Publishing and Linking WordNet using lemon and RDF

♣John M^cCrae, ♦Christiane Fellbaum, ♣Philipp Cimiano

♣Semantic Computing Group, University of Bielefeld. ♦Princeton University

What is WordNet?

WordNet is one of the oldest and most widely used lexico-semantic resources
Has been used for
- Word sense disambiguation
- Query expansion
- Entity recognition
- Event processing
- Sentiment analysis
- etc., etc.,

WordNet as a global project

WordNets have been developed for over 70 languages
Has been extended by
- FrameNet
- VerbNet
- OntoNotes
- etc., etc.,
It is difficult to link back to the original
More so as WordNet synsets are not stable cross-version!
- Actually, it is not clear which synsets map

WordNet-RDF

WordNet needs RDF
To be a stable center of the LLOD cloud
To enable easy interversion linking
To integrate with the Semantic Web community

Wait, hasn't this been done already?

Yes:
- W3C WordNet
- VU Wordnet 3.0
- As part of UBY, BabelNet, etc.
But:
- Not involved with development of WordNet
- Not hosted by data producer (Princeton)
- Not updated with new releases

So..., what is a (ontology-)lexicon?

A lexicon is a collection of lexical information
We do not need to define semantics within the lexicon
“An ontology-based semantic lexicon would leave the semantics to the ontology, focusing instead on providing domain-specific terms and object descriptions in the ontology.” (Buitelaar, 2010)

lemon Design goals

RDF(S)
Conciseness
Not prescriptive
- i.e., uses data categories
Semantics by reference
- i.e., uses ontologies
Extensible

Forms

lemon allows us to distinguish between different forms:

“edema” (singular) “edemata” (plural)

Hence, forms are also nodes:

Senses

lemon allows us to say something about why a particular word is used

“edema” (modern) “dropsy” (antiquated)

Hence we introduce a sense to describe the usage of a word with a given meaning

lemon and WordNet

lemon as the W3C Model for lexica should be a natural choice
WordNet strongly influenced the development of the Lexical Markup Framework (LMF)
LMF strongly influenced lemon
Thus, lemon fits closely to WordNet

WordNet as an ontology

But... lemon requires reference to an ontology
We treat the WordNet synset graph as the ontology
We do not assign the synsets OWL types
- Instead they are skos:Concepts

WordNet stable identifiers

Synsets:

http://wordnet-rdf.princeton.edu/wn31/103547513-n

Version: wn31
Part-of-speech: 1********-n
Offset identifier: 03547513

Words:

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

Version: wn31
Lemma: hotel
Part-of-speech: n

Wordnet data example

Data available as HTML+RDFa, RDF/XML, Turtle, N-Triples and JSON-LD

External data linking

Open Multilingual Wordnet
lemonUBY
VerbNet
W3C WordNet

Including translations

There are three ways of including translations in lemon

Including translations

By Reference
- Ontology-based semantics
By Sense Relationship
- Allows exact correspondence, e.g., "NATO"@en => "OTAN"@fr, "North Atlantic Treaty Organization"@en => "Organisation du traité de l'Atlantique Nord"@fr
- Explicit links
As datatype property
- No further annotation
- Easy to query

Links to other resources

lemonUBY
- To WordNet 3.0 data on lexical entries and synsets
VerbNet
- To frames on a sense level
W3C WordNet
- Links on a lexical entries and synsets
Your data?

Linking to previous version

URI schemes are also supported for previous version
- Implemented by 303 redirect

e.g.,

http://wordnet-rdf.princeton.edu/wn30/03542333-n

⇓

http://wordnet-rdf.princeton.edu/wn31/103547513-n

Summary

First Princeton WordNet version
Provides stable identifiers
- i.e., will support the interlingual index
Linked already to many resources
Includes extra data (e.g., translations) from other open licensed resources

These slides http://jmccrae.github.io/wn-rdf-paper/slides/slides.html (CC-BY)