OWL: a Description Logic Based Ontology Language for the Semantic Web

grotesqueoperationInternet και Εφαρμογές Web

21 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

70 εμφανίσεις

14
OWL:aDescriptionLogicBasedOntology
LanguagefortheSemanticWeb
IanHorrocks
PeterF.Patel-Schneider
DeborahL.M
cGuinness
ChristopherA.Welty
Abstract
Ithaslongbeenrealizedthatthewebcouldbenefitfromhavingitscontent
understandableandavailableinamachineprocessableform.TheSemantic
Webaimstoachievethisviaannotationsthatusetermsdefinedinontologies
togivewelldefinedmeaningtoWebaccessibleinformationandservices.
OWL,theontologylanguagerecommendedbytheW3Cforthispurpose,
washeavilyinfluencedbyDescriptionLogicresearch.Inthischapterwe
reviewbrieflysomeearlyeffortsthatcombineDescriptionLogicsandthe
Web,includingpredecessorsofOWLsuchasOILandDAML+OIL.We
thengoontodescribeOWLinsomedetail,includingthevariousinfluences
onitsdesign,itsrelationshipwithRDFS,itssyntaxandsemantics,anda
rangeoftoolsandapplications.
14.1Backgroundandhistory
TheWorldWideWeb,whilewildlysuccessfulingrowth,maybeviewed
asbeinglimitedbyitsrelianceonlanguageslikeHTMLthatarefocused
onpresentation(i.e.,textformatting)ratherthancontent.Languagessuch
asXMLdoaddsomesupportforcapturingthemeaningofWebcontent
(insteadofsimplyhowtorenderitinabrowser),butmoreisneededin
ordertosupportintelligentapplicationsthatcanbetterexploittheever
increasingrangeofinformationandservicesaccessibleviatheWeb.Such
applicationsareurgentlyneededinordertoavoidoverwhelminguserswith
thesheervolumeofinformationbecomingavailable.
TheSemanticWebhasbeenenvisagedasanevolutionoftheexistingWeb
fromalinkeddocumentrepositoryintoanapplicationplatformwhere“infor-
mationisgivenwell-definedmeaning,betterenablingcomputersandpeople
1
2I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
toworkincooperation”
[
Berners-Leeetal.,2001
]
.Thisistobeachievedby
augmentingtheexistinglayoutinformationwithsemanticannotationsthat
adddescriptivetermstoWebcontent,withthemeaningofsuchtermsbeing
definedinontologies.
Inorderforthemeaningofsemanticannotationstobeaccessibletoappli-
cations(aswellashumans),theontologylanguagebeingusedshouldhavea
preciselydefinedsemanticsandshouldbeamenabletoautomatedprocess-
ing.DescriptionLogicsappeartobeideallysuitedtothisrole:theyhavea
formallogic-basedsemantics,andareoftenequippedwithdecisionproce-
duresthathavebeendesignedwithaviewtoimplementationinautomated
reasoningsystems.ThisviewofthepotentialplaceofDescriptionLogics
intheSemanticWebledtothedevelopmentofanumberoflanguagesthat
broughtDescriptionLogicconceptstotheSemanticWeb,culminatinginthe
developmentoftheWebOntologyLanguageOWL.OWListheWorldWide
WebConsortium(W3C)recommendedontologylanguagefortheSemantic
Web,andexploitsmanyofthestrengthsofDescriptionLogics,including
welldefinedsemanticsandpracticalreasoningtechniques.
InthischapterwefirstreviewbrieflythehistoryofDescriptionLogic
effortsrelatedtotheSemanticWeb,inparticularOILandDAML+OIL.
WethengoontodescribeOWLinsomedetail,andtoshowhowitbrings
DescriptionLogicconceptsfullyintotheSemanticWeb.
14.1.1EarlyUsesofDescriptionLogicsintheSemanticWeb
BeforethedevelopmentofDescriptionLogic-relatedlanguagesdesignedfor
theSemanticWeb,therewereseveralsystemsthatusedDescriptionLogics
inthecontextoftheweb.Wewilldescribesomesalientfeaturesoftwo
systems,UntangleandFindUR,thatillustrateearlyDescriptionLogic
usageontheweb.
Therelationshipbetweenhypertextandsemanticnetworkshaslongbeen
realized,butoneoftheearliestDescriptionLogicsystemstorealizethisre-
lationshipwastheUntanglesystem
[
WeltyandJenkins,2000
]
,aDescrip-
tionLogicsystemforrepresentingbibliographic(card-catalog)information.
TheUntangleprojectbeganasabitofexploratoryresearchinusingDe-
scriptionLogicsfordigitallibraries
[
Welty,1994
]
,butoutofsheertemporal
coincidencewiththeriseoftheweb,awebinterfacewasaddedandthefirst
web-basedDescriptionLogicsystemwasborn.
TheoriginalUntanglewebinterfacewasdevelopedin1994
[
Welty,
1996a
]
,andcombinedLisp-ClassicandtheCommonLispHypermedia
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb3
Server(CL-HTTP)
[
Mallery,1994
]
toimplementahypertextviewofthe
ABoxandTBoxsemanticnetworks,andusednestedbulletliststoviewthe
concepttaxonomy,within-pagecrossreferencesforconceptshavingmul-
tipleparents.Theinterfacewasinterestinginsomerespectsasatoolto
visualizeDescriptionLogicandsemanticnetworkinformation,thoughthis
aspectwasneverfullydeveloped.
AstheWorldWideWeb(WWW)becametheprimarymeansofdissem-
inationofcomputerscienceresearch,thegoalsoftheUntangleproject
shiftedin1995tocatalogingandclassifyingpagesontheweb
[
Welty,1996b
]
,
whichwasviewedasamassiveandunstructureddigitallibrary
[
Welty,1998
]
.
AnotherearlyprojectusingDescriptionLogicsforthewebwasthe
FindURsystematAT&T
[
McGuinness,1998;McGuinnessetal.,1997
]
.
ThebasicnotionofFindURwasqueryexpansion,1
thatis,takingsyn-
onymsorhyponyms(morespecificterms)andincludingthemintheinput
terms,therebyexpandingthequery.
TheFindURsystemrepresentedasimplebackgroundknowledgebase
containingmostlythesaurusinformationbuiltinaDescriptionLogic(Clas-
sic)usingthemostbasicnotionsofWordnet(synsetsandhyper/hyponyms)
[
Miller,1995
]
.Conceptscorrespondingtosetsofsynonyms(synsets)were
arrangedinataxonomy.Thesesynsetsalsocontainedaninformallistof
relatedterms.Sitespecificsearchengines(builtonVerity—acommercial
searchengine)werehookeduptotheknowledgebase.Anysearchterm
wouldfirstbecheckedintheknowledgebase,andifitwascontainedinany
synset,anewquerywouldbeconstructedconsistingofthedisjunctionofall
thesynonymousterms,aswellasallthemorespecificterms(hyponyms).
ThebackgroundknowledgewasrepresentedinClassic,buttheDescrip-
tionLogicwasnotitselfpartoftheon-linesystem.Instead,theinformation
usedbythesearchenginewasstaticallygeneratedonaregularbasisand
usedtopopulatethesearchengine.ThetruepowerofusingaDescrip-
tionLogicasthesubstratefortheknowledgebasewasrealizedmainlyin
themaintenancetask.TheDescriptionLogicallowedthemaintainerofthe
knowledgebasetomaintainsomeamountofconsistency,suchasdiscovering
cyclesinthetaxonomyanddisjointsynsets.Thesesimpleconstraintsproved
effectivetoolsformaintainingtheknowledgesincetheknowledgeitselfwas
verysimple.
AdditionaluseoftheDescriptionLogicapproachinFindURwasrealised
1
WorkonasubsequentDescriptionLogicbasedapproachtoqueryexpansionaddressedsome
formalissuesinevaluatingthesoundnessandcompletenessofalternativeapproaches
[
Rousset,
1999
]
.OtherworkonDescriptionLogic(orDescriptionLogicinspired)approachestoretrieval
alsoexists,e.g.,
[
Meghinietal.,1997
]
.
4I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
inapplicationsthatexposedmorestructuredsearches,exploitingsubclass
hierarchiesandpropertyrelationships,suchasthemedicalapplicationsof
FindURintheP-CHIPPrimaryCareSearchApplication
[
Brachmanetal.,
1999
]
.Thistypeofstructuredsearch,exploitingbackgroundontologiesand
relationshipsbetweenterms,canalsobeseeninlaterwork,e.g.,intheSHOE
project
[
Heflinetal.,2003
]
.
14.2StepsTowardsIntegrationwiththeSemanticWeb:OIL
andDAML+OIL
ThefirstmajorefforttobuildalanguagethatcombinedDescriptionLogics
andtheSemanticWebwasOIL(theOntologyInferenceLayer)
[
Horrocks
etal.,2000a
]
,apartoftheOn-To-Knowledgeresearchprojectfundedby
theEuropeanUnion.TheOILlanguagewasexplicitlydesignedas“aweb-
basedrepresentationandinferencelanguageforontologies[combining]the
widelyusedmodelingprimitivesfromframe-basedlanguageswiththeformal
semanticsandreasoningservicesprovidedbydescriptionlogics”(http://
www-ontoknowledge.org/oil/oilhome.shtml).
DescriptionLogicsprovidethesemanticsforOIL,somuchsothatthe
semanticsofOILisspecifiedviaamappingtotheDescriptionLogicSHIQ
[
Fenseletal.,2001;Horrocksetal.,1999
]
.OILhasasyntaxbasedonthe
ResourceDescriptionFramework(RDF),aswellasanXMLsyntax,that
providedtheconnectiontotheSemanticWebofthetime.
2
OILallowsthe
groupingofDescriptionLogicconstructsinawaysimilartoframesystems,
providingamoreintuitivefeeltothelanguageasopposedtothelogically
inspiredsyntaxusuallyusedforDescriptionLogics.Thesethreeinfluences—
DescriptionLogics,frames,andtheSemanticWeb—arepresentnotonlyin
OIL,butalsoinallofitssuccessors.
14.2.1OIL
TheOILlanguageisdesignedtocombineframe-likemodelingprimitives
withtheincreased(insomerespects)expressivepower,formalrigorand
automatedreasoningservicesofanexpressiveDescriptionLogic
[
Fenselet
al.,2000
]
.OILalsocomes“webenabled”byhavingbothXMLandRDFS
basedserializations(aswellasaformallyspecified“humanreadable”form,
whichwewillusehere).Inframelanguages,classes(concepts)aredescribed
byframes,whosemaincomponentsconsistofalistofsuperclassesandalist
2
AtthetimethatOILwasdeveloped,RDF—thebaselanguageoftheSemanticWeb—was
withoutafullyspecifiedsemanticfoundation.
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb5
ofslot-fillerpairs.AslotcorrespondstoaroleinaDescriptionLogic,anda
slot-fillerpaircorrespondstoeitheravaluerestriction(aconceptoftheform
∀R.C)oranexistentialquantification(aconceptoftheform∃R.C)—one
ofthecriticismsleveledatframelanguagesisthattheyareoftenunclearas
toexactlywhichoftheseisintendedbyaslot-fillerpair.
OILextendsthisbasicframesyntaxsothatitcancapturethefullpower
ofanexpressiveDescriptionLogic.
Inordertoallowuserstochoosetheexpressivepowerappropriatetotheir
application,andtoallowforfutureextensions,alayeredfamilyofOIL
languageswasdescribed.Thebaselayer,called“CoreOIL”
[
Bechhoferet
al.,2000
]
,isacut-downversionofthelanguagethatcloselycorresponds
withRDFS(i.e.,itincludesonlyclassandslotinclusionaxioms,andslot
rangeanddomainconstraints).Thestandardlanguageiscalled“Standard
OIL”,andwhenextendedwithABoxaxioms(i.e.,theabilitytoassert
thatindividualsandtuplesare,respectively,instancesofclassesandslots),
iscalled“InstanceOIL”.Finally,“HeavyOIL”wasthenamegiventoa
furtherlayerthatwastoincludestillunspecifiedlanguageextensions.
Figure14.1illustratesanOILontology(usingthehumanreadableserial-
ization)correspondingtoanexampleterminologyfromChapter??.Afull
specificationofOIL,includingDTDsfortheXMLandRDFSserializations,
canbefoundin
[
Horrocksetal.,2000a
]
.
StandardOILcanbeseenasasyntacticvariantoftheDescriptionLogic
SHIQ
[
Horrocksetal.,1999
]
extendedwithsimpleconcretedatatypes
[
BaaderandHanschke,1991;HorrocksandSattler,2001
]
;wewillcall
thisDescriptionLogicSHIQ(D).Ratherthanprovidingtheusualmodel-
theoreticsemantics,OILdefinesatranslationthatmapsanOILontol-
ogyintoanequivalentSHIQ(D)terminology.Fromthismapping,OIL
derivesbothaclearsemanticsandameanstoexploitthereasoningser-
vicesofDescriptionLogicsystemssuchasFact
[
Horrocks,1998b
]
,Racer
[
HaarslevandM¨oller,2001
]
andPellet
[
Pellet,2003
]
thatimplement(most
of)SHIQ(D).
14.2.2TheDAMLprojectandDAML+OIL
AtaboutthesametimeasOILwasbeingdeveloped,theDARPAAgent
MarkupLanguage(DAML)programwasstartedintheUnitedStates.
DAMLwasinitiatedinordertoprovidethefoundationforthenextgener-
ationofthewebwhich,itwasanticipated,wouldincreasinglyutilizeagents
andprogramsratherthanrelyingsoheavilyonhumaninterpretationofweb
information
[
HendlerandMcGuinness,2000
]
.
6I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
name“Family”
documentation“Exampleontologydescribingfamilyrelationships”
definitions
slot-defhasChild
inverseisChildOf
class-defdefinedWoman
subclass-ofPersonFemale
class-defdefinedMan
subclass-ofPersonnotWoman
class-defdefinedMother
subclass-ofWoman
slot-constrainthasChild
has-valuePerson
class-defdefinedFather
subclass-ofMan
slot-constrainthasChild
has-valuePerson
class-defdefinedParent
subclass-oforFatherMother
class-defdefinedGrandmother
subclass-ofMother
slot-constrainthasChild
has-valueParent
class-defdefinedMotherWithManyChildren
subclass-ofMother
slot-constrainthasChild
min-cardinality3
class-defdefinedMotherWithoutDaughter
subclass-ofMother
slot-constrainthasChild
value-typenotWoman
Fig.14.1.OIL“family”ontology.
Oneoftheearlywidely-distributedcontributionsoftheDAMLpro-
gramwasDAML-ONT—aproposalforanontologylanguagefortheweb
[
HendlerandMcGuinness,2000;McGuinnessetal.,2002
]
.Thislanguage
beganwiththerequirementtobuildonthebestpracticeinweblanguages
ofthetime,andinparticulartoextendW3C’sResourceDescriptionFrame-
work,withtheaimofaddingexpressivepowersuitedtoagentandservice
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb7
interoperation.
ItbecameobviousthatthegoalsofDAML-ONTandOILweresosimilar
thattheseobjectivescouldbestbeservedbycombiningthetwoefforts.The
resultinglanguage,DAML+OIL,hasaformal(model-theoretic)semantics
thatprovidesmachineandhumanunderstandability
[
vanHarmelenetal.,
2001
]
,anaxiomatization
[
FikesandMcGuinness,2001
]
,andareconciliation
ofthelanguageconstructorsfromthetwoprecursorlanguages.
DAML+OILissimilartoOILinmanyrespects,butismoretightlyin-
tegratedwithRDFS,whichprovidestheonlyspecificationofthelanguage
anditsonlyserialization.WhilethedependenceonRDFShassomead-
vantagesintermsofthere-useofexistingRDFSinfrastructureandthe
portabilityofDAML+OILontologies,usingRDFStocompletelydefine
thestructureofDAML+OILisquitedifficultas,unlikeXML,RDFSis
notdesignedfortheprecisespecificationofsyntacticstructure.Forexample,
thereisnowayinRDFStostatethatarestriction(slotconstraint)should
consistofexactlyoneproperty(slot)andoneclass.
ThesolutiontothisproblemadoptedbyDAML+OIListodefinethe
semanticsofthelanguageinsuchawaythatitgivesameaningtoany(parts
of)ontologiesthatconformtotheRDFSspecification,including“strange”
constructssuchasslotconstraintswithmultipleslotsandclasses.Thisis
madeeasierbythefactthat,unlikeOIL,thesemanticsofDAML+OILis
directlydefined.
AnothereffectofDAML+OIL’stightintegrationwithRDFSisthat
theframestructureofOIL’ssyntaxismuchlessevident:aDAML+OIL
ontologyismoreDescriptionLogiclikeinthatitconsistslargelyofarel-
ativelyunstructuredcollectionofsubsumptionandequalityaxioms.This
canmakeitmoredifficulttouseDAML+OILwithframe-basedtoolssuch
asProt
´
eg
´
e
[
Grossoetal.,1999
]
orOilEd
[
Bechhoferetal.,2001b
]
,be-
causetheaxiomsmaybesusceptibletomanydifferentframe-likegroupings
[
Bechhoferetal.,2001a
]
.
Fromthepointofviewoflanguageconstructs,thedifferencesbetween
OILandDAML+OILarerelativelytrivial.Althoughthereissomedif-
ferencein“keyword”vocabulary,thereisusuallyaone-to-onemappingof
constructors,andinthecaseswheretheconstructorsarenotcompletely
equivalent,simpletranslationsarepossible.
TheinitialreleaseofDAML+OILdidnotincludeanyspecificationof
datatypes.Thelanguagewas,however,subsequentlyextendedwitharbi-
trarydatatypesfromtheXMLSchematypesystem,whichcanbeusedin
restrictions(slotconstraints)andrangeconstraints.AsinSHOQ(D)
[
Hor-
rocksandSattler,2001
]
,acleanseparationismaintainedbetweeninstances
8I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
of“object”classes(definedusingtheontologylanguage)andinstancesof
datatypes(definedusingtheXMLSchematypesystem).Inparticular,itis
assumedthatthedomainofinterpretationofobjectclassesisdisjointfrom
thedomainofinterpretationofdatatypes,sothataninstanceofanobject
class(e.g.,theindividualItaly)canneverhavethesameinterpretationasa
valueofadatatype(e.g.,theinteger5),andthatthesetofobjectproperties
(whichmapindividualstoindividuals)isdisjointfromthesetofdatatype
properties(whichmapindividualstodatatypevalues).
14.3FullIntegrationintotheSemanticWeb:OWL
WithitstighterintegrationwithRDFandRDFS,DAML+OILwasthe
firstDescriptionLogic-inspiredlanguagetobecompletelyintegratedintothe
fabricoftheSemanticWeb(asitwasthendefined).Therewas,however,
noformalsemanticintegrationofDAML+OILwithRDFandRDFS,
asRDFandRDFSdidnothaveaformalsemanticsatthetimewhen
DAML+OILwasbeingdeveloped.DAML+OILwasalsoonlytheproduct
ofanad-hocgroupofresearchers,andwasthusnotanofficialpartofthe
SemanticWeb.
Thefixtheseproblems,theW3Ccharteredtwoworkinggroupsin2001.
TheRDFCoreWorkingGroupwastoberesponsibleforupdatingtheRDF
recommendationandprovidingaformalsemanticsforRDFandthusforthe
SemanticWeb.TheWebOntologyWorkingGroupwastoberesponsiblefor
designinganontologylanguagefortheweb,compatiblewiththenewversion
ofRDF.TheWebOntologyWorkingGroupdevelopedtheWebOntology
LanguageOWL
[
Bechhoferetal.,2004;Patel-Schneideretal.,2004
]
,which
becameaW3Crecommendationon10February2004.Inparallel,theRDF
CoreWorkingGroupdevelopedaformalsemanticsforRDFandRDFS
[
Hayes,2004
]
.
14.3.1InfluencesontheDesignofOWL
ThedesignofOWLwassubjecttoavarietyofinfluences.Thesein-
cludedinfluencesfromestablishedformalismsandknowledgerepresentation
paradigms,influencesfromexistingontologylanguages,andinfluencesfrom
existingSemanticWeblanguages.
SomeofthemostimportantinfluencesonthedesignofOWLcame,
viaitspredecessorDAML+OIL,fromDescriptionLogics,fromtheframes
paradigm,andfromRDF.Inparticular,theformalspecificationofthelan-
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb9
guagewasinfluencedbyDescriptionLogics,thesurfacestructureofthe
language(asseenintheabstractsyntax)wasinfluencedbytheframes
paradigm,andtheRDF/XMLexchangesyntaxwasinfluencedbyare-
quirementforupwardscompatibilitywithRDF.
DescriptionLogics,andinsightsfromDescriptionLogicresearch,hada
stronginfluenceonthedesignofOWL,particularlyontheformalisation
ofthesemantics,thechoiceoflanguageconstructors,andtheintegration
ofdatatypesanddatavalues.InfactOWLDLandOWLLite(twoof
thethreespeciesofOWL)canbeviewedasexpressiveDescriptionLogics,
withanontologybeingequivalenttoaDescriptionLogicknowledgebase.
Moreprecisely,OWLDLandOWLLiteareequivalenttoSHIF(D)and
SHOIN(D)respectively(seeSections14.3.3and14.3.5formoredetails).
Thisdesignwasmotivatedbypracticalconsiderations.Thedesignersof
OWLwantedtohavesomeideaastohowdifficultitwouldbefortools
andapplicationstosupportthelanguage.Itwasthereforeimportantto
understanditsformalproperties,e.g.,withrespecttothedecidabilityand
complexityofkeyinferenceproblems.Thesepropertiesfolloweddirectly
fromthecorrespondenceswithDescriptionLogics.Thesecorrespondences
wouldallowtoolsandapplicationstoexploitknownreasoningalgorithms
andeven(highlyoptimised)implementations.
IntheSemanticWebcontext,whereuserswithawiderangeofexpertise
mightbeexpectedtocreateormodifyontologies,readabilityandgeneral
easeofuseareimportantconsiderationsforanontologylanguage.Inthede-
signofOIL
[
Fenseletal.,2001
]
,oneofthelanguagesonwhichDAML+OIL
wasbased,theserequirementswereaddressedbyprovidingasurfacesyntax
basedontheframesparadigm.Framesgrouptogetherinformationabout
eachclass,makingontologieseasiertoreadandunderstand,particularlyfor
usersnotfamiliarwith(Description)Logics.Theframesparadigmhasbeen
usedinanumberofwellknownknowledgerepresentationandontologyen-
vironmentsystemsincludingtheProt
´
eg
´
eontologydesigntool
[
Grossoet
al.,1999
]
,theOntolinguaontologyenvironmenttool
[
Farquharetal.,1996
]
,
theOKBCknowledgemodel
[
Chaudhrietal.,1998
]
,andtheChimaeraOn-
tologyEvolutionEnvironment
[
McGuinnessetal.,2000
]
.ThedesignofOIL
wasinfluencedbyXOL
[
Karpetal.,1999
]
—aproposalforanXMLsyntax
forOKBCLite(acutdownversionoftheOKBCknowledgemodel).
Inframebasedlanguages,eachclassisdescribedbyaframe.Theframe
includesthenameoftheclass,identifiesthemoregeneralclass(orclasses)
thatitspecialises,andlistsasetof“slots”.Aslotmayconsistofaproperty-
valuepair,oraconstraintonthevaluesthatcanactasslot“fillers”(inthis
10I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
context,valuemeanseitheranindividualoradatavalue).Thisstructurewas
usedintheOILlanguage,withsomeenrichmentofthesyntaxforspecifying
classesandslotconstraintssoastoenablethefullpowerofaDescription
Logicstylelanguagetobecaptured.Inaddition,propertyframeswereused
todescribeproperties,e.g.,specifyingmoregeneralproperties,rangeand
domainconstraints,transitivityandinversepropertyrelationships.
AclassframeissemanticallyequivalenttoaDescriptionLogicaxiomas-
sertingthattheclassbeingdescribedbytheframeisasubclassofeachof
theclassesthatitspecialisesandofeachofthepropertyrestrictionscorre-
spondingtotheslots.Aswellasaricherslotsyntax,OILalsoofferedthe
possibilityofassertingthattheclassbeingdescribedbytheframewasex-
actlyequivalenttotherelevantintersectionclass,(i.e.,thattheyweremutu-
allysubsuming).Apropertyframeisequivalenttoasetofaxiomsasserting
therelevantsubpropertyrelationships,rangeanddomainconstraintsetc.
OILwasdesignedsothatOILframescouldeasilybemappedtoequivalent
axiomsintheSHOQ(D)DescriptionLogic
[
Deckeretal.,2000
]
.
TheformalspecificationandsemanticsofOWLaregivenbyanabstract
syntax
[
Patel-Schneideretal.,2004
]
thathasbeenheavilyinfluencedby
framesingeneralandbythedesignofOILinparticular.Intheabstract
syntax,axiomsarecompoundconstructionsthatareverylikeanOIL-style
frame.Forclasses,theyconsistofthenameoftheclassbeingdescribed,a
modalityof“partial”or“complete”(indicatingthattheaxiomisassertinga
subclassorequivalencerelationshiprespectively),andasequenceofproperty
restrictionsandnamesofmoregeneralclasses.Similarly,apropertyaxiom
specifiesthenameofthepropertyanditsvariousfeatures.
Theframestyleoftheabstractsyntax,whichborrowsheavilyinspirit
fromthehumanreadablesyntaxforOIL,makesitmucheasiertoread
(comparedtotheRDF/XMLsyntax),andalsoeasier(fornon-logicians)
tounderstandandtouse.Moreover,abstractsyntaxaxiomshaveadirect
correspondencewithDescriptionLogicaxioms,andtheycanalsobemapped
toasetofRDFtriples.
ThethirdmajorinfluenceonthedesignofOWLwastherequirementto
maintainthemaximumupwardscompatibilitywithexistingweblanguages,
andinparticularwithRDF
[
ManolaandMiller,2004
]
.Onthefaceofit
thisrequirementmadegoodsenseasRDF(andinparticularRDFSchema)
alreadyincludedseveralofthebasicfeaturesofaclassandpropertybased
ontologylanguage,e.g.,itallowssubclassandsubpropertyrelationshipsto
beasserted.Moreover,thedevelopmentofRDFprecededthatofOWL,
anditseemedreasonabletotrytoappealtoanyusercommunityalready
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb11
establishedbyRDF.
ItmayseemeasytomeetthisrequirementsimplybygivingOWLan
RDF-basedsyntax,but,inordertoprovidemaximumupwardscompati-
bility,itwasalsothoughtnecessarytoensurethatthesemanticsofOWL
ontologieswasconsistentwiththesemanticsofRDF.Thisprovedtobe
difficult,however,giventhegreatlyincreasedexpressivepowerprovidedby
OWL.ThiswillbediscussedinmoredetailinSection14.3.2.
14.3.2Layering(onRDFS)andthree“species”ofOWL
Asmentionedabove,thethedesignofOWLincludedarequirementtomain-
tainthemaximumupwardscompatibilitywithRDF
[
ManolaandMiller,
2004
]
.Onthefaceofitthisrequirementmadegoodsense,butitledtoa
numberofproblemsinthedesignofOWL.
Inthefirstplace,RDF/XMLisextremelyverbose.Compare,forexample,
informationaboutaclassasitwouldbegiveninaDescriptionLogicsyntax
StudentDPerson￿1enrolledIn
(aStudentisaPersonwhoisenrolledInatleast1thing),withhowit
wouldmostnaturallybewrittenusingtheOWLRDF/XMLsyntax
3
<owl:Classrdf:ID="Student">
<owl:intersectionOfrdf:parsetype="Collection">
<owl:Classrdfs:about="Person"/>
<owl:Restriction>
<owl:onPropertyrdf:resource="enrolledIn"/>
<owl:minCardinalityrdfs:datatype="&xsd;Integer">
1
</owl:minCardinality>
</owl:Restriction>
<owl:intersectionOf>
</owl:Class>
ThisverbosityofOWL’sRDF/XMLsyntaxmaynot,initself,bease-
riousproblemgiventhecapabilitiesandbandwidthsofmoderncomputers
andcommunicationsystems,andthatthissyntaxismainlyintendedfor
exchangingdatabetweenOWLapplications.TheRDF/XMLsyntaxdoes,
however,leadtosomemoreseriousproblems.RDFisitselfagraphbased
formalism,withgraphsexpressedassetofsubject-predicate-objecttriples,
3
FulldetailsontheOWLRDF/XMLsyntaxcanbefoundintheOWLReferencedocu-
ment
[
Bechhoferetal.,2004
]
.
12I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
whereeachtriplerepresentsalabellededge(thepredicate)connectingtwo
vertices(thesubjectandobject).ThismeansthatmanyOWLconstructs,
suchaspropertyrestrictions,havetobeencodedasseveraltriples.Thereis
norequirementthatthesetriplesoccurtogether,soparsingbecomesdifficult
asitmaybenecessarytoscantheentireinputinordertolocateallofthe
componentsofagivenconstruction.Moreover,circularandotherunusual
structureswithilldefinedmeaningscannotberuledout.
Evenmoreproblematicalistherelationshipbetweenthesemanticsofan
OWLontologyandthesemanticsoftheRDFtriplesusedtoencodeit.This
wasnotasmuchofanissuewhenOILandDAML+OILweredesigned,
asatthattimethemeaningofRDFwasnotpreciselyspecified.OILin
particulardidnotbothertorelatetheRDFmeaningofitsRDF/XML
syntaxtotheOILmeaningofthissyntax—theRDF/XMLsyntaxforsome
OILconstructsdoesmore-or-lesslineupwiththeRDFmeaningofthese
constructs,butthisisbynomeansthecaseforallsuchconstructs.
DAML+OILdidabetterjobofabidingbytheRDFmeaningofitssyn-
tax.TheDAML+OILmodeltheory
[
vanHarmelenetal.,2001
]
includeda
semanticconditionfortriplesthatwasclosetotheRDFmeaning(asde-
finedatthattime)fortriples.Moreover,DAML+OILusedthebuilt-inRDF
andRDFSvocabularytoagreaterextentthandidOIL,anduseditina
waygenerallycompatiblewiththeRDForRDFSmeaning.Forexample
DAML+OILusesrdfs:subClassOftorelateclassestosuperclasses.
BythetimeOWLwasbeingdesigned,RDFwasbeinggivenaformal
meaningviatheRDFmodeltheory
[
Hayes,2004
]
.Itprovedtobeextremely
difficulttodesignanRDFsyntaxforOWLsuchthattheDescriptionLogic
semanticsassignedbyOWLtoitsvariousconstructswascompatiblewith
thesemanticsoftheRDFtriplesusedtoencodethem.Moreover,therewas
afundamentalincompatibilitybetweentherequirementtobefullyback-
wardscompatiblewithOWL(i.e.,toallowarbitraryRDFgraphstobe
interpretedasOWLontologies),andtherequirementforOWLtobeequiv-
alenttoanexpressivedescriptionlogic.(See
[
Horrocksetal.,2003
]
fora
detaileddiscussionoftheseissues.)
AftermuchargumentwithintheOWLworkinggroup,theseproblems
wereeventually“resolved”bydefiningtwo“species”ofOWL:OWLFull
andOWLDL.OnlyasubsetofRDFgraphscorrespondtoOWLDLon-
tologies.Inparticular,thegraphcannotincludecyclicalandotherprob-
lematicalconstructions(including“malformed”OWLsyntax),andthesets
ofclass,property,andindividualnamesmustbedisjoint.ForRDFgraphs
thatsatisfythesesyntacticrestrictions,afairlystandardDescriptionLogic
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb13
stylemodeltheoryisusedtodefinethesemanticsoftheresultingOWLcon-
structs.Incontrast,OWLFullusesanRDFstylemodeltheorytogivea
semanticaccountoftheuseofOWLconstructorsinarbitraryRDFgraphs.
(SeeSections14.3.4and14.3.5formoredetails.)
AswellasargumentsabouttheimportanceofcompatibilitywithRDF,
therewerealsotensionswithintheworkinggroupregardingtheexpressive
powerthatwasappropriateforOWL,withsomemembersarguingthat
OWLDLwastoocomplextobeunderstoodbynewusersorimplemented
byapplicationdevelopers.Weightwaslenttotheseargumentsbythefact
that,atthetime,notableauxdecisionprocedurewasknownforSHOIN(D)
(thedescriptionlogiclanguageunderlyingOWLDL),andnoimplemented
systemsupportedit.Inresponsetotheseconcerns,athirdOWLspecies
wasdefined,namelyOWLLite.
OWLLiteisasyntacticsubsetofOWLDLthatprohibitsand/orrestricts
theuseofcertainconstructorsandaxiomswiththeaimofmakingthe
languageeasiertounderstandand/orimplement.Ithassubsequentlybeen
shown,however,thatbycombiningotherconstructorsandaxioms,mostof
theexpressivepowerofOWLDLcanberegained,andthatOWLLiteis
expressivelyequivalenttoSHIF(D)
[
HorrocksandPatel-Schneider,2003
]
.
GiventhatOWLFulldoesnotcorrespondeithersyntacticallyorseman-
ticallytoaDescriptionLogic,intheremainderofthischapterwewillfocus
ourattentiononOWLDLandOWLLite.
14.3.3OWLDLabstractsyntaxandsemantics
OWLDLhassomedifferencesfromstandardDescriptionLogics.These
differencesprovideabridgebetweentheformalDescriptionLogicworldand
theSemanticWebworld.
r
OWLusesURIreferencesasnames,andconstructstheseURIreferencesinthe
samemannerasthatusedbyRDF.ItisthuscommoninOWLtousequalified
namesasshorthandsforURIreferences,using,forexample,thequalifiedname
owl:ThingfortheURIreferencehttp://www.w3.org/2002/07/owl#Thing.
r
OWLgathersinformationintoontologies,whicharegenerallystoredasWeb
documentswritteninRDF/XML.Ontologiescanimportotherontologies,adding
theinformationfromtheimportedontologytothecurrentontology.
r
OWLallowsRDFannotationpropertiestobeusedtoattachinformationto
classes,properties,andontologies,suchasowl:DeprecatedClass.Theseannota-
tionsareRDFtriples,andarethereforerequiredtocarryafullsemanticweightin
RDF.InOWLDL,however,suchannotationscarryaseparate,limitedmeaning.
14I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
AbstractSyntaxDLSyntax
Descriptions(C)
A
A
owl:Thing

owl:Nothing

intersectionOf(C1
...C
n)
C1
...Cn
unionOf(C1
...C
n
)
C1
...Cn
complementOf(C)
¬C
oneOf(o1
...o
n)
{o1
}...{on}
restriction(RsomeValuesFrom(C))
∃R.C
restriction(RallValuesFrom(C))
∀R.C
restriction(RhasValue(o))
R:o
restriction(RminCardinality(n))
￿nR
restriction(RminCardinality(n))
￿nR
restriction(UsomeValuesFrom(D))
∃U.D
restriction(UallValuesFrom(D))
∀U.D
restriction(UhasValue(v))
U:v
restriction(UminCardinality(n))
￿nU
restriction(UmaxCardinality(n))
￿nU
DataRanges(D)
D
D
oneOf(v1
...v
n)
{v1
}...{vn}
ObjectProperties(R)
R
R
inv(R)
R−
DatatypeProperties(U)
U
U
Individuals(o)
o
o
DataValues(v)
v
v
Fig.14.2.OWLDLDescriptions,DataRanges,Properties,Individuals,andData
Values
r
OWLusesthefacilitiesofRDFdatatypesandXMLSchemadatatypestoprovide
datatypesanddatavalues(averyrestrictedformofconcretedomains??).
r
OWLDLandOWLLitehaveaframe-likeabstractsyntax,whereasRDF/XML
istheofficialexchangesyntaxforallthreespeciesofOWL.
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb15
Asmentionedabove,OWLDLisverycloselyrelatedtoSHOIN(D),
whichextendsSHOIQ
[
HorrocksandSattler,2005
]
withdatatypeslike
thoseinSHOQ(D)
[
HorrocksandSattler,2001
]
,butallowsonlyunqual-
ifiednumberrestrictions(seeChapter??).OWLDLcanformdescrip-
tionsofclasses,datatypes,individualsanddatavaluesusingthecon-
structsshowninFigure14.2.InthistablethefirstcolumngivestheOWL
abstractsyntaxfortheconstruction,whilethesecondcolumngivesthe
equivalentDescriptionLogicsyntax.ThelettersA,D,R,U,oandv
represent,respectively,namesforclasses(concepts),dataranges,object
properties(abstractroles),datatypeproperties(concreteroles),individu-
als(nominals)anddatavalues;C,possiblysubscripted,representsanar-
bitraryclassdescription.InOWL,datavaluesareRDFliterals(i.e.,in-
stancesofdatatypessuchasstringorinteger),andallothernamesareURI
references.Asmentionedabove,owl:Thingandowl:Nothingareshort-
handfortheURIreferenceshttp://www.w3.org/2002/07/owl#Thingand
http://www.w3.org/2002/07/owl#Nothingrespectively.
ThetreatmentofconcretevaluesinOWLissomewhatdifferentfrom
theusualtreatmentofconcretevaluesinDescriptionLogics.OWL
usesdatatypesfromXMLSchemadatatypesasitsconcretetypes,so
xsd:integerisaconcretetypeinOWL,namelythetypeofintegers.
OWLusestheRDFsyntaxfortheseconcretevalues,so"2"^^xsd:integer
isthewaytowritetheinteger2.OWLalsoallowsplainRDFlit-
erals,whichareacombinationofastringandanoptionallanguage
tag.Theseplainliteralsbelongtothedatatyperdfs:Literal.OWLalso
allowssetsofdatavaluestobeusedinconceptexpressions,asin
oneOf("1"^^xsd:integer"2"^^xsd:integer"3"^^xsd:integer).
OWLusesthesedescription-formingconstructsinaxiomsthatprovidein-
formationaboutclasses,properties,andindividuals,asshowninFigure14.3.
Again,theframe-likeabstractsyntaxisgiveninthefirstcolumn,andthe
standardDescriptionLogicsyntaxisgiveninthesecondcolumn.Theletters
A,D,R,U,oandv(ineachcasepossiblysubscripted)represent,respec-
tively,namesforclasses(concepts),dataranges,objectproperties(abstract
roles),datatypeproperties(concreteroles),individuals(nominals)anddata
values;C(possiblysubscripted)representsanarbitraryclassdescription.
Eitherpartialorcompleteinformationcanbestatedaboutaclass,asin
Class(ex:Countrypartialowl:Thing)
Class(ex:Personpartialowl:Thing)
Class(ex:Studentpartialex:Person)
16I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
AbstractSyntaxDLSyntax
Class(ApartialC1
...C
n)
AC1
...Cn
Class(AcompleteC1
...C
n)
A≡C1
...Cn
EnumeratedClass(Ao
1
...o
n)
A≡{o1
}...{on}
SubClassOf(C1
C2)
C1
C2
EquivalentClasses(C1
...C
n
)
C1
≡...≡Cn
DisjointClasses(C1
...C
n
)
Ci
Cj
⊆⊥,i

Dj
Datatype(D)
ObjectProperty(Rsuper(R1)...super(Rn)
RRi
domain(C1)...domain(Cm
)
￿1RCi
range(C1)...range(C)
∀R.C
i
[inverseOf(R0)]
R≡R−
0
[Symmetric]
R≡R−
[Functional]
￿1R
[InverseFunctional]
￿1R−
[Transitive])
Tr(R)
SubPropertyOf(R1
R2
)
R1
R2
EquivalentProperties(R1
...R
n)
R1
≡...≡Rn
DatatypeProperty(Usuper(U1)...super(Un)
UUi
domain(C1)...domain(Cm
)
￿1UCi
range(D1)...range(D
)
∀U.Di
[Functional])
￿1U
SubPropertyOf(U1
U2)
U1
U2
EquivalentProperties(U1
...U
n)
U1
≡...≡Un
AnnotationProperty(S)
OntologyProperty(S)
Individual(otype(C1
)...type(Cn
)
o∈Ci
value(R1
o1
)...value(Rn
on)
o,o
i ∈Ri
value(U1
v1)...value(Un
vn))
o,v
i ∈Ui
SameIndividual(o1
...o
n)
{o1}≡...≡{on
}
DifferentIndividuals(o1
...o
n)
{oi}¬{oj},i

Dj
Fig.14.3.OWLDLAxiomsandFacts
Class(ex:Canadiancompleteex:Person
hasValue(ex:nationalityex:Canada))
whichmakesex:Countryandex:Personclasses,ex:Studentasub-
classofex:Person,andex:Canadianpreciselythosepeoplewhohave
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb17
ex:nationalityex:Canada.4
InOWLDLpropertiesaredividedintoobjectproperties,like
ex:nationality,whichrelateindividualstootherindividuals,datatype
properties,likeex:age,whichrelateindividualstodatavalues,andannota-
tionproperties,whichcanbeusedtoadduninterpretedinformation(suchas
versioninginformation)toindividuals,classes,andproperties.Constraints,
suchasdomainsandranges,canbegivenforobjectpropertiesanddatatype
properties,butnotforannotationproperties.
DatatypeProperty(ex:agedomain(ex:Person)
range(xsd:integer))
ObjectProperty(ex:nationalitydomain(ex:Person)
range(ex:Country))
Objectpropertiescanalsobespecifiedtobetransitive,symmetric,func-
tional,andinversefunctional.Notallofthesecanbespecifiedforaparticu-
larobjectproperty,astoretaindecidabilityofOWLDLpropertiesthatare
specifiedasbeingtransitive,andtheirsuper-propertiesandtheirinverses,
cannothavetheircardinalityrestricted,eitherviathefunctionalpartof
propertyaxiomsorincardinalityrestrictions(see
[
Horrocksetal.,1999
]
).
Annotationpropertiesareawayofassociatinguninterpretedinforma-
tionwithclasses,properties,andindividuals.Thesyntaxforannota-
tionsisnotgiveninFigure14.3.Manyaxioms(Class,EnumeratedClass,
Datatype,ObjectProperty,DatatypeProperty,AnnotationProperty,
OntologyProperty,andIndividualaxioms)canhaveanannotationthat
providesuninterpretedinformationaboutthatclass,property,orindividual.
TopreventannotationsinfluencingthesemanticsofOWLDL,littlecanbe
saidaboutannotationproperties:theycannotparticipateinrestrictions,
norcantheybegivendomains,ranges,oranyotheraspectofotherkinds
ofproperties.Theselimitationsresultinnonon-trivialinferencerelatedto
annotationpropertiesinOWLDL.
AnyaxiomthatstartswithaURIreferencecanhaveanannotationthat
providesuninterpretedinformationaboutthatURIreference.Toprevent
undueinfluencesonthesemanticsofOWLDL,littlecanbesaidabout
annotationproperties.Annotationpropertiescannotparticipateinrestric-
tions,norcantheybegivendomains,ranges,oranyotheraspectofother
kindsofproperties.
4
ForamoreextensiveexampleofhowtouseOWL,seetheOWLGuide
[
Smithetal.,2004
]
.
18I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
Oneuseofannotationpropertiesistoprovidecommentinformationfor
classesandproperties,asin
Class(ex:Countrypartial
annotation(rdfs:comment"Countriesoftheworld")
owl:Thing)
Thisis,perhaps,themostusefulpurposeforannotationsinOWLDL.
InformationaboutindividualscanalsobeprovidedinOWL,foreither
namedindividualsoranonymousindividuals,asin:
Individual(ex:Canadatype(ex:Country))
Individual(ex:Englandtype(ex:Country))
Individual(ex:Petertype(ex:Canadian)
value(ex:age"48"^^xsd:integer))
Individual(value(ex:nationalityex:England)
value(ex:age"44"^^xsd:integer))
InOWL,axiomsandfactsaregroupedintoontologies,withtheresultthat
anOWLDLontologyisequivalenttoaDescriptionLogicKnowledgeBase,
i.e.,aTboxplusanAbox(seeSection??).Thisisnotcompletelystandard,
asontologiesaremoretypicallythoughtofasdescribingonlythestructure
ofadomain(intermsofclassesandproperties),andnotasdescribinga
particularsituation(intermsofinstancesofclassesandproperties);inthis
morecommonusage,anontologyisthereforeequivalenttoaDescription
LogicTbox,andnottothecombinationofaTboxandanAbox.
Ontologiescanbegivenannotations,justlikeclasses,properties,andin-
dividuals.Ontologiescanalsohaveontologypropertiesasannotations.On-
tologypropertiesarejustlikeannotationsexceptthattheyrelateontologies
tootherontologies.Mostontologypropertiesactjustlikeannotationprop-
erties(i.e.,theyprovideuniterpretedinformation),butonespecialproperty,
owl:imports,canbeusedto“import”otherontologies.Importinganon-
tologyeffectivelytreatsthecontentofawebpageaspartofthecurrent
ontology,asin:
Ontology(Simple
Annotation(rdfs:comment"Asimpleontologyfornationality")
Annotation(owl:importshttp://www.foo.ex/simpler.owl)
....
)
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb19
Themeaningoftheaboveimportsstatementisthatthesetofaxiomsin
theSimpleontologyistakentoincludethesetofaxiomsinthesimpler.owl
ontology.
14.3.4SemanticsforOWLDL
Aformalsemantics,verysimilartothesemanticsprovidedforDescription
Logics(seeSection??),isprovidedforthisstyleofusingOWL.Fulldetails
onthismodeltheorycanbefoundintheOWLSemanticsandAbstract
Syntax
[
Patel-Schneideretal.,2004
]
.
BecauseOWLincludesdatatypes,thesemanticsforOWLisverysimilar
tothatofDescriptionLogicsthatalsoincorporatedatatypes,inparticular
SHOQ(D).However,theparticulardatatypesusedinOWLaretakenfrom
RDFandXMLSchemaDatatypes
[
BironandMalhotra,2001
]
.Datavalues
suchas"44"^^xsd:integerthusmeanwhattheywouldmeanasXML
Schemadatavalues.
ThesemanticsforOWLDLdoesincludesomeaspectsthatmaybeviewed
asunusualfromadescriptionlogicperspective.Annotationsaregivena
simpleseparatemeaning,notshownhere,thatcanbeusedtoassociatein-
formationwithclasses,properties,andindividualsinamannercompatible
withtheRDFsemantics.Ontologiesalsolivewithinthesemanticsandcan
begivenannotationinformation.Finally,owl:importsisgivenameaning
thatinvolvesfindingthereferencedontology(ifpossible)andaddingits
meaningtothemeaningofthecurrentontology.Inotherrespects,however,
themeaningofanOWLontologyshouldbeexactlyequivalenttothemean-
ingoftheDescriptionLogicknowledgebasederivedviathecorrespondences
giveninFigures14.2and14.3.
WhatmakesOWLDL(andOWLLite)aSemanticWeblanguage,there-
fore,isnotitssemantics,whicharequitestandardforaDescriptionLogic,
butinsteadtheuseofURIreferencesfornames,theuseofXMLSchema
datatypesfordatavalues,andtheabilitytoconnecttodocumentsinthe
WorldWideWeb.
14.3.5OWLLiteandOWLFull
Aswehaveseen,OWLDLisrelatedtoSHOIN(D),averyexpressive
DescriptionLogic.ThisDescriptionLogicissomewhatdifficulttopresent
tonaiveusers,asitispossibletobuildcomplexbooleandescriptionsusing,
20I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
forexample,unionandcomplement.SHOIN(D)isalsodifficulttoreason
with,askeyinferenceproblemshaveNExpTimecomplexity,andsomewhat
difficulttobuildevennon-reasoningtoolsfor,becauseofthecomplexde-
scriptions.
Forthesereasons,asubsetofOWLDLhasbeenidentifiedthatshould
beeasieronalltheabovemetrics;thissubsetiscalledOWLLite.OWL
Liteprohibitsunionsandcomplements,restrictsintersectionstotheimplicit
intersectionsintheframe-likeclassaxioms,limitsallembeddeddescriptions
toclassnames,doesnotallowindividualstooccurindescriptionsorclass
axioms,andlimitscardinalitiesto0or1.
TheserestrictionsmakeOWLLiteexpressivelyequivalenttoSHIF(D).
AsinSHIF(D),keyinferencesinOWLLitecanbecomputedinworstcase
exponentialtime(ExpTime),andtherearealreadyseveraloptimizedreason-
ersforlogicsequivalenttoOWLLite(seeSection14.3.7).Thisimprovement
intractabilitycomeswithrelativelylittlelossinexpressivepower—although
OWLLitesyntaxismorerestrictedthanthatofOWLDL,itisstillpossible
toexpresscomplexdescriptionsbyintroducingnewclassnamesandexploit-
ingtheimplicitnegationsintroducedbydisjointnessaxioms
[
Horrocksand
Patel-Schneider,2003
]
.Usingthesetechniques,allOWLDLdescriptionscan
becapturedinOWLLiteexceptthosecontainingeitherindividualnames
orcardinalitiesgreaterthan1.
OWLDLandOWLLiteareextensionsofarestricteduseofRDFand
RDFS,because,unlikeRDFandRDFS,theydonotallowclassestobe
usedasindividuals,andthelanguageconstructorscannotbeappliedtothe
languageitself.Foruserswhowantthesecapabilities,aversionofOWLthat
isupwardcompatiblewithRDFandRDFShasbeenprovided;thisversion
iscalledOWLFull.InOWLFull,allRDFandRDFScombinationsare
allowed.Forexample,inOWLFull,itispossibletoimposeacardinality
constraintonrdfs:subClassOf,ifsodesired.
OWLFullcontainsOWLDL,butgoeswelloutsidethestandardDe-
scriptionLogicframework.Thepenaltytobepaidhereistwo-fold.First,
reasoninginOWLFullisundecidable.(Showingtheundecidabilityistrivial,
becauserestrictionsrequiredinordertomaintainthedecidabilityofOWL
DLdonotapplytoOWLFull
[
Horrocksetal.,1999
]
,butalsoasaresultof
theabilitytoapplyOWL’sexpressivepowertoRDFsyntax,asexemplified
above
[
Motik,2005
]
.)Second,theabstractsyntaxforOWLDLisinadequate
forOWLFull,andtheofficialOWLexchangesyntax,RDF/XML,mustbe
used.
OWLFullhasbeengivenamodel-theoreticsemanticsthatisavocabulary
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb21
extensionoftheRDFmodeltheory
[
Patel-Schneideretal.,2004;Hayes,
2004
]
.AcorrespondencebetweenthissemanticsandthesemanticsofOWL
DLhasalsobeenestablished:ithasbeenshownthatthemodeltheoryfor
OWLDLhasverysimilarconsequencestothisRDF-stylemodeltheory
forthoseOWLontologiesthatcanbewrittenintheOWLDLabstract
syntax
[
Patel-Schneideretal.,2004
]
.
Moreformally,giventwoOWLDLontologiesO1
andO2,writteninthe
abstractsyntax,ifO1
entailsO2
accordingtotheOWLDLmodeltheory
thenthemappingofO
1
intoRDFtripleswillentailthemappingofO2
into
RDFtriplesaccordingtotheOWLFullmodeltheory.
Theconverse,however,isnottrue.Althoughthecorrespondenceisusually
exact,ithasbeenshownthatthereareatleastsomepathologicalcaseswhere
thecorrespondencebreaksdown.Forexample,inOWLFullowl:Thing
containsindividualscorrespondingtotheRDFSandOWLvocabularies
(suchasrdf:Property,rdfs:subClassOf,andowl:hasValue).(Thereareno
suchindividualsinthemodeltheoryforOWLDL.)Limitingtheextentof
owl:Thing,forexampleasin
SubClassOf(owl:ThingoneOf(ex:fooex:bar))
forcesequalitiesinthisvocabulary,whichhavetruelyunusualeffects,inthe
OWLFullmodeltheory.
InordertoavoidanypossibleconfusionastothemeaningofOWLDL,
theOWLFullmodeltheoryhasbeengiven“non-normative”status(i.e.,it
isonlyinformative)forOWLontologiesthatcanbewrittenintheabstract
syntax—forsuchontologies,thedefinitivesemanticsisgivenbytheOWL
DLmodeltheory.
14.3.6OWLDatatypes
Aswellasdealingwith“abstract”classessuchasPersonandAnimal,many
practicalapplicationsneedtorepresentandreasonaboutdatatypesand
valuessuchasintegersandstrings.Theintegrationofdatatypesinthe
OWLlanguageisagainheavilyinfluencedbyDescriptionLogicresearch,
whichhasdemonstratedthatcareisrequiredinordertoavoidcomplexity
blow-upsorevenundecidabilitybeingcausedbydatatypes
[
Lutz,2002
]
.In
theSHOQ(D)DescriptionLogicitwasshownthatthiscouldbeachieved
bystrictlyseparatingtheinterpretationofdatatypesandvaluesfromthat
ofclassesandindividuals:SHOQ(D)interpretationsincludeanadditional
interpretationdomainfordatavaluesI
D
whichisdisjointfromthedomain
ofindividualsI.Datatypes,suchasinteger,areinterpretedasasubset
22I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
ofI
D
,andvaluessuchastheinteger“35”areinterpretedaselementsof
I
D.Theseparationisfurtherstrengthenedbydividingpropertiesintotwo
disjointsetsofabstractanddatatypeproperties.Abstractpropertiessuchas
brotherareinterpretedasbinaryrelationsonI
(i.e.,subsetsofI
×I
),
whiledatatypepropertiessuchasageareinterpretedasbinaryrelations
betweenI
andI
D
(i.e.,subsetsofI
×I
D
).
Thisdesignhastheadvantagethatreasoningwithdatatypesandval-
uescanbealmostentirelyseparatedfromreasoningwithclassesand
individuals—aclassbasedreasonersimplyneedsaccesstoadatatype“or-
acle”thatcananswersimplequestionswithrespecttodatatypesandval-
ues(e.g.,“is-5anonNegativeInteger?”).Moreover,thelanguageremains
decidableifdatatypeandvaluereasoningisdecidable,i.e.,iftheoracle
canguaranteetoanswerallquestionsoftherelevantkindforsupported
datatypes.Thiscaneasilybeachievedforarangeofcommondatatypes
suchasintegers,decimalsandstrings
[
Lutz,2002
]
.
Aswellasthesepracticalconsiderations,itcanalsobearguedthatthe
separationofclassesanddatatypesmakessensefromaphilosophicalstand-
pointasdatatypesarealreadystructuredbybuiltinpredicatessuchas
greater-thanandless-than.Fromthispointofview,itdoesnotmakesense
touseontologyaxiomstoaddfurtherstructuretodatatypesortoform
“hybrid”classessuchastheclassofredintegers.
14.3.7Reasoners,toolsandapplications
AsdiscussedinSection14.3.1,animportantmotivationforthedesignof
OWLDLandOWLLitewastheabilityforapplicationstoexploitknown
reasoningalgorithmsandexisting(highlyoptimised)reasonerimplementa-
tions.Thismeantthat,evenbeforetheOWLspecificationwasfinalised,
prototypetoolsandapplicationscouldmakeuseofDescriptionLogicrea-
soningsystemssuchasFaCT
[
Horrocks,1998a
]
,Pellet
[
Pellet,2003
]
and
Racer
[
HaarslevandM¨oller,2001
]
.TheuseofFaCTandRacerwasalso
facilitatedbythefactthatbothsystemsprovideastandardapplication
interfacedesignedbytheDescriptionLogicImplementationGroup(DIG)
[
Bechhoferetal.,1999
]
.5
AtthetimewhenOWLachievedrecommendationstatus(thefinalstage
intheW3Cstandardisationprocess),theabovesystemswereonlyableto
supportOWLLitereasoning.Thiswasbecause,atthattime,nosuitable
algorithmwasknownforSHOIN,andalloftheimplementationswere
5
ThePelletsystemnowalsoprovidesaDIGcompliantinterface.
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb23
basedontheSHIQalgorithmsdescribedin
[
Horrocksetal.,1999
]
and
[
Horrocksetal.,2000b
]
.InordertoprocessOWLDLontologies,reasoning
systemstypicallyappliedsome“work-around”withrespecttonominals,
e.g.,bytreatingthemasprimitiveclasses.Thiswork-aroundissoundbut
incompleteforsubsumption.I.e.,givenontologiesOandO,whereO

has
beenderivedfromObyreplacingeachoccurrenceofanominalowitha
primitiveclassCo,aclassCissubsumedbyaclassDwithrespecttoOifC
issubsumedbyaDwithrespecttoO
,butifCisnotsubsumedbyaDwith
respecttoO,thenwecannotbesurethatCisnotsubsumedbyaDwith
respecttoO.Clearly,thework-aroundisunsoundforsatisfiability,i.e.,there
maybeconceptsthataresatisfiablewithrespecttoO
,butunsatisfiablewith
respecttoO.
Thissituationwasclearlyveryunsatisfactorygiventhemotivationfor
OWL’sDescriptionLogicbaseddesign,anditwasalwaysanticipatedthat
existingtableauxdecisionproceduresforSHIQandSHOQwouldsoonbe
extendedtoSHOIQ.Althoughthistookalittlelongerthananticipated,a
tableauxdecisionprocedureforSHOIQwaseventuallydeveloped
[
Horrocks
andSattler,2005
]
,andthePelletsystemnowusesthisalgorithmtofully
supportOWLDL.Itisexpectedthatotherreasoners,includingFaCT++
[
TsarkovandHorrocks,2005
]
(thesuccessoroftheFaCTsystem),willsoon
followsuit.
ItisinterestingtonotethatOWLisnowsupportedbycommercialDe-
scriptionLogicsystems.TheseincludetheCerebrasystemfromCerebraInc
(formerlyNetworkInference),andRacerPro,acommercialversionofthe
Racersystem.TherearealsoseveralOWLreasonersthatarenotbased
ontableauxdecisionprocedures.TheseincludeKAON2,asystemthatuses
areductionofSHIQ(D)todisjunctivedatalog
[
Hustadtetal.,2004
]
,and
Hoolet,asystemthatusestheVampireFirstOrderTheoremproverviaa
translationofSHOIN(D)intoFOL
[
Tsarkovetal.,2004
]
.
Thegrowingimportanceofontologies,andtheemergenceoftheOWL
standard,hasalsogivenimpetustothedevelopmentofontologyengineering
tools,includingtoolsforediting,validating,visualising,merginganddebug-
gingOWLontologies.SeveralApplicationProgrammingInterfaces(APIs)
forOWLarealsoavailable.
OftheavailableOWLEditingtools,probablythebestknownandmost
widelyusedisProt
´
eg
´
e
[
Gennarietal.,2003
]
.Prot
´
eg
´
eisaframebased
editorthatsupportsOWLviaanOWLPlugin.ThePluginusesarangeof
techniques(someofwhichwerefirstdevelopedintheOilEdeditor
[
Bech-
hoferetal.,2001b
]
)toextendthelanguagethatcanbedealtwith,e.g.,
24I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
byexplicitlyspecifyingquantificationwithslotsandallowingformultiple
necessaryandsufficientconditionsinclassdefinitions.Evenso,Prot
´
eg
´
e
stillhassomerestrictionswithrespecttotheOWLDLlanguage;itdoes
not,forexample,supportarbitraryclassesinrestrictions(butonlynamed
classes).Inadditiontotheframeeditor,thereareadditionalProt
´
eg
´
eplu-
ginssupporting,e.g.,ontologyvisualisation,ontologydocumentationand
“wizards”thatcanautomatesomebasicstepsintheontologydevelopment
process.Prot
´
eg
´
ecanconnecttoanyDescriptionLogicreasonerwitha
DIGcompliantinterface,andusesthereasonertocheckclassconsistency,
tocomputetheclasshierarchy,andtocomputethemostspecificclass(es)
thateachindividualisaninstanceof.
SeveralotherOWLeditingtoolsarealsoavailable.TheseincludeOilEd
[
Bechhoferetal.,2001b
]
(fromManchesterUniversity),SWOOP
[
Kalyan-
puretal.,2005a
]
(fromthePelletteam)andConstruct(fromCerebra).The
designofOilEdisbasedonthatofProt
´
eg
´
e,butitprovidesmorecomplete
supportforOWLDL.LikeProt
´
eg
´
e,OilEdcanuseanyDIGcompliant
reasonertoreasonovertheontology.SWOOPisbrowserbased,andismuch
moretightlylinkedtoOWL’ssyntacticstructure;itprovidesbothabstract
syntaxandRDF/XMLsyntaxeditingmodes,andfullysupportsOWLDL.
SWOOPusesthePelletsystemforreasoningsupport,andalsohasaninte-
grateddebugger
[
Kalyanpuretal.,2005b
]
.Constructisagraphicaltoolthat
usesaUMLlikenotation;itusestheCerebrareasonertoprovidereasoning
support.
OWLeditingtoolsarealsoexpandingintothesoftwareengineeringrealm,
astoolssuchasSandpiperSoftware’sMediusVisualOntologyModelersup-
portsontologydevelopmentusingUMLmodellingtoolswithoutputinOWL
(seehttp://www.sandsoft.com/).Thisenablesbroadercommunitiesto
model,edit,andintegratewithOWLontologies.
AlthoughOWLwasinitiallydevelopedasanontologylanguageforthe
SemanticWeb,OWLhasalsobeenwidelyadoptedasadefactostandard
forontologybasedapplications.Theseinclude“traditional”applicationsin
e-ScienceandMedicine,aswellasapplicationsinindustryandgovernment.
TheadvantagesofusingOWLinsuchapplicationsincludetherelativesta-
bilityandinteroperabilityconferredbyaW3Cstandard,andtheavailability
ofanexpandingrangeofreasonersandtools.
ExamplesofOWLontologyapplicationsinclude:
r
OntologiesdevelopedbymembersoftheOpenBiomedicalOntologiesConsortium
(seehttp://obo.sourceforge.net/),whichrecommendsOWLastheexchange
languageforallLifeScienceontologies.TheseincludethewidelyusedGeneOn-
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb25
tology(GO)andMicroarrayGeneExpressionData(MGED)ontology.
r
TheUSNationalCancerInstitute(NCI)“thesaurus”,anontologycontainingthe
workingvocabularyusedinNCIdatasystem(seehttp://ncicb.nci.nih.gov/
NCICB/core/EVS/).
r
UnitedNationsFoodandAgricultureOrganization(FAO)isusingOWLtode-
veloparangeofontologiescoveringareassuchasagricultureandfisheries(see
http://www.fao.org/agris/aos/Applications/intro.htm).
r
TheSemanticWebforEarthandEnvironmentalTerminology(SWEET)on-
tologiesdevelopedattheUSNationalAeronauticsandSpaceAdministra-
tion(NASA)JetPropulsionLaboratory(seehttp://sweet.jpl.nasa.gov/
ontology/).Theseincludeontologiesdescribingspace,thebiosphereandthe
sun.
r
AnontologyusedatGeneralMotorsinaprojecttohelpqualityimprovement
activitiesforassemblylineprocessesindifferentproductionsites
[
Morganetal.,
2005
]
.
14.4Summary
TheSemanticWebisenvisagedasanevolutionoftheexistingWebwhere
termsdefinedinontologieswillbeusedtogivewelldefinedandmachine-
processablemeaningtoWebaccessibleinformationandservices.OWL,a
DescriptionLogicbasedontologylanguage,hasbeendesignedforthispur-
pose.
BecauseoftheambitiousdesigngoalsforOWL,becauseofthemulti-
pleinfluencesonOWL,andalsobecauseofthestructuralrequirements
constrainingOWL,thedevelopmentofOWLhasnotbeenwithoutprob-
lems.Throughhardworkandcompromise,theseproblemshavelargelybeen
overcome,resultinginaontologylanguagethatistrulypartoftheSemantic
Web.
ItwasnotpossibletosimultaneouslysatisfyalloftheconstraintsonOWL,
sotwostylesofusingOWLhavebeendeveloped,eachappropriateunder
differentcircumstances.
OnestyleofOWLusageismotivatedbytheneedunambiguouslyrepresent
informationinanexpressivelanguage,butonethatcanstillbereasoned
withpredictably.Whenthisisaprimarygoal,OWLDLwillbethetarget
language,andithasbeenusedinanumberofexistingapplications.When
usingOWLDL,somecompatibilitywithRDFislost,mostlyhavingto
dowithusingclassesandpropertiesasindividuals.Ontheotherhand,
usersofOWLDLbenefitfromdecidableinference,andtheavailabilityof
anincreasinglywiderangeoftoolsandinfrastructure,includingefficient
reasoningsystemsandsophisticatedontologydevelopmentenvironments.
26I.Horrocks,P.F.Patel-Schneider,D.L.McGuinness,andC.Welty
OWLDLalsohasaframe-likealternativesyntaxthatcanbeusedtomake
workingwithOWLeasier.
EventhoughOWLDLis,essentially,adescriptionlogic,italsoin-
cludesfeaturesthatplaceitfirmlyintheSemanticWeb.OWLDLusesthe
datatypingmechanismsfromRDFandmanyofthebuilt-inXMLSchema
datatypes.OWLDLusesRDFURIreferencesasnames,includingthe
namesfromRDF,RDFS,andXMLSchemadatatypesthatarerelevant.
EntailmentinOWLDLiscompatiblewithentailmentinRDFandRDFS.
Foruserswhostillneedunambiguousrepresentationandpredictablerea-
soning,butforwhomsimplicityismoreimportantthanexpressivepower,
theOWLLitesubsetofOWLDLmaybeagoodchoice.Thissublanguage
rulesoutsomeofthethingsthatcanbesaidinOWLDL,butstillre-
tainsconsiderableexpressivepower.Moreover,OWLLiteissupportedbya
widerrangeofreasoningtools,andaskeyreasoningtasksareoflowerworst
casecomplexity,thesereasonersmightbeexpectedtobemoreefficient,in
general,thanOWLDLreasoners.
TheotherstyleofOWLusageisonewherecompatibilitywithRDFis
theoverarchingconcern.Inthiscase,OWLFullwouldbeanappropriate
choice.OWLFullextendsRDFandRDFStoafullontologylanguage,with
awell-specifiedentailmentrelationshipthatextendsentailmentinRDFand
RDFS,whileavoidinganyparadoxesthatmightarise.However,entailment
inOWLFullisundecidable,whichisasignificantissueinmostcircum-
stances,andnoeffectivetoolsforreasoningareavailableforOWLFull,nor
aretheyexpectedtoappear.Also,theuser-friendlyalternativesyntaxisnot
adequateforOWLFull,soRDF/XMLmustbeused.
Inpractice,relativelyfewOWLFullapplicationshaveemergedtodate,
andwhereOWLFullontologiesarefound,theyoftenturnouttobeoutside
theOWLDLsubsetonlyastheresultofminorsyntacticerrors.Fragments
ofOWLare,however,sometimesusedasadhocextensionstoRDFS.A
commonexampleistheuseofOWLfunctionalproperties,andexplicitequiv-
alencesand(in)equalities,inwhatwouldotherwisebeanRDFSontology.
Thereremain,ofcourse,significantissuesthataredeliberatelynothan-
dledbyOWL,butwhicharedefinitelyrelevanttomanySemanticWebuse
cases:
r
OWLavoidsanythingrelatedtononmonotonicity(suchasdefaultreasoningand
localisedclosedworldassumptions);
r
OWL’slimitedexpressivenessexcludesoperationssuchasproperty-chaining,or,
moregenerally,axiomswithvariables,suchasrules(althoughtherearealready
proposalsforextensionsinthisdirection
[
Horrocksetal.,2005;Eiteretal.,2004;
OWL:aDescriptionLogicBasedOntologyLanguagefortheSemanticWeb27
Motiketal.,2004
]
);
r
OWL’simportmechanismislimited,anddoesnotsupportfine-grainedoperations
(suchastheimportationofpartsofontologies);
r
OWLintegratesdatatypesinaverycleanway,butthereisnonotionofoperations
onthesedatatypes(suchintegerarithmeticorstringoperations).
ExtendingthecurrentSemanticWebwithsomeorallofthesefeatures
willrequirenotonlyastandardisationeffort,butsetsasignificantresearch
challengetothecommunity.
Bibliography
[
BaaderandHanschke,1991
]
FranzBaaderandPhilippHanschke.Aschemafor
integratingconcretedomainsintoconceptlanguages.InProc.ofthe12thInt.
JointConf.onArtificialIntelligence(IJCAI’91),pages452–457,1991.
[
Bechhoferetal.,1999
]
SeanBechhofer,IanHorrocks,PeterF.Patel-Schneider,
andSergioTessaris.Aproposalforadescriptionlogicinterface.InProc.of
the1999DescriptionLogicWorkshop(DL’99),pages33–36.CEUR
ElectronicWorkshopProceedings,http://ceur-ws.org/Vol-22/,1999.
[
Bechhoferetal.,2000
]
SeanBechhofer,JeenBroekstra,StefanDecker,Michael
Erdmann,DieterFensel,CaroleGoble,FrankvanHarmelen,IanHorrocks,
MichelKlein,DeborahL.McGuinness,EnricoMotta,PeterF.
Patel-Schneider,SteffenStaab,andRudiStuder.Aninformaldescriptionof
OIL-CoreandStandardOIL:alayeredproposalforDAML-O.Technical
ReportKSL-00-19,StanfordUniversityKSL,November2000.Availableat
http://www.ontoknowledge.org/oil/downl/dialects.pdf.
[
Bechhoferetal.,2001a
]
SeanBechhofer,CaroleGoble,andIanHorrocks.
DAML+OILisnotenough.InProc.ofthe2001Int.SemanticWebWorking
Symposium(SWWS2001),pages151–159,2001.Availableat
http://www.semanticweb.org/SWWS/program/full/SWWSProceedings.pdf.
[
Bechhoferetal.,2001b
]
SeanBechhofer,IanHorrocks,CaroleGoble,andRobert
Stevens.OilEd:AReason-ableontologyeditorforthesemanticweb.InProc.
oftheJointGerman/AustrianConf.onArtificialIntelligence(KI2001),
number2174inLectureNotesinArtificialIntelligence,pages396–408.
Springer,2001.AppearedalsoinProc.ofthe2001DescriptionLogic
Workshop(DL2001).
[
Bechhoferetal.,2004
]
SeanBechhofer,FrankvanHarmelen,JimHendler,Ian
Horrocks,DeborahL.McGuinness,PeterF.Patel-Schneider,and
LynnAndreaStein.OWLwebontologylanguagereference.W3C
Recommendation,10February2004.
[
Berners-Leeetal.,2001
]
TimBerners-Lee,JamesHendler,andOraLassila.The
semanticweb.ScientificAmerican,284(5):34–43,May2001.
[
BironandMalhotra,2001
]
PaulV.BironandAshokMalhotra.XMLschema
part2:Datatypes.W3CRecommendation,May2001.Availableat
http://www.w3.org/TR/xmlschema-2/.
[
Brachmanetal.,1999
]
RonaldJ.Brachman,AlexanderBorgida,DeborahL.
McGuinness,andPeterPatel-Schneider.ReducingCLASSICtopractice:
28
Bibliography29
Knowledgerepresentationtheorymeetsreality.ArtificialIntelligence,
114(1–2):203–237,1999.
[
Chaudhrietal.,1998
]
VinayK.Chaudhri,AdamFarquhar,RichardFikes,
PeterD.Karp,andJamesRice.OKBC:Aprogrammaticfoundationfor
knowledgebaseinteroperability.InProc.ofthe15thNat.Conf.onArtificial
Intelligence(AAAI’98),pages600–607,1998.
[
Deckeretal.,2000
]
StefanDecker,DieterFensel,FrankvanHarmelen,Ian
Horrocks,SergeyMelnik,MichelKlein,andJeenBroekstra.Knowledge
representationontheweb.InProc.ofthe2000DescriptionLogicWorkshop
(DL2000),pages89–97.CEURElectronicWorkshopProceedings,
http://ceur-ws.org/Vol-33/,2000.
[
Eiteretal.,2004
]
ThomasEiter,ThomasLukasiewicz,RomanSchindlauer,and
HansTompits.Combininganswersetprogrammingwithdescriptionlogics
forthesemanticweb.InProc.ofthe9thInt.Conf.onPrinciplesof
KnowledgeRepresentationandReasoning(KR2004),pages141–151.Morgan
Kaufmann,LosAltos,2004.
[
Farquharetal.,1996
]
AdamFarquhar,RichardFikes,andJamesRice.The
ontolinguaserver:Atoolforcollaborativeontologyconstruction.In
ProceedingsoftheTenthKnowledgeAcquisitionforKnowledge-Based
SystemsWorkshop,November1996.AlsoavailableasStanfordUniversity
KnowledgeSystemsLabTechReport:KSL-TR-96-26.
[
Fenseletal.,2000
]
DieterFensel,IanHorrocks,FrankvanHarmelen,Stefan
Decker,MichaelErdmann,andMichelKlein.OILinanutshell.InR.Dieng,
editor,Proc.ofthe12thEuropeanWorkshoponKnowledgeAcquisition,
Modeling,andManagement(EKAW2000),number1937inLectureNotesin
ArtificialIntelligence,pages1–16.Springer,2000.
[
Fenseletal.,2001
]
DieterFensel,FrankvanHarmelen,IanHorrocks,DeborahL.
McGuinness,andPeterF.Patel-Schneider.OIL:Anontologyinfrastructure
forthesemanticweb.IEEEIntelligentSystems,16(2):38–45,2001.
[
FikesandMcGuinness,2001
]
RichardE.FikesandDeborahL.McGuinness.An
axiomaticsemanticsforRDF,RDFSchema,andDAML+OIL.Technical
ReportKSL-01-01,StanfordUniversityKSL,2001.Availableat
http://www.ksl.stanford.edu/people/dlm/daml-semantics/
abstract-axiomatic-semantics.html.
[
Gennarietal.,2003
]
JohnH.Gennari,MarkA.Musen,WilliamE.Grosso,
MonicaCrub´ezy,HenrikEriksson,NatalyaF.Noy,andSamsonW.Tu.The
evolutionofprot´eg´e.InternationalJournalofHumanComputerStudies,
58(1):89–123,2003.
[
Grossoetal.,1999
]
W.E.Grosso,H.Eriksson,R.W.Fergerson,J.H.Gennari,
S.W.Tu,andM.A.Musen.Knowledgemodellingatthemillenium(The
designandevolutionofProt´eg´e-2000).InProc.ofKnowledgeacqusition
workshop(KAW’99),1999.
[
HaarslevandM¨oller,2001
]
VolkerHaarslevandRalfM¨oller.RACERsystem
description.InProc.oftheInt.JointConf.onAutomatedReasoning
(IJCAR2001),volume2083ofLectureNotesinArtificialIntelligence,pages
701–705.Springer,2001.
[
Hayes,2004
]
PatrickHayes.RDFmodeltheory.W3CRecommendation,10
February2004.
[
Heflinetal.,2003
]
JeffHeflin,JamesHendler,andSeanLuke.SHOE:A
blueprintforthesemanticweb.InD.Fensel,J.Hendler,H.Lieberman,and
30bibliography
W.Wahlster,editors,SpinningtheSemanticWeb.MITPress,Cambridge,
MA,2003.
[
HendlerandMcGuinness,2000
]
JamesHendlerandDeborahL.McGuinness.The
darpaagentmarkuplanguage”.IEEEIntelligentSystems,15(6):67–73,2000.
[
HorrocksandPatel-Schneider,2003
]
IanHorrocksandPeterF.Patel-Schneider.
ReducingOWLentailmenttodescriptionlogicsatisfiability.InDieterFensel,
KatiaSycara,andJohnMylopoulos,editors,Proc.ofthe2003International
SemanticWebConference(ISWC2003),number2870inLectureNotesin
ComputerScience,pages17–29.Springer,2003.
[
HorrocksandSattler,2001
]
IanHorrocksandUlrikeSattler.Ontologyreasoning
intheSHOQ(D)descriptionlogic.InProc.ofthe17thInt.JointConf.on
ArtificialIntelligence(IJCAI2001),pages199–204,2001.
[
HorrocksandSattler,2005
]
IanHorrocksandUlrikeSattler.Atableauxdecision
procedureforSHOIQ.InProc.ofthe19thInt.JointConf.onArtificial
Intelligence(IJCAI2005),2005.
[
Horrocksetal.,1999
]
IanHorrocks,UlrikeSattler,andStephanTobies.Practical
reasoningforexpressivedescriptionlogics.InHaraldGanzinger,David
McAllester,andAndreiVoronkov,editors,Proc.ofthe6thInt.Conf.on
LogicforProgrammingandAutomatedReasoning(LPAR’99),number1705
inLectureNotesinArtificialIntelligence,pages161–180.Springer,1999.
[
Horrocksetal.,2000a
]
I.Horrocks,D.Fensel,J.Broekstra,S.Decker,
M.Erdmann,C.Goble,F.vanHarmelen,M.Klein,S.Staab,R.Studer,and
E.Motta.OIL:TheOntologyInferenceLayer.TechnicalReportIR-479,Vrije
UniversiteitAmsterdam,FacultyofSciences,September2000.
[
Horrocksetal.,2000b
]
IanHorrocks,UlrikeSattler,andStephanTobies.
ReasoningwithindividualsforthedescriptionlogicSHIQ.InDavid
McAllester,editor,Proc.ofthe17thInt.Conf.onAutomatedDeduction
(CADE2000),volume1831ofLectureNotesinComputerScience,pages
482–496.Springer,2000.
[
Horrocksetal.,2003
]
IanHorrocks,PeterF.Patel-Schneider,andFrankvan
Harmelen.FromSHIQandRDFtoOWL:Themakingofawebontology
language.J.ofWebSemantics,1(1):7–26,2003.
[
Horrocksetal.,2005
]
IanHorrocks,PeterF.Patel-Schneider,SeanBechhofer,
andDmitryTsarkov.OWLrules:Aproposalandprototypeimplementation.
J.ofWebSemantics,3(1):23–40,2005.
[
Horrocks,1998a
]
IanHorrocks.TheFaCTsystem.InHarriedeSwart,editor,
Proc.ofthe2ndInt.Conf.onAnalyticTableauxandRelatedMethods
(TABLEAUX’98),volume1397ofLectureNotesinArtificialIntelligence,
pages307–312.Springer,1998.
[
Horrocks,1998b
]
IanHorrocks.Usinganexpressivedescriptionlogic:FaCTor
fiction?InProc.ofthe6thInt.Conf.onPrinciplesofKnowledge
RepresentationandReasoning(KR’98),pages636–647,1998.
[
Hustadtetal.,2004
]
UllrichHustadt,BorisMotik,andUlrikeSattler.Reducing
SHIQ-descriptionlogictodisjunctivedatalogprograms.InProc.ofthe9th
Int.Conf.onPrinciplesofKnowledgeRepresentationandReasoning
(KR2004),pages152–162,2004.
[
Kalyanpuretal.,2005a
]
AdityaKalyanpur,BijanParsia,andJamesHendler.A
toolforworkingwithwebontologies.InternationalJournalonSemanticWeb
andInformationSystems,1(1):36–49,2005.
[
Kalyanpuretal.,2005b
]
AdityaKalyanpur,BijanParsia,EvrenSirin,andJames
Bibliography31
Hendler.Debuggingunsatisfiableclassesinowlontologies.InJournalofWeb
Semantics—SpecialIssueoftheSemanticWebTrackofWWW2005,2005.
(ToAppear).
[
Karpetal.,1999
]
PeterD.Karp,VinayK.Chaudhri,andJeromeThomere.
XOL:AnXML-basedontologyexchangelanguage.TechnicalReportSRIAI
TechnicalNote559,SRIInternational,MenloPark(CA,USA),1999.
[
Lutz,2002
]
CarstenLutz.TheComplexityofReasoningwithConcreteDomains.
PhDthesis,TeachingandResearchAreaforTheoreticalComputerScience,
RWTHAachen,2002.
[
Mallery,1994
]
JohnMallery.ACommonLISPhypermediaserver.InProc.ofthe
1stInt.Conf.onTheWorld-WideWeb.CERN,1994.Availableathttp:
//www.ai.mit.edu/projects/iiip/doc/cl-http/server-abstract.html.
[
ManolaandMiller,2004
]
FrankManolaandEricMiller.RDFprimer.W3C
Recommendation,10February2004.
[
McGuinnessetal.,1997
]
DeborahL.McGuinness,HarleyManning,andTom
Beattie.Knowledgeaugmentedintranetsearch.InProc.ofthe6thWorld
WideWebConferenceCDROMversion,1997.
[
McGuinnessetal.,2000
]
DeborahL.McGuinness,RichardFikes,JamesRice,
andSteveWilder.Anenvironmentformergingandtestinglargeontologies.
InProc.ofthe7thInt.Conf.onPrinciplesofKnowledgeRepresentationand
Reasoning(KR2000),pages483–493,2000.
[
McGuinnessetal.,2002
]
DeborahL.McGuinness,RichardFikes,LynnA.Stein,
andJamesHendler.DAML-ONT:Anontologylanguageforthesemantic
web.InDieterFensel,Hendler,HenryLieberman,andWolfgangWahlster,
editors,TheSemanticWeb:Why,What,andHow.TheMITPress,2002.
[
McGuinness,1998
]
DeborahL.McGuinness.Ontologicalissuesfor
knowledge-enhancedsearch.InProceedingsofFormalOntologyin
InformationSystems.,1998.AlsopublishedinFrontiersinArtificial
IntelligenceandApplications,IOS-Press,1998.
[
Meghinietal.,1997
]
CarloMeghini,FabrizioSebastiani,andUmbertoStraccia.
Modellingtheretrievalofstructureddocumentscontainingtextsandimages.
InCostantinoThanos,editor,Proc.ofthe1stEuropeanConf.onResearch
andAdvancedTechnologyforDigitalLibraries(ECDL’97),volume1324of
LectureNotesinComputerScience.Springer,1997.
[
Miller,1995
]
GeorgeA.Miller.WordNet:AlexicaldatabaseforEnglish.
CommunicationsoftheACM,38(11):39–41,1995.
[
Morganetal.,2005
]
AlexanderP.Morgan,JohnA.Cafeo,KurtGodden,
RonaldM.Lesperance,AndreaM.Simon,DeborahL.McGuinness,and
JamesL.Benedict.Thegeneralmotorsvariation-reductionadviser.AI
Magazine,26(2),2005.
[
Motiketal.,2004
]
BorisMotik,UlrikeSattler,andRudiStuder.Query
answeringforOWL-DLwithrules.InProc.ofthe2004International
SemanticWebConference(ISWC2004),pages549–563,2004.
[
Motik,2005
]
BorisMotik.OnthepropertiesofmetamodelinginOWL.InProc.
ofthe2005InternationalSemanticWebConference(ISWC2005),Lecture
NotesinComputerScience,pageToappear.Springer,2005.
[
Patel-Schneideretal.,2004
]
PeterF.Patel-Schneider,PatrickHayes,andIan
Horrocks.OWLwebontologylanguagesemanticsandabstractsyntax.W3C
Recommendation,10February2004.
[
Pellet,2003
]
PelletOWLreasoner.MarylandInformationandNetworkDynamics
32bibliography
Lab,2003.
[
Rousset,1999
]
Marie-ChristineRousset.Queryexpansionindescriptionlogics
andCARIN.InWorkingNotesoftheAAAIFallSymposiumon“Question
AnsweringSystems”,1999.
[
Smithetal.,2004
]
MichaelK.Smith,ChrisWelty,andDeborahL.McGuinness.
OWLwebontologylanguageguide.W3CRecommendation,10February
2004.
[
TsarkovandHorrocks,2005
]
DmitryTsarkovandIanHorrocks.Ordering
heuristicsfordescriptionlogicreasoning.InProc.ofthe19thInt.JointConf.
onArtificialIntelligence(IJCAI2005),2005.
[
Tsarkovetal.,2004
]
DmitryTsarkov,AlexandreRiazanov,SeanBechhofer,and
IanHorrocks.UsingVampiretoreasonwithOWL.InSheilaA.McIlraith,
DimitrisPlexousakis,andFrankvanHarmelen,editors,Proc.ofthe2004
InternationalSemanticWebConference(ISWC2004),number3298in
LectureNotesinComputerScience,pages471–485.Springer,2004.
[
vanHarmelenetal.,2001
]
FrankvanHarmelen,PeterF.Patel-Schneider,and
IanHorrocks.Amodel-theoreticsemanticsforDAML+OIL(March2001),
March2001.
[
WeltyandJenkins,2000
]
ChristopherWeltyandJessicaJenkins.Untangle:a
newontologyforcardcatalogsystems.InProc.ofthe17thNat.Conf.on
ArtificialIntelligence(AAAI2000),pages1137–1138.AAAIPress/TheMIT
Press,2000.
[
Welty,1994
]
ChristopherWelty.Aknowledge-basedemaildistributionsystem.In
Proc.ofthe1994FloridaAIResearchSymposium.AAAIPress/TheMIT
Press,May1994.
[
Welty,1996a
]
ChristopherWelty.AnHTMLinterfaceforClassic.InProc.ofthe
1996DescriptionLogicWorkshop(DL’96),numberWS-96-05inAAAI
TechnicalReport.AAAIPress/TheMITPress,1996.
[
Welty,1996b
]
ChristopherWelty.Intelligentassistancefornavigatingtheweb.In
Proc.ofthe1996FloridaAIResearchSymposium.AAAIPress/TheMIT
Press,May1996.
[
Welty,1998
]
ChristopherWelty.Theontologicalnatureofsubjecttaxonomies.In
Proc.oftheInt.Conf.onFormalOntologyinInformationSystems
(FOIS’98),FrontiersinArtificialIntelligence.IOSPress,1998.