For the [LPL \LPL
(Ci , Ci+1 ) and negated in (C) is done like the second case above.
), I I B, I = :
For any B r(LPL
(E6) there is some designated i x I , i x = i F [ch] or i x = i S [ch], such that for all i F [ch ],
i S [ch ] in I , ch is a segment of ch ,
)-conjunct of i x PL-derives the r(L )-conjunct of any i I ,
(E7) the r(LPL
PL
(E8) i x is i F [ch] or i S [ch] with ch = D1 , . . . , Dn such that PL B Dn1 .
)-conjunct
Proof. (E6) is immediate from (E2-4) and finiteness of I . For (E7), the r(LPL
of i is Dn , where Dn is the last member of some ch = D1 , . . . , Dn such that i = i F [ch]
)-conjunct of i x is D for some 1
or i = i S [ch]. By (E6) the r(LPL
k n, and by
k
(E1) PL Dk Dn . For (E8), note that Dn1 exists as ch = is excluded by the
construction. If Dn1 = , then PL B Dn1 is trivial. Otherwise there must be some
ch = D1 , . . . , Dn1 such that i (ch ) I , * being F or S. By (E6), i (ch ) cannot be in
205
206
503
)-conjunct D
x
I , though the r(LPL
n1 derives the r(LPL )-conjunct Dn of i by (E1)
]-conjunct is derived
and hence that of any other i I due to (E7), while its [LPL \LPL
by that of i x . So it must be that {Dn1 } {B} PL , and PL B Dn1 .
):
Lemma 1 (Coincidence lemma). For all A, B r(LPL
I |= O F (A/B)
iff O F (A/B)
I |= O S (A/B)
iff O F (A/B)
r(LPL
PL A as the r(LPL )-conjuncts of any
i I are PL-derived by D (E7), and the [LPL \LPL ]-conjuncts are not rel ). i x is i F [ch] or i S [ch] for some ch =
evant for a derivation of A r(LPL
S
}. So
D1 , . . . , Dn with D = Dn = (Dn1 O), O OFDn1 {OD
n1
F
S
{O} PL B A, so O (B A/Dn1 ) or O (B A/Dn1 ) by
(B1), (B2). From P F (A/B) we get P F ((B A)/B) with (CExtF ).
By (E8) PL B Dn1 , so with (FH+F ) or (FH+SSF ) we obtain O F ((B
A) B/Dn1 ) or O S ((B A) B/Dn1 ) respectively. But
then O B follows from minimality of O. So {B} PL Dn1 O, i.e.,
/ I . So by (E6) I = . Then
{B} PL D, and since {i x } PL D we get i x
F
F
{B} PL A. With P (A/B) and (CExt ) we get P F (/B), so by (DN-RF )
B = and I B = , completing the r.a.a.
Coincidence for O S .
Right-to-left. Assume O S (A/B) and for r.a.a. suppose that there is some I
I B: I {B} PL A. Assume I = , so let i x be the designated member of I ,
)-conjunct. i x is i F [ch] or i S [ch] for some ch = D , . . . , D
and D its r(LPL
1
n
S
with D = Dn = (Dn1 O). Either O OFDn1 , then PL O OD
n1
S
. By (E8) PL B Dn1 , so
follows from (DCSF ), or trivially if O = OD
n1
S
S
S
by (Cond ) PL ODn1 (B OB ), and also I {B} PL O. Chaining the
results, we get I {B} PL OBS , and I {B} PL A by definition of OBS ,
), B = , we
contrary to what was assumed. So I = . For any B r(LPL
504
207
P |= O F (A/C)
S
P |= O (A/C)
iff
iff
P P: bestP ( C ) A
P P: bestP ( C ) A
Likewise, for Lewis-type operators, let LDDL{F,S } be like LDDL{F,S} except that O S replaces O S , and the truth definitions now read:
P |= O F (A/C)
P |= O S (A/C)
iff
iff
P P: v C A : v C A : not v P v
P P: v C A : v C A : not v P v
+
The axiomatic system DDL{F ,S} is like DDL{F,S} except that (DNF ) replaces (DN-RF )and
(DP-RF ) replaces (DPF ). Similarly, DDL{F,S } is like DDL{F,S} except that (DN-RS )
208
505
replaces (DNS ) and (DPS ) replaces (DP-RS ). So all systems only differ on the mindboggling [29] question whether everything or nothing is obligatory in impossible cir+
cumstances. DDL{F ,S} and DDL{F,S } are sound (cf. [14], also Arrows axiom: if
bestP ( C D ) C = , then bestP ( C ) = bestP ( C D ) C , is helpful). I have
no counterexample to compactness, so the semantics might just be compact. Weak completeness is easily obtained from the previous constructions, but seems not to have been
stated before, so I shall give the proof in full.
+
Theorem 5 (Completeness of DDL{F ,S} and DDL{F,S } ). The systems DDL{F ,S} and
DDL{F,S } are weakly complete with respect to the above multiplex preference semantics.
Proof. In proving DDL{F,S} -completeness, up till the coincidence lemma no use was made
+
use of unrestricted (DPF ) and (DNS ) missing in DDL{F ,S} and DDL{F,S } respectively.
So we can reuse and continue that construction with all pertaining lemmas in the canon+
ical construction for DDL{F ,S} and DDL{F,S } , with the implicit understanding that for
+ ,S}
{F
+
the index meant is F rather than F , and for DDL{F,S } the index meant is S
DDL
instead of S.
Let F-Chain(), S-Chain(C, ), be defined as before, C C. We only consider ch =
ch = D D
D1 , . . . , Dn that are in such a set. Let OD
i < n. Note that
i
i+1 , for any 1
i
S }, and O D by (CExt ), so
ch
F
Di+1 = Di O for some O ODi {OD
i
PL ODi
i
O. For any 1 i < j n:
(S1) Dj Di
ch = (and O ch O ch = with CExt , j = n)
(S2) Dj OD
Dj
Di
i
(S3) n =
ch O ch
(S4) OD
Dn1 = B
1
Proof. (S1) and (S2) are immediate from (E1) and the definition of ch. n is finite (S3)
) is finite and repetitions in ch are excluded. For (S4),
since r(LPL
ch
ch
ch
OD
= Dn1 (Dn1 Dn )
= Dn1 OD
B\ OD
n
1
n1
and Dn = by definition.
For any ch = D1 , . . . , Dn , v, v B, let Pch be such that
vPch v
iff
ch
v OD
,
i
ch
v OD
,
j
j <n
By (S2) and (S4), each v must belong to exactly one sphere. The index of each Ci is
transitive and connected, so Pch is as well. LA holds due to (S3) and (S4).
Let ch = D1 , . . . , Dn be as described above. Let Di be its smallest A-permitting
ch A = and j , 1
ch A = (we
sphere, i.e., a Di with OD
j < i < n: OD
i
j
write DA for Di ). We then obtain, for any ch and A = :
(S5) There is a DA ch,
506
209
(S6) A DA , and
ch A .
(S7) bestPch ( A ) = OD
A
Proof. (S5) is immediate from (S1), (S4) and D1 = . For (S6) let DA = Di ch, 1 i <
n: If A
DA , then A DA = , so by (S5) there is a DADA = Dj ch. If DA =
ch = O ch
ch
ch
DADA , then OD
DADA , by construction PL ODA DA , so ODADA
A
A DA = contrary to the definition of DADA . If i < j , then DADA DA ,
ch
but then DADA A DA = and again OD
A DA = . So
A DA
ch
ch
j < i, but then ODA A DA = implies ODA A = , so DA was not the
ch and P .
smallest A-permitting sphere. (S7) then follows from the definitions of OD
ch
A
Finally, let P = {Pch | ch F-Chain() C C S-Chain(C, )}. Note that P = : by
S , then , F-Chain(),
(D1) C, so even if O = for all O OF O
, S-Chain(, ), hence P , = B B P. So P is as required.
), A = and P P and saves us from having
The next lemma holds for all A r(LPL
to do separate proofs for the two systems:
(S8) bestP ( A ) B iff v A B : v A B : not v P v.
Proof. Assume bestP ( A ) B : A = , so bestP ( A ) = due to (LA). So v
bestP ( A ) s.t. v A B . Suppose v P v for some v A B : so v A and v
bestP ( A ) by transitivity of P and definition of best. But then bestP ( A ) B = ,
contradicting the assumption. Assume v A B : v A B : not v P v, and
for r.a.a. suppose that bestP ( A ) B = : So v bestP ( A ) B . Then v
A B by definition of best, and not v P v, as assumed. But v A , so v P v by
definition of best, which completes the r.a.a.
):
Lemma 2 (Coincidence lemma). For all A, B r(LPL
P |= O F (A/B)
iff O F (A/B)
P |= O S (A/B)
iff O F (A/B)
210
507
sides of the iff-clause are true, as is the iff-clause. In the case of DDL{F,S } , if
B = , then P S (/) due to (DPS ), so P S (A/) due to (CExtS ),
/ . Also, if B = , then A B = , and
and by definition of , O S (A/)
since P = there is some P for which it is false that v A B : v
A B : not v P v, so it is not true for all P . So both sides of the iff-clause
are false, and the clause true.
Case B = . Right-to-left. Assume O S (A/B) , and for r.a.a. P O S (A/B), so by
(S8) P P: bestP ( B ) A = . By construction there is some ch =
D1 , . . . , Dn such that P = Pch , ch F-Chain() or ch S-Chain(C, ) for
some C C. Since B = , by (S5) there is some smallest B-permitting sphere
ch
ch
B and OD
B = . Either
DB in ch, with bestP ( B ) = OD
B
B
S
ch
ch
F
DB C and ODB ODB , or ODB = ODB : In both cases, since PL B DB
by (S6), we have O S (B A/DB ) from O S (A/B) and (CondS ), and
SF
ch }
ch
F
{OD
PL B A either by (DC ) and minimality of ODB ODB , or by definB
S . So O ch B = best ( B ) A , which completes the r.a.a.
ition of OD
P
DB
B
/ , so P S (A/B) .
Left-to-right. Suppose P |= O S (A/B), O S (A/B)
508
211
212
509
imperative ! (A B), and the sequence represents the ordering <. Though we
do not know which imperative overrides the weakest imperative !(A B), we know
for sure that it is overridden in these circumstances and so should not be included in
Binding(I ,<) (C D), but with Hortys definition it is. I suggest that, for a better definition of the set TriggeredI (C D), we need an operation like Makinson and van der
Torres [32] basic output, which is expressly tailored to process such disjunctive inputs
(triggering conditions) intelligibly.
Secondly, the inconsistency check seems both too rigid and not rigid enough. For the
latter, let I1 = !(A B), !(B C) and I2 = !((A B) D), !D, !(B C) : in
both cases more important imperatives are in conflict with the weakest, but it is rejected
only in the first. For the former, let I = C ! D, C ! (B D) and the situation be
(C D). C ! (B D) is not in Binding(I ,<) (C D), its consequent contradicting
that of a more important imperative. But this has become unfulfillable, which intuitively
clears the way for obligatoriness of B. For a solution, I propose to leave inconsistency
checks entirely to the (credulous or skeptical) reasoning strategy defined via sets consistent
with the circumstances C: let each of these include a maximally C-consistent subset of
the most important triggered imperatives consequents, a maximal subset of the second
most important triggered imperatives consequents that can be C-consistently added to the
former, etc. This is the incremental maximizing employed for belief revision by Brewka
[6] and Nebel [34] (to work, < must be well-founded). Drawing on a parallel result by
Rott [37, Theorem 7], as long as conflicts between incomparable or equally important
imperatives are allowed, the logic for accordingly defined deontic operators should still be
DDL{F,S} .
Acknowledgement
I thank David Makinson for inspiring remarks, and Lou Goble and Leon van der Torre
for comments on an earlier version presented at EON 2004. The paper is part of a project
begun in [16] to relate deontic logic to reasoning about imperatives.
References
[1] C.E. Alchourrn, E. Bulygin, Unvollstndigkeit, Widersprchlichkeit und Unbestimmtheit der Normenordnungen, in: A.G. Conte, R. Hilpinen, G.H. von Wright (Eds.), Deontische Logik und Semantik, Athenaion,
Wiesbaden, 1977, pp. 2032.
[2] A. Bochman, Credulous nonmonotonic inference, in: T. Dean (Ed.), Proceedings of the 16th International
Joint Conference on Artificial Intelligence, IJCAI 99, Stockholm, Sweden, 1999, Morgan Kaufmann, San
Francisco, 1999, pp. 3035.
[3] A. Bochman, A Logical Theory of Nonmonotonic Inference and Belief Change, Springer, Berlin, 2001.
[4] A. Bochman, Brave nonmonotonic inference and its kinds, Ann. Math. Artificial Intelligence 39 (2003)
101121.
[5] S. Brass, On the semantics of supernormal defaults, in: R. Bajcsy (Ed.), Proceedings of the 13th International
Joint Conference on Artificial Intelligence, IJCAI 93, Chambery, France, Morgan Kaufmann, New York,
1993, pp. 578583.
510
[6] G. Brewka, Belief revision in a framework for default reasoning, in: A. Fuhrmann, M. Morreau (Eds.), The
Logic of Theory Change, Workshop, Konstanz, 1989, in: Lecture Notes in Artificial Intelligence, vol. 465,
Springer, Berlin, 1991, pp. 206222.
[7] D.O. Brink, Moral conflict and its structure, Philos. Rev. 103 (1994) 215247.
[8] A. Donagan, Consistency in rationalist moral systems, J. Philos. 81 (1984) 291309.
[9] D. Fllesdal, R. Hilpinen, Deontic logic: An introduction, in [19], pp. 135.
[10] L. Forget, V. Risch, P. Siegel, Preferential logics are X-logics, J. Logic Comput. 11 (2001) 7183.
[11] B. van Fraassen, Values and the Hearts command, J. Philos. 70 (1973) 519.
[12] L. Goble, Multiplex semantics for deontic logics, Nordic J. Philos. Logic 5 (2000) 113134.
[13] L. Goble, Preference semantics for deontic logics: Part Isimple models, Logique Anal. 46 (2003) 383418.
[14] L. Goble, Preference semantics for deontic logics: Part IImultiplex models, Logique Anal. 47 (2004).
[15] L. Goble, A logic for deontic dilemmas, J. Appl. Logic 3 (2005).
[16] J. Hansen, Sets, sentences, and some logics about imperatives, Fund. Inform. 48 (2001) 205226.
[17] J. Hansen, Problems and results for logics about imperatives, J. Appl. Logic 2 (2004) 3961.
[18] B. Hansson, An analysis of some deontic logics, Nos 3 (1969) 373398, reprinted in [19], pp. 121147.
[19] R. Hilpinen, Deontic Logic: Introductory and Systematic Readings, Reidel, Dordrecht, 1971.
[20] J.F. Horty, Nonmonotonic foundations for deontic logic, in: D. Nute (Ed.), Defeasible Deontic Logic,
Kluwer, Dordrecht, 1997, pp. 1744.
[21] J.F. Horty, Reasoning with moral conflicts, Nos 37 (2003) 557605.
[22] D. Jacquette, Moral dilemmas, disjunctive obligations, and Kants principle that Ought implies Can,
Synthese 88 (1991) 4355.
[23] H. Kelsen, Reine Rechtslehre, second ed., Deuticke, Wien, 1960.
[24] H. Kelsen, Allgemeine Theorie der Normen, Manz, Wien, 1979.
[25] A. Kratzer, Conditional necessity and possibility, in: R. Buerle (Ed.), Semantics from Different Points of
View, Springer, Berlin, 1979, pp. 116147.
[26] S. Kraus, D. Lehmann, M. Magidor, Nonmonotonic reasoning, preferential models and cumulative logics,
Artificial Intelligence 44 (1990) 167207.
[27] E.J. Lemmon, Moral dilemmas, Philos. Rev. 71 (1962) 139158.
[28] D. Lewis, Counterfactuals, Basil Blackwell, Oxford, 1973.
[29] D. Lewis, Semantic analyses for dyadic deontic logic, in: S. Stenlund (Ed.), Logical Theory and Semantic
Analysis, Reidel, Dordrecht, 1974, pp. 114.
[30] D. Lewis, Ordering semantics and premise semantics for counterfactuals, J. Philos. Logic 10 (1981) 217
234.
[31] D. Makinson, Bridges between classical and nonmonotonic logic, Logic J. IGPL 11 (2003) 6996.
[32] D. Makinson, L. van der Torre, Input/output logics, J. Philos. Logic 29 (2000) 383408.
[33] R.B. Marcus, Moral dilemmas and consistency, J. Philos. 77 (1980) 121136.
[34] B. Nebel, Syntax-based approaches to belief revision, in: P. Grdenfors (Ed.), Belief Revision, Cambridge
University Press, Cambridge, 1992, pp. 5288.
[35] D. Poole, A logical framework for default reasoning, Artificial Intelligence 36 (1988) 2747.
[36] N. Rescher, An axiom system for deontic logic, Philos. Stud. 9 (1958) 2430.
[37] H. Rott, Belief contraction in the context of the general theory of rational choice, J. Symbolic Logic 58
(1993) 14261450.
[38] P. Siegel, L. Forget, A representation theorem for preferential logics, in: L.C. Aiello, J. Doyle, S.C. Shapiro
(Eds.), Principles of Knowledge Representation and Reasoning (Proceedings of the 5th International Conference, KR 96), Morgan Kaufmann, San Francisco, CA, 1996, pp. 453460.
[39] W. Spohn, An analysis of Hanssons dyadic deontic logic, J. Philos. Logic 4 (1975) 237252.
[40] E. Stenius, The principles of a logic of normative systems, Acta Philos. Fennica 16 (1963) 247260.
[41] L. van der Torre, Reasoning About Obligations, Thesis Publishers, Amsterdam, 1997.
[42] L. van der Torre, Y.-H. Tan, Two-phase deontic logic, Logique Anal. 43 (2000) 411456.
[43] B. Williams, Ethical consistency, Proc. Aristotelian Soc. (supp.) 39 (1965) 103124.
[44] G.H. von Wright, A note on deontic logic and derived obligation, Mind 65 (1956) 507509.
[45] G.H. von Wright, Norm and Action, Routledge & Kegan Paul, London, 1963.
[46] G.H. von Wright, A new system of deontic logic, Danish Yearbook Philos. 1 (1964) 173182, reprinted in
[19], pp. 105115.
213
214
511
[47] G.H. von Wright, A correction to a new system of deontic logic, Danish Yearbook Philos. 2 (1965) 103107,
reprinted in [19], pp. 115120.
[48] G.H. von Wright, An Essay in Deontic Logic and the General Theory of Action, North-Holland, Amsterdam,
1968.
[49] G.H. von Wright, Deontic logicas I see it, in: P. McNamara, H. Prakken (Eds.), Norms, Logics and Information Systems, IOS, Amsterdam, 1999, pp. 1525.
215
Springer 2006
Institut fur Philosophie, Universitat Leipzig, Beethovenstrae 15, D-04107, Leipzig, Germany
e-mail: jhansen@uni-leipzig.de
Abstract. When a conict of duties arises, a resolution is often sought by use of an ordering of
priority or importance. This paper examines how such a conict resolution works, compares
mechanisms that have been proposed in the literature, and gives preference to one developed by
Brewka and Nebel. I distinguish between two cases that some conicts may remain unresolved,
and that a priority ordering can be determined that resolves all and provide semantics and
axiomatic systems for accordingly dened dyadic deontic operators.
Keywords: deontic logic, logic of imperatives, priorities
1. Introduction
W. D. Ross (1930) argued that whenever there appears to be a conict of
duties, through careful study of all aspects of the situation one will arrive
at the conclusion or rather: the considered opinion that one of these
duties is more pressing than others, and this duty is then ones duty sans
phrase, whereas the others were prima facie only. Ross gives the following
example:
216
JORG HANSEN
217
3
hI; f i OA
iff
I f PL A:
denes a normal modal logic, i.e. the set of LDL -sentences dened as true for
all tuples hI; f i equals the axiomatically dened set that contains all LDL instances into tautologies, furthermore all LPL -instances into
Ext
M
C
OA ^ B ! OA ^ OB
OA ^ OB ! OA ^ B
O>
and is closed under modus ponens. Furthermore, the above truth denition
denes standard deontic logic SDL, which adds the deontic scheme (D):
D OA ! PA
i hI; fi is required to be such that I f is consistent (as usual, PA abbreviates :O:A). Requiring I f to be consistent excludes conicts between
imperatives and is thus a severe and in this case unwanted restriction, for to
show how conicts are resolved they must rst be semantically modeled.
But if e.g. two imperatives !p1 and !( p2 ^ :p1 ) can both be in I, then not
only does (D) fail, but also (td-1) is not very useful, making OA true for
any A 2 LPL . Instead, the following denition for a disjunctive ought
4
operator was put forward:
td-2
hI; f i OA
iff
8C 2 If ? : Cf PL A:
218
JORG HANSEN
this solution is easily adapted to the dyadic case and the related problem of
dilemmas. Dyadic deontic logic uses a language LDDL that employs the
additional auxiliary sign / and is like LDL , except that clause (a) now reads
(a) for all A; C 2 LPL ; OA=C 2 LDDL ,
where O(A/C) is read as A is obligatory in the circumstances characterized
by C. The truth denitions for dyadic deontic formulas should allow some
inuence of the circumstances, so e.g. if John must either not impregnate
Suzy Mae or marry her, and she is in fact pregnant by him, then marrying her
seems to be what he must do. But simply putting
td-3
hI; f i OA=C
iff
I f [ fCg PL A:
will not suce if subjects can get (themselves) into dilemmas, i.e. situations
where the norms are collectively satisable at the outset, but due to misfortune or failure they cannot all be satised anymore. To handle such situations, and to e.g. prevent the derivation of O(^/p1) when I contains
!:p1 _ p2 and !:p2 ^ p3 , the truth denition for a disjunctive dyadic
ought operator can be given as:
td-4
8C 2 If:C : Cf [ fCg PL A:
If PL C $ D
219
5
220
JORG HANSEN
becomes involved in a trac accident. The law requires her to stay at the
accident site until the police have recorded it, which wont happen before 6:00.
Then she can only get to one place, the post oce or the day care center, on
time. The law takes priority over her other duties, but a ranking of these is not
obvious; in particular it is dicult to say which violation could have worse
consequences, and Mirjam will have a hard time making up her mind.
To formalize the reasoning about priorities when faced with conicting
demands, Horty (2003) proposed that the priority ordering is used to rst
determine a set of binding imperatives in the set of all imperatives:
iff
I examine how this denition6 copes with the examples and variants:
221
7
222
JORG HANSEN
Thus Hortys set Binding solves simple cases, but is not adequate for
complex hierarchies where more than two imperatives may be in conict, and it
makes life too easy when conicting higher ranking imperatives become unfulllable or are themselves overridden.7 To overcome these diculties when
formalizing that disregard for a lower ranking imperative can (only) be excused
by obedience to higher-ranking ones (and not vice versa), I suggest that neither
remainder sets of I, nor of a xed subset Binding, but an incremental maximizing strategy should be used. For a situation C, the relevant sets are constructed by rst adding a maximal set of the most important imperatives such
that their demands do not derive :C, then adding a maximal subset of the
second most important imperatives that can be added without the corresponding demands now deriving :C, etc. Introduced by Rescher (1964, p. 50),
such incremental maximizing was more rigorously dened and employed for
the purpose of theory revision by Brewka (1989, 1991) and Nebel (1991, 1992).
Both employ a strict partial order, i.e. < is irreexive and transitive. Nebel
additionally assumes < to be the asymmetric part of a complete preorder ,
i.e. obtained from a reexive, transitive and connected ordering via dening
i<j i i j and j 6 i. Both agree that < must be well-founded, i.e. innite
descending chains are excluded.8 For any <, Brewka denes a full prioritization to be any (strict) well-order on the given set that preserves <, i.e. for all i,
j: if i< j then i j. Clearly:
THEOREM 1 (Existence of full prioritizations). For every wellfounded strict partial order < on a set C there is a full prioritization , i.e. a
strict well-order that is order-preserving with respect to <.
Brewka then denes subsets of the set as preferred subtheories. Calling
them preferred remainders (they are not theories here), his denition translates thus:9
(S
S
A
f
f
SA
#j [ fig if ji S#j [ fi g0PL A, and
S
A
otherwise,
ji S#j
ji
i2I
A
SA
#i and C S .
(i) bans the empty set from I + A for tautological A, and (ii) recursively
denes S#i to include all elements of some such set for a prior element j,
adding i if possible without the corresponding set deriving A. C is the union
223
9
of all such sets. I drop superscripts if the meaning is clear. The following is
almost immediate:
iff
9i 2 I : 8j < i : C \ j D \ j and C \ i D \ i
i.e. by preferring D over C i both agree for all priority classes up to some [i],
of which D contains all elements of C plus more. Then choosing a maximally
N-preferred set among all C I with Cf 0PL A equals choosing from I + A:
There is an alternative, non-constructive denition of Brewkas preferred remainders, attributed to Ryan (1992) by Rintanen (1994) and also
appearing in Sakama and Inoue (1996): Let hI; f; <i be a prioritized imperative structure, where < is a well-founded strict partial order on I, and dene
p IfA to be the set
fC 2 IfA j 9 : 8D 2 IfA n fCg : 9i 2 C n D : 8j 2 D n C : i jg:
224
10
JORG HANSEN
td-6
This truth definition fares better in dealing with the above examples:
225
11
overrides her familial duty, so !A < !C < !B, which, being its only full
prioritization, yields I + ={{!A, !B}}. So Mirjam must wait and post the
letters, as it should be.
T
4.1.
Alchourron and Makinson (1981) seem to have been the rst to logically
examine the idea of resolving contradictions in a body of norms, or contradictions that arise from such a body together with some set of true
empirical facts, by imposing an order upon that body. The object of their
study is a set of regulations that is partially ordered by a relation , which
does not necessarily stand for an ordering by priority or importance. Rather,
i j means that j is as much exposed, or more exposed, to the risk of
legislative derogation as i. If a conict occurs between two parts of the code,
or between the code and some empirical facts, the aim is to nd a (possibly
maximal) non-conicting subset that is most secure from the changes which
the law-giver will presumably enact upon learning of this situation. Their
denition translates to the present framework as follows:
iff
D 6 [ and 8i 2 C : 9j 2 D : i < j:
So a subset is strictly less exposed than some other if for any member
of the rst there is a member of the second which is strictly more exposed.
To see how this approach compares to Brewka and Nebels, consider three
cases:
226
JORG HANSEN
Case 1. Let hI; f; <i be !p1 < !p2 ^ :p3 <!p3 , the imperative in the
middle conicting with the lowest. Then we have
If ? ff!p1 ; !p2 ^ :p3 g; f!p1 ; !p3 gg and
I +? ff!p1 ; !p2 ^ :p3 gg.
The exposure criterion yields f!p1 ; !p2 ^ :p3 g AM f!p1 ; !p3 g: for each left
member, a right member is strictly more exposed, namely !p3.
Case 2. Let hI; f; <i be !p1 < !:p1 ^ p2 < !p3 , so the middle now
conicts with the higher-ranking imperative. Then we have
If ? ff!p1 ; !p3 g; f!:p1 ^ p2 ; !p3 gg and
I +? ff!p1 ; !p3 gg.
But f!p1 ; !p3 g 6AM f!:p1 ^ p2 ; !p3 g: from the left set, !p3 is not less exposed
than !:p1 ^ p2 from the right. The authors recognize that a conict between
higher-ranking norms excludes lower-ranking norms from a least exposed set
and propose to use relevant logic for determining conicts as a cure (Alchourron and Makinson 1981, p. 139).
Case 3. Let hI; f; <i be !p1 < !:p1 ^ p2 ^ :p3 < !p3 , the middle now in
conict with both ends of the hierarchy (by whatever logic). Then we have
If ? ff!p1 ; !p3 g; f!:p1 ^ p2 ^ :p3 gg and
I +? ff!p1 ; !p3 gg.
Yet f!:p1 ^ p2 ^ :p3 g AM f!p1 ; !p3 g: from the right set, !p3 is more exposed than any left member. Mediocrity rules! But even if !p3 is more exposed
to legislative change, if that change came about and removed !p3, the right set
would still contain a member that ranks higher than any in the left.
Prakken, pursuing an argumentative approach, wants to employ Alchourron and Makinsons criterion at the heart of his rebuttal mechanism
used to determine justied arguments (derivations from facts and defaults.
But the criterion he presents (Prakken 1997, p. 192) translates dierently:
iff
9j 2 D : 8i 2 C : i<j:
227
13
!p2 < !p4 ^ :p2 . E.g. p1, p2 may be primary targets and p3 ^ :p1 ; p4 ^ :p2
respective secondary ones, where reaching the secondary target includes
failing to reach the (better) primary one. Reaching both primary targets
seems best, and in fact f!p1 ; !p2 g AM f!p3 ^ :p1 ; !p4 ^ :p2 g: for every
member in the left set there is a lower-ranking one in the right. Also
I + ? ff!p1 ; p2 gg, since all four full prioritizations yield this preferred
remainder. But f!p1 ; !p2 g 6P f!p3 ^ :p1 ; !p4 ^ :p2 g as no member in the
right set ranks lower than all in the left. So Prakkens criterion appears even
less suited to our task than Alchourron & Makinsons.11
Sartor (1991) used Alchourron and Makinsons criterion for a
prevailing relation between subsets modulo a rejected sentence A,12 as
follows:
i3 : !p2
228
JORG HANSEN
and the ordering i1 < i2 < [i3, i4]. Intuitively, to satisfy i1 takes priority, and
then the only choice is the one between the equally ranking i3 and i4, and indeed,
I + ? ffi1 ; i3 g; fi1 ; i4 gg. But the remainder {i2, i3, i4}, missing the most
important imperative i1, is also in prefI f ?, as
a fi1 ; i3 g 6 ?
S fi2 ; i3 ; i4 g;
b fi1 ; i4 g 6?
S fi2 ; i3 ; i4 g:
For (a), consider {i4}, which is a subset of the right hand set: it conflicts with the
left hand set, so for the relation to hold, a strictly less exposed subset of the left
set must also conflict with {i4}. The only such subset is the left set itself, but since
for its member i3 there is no strictly more exposed member in {i4}, it is not
strictly less exposed. The refutation of (b) works similarly using {i3}.
So Sartors denition also produces counterintuitive results where Brewka
and Nebels approach does not.13
4.2.
UTILITY-REFLECTING PRIORITIES
Regarding the neighboring realm of epistemic logic, and the related problem
of revising belief sets in the face of conicting information, such information
often nds the reasoner less willing to give up some beliefs than others. In an
attempt to allocate this ordering of epistemic importance a role in determining which of the contradictory beliefs should be given up, Gardenfors
(1984) proposed the following: Let K be a belief set (set of descriptive sentences) that is the logical closure of some nite basis, and a relation (of
epistemic importance) that is a complete preorder on this set, which additionally ranks logically equivalent beliefs equally. For any remainder
C 2 K ? A there is then a spanning sentence SC in C that derives any
element in C. Then for any C; D 2 K ? A:
C G D
iff SC <SD
So a remainder is preferred to some other i its spanning sentence is epistemically at least as important as that of the other. It is essential for the
construction that < is a complete ordering on K, which due to logical closure includes the spanning sentence that is the sum of a remainder. But, the
logical philosophers not being kings, a set of imperative-contents is rarely
logically closed, which precludes a direct parallel. Yet, choosing subsets that
in sum are the most important has an analogue if the ordering of the
imperatives reects not so much their importance or rank of the source, but a
measure of goodness or utility of the outcome when satisfying the
imperative. For this, let the (well-founded, strict partial) order <u correspond to a function u : X ! R, with I f X LPL , that assigns a real
229
15
uD f
8C 2 I u :C : Cf [ fCg PL A:
4.3.
230
JORG HANSEN
(ii) for all D I, if D C then D f 0PL A. < being a strict partial ordering on
I, the following denes the set of all the least important imperatives of
A-kernels of I:
rIgA def fmax< C j C 2 IgAg:
def
InrI gA:
To see how this works, consider again the example of the road accident:
231
17
too easy for the norm subjects, and some moderation appears necessary.
Following the idea that the removal mechanism should somehow be adjusted
to the sets shrinking it brings about, a moderated version of safe contraction
can be dened as follows:
b<a
MA
<;b [ min< rIn
b<a
MA
<;b gA:
ScardI A
Finally MA
M<;a and
.
< def
a1
So if X are the <-minimal elements in rIgA, then the moderated mechanism rst puts X in the set M< of elements to be removed, then the
<-minimal elements in rIn X gA, etc. Thus elements get removed in
each step until there is no A-kernel left in I minus the last version of M< , which
also means that the cardinality of I suces for the indices (I omit superscripts if
the meaning is clear). To see how this works, consider again the above example:
is some full
232
JORG HANSEN
underlying norms. Whether all conicts and dilemmas are thus avoidable is a
matter of dispute. W. D. Ross may be understood as claiming that conicts
are merely apparent and that by weighing all relevant facts and reasons, it can
be decided which of the conicting prima facie duties are really our duties (cf.
Ross 1930, Searle 1980, p. 242). G. H. von Wright stated that an axiological
order can provide a safeguard against any genuine predicament (cf. von
Wright 1968, p. 68, 80). And Hares description of critical moral thinking,
that lets principles override other, less important principles, suggests that this
process can overcome all moral conicts (Hare 1981, p. 43, 50). On the other
hand, Barcan Marcus (1980) and Horty (2003, p. 564) have argued that if it is
the presence of certain facts that determines the ordering, i.e. this is not an
arbitrary hacking through the Gordian knot, then situations might be
incomparable (if all such facts are missing), or be completely symmetrical (e.g.
identical obligations towards identical twins), so conicts remain possible.
Rather than take sides in this controversy, I will examine what is required if
the method of Brewka and Nebel is to resolve all conicts and dilemmas. For
all possible17 situations C; I + :C must then be a singleton: otherwise there are
C; D 2 I + :C such that Cf [ Df PL :C, and there is a dilemma. So I dene:
233
19
Case 2: Let hI; f; <i be !p1 ; !p2 <!:p1 ^ p2 , so the demands of the two
higher-ranking imperatives run contrary to the lower ones. Either both
!p1,!p2 are in S 2 I + :C, so !:p1 ^ p2 cannot be consistently added, or
S#!:p1 ^p2 f [ fCg PL :p1 ^ p2 , so adding !:p1 ^ p2 adds nothing.
Case 3: Let hI; f; <i be !p1 ^ p2 <!p1 ^ :p2 <!p1 ; !p2 . If p1 ^ p2
is consistent with C, then I + :C is ff!p1 ^ p2 ; !p1 ; !p2 gg. Otherwise,
fCg PL :p1 ^ p2 . If p1 ^ :p2 is consistent with C, then I + :C is
ff!p1 ^ :p2 ; !p1 gg. Otherwise, also fCg PL ; :p1 ^ :p2 . Hence fCg PL :p1 .
Then any set in I + :C contains at most !p2, depending on whether C is consistent
with p2.
Unable to make out a necessary and sucient requirement for
cardI + :C 1, without reference to particular C, I can only rephrase its
denition as follows:
Consider the system PD that was sound and (weakly) complete with respect
to prioritized imperative semantics. The system that results when (RMon) is
added is Hanssons (1969) DSDL3 as axiomatized by Spohn (1975):
234
JORG HANSEN
6. Conclusion
Describing how a resolution of normative conicts using priorities works is
surprisingly dicult. A method based on a proposal by Horty could not solve
complex cases where more than two norms conict or overriding norms are
no longer satisable or are themselves overridden. A method developed for
theory revision by Brewka and Nebel, which creates maximally non-conicting sets by starting with a maximal set of what is most important and
235
21
Suppose that b does not mind if you have one drink with c, and c does not
care that you may be driving, and let the three imperatives be ranked in descending order. One may be tempted to reason as follows: consider rst the
imperative in line (1), but it has not yet been triggered as you have not yet
drunk anything, so it is set aside. Regarding (2), its condition is true, so you
must do the driving. Still, only (3) is triggered, so you should have a drink with
c. But satisfying (2) and (3) both triggers and violates the highest-ranking
imperative. Is it not more prudent to violate one of the lower-ranking imperatives instead of the higher-ranking one? For a solution, we need an adequate
denition of triggering (that can handle e.g. disjunctive inputs, like Makinson
and van der Torres basic output (Makinson and van der Torre 2000, 2001),
and to nd a maximizing strategy that is consistent with the above intuition. It
is clear that the present discussion has not provided the tools to properly
address such problems, so these must be left to further study.
236
JORG HANSEN
Acknowledgments
I thank David Makinson for valuable help with an earlier version, and Lou Goble,
John F. Horty and Henry Prakken for helpful discussions and clarifying correspondence. An earlier version of the paper was presented to the working group Law
and Logic of the XXII World Congress of Philosophy of Law and Social Philosophy
(IVR 2005) 2429: May 2005, Granada.
Notes
1
Though some discussed approaches cover conditional imperatives, or entities that can be
interpreted as such, these cause problems that are best considered separately.
2
PL is based on a language LPL , dened from a set of proposition letters Prop ={p1,
p2,...}, Boolean connectives :, ^, _, , M and brackets (,) as usual,
The truth of a LPL -sentence A is dened recursively using a valuation function v :
Prop ! f1; 0g (I write v A), starting with v p i v(p)=1 and continuing as usual. If
A 2 LPL is true for all valuations it is called a tautology. PL is the set of all tautologies,
and this set is used to dene provability, consistency and derivability (I write C PL A) as
usual. > is an arbitrary tautology, and ? is :>.
3
E.g. (td-1) most closely resembles denitions of Kanger (1957) and Alchourron and
Bulygin (1981). For authors belonging to this tradition cf. Hansen (2001), Section 1 and
Hansen (2004), fn. 1, in addition to which Ziemba (1971) must be mentioned.
4
Cf. Horty (1997). The disjunctive ought is more commonly referred to as skeptical
non-monotonic inference. Horty (2003) attributes the proposal to Brink (1994), yet the idea
to use such a denition for (dyadic) deontic logic already appears in Lewis (1981). For
alternatives in the deontic-logical treatment of normative conicts cf. Goble (2005).
5
Cf. Kraus et al. (1990), where, however, the proofs are done in a more general setting. Also
cf. my (Hansen 2005) for constructive proofs in the manner of Spohn (1975) as well as more
comparisons and truth denitions and axioms for an alternative credulous O-operator.
6
Hortys denition only employs circumstances to derive consequents from a set of conditional imperatives, but this has no eect on the solution of the examples.
7
Horty is preparing a rened version of Binding that solves all of the examples (private
correspondence).
8
Brewka (1991) and for Nebel cf. Rott (1993, fn. 9). For the rationale, let the ordered I
be hi0 ; . . . ; i0:125 ; i0:25 ; i0:5 ; i1 i, with i1 !p; i0:5 !:p; i0:25 !p; i0:125 !:p, etc., and i0=!q.
We cannot tell whether p or :p is obligatory, but this is not a case of conict either, all
imperatives !p being overridden by ones demanding :p, and vice versa.
9
I use notation from both, Brewka (1989, 1991), Brewka and Eiter (1999), and Nebel
(1991, 1992).
10
Also cf. Sartor (2005, p. 734): preference must be given to the argument such that its
weakest defeasible subreasons are better than the weakest defeasible subreasons in the other.
11
Prakken could argue that he only compares minimal conict pairs (subarguments), which f!p1 ; !p2 g; f!p3 ^ :p1 ; !p4 ^ :p2 g is not, while f!p1 g P f!p3 ^ :p1 g and
f!p2 g P f!p4 ^ :p2 g hold. But let I f!p1 ; !p2 ; !p3 ^ p4 ! :p1 ^ p2 ; !p4 ^ p3 ! :
p1 ^ p2 g, with !p1 < !p3 ^ p4 ! :p1 ^ p2 and !p2 < !p4 ^ p3 ! :p1 ^ p2 , so
with the primary targets one bonus secondary target is reachable. Still
f!p1 ; !p2 g 6P f!p3 ^ p4 ! :p1 ^ p2 ; !p4 ^ p3 ! :p1 ^ p2 g, and this is a minimal
conict pair. Yet intuitively, the argument for p1 p2 should win over any for :p1 ^ p2 .
237
23
12
Sartor (1991) simply rejects contradictions, yet the adjustment to any A is immediate.
The indicative version of the example also shows that Prakken cannot avoid counterintuitive results by replacing, in his denition of a rebuttal in (1997), his own relation P
by the relation AM of Alchouron and Makinson for the comparison of minimal conict
pairs: then all arguments for q are rebutted by arguments for :q as demonstrated, but
intuitively any consistent argument including i1 cannot be defeated and so the argument
for q should win.
14
For resolving legal arguments by summing up weights of reasons cf. Hage (1991, 1996).
15
Also cf. Iwin (1972, p. 486): When we are in a situation compelled to satisfy two
obligations requiring contradictory actions, then the most natural way out of this diculty consists in comparing the two obligations and not satisfying the less important
one.
16
I=:C is the notation in Alchourron and Makinson (1985), whereas the notation in
Alchourron (1986) would be I/C. Alchourrons own truth denition for deontic operators
employs a deontic logic as basic logic and conditional imperatives that are not treated
here.
17
If C is a contradiction, then by denition I + :C [.
18
I owe this insight to Leon van der Torre (private correspondence).
13
Appendix: Proofs
THEOREM 1 (Existence of full prioritizations). For every wellfounded strict partial order < on a set C there is a full prioritization , i.e. a
strict well-order that is order-preserving with respect to <.
Proof. Let < be a well-founded strict partial order on the set C. Let each
x 2 C be assigned an ordinal ax in the following way: ax=0 for the elements
in min< C, and for any other x 2 C; ax supfay jy<xg 1, where sup denotes
the supremum of a set of ordinals. Transnite induction available for wellfounded partial ordered sets tells us that ax is well-dened for any x 2 C. Let
the equivalence class x fy 2 Cjay ax g. Finally is an arbitrary strict
well-order on elements of the same equivalence class, and x y if ax is
smaller than ay. Clearly is a strict well-order on C. To prove x y if x<y,
for any x; y 2 C, suppose x<y. Then ax is in {az|z<y}, so ay=sup
{az|z<y}+1 is at least ax+1. Hence x y.
238
JORG HANSEN
239
25
240
JORG HANSEN
(c) Non-compactness. Leaving details to the reader, I gave a counterexample to compactness of basic imperative semantics in Hansen (2005, Theorem 3), i.e. a non-satisable set C LDDL of which all nite subsets are
satisable, which applies here as well. The tricky part is to show that the sets
Ip1 [ Ip1 $:p2 and I:p1 [ Ip1 $p2 are in I +?, which works using the complete
preorder induced by ordinal labels from the proof of Theorem 1 (for each
union choose a full prioritization that puts its elements -rst in any
equivalence class and prove that S in I +? constructed from equals the
respective union).
is some full
be the first set in the construction of M that includes next(i) if that is nonempty and otherwise the whole of M . Then
Mi n nexti M \ fj 2 I j j ig:
241
27
Left-to-right:
Suppose
i
2
6
S
ji S#j [
S
Sfi g PL A
f
and ji S#j 0PL A. By the induction hypothesis
ji S#j
S \ fj 2 I j j ig I n M \ fj 2 I j j ig. By the above lemma, M \
fj 2 I j j ig Mi n nexti and so I n M \ fj 2 I j j ig I n M \
fj 2 I j j ig \ fj 2 I j j ig I n Mi n nexti \ fj 2 I j j ig, which,
242
JORG HANSEN
Proof. For the left-to-right direction (in contraposition), suppose there are
j1, j2 2 [i], a consistent D 2 LPL and C 2 I + :D with neither fi0 2 C j i0 <igf
[fDg PL j f1 ! j f2 nor fi0 2 C j i0 <igf [ fDg PL j f2 ! j f1 . Consider C
D ^ :j f1 ^ j f2 . If fi0 2 C j i0 <igf PL :C, then fi0 2 C j i0 <igf PL :D_
j f1 ^ j f2 , which includes fi0 2 C j i0 <igf [ fDg PL j f1 ! j f2 , contrary to what
we supposed. Let 1 be a full prioritization of < that for all i <i is like C,
i.e. it contains fi0 2 C j i0 <ig, and then puts j1 1 -first among the members of
[i]. Likewise let 2 be a full prioritization of < that contains fi0 2 C j i0 <ig
and puts j2 2 -first among the members of [i]. fi0 2 C j i0 <igfSPL :C is excluded, so regarding the constructions S1 ; S2 2 I + :C; k1 j1 S1 #k
S
S
fi0 2 C j i0 <ig k2 j2 S2 #k . If k1 j1 S1 #k f [ fj f1 g PL :C, then fi0 2 C j
i0 <igf [ fj f1 g PL :D _ j f1 ^ j f2
or
equivalently
fi0 2 C j i0 < igf [
S
fDg PL j f1 ! j f2 , but that was excluded. So k1 j1 S1 #k f [ fj f1 g0PL :C and
so j1 2 S1 , and likewise j2 2 S2 is proved. But then S1 6 S2 otherwise
S1 f PL j f1 ^ j f2 and so S1 f PL :C, which is excluded. So card
I + :C 6 1 for some non-contradictory C 2 LPL .
For the right-to-left direction, assume for r.a.a. that cardI + :C 6 1 for
some consistent C 2 LPL , so there are two full prioritizations 1 and 2
243
29
such that S1 6 S2 for the according S1 ; S2 2 I + :C. Then there is a i 2 I
such that S1 \ i 6 S2 \ i: let [i] be the <-first such equivalence class. So
there is a j1 2 [i] with either j1 2 S1 and j1 62 S2 or vice versa. This being
S
equal, assume the former, so k2 j1 S2 #k f [ fj f1 g PL :C. So there is a
S
(j f1 ! :C)-kernel X in k2 j1 S2#k , and X \ i 6 [, else X S1 by choice
of [i] and since j1 2 S1 we obtain S1 f PL :C, which is excluded. We get
fj 2 S2 j j < igf [ fCg PL j f2 ! j f1 or fj 2 S2 j j < igf [ fCg PL j f1 ! j f2
from the right side of the iff-clause for any j2 2 X S
\ i. Suppose the first case
holds for a j2 2 X \ i. Since j2 2 X k2 j1 S2#k we obtain
S
k2 j1 S2#k f PL C ! j f1 , and since the ( j f1 ! :C)-kernel X is in
S
S
f
k2 j1 S2#k , this yields k2 j1 S2#k PL :C, which is excluded. So for all
f
f
j2 2 X \ i: fj 2 S2 j j<igf [ fCg
S PL j 1 ! j 2 . For all other k 2 X, i.e.
k 62 i, we have k 2 i by X k2 j1 S2#k and so k 2 S1 since j1 2 [i] and
due to the choice of i. So since j1 is in S , and also fj 2 S2 j j<ig S by the
choice of i, for any k 2 X : S f [ fCg PL kf . Hence, S f PL :C, which
the construction of the antecedent excludes. This completes the r.a.a.
244
JORG HANSEN
prioritized imperative structure hI; f; <i that models A. We build a disjunctive normal form of A and obtain a disjunction of conjunctions, where each
conjunct is O(B/D) or :OB=D. One disjunct must then be DSDL3-consistent. Let d be that disjunct. Let the d-restricted language LdPL be the PLsentences that contain only proposition letters occurring in d. Let rLdPL be
n
22 mutually non-equivalent representatives of LdPL , n being the number of
proposition letters in d. LPL -sentences now mean their representatives in
rLdPL . Construct a set D & LDDL such that:
(a) Any conjunct of d is in D.
(b) For all B; D 2 rLdPL : either P(B/D) or O:B=D 2 D.
(c) D is DSDL3-consistent.
It then suces to nd a uniquely prioritized imperative structure that
makes true all of D. We identify what Hansson (1969) called the deontic
V basis
~ by letting OC fA 2
in an extension kCk (Spohn 1975 writes C)
rLdPL j OA=C 2 Dg be the sum of everything demanded in the situation
C. From this denition and (b), (c) it follows immediately that OOC =C 2 D.
Furthermore, observe
(O1)
OA=C 2 D iff PL OC ! A:
PL Cj ! Ci
(O3) Cj ^ OCi ?
Proof. Immediate (due to A0 and O1, PL OCj ! Cj so also OCj ^ OCi ?).
(O4) Ci 2 rLdPL ; 1
245
31
PL OA $ OCA ^ A
Proofs. (O7) is immediate from (O6). Regarding (O8) let CA=Ci, so for all
j < i : OCj ^ A ?, so PL A ! > ^ :O> ^ . . . ^ :OCi1 and thus PL A
! Ci . For (O9), if CA exists, then by the construction of D either
PA=CA 2 D or O:A=CA 2 D, but in the second case OCA ^ A ? due to
(O1), which is excluded by the definition of CA. Regarding (O10): due to (O8)
and derivability of (Cond) we obtain OA ! OA =CA 2 D from
OOA =A 2 D, so PL OCA ^ A ! OA , which is the right-to-left direction. If
A ?, then OA ? since PL OA ! A due to (A0) and (O1) and then the leftto-right direction is trivial. Otherwise PA=CA 2 D due to (O9) and
OOCA =A 2 D is obtained from OOCA =CA , (O8) and derivability of
(RMon). Thus PL OA ! OCA and due to (A0) and (O1) also PL OA ! A.
Both include the left-to-right version.
Canonical construction of hI; f; <i: Let I fC ! OC j C 2 Sg; f be identity and < be the strict part of =I I (or [, al gusto). Verification: due to
the constructions of S and I it is immediate that for all i; j 2 I : PL i f ! j f or
PL j f ! i f , so the demands of the imperatives are chained and I is uniquely
prioritized (Theorem 7). All imperatives are equally ranked, so I + A IfA.
Coincidence: We must prove that for all B; D 2 rLdPL , hI; f; <i models
O(B/D) iff OB=D 2 D. Right-to-left: Assume OB=D 2 D. If D =^, then
by definition I + :D If:D [, so (td-6) makes O(B/D) trivially true.
Otherwise, by (O7) there is some CD 2 S. By its definition, D ^ OCD is consistent, so D ^ CD ! OCD is also consistent and so CD ! OCD must be in
(the) C 2 If:D. Since PL D ! CD by (O8) and PL OCD ^ D ! OD by
(O10) we have PL CD ! OCD ^ D ! OD , so C [ fDg PL OD and due to
(O1), C [ fDg PL B and so (td-6) makes O(B/D) true. Left-to-right: Assume
that for C 2 I + :D If:D we have C [ fDg PL B. If D=^, then
OB=D 2 D follows trivially from (A0), (A2) and (A3). Otherwise, by (O7)
there is a CD 2 S: D ^ OCD is by definition consistent, so D ^ CD ! OCD is
consistent and CD ! OCD must be in (the) set C. CD=Ci for some i in the
construction of S. For all j < i; PL D ! Cj due to (O8) and (O2), and
PL D ! :OC j by definition of CD, so Cj ! OC j 62 C. For all
j > i; PL OC D ! :Cj by (O3), so with (O8) we obtain fCD ! OCD g[
fDg PL Cj ! OC j for any Cj ! OC j 2 C. So if C [ fDg PL B, then
fCD ! OCD g [ fDg PL B. Hence PL OCD ^ D ! B by (O8), PL OD ! B
by (O10), so OB=D 2 D by (O1).
246
32
JORG HANSEN
1
in Definition 2. S?
#i 6 S
f
could not be true, so it must be that S?
#i PL p1 ^ . . . ^ pn . But then
p ^...^p
p ^...^p
n
n
S 1
f PL i f ! p1 ^ . . . ^ pn , which includes S 1
f PL i f ! pn .
f
Hence Pi ^ :pn =:p1 ^ . . . ^ pn cannot be true. But PA ^ :pn =
:p1 ^ . . . ^ pn must be true for all A such that fAg0PL pn and if meets the
condition. So D is not satisfiable.
References
Alchourron, C. E. (1986). Conditionality and the Representation of Legal Norms, in Martino,
A. A. and Socci Natali, F. (eds.), Automated Analysis of Legal Texts: Edited Versions of
selected papers from the Second International Conference on Logic, Informatics, Law,
Florence, Italy, September 1985. North Holland, Amsterdam, 173186.
Alchourron, C. E. and Bulygin, E. The Expressive Conception of Norms, in Hilpinen, R.
(1981). New Studies in Deontic Logic. Reidel: Dordrecht, 95124.
Alchourron, C. E. and Makinson, D. Hierarchies of Regulations and Their Logic, in Hilpinen,
R. (1981). New Studies in Deontic Logic. Reidel: Dordrecht 95124, 125148.
247
33
Alchourron, C. E. and Makinson, D. (1985). On the Logic of Theory Change: Safe Contraction. Studia Logica 44: 405422.
Aqvist, L. (1986). Some Results on Dyadic Deontic Logic and the Logic of Preference.
Synthese 66: 95110.
Brewka, G. (1989). Preferred Subtheories: An Extended Logical Framework for Default
Reasoninig, in Sridharan, N. S. (ed.), Proceedings of the Eleventh International Joint
Conference on Articial Intelligence IJCA189, Detroit, Michigan, USA, August 2025,
1989, San Mateo, Calif.: Kaufmann, 10431048.
Brewka, G. (1991). Belief Revision in a Framework for Default Reasoning, in Fuhrmann, A.
and Morreau, M. (eds.), The Logic of Theory Change, Workshop, Konstanz, Germany,
October 1315, 1989. Springer, Berlin, 206222.
Brewka, G. and Eiter, T. (1999). Preferred Answer Sets for Extended Logic Programs. Articial Intelligence 109: 297356.
Brink, D. O. (1994). Moral Conict and Its Structure. Philosophical Review 103: 215247.
Fehige, C. (1994). The Limit Assumption in Deontic (and Prohairetic) Logic, in Meggle, G.
and Wessels, U. (eds.), Analyomen 1: Proceedings of the 1st Conference Perspectives in
Analytical Philosophy, Saarbrucken 1991. de Gruyter, Berlin, 4256.
Gardenfors, P. (1984). Epistemic Importance and Minimal Changes of Belief. Australasian
Journal of Philosophy 62: 136157.
Goble, L. (2005). A Logic for Deontic Dilemmas, Journal of Applied Logic: 961983.
Hage, J. C. (1991). Monological Reason Based Reasoning, in Breuker, J. A., De Mulder, R. V.,
and Hage, J. C., Legal Knowledge Based Systems: Model-based Legal Reasoning (Proceedings of the Fourth International Conference on Legal Knowledge Based Systems JURIX
1991, Lelystad, December 1991), 7791.
Hage, J. C. (1996). A Theory of Legal Reasoning and a Logic to Match. Articial Intelligence
and Law 4: 199273.
Hansen, J. (2001). Sets, Sentences, and Some Logics about Imperatives. Fundamenta Informaticae 48: 205226.
Hansen, J. (2004) Problems and Results for Logics about Imperatives, Journal of Applied
logic, 3661.
Hansen, J. (2005) Conicting Imperatives and Dyadic Deontic Logic, Journal of Applied
Logic, 484511.
Hansson, B. (1969). An Analysis of Some Deontic Logics. Nous 3: 373398, reprinted in
Hilpinen, R. (1971). Deontic Logic: Introductory and Systematic Readings. Reidel:
Dordrecht, 121147.
Hare, R. M. (1981). Moral Thinking. Clarendon Press: Oxford.
Horty, J. F. (1997). Nonmonotonic Foundations for Deontic Logic, in Nute, D. (eds.),
Defeasible Deontic Logic. Kluwer, Dordrecht, 1744.
Horty, J. F. (2003). Reasoning with Moral Conicts. Nous 37: 557605.
Iwin, A. A. (1972). Grundprobleme der deontischen Logik: in Wessel, H. (eds.), Quantoren
ModalitatenParadoxien. VEB Deutscher Verlag der Wissenschaften, Berlin, 402522.
Kanger, S. (1957). New Foundations for Ethical Theory: Part 1, duplic., 42 p., reprinted in
Hilpinen, R. (1971). Deontic Logic: Introductory and Systematic Readings. Reidel:
Dordrecht, 3658.
Kraus, S., Lehmann, D. and Magidor, M. (1990). Nonmonotonic Reasoning, Preferential
Models and Cumulative Logics. Articial Intelligence 44: 167207.
Lewis, D. (1981). Ordering Semantics and Premise Semantics for Counterfactuals. Journal of
Philosophical Logic 10: 217234.
Makinson, D. and van der Torre, L. (2000). Input/Output Logics. Journal of Philosophical
Logic 29: 383408.
248
JORG HANSEN
Makinson, D. and van der Torre, L. (2001). Constraints for Input/Output Logics. Journal of
Philosophical Logic 30: 155185.
Marcus, R. B. (1980). Moral Dilemmas and Consistency. Journal of Philosophy 77: 121136.
McNamara, P. (1995). The Connement Problem: How to Terminate Your Mom With Her
Trust. Analysis 55: 310313.
Nebel, B. (1991). Belief Revision and Default Reasoning: Syntax-Based Approaches, in Allen,
J. A., Fikes, R. and Sandewall, E. (eds.), Principles of Knowledge Representation and
Reasoning: Proceedings of the Second International Conference, KR 91, Cambridge, MA,
April 1991. Morgan Kaufmann, San Mateo, 417428.
Nebel, B. (1992). Syntax-Based Approaches to Belief Revision, in Gardenfors, P. (eds.), Belief
Revision. University Press, Cambridge, 5288.
Prakken, H. (1997). Logical Tools for Modelling Legal Argument. Kluwer: Dordrecht.
Prakken, H. and Sartor, G. (1997). Argument-based Logic Programming with Defeasible
Priorities. Journal of Applied Non-classical Logics 7: 2575.
Rescher, N. (1964). Hypothetical Reasoning. North-Holland: Amsterdam.
Rintanen, J. (1994). Prioritized Autoepistemic Logic, in MacNish, C., Pearce, D. and Pereira,
L. M. (eds.), Logics in Articial Intelligence, European Workshop, JELIA 94, York, September 1994, Proceedings. Springer, Berlin, 232246.
Ross, W. D. (1930). The Right and the Good. Clarendon Press: Oxford.
Rott, H. (1993). Belief Contraction in the Context of the General Theory of Rational Choice.
Journal of Symbolic Logic 58: 14261450.
Ryan, M. (1992). Representing Defaults as Sentences with Reduced Priority, in Nebel, B.,
Rich, C. and Swartout, W. (eds.), Principles of Knowledge Representation and Reasoning:
Proceedings of the Third International Conference, KR 92, Cambridge, MA, October 1992.
Morgan Kaufmann, San Mateo, 649660.
Sakama, C. and Inoue, K. (1996). Representing Priorities in Logic Programs, in Maher, M.
(eds.), Joint International Conference and Syposium on Logic Programming JICSLP 1996,
Bonn, September 1996. MIT Press, Cambridge, 8296.
Sartor, G. (1991). Inconsistency and Legal Reasoning, in Breuker, J. A., De Mulder, R. V.,
and Hage, J. C., Legal Knowledge Based Systems: Model-based Legal Reasoning
(Proceedings of the Fourth International Conference on Legal Knowledge Based Systems
JURIX 1991, Lelystad, December 1991), 92112.
Sartor, G. (2005) Legal Reasoning. A Cognitive Approach to the Law, vol. 5 of A Treatise of
Legal Philosophy and General Jurisprudence, Springer: Dordrecht.
Searle, J. (1980). Prima-facie Obligations, in Straaten, Z. v. (ed.), Philosophical Subjects:
Essays presented to P. F. Strawson. Clarendon Press, Oxford, 238259.
Spohn, W. (1975). An Analysis of Hanssons Dyadic Deontic Logic. Journal of Philosophical
Logic 4: 237252.
von Wright, G. H. (1968). An Essay in Deontic Logic and the General Theory of Action. North
Holland: Amsterdam.
Ziemba, Z. (1971). Deontic Syllogistics. Studia Logica 28: 139159.
249
Abstract The sentences of deontic logic may be understood as describing what an agent
ought to do when faced with a given set of norms. If these norms come into conflict, the best
the agent can be expected to do is to follow a maximal subset of the norms. Intuitively, a
priority ordering of the norms can be helpful in determining the relevant sets and resolve conflicts, but a formal resolution mechanism has been difficult to provide. In particular, reasoning
about prioritized conditional imperatives is overshadowed by problems such as the order
puzzle that are not satisfactorily resolved by existing approaches. The paper provides a new
proposal as to how these problems may be overcome.
Keywords
J. Hansen (B)
Institut fr Philosophie, University of Leipzig, Beethovenstrae 15, 04107 Leipzig, Germany
e-mail: jhansen@uni-leipzig.de
123
250
Auton Agent Multi-Agent Syst
2 Formal preliminaries
To formally discuss problems such as the one presented above, I shall use a simple framework: let I be a set of objects, they are meant to be (conditional) imperatives. Two functions
g and f associate with each imperative an antecedent and a consequentthese are sentences
from the language of a basic logic that here will be the language L P L of propositional
logic.2 g(i) may be thought of as describing the grounds, or circumstances in which the
consequent of i is to hold, and f (i) as associating the sentence that describes what must be
1 Cf. Rintanen [34] p. 234, who in turn credits Gerhard Brewka with its invention.
2 PL is based on a language L
P L , defined from a set of proposition letters Prop = { p1 , p2 , ...}, Boolean
connectives , , , , and brackets (, ) as usual. The truth of a L P L -sentence (I use upper case letters
A, B, C, ...) is defined recursively using valuations v : Prop {1, 0} (I write v | A), starting with v | p
iff v( p) = 1 and continuing as usual. If A L P L is true for all valuations it is called a tautology. PL is the
set of all tautologies, and used to define provability, consistency and derivability (I write P L A) as usual.
is an arbitrary tautology, and is .
123
251
the case if the imperative i is satisfied, its deontic focus or demand.3 In accordance with
tradition (cf. Hofstadter and McKinsey [19]), I write A !B for an i I with g(i) = A and
f (i) = B, and !A means an unconditional imperative !A. Note that A !B is just the
name for a conditional imperative that demands B to be made true in a situation where A is
trueit is not an object that is assigned truth values. A useful construction is the materialization m(i) of an imperative i, which is the material implication g(i) f (i) that may be
thought of as corresponding to a conditional imperative. For any i I and I , instead of
f (i), g(i), m(i), f ( ), g( ) and m( ), I may use the superscripted i f , i g, i m, f , g and
m for better readability.
Let I be a tuple I, f, g , let W L P L be a set of sentences, representing real world
facts, and I be a subset of the imperatives: then we define
Triggered I (W, ) = {i
|W P L g(i)}.
Violated I (W, ) = {i
|W P L i g i f },
|W P L i g i f },
An imperative in Satisfied I (W, ) [Violated I (W, )] is called satisfied [violated] given the
facts W . It is of course possible that an imperative is neither satisfied nor violated given the
facts W . If an imperative is triggered, but not violated, we call the imperative satisfiable:
SatisfiableI (W, ) = {i Triggered I (W, )|W
PL
i f }.
Moreover, the following definition will play a major rle in what follows:
ObeyableI (W, ) = {
| m W
PL
}.
So a subset of is obeyable given W iff it is not the case that for some {i 1 , . . . , i n }
g
f
g
f
we have W P L (i 1 i 1 ) . . . (i n i n ): otherwise we know that whatever we do, i.e.
given any maxiconsistent subset V of L P L that extends W V , at least one imperative in
is violated.5 We speak of a conflict of imperatives when the triggered imperatives cannot all
be satisfied given the facts W , i.e. when Triggered I (W, ) f W P L . More generally
speaking I will also call imperatives conflicting if they are not obeyable in the given situation.
As prioritized conditional imperatives are our concern here, we let all imperatives in I be
ordered by some priority relation < I I . The relation < is assumed to be a strict partial
order on I , i.e. < is irreflexive and transitive, and additionally we assume < to be wellfounded, i.e. infinite descending chains are excluded. For any i 1 , i 2 I, i 1 < i 2 means that
i 1 takes priority over i 2 (ranks higher than i 2 , is more important than i 2 , etc.). A tuple I, f, g
will be called a conditional imperative structure, and I, f, g, < a prioritized conditional
imperative structure. If all imperatives in I are unconditional, we may drop any reference to
the relation g in the tuples and call these basic imperative structures and prioritized imperative structures respectively.
3 In analogy to Reiters default logic one might add a third function e that describes exceptional circumstances
in which the imperative is not to be applied. I will not address this additional complexity here.
4 Cf. Rescher [33], Sosa [38], van Fraassen [9]. Also cf. Greenspan [11]: Oughts do not arise, it seems, until
it is too late to keep their conditions from being fulfilled.
5 Terms differ here, e.g. Downing [8] uses the term compliable instead of obeyable.
123
252
Auton Agent Multi-Agent Syst
3 Deontic concepts
Given a set of imperatives, one may truly or falsely state that their addressee must, or must
not, perform some act or achieve some state of affairs according to what the addressee was
ordered to do. Regarding the drinking and driving example, I think it is true that the agent
ought to do the driving, as this is what the second-ranking imperative, uttered by the agents
best friend, requires her to do, but that it would be false to say that the agent ought to drink
and drive. Statements that something ought to be done or achieved are called normative or
deontic statements, and the ultimate goal of deontic logic is to find a logical semantics that
models the situation and defines the deontic concepts in such a way that the formal results
coincide with our natural inclinations in the matter.
3.1 Deontic operators for unconditional imperatives
For unconditional imperatives, such definitions are straightforward. Given a basic imperative
structure I = I, f , a monadic deontic O-operator, that formalizes it ought to be that A (is
realized) by O A, is defined by
(td-m1) I | O A if and only if (iff) I f P L A.
So obligation is defined in terms of what the satisfaction of all imperatives logically implies.6
With the usual truth definitions for Boolean operators, it can easily be seen that such a definition produces a normal modal operator, i.e. one that is defined by the following axiom
schemes plus modus ponens:
(Ext) If P L A B, then O A O B is a theorem.
(M) O(A B) (O A O B)
(C) (O A O B) O(A B)
(N) O
: f P L A
Quite similarly, a dyadic deontic operator O(A/C), meaning that A ought to be true given
that C is true, can be defined with respect to the maximal subsets of imperatives that do not
conflict in these circumstances:
(td-d1) I | O(A/C) iff I
C : f P L A
6 Such a definition of obligation was proposed e.g. by Alchourrn and Bulygin [1].
123
253
C : f {C} P L A
So A is obligatory given C is (invariably) true iff all maximal subsets of the imperatives
demands (the imperatives associated descriptive sentences) that are consistent with the circumstances C, plus C, derive A. With the usual truth conditions for Boolean operators, a
semantics that employs (td-d1+ ) has a sound and (weakly) complete axiom system PD that
extends the system P of Kraus et al. [22], defined by these axiom schemes
(DExt)
(DM)
(DC)
(DN)
O(/C)
(ExtC)
(CCMon)
(CExt)
(Or)
PL
sible situation Cthis is the same for any dyadic deontic logic since Hansson [15] and Lewis [23] (for the
motivation cf. [26] pp. 158159). Similarly, imperatives like C !C are treated as violated as soon as they
are triggered by the facts. There exist meaningful natural-language imperatives like close the window if it
is open, but I think that in these the proposition in the antecedent is different from the negation of the one
corresponding to the consequent, in that the second refers to a different point of time (see to it that the window
is closed some time in the near future if it is open now), so they should not be represented by C !C.
8 For alternative solutions to the problem of conflicts cf. Goble [10] and Hansen [12], [13].
9 For proofs, and an additional credulous ought that defines O(A/C) true if the truth of A is required to
satisfy all imperatives in some C-remainder, cf. Hansen [13].
123
254
Auton Agent Multi-Agent Syst
be systematic relations governing this moral dynamics, but I can only profess ignorance of
them.
Representing a conditional imperative as an unconditional imperative that demands a material conditional to be made true yields undesired results. Most notorious is the problem of
contraposition: consider a set I with the only imperative !( p1 p2 ), meaning e.g. if it rains,
take an umbrella. (td-d1) makes true O( p2 / p1 ), but also O( p1 / p2 ), so if you cannot
take your umbrella (your wife took it) you must see to it that it does not rain, which is hardly
what the speaker meant you to do. One may think that such problems arise from the fact that
antecedents of conditional imperatives often describe states of the affairs that the agent is not
supposed to, and often cannot, control. But consider the set {!( p1 p2 ), !( p1 p3 )},
it yields O( p2 / p3 ) with (td-d1). Here, p2 is what the consequent of some imperative
demands, so it supposedly describes something the agent can control. Now let the imperatives be interpreted as ordering me to wear a rain coat if it rains, and my best suit if it does
not: it is clear nonsense that I am obliged to wear a raincoat given that I cant wear my best
suit (e.g. it is in the laundry). Such problems are the reason why we use special models for
conditional imperatives that separate antecedents and consequents (conditional imperative
structures), and write p1 ! p2 instead of !( p1 p2 ). But this only delegates the problem
from the level of representation to that of semantics, where now new truth definitions must
be found.
Let I = I, f, g be a conditional imperative structure, and let us ignore for the moment
the further complication of possible conflicts between imperatives. Then the following seems
a natural way to define what ought to be the case in circumstances where C is assumed to be
true:
(td-cd1) I | O(A/C) iff [Triggered I ({C}, I )] f P L A
So dyadic obligation is defined in terms what is necessary to satisfy all imperatives that are
triggered in the assumed circumstances. E.g. if I = { p1 ! p2 }, with its only imperative
interpreted as if you have a cold, stay in bed, then O( p2 / p1 ) truly states that I must stay
in bed given that I have a cold.
Like in the unconditional case, it seems important to be able to use circumstantial reasoning, i.e. employ the information about the situation not only to determine if an imperative
is triggered, but also for reasoning with its consequent. E.g. if the set of imperatives is
{ p1 !( p2 p3 )}, with its imperative interpreted as expressing if you have a cold, either
stay in bed or wear a scarf, one would like to obtain O( p3 / p1 p2 ), expressing that given
that I have a cold and dont stay in bed, I must wear a scarf. So (td-cd1) may be changed
into
(td-cd1+ ) I | O(A/C) iff [Triggered I ({C}, I )] f {C} P L A.
Though the step from (td-cd1) to (td-cd1+ ) seems quite reasonable, such definitions have
also been criticized for defining the assumed circumstances as obligatory. In the above example, (td-cd1+ ) also makes true O( p1 / p1 p2 ), so given that you have a cold it is true that
you ought to have it. The criticism loses much of its edge in the present setting, where one
can point to the distinction between imperatives (there is no imperative that demands p1 )
and ought sentences that describe what must be true given the facts and the satisfaction of all
triggered imperatives: then the truth of O( p1 / p1 ) seems no more paradoxical than the truth
of O that is accepted in most systems of deontic logic.
123
255
the input C (Makinson and van der Torre write A out (G, {C})). Though these authors liken their generating
set G to a body of conditional norms, it should be noted that they do not themselves introduce dyadic deontic
operators.
11 Makinson and van der Torres [25] call the resulting operator basic output, of which a syntactical version
123
256
Auton Agent Multi-Agent Syst
where Triggered I (W, ) means the smallest subset of I such that for all i , if
[Triggered I (W, )] f W P L g(i) then i Triggered I (W, ). Moreover, the two modifications of reasoning by cases and reusable output can be combined to produce the
following definition and its throughput variant:
(td-cd4) I | O(A/C) iff V L P L C : [Triggered I (V, I )] f P L A
12 (td-cd2 ) does not fare much better: though it does not include contraposition, it again makes O( p / p )
2
3
true for I = { p1! p2 , p1! p3 }, which is counterintuitive.
13 With respect to their out -operation that corresponds to (td-cd4), Makinson and van der Torre [25] speak
4
of a ghostly contraposition.
123
257
I|
A,
A,
A,
A,
A,
A,
A,
A.
So e.g. (td pcd1) defines A as obligatory if the truth of A is required to satisfy the triggered imperatives in any preferred subset. Of course, the crucial and as yet missing element
is the decision procedure that determines the set PI ({C}, I ) of preferred subsets. The next
section discusses several proposals to define such subsets; a new proposal is presented in the
section that follows it.
123
258
Auton Agent Multi-Agent Syst
defaults. For any priority relation <, the idea is to consider all the full prioritizations of
< (strict well orders that preserve <), and then work ones way from the top of the strict
order downwards by adding the -next-higher imperative to the thus constructed preferred
subtheory if its demand is consistent with the given facts and the demands of the imperatives
that were added before. For the present setting, the definition can be given as follows:
Definition 1 (Brewkas preferred subtheories) Let I = I, f, g, < be a prioritized conditional imperative structure, be a subset of I , and W L P L be a set of PL-sentences.
Then PIB (W, ) iff (i) W P L , and (ii) is obtained from a full prioritization by
defining
f
i =
for any i
j {i} if W
otherwise,
j i j
j i
, and letting =
j i
j {i}
PL
, and
i .
Clause (i) ensures that for an inconsistent set of assumed facts, no set is preferred. Somewhat roundabout, owed to the possibility of infinite ascending subchains, clause (ii) then
recursively defines a set PIB (W, ) for each full prioritization : take the -first i (the
exclusion of infinite descending subchains guarantees that it exists) and if W {i f } P L
then let i = {i}; otherwise i is left empty.15 Similarly, any -later i is tested for possible
addition to the set j i j of elements that were added in the step for a j that occurs
-prior to i. is then the union of all these sets.
To see how this definition works, consider the set I = {!( p1 p2 ), ! p1 , ! p2 }, with the
ranking !( p1 p2 ) < ! p2 and ! p1 < ! p2 . For an interpretation, let !( p1 p2 )) be your
mothers request that you buy cucumbers or spinach for dinner, ! p1 be your fathers wish
that no cucumbers are bought, and ! p2 your sisters desire that you dont buy any spinach.
We have two full prioritizations !( p1 p2 ) < ! p1 < ! p2 and ! p1 < !( p1 p2 ) < ! p2
let these be termed 1 and 2 , respectively. The construction for 1 adds the imperative
!( p1 p2 ) in the first step and, since no conflict with the situation arises, ! p1 in the second
step. In the third and last step, nothing is added since ! p2 conflicts with the demands of
the already added imperatives. For 2 the only difference is that the first two imperatives
are added in inverse order. Thus PIB (W, I ) = {{!( p1 p2 ), ! p1 }}. Using (td pcd1) we
obtain O( p1 p2 /), which means that you have to buy spinach and not cucumbers, thus
fulfilling your parents requests but not your sisters, which seems reasonable.
As I showed in [14], Brewkas method is extremely successful for dealing with unconditional imperatives. It is provably equivalent for such imperatives to methods proposed
by Ryan [36] and Sakama and Inoue [37], and it avoids problems of other approaches by
Alchourrn and Makinson [2], Prakken [30] and Prakken and Sartor [31]. Moreover, an
equally intuitive maximization method proposed by Nebel [28], [29], that adds first a maximal set of the highest-ranking imperatives, then a maximal set of second-ranking imperatives,
etc., but for its construction requires the ordering to be based on a complete preorder, can
be shown to be embedded in Brewkas approach for such orderings. So my aim will be to
retain Brewkas method for the unconditional casein fact, all proposals that follow meet
this criterion. However, when it is applied without change to conditional imperatives, the
15 As usual, the union of an empty set of sets is taken to be the empty set.
123
259
algorithm may lead to incorrect results. E.g. consider a set I with two equally ranking imperatives { p1 ! p2 , p1 ! p2 }, meaning e.g. if you go out, wear your boots and if you
dont go out, dont wear your boots: since the consequents contradict each other, an unmodified application of Brewkas method produces PIB ({ p1 }, I ) = {{ p1! p2 }, { p1! p2 }},
which fails to make true O( p2 / p1 ) by any truth definition of Sect. 3.4: the right set contains
no imperatives that are triggered by p1 . So we cannot derive that you ought to wear your
boots, given that you are going out. But intuitively there is no conflict, since the obligations
arise in mutually exclusive circumstances only.
4.2 A nave approach
A straightforward way to adopt Brewkas method to the case of conditional imperatives is to
use not all imperatives for the construction, but only those that are triggered by the facts W ,
i.e. to use Triggered I (W, ) instead of :
Definition 2 (The nave approach) Let I = I, f, g, < be a prioritized conditional imperative structure,
be a subset of I , and W L P L be a set of PL-sentences. Then
PIn (W, ) iff PIB (W, Triggered I (W, )).
The change resolves our earlier problems with Brewkas method: consider again the set of
imperatives { p1! p2 , p1! p2 }, where the imperatives were interpreted as ordering me
to wear my boots when I go out, and not wear my boots when I dont. The new definition produces PIn ({ p1 }, I ) = {{ p1 ! p2 }}, its only preferred subset containing just the
one imperative that is triggered given the facts { p1 }. By all truth definitions of Sect. 3.4,
O( p2 / p1 ) is now true, so given that you go out, you ought to wear your boots, which is as
it should be.
The nave approach is similar to Hortys proposal in [20] in that conflicts are only removed
between imperatives that are triggered (though the exact mechanism differs from Hortys).
When I nevertheless call it nave, this is because there are conceivable counterexamples to
this method. Consider the set of prioritized imperatives ! p1 < p1! p2 < ! p2 , and for an
interpretation suppose that your job requires you to go outside p1 , that your mother, who is
concerned for your health, told you to wear a scarf p2 if you go outside, and that your friends
dont want you to wear a scarf, whether you go outside or not. In the default situation
only the first imperative and the third are triggered, i.e. Triggered I ({}, I ) = {! p1 , ! p2 }.
Since their demands are consistent with each other, we obtain PIn ({}, I ) = {{! p1 , ! p2 }},
for which all truth definitions of Sect. 3.4 make O( p1 p2 /) true. So you ought to go
out and not wear a scarf, thus satisfying the first and the third imperative, but violating the
second-ranking imperative. But arguably, if an imperative is to be violated, it should not be
the second-ranking p1! p2 , but the lowest ranking ! p2 instead.
4.3 The stepwise approach
To avoid the difficulties of the nave approach, it seems we must not just take into account
the imperatives that are triggered, but also those that become triggered when higher ranking
imperatives are satisfied. To this effect, the following modification may seem
reasonable:
Definition 3 (The stepwise approach) Let I = I, f, g, < be a prioritized conditional
imperative structure, be a subset of I , and W L P L be a set of PL -sentences. Then
123
260
Auton Agent Multi-Agent Syst
PL
i =
for any i
j {i} if i SatisfiableI (W
otherwise,
j i j
j i
, and letting =
j i
, ), and
i .
So at each step one considers what happens if the imperatives that were included so far are
satisfied, and adds the current imperative only if it is satisfiable given W and the satisfaction
of these previous imperatives. Note that satisfiability of an imperative, like its satisfaction and
violation, presupposes that the imperative is triggered. The new definition not only includes,
at each step, those imperatives that are triggered and can be satisfied given the facts and the
supposed satisfaction of the previously added imperatives: it also includes those that become
triggered when a previously added imperative is satisfied.
This modification avoids the previous difficulty: consider again the set of prioritized
imperatives ! p1 < p1 ! p2 <! p2 . There is just one full prioritization, which for W = {}
yields in the first step the set {! p1 }, and in the second step {! p1 , p1! p2 }, since p1! p2 is
triggered when the previously added imperative ! p1 is assumed to be satisfied. In the third
step, nothing is added: though the imperative ! p2 is triggered, it cannot be satisfied together
with the previously added imperatives. So we obtain PIs ({}, I ) = {{! p1 , p1 ! p2 }}, and
hence O( p1 /), but not O( p1 p2 /), is defined true by all truth definitions of Sect. 3.4.
Operators that accept deontic detachment (td- pcd3(+) , 4(+) ) make true O( p1 p2 /), so
you must go out and wear a scarf, which is as it should be.
However, a small change in the ordering shows that this definition does not suffice: let
the imperatives now be ranked p1 ! p2 < ! p1 < ! p2 . (For the interpretation, assume
that the conditional imperative to wear a scarf when leaving the house was uttered by some
high-ranking authority, e.g. a doctor.) Then again PIs ({}, I ) = {{! p1 , ! p2 }}: in the first
step, nothing is added since p1 ! p2 is neither triggered by the facts nor by the assumed
satisfaction of previously added imperatives (there are none). In the next two steps, ! p1 and
! p2 are added, as each is consistent with the facts and the satisfaction of the previously
added imperatives. So again all truth definitions of Sect. 3.4 make true O( p1 p2 /), i.e.
you ought to go out and not wear a scarf, satisfying the second and third ranking imperatives
at the expense of the highest ranking one. But surely, if one must violate an imperative, it
should be one of the lower-ranking ones instead.
4.4 The reconsidering approach
The merits of the stepwise approach were that it did not only consider the imperatives that are
triggered, but also those that become triggered when already added imperatives are satisfied.
Such considerations applied to those imperatives that follow in the procedure. Yet the satisfaction of already added imperatives might also trigger higher-ranking imperatives, which by
this method are not considered again. So it seems necessary, at each step, to reconsider also
the higher-ranking imperatives. An algorithm that does that was first introduced for default
theory by Marek and Truszczynski [27] p. 72, and later employed by Brewka in [4]; it can
be reformulated for the present setting as follows:
Definition 4 (The reconsidering approach) Let I = I, f, g, < be a prioritized conditional
imperative structure, be a subset of I , and W L P L be a set of PL-sentences. Then
123
261
PL
i =
for i
j i
j min SatisfiableI W
, and letting =
j i
j i
i .16
The definition reconsiders at each step the whole ordering, and adds the -first17 imperative (due to the definition of there is just one) that has not been added previously and is
satisfiable given both the facts W and the consequents of the previously added imperatives.
To see how the definition works, consider again the example which the stepwise approach
failed, i.e. the set of prioritized imperatives p1! p2 < ! p1 < ! p2 . We are interested in the
preferred sets for the default circumstances , i.e. the sets in PIr ({}, I ). I is already fully
prioritized, so there is just one such set. Applying the algorithm, we find the minimal (highest
ranking) element in SatisfiableI ({}, I ) is ! p1 , so this element is added in the first step. In
the second step, we look for the minimal element in SatisfiableI ({} {! p1 } f , I ), other
than the previously added ! p1 . It is p1! p2 , since the assumed satisfaction of all previously
added imperatives triggers it, and its consequent can be true together with {} { p1 }. So
p1 ! p2 is added in this step. In the remaining third step, nothing is added: ! p2 is not in
SatisfiableI ({}{! p1 , p1! p2 } f , I ), and all other elements in this set have been previously
added. So PIr ({}, I ) = {{! p1 , p1! p2 }}. Now all truth definitions of Sect. 3.4 make true
O( p1 /), but not O( p1 ! p2 /), and operators that accept deontic detachment make
true O( p1 p2 /). So, in the given interpretation, you must go out (as your job requires)
and wear a scarf (as the doctor ordered you to do in case you go out), which is as it should
be.
However, again problems remain. Let the imperatives now be prioritized in the order
p1 ! p2 < ! p2 < ! p1 . Let p1 ! p2 stand for the doctors order to wear a scarf when
going outside, let ! p2 stand for your friends expectation that you dont wear a scarf, and let
! p1 represent your sisters wish that you leave the house. Construct the set in PIr ({}, I )
since I remains fully prioritized, there is again just one such set. The minimal element
in SatisfiableI ({}, I ) is ! p2 , and so is added in the first step. The minimal element in
SatisfiableI ({} {! p2 } f , I ), other than ! p2 , is ! p1 which therefore gets added in the
second step. Nothing is added in the remaining step: ! p2 and ! p1 have already been added,
and p1 ! p2 is not in SatisfiableI ({} {! p2 , ! p1 } f , I ): though it is triggered by the
assumed satisfaction of ! p1 , its consequent is contradicted by the assumed satisfaction of
! p2 . So PIr ({}, I ) = {{! p1 , ! p2 }}. Hence all truth definitions of Sect. 3.4 again make
true O( p1 p2 /), so you ought to go out without a scarf, again satisfying the second and
third ranking imperatives at the expense of the first, which seems the wrong solution.
4.5 A fixpoint approach
To eliminate cases in which the reconsidering approach still adds imperatives that can
only be satisfied at the expense of violating a higher-ranking one, a fixpoint approach was
first proposed for default reasoning by Brewka and Eiter [5]. It tests each set that may be
considered as preferred to see if it really includes all the elements that should be included:
16 Note that in , i is used just as an indexit does not mean that i is considered for addition at this step,
i
123
262
Auton Agent Multi-Agent Syst
imperatives that are triggered given the facts and the assumed satisfaction of all imperatives
in the set, and would be added by Brewkas [3] original method that adds the higher ranking
imperatives first. The procedure translates as follows:
Definition 5 (The fixpoint approach)
Let I = I, f, g, < be a prioritized conditional
imperative structure, be a subset of I , and W L P L be a set of PL-sentences. Then
f
Another approach to conditional imperatives by Makinson in [24] has trouble resolving the same example:
for the default circumstances it produces the set {!( p1 p2 ), ! p3 }. p1 ! p2 is not considered, since its
only label (roughly: a conjunction of the circumstances, the imperatives antecedents that would trigger it,
and its consequent) is inconsistent (it is ( p1 p2 ) p2 ). But why should the agent not be free to obey
p1! p2 , and not violate it by satisfying !( p1 p2 )?
123
263
4.6 Discussion
For a discussion of our results so far, let us return to the drinking and driving example from
the introduction. Let the three imperatives:
(1) Your mother says: if you drink anything, then dont drive.
(2) Your best friend says: if you go to the party, then you do the driving.
(3) Some acquaintance says: if you go to the party, then have a drink with me.
be represented by the set of prioritized imperatives p1 p2 < p3 p2 < p3 p1 . Let the
set of facts be { p3 }, i.e. you go to the party. Brewkas original method is not tailored to
be directly employed on conditional imperatives, as it ignores the antecedents altogether.
The next three approaches, the nave, the stepwise and the reconsidering ones, produce
PIn ({ p3 }, I ) = PIs ({ p3 }, I ) = PIr ({ p3 }, I ) = {{ p3 ! p2 , p3 ! p1 }}, which by all truth
definitions of Sect. 3.4 makes true O( p1 p2 / p3 ), so you ought to drink and drive. The
fixpoint approach produces PIf ({ p3 }, I ) = {{ p1! p2 , p3! p1 }}, so all truth definitions
make true O( p1 / p3 ), which means you ought to drink. Truth definitions with deontic detachment additionally make true O( p1 p2 / p3 ), so you ought to drink and not drive. But
being obliged to drink runs counter to our intuitions for the drinking and driving example.
So we have to look for a different solution.
Before we do that, I will, however, question again our intuition in this matter. Horty [21]
has recently used a structurally identical example to argue for just the opposite, that the solution by the fixpoint approach is correct. His example is that of three commands, uttered by
a colonel, a major and a captain to a soldier, Corporal OReilly. The Colonel, who does not
like it too warm in the cabin, orders OReilly to open the window whenever the heat is turned
on. The Major, who is a conservationist, wants OReilly to keep the window closed during
the winter. And the Captain, who does not like it to be cold, orders OReilly to turn the heat
on during the winter. OReilly is trying to figure out what to do. The intended representation
is again the prioritized conditional imperative structure employed above for the drinking
and driving example, where p1 now means that the heat is turned on, p2 means that the
window is closed, and p3 means that it is winter. We saw that the fixpoint approach yields
the preferred subset { p1 ! p2 , p3 ! p1 }, making true O( p1 / p3 ) for all truth definitions,
and O( p1 p2 / p3 ) for truth definitions that accept deontic detachment, so OReilly must
turn on the heat and then open the window, and thus violate the Majors order. Horty argues
as follows in support of this choice:
OReillys job is to obey the orders he has been given exactly as they have been
issued. If he fails to obey an order issued by an officer without an acceptable excuse,
he will be court-martialed. And, let us suppose, there is only one acceptable excuse for
failing to obey such an order: that obeying the order would, in the situation, involve
disobeying an order issued by an officer of equal or higher rank. (...) So given the set
of commands that OReilly has been issued, can he, in fact, avoid court-martial? Yes
he can, by (...) obeying the orders issued by the Captain and the Colonel (...). OReilly
fails to obey the Majors order, but he has an excuse: obeying the Majors order would
involve disobeying an order issued by the Colonel.
Hortys principle seems quite acceptable: for each order issued to the agent, the agent may
ask herself if obeying the order would involve disobeying an order of a higher ranking officer
(then she is excused), and otherwise follow it. The result is a set of imperatives where each
imperative is either obeyed, or disobeyed but the disobedience excused. When I nevertheless
think the argument is not correct, it is because I think it confuses the status quo and the status
123
264
Auton Agent Multi-Agent Syst
quo posterior. Obeying the Majors order does not, in the initial situation, involve disobeying
the Colonels order. Only once OReilly follows the Captains order and turns on the heat, it
is true that he must obey the Colonel, open the window and thus violate the Majors order. But
this does not mean that he should follow the Captains order in the first placeas by doing
so he brings about a situation in which he is forced, by a higher ranking order, to violate a
command from another higher ranking officer. Quite to the contrary, I think that being forced
to violate a higher ranking order when obeying a lower ranking one is a case where following
the lower one involves such a violation, and so the only order the agent is excused from
obeying is the lowest ranking command.
Consider finally this variant: suppose that if I am attacked by a man, I must fight him (to
defend my life, my family etc.). Furthermore, suppose I have pacifist ideals which include
that I must not fight the man. Now you tell me to provoke him, which in the given situation
means that he will attack me. Let self-defense rank higher than my ideals, which in turn rank
higher than your request. Should I do as you request? By the reasoning advocated by Horty,
there is nothing wrong with it: I satisfy your request, defend myself as I must, and though I
violate my ideals, I can point out to myself that the requirement to fight back took priority.
But I think if I really do follow your advice, I would feel bad. I think this would not just be
some irrational regret for having to violate, as I must, my ideals, but true guilt for having
been tempted into doing something I should not have done, namely provoking the man: it
caused the situation that made me violate my ideals. So I think our intuitions in the drinking
and driving example and the other cases have been correct.
123
265
(cf. [6] def. 5.5). This test for conflicts may then be employed within a variant of Brewkas
original method.
Definition 6 (Preferred Subsets with Consistent Extensions) Let I = I, f, g, < be a prioritized conditional imperative structure,
be a subset of I , and W L P L be a set of
PL-sentences, and let E(W, ) be defined as described above. Then PIe (W, ) iff (i)
W P L , and (ii) is obtained from a full prioritization by defining
i =
for any i
j {i} if E(W,
otherwise,
j i j
j i
, and letting =
j i
j {i})
PL
, and
i .
To see how this definition works, consider the drinking and driving example, where the set
of prioritized imperatives is p1 ! p2 < p3 ! p2 < p3 ! p1 and the situation W = { p3 }.
There is only one full prioritization which is identical with <. In the first step, p1 ! p2 is
added, since E({ p3 }, { p1 ! p2 }) = { p3 } which is consistent. In the second step, p3 ! p2
is added, as E({ p3 }, { p1 ! p2 , p3 ! p2 }) = { p3 , p2 }, which is again consistent. In
the third step, p3 ! p1 is rejected, since E({ p3 }, { p1 ! p2 , p3 ! p2 , p3 ! p1 }) =
{ p3 , p2 , p1 , p2 }, which is inconsistent. So we have PIe ({ p3 }, I ) = {{ p1 ! p2 , p3
! p2 }}, making true O( p2 / p3 ) for all truth definitions of Sect. 3.4, so given that I go to the
party I must do the driving, which is as it should be.
However, there is a problem for the test using consistent extensions, as for some truth definitions it delivers sets of imperatives that are clearly conflicting, in the sense that they make
O(/C) true for a consistent fact C: Consider the set of facts W = { p1 p2 } and the set of
imperatives I = { p1 ! p3 , p2 ! p3 , ! p3 }. We have E(W, I ) = { p1 p2 , p3 }, which
is consistent, and so all imperatives in I are added to the preferred subset, regardless of their
ordering. But for any truth definition that allows for reasoning by cases (td-pcd2(+), 4(+) )
we then have both O( p3 / p1 p2 ) and O( p3 / p1 p2 ), and so also O(/ p1 p2 ).
This is simply because the construction of extensions does not take care of reasoning by
cases, i.e. it does not add p3 to the extension in case we both have p1 p2 in W and
{ p1 ! p3 , p2 ! p3 } I . Perhaps the definition of consistent extensions can be amended,
but to avoid delivering preferred sets of imperatives that make true O(/C) for some truth
definition and some consistent fact C suggests a different solution that is explored in the next
section.
5.2 Deontically tailored preferred subsets
In the unconditional case, the reason to move from definition (td-m1) to (td-m2) was that
when there are conflicts between imperatives, the former makes true the monadic deontic
formula O, i.e. the agent ought to do the logically impossible. This result was avoided by
considering only maximal sets of imperatives with demands that are collectively consistent,
i.e. sets that do not make O true. When faced with the question what dyadic deontic formula should not be true when conflicts are resolved for arbitrary situations C, the formula
O(C/C) appears to be the dyadic equivalent: a mechanism for conflict resolution should
not result in telling the agent to change the supposed, unalterable facts.19 So to define the set
PI ({C}, I ) required by the truth definitions (td- pcd1(+) 4+ ), we can modify Brewkas
19 This test is identical to the one used by Makinson and van der Torre [26] p. 158/159 to determine consistency of output (cf. also for arguments why O(C/C) should be used, i.e. for their setting, the output should
be consistent with the input, rather than the formula O(/C) and thus consistency of output simpliciter).
123
266
Auton Agent Multi-Agent Syst
original method in such a way that it tests, at each step, for each of the constructed subsets, if
the corresponding truth-definition (td-cd1(+) 4+ ) does not make O(C/C) true for this
set.20 Formally:
Definition 7 (Deontically Tailored Preferred Subsets) Let I = I, f, g, < be a prioritized
conditional imperative structure, and C LPL describe the given situation. Let (td- pcd) be
any of the truth definitions (td- pcd1(+) 4(+) ). Then is in the set PI ({C}, I ) employed
by this truth definition iff (i) {C} P L , and (ii) is obtained from a full prioritization
by defining
i =
j {i} if
j i j {i}, f, g
otherwise,
j i j
j i
iI
O(C/C)by(td-cd),
i .
By this construction, each of the preferred subsets contains a maximal number of the imperatives such that they do not make true O(C/C) for the situation C and the truth definition
that is employed, and so the resulting truth definition likewise avoids this truth. Such a construction of the preferred subsets might be considered tailored to the truth definition in
question, and any remaining deficiencies might be seen as stemming from the employed
truth definition. But this being so, the method reveals a strong bias towards truth definitions
that accept deontic detachment, and in particular truth definitions (td-pcd2+, 3(+), 4(+) ):
Consider the set of imperatives I = {! p1 , p1 ! p2 , ! p2 } with the ranking ! p1 <
p1 ! p2 < ! p2 , that was used to refute the nave approach. As can be easily checked,
PI ({}, I ) = {I } for all truth definitions (td-pcd1, 1+, 2). So by all these truth definitions,
O( p1 p2 /) is true. So they commit us to violating the second-ranking imperative,
whereas intuitively, the third-ranking imperative should be violated instead. By contrast,
all truth definitions (td-pcd3(+), 4(+) ), that employ reusable output, and of course likewise
(td-pcd2+ ) that is equivalent to (td-pcd4+ ), handle all given examples exactly as it was suggested they should. In particular, consider the ordered imperatives p1! p2 < ! p2 < ! p1 ,
that were used to refute both the reconsidering and the fixpoint approaches: for =
2+, 3(+) , 4(+), PI ({}, I ) is {{ p1 ! p2 , ! p2 }}, making O( p2 /) true by all these truth
definitions, which thus commit us to satisfying the second ranking imperative, and not to
violating it in favor of satisfying the third ranking imperative as these approaches did. The
drinking and driving example is also handled correctly: the set of prioritized imperatives
p1 ! p2 < p3 ! p2 < p3 ! p1 produces, for the situation p3 , the set PI ({ p3 }, I ) =
{{ p1! p2 , p3! p2 }}. So the third ranking imperative, that commits the agent to drinking
and thus, by observation of the highest ranking imperative, prevents the agent from driving,
is disregarded. Instead, the truth definitions make true O( p2 / p3 ), so the agent must do the
driving if she goes to the party, as her best friend asked her to.
Is this the solution, then? Some uneasiness remains as to the quick way with which definitions (td-pcd1, 1+ , 2) are discharged as insufficient. Why should it not be possible to
maintain, as these definitions do, that conditional imperatives only produce an obligation
if they are factually triggered, while at the same time maintaining that the above examples
should not be resolved the way they are? The purpose of a truth definition for the deontic
O-operator is to find a formal notion of ought that reflects ordinary reasoning, and our
intuitions on that matter may differ from our ideas as to what may constitute a good choice
from a possibly conflicting set of prioritized conditional imperatives. I will now make a new
proposal how to construct the preferable subsets, that keeps the positive results without
20 The preferred subsets are thus a choice from the maxfamilies defined in [26].
123
267
committing us to prefer any of the truth definitions of Sect. 3.4 by virtue of their handling of
prioritized imperatives alone.
5.3 Preferred maximally obeyable subsets
What made Brewkas approach so successful is that it maximizes the number of higher ranking imperatives in the preferred subsets of a given set of unconditional imperatives: for each
rank, a maximal number of imperatives are added that can be without making the sets
demands inconsistent in the given situation. As was shown, Brewkas approach cannot be
directly applied to conditional imperatives, since it makes no sense to test the demands of
imperatives for inconsistencies if these imperatives may not be triggered in the same circumstances. Just considering triggered imperatives is also not enough, as was demonstrated for
the nave approach. But if the maximization method is to include imperatives that are not
(yet) triggered, then we must look for something else than inconsistency of demands to take
the role of a threshold criterion for the maximization process.
To do so we should ask ourselves why, for the unconditional case, the aim was to find a
maximal set of imperatives with demands that are collectively consistent with the situation.
I think that by doing so we intend to give the agent directives that can be safely followed.
While in the unconditional case this means that the agent can satisfy all the chosen imperatives, the situation is different for conditional imperatives: here an agent can also obey
imperatives without necessarily satisfying their demands. If you tell me to visit you in case I
go to Luxembourg next month, I can safely arrange to spend all of next month at home and
still do nothing wrong. If we think not so much of imperatives, but of legal regulations, then
I can obviously be a law-abiding citizen by simply failing to trigger any legal norm (even
though this might imply living alone on an island): whether I do that or boldly trigger some
of the regulations antecedents and then satisfy those I have triggered seems not a question of
logic, but of individual choice. So I think the threshold criterion to be used should be that of
obeyability: we should maximize the set of imperatives the agent can obey, and only disregard
an imperative if its addition to the set means that at least one of the added imperatives must
(now) be violated, given the facts.21
For a given set of conditional imperatives and a set of factual truths W , the subsets of
imperatives that can be obeyed are described by ObeyableI (W, ), i.e. they are those subsets
such that W m P L . To maximize not by collective consistency of demands, but
by collective obeyability, Brewkas original approach can therefore be changed as follows:
Definition 8 (Preferred Maximally Obeyable Subsets) Let I = I, f, g, < be a prioritized
conditional imperative structure, be a subset of I , and W L P L be a set of PL-sentences.
Then PIo (W, ) iff (i) W P L , and (ii) is obtained from a full prioritization by
defining
i =
for any i
otherwise,
j i j
j i
, and letting =
i .
The change from Brewkas original definition is only minute: we test not the demands of the
imperatives for consistency, but their materializations. Note that this is a conservative extension of Brewkas method, since for any unconditional imperative i we have P L f (i)
21 While Hansson, in [16] p. 200, also advocates a move from consistency to obeyability, what is meant
123
268
Auton Agent Multi-Agent Syst
m(i). As can easily be seen, the new construction solves all of the previously considered
difficulties, regardless of the chosen truth definition for the deontic O-operator:
To refute a direct application of Brewkas original method, we used the set I = { p1 p2 ,
p1 p2 } with no ranking imposed. m(I ) is consistent and so PIo ({ p1 }, I ) = {I },
making O( p2 / p1 ) true for all definitions of Sect. 3.4 So you ought to wear your boots in
case you go out, as it should be.
To refute the nave approach, we used the set of prioritized imperatives p1 < p1 p2 <
p2 . Since < is already fully prioritized, the construction produces just one maximally
obeyable subset, which is {! p1 , p1! p2 }, as its two imperatives get added in the first two
steps, and nothing is added in the third since m(I ) is inconsistent. All truth definitions
make true O( p1 /), none makes true the non-intuitive formula O( p1 p2 /), and
definitions that accept deontic detachment make true O( p1 p2 /). So you must go
out and wear a scarf, which is as it should be.
To refute the stepwise approach the ordering of the imperatives was changed into p1 p2 <
p1 < p2 . Still PIo (}, I ) = {{! p1 , p1 ! p2 }}, so the sentences made true by the truth
definitions of Sect. 3.4 likewise do not change, and in particular the non-intuitive formula
O( p1 p2 /) is still false, and definitions that accept deontic detachment make true
O( p1 p2 /), so again you must go out and wear a scarf, which is as it should be.
To refute the reconsidering and the fixpoint approaches the ordering of the imperatives
was again changed into p1 ! p2 < ! p2 < ! p1 . Now PIo (}, I ) = {{ p1 ! p2 , ! p2 }}.
All truth definitions make true O( p2 /) but not O( p1 /) so the agent must satisfy the
second ranking imperative, but not the third ranking imperative, which otherwise would
include violating the highest ranking imperative, which is as it should be.
Troublesome for the fixpoint approach was also the set of prioritized imperatives p1
! p2 < !( p1 p2 ) < ! p3 : no fixpoint could be made out and so the approach produced no
preferred subset, making everything obligatory. The preferred maximally obeyable subset
is { p1! p2 , ! p3 }, eliminating the second ranking imperative that demands a violation of
the first, and making O( p3 /) true for all truth definitions, which again is as it should be.
Finally, consider the drinking and driving example: the set of prioritized imperatives
p1! p2 < p3! p2 < p3! p1 produces, for the situation p3 , the set of preferred maximally obeyable subsets PIo ({ p3 }, I ) = {{ p1! p2 , p3! p2 }}, making true O( p2 / p3 )
for all truth definitions of Sect. 3.4, so given that I go to the party I must do the driving,
which is as it should be.
As could be seen, all truth definitions now produce the right results in the examples
used. Moreover, since all truth definitions refer to the same preferred subsets PIo ({C}, I ), it
is possible to index the O-operators according to the truth definition employed, and e.g. state
truths like O 1 (A/C) O 3 (B/C) O 4 (A B/C), meaning that if, for any maximal set of
imperatives that I can obey in the situation C, imperatives are triggered that demand A, and
that if I satisfy all such triggered imperatives, I will have to do B, then obeying a maximal
number of imperatives includes having to do A B. It may well be that natural language
ought-statements are ambiguous in the face of conditional demands, the discussion in Sect.
3 suggested this. If maximal obeyability is accepted as the threshold criterion that limits what
norms an agent can be expected to conform to in a given situation, then definition 8 leaves
the philosophical logician with maximal freedom as to what deontic operator is chosen.
123
269
6 Theorems
Truth definitions (td-pcd1(+) 4(+) ) define when a sentence of the form O(A/C) is true
or false with respect to a prioritized conditional structure I and a situation C. So I briefly
consider what sentences are theorems, i.e. hold for all such structures, given the usual truth
definitions for Boolean operators. It is immediate that for all truth definitions, (DExt), (DM),
(DC), (DN) and (DD-R) are theorems (cf. Sect. 3.1). (DD-R) states that there cannot be both
an obligation to bring about A and one to bring about A unless the situation C is logically
impossible, so our truth definitions succeed in eliminating conflicts. All these theorems are
monadic in the sense that C is kept fixed; in fact, they are the C-relative equivalents of
standard deontic logic SDL. More interesting are theorems (known from the study of nonmonotonic reasoning) that describe relations between obligations in different circumstances.
Obviously we have
(ExtC)
for all truth definitions, i.e. for equivalent situations C, the obligations do not change. As
long as truth definitions are not sensitive to conflicts, e.g. for (td cd (+) 4(+) ), we have
strengthening of the antecedent, i.e. for these definitions
(SA) O(A/C) O(A/C D)
holds. When only maximally obeyable subsets are considered, i.e. for truth definitions (tdpcd1(+) 4(+) ), both (SA) and the weaker rational monotonicity theorem
(RM) O(D/C) O(A/C) O(A/C D)
are refuted e.g. by a set I = {!( p1 p2 ), !( p1 p2 ), p2 p1 } of equally ranking imperatives: though O( p1 /) is true and O( p2 /) false, O( p1 / p2 ) is false. However, for all
definitions of Sect. 3.4, (conjunctive) cautious monotonicity
(CCMon) O(A D/C) O(A/C D)
holds, which states that if you should to two things and you do one of them, you still have
the other one left.22 Moreover, truth definitions (td-pcd1+ , 2+ , 3+ , 4+ ) validate the circumstantial extensionality rule
(CExt)
123
270
Auton Agent Multi-Agent Syst
a theorem. (Cut) is derivable given (Cond) (use Cond on the first conjunct O(A/C D) to
obtain O(D A/C), agglomerate with O(D/C), and from O(D (D A)/C) derive
O(A/C)), which syntactically mirrors the semantic equivalence of definitions (td-pcd2+ )
and (td-pcd4+ ). Theoremhood of all of the above for semantics that employ the respective
truth definitions is easily proved and left to the reader (cf. Hensen [13] and [14] as well as
Makinson and van der Torre [25] for the general outline). Makinson and van der Torres results
also suggest that these theorems axiomatically define complete systems of deontic logic with
respect to semantics that employ the respective truth definitions (td-pcd1(+) 4(+) ), but this
remains a conjecture that further study must corroborate.23
123
271
8 Conclusion
Reasoning about obligations when faced with different and possibly conflicting imperatives
is a part of everyday life. To avoid conflicts, these imperatives may be ordered by priority
and then observed according to their respective ranks. The drinking and driving case in the
introduction presented an example of such natural reasoning. Providing a formal account is,
however, additionally complicated by the fact that there are various and mutually exclusive
intuitions about what belongs to the right definition of an obligation in the face of conditional
imperatives, i.e. the definition of a deontic O-operator. Based on similar definitions of operators by Makinson and van der Torre [25], [26] for their input/output logic, but leaving
the choice of the right operator to the reader, I presented several proposals in Sect. 3 for
definitions of a dyadic O-operator, namely (td-pcd1(+) 4(+) ). These were dependent on
a choice of preferred subsets among a given set of prioritized conditional imperatives. A
particularly successful method to identify such subsets, but applying to unconditional imperatives only, was Brewkas [3] definition of preferred subtheories within a theory. In Sect. 4
I discussed various approaches that extend this method to conditional imperatives, but these
failed to produce satisfactory results for a number of given examples. In Sect. 5 I proposed
that the maximality criterion used to construct the preferred subsets should be the avoidance
of conflict. A recent approach to this task by consistent extensions was found to be biased
towards definitions of obligation that do not accept reasoning by cases; another, that aims
to avoid the truth of O(C/C) for possible circumstances C, produced satisfactory results
only for truth definitions that accept deontic detachment. I then argued that the solution is
to adapt Brewkas method in such a way that it constructs, instead of maximal subsets of
imperatives that are collectively satisfiable by an agent, maximally obeyable subsets of the
imperatives, in the sense that the facts do not derive that some imperative of the set must be
violated. I showed that this new proposal provides adequate solutions to all of the examples,
and in particular the drinking and driving example is resolved in a satisfactory fashion for
all of the discussed deontic operators.
Acknowledgments I am grateful to Lou Goble, John F. Horty, David Makinson and Leon van der Torre
for helpful comments and discussions. Also, I thank four anonymous referees for pointed remarks and useful
suggestions. A preliminary version of this paper was presented to the participants in the seminar Normative
Multi-agent Systems (NorMAS07) held in Dagstuhl, Germany, from 18 to 23 March 2007.
References
1. Alchourrn, C. E., & Bulygin, E. (1981). The expressive conception of norms. In [18], 95124.
2. Alchourrn, C. E., & Makinson, D. (1981). Hierarchies of regulations and their logic. In R. Hilpinen
(Ed.), New Studies in Deontic Logic (pp. 125148). Dordrecht: Reidel.
3. Brewka, G. (1989). Preferred subtheories: An extended logical framework for default reasoning. In
N. S. Sridharan (Ed.), Proceedings of the eleventh international joint conference on artificial intelligence
IJCAI-89, Detroit, Michigan, USA, August 2025, 1989 (pp. 10431048). San Mateo, Calif.: Kaufmann.
4. Brewka, G. (1994). Reasoning about priorities in default logic. In B. Hayes-Roth & R. E. Korf (Eds.),
Proceedings of the 12th national conference on artificial intelligence, Seattle, WA, July 31stAugust 4th,
1994 (Vol. 2 pp. 940945). Menlo Park: AAAI Press.
5. Brewka, G., & Eiter, T. (1999). Preferred answer sets for extended logic programs. Artificial Intelligence,
109, 297356.
6. Broersen, J., Dastani, M., & van der Torre, L. (2002). Realistic desires. Journal of Applied Non-Classical
Logics, 12, 287308.
7. Caminada, M., & Sakama, C. (2006). On the existence of answer sets in normal extended logic programs.
In G. Brewka, S. Coradeschi, A. Perini, & P. Traverso (Eds.), Proceedings of the 17th European con-
123
272
Auton Agent Multi-Agent Syst
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
ference on artificial intelligence (ECAI 2006), August 29September 1, 2006, Riva del Garda, Italy (pp.
741742). Amsterdam: IOS Press.
Downing, P. (1959). Opposite conditionals and deontic logic. Mind, 63, 491502.
van Fraassen, B. (1973). Values and the hearts command. Journal of Philosophy, 70, 519.
Goble, L. (2005). A logic for deontic dilemmas. Journal of Applied Logic, 3, 461483.
Greenspan, P. (1975). Conditional oughts and hypothetical imperatives. Journal of Philosophy, 72, 259
276.
Hansen, J. (2004). Problems and results for logics about imperatives. Journal of Applied Logic, 2, 3961.
Hansen, J. (2005). Conflicting imperatives and dyadic deontic logic. Journal of Applied Logic, 3, 484511.
Hansen, J. (2006). Deontic logics for prioritized imperatives. Artificial Intelligence and Law, 14, 134.
Hansson, B. (1969). An analysis of some deontic logics. Nus, 3, 373398. Reprinted in [17], 121147.
Hansson, S. O. (2001). The structure of values and norms. Cambridge: Cambridge University Press.
Hilpinen, R. (Ed.) (1971). Deontic logic: Introductory and systematic readings. Dordrecht: Reidel.
Hilpinen, R. (Ed.) (1981). New studies in deontic logic. Dordrecht: Reidel.
Hofstadter, A., & McKinsey, J. C. C. (1938). On the logic of imperatives. Philosophy of Science, 6,
446457.
Horty, J. F. (2003). Reasoning with moral conflicts. Nos, 37, 557605.
Horty, J. F. (2007). Defaults with priorities. Journal of Philosophical Logic, 36, 367413.
Kraus, S., Lehmann, D., & Magidor, M. (1990). Nonmonotonic reasoning, preferential models and cumulative logics. Artificial Intelligence, 44, 167207.
Lewis, D. (1974). Semantic analysis for dyadic deontic logics. In S. Stenlund (Ed.), Logical theory and
semantic analysis (pp. 114). Dordrecht: Reidel.
Makinson, D. (1999). On a fundamental problem of deontic logic. In P. McNamara & H. Prakken (Eds.),
Norms, logics and information systems (pp. 2953). Amsterdam: IOS.
Makinson, D., & van der Torre, L. (2000). Input/output logics. Journal of Philosophical Logic, 29, 383
408.
Makinson, D., & van der Torre, L. (2001). Constraints for Input/output logics. Journal of Philosophical
Logic, 30, 155185.
Marek, V. W., & Truszczynski, M. (1993). Nonmonotonic logic context-dependent reasoning. Berlin:
Springer.
Nebel, B. (1991). Belief revision and default reasoning: Syntax-based approaches. In J. A. Allen, R.
Fikes & E. Sandewall (Eds.), Principles of knowledge representation and reasoning: Proceedings of the
second international conference, KR 91, Cambridge, MA, April 1991 (pp. 417428). San Mateo: Morgan
Kaufmann.
Nebel, B. (1992). Syntax-based approaches to belief revision. In: P. Grdenfors (Ed.), Belief revision (pp.
5288). Cambridge: Cambridge University Press.
Prakken, H. (1997). Logical tools for modelling legal argument. Dordrecht: Kluwer.
Prakken, H., & Sartor, G. (1997). Argument-based logic programming with defeasible priorities. Journal
of Applied Non-classical Logics, 7, 2575.
Rescher, N. (1964). Hypothetical reasoning. Amsterdam: North-Holland.
Rescher, N. (1966). The Logic of commands. London: Routledge & Kegan Paul.
Rintanen, J. (1994). Prioritized autoepistemic logic. In: C. MacNish, D. Pearce, & L. M. Pereira (Eds.),
Logics in artificial intelligence, European Workshop, JELIA 94, York, September 1994, Proceedings (pp.
232246). Berlin: Springer.
Ross, W. D. (1930). The right and the good. Oxford: Clarendon Press.
Ryan, M. (1992). Representing defaults as sentences with reduced priority. In B. Nebel, C. Rich, &
W. Swartout (Eds.), Principles of knowledge representation and reasoning: Proceedings of the third
international conference, KR 92, Cambridge, MA, October 1992. (pp. 649660). San Mateo: Morgan:
Kaufmann.
Sakama, C., & Inoue, K. (1996). Representing priorities in logic programs. In: M. Maher (Ed.), Joint
international conference and symposium on logic programming JICSLP 1996, Bonn, September 1996
(pp. 8296). Cambridge: MIT Press.
Sosa, E. (1996). The logic of imperatives. Theoria, 32, 224235.
Swirydowicz,
K. (1994). Normative consequence relation and consequence operations on the language
of dyadic deontic logic. Theoria, 60, 2747.
von Wright, G. H. (1968). An essay in deontic logic and the general theory of action. Amsterdam: North
Holland.
von Wright, G. H. (1984). Bedingungsnormen, ein Prfstein fr die Normenlogik. In W. Krawietz,
H. Schelsky, O. Weinberger, & G. Winkler (Eds.), Theorie der Normen (pp. 447456). Berlin: Duncker & Humblot.
123
273
123
Zusammenfassung
zur Dissertation
Imperatives and Deontic Logic:
On the Semantic Foundations of Deontic Logic
eingereicht an der Fakultat f
ur Sozialwissenschaften und Philosophie der Universitat Leipzig
von Herrn M. A. Jorg Hansen
angefertigt am Institut f
ur Philosophie
offentlich verteidigt am 18. November 2008
jene u
bertragen werden konnen. Mit seinem Aufsatz Deontic Logic stellte
ANHANG: ZUSAMMENFASSUNG
275
anders als das Notwendige zwar nicht zwingend auch der Fall, aber zumindest ist das, was geboten ist, zugleich auch erlaubt. (D) ist in diesem System
aquivalent zu (OA OA): Wenn A geboten ist, kann nicht zugleich non-A
geboten sein. Eine interessante weitere Beziehung zwischen der deontischen
Logik und der alethischen Modallogik wurde 1957 von Alan Ross Anderson
aufgezeigt: Die deontische Logik kann auf die alethische Modallogik reduziert
werden, wenn OA als (A S) definiert wird A ist geboten, wenn non-A
notwendig S nach sich zieht, wobei die Konstante S f
ur das Eintreten einer
Sanktion steht. Unterstellt man, da S moglich ist, also das Eintreten einer
Sanktion vermieden werden kann, so wird das Standardsystem der deontischen Logik mittels einer solchen Definition zu einem Fragment des (um die
Konstante S bereicherten) alethischen Modalkalk
uls.
Wie die Ausdr
ucke der alethischen Modallogik werden auch die Ausdr
ucke der deontischen Logik u
berwiegend mithilfe einer Mogliche-Welten
Semantik interpretiert: OA, der mit A beschriebene Zustand ist geboten,
ist wahr, wenn A in allen Idealwelten zu unserer aktuellen Welt wahr ist. Mithilfe einer logischen Semantik, deren Modelle eine Menge moglicher Welten,
eine zwischen ihnen bestehenden Idealitatsrelation, sowie eine Wahrheitsfunktion angeben, die den Zustand in jeder Welt beschreibt, kann sodann
die G
ultigkeit aller Satze und die Vollstandigkeit des Standardsystem der
deontischen Logik und seiner Abwandlungen bewiesen werden.
Die deontische Logik wird bis heute gelegentlich als eine Logik der Nor
men charakterisiert. Diese Darstellung ist freilich falsch, denn Normen konnen weder wahr noch falsch sein, sie wollen die Welt nicht beschreiben, sondern etwas in ihr bewirken. Nur wahre oder falsche Satze konnen aber Einsetzunginstanzen f
ur OA und P A etwa in dem obigen Schema (D) sein, denn
die durch den Pfeil ausgedr
uckte wenn ... dann Beziehung ist wahr
heitsfunktional: Der Satz ist falsch genau dann wenn A geboten ist, aber A
gleichwohl nicht erlaubt ist (weil z.B. A ebenfalls geboten ist). Wenn OA
deshalb keine Norm sein kann, sondern ein beschreibender Satz ist, so gelten
die normalen Regeln: OA, also A ist geboten, ist genau dann wahr wenn
A geboten ist, wenn es also eine Norm gibt, die A gebietet. Die deontische
Logik versteht daher ihre Ausdr
ucke u
blicherweise als wahre oder falsche Behauptungen dar
uber, was gema einem (meist nicht naher beschriebenen)
Normsystem geboten, verboten oder erlaubt ist.
Wenn sich aber die Wahrheit der Satze der deontischen Logik nach dem
richtet, was gema tatsachlich bestehender Normen als geboten, verboten
oder erlaubt anzusehen ist, mu sie die Eigenheiten bestehender Normsy-
ANHANG: ZUSAMMENFASSUNG
276
steme abbilden konnen. Hierzu gehort zunachst die Begleitung von Primarnormen durch Sekundarnormen (oder contrary-to-duty imperatives), die
Pflichten f
ur solche Falle statuieren, in denen die Primarnormen verletzt
werden. Solche Sekundarpflichten sind etwa: sich entschuldigen, den Schaden wiedergutmachen, eine Strafe zahlen usw. Hierzu gehort aber auch die
Moglichkeit eines Normenkonflikts: Es mag, durch einen Fehler des Gesetzgebers oder den Irrtum eines Richters, zwei oder mehr Normen geben, die
Gegensatzliches verlangen. Haufig gibt es aber in Fallen von Normenkonflikten zugleich Losungen mithilfe von Prioritatsregelungen: die Norm des Verfassungsrechts hat Vorrang vor der des normalen Gesetzesrecht, die spatere
Norm vor der fr
uheren, die speziellere vor der allgemeinen, diejenige, die bedeutende Rechtsg
uter sch
utzt, hat Vorrang vor bloen Ordnungsregeln, usw.
Sowohl axiomatische wie semantische Ansatze der traditionellen deontischen Logik geraten bei der Bewaltigung dieser Besonderheiten in Schwierigkeiten. Das Standardsystem der deontischen Logik hat sich zur Beschreibung
von Sekundarnormen als unbrauchbar erwiesen und wurde durch Systeme
mit zweistelligen deontischen Operatoren ersetzt: In der sogenannten dyadischen deontischen Logik bedeutet O(A/B): A ist geboten in der Situation
B. In der logischen Semantik wurden die Idealwelten durch eine Praferenzrelation zwischen moglichen Welten ersetzt: O(A/B) ist nun wahr, wenn A
in den besten aller moglichen B-Welten der Fall ist. Noch schwieriger ist
die Darstellung von Normenkonflikten: wenn es moglich ist, da sowohl OA
wie auch OA wahr sind, ist (D) als Axiom widerlegt. Aber auch die u
brigen Axiome des Standardsystems m
ussen modifiziert werden, damit aus OA
und OA nicht OB folgt (sogenannte deontische Explosion). Semantisch
reichen die Abhilfevorschlage vom Einsatz minimaler Modelle (sog. Nachbarschaftsemantiken, bei denen nicht eine, sonderen mehrere Mengen von
Welten gegen
uber einer anderen ideal sind) bis zu Multipraferenzseman
tiken, bei denen nicht eine, sondern mehrere Praferenzrelationen die jeweils
besten aus einer gegebenen Menge von Welten auswahlen. Noch schwieriger ist die Rekonstruktion von Konfliktlosungsstrategien mittels Prioritaten:
Hier m
uten konsequenterweise Praferenzrelationen zwischen Mengen idealer
Welten ( beste Nachbarschaften) oder zwischen anderen Praferenzen zum
Einsatz kommen, selbst wenn dies meines Wissens noch nirgendwo versucht
worden ist. Alle diese Semantiken haben jedenfalls den Nachteil, da sie nicht
sonderlich intuitiv sind.
In der Behandlung von Beispielen hat es sich als besonders unvorteilhaft
erwiesen, da es in der Sprache der deontischen Logik unmoglich ist, unmittel-
ANHANG: ZUSAMMENFASSUNG
277
weil zwei Normen AC und B C fordern es macht dann keinen Sinn, von
den Normadressaten sowohl die Realisierung von A als auch B zu fordern,
wenn doch nur eine dieser beiden Normen erf
ullt werden kann: Das Tun von
entweder A oder B ware u
ussig.
berfl
In dieser Situation ist es kein Wunder, wenn sogar in Standardwerken zur
deontischen Logik aus Gebots- und Erlaubnissatzen klammheimlich Imperative oder deontische Normen werden, etwa gesagt wird, da A die deon
tische Norm OA verletze. Da dies unzulassig ist, habe ich oben festgestellt.
Und so war auch meine eigene Ausdrucksweise im vorigen Absatz irref
uhrend,
denn OA ist kein Gebot, sondern ein Satz, da etwas Bestimmtes geboten
ist, er kann nicht verletzt oder erf
ullt, sondern nur wahr oder falsch sein;
er ist bestenfalls eine Pramisse. David Makinson sah sich deshalb wahrend
des Workshops EON98 in Bologna 1998 zu der Feststellung veranlat, die
Arbeiten zur deontischen Logik erweckten vielerorts den Eindruck, als habe
der Autor noch nie von einer Unterscheidung zwischen Normen und deontischen Satzen u
ber Normen gehort. Die deontische Logik sei in Einzelteile
zerfallen: in zahlreiche Axiomatisierungen, und in noch mehr unterschiedliche Mogliche-Welten-Semantiken mit ihren unzahligen Detailabweichungen.
Er forderte dazu auf, die deontische Logik zu rekonstruieren : als eine Lo
gik, die mit Normen befat ist, aber respektiert, da Normen weder wahr
noch falsch sein konnen.
Diesem Ruf folgend habe ich seither in einer Reihe von Forschungsarbeiten
untersucht, wie sich mithilfe einer imperativistischen Semantik die deonti
schen Operatoren neu definieren lassen, und welche alten oder neuen Systeme
ANHANG: ZUSAMMENFASSUNG
278
ANHANG: ZUSAMMENFASSUNG
279
Kapitel 1 ergibt jedoch, da die Existenz unmittelbarer logischer Beziehungen zwischen Imperativen nicht unterstellt werden kann. Hierf
ur liefere ich ein
Argument in der Tradition der Ordinary Language Philosophy, und damit
zu, denn einige Autoren haben bereits in Abweichung von der herkommlichen Mogliche-Welten-Semantik versucht, die deontischen Operatoren mittels explizit angegebener Normsysteme oder Imperativmengen zu definieren.
Sodann erlautere ich die zur Rekonstruktion der traditionellen deontischen
Systeme verwendete Semantik und beschreibe zwei wichtige Grundvoraussetzungen: die Unabhangigkeit und die Untrennbarkeit der in der logischen
Semantik modellierten Imperative. Die oben bereits kurz erlauterten Ergebnisse meiner Forschungsarbeiten zu dieser Semantik werden anschlieend zusammengefat. Abschlieend wende ich mich den sogenannten Paradoxien
der deontischen Logik zu und untersuche, wie der neue semantische Ansatz
mit diesen Paradoxien umgehen kann.
In gewisser Weise bedeutet eine imperativistische Semantik einen Pa
radigmenwechsel f
ur die deontische Logik. Verstand sich die deontische Logik
zuvor, etwa im Spatwerk Georg Henrik von Wrights, als Ratgeber eines ra-
ANHANG: ZUSAMMENFASSUNG
280
Eine Rationalitat des Gesetzgebers wird nicht vorausgesetzt, wohl aber ein
u
berlegendes Normsubjekt. Neben solchen grundsatzlichen Erwagungen zu
unserem Projekt wird im Schluteil auf fortbestehende Schwierigkeiten hingewiesen. Gebotsausdr
ucke lassen sich in Bezug auf konditionale Imperative in
verschiedener Weise definieren, konnen z.B. die sogenannte Wahlfeststellung
zulassen (wenn A B der Fall ist, und A und B beide Imperative auslosen,
ist dann auch geboten, was ein Imperativ verlangt, der durch B ausgelost
B C erforderlich ist, ist dann C geboten?). Diese Argumentationsweisen konnen jedoch nur in bestimmten Kombinationen zugelassen werden,
und f
uhren in anderen Kombinationen zu Paradoxien: Welches ist also die
richtige Gebotsdefinition f
ur konditionale Imperative? Erlaubende Normen
Bibliographische Beschreibung
Hansen, Jorg
Imperatives and Deontic Logic: On the Semantic Foundations of
Deontic Logic
Universitat Leipzig, Diss., 120 S., 260 Lit.
Referat:
Die Satze der deontischen Logik werden traditionell mithilfe der Mogliche
Welten-Semantik interpretiert. Andererseits versteht die deontische Logik,
in Abgrenzung von einer Normenlogik, ihre Satze als wahre oder falsche
Aussagen dar
uber, was gema einem meist nicht naher beschriebenen System von Normen geboten, verboten oder erlaubt ist. Dieser Erklarung ihres Gegenstands folgend haben einige Autoren versucht, die Wahrheit oder
Falschheit der Satze der deontischen Logik nicht mit R
ucksicht auf mogliche
Welten zu interpretieren, sondern mittels einer explizit angegebenen Normmenge oder Menge von Imperativen, und was zu ihrer Erf
ullung erforderlich
ist. Es kann dann gezeigt werden, da sich im Sinne dieser imperativisti
schen Tradition der deontischen Logik die traditionellen axiomatischen Systeme der monadischen und dyadischen Logik rekonstruieren lassen. Zu
gleich ermoglicht dieser logisch-semantische Ansatz die Konstruktion anderer Systeme, die etwa Konflikte zwischen Imperativen zulassen oder solche
Konflikte mittels Vorrangbeziehungen zwischen den Imperativen ganz oder
teilweise losen. Die Arbeit erlautert die Grundidee dieser imperativistischen
281
Lebenslauf
09.09.1967
01/1985 12/1985
21.06.1988
10/1988 03/1991
05/1989 03/1991
04/1991 06/1997
05/1991 09/1993
08/1992 10/1999
05.08.1994
04/1995 09/1995
10/1995 03/1996
02/1997 06/1999
02.06.1997
04.06.1999
14.09.1999
09/1999
10/1999
282
ANHANG: LEBENSLAUF
10/2000 03/2001
283
Hilfskraft am Institut f
ur Logik und Wissenschaftstheorie (Dr. Peter Steinacker)
07/2001 12/2001 Geschaftsf
uhrer Bachhaus Eisenach gGmbH
12/2005
Geschaftsf. u. Direktor Bachhaus Eisenach gGmbH
21.11.2007
Fachanwalt f. Arbeitsrecht
Publikationen
Paradoxes of Commitment, in: Meggle, G. (Hrsg.), Actions, Norms, Values.
Discussions with Georg Henrik von Wright, De Gruyter: Berlin, New York,
1999, 255263.
On Relations between
Aqvists Deontic System G and Van Ecks Deontic
221232.
Prioritized Conditional Imperatives: Problems and a New Proposal, Auto
nomous Agents and Multi-Agent Systems, 2007, 17, 2008, 1135.
Mit Gabriella Pigozzi und Leon van der Torre:
Ten Philosophical Problems in Deontic Logic, in: Boello, G., van der Torre,
284
Vortr
age
10.01.1998
On Relations between
Aqvists Deontic System G and Van Ecks
22.01.2000
22.05.2002
28.05.2004
27.05.2005
20.03.2007
20.03.2007
05.08.2007
285
ANHANG: VORTRAGE
286
28.04.2008
04.-15.08.2008
Torre, ESSLI 2008, 20th European Summer School in Logic, Language and Information, Hamburg)
Versicherung
Hiermit versichere ich, da ich die vorliegende Arbeit ohne unzulassige Hilfe
Dritter und ohne Benutzung anderer als der angegebenen Hilfsmittel angefertigt habe; die aus fremden Quellen direkt oder indirekt u
bernommenen
Gedanken sind als solche kenntlich gemacht.
Bei der Auswahl und Auswertung des Materials sowie bei der Herstellung
des Manuskriptes habe ich Unterst
utzungsleistungen von folgenden Personen
erhalten: dem englischsprachigen Korrekturleser Herrn Jason F. Ortmann.
Weitere Personen waren an der geistigen Herstellung der Arbeit nicht beteiligt. Insbesondere habe ich nicht die Hilfe eines Promotionsberaters in
Anspruch genommen. Dritte haben von mir weder unmittelbar noch mittelbar geldwerte Leistungen f
ur Arbeiten erhalten, die im Zusammenhang mit
dem Inhalt der vorgelegten Dissertation stehen.
Die Arbeit wurde bisher weder im In- noch im Ausland in gleicher oder
ahnlicher Form einer anderen Pr
ufungsbehorde vorgelegt und ist auch noch
nicht veroffentlicht worden.
287