From thesauri to ontologies: semantic standards for law

1
From thesauri to ontologies semantic standards
for law

• Daniela Tiscornia
• tiscornia_at_ittig.cnr.it

2
Index of arguments

• 1. Linguistic barriers
• 2. Language-dependent approaches
• - traditional tools (thesauri, keywords)
• - metadata
• - lexicons
• 3. Language-independent tools ontologies
• 4.The Lois database
• 5. Conclusions

3
Linguistic barriers

• hamper
• access to content for non expert users
• semantic interoperability in e-government
• cross-lingual Legal Information Searching
• commercial Exploitation of Public Sector
  Information

4
References

• Fellbaum C. (editor), WordNet An electronic
  lexical database, Cambridge, MA The MIT Press,
  1998, 305, downloadable from http//mitpress.mit.
  edu/book-home.tcl?isbn026206197X.
• Gangemi A., Guarino N., Masolo C., Oltramari, A.
  Sweetening WordNet with DOLCE, AI Magazine 24(3)
  Fall 2003, 13-24Legal Knowledge and Information
  Systems, Proceedings of JURIX Conferences,
  Amsterdam, IOS Press.
• Gangemi A., Sagri M.T., Tiscornia D., Metadata
  for Content Description in Legal Information,
  Workshop Legal Ontologies, ICAIL2003, Edinburgh.
  In press for Journal of Artificial Intelligence
  and Law, Kluwer.
• Hirst G., Ontology and the lexicon In Staab,
  Steffen and Studer, Rudi (editors) Handbook on
  Ontologies in Information Systems, Berlin
  Springer, 2003, p.14.

5
Limits of the language-based retrieval tools

• Terminology vs common language the Italian Code
  on Data Protection doesnt contain the term
  privacy
• Polisemy the Italian term ordine (order) has 4
  legal senses
• Cross lingual IR the Italian term diritto means
  right and law.

6
Thesauri

• Vertical (systematic) Thesauri (e.g.Eurovoc)
• mono hierarchic tree structure of terms
  interlinked with broad and basic relationship
  (BT,NT,RT) no distinction is made between terms
  and concepts, semantic specification of relations
  is missing.
• Horizontal thesauri (e.g. The Italgiure semantic
  area) unstructured collection of terms without
  any distinction among words, concepts, types,
  part of speech .

7
Semantic metadata

• Semantic metadata are expected to support search
  engines for legal information retrieval,
  providing legal knowledge to include into their
  search strategies
• Conceptual search strategies based on keywords
  are still missing a clear semantics of terms, and
  this does not allow a conceptual query expansion
• there is no semantic relationship between
  information needs of the user and the information
  content of documents, apart from text pattern
  matching

8
Sense distinction

• From EU Legislation texts, four senses of
  'worker' are defined
• any worker as defined in Article 3 (a) of
  Directive 89/391/EEC who habitually uses display
  screen equipment as a significant part of his
  normal work.
• any person employed by an employer, including
  trainees and apprentices but excluding domestic
  servants
• any person carrying out an occupation on board a
  vessel, including trainees and apprentices, but
  excluding port pilots and shore personnel
  carrying out work on board a vessel at the
  quayside
• any person who, in the Member State concerned, is
  protected as an employee under national
  employment law and in accordance with national
  practice
• The corresponding lexical entry is defined as
  follows
• a person who works at a specific occupation

9
From words to concepts

• A semantic theory requires an ontology of all the
  concepts or predicates expressed by the words of
  a language
• Concepts are organized in structure that
  represent knowledge about the world
• Lexicons map words to concepts words are
  lexicalizations of a concept a concept can be
  represented by many terms (words or phrases) in
  multiple languages one term can identify several
  concepts
• A lexicon it is a bridge between a language and
  the knowledge expressed in that language (Sowa
  2000), but it is still language dependent!

10
Describing concepts

• By Lexical and semantic relations (e.g. Wordnet)
• By semantic roles among predicates (verbs) and
  their arguments(FrameNet)
• By properties linked by formal relations
  (ontologies)

11
WordNets Family

• WordNet (WN)(freeware, American English)
  (Cognitive Science Laboratory Princeton
  University)
• EuroWordNet (EUW) (proprietary, European
  languages) (ILC Institute of Computational
  Linguistic-Pisa for Italian language)
• ItalWordNet (IWN)(Italian part of EWN)
  (IRST-ICT-Trento)
• Jur-(Ital)WordNet (JWN) (C.N.R. Project ITTIG
  Institute of Theory and Techniques for Legal
  Information, ILC, LOA Laboratory of Applied
  Ontology)

12
Frame Net (Fillmore 1997)

• FrameNet is a frame-semantic description of
  lexical items based on semantically tagged
  corpora. Semantic roles (case roles, thematic
  roles, theta roles) characterize the semantic
  relation that a predicate can have to its
  arguments
• Mapping between the syntactic constituents of a
  sentence and the frame semantic elements

13
Thematic roles

• Based on
• syntactic patterns subj verb object
• semantic patterns agent action (action or
  state associated with the verb the participants
  the roles of participants)
• ontological assumptions role as participant
  relation and roles as ontological classes
  (Guarino 2004)

14
Conceptual-oriented vscontext-sensible
representation

• The traditional 'standardisation oriented' and
  'concept centred' approach, where (ideally) only
  one term is assigned to a concept, has proved to
  fail in cross-lingual conceptualizations
• Termino-ontographers' need an intermediate
  structure of the dominion, to distinguish
  language-independent concepts and relations from
  concepts and relations which are not (Kerremans
  and Temmerman, 2004)

15
The importance of context

• It is necessary anchoring of term extraction,
  term definition and inter-term relation
  identification on the contexts of use
• In law, legislative definitions are contexts
  which have a prescriptive force. This fact
  influences the determination of the number of
  senses of terms, and the equivalence setting
  between legal concepts and lexical concepts

16
Lightweight ontologies

• Lexicons are considered lightweight ontologies,
  linguistic expansions of the description of a way
  of perceiving reality, with limited formal
  modelling.
• It is possible that a lexicon with a semantic
  hierarchy might serve as the basis for a useful
  ontology, and that an ontology may serve as a
  grounding for a lexicon. This is particularly the
  case in technical domains, in which vocabulary
  and ontology are more closely tied than in more
  general domains (Hirst 2003).

17
The proposed approach

• Define a shareable conceptual model based on a
  semantic structure (classes of concepts and of
  semantically constrained relations).
• Concepts in the model are lexicalized by a
  multilingual lexicon which provide a source of
  legal semantic metadata (e.g. The Lois data
  base),
• locally and dynamically incremented,
• integrated by existing resources,
• to support a semantically structured Google for
  Law.

18
The Lois project

• The Lois project (EDC 22161) aims at developing
  a multi-language legal thesaurus based on WordNet
  and EuroWordNet technology
• WordNets lexicons pertain to the class of
  computational lexicons that aim at making word
  content machine-understandable via the highly
  structured semantic representation of concepts.
  These are represented by synsets, a set of all
  the terms expressing the same conceptual area,
  linked by a semantic relation of meaning
  equivalence. A synset is a set of one or more
  uninflected word forms (lemmas) with the same
  part-of-speech that can be interchanged in a
  certain context.
• Cross-lingual equivalence relations are made
  explicit in the so-called Inter-Lingual-Index
  (ILI). The ILI is the superset of all concepts
  from all wordnets, and the concepts from
  indigenous wordnets are linked into one or more
  ILI records by means of equivalence relations.
• ILI is an unordered list of concepts, i.e., it
  does not have any internal structuring. The
  reason behind this is that we assume that each
  language imposes its own language specific
  structural constraints on the concepts.
  Therefore, any ordering of ILI concepts needs to
  be retrieved from knowledge bases that link into
  the ILI (or from ontological classification).

19
Lois Architecture
20
Multiple Levels in Legal Language

Philosophy of Law
Lexical Data Base
Judges discourse
EU-National Legal Concept
Legislators language
21
National legal WNs

• The Lexical Data Base conceptualizes general
  language entities pertaining to legal theory and
  legal dogmatics (structured according to the EWN
  methodology).
• The Legislative data base (EU-National Legal
  Concept) is populated by concepts defined in
  European and national legislations.

22
LEXDB Lexical DB

• Lexical legal concepts 1944 ILI records.
• first nucleus translated from the Italian
  JurWN739 synsets
• new concepts selected by legal expert provided by
  Universities of Vienna, Evora, Praha and
  Sheffield.

23
The Lexical Data Base

• synsets are linked by
• internal relations
• cross-lingual relations
• eq_synonym,eq_near_synonym,eq_has_hyperonym,
• eq-has_hyponym, etc.

Lexical relations (syn, antonym,
near-syn) Semantic relations (Hyper, Hypo, role,
instance, etc.)
24
EULX EU lexical concepts

• Terminology automatically extracted from EU texts
  which do not occur as explicitly defined
• The selection has been automated by analysing the
  English EU directives, and extracting salient
  terms, mapping them to WordNet and selecting
  only the ones with one legal meaning in WordNet
• This process has created an automatic import of
  terms with gloss, and a plug-in synonym relation
  into WordNet

25
EULG EU legal concepts

• Concepts from EU directives with explicit
  definition, obtained by a process of
  semi-automatic alignment of the EU directives in
  the different languages 2332 ILI records

26
NATLG National legal concepts

• Concepts defined in national legislation within
  the domain of consumer law or implementing the
  EU legislation in the the domain automatically
  or manually extracted.

27
EU-National Legal Concept
EU-National legal document
National Legislation
ID
Celex Def.s about 2478 concepts
Implemented_as
National Legal Concepts
eq_synonym ,eq_near_synonym,has_hyperonym
28
Kinds of equivalence in Lois

• 1. Between lexical concepts
• near-equivalence
• hypo/hyper-equivalence
• functional equivalence
• 2. Between legal concepts
• legal equivalence

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
Consumer Protection Law structuring the domain
(I)

• Lexical Def. ILI GLOSS - worker_1 a person who
  works at a specific occupation.
• EU Def.s
• 8.2005-02-02 worker_2 any person who, in the
  Member State concerned, is protected as an
  employee under national employment law and in
  accordance with national practice
• 23 2005-02-02 worker_3 any person carrying out
  an occupation on board a vessel, including
  trainees and apprentices, but excluding port
  pilots and shore personnel carrying out work on
  board a vessel at the quayside
• 22. 2005-02-02 worker 4 any person employed by
  an employer, including trainees and apprentices
  but excluding domestic servants21. 2005-02-02
  worker_5 any worker as defined in Article 3 (a)
  of Directive 89/391/EEC who habitually uses
  display screen equipment as a significant part of
  his normal work.

Has_hyper
Has_hyper
33
Consumer Protection Law structuring the domain
(II)
Implemented-as
EU concept device National concept medical
device National concept active implantable
medical device
Near-syn
Has_hyper
Has_hyper
Has_hyper
34
Consumer Protection Law structuring the domain
(III)
Core Ontology physical object social
object EU concept device National
concept medical device National concept
active implantable medical device
Has_hyper
Has_hyper
35
The Core Legal Ontologyas ordering principle

• Creating ILI records from WordNET high-level
  concepts.
• Creating ILI records from the upper concepts of
  the IT-LEXDB linked to the Core Legal Ontology
  (together with their LCO links), used as
  hypernyms in local hierarchies.
• Link WordNET high-level to CLO categories.

36
Why do we need a core legal ontology as ordering
principle

• Disadvantages
• Manually performed
• Limited improvement of searching capabilities
• Advantages
• Aid in harmonizing lexical concepts proposed by
  national legal experts and existing lexical
  resources
• Added value future use of the lexical resources
  in Semantic Tagging, Information Extraction,
  Ontologies building, Knowledge-Based Systems.

37
Dolce D S and the Core Legal Ontology (CLO)

• DOLCE (a Descriptive Ontology for Linguistic and
  Cognitive Engineering) is a foundational ontology
  (FO) developed originally in the EU WonderWeb
  project
• DOLCE, extended by means of the Description
  and Situation(DS). ontology, is suited to
  conceptualize domains (such as Law) that are
  mainly constituted by non Physical (Mental,
  Social) objects.
• A Description in Dolce DS is a social object,
  which represents a conceptualization.
  Differently from physical objects, social objects
  are dependent on some agentive physical object
  that is able to conceive them. Descriptions have
  typical components, called concepts. Concept is
  also a social object, which is defined by a
  description and can be used in other
  descriptions. Figures, or social individuals
  (either agentive or not) are other social
  objects, defined by descriptions. Typical
  agentive figures are societies, organizations,
  and in general all socially constructed persons.
  (Gangemi et al.2005)

38
Dolce

39
CLO Cathegories
In CLO a norm is a Legal Description which has
components such as a Task (the set of actions
the norm aims to regulate) legal roles (played
by legal subjects involved) and parameters, as
temporal and physical locations. Legal
Descriptions are satisifed by Situations
(Fattispecie) composed by entities pertaining to
real word (Legal Subjects as Persond, Bodies,
etc.) and by Behaviours performed by them.

40
'Translating' legal concept

• The Italian term contratto is, in terms of CLO
  concepts, a legal description, an information
  content and a physical object (the material
  support of the information content).
• A legal institution, for instance the Prime
  Minister, is a figure, created by norms, but it
  is also a social role.

41
Comparing WordNet High-level and CLO classes
WN
CLO Artificial Person
Artificial Person .
Social Figure Person
Social concept
Being2
Non-Physical Object Living thing1
Endurant
Object1
Entity Physical entity1
Entity

42
Comparing WordNet High-level and CLO classes
WN CLO Lease
Lease contract
contract written
agreement social description
agreement 1
social concept statement 1
non physical Endurant
message2
Endurant communication2
Entity
abstraction abstract
entity

43
Comparing WordNet High-level and CLO classes
WN CLO Consumer
Consumer User1
Social Role Person
Social Concept Being2
Non-Physical
Endurant Living thing1
Endurant Object1
Entity Physical
entity1 Entity

44
Conclusions (I) the importance of semantic
metadata

• Structural documentary standards (Legal XML) must
  be integrated with semantic ones for the
  description of content, to achieve a high level
  of semantic interoperability between sectors in
  order to
• improve communication between areas and services
  of the Public Administration
• make it possible for the user to access
  information and to make that information
  available for further use by other sections of
  the Public Administration
• develop easy-to-access tools to incorporate and
  organize the data the users themselves are asked
  to supply.

45
Conclusions(II) the semantic lexicon role

• A semantic lexicon for law should be a source of
  semantic metadata, shared by multilingual and
  multinational legal information systems.
• It needs to be based on a common conceptual model
  of legal and world knowledge
• The Lois project aims at defining a methodology
  to achieve this goal.

46
Conclusions(III) lesson learned

• One of the main methodological point to be faced
  is the harmonization between
• lexical and legislative concepts,
• linguistic and ontological levels
• domain and world entities
• and the integration between
• new and existing resources
• manual and semi-automatic procedures.

47
References

• Breuker, J. and Hoekstra, R. (2004) Epistemology
  and ontology in core ontologies exemplified by
  two core ontologies for law FOLaw and LRI-Core.
  In Coront-Wes Ekaw 2004.
• Gangemi, A., Sagri, M.-T., Tiscornia, D.,
  (2005), A Constructive Framework for Legal
  Ontologies . In Law and the Semantic Web (
  Benjamins, Casanovas, Breuker and Gangemi eds.)
  Springer Verlag, 2005.
• Gangemi, A., Guarino, N., Masolo, C., Oltramari,
  A., Schneider, L. (2002), Sweetening Ontologies
  with DOLCE. In proceedings of EKAW 2002.
• Hirst, G. (2004), Ontology and the Lexicon, in
  (Staab and Studer eds.)HAndbook on Ontologies,
  Springer, 2004.
• Kerremans K. and Temmerman R.(2004) Towards
  Multilingual, Termontological Support in Ontology
  Engineering. In Proceeding of Termino 2004 ,
  workshop on Terminology, (2004).
• Peters W., M. T. Sagri,Tiscornia D.,The
  Structuring of Legal Knowledge in LOIS, in
  Artificial Intelligence and Law Journal ,
  forthcoming.
• Vossen, P., Peters, W. and Díez-Orzas, P. (1997),
  The Multilingual design of the EuroWordNet
  Database, in Mahesh, K. (ed.), Ontologies and
  multilingual NLP, Proceedings of IJCAI-97
  workshop, Nagoya, Japan, August 23-29.