Introducing OWL

Web Ontology Language

Elias Ponvert <efp@uts.cc.utexas.edu>

Department of Linguistics The University of Texas at Austin

Motivation

• You want computers to reason effectively about the single greatest sport of all time, international soccer
• Why?
  • To program a robot that places bets for you
  • To keep your personalized news summary updated
    • and aware of fine-grained consequences of Uruguay tying Costa Rica in the qualifiers
  • As a medium for Web-Service enabled multi-lingual news wire services about match results
• What do you need computers to understand?
  • There are soccer teams (or sides)
  • Some soccer teams are national teams
  • National teams are associated with countries

Motivation (cont'd)

• What do you need computers to understand? (cont'd)
  • DaMarcus Beasley is a player for the US National Team
  • Soccer teams compete in matches
  • The FIFA World Cup consists of matches
  • There are two stages to the FIFA World Cup tournament. Group play and the Finals
  • The top two teams of each group advance to the Finals
  • Etc. etc. etc.
• A digital representation of structured information of this sort is called an ontology

Ontologies

• Traditionally, ontology is the philosophical study of what actually exists (and doesn't)
• Typified by W.V.O. Quine's article "On What There Is"
• In computer science, structured digital representations constitute ontologies
• Ontologies may use tools from philosophy to articulate representations of the world
  • E.g. set-theory, propositional logic

Ontologies & KBs

• Is an ontology a database?
  • Strictly speaking, sure.
    • Databases are (often) representations of the world
    • Ontologies are (often) implemented on top of relational DBs
  • Typically, ontologies
    1. are more flexible and expressive than a DB
    2. contain greater meta-data than (typical) DBs
• Ontology + Inference Engine = Knowledge Base
  • Inference engines use logic (i.e. automated reasoning) to deduce new facts from material already known

Languages for Ontologies

• Many knowledge representation languages have been developed
• Requirements for an ontology language (Antoniou & van Harmelen):
  • Well-defined syntax
  • Formal semantics
  • Convenience of expression
  • Efficient reasoning support
  • Sufficient expressive power
• Those developed for the Semantic Web
  • RDF/RDFS
  • OWL
  • OWL Lite, OWL DL, OWL Full

RDF/RDFS

• RDF Schema: Built on top of RDF
• RDF: Makes statements about resources
  • {USNationalTeam winnerOf Match33214}
  • {MexicanNationalTeam loserOf Match33214}
  • {RoundOf16 levelOf Match33214}
  • {Match33214 occuredIn FIFAWorldCup2004}
• Includes reification: reasoning about statements
  • {EliasPonvert believes CrazyNotion3325}
  • where CrazyNotion3325 is identified with {DaMarcusBeasley bestMidfielder USNationalTeam}

RDF/S (cont'd)

• A collection of RDF statements is considered a graph

  

RDF/S Classes

• rdfs:Resource The class of resources
  • Example: USNationalTeam
• rdfs:Class The class of classes
  • Example: NationalTeam(s)
• rdf:Property The class of properties
  • Example: winnerOf
• rdfs:Literal A literal value—a string or number
  • Example: "¡Gooooooooooooo!"
• rdf:Statement The class of RDF statements
  • Example: {USNationalTeam winnerOf Match33214}

RDF/S Relations

• rdf:type Gives the class of a resource
  • Example: {USNationalTeam type NationalSoccerTeam}
  • i.e. USNationalTeam is a NationalSoccerTeam
• rdfs:subClassOf States one class is a sub-class of another
  • Example: {NationalSoccerTeam subClassOf SoccerTeam}
  • i.e. if {TeamX type NationalSoccerTeam} then {TeamX type SoccerTeam}
• rdfs:subPropertyOf States one property is a subProperty of (or entails) another
  • Example: {winnerOf subPropertyOf playedIn}
  • i.e. if {TeamX winnerOf MatchY} then {TeamX playedIn MatchY}

RDF/S Relations (cont'd)

• rdfs:domain Gives the domain of a property
  • Example: {winnerOf domain SoccerTeam}
  • i.e. if {TeamX winnerOf MatchY} then {TeamX type SoccerTeam}
• rdfs:range Gives the range of a property
  • Example: {winnerOf range SoccerMatch}
  • i.e. if {TeamX winnerOf MatchY} then {MatchY type SoccerMatch}
• RDF Reification: rdf:subject, rdf:predicate, rdf:object
  • {CrazyNotion3325 subject DaMarcusBeasley}
  • {CrazyNotion3325 predicate bestMidfielder}
  • {CrazyNotion3325 object USNationalTeam}

What Can RDF/S Do?

• Define classes of objects, and the objects themselves
• Build taxonomies via classes and subclasses
• The curiously powerful feature of reification
• Make statements about statements
• Queries in RQL.

What Can't RDF/S Do?

• Express a relation between more than 2 elements at once
• Express local scope of properties
  • Example: rdfs:range applies globally.
  • We can declare {playsFor range SoccerPlayer}
  • Cannot stipulate that only amateur players can play for Olympic teams, or male soccer players for men's national teams
• Express disjointedness of classes
  • Example: that no team can be both a MensNationalSoccerTeam and a WomensNationalSoccerTeam.
• Boolean combinations of classes
  • Example: persons that are both a SoccerPlayer and a ProfessionalAthlete

What Can't RDF/S Do? (cont'd)

• Express cardinality conditions
  • Example: a TournamentGroup consists of exactly 4 NationalSoccerTeams
• Express relations between relations
  • (other than subPropertyOf)
  • Example: A team has a unique coach. That is, BruceArena is the only individual ?X such that {?X coaches USNationalSoccerTeam}
  • Another example: There is an inverse relation between a nation's national team and a team's home nation:

    {USNationalTeam nationalTeamFor UnitedStates}

    if and only if

    {UnitedStates homeCountryFor USNationalTeam}

(Antoniou & van Harmelen 171–172)

Introducing OWL

• Web Ontology Language
• Regarding the acronym, from the Wikipedia article:

  Some may claim that the correct acronym for Web Ontology Language should be WOL instead of OWL. Others believe that the order was chosen in honor of the character Owl from Winnie the Pooh, who wrote his name WOL instead of OWL. In truth, OWL was proposed as an acronym that would be easily pronounced, yield good logos, suggest wisdom and honor Bill Martin's One World Language KR project from the 1970s. And, quoting Guus Schreiber: Why not be inconsistent in at least one aspect of a language which is all about consistency.

OWL Goals

• To enrich RDF to express the complex relations it otherwise cannot
• W3C Stated Goals:
  • Shared ontologies
  • Ontology evolution
  • Ontology interoperability
  • Inconsistency detection
  • Balance of expressivity and scalability
  • Ease of use
  • Compatibility with other standards
  • Internationalization

OWL as an Extension of RDF/S

owl:Class is a subclass of rdf:Class
owl:DatatypeProperty and owl:ObjectProperty are subclasses of rdf:Property

Classes in OWL

• Same idea as rdf:Class
• OWL introduces 3 special classes:
  • ThingThe class of everything.
  • Nothing The empty set.
  • Individual Everything that is not a class.
• New relations on classes:
  • disjointWith = The two classes have no members in common
  • equivalentClass = The two classes contain the same members

A Taste of OWL

{owl:Class,
  {rdf:ID, NationalSoccerTeam}
  {rdfs:subClassOf, SoccerTeam}}

{NationalSoccerTeam
  {rdf:ID, USNationalSoccerTeam}
  {nationalTeamFor
    {Country
      {rdf:ID, UnitedStates}
      {homeCountry, #USNationalSoccerTeam}}}}

{owl:ObjectProperty,
  {rdf:ID, memberOf}
  {rdfs:domain, #Person}
  {rdfs:range, #SoccerTeam}}

The XML is worse, believe me.

<rdf:RDF
  xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-sn#"
  xmlns:xsd="http://www.w3.org/2001/SXMLSchema#"
  xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
  xmlns:owl="http://www.w3.org/2002/07/owl#"
  xml:base="http://example.com/worldcup.owl"> ...

Properties in OWL

• DatatypeProperty and ObjectProperty
• equivalentProperty A property holds for all the same objects as another property

  {owl:ObjectProperty
    {rdf:ID, championOf}
    {owl:equivalentProperty, #winnerOfWorldCupFinal}}
  

• inverseProperty if one property applies to the opposite objects as the other
  • Example: homeCountryFor and nationalTeamFor

Restrictions

• You can restrict the application of a property given the class of its arguments
• Example: All players on a womens' soccer team are, women

  {owl:Class
    {rdf:about, #WomensSoccerTeam}
    {owl:Restriction
      {rdf:onProperty,
        {rdf:about, #memberOf}
        {owl:allValuesFrom #Woman}}}}
  

Restrictions by Cardinality

• owl:minCardinality the least number of individuals that apply to a property
• owl:maxCardinality the greatest number of individuals
• Example: Every team has exactly one coach

  {#SoccerTeam,
    {owl:Restriction
      {owl:onProperty, #coach}
      {owl:minCardinality, "1"}
      {owl:maxCardinality, "1"}}
  

Special Properties

• owl:TransitiveProperty If a property holds of individuals A and B, and of B and C then it holds of A and C.
  • Example: olderThan
• owl:SymmetricProperty If a property holds of A and B, then it holds of B and A
  • Example: played (two teams), isSiblingOf
• owl:FunctionalProperty A property holds of A and B if B is the only one
  • Example: age, homeCountryFor
• owl:InverseFunctionProperty Like FunctionalProperty. The other way around.

Boolean Combinators

• Boolean = Logical operators: and, or, not etc.
• owl:complementOf
• owl:unionOf
• owl:intersectionOf

OWL Extras

• Enumerations with owl:oneOf
• Primitive Data Types
  • Does not contain the derived datatypes in RDFS
• Versioning information
• owl:import statement

Layers of OWL

• There are 3 dialects of OWL
  • OWL Lite
  • OWL DL
  • OWL Full
• Each is properly contained within the other—that is
  • An OWL Lite document is an OWL DL document
  • An OWL DL document is an OWL Full document
• Each layer is more expressive than the one beneath it

Motivation for the Layers

• Balance of expressivity and efficiency
• Some applications only require expressive power of OWL Lite
• Worst case scenario OWL Full can be very difficult to reason with

OWL Lite

• No enumerated classes
• Cardinality (min or max) restricted to 1 or 0
• No boolean combinators
• No disjoint classes
• Class equivalence only valid on defined classes—not anonymous classes

OWL DL

• "Description Logic"
• Vocabulary partitioning: A resource must be an individual, or a class, or an object property, etc., but not two of these at the same time
• Explicit typing
• Given the first constraint, the following do not apply to datatype properties
  • owl:inverseOf
  • owl:FunctionalProperty
  • owl:InverseFunctionalProperty
  • SymmetricProperty
• No cardinality restrictions may be put on transitive properties
• Anonymous classes only allowed:
  • in owl:equivalentClass statements
  • in owl:disjointWitth statements
  • as the range of rdfs:subClassOf

OWL Full

• The Whole Banana

OWL Background

• W3C revision of DAML+OIL
  • DAML+OIL developed jointly (and independently) by DARPA and the EU
  • Also built on RDF/S
  • Extremely similar goals to OWL
  • Important influence and model for OWL
• Key differences (Daconta et al. 235)
  • New RDF/S
  • No qualified restrictions
  • daml:hasClass → owl:someValuesFrom and other name changes
    • owl:equivalentTo → owl:sameAs
  • Some RDF/S classes and properties removed
  • daml:disjointUnionOf removed
  • owl:symmetricProperty added
  • functionalProperty, inverseFunctionalProperty treated as global cardinality restrictions

Summary

• OWL is a language intended to provide SWT with a logical language rich enough to apply knowledge representation techniques to Web semantic content
• OWL builds upon the foundation laid by RDF/S
• OWL Dialects: Lite, DL and Full

Potential for OWL

• OWL is unwieldy and messy to work with by hand
• Authoring tools seem like fun, but need integration with Web Services and application development
• OWL under the covers or "powered by semantics" hopefully someday
• OWL does not include (propositional) logic within it—probably a good thing

Bibliography

Thanks!