linked data and language - cilc...

Linked Data and

Language

A. Gómez-Pérez

Universidad Politécnica de Madrid

[email protected]

Acknowledgements: Jorge Gracia, Victor Rodríguez Doncel, LIDER Consortium

members

Linked Data and Language © Asunción Gómez-Pérez 7th International Conference on Corpus Linguistic , 5-7 th March, 2015, Valladolid, Spain

License

• This work is licensed under the Creative Commons

Attribution – Non Commercial – Share Alike License

• You are free:

- to Share — to copy, distribute and transmit the work

- to Remix — to adapt the work

• Under the following conditions

- Attribution — You must attribute the work by inserting

• “[source http://www.oeg-upm.net/]” at the footer of each

reused slide

• a credits slide stating: “Linked Data and Language” by A.

Gómez-Pérez ”

- Non-commercial

- Share-Alike

2


Table of Content

• Linked Data

• Linguistic Linked Data

• Multilingual Linked Data

• The LIDER project

Linked data

Linguistic

Linked Data

Multilingual

Linked Data

www.lider-project.eu


Linked Data Foundations


LD domains in August 2014

Media

Geographic

Life Sciences

Publications Goverment

Social

Networking Cross-domains

User Generated

Content Linguistics


Foundations

Unique identifiers: URI identify or name a resource

RDF(S) models

El Quijote Cervantes Is creator of

Work Person Is creator of

Is a Is a

http://datos.bne.es/resource/XX1718747 http://datos.bne.es/resource/XX3383563

http://iflastandards.info/ns/fr/frbr/frbrer/C1005 http://iflastandards.info/ns/fr/frbr/frbrer/C1001

Equivalence links to other datasets Same As

http://viaf.org/viaf/17220427

Cervantes

Same As Same As

http://dbpedia.org/resource/Miguel_de_Cervantes

Cervantes

Data navigation

http://datos.bne.es/resource/XX1718747





The model (Ontology) and the data

9

Work

Idiom

translation

Year

Publication date

Library

Located at

Person

Is creator of

Has subject

El Quijote Cervantes

Is creator of

Catalán

translation

1960

Publication date

BNE

Located in

Has subject

Vida de Cervantes

birthPlace Place

birthPlace Alcalá de Henares

Ontology

Data


10

http://iflastandards.info/ns/fr/frbr/frbrer/C1001


translation

Año

Publication date

http://xmlns.com/foaf/0.1/Organization

Located in


Is creator of

Has subject

http://datos.bne.es/resource/XX3383563 http://datos.bne.es/resource/XX1718747

Es autor


translation

1960

Publication date

BNE

Located in

Has subject

http://datos.bne.es/resource/bimo0002045496

Vida de Miguel de Cervantes Saavedra

Don Quijote de la Mancha Cervantes Saavedra, Miguel de

Catalán

Ontology

Data

http://datos.bne.es/#

Language

work

Biblioteca

Person

http://geo.linkeddata.es/ontology/Municipio

birthPlace

http://geo.linkeddata.es/resource/Alcalá de Henares

birthPlace

Linked data is full of URIs


Linked Data without ontologies

http://www.server1.org/resource/Cervantes

http://www.server2.es/resource/Cervantes


http://d-nb.info/gnd/11851993X

http://geo.linkeddata.es/page/resource/Municipio/Cervantes

Same as

Same as

Same as

Same as

URI URI

URI URI

URI

914 296 093

276,4 km²

Phone

Size

1547

#People

1547

Date of Birth

Author

D. Quijote

Cervantes


Linked Data and ontologies

http://www.server1.org/resource/Cervantes

http://www.server2.es/resource/Cervantes


http://d-nb.info/gnd/11851993X

http://geo.linkeddata.es/page/resource/Municipio/Cervantes

Same as

Person rdf:type

rdf:type

Retaurant rdf:type

Street rdf:type

Municipality rdf:type

URI URI

URI URI

URI

1547

Date of Birth

Author

D. Quijote

Cervantes

(Person)


Linked Data allows uniform access

1. Agree on vocabularies for

describing metadata and domain

data

2. Unified and standardized language

for describing resources ( RDF(S))

3. Unified and standardized query

language (SPARQL)

4. Standardized non-proprietary APIs

5. Links to other resources


Uses of Linked Data

1. Programmers built

applications using

make queries in

SPARQL and get RDF

Culture

(@BNE)

Geograhical

(@IGN)

Metereological

(@AEMET)

Smart Cities 2. Citizens/Users access

LD through a user

interface (they do not

see RDF)

3. Machine – Machine

data exchange and

semantic

interoperability in RDF

http://aemet.linkeddata.es/

http://datos.bne.es/

http://map4rdf.linkeddata.es/latest/


The new Linked Data Ecosystem

Culture

(@BNE)

Geograhical

(@IGN)

Metereological

(@AEMET)

Smart Cities





Linguistic Linked Licensed Data

Linked data

Linguistic

Linked Data


The problem

Finding and reusing LR in third party applications is manual and time consuming


Lack of interoperability of Language resources

• Ecosystem of

- Open and Closed resources

- Silos of LRs

- Complementary resources

• Lexicon, Corpora, Dictionaries, Grammars, ….

- Heterogeneous formats

• E.g, for Lexicons: Lexinfo, LMF, LIR, Lemon, …

- Several repositories with different metadata and schemas

- Many APIs and services for querying

Discovery and reuse LR in third party applications is hard, manual and time consuming


Use cases for LR Discovery

• Language metadata content

- Give me bilingual dictionaries in

Spanish, German , that

accounts for grammatical

number and gender with

Creative Common licenses

• Language Resources content

- Give me all occurrences in

corpora of the token “bank”

disambiguated as the WorNet

synset http://wordnet-

rdf.princeton.edu/wn31/1084372

35-n

• Language Services

- Give me all RESTfull

services that can extract

terms from text in Latvian.

http://wordnet-rdf.princeton.edu/wn31/108437235-n







20

http://rae.es

Motivation


21 *Picture attribution: http://commons.wikimedia.org/wiki/User:Gugerell

http://es.wiktionary.org

http://rae.es

An example




http://rae.es

http://www.wikilengua.org/

index.php/Terminesp:red

An example




http://rae.es



http://www.wordreference.

com/sinonimos/

An example




http://rae.es



http://es.wikipedia.org


com/sinonimos/

An example



http://rae.es



http://es.wikipedia.org


com/sinonimos/

An example

“Red”

(computer

network)


*Picture attribution: http://commons.wikimedia.org/wiki/User:Gugerell

“Red”

Etimologiy Del latin “rete”

Gender: “f”

Definition.: “Conjunto de

ordenadores o de equipos

informáticos conectados entre

sí….”

“Red”

Sinonyms: “sistema”, “malla”,” distribución”

“Red”

Norm: UNE 21302-131

English: network

German: Netzwerk

“Red”

Pronunciation: [red]

Grammar category: sustantivo femenino

Singular: “red”

Plural: “redes”

“Red_de_computadores”

Category: redes informáticas

Image

Complementary

but not connected


LD allows linguistic data integration

Red

Phonetic form

Form

number singular

[RED]

Form

plural

[REDES]

Phonetic form

number

Red

Sense

written form

“red”

Sense

written form

“malla”

equivalent

Red

image

Red

Sense Sense

translation

es - en

written form

“red” “network”

written form

Red

written form

Form

gender

femenine

“red”



3LD Linguistic Linked Licensed Data

Language resources

such as:

- Lexica

- Corpora

- Dictionaries ..

NIF

NLP Interchange Format

Using RDF and standard data

models (vocabularies):

- Lexica

- Corpora

ODRL Open Digital Rights Language

Published along with

a machine-readable license.




Subset of LOD

Linguistic domain

Open License

Resources in RDF

Interconnected with

other LD resources

Requirements: Keep track of the License (open or closed) information

Keep track of the Provenance of the resource

Keep track of the use of the resource

www.lider-

project.eu


Linguistic Linked Licensed Data Evolution

Jan. 2013

2014

Sept. 2014


Sept. 2013

• 65 Resources • 82 Links • Unbalanced: mostly lexicons • Centralized: 30% of links to/from DBpedia


Linguistic Linked Licensed Data @ Nov 2014

LLOD Cloud in November 2014 • 103 Resources (+58%) • 165 Links (+101% increase) • More balanced (14 Corpora,

+367%) • Less Centralized: Babelnet, LexVo

and LexInfo new hubs

Criteria for inclusion: • Resolvable: URLs that resolve • RDF: resolve to RDF • 1000 Triples: self-explaining • Links: to one resource from the

cloud or other 50 links • Crawlable: get the whole

resource by crawling • Linguistic: data must be a

language resources • Registered: at CKAN




1. Agree on vocabularies for

describing

• LR metadata and content (Lemon-

Ontolex, NIF, …)

2. Unified and standardized language

for describing resources ( RDF(S))

3. Unified and standardized query

language (SPARQL)

4. Standardized non-proprietary APIs

5. Links to other resources


BabelNet



How do we represent license information?



…

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .

@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .

@prefix odrl: <http://www.w3.org/ns/odrl/2/> .

<http://example.com/nc-nored-nd-ff> a odrl:Policy ;

rdfs:label "NC-NoReD-ND-FF" ;

rdfs:comment "MetaShare NonCommercial, No Redistribution, No Derivatives, for a fee.

Perpetual, worldwide, allowing no redistribution of the original. "@en ;

rdfs:seeAlso <http://www.meta-net.eu/meta-share....pdf> ;

odrl:permission [ a odrl:Permission ;

odrl:action odrl:reproduce;

odrl:duty [ a odrl:Duty ;

odrl:action odrl:pay ;

odrl:target "XXX EUR"

]

] ;

odrl:prohibition [ a odrl:Prohibition ;

odrl:action odrl:commercialize, odrl:distribute, odrl:derive

] .

META-SHARE NonCommercial NoRedistribution NoDerivatives For-a-Fee Licence

See demo at http://conditional.linkeddata.es


The new Linked Data Ecosystem

Culture

(@BNE)

Geograhical

(@IGN)

Metereological

(@AEMET)

Smart Cities





Linked Data is multilingual

Linked data

Linguistic

Linked Data

Multilingual

Linked Data


LOD is dominated by the English Language

Some questions:

1. Distribution of natural languages across RDF

datasets?

2. Usage of language tags to indicate the natural

language of RDF tags?

1. Distribution of usage of language tags

2. Distribution of literals tagged as English vs other languages

3. Distribution of literals tagged in languages other than

English

42

2007 2009 2014


LOD is dominated by the English language

4. Evolution of top-10 languages (non Eglish)

0

10.000

20.000

30.000

40.000

50.000

60.000

70.000

es de zh fr it ru pl nl pt sv

jan2014 jan2015

1. Number of Monolingual and multilingual datasets

JAN 2014

JAN 2015

67%

33%

RDFliteralsEnglish

RDFliteralsotherthanEnglish

71%

29%

RDFliteralsEnglish

RDFliteralsotherthanEnglish

3. English tags versus other languages' tags

JAN 2014

JAN 2015

JAN 2014

7%

93%

RDFliteralswithlangtag

RDFliteralswithoutlangtag

9%

91%

RDFliteralswithlangtag

RDFliteralswithoutlangtag

2. Current usage of language tagging capabilities in RDF

JAN 2014

JAN 2015


Multilingualism and the Linked Data

How to represent multilingual Linked Data?

Traditional annotation properties for most cases

Richer models for more demanding applications

dbpedia:Miguel_de_Cervantes

rdfs:label "Miguel de Cervantes"@es .

"ミゲル・デ・セルバンテス"@ja .

"미겔 데 세르반테스"@ko .

# LEMON

isbd:T1001 lemon:isReferenceOf [lemon:isSenseOf :cartographic].

:cartographic a lemon:LexicalEntry;

lemon:form [lemon:writtenRep “cartográfico”@es;

isocat:grammaticalGender isocat:masculine];

lemon:form [lemon:writtenRep “cartográfica”@es;

isocat:grammaticalGender isocat:feminine].

isocat:grammaticalGender rdfs:subPropertyOf lemon:property.

Association of the vocabulary to an external lexicon model


The LIDER project


LIDER Vision

LOD-aware NLP services

Metadata as

LD

...

Language Resources (Lexicon, corpora, ...)

some of them are open other are closed

Linguistic LOD generation

(Metadata and Content)

Language resources

as LD

Producers

Multimedia and

Multilingual Content

Metadata

Generation

Consumers

Content

Analytics


The use of LOD for NLP in Content Analytics

• Which extensions to the LOD are needed to support a new generation of large-scale content analytics applications that will overcome language barriers. - identification of key NLP tasks

that require background knowledge

- Specification of a new generation of NLP services that are LOD-aware and can exploit LOD

- Licensed linguistic linked data (3LD)


Industry use cases

Technical

activities

Community building

networking

LD4LT W3C- CG

BP-MLOD W3C-CG

OntoLex W3C-CG

1. Surveys

2. Requirements

1. Use cases

2. Industry board

1. Vocabularies

2. Guidelines

3. Roadmap

4. Reference Architecture

http://www.w3.org/community/ld4lt/

http://www.w3.org/community/bpmlod/

http://www.w3.org/community/ontolex/


http://www.w3.org/community/ld4lt/

http://www.w3.org/community/bpmlod





Key Use Cases and Requirements

49



LIDER Reference Architecture proposed to LD4LT

LLD Publishing

Vocabularies Hosting

Metadata (author, language, structure,…)

Licensing (terms and conditions of use)

Provenance ( origin

and history of data)

Multilingual Data Multilingual Data: Terminologies, (Multimodal) Corpora; Bilingual Dictionaries; Para

llel Data; Translation Memories; Ontologies; Glossaries, Classification Schemas

LLD-aware Services

Scalability Streaming Interoperability

LLD Linking Service Composition

Discovery (indexing and aggregation services)

Benchmarking & Validation of datasets and services

Certification

Gu

idelin

es a

nd S

tandard

ization


Linguistic Linked Licensed Data life cycle

1. Clear methodologies,

methods and tools for

monolingual LD

generation and

publication

2. Initial guidelines for:

- including the language

dimension

- Multilingual LD

- Linguistic LD

Villazón-Terrazas, B.; Vilches. L.; Corcho, O.; Gómez-Pérez, A.

Methodological Guidelines for Publishing Government Linked

Data. In D. Wood, ed. Linking Government Data. Springer. (pp,

27-49). 2011

Specification

Modelling

Generation Publication

Exploitation

Linking


Best practices and guidelines (BPMLOD @ W3C)

1. Best practices for Multilingual Linked Data Publication (BPMLOD @ W3C)

- Practices for Naming (URIs)

- Practices for Dereferencing

- Practices for Textual Information

- Practices for Linking

- Practices for Language Identification

2. Guidelines for Linguistic Linked License Data

- Wordnets,

- Multilingual Lexicographic resources

- Bilingual Dictionaries

- Terminologies in TBX

- NIF-based NLP Web services

How many Linguistic

Resources are exposed in

RDF?



Models for metadata and data

1. Metadata (LD4LT@W3C)

- Definition of the metadata OWL

ontology @ LD4LT W3C group

• UPF’s Metashare model as starting point

• Expanded with data and process

PROVENANCE and LICENSE modules

• Backwards compatible with MS and

LREMap models

• In agreement with members of LD4LT

W3C group

- Exposure of Metashare, Clarin,

LREMap, datahub metadata as LD

- Metadata Linguistic Observatory

(linghub)

http://linghub.lider-project.eu/


Relation between W3C groups

Linked Data for Language

Technologies (LD4LT)

Best practices

(BPMLOD)

Ontology lexica (Ontolex)

Data on the Web Best Practices

Use Cases BP for LD in LT lemon

specification

BP for

using lemon

BP for Multlingual Data on the Web BP for Data on the Web

55



56

http://datathon.lider-project.eu/


Follow Linguistic LD for Language Technologies


twitter.com/multilingweb

Hashtag: #LiderEU

Join the community

www.w3c.org/community/ld4lt

http://datathon.lider-project.eu/


Thanks for your attention !

58

linked data and language - cilc...

Documents