multilingual ontologies – standards and technologies gerhard budin university of vienna chair,...

Multilingual Ontologies – Standards and Technologies

Gerhard BudinUniversity of Vienna

Chair, ISO/TC 37/SC 2

22nd APAN MeetingNUS, Singapore

20 July, 2006

Outline

• Problem description• Methods integration• Multi-layer data modeling• Multi-standard frameworks• Multiple representation languages• Interlinking and harmonization of standards and

specifications• Interoperability frameworks• Integration of tools• Examples from risk management ontology engineering

Problem Description1. There is (still) a communication gap between formalized

knowledge representations such as ontologies and users of information and communication systems, where such ontologies are used, also on user interfaces.

2. Although the Semantic Web has been designed primarily for machine-to-machine-communication, we need seamless natural language interaction workflows in (semantic) web services of any kind

3. While the Semantic Web is (still) essentially monolingual and the international lingua franca is English, there is a growing need for multilingual ontology resources as well as ontology-based translation services that overcome communication barriers arising from cultural-linguistic differences, lack of excellent command of English, need for high precision in communication, etc.

Need for integration of diverse methods• As expressed in standards and implemented in technologies, the

following “traditions” increasingly merge:– Ontology engineering standards, frameworks, technologies

• e.g. OWL (based on RDF), SKOS (also on RDF) (W3C), DOLCE/SUMO, description logic, frame logic, unified logic, annotation

• Types of ontologies (e.g. domain o., upper o., application o., task o.)• Editors such as Protégé, Altova, OntoEdit, div. merging/annotation tools

– Translation engineering standards• i.e. various paradigms in machine translation and computer-assisted

translation (language-based, statistical MT, Transl. Memories, patterns)– Terminology and language engineering standards (as the pre-requisite for

and interface between ontology and translation)• Terminology and lexical markup frameworks: TMF, LMF (ISO)• Markup languages such as TBX (language industry+ISO)• Lexical databases/ling.ontol: WordNet, Ontowordnet, EuroWordNet• Linguistic enrichment of ontologies (e.g. FrameNet)• Interaction mechanisms, translation of ontologies• Integration of multilingual ontologies in machine translation processes

Diversity and interoperability• Strong diversity of lexico-terminological resources

– Data models, data structures + data semantics– Diversity of semantic, linguistic/cultural complexity and semantic

depth/richness

• Diversity of user groups and their requirements• Sheer quantity of resources• Data interchange between organizations (within and

across domains) as well as (distributed) data integration – early needs asking for immediate solutions

• History of data modeling• History of interchange standards• History of semantic interoperability management

Need for multi-level modeling architectures

generic

interoperability

framework

terminological

interoperability

Developing the Terminology Markup Framework in order to cope with this complexity-diversity

• Based on empirical studies and practical user-driven requirements analysis

• Markup/representation/modeling: XML, XMLS, RDF, UML• Open standards strategy (ISO TC 37)

– ISO 12620 Data categories – meta-model element + semantics registry (RDF)

– ISO 16642 Terminology Markup Framework (TMF) – meta-model architecture and specifications (UML)

– ISO 12200 – Terminology Markup Language (XML)• Instance for language industry: TBX Termbase Exchange Format

(XML) • Instance for lexicography/publishing: LexML ISO 1951

– Lexical Markup Framework (LMF) (UML) – ISO 704 and ISO 1087 (foundational level)– ISO 15188 (workflow and collaborative issues)– Alignment with ISO 11179, W3C, OASIS, etc.

Introduction to TBX

• TBX® stands for TermBase eXchange• TBX is a Terminological Markup Framework (TMF)

markup language– TMF is an ISO standard (16642)

• TBX is consistent with ISO 12200 (MARTIF)• TBX is maintained by OSCAR (www.lisa.org)• The TBX specification is free• Serving portability of resources across proprietary

terminology management systems, as well as interoperability of application-specific resources

TBX structure

• A TBX file is an XML document• A TBX file consists of:

– A header that describes the file– A set of entries, one per concept in the termbase– For each concept, a set of terms, grouped by

language, that designate the concept• A terminological concept entry (termEntry)

– Can be multilingual– Can be monolingual

TBX and Other Standards

• (1) TBX and ISO 16642 (TMF)• (2) TBX and ISO 12620 (Data Categories)• (3) TBX and SKOS

1: TBX and ISO 16642

• TBX is a TML (Terminological Markup Language) of TMF (ISO 16642) (see Annex B)

• TBX maps to the TMF meta-model– A TBX file is a TDC (terminological data collection)– martifHeader provides GI (global information)– termEntry: TE (terminological entry)– langSet: LS (language section)– tig/ntig: TS (term section)

• A TMF DCS (Data Category Selection) in TBX is in XCS (eXtensible Constraint Specification) format

• TBX uses ISO 12200 for its XML style

TMF Metamodel

GlobalInformation

(GI)

ComplementaryInformation

(CI)

Term Section(s)(TS)

Term Component Section(s)

(TCS)

Language Section(s)(LS)

Terminological(Concept) Entry/Entries

(TE)

Terminological Data Collection (TDC)

TMF and lexical resources

• In general, a terminological resource is organized into concept entries, each of which includes one or more terms designating a particular concept

• In general, a lexical resource is organized into lexical entries, each of which includes one or more senses of a particular lexical item (a word or phrase)

• A concept entry containing multiple terms can be split into multiple lexical entries, one per term, and multiple lexical entries associated with the same concept can be combined into one concept entry

• Link to Lexical Markup Framework (LMF)

2: TBX and ISO 12620

• All data categories in the default TBX DCS are taken from ISO 12620

• ISO 12620 is organized as an online registry and serves as a meta-ontology for resource modeling and for resource interoperability

3: TBX and SKOS

• A typical concept entry will contain a subject field to specify the domain of the concept.

• However, the subject field is typically some kind of hierarchy that is flattened into a string within TBX

• SKOS makes it possible to represent the subject field hierarchy as a hierarchy and then create a link within TBX

Simple Knowledge Organization System (SKOS)

• “SKOS is an area of work developing specifications and standards to support the use of knowledge organisation systems (KOS) such as thesauri, classification schemes, subject heading lists, taxonomies, other types of controlled vocabulary, and perhaps also terminologies and glossaries, within the framework of the Semantic Web.”

- http://www.w3.org/2004/02/skos/ (Accessed on 3/17/06)

Sample SKOS• <skos:Concept rdf:about="#s71">• <skos:prefLabel>Food</skos:prefLabel>• <skos:narrower rdf:resource="#s81"/>• <skos:narrower rdf:resource="#s79"/>• </skos:Concept>

• <skos:Concept rdf:about="#s81">• <skos:prefLabel>Recipe Ingredient</skos:prefLabel>• <skos:broader rdf:resource="#s71"/>• </skos:Concept>

• <skos:Concept rdf:about="#s79">• <skos:prefLabel>Restaurant Menu Item</skos:prefLabel>• <skos:broader rdf:resource="#s71"/>• </skos:Concept>

Visual Representation of SKOS

Food

Recipe Ingredient Restaurant Menu Item Grocery Store Item Homemade Item

Appetizer Entree Salad Soup

GEvTerm Initiative

• The information previously used dealing with food has been taken from FooNaVar, a project of the GEvTerm Initiative.

• The GEvTerm Initiative is a terminological database that has committed to being fully TBX and SKOS compliant

C: Multilingual Thesaurus for Medieval Studies (MLTMS)

• “Imagine the ability to search across web-resources using your native modern european language and find appropriate primary and secondary sources in Latin, French, Italian, German, Spanish, English, etc., based upon the meaning rather than the form of the search term. Imagine having a tool that would enable you to search for a concept and be able to construct the forms it has taken historically as well as the ability to link outward for both evidence and argument. Imagine a tool that would enable you to study the slippage of concept which is beyond naming. Imagine having a tool that can deconstruct ontological orders asking for different kinds of readings.”http://www.mith2.umd.edu/thes/ (Accessed on 3/17/06)

Why did MLTSM use TBX?

• integration of terminological data from multiple sources;

• querying multiple termbases through a single user interface by passing data through a common intermediate format on a batch or dynamic basis;

• placing data on an FTP site for download by interested parties;

• peer review by colleagues of tentative entries

- http://www.mith2.umd.edu/thes/ytbx.html (Accessed on 3/17/06)

MLTSM Sample<termEntry id='eid-VocCod-211.01'> <descrip type='subjectField'>personnel</descrip> <descrip type='definition'>personne qui accomplit un travail copie ou d'&#x00E9;criture</descrip> <langSet xml:lang='fr'> <ntig> <termGrp> <term id='tid-voccod-211.01-fr1'>copiste</term> <termNote type='termType'>entryTerm</termNote> </termGrp> </ntig> <ntig> <termGrp> <term id='tid-voccod-211.01-fr3'>&#x00E9;crivain</term> <termNote type='termType'>synonym</termNote> </termGrp> </ntig> </langSet> <langSet xml:lang='en'> <ntig> <termGrp> <term id='tid-voccod-211.01-en1'>scribe</term> <termNote type='termType'>entryTerm</termNote> </termGrp> </ntig> </langSet></termEntry>

MLTSM Sample(Rendered with XSLT)

TBX HTML

• The last few slides have provided an example of rendering HTML from a TBX file. Here is a brief diagram of the process.

TBX XSLT HTML

Processed by Results in

D: Other Standards

• ISO 11179 and XCS, which defines a flavor of TBX, both provide a list of data element types

• XMDR• RDF• OWL• Topic Maps/XTM

E: Tasks for TBX

• Encourage translation technology vendors to implement TBX

• Revise the specification• Compare ISO 11179 to XCS• Render TBX in RDF -> TBR for TBX-SKOS

interoperability implementation• TBR -> OWL• TBX – TMX (translation memory exchange standard)• TBX in Machine Translation applications

Ontology EditorProtege11179 OWL Ontology

XMDR Prototype Architecture: Initial Implemented Modules

MetadataValidator (defer) schema-driven syntax checker

Authentication Service (defer)

MappingEngine (defer)

RegistryExternalInterface

Generalization Composition (tight ownership) Aggregation (loose ownership)

Jena, Xerces

Java

RetrievalIndex

FullTextIndex

Lucene

LogicBasedIndexJena, OWI KSRacer,Kowari

RegistryStore

WritableRegistryStore

Subversion

OWL, RDF & XML Schema used to specify XMDR as UML used for 11179 Edition 2

UML11179Metamodel

11179 Relational Schema

Relational Metadata

OWL XMDROntology &annotations XMDR’s

Relax NG Schema

XMDRXML Schema

RDF Spec

TRang

XML SchemaLanguage spec

XML Objects

Types &Cardinalities

What things go in own files?Which property direction stored?Sequential ordering of properties

Triples: binarylabeled relationships

XMDR XML schema provides a number of important benefits…

• Schema specifies what is required as well as what is legal

• Divides metadata into files conforming to XML schema

• Normalizes data (ala’ relational “one fact in one place”)

• Facilitates XSLT transformations by reducing degrees of freedom to a canonical encoding within the RDF standard

• Relax NG used to create and check XMDR-it schema

• RNG validator enforces many OWL ontology constraints

• TRang automatically translates into XML schema syntax

From texts and terminologies to ontologies

• Using the Risk scenario– Termbase

• Export XML• Domain Models – meta-models -> patterns

– Text corpus• Term extraction – comparative testing ProTerm, MultiTerm

Extract, MultiCorpora• Aligning with termbase• Convert to RDF

– Ontology import -> editor

• The MULTH-WIN Project as an example of methods integration:

Bornemisza

Terminological frame semantics

• INTERVENTION (ACTOR(S), ACTIVITIES/PHASES):• RISK DETECTING (PRE-EVENT)• - R-ASSESSMENT• - R-PERCEPTION (X is risk)• - EXPERIENCE (statistics, case studies)• - OBSERVATION (monitoring)• - METHOD • - SATELLITE• - PROGNOSES• - R-ANALYSIS• - R-FEATURES• - SITUATION/CONTEXT (danger/hazard)• - SIMULATION (course of events)• - PROBALISTIC METHODS (safety)• - RELIABILITY• - R-IDENTIFICATION (DAMAGE)• - R-SOURCE• - DAMAGE CAUSE• - VULNERABILITY (DAMAGE TARGET)• - SUSCEPTABILITY (capacity/people)

Terminological frame semantics

I. Pre-event B. Public awareness and planning, II. In-event: C. Events and response

afflux/Hochwasser durch AufstauBE [[TYPE=flood], [PLACE=], [TIME=]], HAVE [CAUSE [[ORIGIN=], [NIEDERSCHLAG [TYPE=]], [STAU

[TYPE= Aufstau]]], DAMAGE [TARGET=, SOURCE=, DEGREE=]], HAPPEN [STATES=, PROCESSES=]]backwater/RückstauBE [[TYPE=flood], [PLACE=], [TIME=]], HAVE [CAUSE [[ORIGIN=], [NIEDERSCHLAG [TYPE=]], [STAU

[TYPE= Rückstau]]], DAMAGE [TARGET=, SOURCE=, DEGREE=]], HAPPEN [STATES=, PROCESSES=]]

Relationship modeling

disaster

general

rain

hail

snow

type

origin

cause

percipitationman-made

natural

„Stau“

Aufstau`afflux` Rückstau

`backwater`

im Entwässerungssystem`drainage flood´

TBX-SKOS interoperability• Differences

– XML vs. RDF (-> TBX will be turned into TBR)– Inherent flexibility + ”open” data modeling for a large

variety of resources vs. traditional thesaurus data model as a default for a KOS (diff. scopes)

– TBX has documented use cases and mapping tools -> language industry standard

– Different semantics + vocabularies (12620 vs. thesaurus standard)

• Commonalities– Conceptual approach– W3C – Integrated applications in the future

• Vocabulary mapping (RDF)

GlobalInformation

(GI)

ComplementaryInformation

(CI)

Term Section(s)(TS)

TMF Metamodel

Term Component Section(s)

(TCS)

Language Section(s)(LS)

Terminological(Concept) Entry/Entries

(TE)

Terminological Data Collection (TDC)

GlobalInformation

(GI)

ComplementaryInformation

(CI)

Term Entry Level (Level 1)

Terminological(Concept) Entry/Entries

(TE)

Terminological Data Collection (TDC)

Concept-RelatedDat-cats

Subject Field Note

Definition

SourceID

Responsibility

Date

Transaction

Adminis-trative

Dat-catsNotes

ConceptSystemDatCats

Language Section Level (Level 2)

Language Section(s)(LS)

Language Section(s)(LS)Language Section(s)

(LS)(LS * n …)

Concept-RelatedDat-cats

NoteDefinition

SourceID

Responsibility

Date

Transaction

Language-RelatedDat-cats

NotesAdminis-

trativeDat-cats

xml:lang

Transfer-comment

Equivalence

ConceptSystem

Dat-cats

Terminological Entry

Term-Level 3

Language Section(s) (LS)

Term Section(s)(TS)

Term Section(s)(TS)

(TS * n …)

Definition

Term-relatedDatCats (TRD)

Term

Context

Note

SourceID

Responsibility

Date

Transaction

NotesConcept-RelatedDat-cats

Adminis-trative

DatCats

Term Section(s)(TS)

Transfer-comment

Transfer-comment

SKOS Vocabulary

• SKOS Core is a model for expressing the structure and content of concept schemes (thesauri, classification schemes, subject heading lists, taxonomies, terminologies, glossaries and other types of controlled vocabulary).

• The SKOS Core Vocabulary is an application of the Resource Description Framework (RDF), that can be used to express a concept scheme as an RDF graph. Using RDF allows data to be linked to and/or merged with other RDF data by semantic web applications.

SKOS Graphs

SKOS Graphs

RDF Representation of SKOS Graph

<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:skos="http://www.w3.org/2004/02/skos/core#" xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"> <skos:Collection> <rdfs:label>milk by source animal</rdfs:label> <skos:member rdf:resource="http://www.example.com/concepts#buffalomilk"/> <skos:member rdf:resource="http://www.example.com/concepts#cowmilk"/> <skos:member rdf:resource="http://www.example.com/concepts#goatmilk"/> <skos:member rdf:resource="http://www.example.com/concepts#sheepmilk"/> </skos:Collection> <skos:Concept rdf:about="http://www.example.com/concepts#buffalomilk"> <skos:prefLabel>buffalo milk</skos:prefLabel> </skos:Concept> <skos:Concept rdf:about="http://www.example.com/concepts#cowmilk"> <skos:prefLabel>cow milk</skos:prefLabel> </skos:Concept> <skos:Concept rdf:about="http://www.example.com/concepts#goatmilk"> <skos:prefLabel>goat milk</skos:prefLabel> </skos:Concept> <skos:Concept rdf:about="http://www.example.com/concepts#sheepmilk"> <skos:prefLabel>sheep milk</skos:prefLabel> </skos:Concept> </rdf:RDF>

Mapping TBX/12620 DatCatsto SKOS Vocabulary

• TBX data categories (data element concepts in the sense of ISO/IEC 11179-3) contain instantiations of information that are expressed in SKOS using SKOS core vocabulary.

• Interoperability (a cross-walk between the two standards) depends on mapping between the two systems

Terminological knowledge engineering framework

1. meta modeling level:Terminology Markup Framework (TMF) Lexical Markup Framework (LMF)

UML, RDF, XML

DC selection ISO 12620 in RDF is the meta-ontology DC selectionsubsets, value sets subsets, value sets

2. modeling levelTerminology Markup Languages, e.g. TBX Lexical Markup, e.g. LexML (ISO 1951)

XML

3. resource levelTerminological resources Lexical resources

Markup, Annotation, Alignment, Analysis, Term Extraction

4. workflow level

ISO 15188 and other workflow specifications govern resource management processes (logistics, organizational measures, maintenance, quality assurance, etc.)

Framework integrationontology translationengineering

engineeringframework framework

interoperableintegrative

multilingual applications

e.g. MULTH-WIN project

terminology and language engineering framework

Thank you for your attention

Acknowledgements:

Slides 9-27 together with Alan Melby, Sue Ellen WrightSlides 28-29 Bruce BargmeyerSlide 33 Entry from the WordNet databaseSlides 35-37 WIN project (Rothkegel)Slides 39-42 Flood Risk ProjectSlides 44 WIN, 45: ThesShow Legat/Stallbaumer46: GEMET, 47: Bandholtz, 48/49: Gangemi, 56-59: Miles/SKOS