From Event to Action: Accelerate Your Decision Making with Real-Time Automation
Representing Translations on the Semantic Web
1. Representing Translations on
the Semantic Web
Elena Montiel-Ponsoda, Jorge Gracia,
Guadalupe Aguado-de-Cea, Asunción Gómez-Pérez
Ontology Engineering Group (OEG)
Facultad de Informática
Universidad Politécnica de Madrid
http://www.oeg-upm.net
{emontiel, j
{ ti l jgracia, l
i lupe, asun}@fi.upm.es
}@fi
2. The (Multilingual) Web of Data
• We know that the Web is multilingual….
• Is the Web of Data also multilingual?
• Ell, B., Vrandecic, D., and Simperl, E. (2011). Labels in the Web
of Data
English: 44.72%
Most used
1 language specified: 2.2% language tags: German: 5.22 %
Nl languages specified: 0 7%
ifi d 0.7% French: 5.11%
F h 11%
2
3. The (Multilingual) Web of Data
data.bnf.fr – Bibliothèque national de France
GeoLinkedData.es – Spanish geospatial data
p g p
Rechtspraak.nl – Netherlands Council of the Judiciary
FAO geopolitical ontology – with labels in en, fr, es, ar, zh, ru, it
AGROVOC Linked Open Data – AGROVOC agricultural thesaurus
3
5. Our proposal
• To propose a representation mechanism for explicit
p p p p
translation relations between natural language
descriptions associated to ontology elements and data.
• To implement it as a metamodel in OWL offered as a
module of the lemon model, lexicon-ontology model to
account for the linguistic descriptions associated to
g p
ontologies and linked data.
5
6. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
6
7. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
7
9. SKOS
SKOS enables a simple form of multilingual labeling:
ifrs:FinancialAssets skos:prefLabel
p
“financial assets”@en
skos:prefLabel
“activos financieros”@es
What happens when we have more than one label per
language? Food and Agriculture Organization and FAO?
How can we create explicit links between labels?
Say that one is translation, acronym of the other?
9
13. Limitations
These solutions work! …
Th l ti k!
…but with some limitations
Rigid models
Simple translation relation insufficient for:
p
original vs. target label
type of translation relation
source of the translation
adequacy or reliability of translations
13
14. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
14
15. The lemon model
An RDF‐based ontology‐lexicon model for ontologies
An RDF‐based ontology‐lexicon model for ontologies
Main features:
• Semantics by reference
• Rich lexical and terminological description of ontology
elements
• Concise (i.e., trade off between complexity and
expressivity)
• Descriptive not prescriptive (i.e., uses data categories)
• Modular and extensible
15
16. The lemon model
But this is also quite
Not so much… remember its
complex, isn’t it?
modular nature
16
18. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
18
19. Typology of translation relations
Ontology Localization
Multilingual Ontology
(an ontololgy in which labels are
documented in multiple NLs)
but…
b t
Does a 1 to 1 correspondence between always exist?
19
20. Typology of translation relations
Types of domains
Internationalized Culturally
C
or standardized influenced
domains domains
Types of conceptualizations
Conceptualizations Conceptualizations that
shared among the represent mismatches
languages represented between cultures and
in the ontology languages
20
21. Literal vs. Cultural equivalence Translation
Ontology A Ontology B
(German) (English)
Concept A
p Concept B
p
Sparkasse German savings institution Savings bank
Literal translation Cultural equivalence
translation
21
22. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
22
23. lemon module for translations
Lexicon
language:String
entry
isSenseOf reference
LexicalEntry LexicalSense Ontology term
lexicalForm
sourceLexicalSense targetLexicalSense
Form Translation translationOrigin
Resource
representation:String confidenceLevel:double
LiteralTranslation CulturalEquivalenceTranslation
q
23
24. Example of literal translation
LEXICONEN
LexicalEntry LexicalSense
ONTOLOGY
“payment method”
p y
http://purl.org/goodrelations/v1#PaymentMethods
Translation
LexicalEntry LexicalSense
“medio de pago”
medio pago
LEXICONES
25. Example of literal translation
LEXICONEN
LexicalEntry LexicalSense
ONTOLOGY
“Cabinet of Spain”
p
http://dbpedia.org/page/Consejo_de_Ministros
LiteralTranslation
LexicalEntry LexicalSense
“Consejo de Ministros”
Consejo Ministros
LEXICONES
26. Example of cultural equivalence translation
LEXICONEN
ONTOLOGY
LexicalEntry LexicalSense
http://www.oegov.us/democracy/us/core/owl/usgov#CABINET
“Cabinet”
CulturalEquivalenceTranslation
http://dbpedia.org/page/Cabinet_of_Spain
LexicalEntry LexicalSense
“Consejo de Ministros”
Consejo Ministros
ONTOLOGY
LEXICONES
27. Outline
1. Current mechanisms for translation relations
2. lemon
3. Typology of translation relations
4.
4 Proposed lemon module for translations
Examples of use
5. Conclusions
27
28. Conclusions
Benefits of the approach:
Direct,
Direct explicit translations can be represented
Distinction between literal/culturally equivalent translation
Translation metadata can be accounted for
Moderate complexity
Expressivity of lemon model
Conceptual/lexical layers remain separate
Future work:
Test this with more real examples
Algorithms to distinguish literal/culturally equivalent
translations
28