Problem

• Ontologies have become popular.
• Use several formalisms: RDFS, OWL, F-Logic, etc.
• Ontologies do not have much linguistic information.

:Cat a owl:Class ;
  rdfs:label "cat"@en ;
  rdfs:label "Katze"@de .

• What is the plural? Easy for English, not for German

Ontologies

Ontologies

Ontology labels

Inflection and Synonyms

SKOS-XL

• Similar to SKOS-XL
  • eXtended Labels for the Simple Knowledge Organization System
  • W3C Recommendation since 2009
• SKOS-XL does not allow multiple forms of the same label
  • No grouping of “edema” and “edemata”
• “We [TopQuadrant] have yet to hear a use case that cannot be supported by SKOS alone” (Who need SKOS-XL? Maybe no-one. (Polikoff, 2013))

Forms

Senses

The core of lemon

What is an (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)

Dictionaries as lexica

• In fact, a lexicon represents much of the information already found in a dictionary
• That is words, their forms and their meaning
• Must be machine-readable
• Take Wiktionary as an example

Wiktionary as a lexicon

Wiktionary as a lexicon

Wiktionary as a lexicon

Wiktionary as a lexicon

lemon's origins

Design goals

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

Why lemon: RDF(S)

Why lemon: Conciseness

Why lemon:
Semantics by Reference

Why lemon:
Modular and extensible

Modules

Design Patterns

Why design patterns?

• Simplify usage of lemon
• Factorize common patterns
• Cover 95% of entries
• Expressable as simple language

Using the compiler

The Design Patterns Language

• Header consisting of prefix declarations
• Any number of lexicons
  • With a URI
  • And a language
• Any number of patterns

Example

@prefix example: <http://www.example.org/example#> .

Lexicon(<http://www.example.org/lexicon>, "en",
  ClassNoun("test", example:Test)
)

Noun Patterns

• Name, a proper noun referring to an individual
  • Name("Microsoft", <http://www.microsoft.com/>)
• Class Noun, a noun referring to a class in the ontology
  • ClassNoun("test", dbpedia:Test)
    

Noun Patterns

• If there is no class we may refer to a particular property value
  • ObjectPropertyNoun("oenophile", ontology:loves, ontology:wine)
• Relational Noun, a noun with a prepositional phrase referring to a property in the ontology
  • RelationalNoun("father", dbpedia:father, propObj = PossessiveAdjunct)
  • We use PossessiveAdjunct to allow "X is father of Y" and "X is Y's father"

Verb Patterns

• State Verb, a verb indicating a single property in the ontology
  • StateVerb("know", foaf:knows)
• Consequence Verb, a verb causing a property in the ontology
  • ConsequenceVerb("die", dbpedia:deathDate, propObj = PrepositionalObject("on"))
• Event Verb, a verb describing an event with multiple roles

Verb Pattern Example

Adjective Patterns

• Intersective Adjectives, an adjective referring to a class
  • IntersectiveAdjective("Belgian", ontology:Belgian)
  • IntersectiveObjectPropertyAdjective( "Belgian", dbpedia-owl:nationality, dbpedia:Belgium)
• Relational Adjective, like relational nouns
• Scalar Adjectives
  • ScalarAdjective("big", [ontology:size > 5.0 for ontology:Building])

Modifiers

• Inflectional form, e.g.,
  • ClassNoun("Katze", onto:Cat) with plural "Katzen"
• Gender (nouns-only)
  • ClassNoun("Katze", onto:Cat) feminine (with ...)
• Register
  • StateVerb("knows", foaf:knows) neutralRegister

Arguments

• Indicated with propSubj and/or propObj
• Subject (subject of a verb only!)
• CopulativeArg (subject for nouns and adjectives!)
• DirectObject
• IndirectObject
• PrepositionalObject("for") (not normally used for "of"!)
• PossessiveAdjunct

Multiword Expressions

• Multiword terms can be specified using the following syntax
  • ClassNoun(["blood"/noun "vessel"/noun], dbpedia:BloodVessel)
• This gives a decomposition of a term

Task:

In your language model the following DBpedia properties:

• Person
• Surname
• GivenName
• profession = Historian
• Germany
• nationality = Germany
• stateOfOrigin
• relative
• predecessor
• spouse

Solution

@prefix dbo: <http://dbpedia.org/ontology/> .
@prefix dbr: <http://dbpedia.org/resource/> .
@prefix dbp: <http://dbpedia.org/property/> .

Lexicon(<>, "en",
  ClassNoun("person", dbo:Person),
  ClassNoun("surname", dbo:Surname),
  ClassNoun(["given"/noun "name"/noun], dbo:GivenName),
  ObjectPropertyNoun("historian", dbp:profession, dbr:Historian),
  Name("Germany", dbr:Germany),
  IntersectiveObjectPropertyAdjective("German", dbp:nationality, dbr:Germany),
  StateVerb("come", dbp:stateOfOrigin, propObj = PrepositionalObject("from")) 
    with imperfect "came" with past participle "come",
  RelationalAdjective("related", dbp:relative, relationalArg = PrepositionalObject("to")),
  StateVerb("preceed",dbp:predecessor, propSubj = DirectObject, propObj = Subject),
  ConsequenceVerb("marry",dbp:spouse),
  RelationalNoun("husband",dbp:spouse,propObj = PossessiveAdjunct)
)

Summary

• We introduced lemon
  • A model for adding lexical information to ontologies
  • A concise, modular, extensible model
• lemon adds the following key elements
  • Lexical entries
  • Forms
  • Senses
  • Reference to ontology classes
• Ontology-lexica build on existing lexicography (dictionaries, wordnets, etc.)

Summary

• Design patterns for lemon simplify most common usage
• Implemented as a standalone language
• Can also handle
  • Morphosyntactic annotations
  • Argument mapping
  • Multiword expressions