D2 - MIning and REasoning with Legal texts · 5.1.1 Ontological extension of the LegalRuleML Meta...
Transcript of D2 - MIning and REasoning with Legal texts · 5.1.1 Ontological extension of the LegalRuleML Meta...
D2.2
Computational ontologies for normative reasoning
Grant Agreement nº: 690974 Project Acronym: MIREL Project Title: MIning and REasoning with Legal texts Website: http://www.mirelproject.eu/ Contractual delivery date: 31/12/2017 Actual delivery date: 31/12/2017 Contributing WP WP2 Dissemination level: Public Deliverable leader: UNITO Contributors: UNITO, INRIA, UL
This project has received funding from the European Union’s Horizon 2020 research and innovation
programme under the Marie Skłodowska-Curie grant agreement No 690974
Ref. Ares(2017)6380665 - 28/12/2017
MIREL- 690974 Page 2 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Document History
Version Date Author Partner Description
0.1 4/12/2017 Luigi Di Caro UNITO Initial draft
1.0 31/12/2017 Luigi Di Caro UNITO Final Version
Contributors
Partner Name Role Contribution
UNITO Luigi Di Caro Editor, Author, Reviewer
Editor, author and reviewer of the deliverable
INRIA Serena Villata Author Author of the deliverable
UL Livio Robaldo Reviewer Reviewer of the deliverable
Disclaimer: The information in this document is provided “as is”, and no guarantee or warranty is
given that the information is fit for any particular purpose. MIREL consortium members shall have
no liability for damages of any kind including without limitation direct, special, indirect, or
consequential damages that may result from the use of these materials subject to any liability which
is mandatory due to applicable law.
MIREL- 690974 Page 3 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Table of Contents
Executive Summary ............................................................................................................................. 5
Section 1: Introduction ........................................................................................................................ 5
Section 2: Background ......................................................................................................................... 5
Section 3: Selected Ontologies and Feature Analysis ......................................................................... 6
3.1 LKIF ................................................................................................................................................ 7
3.1.1 Description ......................................................................................................................... 7
3.1.2 Analysis of the features ...................................................................................................... 8
3.1.3 Considerations .................................................................................................................. 12
3.2 ODRL ............................................................................................................................................ 12
3.2.1 Description ....................................................................................................................... 12
3.2.2 Analysis of the features .................................................................................................... 13
3.2.3 Considerations .................................................................................................................. 15
3.3: L4LOD ......................................................................................................................................... 17
3.3.1 Description ....................................................................................................................... 17
3.3.2 Analysis of the features .................................................................................................... 17
3.3.3 Considerations .................................................................................................................. 18
3.4 LegalRuleML ................................................................................................................................ 19
3.4.1 Description ....................................................................................................................... 19
3.4.2 Feature Analysis ............................................................................................................... 19
3.4.3 Considerations .................................................................................................................. 19
3.5 CLO .............................................................................................................................................. 20
3.5.1 Description ....................................................................................................................... 20
3.5.2 Feature Analysis ............................................................................................................... 20
3.5.3 Considerations .................................................................................................................. 21
Section 4: Features Analysis of legal ontologies ............................................................................... 21
4.1 Types of features for legal ontologies ......................................................................................... 21
4.2. Features classification ................................................................................................................ 22
Section 5: MIREL-specific achievements ........................................................................................... 25
5.1 Normative Requirements as Linked Data.................................................................................... 25
5.1.1 Ontological extension of the LegalRuleML Meta Model .................................................. 25
MIREL- 690974 Page 4 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
5.1.3 Core primitives ................................................................................................................. 26
5.1.3 Formalization .................................................................................................................... 27
5.1.4 Requirements as Linked Data ........................................................................................... 29
5.1.5 State of affairs as named graphs. ..................................................................................... 30
5.1.6 Deontic reasoning as SPARQL rules.................................................................................. 31
5.1.7 Proof of concept and experimentation ............................................................................ 33
5.2 European Legal Taxonomy Syllabus ............................................................................................ 33
5.2.1 Multi-linguality and Multi-jurisdictionality ...................................................................... 33
5.2.2 European Legal Taxonomy Syllabus - The ELTS Ontology Schema .................................. 35
Section 6: Conclusions and Research Challenges .............................................................................. 39
References ......................................................................................................................................... 39
MIREL- 690974 Page 5 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Executive Summary
The main objective of Deliverable 2.2 is to provide a description of the state of the art
regarding the main existing ontologies for enabling legal reasoning in the scope of the
MIREL objectives and current activities, together with the effort and the achievements
already reached within the project work.
In Work Package n.2 of the MIREL project, legal ontologies represent crucial building blocks
as they represent machine-readable conceptualisations of the domain allowing forms of
reasoning to be populated (manually, semi-automatically or automatically) and used in Work
Package n.3.
Introduction, background and related work on legal ontologies are the first sections of this
deliverable. Considerations on the key features of ontologies in the legal domain are
identified and illustrated. Then, MIREL-specific approaches for ontological representation
are presented. Finally conclusions and research challenges identified are discussed.
Section 1: Introduction
There is a large body of research and practice in building and reusing ontologies in the legal
domain. This research effort has been growing in importance in the last few years, due to the
rise of concepts such as Linked Open Data, and Legal Linked Open Data, while the growing
research on Semantic Web enabled reasoning mechanisms among data and resources.
In this document, legal ontologies will be examined in the first part of this deliverable. This
will be followed by an analysis of common and discriminative features of the existing
ontologies to be possibly used within the scope of the MIREL project. Finally, some effort is
then presented in terms of MIREL-specific solutions in specific legal contexts. Conclusions
and research questions end the deliverable.
Section 2: Background
Some parts of this section are shared with Deliverable 3.1.
An ontology can be defined as “A formal specification of a shared conceptualization of a
domain of interest”. Being formal means that the ontology should be machine-readable, so
as to allow for automatic processing. It must also be shared i.e., accepted by a community
of users. Typically an ontology is restricted to a given domain of interest and model concepts
MIREL- 690974 Page 6 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
and relations that are relevant to a particular application domain or a particular task. Using
formal axioms allows for reasoning, and typically a concept rather than a term based
representation is language independent.
Semantic representations that can be used in the legal domain are lexicons, thesauri or
lightweight ontologies such as taxonomies offering a simple representation based on lexical
terms but without complex reasoning capabilities. Sources are legal texts in natural language.
In legal language the meaning of terms in a legal concept often differs from that in everyday
language. In addition, common problems in Natural Language Processing (NLP) such as
polysemy of terms, also appear. By providing formal definitions of concepts a semantic
representation can be useful for tasks such as data access and information retrieval,
publication exchange interoperability and harmonization Interoperability in particular is very
important in case of multilingual corpora, and comparison/ integration of sources, especially
from different legal systems.
Structured vocabularies such as taxonomies and thesauri are lists of terms organized in
hierarchies of broader and narrower terms and also associated terms using related term
relations. The definition of terms is restricted to the relationships with other terms into the
taxonomy, without complex concept constructs and semantic constraints. Thus complex
reasoning tasks are not supported, but tasks such as document tagging and classification as
part of retrieval of information are. Controlled vocabularies, taxonomies and thesauri,
referred to as Knowledge Organization Systems (KOS), have been used in digital libraries
among others and a recent development is the introduction of standards for their
representation and exchange on the web.
Section 3: Selected Ontologies and Feature Analysis
In the following sections, a number of selected relevant ontologies for enabling legal
reasoning are presented and analysed. In particular, each resource is described in terms of a
short description, its features, together with some further discussion points. This knowledge
has been then reorganized and evaluated in order to promote a list of candidate features which
can be considered for further research on legal ontologies within the scope and actual
activities of the MIREL consortium.
MIREL- 690974 Page 7 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3.1 LKIF
3.1.1 Description
The LKIF-Core ontology was developed within the ESTRELLA project1 for defining basic
legal concepts. The LKIF core legal ontology2 “consists of 13 modules, each of which
describes a set of closely related concepts from both legal and commonsense domains”. Thus
the LKIF core ontology is a library of ontologies relevant for the legal domain. The most
abstract concepts are defined in modules: top3, place4, mereology5, time/spacetime6. Basic-
level concepts are distributed across four modules: process7, role8, action9 and expression10.
These modules are extended by three modules that form the legal ontology: legal action11,
legal role12 and norm13. In addition to these legal modules, two modules are provided that
cover the basic vocabulary of two frameworks: modification14 and rules 15 . Finally, the
modules of the abstract, basic and legal level are integrated in the LKIF Core ontology
module16. The two framework modules are accessible through the LKIF Extended ontology
module17 which also imports the LKIF Core module.
1 http://www.estrellaproject.org/ 2 http://www.estrellaproject.org/lkif-core/ 3 https://github.com/RinkeHoekstra/lkif-core/blob/master/lkif-top.owl 4 https://github.com/RinkeHoekstra/lkif-core/blob/master/relative-places.owl 5 https://github.com/RinkeHoekstra/lkif-core/blob/master/mereology.owl 6 https://github.com/RinkeHoekstra/lkif-core/blob/master/time.owl 7 https://github.com/RinkeHoekstra/lkif-core/blob/master/process.owl 8 https://github.com/RinkeHoekstra/lkif-core/blob/master/role.owl 9 https://github.com/RinkeHoekstra/lkif-core/blob/master/action.owl 10 https://github.com/RinkeHoekstra/lkif-core/blob/master/expression.owl 11 https://github.com/RinkeHoekstra/lkif-core/blob/master/legal-action.owl 12 https://github.com/RinkeHoekstra/lkif-core/blob/master/legal-role.owl 13 https://github.com/RinkeHoekstra/lkif-core/blob/master/norm.owl 14 https://github.com/RinkeHoekstra/lkif-core/blob/master/time-modification.owl 15 https://github.com/RinkeHoekstra/lkif-core/blob/master/lkif-rules.owl 16 https://github.com/RinkeHoekstra/lkif-core/blob/master/lkif-core.owl 17 https://github.com/RinkeHoekstra/lkif-core/blob/master/lkif-extended.owl
MIREL- 690974 Page 8 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Figure 3: Top concepts of LKIF-core ontology
3.1.2 Analysis of the features
An analysis of the features of the LKIF ontology is reported in the scheme below. As with
the next sections on other ontologies, features are organized along some main categories. In
this case, they are the following three:
1. Purpose-oriented features
2. Modeling features
3. Semantic features
PURPOSE
Role: understanding the domain (core ontology), but also interoperability
between existing legal knowledge systems (it is the general purpose of the
system inside which LKIF is located)
MIREL- 690974 Page 9 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
application: “translation of legal knowledge bases written in different
representation formats and formalisms”, provide a knowledge representation
formalism
granularity: core, but it counts several integrated modules
MODELING FEATURES
Methodologies of development: different. The methodology is elaborated starting
from the following works:
Hayest (1985) “Formal theories of common sense world”: instead of creating
the ontologies with a top-down approach, he suggests the use of (as much as
possible) independent clusters of interrelated concepts.
Uschold and King (1995) “Towards a methodology for building ontologies”:
they refer to the basic concepts and basic levels theory of Lakoff
mainly composed by 4 steps: (i) identify purpose and scope, (ii) ontology
capture and coding, (iii) integration with existing ontologies (iv) evaluation
construction: manual
Knowledge sources for terms extraction: LRI-Core, LLD, CLO
MIREL- 690974 Page 10 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
SEMANTIC FEATURES
The ontology contains three levels:
1. top level In this level the classes are taken from the LRI-core ontology, so they are
taken from an existing ontology This level models basic concepts which are independent from the legal
concepts, but which are essential to describe any legal account. The basic concepts modelled in LKIF are: - Mental_Concept - Physical_Concept: models the involuntary changes. A change is “a
difference between the situation before and after the change”. Into a Change we can distinguish three types of changes: Initiation Continuation Termination Processes: they occur according to a certain procedure - Abstract_Concept - Occurrence: models spatio-temporal aspects as: Relative_Place Abstract_Place Interval Moment Some spatial relations: cover, coincide Some temporal relations: before, after, during
2. intentional level: this level model an agent’s intelligent behaviour which must be governed,
predicted and explained by law. The main concepts contained in this level are: - Action - Agen - Role - Propositional_Attitudes Intention, Belief and Desire - Expressions
MIREL- 690974 Page 11 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3. legal level In this level there is a distinction between: - the norm: is a situation put in words Obligation and Prohibition are equivalent classes (relation of equivalence) Permission: is in a higher level than Obligation and Prohibition because it doesn’t prohibit nothing - the situation the norm
applies to (indicated in the ontology as Qualified) Disallowed Allowed
The figure below represent an overview of the ontology structure.
MIREL- 690974 Page 12 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3.1.3 Considerations
The legal module of LKIF has some differences from ODRL and L4LOD, later discussed.
Here are some points for further analysis and evaluation:
1. LKIF models the notion of equivalence between the Obligation and Prohibition
concepts, while in ODRL the equivalence relationship between Obligation and Duty
is not present. The absence of this equivalence in ODRL subtends a different view of
the way in which the duty and the obligations are intended: this is highlighted by the
remedy relation which links a prohibition to a duty. The directionality of the remedy
relation “expresses an agreed obligation that must be fulfilled in the case that the
Prohibition has been infringed”, but the contrary doesn’t hold: in ODRL an obligation
(duty) is not complementary to a prohibition
2. In ODRL the concepts of Permission is in a higher level
3. Differently from ODRL (and L4LOD), LKIF models the situations the norms
applies to. So the definition of a Permission/Prohibition/Obligation goes hand-in-
hand with the definition of the situation it permits/prohibits/obliges.
Even if the Prohibition and Obligation concepts are modelled as equivalent norms, the
same parallelism is not maintained in the modelling of the Qualified concepts. Moreover
the Disallowed situation is not modelled as a subclass of Allowed, as it happens in the
norm hierarchy.
3.2 ODRL
3.2.1 Description
The Open Digital Rights Language (ODRL) is a policy expression language that provides a flexible
and interoperable information model, vocabulary, and encoding mechanisms for representing
statements about the usage of content and services. The ODRL Information Model describes the
underlying concepts, entities, and relationships that form the foundational basis for the semantics of
the ODRL policies.
Policies are used to represent permitted and prohibited actions over a certain asset, as well as the
obligations required to be meet by stakeholders. In addition, policies may be limited by constraints
(e.g., temporal or spatial constraints) and duties (e.g. payments) may be imposed on permissions.
The ODRL Information Model defines the underlying semantic model for permission, prohibition,
and obligation statements describing content usage. The information model covers the core concepts,
entities and relationships that provide the foundational model for content usage statements. These
machine-readable policies may be linked directly with the content they are associated to with the aim
to allow consumers to easily retrieve this information.
MIREL- 690974 Page 13 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
The primary aim of the ODRL Information Model is to provide a standard description model and
format to express permission, prohibition, and obligation statements to be associated to content in
general. These statements are employed to describe the terms of use and reuse of resources. The
model should cover as many permission, prohibition, and obligation use cases as possible, while
keeping the policy modelling easy even when dealing with complex cases.
The ODRL Information Model is a single, consistent model that can be used by all interested parties.
A single method of fulfilling a use case is strongly preferred over multiple methods, unless there are
existing standards that need to be accommodated or there is a significant cost associated with using
only a single method. While the ODRL Information Model is built using Linked Data principles, the
design is intended to allow non-graph-based implementations.
3.2.2 Analysis of the features
An analysis of the features of the ODRL ontology is reported in the scheme below, organized in 4
categories:
1. Purpose-oriented features
2. Modeling features
3. Usability features
4. Semantic features
PURPOSE
Role: (flexible and interoperable information model, vocabulary, and encoding
mechanisms)
organize and structure information, semantics integration and interoperation
Application: representing statements about the usage of content and services
Granularity: domain
MODELING FEATURES
Language: English
Ground ontology: not present
Level of structure: high
Temporal clauses: can be expressed in the form of logical expressions
MIREL- 690974 Page 14 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
USABILITY FEATURES
Updates number and frequency: high
SEMANTIC FEATURES
taxonomical relations
abstract relations which are implemented by other relations
abstract concepts
support to the formulation of logical and boolean expressions
supported standards: json-ld, IRI (Internationalized Resource Identifier), DublicCode
for metadata
meronymy relations
ORDL defines a core vocabulary but it also allows some extensions
The scheme below presents an overview of the ODRL ontology.
MIREL- 690974 Page 15 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3.2.3 Considerations From the analysis of the resource, here are the points that can be considered as important for the
project:
allowed the presence of taxonomical (subClass)
allowed abstract relations (see: function [implemented by: assigner and assignee] and
relation [implemented by target])
allowed also abstract concepts, for example the concept Rule which is instantiated by
Permission, Duty and Prohibition.
Logic and Boolean expressions in the form od entity (see Contraint/LogicalContraint)
When policies are expressed with jslon-ld,the rows of the data model become fiel names
associated with the specific class from which the row starts
Meonymy relation (partOf) for the classes Party/PartyCollection and Asset/AssetCollection
Relations may have different cardinality values
The ODRL Information Model has the following classes:
Policy - A non-empty group of Permissions (via the permission property) and/or
Prohibitions (via the prohibition property) and/or Duties (via the obligation property). The
Policy class is the parent class to the Set, Offer, and Agreement subclasses:
o Set - a subclass of Policy that supports expressing generic Rules.
o Offer - a subclass of Policy that supports offerings of Rules from assigner Parties.
o Agreement - a subclass of Policy that supports granting of Rules from assigner to
assignee Parties.
Asset - A resource or a collection of resources that are the subject of a Rule (via the abstract
relation property). The Asset class is the parent class to:
o AssetCollection - a subclass of Asset that identifies a collection of resources.
Party - An entity or a collection of entities that undertake Roles in a Rule (via the abstract
function property). The Party class is the parent class to:
o PartyCollection - a subclass of Party that identifies a collection of entities.
Action - An operation on an Asset.
Rule - An abstract concept that represents the common characteristics of Permissions,
Prohibitions, and Duties.
o Permission - The ability to exercise an Action over an Asset. The Permission may
also have the duty property that expresses an agreed Action that must be exercised
(as a pre-condition to be granted the Permission).
o Prohibition - The inability to exercise an Action over an Asset.
o Duty - The obligation to exercise an Action.
Constraint/LogicalConstraint - A boolean/logical expression that refines an Action and
Party/Asset collection or the conditions applicable to a Rule.
The ODRL Information Model includes property relationships between the classes. Most are
explicitly named properties and some are abstract properties (specifically, relation, function, operand,
and failure). The abstract properties are generic parent properties that are designed to be represented
by child properties (sub-types) with explicit semantics.
For example, the two properties relation and function are designed to represent the conceptual relation
between the Rule and the Asset and Party classes.
MIREL- 690974 Page 16 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
The Policy class has the following properties:
A Policy must have one uid property value (of type IRI [rfc3987]) to identify the Policy.
A Policy must have at least one permission, prohibition, or obligation property values of
type Rule. (See the Permission, Prohibition, and Obligation sections for more details.)
A Policy may have none, one, or many profile property values (of type IRI [rfc3987]) to
identify the ODRL Profile that this Policy conforms to. (See the ODRL Profiles section for
more details.)
A Policy may have none, one, or many inheritFrom property values (of type IRI [rfc3987])
to identify the parent Policy from which this child Policy inherits from. (See the ODRL
Inheritance section for more details.)
A Policy may have none or one conflict property values (of type ConflictTerm) for Conflict
Strategy Preferences indicating how to handle Policy conflicts.(See the Policy Conflict
Strategy section for more details.)
An ODRL Policy must either:
Only use terms defined in the ODRL Core Vocabulary [odrl-vocab], or
Use an ODRL Profile that declares the supported vocabulary used by expressions in the
Policy.
In the former case, the profile property must not be used. In the latter case, the profile property must
be used to indicate the IRIs of the ODRL Profile(s).
An ODRL Policy may be subclassed to more precisely describe the context of use of the Policy that
may include additional constraints that ODRL processors must understand. Additional Policy
subclasses may be documented in the ODRL Common Vocabulary [odrl-vocab] or in ODRL Profiles.
Below, an example of use case.
MIREL- 690974 Page 17 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
On this example, we can make some considerations:
json-ld: Json format for Linked data
uid is inherited direttamente da Policy
permission: in line with the requirements of class permission. That is:
“A Permission must have one target property value of type Asset”
“action” instead is specified because an action is always done on a target / resource
3.3: L4LOD
3.3.1 Description
L4LOD (Licenses for Linked Open Data) is a lightweight vocabulary for expressing the
licensing terms in the Web of Data. The vocabulary is not intended to propose yet another
license, but it is intended to provide the basic means to define in a machine-readable format,
i.e., RDF, the existing licensing terms. The vocabulary does not provide an exhaustive set of
properties for licenses definition. Implementations are free to extend L4LOD to add further
elements.
3.3.2 Analysis of the features
An analysis of the features of the L4LOD ontology is reported in the scheme below, organized in 4
categories:
1. Purpose-oriented features
2. Modeling features
3. technical features
4. usability features
PURPOSE
Role: organize and give a structure to the information
Application: vocabulary to express the license terms
Granularity: domain
MODELING FEATURES
Language: English
MIREL- 690974 Page 18 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Knowledge sources for terms extraction
Ground ontology
Level of structure: low
Temporal clauses: not included
TECHNICAL FEATURES
Knowledge representation formalism: RDF
USABILITY FEATURES
Updates number and frequency: the updates stopped on 13 March 2013
Ontology accessibility/availability: CC-By-SA
There are some similarities between some classes in ODRL and L4LOD as outlined in the
following image:
3.3.3 Considerations
In both ODRL and L4LOD there is a root class (Policy and License) which is linked to the
classes Prohibition, Permission and Duty/Obligation through similar relations:
permission/permits, obligation/oblige, prohibition/prohibits. The difference is that ODRL
uses an abstract class to group the three classes. Both the resources have taxonomical
relations.
There is some weakness of L4LOD, among them the lack of a concept to model the exception
which is always present in all real licenses.
Below, an overview of the vocabulary references.
MIREL- 690974 Page 19 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3.4 LegalRuleML
3.4.1 Description
LegalRuleML takes inspiration from LKIF, particularly in terms of closely representing legal
knowledge and legal reasoning. LegalRuleML, derived from RuleML [7] encourages the
effective exchange and sharing of such semantic information between legal documents,
business rules, and software applications [6].
3.4.2 Feature Analysis
LegalRuleML has 2 modules:
1. Legal_metadata.xml: models the legal metadata concerning the legal rules
1. identification: information about the authors of the rules (because a norm can
have different interpretations equally legitimate under the legal point of view)
2. references: identification of the textual fragments involved in the rules modeled
3. sources: models the connection between textual fragments and rules (it is strictly
connected with references and allow the isomorphism requirement).
4. events: defines temporal events (without any sematic, which is given by
timesInfo)
5. timesInfo: provide semantic information about the events
6. rulesInfo: meta-information about the rule
7. hierarchy: ranging of the rules in the defeasibility logic
2. Legal_operators.xml: defines the legal operators (deontic operators and behaviours)
3.4.3 Considerations
LegalRuleML [6] has key features which do not appear in other ontologies such as LKIF:
MIREL- 690974 Page 20 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
temporal aspects are not linkable to any part of a rule with arbitrary granularity;
the temporal aspect is not managed with appropriate operators;
violation-reparation and, in general, behaviors are not richly modelled;
defeasibility is not linked to the temporal parameters;
suborder lists of atoms, where the order is determinant, are not possible.
3.5 CLO
3.5.1 Description
CLO organises juridical concepts and relations on the basis of formal properties defined in
DOLCE+. The development of the Core Legal Ontology (CLO) takes into account
methodologies proper of foundational ontologies [19], proposals in the field of legal
ontologies (e.g. [20], Breuker et al. in [21]), as well as a large literature on legal knowledge
representation and legal philosophy.
3.5.2 Feature Analysis
An analysis of the features of the CLO ontology is reported in the schemes below:
PURPOSE
Role: understanding the domain
Application: supports three kinds of legal tasks in the Civil Law countries (i)
conformity checking, legal advice and norm comparison
MODELING FEATURES
Language: English
Ground ontology: DOLCE+
Level of structure: high
MIREL- 690974 Page 21 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
3.5.3 Considerations
There exists a distinction between:
• legal description or conceptualization: which includes norms, regulations crime
types, etc.
• situations or legal facts or cases: legal state of affairs, non-legal state of affairs that
are relevant to the Law, and purely juridical state of affairs.
Then, a dependence between descriptions and situations is considered: a descriptions is a
reification of a theory that formalises the content of a norm. That is, every legal description
classifies and constrains a state of affair. This means that a description is satisfied by a
situation when at least some entity in the situation is classified by at least some concept in
the description. In CLO the satisfy (SAT) relations is implemented. Moreover, a legal case is
a reification of a state of affairs that is a logical model of the theory.
Section 4: Features Analysis of legal ontologies
The objective of this section is the interpretation and analysis of the existing legal ontologies
to identify common rather than task-specific features to be used and revisited within the
objectives of the MIREL project.
4.1 Types of features for legal ontologies
Candidate features have been extracted from existing ontologies, surveys, and research
articles. Here we present a list of the sources together with the extrapolated features.
Source 1 (S1): University of Loughborough survey18
scope/goal [also in S2]
standard methodologies of development (e.g., NEON, METHONTOLOGY)
naming and spelling consistency all over the ontology
the ontology is based on (or has reused) other ontologies
the language the ontology is built in (e.g. OWL) [also in S2,S3]
updates number and frequency
ontology accessibility/availability (user license)
18 https://lborobusiness.eu.qualtrics.com/jfe/form/SV_cN5scAnvfy8QLoF
MIREL- 690974 Page 22 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Source 2 (S2): “Types and roles of legal ontology” by Andre Valente [1]
role (chosen from: organize and structure information, reasoning and problem
solving, semantic indexing and search, semantic integration and interoperation,
understanding the domain) [also in S1]
application
type
o knowledge representation formalism: the language the ontology is built in
(RDF, ONTOLINGUA, DOLCE-DAML, OWL, KIF, PROTEGE) [also in
S1,S3]
o level of structure (i.e. number of relation)
type of reasoning (e.g. instance classification, rule base reasoning, class or frame
reasoner)
Source 3 (S3): “Theory and methodology in Legal Ontology Engineering: experiences and
future direction” by Casanovas, Sartor et al. in “Approaches to legal ontologies” [4]
applications (which seem the role of S2)
granularity (domain specific vs core)
degree of formality (highly axiomatized vs. lexical or language-oriented) [also in
S1, S2]
methodologies of development (top-down, bottom-up, middle-out)
knowledge sources for concept and term extraction (official legal sources vs. legal
expert interview and ethnographic work)
construction
language
4.2. Features classification In this section, the types of features are analyzed across the selected sources S1, S2 and S3
of the previous section. The table below presents an overview of this comparison,
highlighting common vs specific types of features.
Sources S1 S2 S3
Role (chosen from 5 types)
scope/goal x
role x
Application (specifies the scope of application inside the role) x x
MIREL- 690974 Page 23 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Granularity (core/domain) x
Methodologies of development
standard methodologies x
top-down, bottom-up, middle-out x
Construction (manual, semi-automated, automated) x
Language (the human language used, not the KRF)
Knowledge sources for terms extraction x
Ground ontology (the ontology is based on (or has reused) other
ontologies)
x
Knowledge representation formalism
knowledge representation formalism x
language the ontology is built x
degree of formality x
Level of structure (number of relations) x
Type of reasoning x
Temporal clauses
Updates’ number and frequency x
Ontology accessibility/availability (user licence) x
MIREL- 690974 Page 24 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
A proposal of classification of the features may be described by the following labels: purpose,
modeling, technical, usability, and semantic. The table below presents such classification.
PURPOSE-CENTERED features
Role
Application
Granularity
MODELING features
Methodologies of development
Construction
Language
Knowledge sources for terms extraction
Ground ontology
Level of structure
Temporal clauses
TECHNICAL features
Knowledge representation formalism
Type of reasoning
MIREL- 690974 Page 25 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
USABILITY features
Updates number and frequency
Ontology accessibility/availability
SEMANTIC features
Semantic relations (taxonomical, etc.)
Abstract concepts / relations
Section 5: MIREL-specific achievements
In this section, two MIREL-focused works on ontological representation are presented. The
first contribution has been published at the 30th international conference on Legal
Knowledge and Information Systems (JURIX) 2017 [5]. The second contribution has been
published in the Journal of Applied Ontologies [22].
5.1 Normative Requirements as Linked Data
The work is about a proposal of a proof of concept for the ontological representation of
normative requirements as Linked Data on the Web. More precisely, starting from
LegalRuleML, we presented an extension of this ontology to model normative requirements
and rules. An operational formalization of the deontic reasoning over these concepts on top
of the Semantic Web languages is then developed.
5.1.1 Ontological extension of the LegalRuleML Meta Model
In this section, a description of the competency questions that motivate an extension of the
LegalRuleML ontology, and then we detail the core concepts of our new legal ontology as
well as their formalization in OWL.
Among the many approaches to design an ontology [2], the writing of motivating scenarios
is a very usual initial step of specifications to capture problems that are not adequately
MIREL- 690974 Page 26 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
addressed by existing ontologies [3]. The motivating scenario for us here is to support the
annotation, detection and retrieval of normative
requirements and rules. We want to support users in information retrieval with the ability to
identify and reason on the different types of normative requirements and their statuses. This
would be possible through ontology population approaches, but the lack of an existing
ontology covering these aspects slows this process, as well as the further development of
more advanced applications in legal computer
science.
In a second step of ontology specification, a standard way to determine the scope of the
ontology is to extract from the scenarios the questions and answers it should be able to
support if it becomes part of knowledge-based system. These so-called competency questions
[3] place demands on the targeted ontology, and
they provide expressiveness requirements. The competency questions we target for this
ontology are:
What are the instances of a given type of requirements (and its sub-types), e.g.,
obligation?
Is a requirement violated by one or more states of affairs, and if so, which ones?
Is a given description of rules and states of affairs coherent?
What are the rules, documents and states of affairs linked to a requirement, and by
what relations?
5.1.3 Core primitives
To support the competency questions and relying on definitions from Legal-RuleML [9] and
deontic reasoning [10, 11], we identified a set of core primitives for an ontology capturing
the different aspects of normative requirements, and supporting the identification and
classification tasks. We called that ontology Normative Requirement Vocabulary (NRV),
and made it available and dereferenceable following the Linked Data principles. The
namespace is http://ns.inria.fr/nrv# with the preferred prefix nrv respectively submitted both
to LOV [8] and to http://prefix.cc.
The top class of the ontology is the Normative Requirement which is defined as the set of the
requirements implying, creating, or prescribing a norm. Then we have a number of upper
classes to capture different features of the requirements:
the classes Compensable Requirement, Non Compensable Requirement,
Compensated Requirement characterize the requirements with respect to their
relation to compensation.
the classes Violable requirement, Non Violable Requirement, Violated Requirement
and Compliant Requirement characterize the requirements with respect to their
relation to a Compliance or a Violation.
MIREL- 690974 Page 27 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
the other classes follow the same logic, and they distinguish requirements with respect
to their perdurance, persistence, co-occurance and preemptiveness.
Using these upper classes, we positioned and extended three primitives from the
LegalRuleML Meta Model (i.e., Prohibition, Permission, Obligation), each one inheriting
from the appropriate super classes we introduced. For instance, Permission inherits from Non
Violable Requirement and Non Compensable Requirement, while Obligation inherits from
Violable Requirement and Compensable Requirement. Specializations of these classes are
then used to introduce the notions of Achievement, Maintenance and Punctual. For the
complete list of classes and their definitions, we refer the reader to the online documentation
available at the namespace URL. These primitives and definitions provide the taxonomic
skeleton of our NRV ontology.
5.1.3 Formalization
In this section, we provide some formalization details (ontological commitment) and their
translation into OWL (computational commitment). We will use the TriG syntax [14] for
RDF, and the prefixes we use in the rest of this article are:
lrmlmm: http://docs.oasis-open.org/legalruleml/ns/v1.0/metamodel#
owl: http://www.w3.org/2002/07/owl#
rdf: http://www.w3.org/1999/02/22-rdf-syntax-ns#
rdfs: http://www.w3.org/2000/01/rdf-schema#
rulemm: http://docs.oasis-open.org/legalruleml/ns/v1.0/rule-metamodel#
xml: http://www.w3.org/XML/1998/namespace
xsd: http://www.w3.org/2001/XMLSchema#
nrv: http://ns.inria.fr/nrv#
nru: http://ns.inria.fr/nrv-inst#
We captured the disjointedness expressed in the upper classes representing exclusive
characteristics of normative requirements (compensable / non-compensable, violable / non-
violable, persistent / non persistent):
:NormativeRequirement a rdfs:Class ;
owl:disjointUnionOf ( :CompensableRequirement :NonCompensableRequirement )
;
owl:disjointUnionOf ( :ViolableRequirement :NonViolableRequirement ) ;
owl:disjointUnionOf ( :PersistentRequirement :NonPersistentRequirement ) .
We initially considered the disjointedness of a compliant requirement and a violated
requirement, however this disjointedness is not global but local to a state of affairs and
therefore it does not translate to a general disjointedness of classes, i.e., a requirement may
be violated by a state of affairs but compliant with an
MIREL- 690974 Page 28 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
other one at the same time. However, this led us to capture this issue as a property
disjointedness, since a requirement cannot be violated and be compliant with the same state
of affairs at the same time:
:hasCompliance a owl:ObjectProperty ; rdfs:label "has for compliance"@en ;
rdfs:domain :ViolableRequirement ; rdfs:range lrmlmm:Compliance ;
owl:propertyDisjointWith :hasViolation .
Obligations are an example of non-disjoint union between achievements and maintenances,
since a punctual requirement is both an achievement and a maintenance:
lrmlmm:Obligation a rdfs:Class ;
rdfs:subClassOf :ViolableRequirement ;
rdfs:subClassOf :CompensableRequirement ;
owl:unionOf ( :Achievement :Maintenance ) .
:Achievement a rdfs:Class ; rdfs:label "achievement"@en ;
owl:disjointUnionOf ( :PreemptiveAchievement :NonPreemptiveAchievement ) ;
owl:disjointUnionOf ( :PerdurantAchievement :NonPerdurantAchievement ) ;
rdfs:subClassOf lrmlmm:Obligation .
:Maintenance a rdfs:Class ; rdfs:label "maintenance"@en ;
rdfs:subClassOf lrmlmm:Obligation .
Violated and compensated requirements could be defined with restrictions on the properties
hasViolation and hasCompensation:
:ViolatedRequirement a rdfs:Class ;
rdfs:subClassOf :ViolableRequirement ;
owl:equivalentClass [ a owl:Restriction ;
owl:onProperty :hasViolation ;
owl:minCardinality 1 ] .
:CompensatedRequirement a rdfs:Class ;
rdfs:subClassOf :CompensableRequirement ;
owl:equivalentClass [ a owl:Restriction ;
owl:onProperty :hasCompensation ;
owl:minCardinality 1 ] .
Likewise we could be tempted to define a compliant requirement with the following two
restrictions:
1 :CompliantRequirement a rdfs:Class ; rdfs:label "compliant requirement"@en ;
2 rdfs:subClassOf :ViolableRequirement ;
3 owl:equivalentClass [ a owl:Restriction ;
4 owl:onProperty :hasCompliance ;
5 owl:minCardinality 1 ] .
MIREL- 690974 Page 29 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
6 owl:equivalentClass [ a owl:Restriction ;
7 owl:onProperty :hasViolation ;
8 owl:maxCardinality 0 ]
.
However we removed the second part (lines 6-8) of the restriction since it reintroduces a
disjunction between the compliant and violated requirement classes.
The notions of compliance and violation are not generally disjoint but only disjoint locally
to a state of affair, i.e., a normative requirement can be violated and compliant at the same
time but with respect to different states of affairs.
However, OWL definitions cannot rely on RDF 1.1 named graphs, which we will use for
representing states of affairs. Therefore we will need another mechanism to capture this kind
of constraints.
Because we used disjoint unions, the ontology is in OWL DL, i.e., SHOIN(D), and more
precisely, in the AL(U)C(H)RN family, i.e., AL attributive language, (U concept union), C
complex concept negation, (H role hierarchy), R limited complex role inclusion axioms,
reexivity, irreexivity, role disjointedness, and N
cardinality restrictions.
We decided to declare the signature of properties (e.g., hasViolation, hasCompensation) at
the ability level (e.g., violable requirement, compensable requirement), and not at the
effective status level (e.g., violated requirement, compensated requirement) because each
status will be local to a state of affairs.
Therefore, in the end, we avoided too strong restrictions and signatures. If we remove
cardinality restrictions, unions and disjointedness, the ontology becomes compatible with
OWL EL and OWL RL which could be interesting for implementations relying on rule-based
systems, especially when we consider the extensions
proposed in the following sections.
5.1.4 Requirements as Linked Data
Using the LegalRuleML Meta Model and the NRV ontology we can now start to represent
normative requirements as Linked Data. Let us introduce two examples.
The first one is a rule stating that according to Australian law one cannot drive over
90km/h:
<http://gov.au/driving-rule> a lrmlmm:Source ;
rdfs:label "driving rules in Australia"@en .
nru:LSS1 a lrmlmm:Sources ;
lrmlmm:hasLegalSource <http://gov.au/driving-rule> .
MIREL- 690974 Page 30 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
nru:LRD1 a lrmlmm:LegalRuleMLDocument ;
lrmlmm:hasLegalSources nru:LSS1 ;
lrmlmm:hasAlternatives [ lrmlmm:fromLegalSources nru:LSS1 ;
lrmlmm:hasAlternative nru:PS1 ] ;
lrmlmm:hasStatements nru:SS1 .
nru:SS1 a lrmlmm:Statements ;
lrmlmm:hasStatement nru:PS1 .
nru:PS1 a lrmlmm:PrescriptiveStatement, lrmlmm:Prohibition ;
rdfs:label "can't drive over 90km"@en .
The second example is a rule stating that employees of CSIRO must wear their badge when
they are inside CSIRO facilities:
<http://csiro.au/security-rule> a lrmlmm:Source ;
rdfs:label "security rules in CSIRO"@en .
nru:LSS2 a lrmlmm:Sources ;
lrmlmm:hasLegalSource <http://csiro.au/security-rule> .
nru:LRD2 a lrmlmm:LegalRuleMLDocument ;
lrmlmm:hasLegalSources nru:LSS2 ;
lrmlmm:hasAlternatives [ lrmlmm:fromLegalSources nru:LSS2 ;
lrmlmm:hasAlternative nru:PS2 ] ;
lrmlmm:hasStatements nru:SS2 .
nru:SS2 a lrmlmm:Statements ;
lrmlmm:hasStatement nru:PS2 .
nru:PS2 a lrmlmm:PrescriptiveStatement, lrmlmm:Obligation ;
rdfs:label "you must wear your badge inside CSIRO facilities"@en .
5.1.5 State of affairs as named graphs.
The ability to define contexts and group assertions was one of the main motivations for
having named graphs in RDF 1.1 [15]. The notion of state of affairs at the core of deontic
reasoning is naturally captured by named graphs where all the statements of each state of
affairs are encapsulated as RDF triples in a named graph, identifying that precise state of
affairs. We provide here four examples of states of affairs respecting (2 and 3) or breaking
(1 and 4) the rules of the normative statements described above. The core idea is to represent
each state of affairs as a named graph typed as a factual statement of LegalRuleML.
:StateOfAffairs1 a lrmlmm:FactualStatement .
GRAPH :StateOfAffairs1 { rdfs:label "Tom" ;
:Tom :activity [ a :Driving ;
:speed "100"^^xsd:integer ;
rdfs:label "driving at 100km/h"@en ] .
}
MIREL- 690974 Page 31 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
:StateOfAffairs2 a lrmlmm:FactualStatement .
GRAPH :StateOfAffairs2 {
:Jim :activity [ a :Driving ; rdfs:label "Jim" ;
:speed "90"^^xsd:integer ;
rdfs:label "driving at 90km/h"@en ] .
}
:StateOfAffairs3 a lrmlmm:FactualStatement .
GRAPH :StateOfAffairs3 { rdfs:label "Jane" ;
:Jane :location [ rdf:value :CSIRO ;
:start "2017-07-18T09:30:10+09:00"^^xsd:date ;
:end "2017-07-18T17:00:10+09:00"^^xsd:date ] ;
:badge [ rdf:value :CSIRO ;
:start "2017-07-18T09:30:10+09:00"^^xsd:date ;
:end "2017-07-18T17:00:10+09:00"^^xsd:date ] .
}
:StateOfAffairs4 a lrmlmm:FactualStatement .
GRAPH :StateOfAffairs4 { rdfs:label "Steve" ;
:Steve :location [ rdf:value :CSIRO ;
:start "2017-07-18T09:30:10+09:00"^^xsd:date ;
:end "2017-07-18T17:00:10+09:00"^^xsd:date ] ;
:badge [ rdf:value :CSIRO ;
:start "2017-07-18T10:30:10+09:00"^^xsd:date ;
:end "2017-07-18T17:00:10+09:00"^^xsd:date ] . }
5.1.6 Deontic reasoning as SPARQL rules
Since the notion of named graph that appeared with RDF 1.1 is absent from OWL 2 and its
constructors, we need to implement the reasoning on states of affairs by other means. The
SPARQL language is both a standard and a language able to manipulate named graphs so we
propose to use SPARQL rules.
In this section, we explore the coupling of OWL reasoning with SPARQL rules to formalize
and implement some deontic reasoning. Description Logics (DL) support reasoning on the
description of concepts and properties of a domain (terminological knowledge or T-Box) and
of their instances (assertional knowledge or
A-box). They are the basis of the Web Ontology Language (OWL). The classical inferences
supported by DL are instance checking, relation checking, subsumption checking, and
consistency checking [16]. While these inferences are useful to reason about deontic
knowledge (e.g., a compensable requirement must also be
a violable requirement), they do not cover all the inferences we want to support here in
particular deontic rules (e.g., a requirement is violated by a state of affairs if, during a specific
period of time, a given constraint does not hold). These rules rely on complex pattern
MIREL- 690974 Page 32 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
matching including, for instance, temporal interval comparison that go beyond OWL
expressiveness.
As a proof of concept, the following rules check the violation or compliance of the statements
made by the previous states of affairs. The core idea is to add to each named graph of each
state of affairs the deontic conclusions of the legal rules relevant to it. The following rules
update compliance and violation for the driving speed requirement:
# Driving rules
DELETE { graph ?g { nru:PS1 nrv:hasCompliance ?g } }
INSERT { graph ?g { nru:PS1 a nrv:ViolatedRequirement ;
nrv:hasViolation ?g } }
WHERE { graph ?g { ?a a :Driving ; :speed ?s . }
FILTER (?s>90) }
;
DELETE { graph ?g { nru:PS1 a nrv:ViolatedRequirement ;
nrv:hasViolation ?g } }
INSERT { graph ?g { nru:PS1 nrv:hasCompliance ?g } }
WHERE { graph ?g { ?a a :Driving ; :speed ?s . }
FILTER (?s<=90) }
The following rules update compliance and violation for the CSIRO badge requirement:
# Badges rules
INSERT { graph ?g { nru:PS2 a nrv:ViolatedRequirement ; nrv:hasViolation ?g }}
WHERE { graph ?g { ?x :location [ rdf:value ?o ; :start ?ls ; :end ?le ]
optional { ?x :badge [ rdf:value ?o ; :start ?bs ; :end ?be ] .
FILTER (?bs<=?ls && ?be>=?le) } }
FILTER ( ( ! bound (?bs)) ) }
;
INSERT { graph ?g { nru:PS2 nrv:hasCompliance ?g } }
WHERE { graph ?g { ?x :location [ rdf:value ?o ; :start ?ls ; :end ?le ]
?x :badge [ rdf:value ?o ; :start ?bs ; :end ?be ] . }
FILTER (?bs<=?ls && ?be>=?le) }
The following rules update compliance for the state of affairs after violations were checked:
# Housekeeping: compliance rules
INSERT { graph ?g {?n a nrv:CompliantRequirement } }
WHERE { ?g a lrmlmm:FactualStatement .
?n a nrv:ViolableRequirement .
graph ?g { ?n nrv:hasCompliance ?g }
minus { graph ?g { ?n nrv:hasViolation ?g } } }
;
MIREL- 690974 Page 33 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
DELETE { graph ?g {?n a nrv:CompliantRequirement } }
WHERE { ?g a lrmlmm:FactualStatement .
?n a nrv:ViolableRequirement .
graph ?g { ?n nrv:hasViolation ?g } }
5.1.7 Proof of concept and experimentation
To validate and experiment with the ontology, the Linked Data and the rules, we used two
established tools: the latest version of the Protege platform -17] and the reasoners it includes
were used to check the NRV OWL ontology which was found coherent and consistent; the
latest version of CORESE [18] was used to load the LegalRuleML and NRV ontologies, the
Linked Data about the rules and the states of affairs, and the SPARQL rules to draw the
conclusions for the two first states of affairs concerning speed limitation.
5.2 European Legal Taxonomy Syllabus
This contribution described a new concept of legal ontology together with an ontology
development tool, called European Legal Taxonomy Syllabus (ELTS). The tool is used to
model the legal terminology created by the Uniform Terminology project on EU consumer
protection law as an ontology.
5.2.1 Multi-linguality and Multi-jurisdictionality
Achieving shared conceptualisations of law is difficult in any legal system. The problem is
confounded in Europe, which is increasingly governed by multiple jurisdictions - European,
national and sometimes regional as well - and with many official languages. The last
amendment to Regulation No. 1 of 15 April 1958 recognises twenty-four languages as having
the status of official and working languages in European institutions. This poses a significant
challenge, since each of the twenty-eight Member States that make up the European
Community have their own cultural baggage that no one, let alone the Community legislature,
is able to escape. The Sapir-Whorf hypothesis (Hoijer, 1954) maintains that it is impossible
for a concept in one language to be imported wholesale into another language due to linguistic
relativity.
Nevertheless, the European Union steadfastly seeks to achieve ‘harmonisation’ of laws in
whole sectors of diverse legal disciplines. Harmonisation of EU law is a complicated matter.
For Regulations, the implementation is automatically ‘binding in its entirety and directly
applicable in all Member States’, while Directives are ‘binding, as to the result to be
achieved, upon each Member State to which it is addressed, but shall leave to the national
authorities the choice of form and methods. The procedure of creating European Union laws
in a multi-lingual environment has its own complications.
MIREL- 690974 Page 34 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
Although debates on legislation in the European Parliament and Council can take place in all
official languages with the assistance of interpreters, the working draft legislation under
discussion is usually only available in one language - English or French or occasionally
German. At the end of the drafting phase, a team of specialist legal translators translate the
text into the other official community languages, subject to consistency checks by the EU’s
General Translation Team.
A research group at the European Commission, aiming to consolidate existing EU law,
worked on
the ‘Principles of the Existing EC Private Law’ or ‘Acquis Principles’ (ACQP), which would
provide a common terminology as well as common principles to guide uniform
implementation and interpretation of European law. The Acquis Principles were sketched by
scholars in European Private Law from the socalled Acquis Communautaire, a collection of
the existing body of EU primary and secondary legislation as well as European Court of
Justice decisions. This glossary aims to minimise conceptual differences and semantic
ambiguity in the EU legal process, and is of great importance in the activities of translators
as well as legal professionals.
Whether the Acquis Principles can truly address the considerable challenges it seeks to
address is a moot question. For instance, it does not solve the problem of conceptual and
terminological misalignment altogether, since Directives need to be transposed into national
law using terms that
make sense within the national legal system. In fact, it is precisely this second level of
translation that causes most problems. Transposing a rule often means having to use and
adapt a different, and sometimes conflicting, lexical baggage to the traditional national one.
The incoherent mix of different cultures and traditions results in translations that are often
inaccurate or insufficiently precise from a legal point of view.
There are several possibilities in the transposition process, as depicted in Figure 1, where the
lines connecting concepts from the EU level to national ones represent the “implement”
relation. The figure below illustrates possible misalignments (1)-(5) between the European
MIREL- 690974 Page 35 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
and national levels. Details are published in [22].
5.2.2 European Legal Taxonomy Syllabus - The ELTS Ontology Schema
As stated in the Introduction, ELTS is composed of a legal ontology schema, a web-based
legal ontology tool conforming to the ontology schema, and a multi-lingual legal ontology
on European consumer law constructed with the tool. In this section, we present the legal
ontology schema by describing the motivations behind it resulting from the legal analysis in
Section 2 of [22].
In this section, we will present the online tool to construct and browse the ontology, showing
for each feature of the tool relevant examples from the consumer law ontology derived from
the terminology of the Uniform Terminology project we use as a benchmark to show the
feasibility and expressivity of our approach.
The European Legal Taxonomy Syllabus is an ontology framework designed to address the
issues raised in Section 2 of [22]. The most important insight from lawyers, which informed
our design, was that the meaning of a legal term depends on its context (jurisdiction, domain,
legislation, timeframe, interpretation).
We designed an ontology schema that aims to make these considerations explicit. Our system
attempts to model interpretations beyond the letter of the law as well as temporal evolution
of concepts in an intuitive way, allowing users to traverse different definitions and determine
which definitions are most relevant to their query.
From a pragmatic need to model European law and national transpositions, the ontology
framework
MIREL- 690974 Page 36 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
must be both multi-lingual, multi-jurisdictional and multi-level. This allows links to be made
between different national ontologies, so that users can find similar terms in other languages
and other jurisdictions, and compare their meaning.
The schema has been designed to support the definition of concepts on the basis of a
comparative law methodology. We chose to adopt a bottom-up approach to ontology
creation, i.e., to compare low-level concepts among different legal systems. In our view, a
comparative law methodology ensures a non-superficial understanding of legal terms.
Starting from low-level elements rather than
abstract or composite concepts fosters evidence-based conceptualisations and generally gives
rise to less disagreement among ontology contributors.
We also adopt the view of comparative law that legal concepts are influenced by formants
other than legislation, and ensure that the ontology should provide space for annotation and
citations of case law and doctrine.
The main purpose of the European Legal Taxonomy Syllabus tool is to support the work of
legal practitioners, scholars and translators in multi-lingual and multi-jurisdictional contexts
such as the European Union, to help share technical knowledge and analyse the law in all its
complexity. As a secondary aim, the system can be used to build automated tools, e.g., for
information retrieval and translation.
Since the ontology framework is primarily designed for human reference, it supports
lightweight rather than axiomatic ontologies. In the classification of Giunchiglia and
Zaihrayeu [23] we use (informal) lightweight ontologies. Designing a full-fledged ontology
(expressed, for example, in OWL-DL) is a difficult and error-prone task even for experienced
users.
The choice of building a lightweight ontology was motivated by the need to develop a more
user-friendly system, thereby enlarging the possible audience of contributors and users, and
at the same time, reducing the costs of building an ontology. It was also driven by the
consideration that many peculiarities of law, such as interpretation, penumbra, interaction
with social values, metaphors and dynamics are far from having commonly accepted
solutions in logic.
Our analysis of the legal domain led us to identify the following features in the ontology
schema to allow the representation of relevant information in the legal domain. Such features
are not always straightforward to represent using standard approaches to ontology design.
Terms and concepts: the varying and highly contextualised meaning of legal terms means
that there needs to be a structured way to allow more than one meaning for terms in a legal
ontology. ELTS separates terms and concepts, allowing terms to be mapped to different
concepts and to have concepts mapped to more than one term (in the same language, or in
different languages in the case of multi-lingual nations and of EU law). Terms can be either
MIREL- 690974 Page 37 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
single words or multi-words (cf. examples below). Therefore we have many-to-many
relations between terms and concepts, thereby allowing both synonymity and polysemy.
Since we are in a multi-lingual context, a term in our system is structured as the term itself
together with the jurisdiction identifier and the relevant language, in order to account for
multiple languages in the same jurisdiction. The idea of neatly separating the lexicon from
the conceptual level is of course not new and is the foundation of several models, including
the well- known Lemon lexicon model.
Sources: each concept is linked to its legal source(s), possibly more than one, since a concept
can arise from multiple parts of legislation and also from the interaction of legislation, case
law and doctrine. However, listing the sources is not enough. For the sake of clarity, concepts
are associated with a description in natural language. Nevertheless, it is important to identify
the legal sources, since they contain important information about scope and purpose.
Domains: in addition to the contextual information of legal sources, the concepts are classed
in domains, traversal with respect to the jurisdictions and levels, in order to organize
knowledge and improve search and browsing. Each concept can be associated with more than
one domain.
Multi-lingual and multi-jurisdictional: the multi-jurisdictional nature of the EU requires
not only a multi-lingual ontological framework associating concepts with terms in different
languages, but
also a multi-jurisdictional one. The ELTS schema involves separate ontologies for each
jurisdiction
whose concepts are in turn mapped to terms in relevant languages. Specific relations connect
concepts from the ontologies of different jurisdictions, which are separate from the relations
within the same ontologies. In particular, the relation “implement”, described in more detail
below, connects concepts in the EU ontology with concepts in the national ontologies.
Multi-level: besides being multi-lingual and multi-jurisdictional, the ELTS schema
distinguishes between the EU level and national levels: these constitute separate ontologies.
Note that the EU level contains a single ontology, where all concepts have associated terms
in the different languages of the Member States considered. This, however, is a
simplification, since it is possible that there are unwanted divergences among different
languages even at the EU level. Concepts in the EU level ontology can be associated with
terms in all the languages of the Member states. Concepts in the different ontologies at the
national level can be associated only with the terms of the languages of the nation they belong
to.
Ontological relations: due to the holistic nature of the law, legal concepts are better
understood in relation to others. Therefore within each ontology, the concepts are linked via
ontological relations such as “is-a”, “part-of” and by more specific legal relations such as
legal “purpose”, expressing the legal goal (e.g., “consumer protection”) that the legal system
aims to achieve with that concept. Relations among concepts may change over time (as stated
below).
Implementation relation: Concepts at the EU level can be connected to national level
concepts by an implementation relation, representing how the concept has been transposed
into one national legal system. Given the separation of terms and concepts, the term
MIREL- 690974 Page 38 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
associated with an EU level concept is not necessarily the same term used to express the same
concept in the implementing legislation at national level. The relation is many-to-many, since
a national level concept may be the fusion of more than one EU level concept and/or a
Member State may express an EU level concept in multiple ways.
Dynamic nature of meaning: the ontology schema must account for the fact that almost
every legislation brings new definitions of terms that effectively replace prior
conceptualisations. Therefore the ontology must represent the current legal situation, and yet
researchers may still need to refer to deprecated conceptualisations for historical purposes or
to trace the evolution of terms. This also raises the problem of what happens to the ontological
relations associated with the replaced concept, a sort of frame problem. Since we are dealing
with a semi-automated context, the proposed solution is that the new concept should inherit
all the relations of the replaced one and it is the responsibility of the knowledge engineer to
remove the outdated ones and possibly introduce new relations in accordance with
authoritative interpretations.
Interpretation: there is a tension in the law between the highly contextual character of
meaning, which leads to multiple meanings for one term, each one associated with specific
sources, and the need to systematise legal knowledge. The legislator can introduce in a new
legislation a new meaning for a term, whose utility can go beyond the context of that
legislation. Due to interpretation, the meaning of the term can be extended also to other
concepts denoted by the term in other contexts. The frequent merging of meanings assigned
to legal terms that takes place in legal reasoning or simultaneous transposition of multiple
directives means that more complex concepts can emerge which do not necessarily replace
contextual meanings in all situations. This situation cannot be simply modelled by “is-a”
relations, since the concepts resulting from the interpretation are neither necessarily more
general in meaning, nor necessarily the simple intersection of the more contextualized
meanings. Rather, what is generalized is the context of use of the concept. Moreover, the
original contextual meaning of a term must always be available to the user and not only the
merged one deriving from interpretation.
Conceptual drafts: the ontology must be able to accomodate the conceptualization also of
draft legislation, to compare the resulting “draft” ontology. Glossaries created to achieve
consistency in legal terminology, such as the CFR and ACQP above, may contain
conceptualisations that are yet to be accepted officially. ELTS allows the creation of
temporary legal ontologies whose concepts are linked to current legal ontologies until such
time as the old concepts are replaced when the draft
legislation becomes law.
MIREL- 690974 Page 39 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
The ETLS ontology UML schema is represented below. Details are published in [22].
Section 6: Conclusions and Research Challenges
In this deliverable, an overview of the background and the existing ontologies for legal
reasoning are examined and evaluated in terms of definitions and expressivity.
Then, a work of analysis on the common and specific features is presented, in order to trace
a line across domains, requirements and the representation of legal norms in conjunction with
an ontology based representation.
Finally, a specific work on an ontology extension is presented. This contribution has been
published at the 30th international conference on Legal Knowledge and Information Systems
(JURIX) 2017 and represents a first effort in the scope of the MIREL project on Work
Package 2, taking into account Work Packages 3 and 4 and their requirements.
References
[1] Valente, Andre. "Types and roles of legal ontologies." Law and the semantic web.
Springer Berlin Heidelberg, 2005. 65-76.
[2] F. Gandon, Ontology Engineering: a Survey and a Return on Experience, RR-4396,
INRIA. 2002.
[3] M. Uschold, M. Gruninger, Ontologies: Principles, methods and applications, The knowl-
edge engineering review, 11(2), 93{136, 1996, Cambridge University Press.
MIREL- 690974 Page 40 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
[4] Sartor, Giovanni, et al. "Approaches to Legal Ontologies: Theories, Domains,
Methodologies. Law." Governance and Technology series. Springer (2011).
[5] Gandon, Fabien, Guido Governatori, and Serena Villata. "Normative Requirements as
Linked Data." The 30th international conference on Legal Knowledge and Information
Systems (JURIX 2017). 2017.
[6] Athan, Tara, et al. "Oasis legalruleml." Proceedings of the Fourteenth International
Conference on Artificial Intelligence and Law. ACM, 2013.
[7] Boley, Harold, Adrian Paschke, and Omair Shafiq. "RuleML 1.0: the overarching
specification of web rules." International Workshop on Rules and Rule Markup Languages
for the Semantic Web. Springer, Berlin, Heidelberg, 2010.
[8] Vandenbussche, Pierre-Yves, et al. "Linked Open Vocabularies (LOV): a gateway to
reusable semantic vocabularies on the Web." Semantic Web 8.3 (2017): 437-452.
[9] Athan, Tara, et al., OASIS LegalRuleML, Proceedings of the Fourteenth International
Conference on Articial Intelligence and Law, ACM, 2013.
[10] M. Hashmi, G. Governatori, M.T. Wynn, Normative requirements for regulatory
compliance: An abstract formal framework. Omnes Information Systems Frontiers 18(3), pp.
429-455, 2016
[11] K. Efstratios, N. Bassiliades, G. Governatori, G. Antoniou, A Modal Defeasible
Reasoner
of Deontic Logic for the Semantic Web, International Journal on Semantic Web and
Information Systems, (IJSWIS) 7 (2011): 1, doi:10.4018/jswis.2011010102
[14] World Wide Web Consortium, RDF 1.1 TriG, W3C Recommendation, 25 February
2014.
[15] F. Gandon, O. Corby, Name That Graph or the need to provide a model and syntax
extension to specify the provenance of RDF graphs., W3C Workshop | RDF Next Steps, Jun
2010, Palo Alto, United States.
[16] F. Baader, D. Calvanese, D. L. McGuinness, D. Nardi, P. F. Patel-Schneider, The
Description Logic Handbook: Theory, Implementation, Applications. CUP, 2003.
[17] N.F. Noy, M. Sintek, S. Decker, M. Crub�ezy, R. W. Fergerson, and M. A. Musen,
Creating
semantic web contents with protege-2000, IEEE Intelligent Systems, 16.2 (2001): 60-71.
MIREL- 690974 Page 41 of 41 27/12/2017
D2.2
Computational ontologies for normative reasoning
[18] O. Corby, R. Dieng-Kuntz and C. Faron-Zucker, Querying the semantic web with corese
search engine, In Proc. of the 16th European Conf. on Articial Intelligence, pp. 705-709
(2004), IOS Press.
[19] Masolo C., Borgo S., Gangemi A, Guarino N, Oltramari A. The WonderWeb Library of
Foundational Ontologies”, IST 2001-33052 Wonder Web.
http://wonderweb.semanticweb.org/deliverables/documents/D18.pdf, 2003.
[20] Visser P., T. Bench Capon, Ontologies in the Design of Legal Knowledge Systems,
towards a Library of Legal Domain Ontologies, in Proceedings of Jurix 99, Leuven,
Belgique.
[21] Benjamins R., Casanovas P., Breuker J., Gangemi A. Law and the Semantic Web. Berlin,
Springer, 2004.
[22] Ajani, Gianmaria, et al. "The european legal taxonomy syllabus: a multi-lingual, multi-
level ontology framework to untangle the web of european legal terminology." Applied
Ontology (2017).
[23] Giunchiglia, F. and Zaihrayeu, I. (2009). Lightweight ontologies. In Encyclopedia of
Database Systems, pages 1613–1619. Springer, Berlin.