Identifying Indirect Normative Conflicts using the WordNet Database

Jéssica Soares dos Santos

and Viviane Torres da Silva

Computer Science Department, Universidade Federal Fluminense, Niterói, Brazil

IBM Research (on leave from Universidade Federal Fluminense), Rio de Janeiro, Brazil

Keywords: Multi-agent Systems, Normative Conflicts, Conflict Detection, WordNet.

Abstract: A challenging issue in Multi-agent systems governed by multiple norms is to deal with normative conflicts,

which are situations where the fulfilment of a norm violates another one. There are several approaches in

the literature to detect conflicting norms. Some kinds of conflicts, here called direct conflicts, can be

detected through a simple comparison between the elements of the norms in order to check if they apply to

the same elements. For instance, if an obligation and a prohibition are applied to the same entity and govern

the same behavior in periods of time that intersects, these norms are in conflict. However, there are

conflicts, here called indirect conflicts, that can only be detected if the relationships among the elements of

the norms are taken into account. The majority of approaches that are able to detect indirect conflict

considers the relationships of the application domain that have been previously defined by the designer.

Different from those approaches, this paper focuses on the detection of indirect conflicts by taking into

account domain-independent relationships that have not been declared by the designer of the multi-agent

system. Our proposal searches for domain-independent relationships among the elements of the norms in the

WordNet database.

1 INTRODUCTION

In open Multi-Agent Systems (MASs) norms have

been applied as a means to restrict and guide the

behavior of autonomous and independently designed

software agents in order to avoid undesirable

behaviour. A norm is commonly associated with a

deontic concept (obligation, permission or

prohibition) and defines which actions an agent can

or cannot perform (Grossi et al., 2010). Due to the

numeral norms that may be necessary to govern a

normative MAS, conflicts between the norms may

arise. A conflict between two norms occurs when the

fulfilment of a norm violates the other norm. For this

reason, to guarantee the proper functioning of a

normative MAS, mechanisms for dealing with

normative conflicts are needed. A normative conflict

can be classified as:

• Direct Conflict: it involves two norms that are

associated with the same addressee, regulate the

same behavior and have opposite or contradictory

deontic modalities (obligation versus prohibition or

permission versus prohibition). This kind of conflict

can be detected through a simple analysis of the

norm elements (i.e., the addressee, behaviour and

context of the norm). Thus, a direct normative

conflict arises, for instance, when a norm obliges an

agent to perform an action in a given organization

and another norm prohibits the same addressee to

execute the same action in a same organization.

• Indirect Conflict: it involves two norms whose

elements are not the same but are related. It can only

be detected when relationships among elements of the

norms are identified. The deontic concepts associated

with the norms involved in an indirect conflict can be

opposite, contradictory or equal. For instance, an

indirect normative conflict arises between two norms

when both norms are addressed to the same agent and

one is prohibiting the execution of an action and the

other is obligating the execution of another action that

is a specialization of this action. Since the more

general action are being prohibited the more concrete

action cannot be executed.

The aim of our research is to develop a

mechanism able to detect indirect normative

conflicts by using the WordNet database (Miller,

1995) to identify the domain-independent

relationships among the elements of the norms.

WordNet is a public lexical database that stores

relationships among words.

The main difference between our approach and

others also able to detect indirect conflicts is that our

186

Santos, J. and Silva, V.

Identifying Indirect Normative Conﬂicts using the WordNet Database.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 2, pages 186-193

ISBN: 978-989-758-187-8

approach detects domain-independent relationships.

Other approaches, such as (da Silva and Zahn, 2014;

Aphale et al., 2012; Sensoy et al., 2012), focus on

the identification of domain-dependent relationships

that have been defined by the application designer.

In fact, our approach and the ones published in the

literature are complementary since the identification

of all kinds of indirect conflicts can only be detected

if both kinds of relationships are considered.

This paper is organized as follows. Section 2

presents related work. Section 3 describes the kinds

of relations that are defined in the WordNet

database. Section 4 presents the norm definition

adopted in this paper. Section 5 describes the kinds

of relationships explored by our approach and

formalizes the rules for conflict detection. Section 6

presents the detection algorithm and describes our

mechanism. Finally, Section 7 concludes and

presents some future work.

2 RELATED WORK

Several techniques have been proposed to detect

normative conflicts. Some of them only are able to

detect direct conflicts (Li et al., 2014; Vasconcelos

et al., 2012; dos Santos Neto et al., 2013; Gunay and

Yolum, 2013) while others can also detect indirect

conflicts (da Silva and Zahn, 2014; Aphale et al.,

2012; Sensoy et al., 2012; Fenech et al., 2009;

Giannikis and Daskalopulu, 2011). Each approach

presents a different definition for norms and can

detect different kinds of normative conflicts. The

majority of approaches to detect indirect conflicts

only deals with relationships among actions, i.e.,

they do not consider relationships between other

elements of the norm.

Since our approach focuses on the detection of

indirect conflicts, we will present a brief description

about the kinds of relationships that related

approaches consider in order to detect indirect

normative conflicts. Note that none of the

approaches are able to detect domain-independent

relationships. All of them assume that the domain-

dependent relationship are provided to the conflict

checker in order to check for conflicts.

2.1 Action Relationships

This subsection describes the action relationships

that have been found in the literature.

• Refinement: it defines a relation of specialization

among actions, i.e., it relates a sub action to a super

action. It is described in the following approaches

(da Silva and Zahn, 2014; Zahn, 2015);

• Composition: it defines a relation of composition

among actions, i.e., it determines that an action is

composed of another one. It is described in (da Silva

and Zahn, 2014; Zahn, 2015; Vasconcelos et al.,

2009; Aphale et al., 2012; Sensoy et al., 2012);

• Orthogonality: it defines actions that cannot be

performed simultaneously by the same entity, as

presented in (da Silva and Zahn, 2014; Zahn, 2015;

Fenech et al., 2008, 2009; Giannikis and

Daskalopulu, 2011, 2009);

• Dependency: it defines actions that are

preconditions of other actions. It is described in the

following approaches (da Silva and Zahn, 2014;

Zahn, 2015; Aphale et al., 2012; Sensoy et al.,

2012);

• Actions’ Side Effects: it determines the side-

effects of the execution of an action, as detailed in

(Aphale et al., 2012; Sensoy et al., 2012;

Kollingbaum and Norman, 2004; Kollingbaum et al.,

2006; Kollingbaum and Norman, 2006;

Kollingbaum et al., 2007).

Additionally, the approach presented in

(Vasconcelos et al., 2009) also can detect normative

conflicts that occur due to the delegation of tasks

among agents. Moreover, the work in (Zahn, 2015)

is able to analyze relationships (such as refinement,

dependency, orthogonality and composition)

between: (i) two actions; (ii) two states; or (iii) an

action and a state.

2.2 Entity Relationships

This subsection describes the entity relationships

that have been found in the literature.

• Play: it relates an agent to the roles it can play. It

is described in the following approaches (da Silva

and Zahn, 2014; Zahn, 2015; Cholvy and Cuppens,

1995, 1998);

• Play-in: it relates an agent to the organization

where it is playing a role, as presented in (da Silva

and Zahn, 2014;);

• Ownership: it relates roles to the organizations

where they were defined and can be played, as

detailed in (da Silva and Zahn, 2014; Zahn, 2015);

• Hierarchy: it relates a sub role to a super role. It is

defined in (da Silva and Zahn, 2014; Zahn, 2015).

2.3 Context Relationships

This subsection describes the context relationships

that have been found in the literature. The context of

Identifying Indirect Normative Conﬂicts using the WordNet Database

187

a norm represents the scope where it is defined, that

is, where the norm must be applicable. For instance,

a norm can be applied to an environment or to a

given organization.

• Inhabit: it relates an entity to the environment that

is its habitat. It is described in the following

approaches (da Silva and Zahn, 2014; Zahn, 2015);

• Hierarchy: it relates a sub context to its super

context, as presented in (da Silva and Zahn, 2014;

Zahn, 2015).

3 WORDNET

The WordNet is a lexical database that stores

relationships among words, which can be nouns,

verbs, adjectives and adverbs. WordNet groups

words that share the same meaning in a given

context. Each group of synonymous words is called

synset. A synset is the building block of WordNet

and all words in a synset denote the same concept.

Each synset has a brief description (called “gloss”)

of its meaning and may be associated with short

sentences that exemplify the use of synset members.

For instance, the nouns “error” and “mistake” are

grouped in a same synset whose gloss is “part of a

statement that is not correct”. Synsets are interlinked

through semantical and lexical relations. We will use

WordNet to identify relationships among the

elements of the norms that have not been defined in

the domain application. The relations that are stored

in WordNet are described in the next subsection.

3.1 WordNet Relations

WordNet defines six relationships, as follows:

• Synonymy: it is the main relation of WordNet.

Synonymous are words with the same meaning in a

given context. As stated in the beginning of this

Section, in WordNet words are grouped into synsets

(synonymous sets) and all words of a synset denote

the same concept. A word may appear in more than

one synset if it is associated with different meanings.

This relation is defined to nouns, verbs, adjectives

and adverbs.

• Hyponymy/Hypernymy: it relates a subset to a

superset since it links one synset more general to

another one more specific. For instance, whale is a

hyponym of mammal because whale is a kind of

mammal. Similarly, mammal is a hypernym of

whale. This relationship is similar to hierarchy

described in Section 2. This relation is defined to

nouns.

• Meronymy/Holonymy: it corresponds to the

relation part-whole. For instance, seat is a meronym

of chair because seat is part of chair. Similarly,

chair is a holonym of seat. This relation is only

defined to nouns.

• Entailment: it defines that a verb entails another

one. This relation is unilateral. For instance, to buy

entails to pay because to buy we need to pay. Then,

if two verbs are related by an entailment relation and

the event denoted by the first verb occur, the action

denoted by the second verb will also occur. This

relation is only defined to verbs.

• Antonymy: it relates a word to another one with

opposite meaning. For instance, clean is antonym of

dirty and vice-versa. This relation can be regarded a

special case of the orthogonality cited in Section 2.

This relation is defined to nouns, verbs and adverbs.

• Troponymy/Hypernymy: it relates a verb

denoting an action to another one denoting a manner

of doing this action. For instance, to lisp is a manner

of to talk. Thus, to lisp is a troponym of to talk. This

relationship is similar to refinement described in

Section 2. This relation is only defined to verbs.

4 NORM DEFINITION

The norm definition used in this paper is based on

the definition presented in (Zahn, 2015). We

consider that norms oblige, prohibit or permit an

entity to perform an action in a given context during

a period of time. Additionally, we consider that

agents inhabit environments and play roles in

organizations.

Our definition of norm is more expressive than

the one described in (Zahn, 2015) because we extend

it including an optional field “obj”, which

corresponds to a parameter of the action being

regulated by the norm that describes an object

applied to the action. In order to exemply, we can

consider the following actions: dress(skirt) and

eat(banana).

A norm is a tuple of the form:

Norm = 〈id, d, c, e, act (obj), ac, dc〉,

where id is the norm identifier, d is a deontic

concept from the set {obligation, permission,

prohibition}; c ∈ C is the context where the norm is

defined (i.e., an organization org ∈ O or an

environment env ∈ Env.); e ∈ E is the entity whose

action is being regulated by the norm (i.e., an agent

a ∈ A, an organization org ∈ O or a role r ∈ R. We

use the symbol “_” to determine that a norm is

addressed to all entities of a given context); act ∈

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

188

Act is the action being regulated; obj ∈ Obj is an

object associated with the action (it is an optional

field. We will use the symbol “_” to represent that

an action is not associated with any object.); ac ∈

Cnd is the condition that activates the norm; dc ∈

Cnd is the condition that deactivates the norm. These

conditions are dates represented in the format:

MM/dd/yyyy HH:mm:ss.

5 KINDS OF RELATIONSHIPS

In this section we present the relationships that our

proposal will analyze in order to detect indirect

normative conflicts. First, the elements of a norm are

mapped to nouns or verbs and then we identify the

relationships among such elements.

We adopt the norm propagation method

described in (Zahn, 2015; da Silva et al., 2015),

which determines propagation rules to create new

norms in order to facilitate the detection process.

Propagation rules propagate norms according to the

relationships between entities or contexts. After the

norm propagation, the detection method only need to

verify, for each pair of norms addressed to the same

entity and context, if the actions regulated by the

norms are related. To avoid the unnecessary

propagation of norms addressed to synonymous

entities/contexts, we identify and locally rewrite

synonymous entities/contexts before propagating

norms. We use First-order logic (Smullyan, 1995) to

define norm propagation constraints. For the sake of

simplicity, we omit the norm identifier id.

Since our norm definition (Section 4) considers

that an action may be applied over an object, norms

associated with related actions only can conflict with

each other if both actions are associated with the

equivalent objects (i.e., synonyms or equal) or they

are not associated with objects.

The identification of relationships and the

propagations do only consider the entities, contexts,

and actions defined in the set of norms being

considered.

5.1 Context Relationships

In this subsection we describe the context

relationships that can be captured by our approach.

All contexts are mapped to nouns and the

relationships defined in the WordNet that applied to

nouns are investigated. We do not consider the

antonymy relationship because there is not a case

where a norm must be propagated due to an

antonymy relation between contexts.

• Synonymy Among Contexts: we use the

WordNet relation “Synonymy” to identify

synonymous contexts. For instance, the context USA

is equivalent to the context United States of

America.

Rewriting Rule: Synonymous contexts are locally

rewritten in order to unify the norm elements that

are equivalent. Thus, we iterate over the set of

contexts used in the set of norms and create a list

where each item is a pair of synonyms contexts.

After that, we replace each context for its synonym

stored in the synonyms list. For instance, if there is

a norm n1 that is associated with the context USA

and there is a norm n2 associated with the context

United States of America, since both contexts are

synonyms, norm n2 is locally rewritten to be

associated with the context USA.

•

Specialization Among Contexts: we use the

WordNet relation “Hyponymy/ Hypernymy” to

identify related contexts. For instance, the context

Brazil is a hyponym of the context South America.

Propagation Rule: If a norm is addressed to an

organization and does not specify a specific entity,

so it must be addressed to all its sub organizations.

∃o

(〈d, o

, _, act (obj), ac, dc〉 ∈ Norm →

∀o

(HyponymyHypernymy (o

, o

) →

〈d, o

, o

, act (obj), ac, dc〉 ∈ Norm))

where d ∈ {obligation, permission, prohibition}; o

, o

∈

O; act ∈ Act; obj ∈ Obj; ac ∈ Cnd; dc ∈ Cnd.

• Part-whole Among Contexts: we use the

WordNet relation “Meronymy/ Holonymy” to

identify related contexts. For instance, the context

intensive care unit (ICU) is part of the context

hospital.

Propagation Rule: If a norm is addressed to an

organization and does not specify a specific entity,

it must be addressed to all organizations that are

part of the given organization.

∃o

(〈d, o

, _, act, ac, dc〉 ∈ Norm →

∀o

(MeronymyHolonymy (o

, o

) →

〈d, o

, o

, act, ac, dc〉 ∈ Norm))

where d ∈ {obligation, permission, prohibition}; o

, o

∈

O; act ∈ Act; obj ∈ Obj; ac ∈ Cnd; dc ∈ Cnd.

5.2 Entity Relationships

In this subsection we describe the entity

relationships that can be captured by our approach.

All entities are mapped to nouns and the

relationships defined in the WordNet that applied to

nouns are investigated. We do not consider the part-

whole relationship among entities because we did

not find cases where an entity is composed of parts.

Our approach can be easily extended to support the

Identifying Indirect Normative Conﬂicts using the WordNet Database

189

relationship part-whole among entities if necessary.

We also do not consider the relationship antonymy

among entities because it does not imply that a norm

must be propagated.

• Synonymy Among Entities: we use the WordNet

relation “Synonymy” to identify synonymous

entities. For instance, the entity child is equivalent to

the entity kid.

Rewriting Rule: Synonymous entities are locally

rewritten in order to unify the norm elements that

are equivalent. This process is similar to the one

described in the “synonymy among contexts”, in

Section 5.1.

• Specialization Among Entities: we use the

WordNet relation “Hyponymy/ Hypernymy” to

identify related entities. For instance, the role

angiologist is hyponym of the role doctor.

Propagation Rule: If a norm is addressed to a

hypernym entity (super role), then it must apply to

all sub roles of the given super role. Only the roles

defined in the domain ontology are compared.

∃r

(〈d, c, r

, act (obj), ac, dc〉 ∈ Norm →

∀r

(HyponymyHypernymy (r

, r

) →

〈d, c, r

, act (obj), ac, dc〉 ∈ Norm))

where d ∈ {obligation, permission, prohibition}; c ∈ C; r

∈ R; act ∈ Act; obj ∈ Obj; ac ∈ Cnd; dc ∈ Cnd.

5.3 Action Relationships

In this subsection we describe the action

relationships that can be captured by our approach.

All actions are mapped to verbs and the relationships

defined in the WordNet that applied to verbs are

investigated.

• Synonymy Among Actions: we use the WordNet

relation “Synonymy” to identify that an action is

equivalent to another one. For instance, the actions

to cooperate and to collaborate are synonyms.

Conflict Rule: Norms associated with

synonymous actions are in conflict if one is a

prohibition and the other is a permission or

obligation, and both actions are applied to the

equivalent objects (synonym or equal) or they are

not associated with objects.

∃act

((〈prohibition, c, e, act

(obj

), ac

,dc

〉 ∈ Norm

∧ 〈d, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ (Synonymy (act

, act

)) →

conflict (〈prohibition, c, e, act

(obj

), ac

, dc

〉,

〈d, c, e, act

(obj

), ac

, dc

〉))

where d ∈ {obligation, permission}; c ∈ C; e ∈ E; act

act

∈ Act; obj

∈ Obj; ac

, ac

∈ Cnd; dc

, dc

∈ Cnd; dc

≥ ac

; and dc

≥ ac

• Specialization Among Actions: we use the

WordNet relation “Troponymy/Hypernymy” to

identify related actions. For instance, the action to

talk is a troponym of the action to communicate.

Conflict Rule: Norms associated with

specialization actions are in conflict if the super

action (hypernym) is prohibited and the sub action

(troponym) is permitted or obliged, and both

actions are applied to equivalent objects (synonym

or equal) or they are not associated with objects.

∃act

((〈prohibition, c, e, act

(obj

), ac

, dc

〉 ∈

Norm

∧ 〈d, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ TroponymyHypernymy (act

, act

)) →

conflict (〈prohibition, c, e, act

(obj

), ac

, dc

〉,

〈d, c, e, act

(obj

), ac

, dc

〉))

where d ∈ {obligation, permission}; c ∈ C; e ∈ E; act

act

∈ Act; obj

∈ Obj; ac

, ac

∈ Cnd; dc

, dc

∈ Cnd; dc

≥ ac

; and dc

≥ ac

•

Entailment: we use the WordNet relation

“Entailment” to identify related actions. For

instance, the action to buy entails the action to pay.

Conflict Rule:

Norms associated with entailment

actions are in conflict if the entailed action is a

prohibition and the other action is an obligation or

permission, and both actions are applied to

equivalent objects (synonym or equal) or they are

not associated with objects.

∃act

((〈d, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ 〈prohibition, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ Entailment (act

, act

)) →

conflict (〈d, c, e, act

(obj

), ac

, dc

〉,

〈prohibition, c, e, b

(obj

), ac

, dc

〉))

where d ∈ {obligation, permission}; c ∈ C; e ∈ E; act

act

∈ Act; obj

∈ Obj; ac

, ac

∈ Cnd; dc

, dc

∈ Cnd; dc

≥ ac

; and dc

≥ ac

• Antonymy Among Actions: we use the WordNet

relation “Antonymy” to identify that an action is

opposite to another one. For instance, the action to

enter is antonym of the action to leave.

Conflict Rule: Norms associated with antonym

actions are in conflict if they are obligations, and

both actions are applied to equivalent objects

(synonym or equal) or they are not associated with

objects. Note that a conflict does not occurs if the

norms are permissions or are prohibitions.

∃b

((〈obligation, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ 〈obligation, c, e, act

(obj

), ac

, dc

〉 ∈ Norm

∧ Antonymy (act

, act

)) →

conflict (〈obligation, c, e, act

(obj

), ac

, dc

〉,

〈obligation, c, e, act

(obj

), ac

, dc

〉))

where c ∈ C; e ∈ E; act

, act

∈ Act; obj

∈ Obj; ac

, ac

∈

Cnd; dc

, dc

∈ Cnd; dc

≥ ac

; and dc

≥ ac

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

190

6 ALGORITHM

In this section we present pseudocodes and describe

our detection method. The WordNet Conflict

Checker

checkes for conflicts between a set of

norms by considering the relationships between their

elements (entities, contexts, actions) that are defined

in the WordNet database. The conflict checker was

implemented by using Java language. To execute

searches over the WordNet database, we use JWNL

(Walenz and Didion, 2011), a Java library that allow

us to connect the information from an offline

WordNet database to a Java program. The main

steps of our detection method (see Algorithm 1) are

described as follows

. Due to the limitation of space

we omit the algorithms of identification of

synonyms and norm propagation.

• Identify Synonymous: We use the WordNet

database (synonymy relationship) to identify the

synonymous contexts and entities and locally rewrite

the norms to unify the syntax of the norms and avoid

the propagation of equivalent norms. In this step we

store the pairs of synonymous entities/contexts in a

list. Each item of the list is a pair of the form (key,

value), where the key is the entity to be replaced for

the entity stored in the value. For instance, if the

domain ontology has the entities: child, kid, minor,

the algorithm will iterate sequentially through the

entities and create the list: {(kid, child), (minor,

child)}. Then, norms addressed to the entity kid or to

the entity minor will be locally rewritten to be

addressed to the entity child.

• Norm Propagation: The second step of the

algorithm is to propagate norms from more general

contexts (or entities) to more specific contexts (or

entities). This process uses the WordNet information

in order to identify the relationships hierarchy and

part-whole between contexts and entities. In this step

new norms are created according to the relations

identified. This step is described in Section 5.1 and

5.2.

• Get related Norms: In the third step, the detection

mechanism gets each pair of norms and verifies

whether the norms are addressed to the same entity,

to the same context and if their period of validity

(activation and deactivation conditions) intersect.

This step is described in Algorithm 1. After that, the

mechanism verifies if there is a direct conflict.

Otherwise, the mechanism uses the WordNet

database to verify if there is an indirect conflict (see

Algorithm 2).

Available at https://goo.gl/KeF5M3

• WordNet Conflict Rules: In the fourth step, the

conflict rules of the WordNet relationships (synonymy,

specialization, antonymy, entailment) are applied in

order to detect conflicts. These rules were specified in

Section 5. The algorithm analyzes the deontic concept

of the norms and searches for relationships among the

actions or objects using WordNet database. The

algorithms 3, 4, 5, 6 detail the application of conflict

rules. In our implementation available to download, the

program informs to the user whether each pair of

compared norms are in conflict or not and presents the

reason of the conflict. For the sake of simplicity,

Algorithm 1 only informs to the user whether there is a

normative conflict or not.

So, in summary, the conflict checker algorithm

uses the WordNet database in order to: (i) unify the

syntax of the elements defined in the set of norms;

(ii) propagate norms addressed to more general

entities (or contexts) to more specific entities (or

contexts); and (iii) search relationships between the

actions defined in the set of norms to detect possible

indirect normative conflicts.

Algorithm 1: Check Conflicts.

Require: N: set of norms, E: set of entities, C: set of contexts,

W: WordNet database

function EXECUTE (N,

)

conflictTime

← false

conflictAction

← false

synonymEntities

← GETSYNONYMENTITIES (E, W)

synonymContexts

← GETSYNONYMCONTEXTS (C, W)

← PROPAGATEENTITY (N, synonymEntities, E, W)

← PROPAGATECONTEXT (N, synonymContexts,

synonymEntities, E, W)

for all n1

∈ N do

for all n2 ∈ N do

conflictTime

← timeIntersect (n1, n2)

if ((conflictTime = true) ∧

(n1

e = n2

e) ∧

(n1

c = n2.c)) then

conflictAction

← CHECKACTION (n1, n2, W)

return conflictAction

Algorithm 2: Check Action.

Require: n1, n2: norms, W: WordNet database

function CHECKACTION (n1, n2, W)

if ((n1.ac

= n2.ac

) ∧ (n1.ob

= n2.ob

) ∧

(n1.

= “FORBIDDEN”) ∧

(n2.

= (“OBLIGED” ∨ “PERMITTED”))) then

return true

if (SYNONYMY

CR (n1, n2, W) ∨

ANTONYMY

CR (n1, n2, W) ∨

SPECIALIZATION

CR (n1, n2, W) ∨

ENTAILMENT

CR (n1, n2, W) then

return true

return false

Identifying Indirect Normative Conﬂicts using the WordNet Database

191

Algorithm 3: Synonymy Conflict Rule.

Require: n1, n2: norms, W: WordNet database

function SYNONYMYCR (n1, n2, W)

if (SYNONYMY (n1.ac

, n2.ac

) ∧

((n1.obj = n2.obj) ∨ SYNONYMY (n1.obj, n2.obj, W)) ∧

(n1.d = “FORBIDDEN”) ∧

(n2.

= (“OBLIGED” ∨ “PERMITTED”))) then

return true

return false

Algorithm 4: Antonymy Conflict Rule.

Require: n1, n2: norms, W: WordNet database

function ANTONYMYCR (n1, n2, W)

if (ANTONYMY (n1.act, n2.ac

) ∧

((n1.obj = n2.obj) ∨ SYNONYMY (n1.obj, n2.obj, W)) ∧

(n1.d = ”OBLIGED”) ∧ (n2.d = “OBLIGED”)) then

return true

return false

Algorithm 5: Specialization Conflict Rule.

Require: n1, n2: norms, W: WordNet database

function SPECIALIZATIONCR (n1, n2, W)

if (

TROPONYMHYPERNYM (n2.act, n1.act, W) ∧

((n1.obj = n2.obj) ∨ SYNONYMY (n1.obj, n2.obj, W)) ∧

(n1.d = “FORBIDDEN”) ∧

(n2.d = (“OBLIGED” ∨ “PERMITTED”))) then

return true

return false

Algorithm 6: Entailment Conflict Rule.

Require: n1, n2: norms, W: WordNet database

function ENTAILMENT

CR (n1, n2, W)

if (ENTAILMENT (n1.act, n2.act, W) ∧

((n1.obj = n2.obj) ∨ SYNONYMY (n1.obj, n2.obj, W)) ∧

(n2.d = “FORBIDDEN”) ∧

(n1.d = (“OBLIGED” ∨ “PERMITTED”))) then

return true

return false

7 CONCLUSIONS

The detection of normative conflicts is an essential

key in a MAS governed by multiple norms. In the

literature, there are several approaches that deal with

conflicts among norms. The majority is only able to

detect direct conflicts, but other ones can also detect

indirect normative conflicts. The kinds of

relationships that can be identified are different

according to each approach. For the best of our

knowledge, all approaches described in the literature

only can detect indirect normative conflicts when the

designer explicitly defines relationships among

entities, contexts, actions and states. However, there

are relationships that are independent of the domain.

Our research focuses on the detection of indirect

normative conflicts using relationships already

defined in WordNet database to identify

relationships among elements of the norms. We

formalize the relationships explored and describe

rules to detect normative conflicts. Our approach can

detect relationships among the elements of the

norms without analyzing relationships of the domain

application. In order to improve our mechanism and

make it more complete, we are studying the

possibility of extending our approach to capture

others relationships. We are in process of integrating

detection methods that analyze domain application

relationships.

REFERENCES

Aphale, M. S., Norman, T. J., and Sensoy, M. (2012).

Goal directed conflict resolution and policy

refinement. In 14th International Workshop on

Coordination, Organizations, Institutions and Norms

in Agent Systems, Valencia, Spain. (COIN@AAMAS

2012).

Cholvy, L. and Cuppens, F. (1995). Solving normative

conflicts by merging roles. In Proceedings of the 5th

international conference on Artifitial intelligence and

law, pages 201–209. ACM.

Cholvy, L. and Cuppens, F. (1998). Reasoning about

norms provided by conflicting regulations. Norms,

logics and information systems: new studies in deontic

logic and computer science, pages 247–264.

da Silva, V. and Zahn, J. (2014). Normative conflicts that

depend on the domain. In Balke, T., Dignum, F., van

Riemsdijk, M. B., and Chopra, A. K., editors,

Coordination, Organizations, Institutions, and Norms

in Agent Systems IX, volume 8386 of Lecture Notes

in Computer Science, pages 311–326. Springer

International Publishing.

da Silva, V., Braga, C., and Zahn, J. (2015). Indirect

normative conflict - conflict that depends on the

application domain. In ICEIS 2015 - Proceedings of

the 17th International Conference on Enterprise

Information Systems, Volume 1, Barcelona, Spain, 27-

30 April, 2015, pages 452–461.

dos Santos Neto, B., da Silva, V., and de Lucena, C.

(2012). An architectural model for autonomous

normative agents. In Barros, L., Finger, M., Pozo, A.,

Gimennez-Lugo, G., and Castilho, M., editors,

Advances in Artificial Intelligence - SBIA 2012,

Lecture Notes in Computer Science, pages 152–161.

Springer Berlin Heidelberg.

dos Santos Neto, B., da Silva, V., and de Lucena, C.

(2013). Developing goal-oriented normative agents:

The nbdi architecture. In Filipe, J. and Fred, A.,

editors, Agents and Artificial Intelligence, volume 271

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

192

of Communications in Computer and Information

Science, pages 176–191. Springer Berlin Heidelberg.

Fenech, S., Pace, G., and Schneider, G. (2009). Automatic

conflict detection on contracts. In Leucker, M. and

Morgan, C., editors, Theoretical Aspects of

Computing - ICTAC 2009, volume 5684 of Lecture

Notes in Computer Science, pages 200–214. Springer

Berlin Heidelberg.

Fenech, S., Pace, G. J., and Schneider, G. (2008).

Detection of conflicts in electronic contracts. NWPT

2008, page 34.

Gaertner, D., Garcia-Camino, A., Noriega, P., Rodriguez-

Aguilar, J.-A., and Vasconcelos, W. (2007).

Distributed norm management in regulated multiagent

systems. In Proceedings of the 6th international joint

conference on Autonomous agents and multiagent

systems, page 90. ACM.

Giannikis, G. K. and Daskalopulu, A. (2009). Normative

conflicts-patterns, detection and resolution. In

WEBIST, pages 527–532.

Giannikis, G. K. and Daskalopulu, A. (2011). Normative

conflicts in electronic contracts. Electronic Commerce

Research and Applications, 10(2):247–267.

Grossi, D., Gabbay, D., and van der Torre, L. (2010). The

norm implementation problem in normative multi-

agent systems. In Dastani, M., Hindriks, K. V., and

Meyer, J.-J. C., editors, Specification and Verification

of Multiagent Systems, pages 195–224. Springer US.

Gunay, A. and Yolum, P. (2013). Engineering conflict-

free multiagent systems. In First International

Workshop on Engineering Multiagent Systems

(EMAS).

Kollingbaum, M. and Norman, T. (2004). Strategies for

resolving norm conflict in practical reasoning. In

ECAI Workshop Coordination in Emergent Agent

Societies, volume 2004.

Kollingbaum, M. and Norman, T. (2006). Informed

deliberation during norm-governed practical

reasoning. In Boissier, O., Padget, J., Dignum, V.,

Lindemann, G., E. Matson, S. Ossowski, J. Sichman,

and J. Vázquez-Salceda, editors, Coordination,

Organizations, Institutions, and Norms in Multi-Agent

Systems, volume 3913 of Lecture Notes in Computer

Science, pages 183–197. Springer Berlin Heidelberg,

2006.

Kollingbaum, M., Norman, T., Preece, A., and Sleeman,

D. (2006). Norm refinement: Informing the re-

negotiation of contracts. In ECAI 2006 Workshop on

Coordination, Organization, Institutions and Norms in

Agent Systems, COIN@ ECAI, volume 2006, pages

46–51.

Kollingbaum, M., Norman, T., Preece, A., and Sleeman,

D. (2007). Norm conflicts and inconsistencies in

virtual organisations. In Noriega, P., Vázquez-Salceda,

J., Boella, G., Boissier, O., Dignum, V., Fornara, N.,

and Matson, E., editors, Coordination, Organizations,

Institutions, and Norms in Agent Systems II, volume

4386 of Lecture Notes in Computer Science, pages

245–258. Springer Berlin Heidelberg.

Li, T. (2013). Normative conflict detection and resolution

in cooperating institutions. In Proceedings of the

TwentyThird international joint conference on

Artificial Intelligence, pages 3231–3232. AAAI Press.

Li, T., Jiang, J., Aldewereld, H., De Vos, M., Dignum, V.,

and Padget, J. (2014). Contextualized institutions in

virtual organizations. In Balke, T., Dignum, F., van

Riemsdijk, M. B., and Chopra, A. K., editors,

Coordination, Organizations, Institutions, and Norms

in Agent Systems IX, volume 8386 of Lecture Notes

in Computer Science, pages 136–154. Springer

International Publishing.

Miller, G. A. (1995). Wordnet: a lexical database for

english. Communications of the ACM, 38(11):39–41.

Sensoy, M., Norman, T., Vasconcelos, W., and Sycara, K.

(2012). OWL-POLAR: A framework for semantic

policy representation and reasoning, volume 12.

Elsevier.

Smullyan, R. M. (1995). First-order logic. Courier

Corporation.

Vasconcelos, W., Kollingbaum, M., and Norman, T.

(2009). Normative conflict resolution in multi-agent

systems. Autonomous Agents and Multi-Agent

Systems, 19(2):124–152.

Vasconcelos, W., García-Camino, A., Gaertner, D.,

Rodríguez-Aguilar, J. A., and Noriega, P. (2012).

Distributed norm management for multi-agent

systems. Expert Systems with Applications,

39(5):5990–5999.

Vasconcelos, W. and Norman, T. (2009). Contract

formation through preemptive normative conflict

resolution. Leibniz-Zentrum fur Informatik.

Walenz, B. and Didion, J. (2011). Jwnl: Java wordnet

library.

Zahn, J. (2015). Um Mecanismo de Verificação de

Conflitos Normativos Indiretos. Master’s thesis,

Instituto de Computação - Universidade Federal

Fluminense (IC/UFF), Niterói, Brasil.

Identifying Indirect Normative Conﬂicts using the WordNet Database

193