Y. Domain ontologies a database-oriented analysis

DOMAINONTOLOGIES:ADATABASE-ORIENTEDANALYSIS

St´ephaneJean,GuyPierra,YamineAit-Ameur

LaboratoryofAppliedComputerScience(LISI)

NationalEngineeringSchoolforMechanicsandAerotechnics(ENSMA)-Poitiers

86960FuturoscopeCedex-FRANCE

Email:{jean,pierra,yamine}@ensma.fr

Keywords:Abstract:

Ontology,SemanticWeb,OntologyBasedInformationSystems,SemanticIntegration,DataExchange,PLIB.Ifthewordontologyismoreandmoreusedinanumberofdomain,thecapabilitiesandbenefitsofontologyforInformationSystemsmanagementarestillunclear.Therefore,theusageofontology-basedInformationSystemsinindustryandservicesisnotwidespread.Thispaperanalysestheconceptofadomainontologyfromadatabaseperspective.Asaresult,firstly,weprovidethreecriteriathatdistinguishdomainontologyfromotherexistingdomainmodelingapproachwhichleadustoproposeanewdefinitionofdomainontolo-gies.Secondly,basedonthevariousapproachesofontologymodelingfollowedbydifferentcommunities,weproposeataxonomyofdomainontology.Weshowhowtheymaybeorganizedintoalayeredmodel,calledtheonionmodel,allowingtodesignandtousethecapabilitiesofeachcategoryofontologyinanintegratedenvi-ronment.Finally,thispaperpresentsseveralinformationsystemsbasedonontologytechnologiesanddescribethekindsofservicesthatshouldbeprovidedtoallowapowerfulusageofontologyindatamanagement.

1Introduction

DefinedbyT.Gruber(Gruber,1993)asanexplicitspecificationofaconceptualization,anontologymaybeconsideredasaquitenewandexcitingartefactincomputerscienceallowingtorepresentexplicitlymeaning.Nowadays,thewordontologyisusedinalotofdiverseresearchfieldsincludingnaturallan-guageprocessing,informationretrieval,electroniccom-merce,WebSemantic,softwarecomponentspecifica-tionandinformationsystemsintegration.Inthiscon-text,severalproposalsforontologymodelsandlan-guagesandcorrespondingoperationalsystemshavebeendevelopedinthelastdecade.Thegrowthofboththenumberandthediversityofsuchmodelsforontologiesleadstosomedifficultiesencounteredbyengineerswhentheyneedtoidentifytherightontol-ogy(ies)andtherightontologymodel(s)touseortoapplyinpracticalengineeringareas.

Duetothewidedomainofusage,themeaningofthewordontologyisofcoursecontext-dependent.Bor-rowedfromphilosophy,whereitstandfor”asys-tematicaccountofexistence”1,thetermontologygotaquitenewmeaningintechnicalandcomputersci-encefields.Inthisnewcontext,onemaydistinguish

1

upper-levelorfoundationontologies,thegoalofwhichistoprovidedefinitionforgeneral-purposeconcepts,suchthatprocess,objectorevent,andtoactasfoun-dationformorespecificdomainontologies(NilesandPease,2001;Gangemietal.,2003),anddomainon-tologiesthataretiedtoaspecificuniverseofdiscourseandmodelthecorrespondingdomainknowledge.Thegoalofthispaperistoanalysetheconceptofadomainontologyinadatabaseperspective.Mostoftheusualdefinitions,suchthattheoneoftheFreeOn-lineDictionaryofComputing”anexplicitformalspecificationofhowtorepresenttheobjects,conceptsandotherentitiesthatareassumedtoexistinsomeareaofinterestandtherelationshipsthatholdamongthem”aresobroadthattheycoversmostofthepre-viousinformationmodellingartefactssuchthatcon-ceptualmodels,knowledgemodelorspecificationofinformationexchangeformats.Asaresultthehighpotentialofontologiesforsemanticintegrationishid-den,andanumberofengineersconsiderontologyasabuzzword.

Inthispaper,wesuggestthattheabovedefinitionshouldberefinedandweproposethreecriteriatodis-tinguishdomainontologiesfromotherinformationmod-ellingartefacts.Suchdomainontologiesintroduceanewmodellinglevelinthedatabasefieldandwepro-

FOLDOC:http://foldoc.org

poseataxonomyofthevariouspossibledomainon-tologiestogetherwithintegrationscenariosthatshowhowthistaxonomymaybehelpfulforaddressingvar-iousdatamanagementissues.Thenwediscusswhatkindoftools,calledOntology-basedDataManage-mentSystem(OBDMS)wouldbeusefulforpromot-ingontologyusageinthedataprocessingcommunity.Itisexpectedthatthisanalysiswillpromotethede-velopmentofnewOBDMSandhelpengineersandpractitionerstochooserelevantOBDMSinordertosolvetheirbusinessproblems.Ourworkisdiffer-entfrompreviousrelatedwork(Cullotetal.,2003;Meersman,2001)aimingatclarifyingthedifferencesbetweenontologyanddatabasetechnologies.Themaincontributionofthispaperarethefollowing:•aproposalforcriteriathatdistinguishdomainon-tologiesfromotherdomainmodelingapproaches;•anewdefinitionofdomainontology;

•ataxonomyofdomainontologyandalayeredmodel(theonionmodel)thatshowshowthesedifferentkindsofontologymaycooperateforsolvingdataprocessingissues.

Thispaperisorganizedasfollows.Nextsectionpresentstheconceptofanontologybyfocussingonthreecriteriathatdistinguishontologyfromotherex-istingmodelingapproach.Thissuggestanewdefi-nitionofdomainontologies.Section3describesthevariouspossibleusagesofontologiesindatamanage-ment.Adatabase-orientedtaxonomyofontologiesisproposedinsection4andsection5proposesaninte-gratedviewofthesedifferentkindsofdomainontol-ogyforaddressingvariousdataprocessingproblems.Finally,section6describesseveralOBDMSandcom-parestheircapabilities.

2

Specificityofdomainontologyasdomainmodels

Weproposeinthissectionthreecriteriathatchar-acterizedomainontologies.Thesecriteriasuggestanewdefinitionofdomainontology.Finally,wedis-cussthedifferencebetweendomainontologiesandconceptualmodels.

2.1Ontologycriteria

Fromourpointofview,adomainontologyisado-mainconceptualizationobeyingtothethreefollowingcriteria.

formal.Anontologyisaconceptualizationbasedonaformaltheorywhichallowstochecksomelevelofconsistencyandtoperformsomelevelof

automaticreasoningovertheontology-definedcon-ceptsandindividuals.Wenotethatthiscriterionexcludesmostmeta-modelsthatdonotprovideau-tomaticreasoningcapabilities.

2.consensual.Anontologyisaconceptualizationagreeduponbyacommunitylargerthanthemem-bersinvolvedinoneparticularapplicationdevel-opment.Forinstance,TheGeneOntology(GO)project2isacollaborativeeffortbetweenmorethan10organismstoaddresstheneedforconsistentde-scriptionsofgeneproducts.Moreover,usersarein-vitedtosubmitsuggestionsforimprovingtheGOontologies.ISO13584-compliant(PLIB)productontologiesfollowaformalstandardizationprocessandarepublishedasISOorIECinternationalstan-dards.Wenotethatthiscriterionexcludesmostdatabase,conceptualmodelswhicharejusttailoredforaparticulardatabaseapplication.3.capabilitytobereferenced.Eachontology-definedconceptisassociatedwithanidentifierallowingtorefertothisconceptfromanyenvironment,inde-pendentlyoftheparticularontologymodelwherethisconceptwasdefined.Wenotethatthiscrite-rionexclude,inparticular,allspecificationofin-formationexchangeformats,suchthatSTEP(Stan-dardfortheExchangeofProductModelData)Ap-plicationProtocols(ISO10303,1994),whereenti-tiesandattributesmayonlybereferencedfromthespecifiedexchangestructure.

2.2

Aproposeddefinitionfordomainontology

Thesethreecriterialeadustoproposeanewdefini-tionfordomainontology.Forus,adomainontologyisaformalandconsensualdictionaryofcategoriesandpropertiesofentitiesofadomainandtherela-tionshipsthatholdamongthem.Byentitywemeanbeing,i.e,anythingthatcanbesaidtobeinthedo-main.Thetermdictionaryemphasizesthatanyentityoftheontologyandanykindofdomainrelationshipdescribedinthedomainontologymaybereferenceddirectly,foranypurposeandfromanycontext,inde-pendentlyofotherentitiesorrelationships,byasym-bol.Thisidentificationsymbolmaybeeitheralan-guageindependentidentifier,oralanguage-specificsetofwords.But,whateverbethesymbol,andun-likeinlinguisticdictionary,thissymboldenotesdi-rectlyadomainentityorrelationship,thedescriptionofwhichisformallystatedprovidingforautomaticreasoningandconsistencychecking.

Weshowinthenextsectionthatthecriteriausedforcharacterizingdomainontologiesallowtodistin-2

http://www.geneontology.org/

1.guishthemwithpreviouskindofconceptmodelinglikeconceptualmodelsandknowledgemodels.

2.3OntologiesVsconceptualmodels

Asbothanontologyandaconceptualmodeldefineaconceptualizationofapartoftheworld,anontol-ogyseemssimilartoaconceptualmodel.Concep-tualmodelsrespecttheformalcriterion.Indeed,aconceptualmodelisbasedonarigorouslyformal-izedlogicaltheoryandreasoningisprovidedbyviewmechanisms.However,aconceptualmodelisappli-cationrequirementdriven:itprescribesandimposeswhichinformationwillberepresentedinaparticu-larapplication(logicalmodel).Twodifferentappli-cationsystemshavingalwaysatleastslightlydiffer-entapplicationrequirements,conceptualmodelsarealwaysdifferentfromsystemstosystems.Thus,con-ceptualmodelsdonotfulfilltheconsensualcriterion.Moreover,anidentifierofaconceptualmodeldefinedconceptisanamethatcanonlybereferencedunam-biguouslyinsidethecontextofaninformationsystembasedonthisparticularconceptualmodel.Thus,con-ceptualmodelsalsodonotfulfillthecapabilitytobereferencedcriterion.

Inthesamemanner,aconceptualizationdefinedinKnowledgeRepresentationandArtificialIntelligenceusinglogicconstructorsarenot,ingeneral,anontol-ogy.Suchconceptualizationssatisfytheformalcrite-rion.Indeed,logicisequippedwithformalsemanticsthatenablesautomaticreasoning.However,insuchknowledgemodels,themaingoalsaretheinferencecapabilitiesofthemodels.Beforethatthenotionofanontologyemerged,neithermechanismsforreferenc-ingeachparticularconceptofaknowledgemodel,norprocessesforensuringaconsensusontheconceptswereconsidered.Therefore,likeconceptualmodels,suchusualknowledgemodelsdonotfulfillthecon-sensualandthecapabilitytobereferencedcriterion.However,datamodelconstructsissuedfromdatabasedesignandlogicaresuitableforontologymodelsdef-initions.Indeed,severalontologymodels,likeOWL(Bechhoferetal.,2004),RDFS(BrickleyandGuha,2004)andKAON(Bozsaketal.,2002)fordescriptionlogicandPLIB(Pierra,2003),DOGMA(JarrarandMeersman,2002)andMADS(Parentetal.,1999)fordatabasedesign,arebasedonconstructorsprovidedeitherbydatabaseconceptualmodelsorbyartificialintellligenceknowledgebasemodels.Thesemodelsaddotherconstructorsthatenabletosatisfythecon-sensualcriterion(contextdefinition,multiinstantia-tion,separationbetweenconceptdefinitionanddatastructureprescription)andthecapabilitytoberefer-encedcriterion(URI,GUI).

Onthebasisofthisdistinctionbetweenontologyandotherconceptsmodels,westudyinnextsectionwhatontologiesaregoodfor.

3Whatareontologiesgoodfor?

Asstatedintheintroduction,ontologytechnolo-giesiswidespreadinalotofdiverseapplicationdo-mainsanditmaybeusedinvariousengineeringstepslikespecification,dataexchange,dataintegrationandsearch.

3.1Specification

Twousagesofontologiesasspecificationarereported.TheusageofaconceptualizationasaspecificationisthebasisoftheModel-DrivenArchitecture(MDA).Amodeloftheapplicationisfirstdefined.Thismodelisthenusedtogeneratethecodeoftheapplication.Theexistingformallinkbetweenthespecificationandthesoftwareenablestoevolvethecodewhenthespec-ificationevolves.Currently,severalsoftwaresaddress-ingsimilarproblemsonthesamedomainaredefinedusingdifferentconceptualizations.Thismakesdiffi-cultinteroperationbetweenthesesoftwares.Ontol-ogyusageisasolutiontothisproblem.Becauseon-tologiessatisfytheconsensualcriterion,thevariousconceptualizationcorrespondingtovariousdomainsoft-waresmaybeconnectedtoadomainontology.Then,softwarescaninteroperateusingaccessorsprovidedbythisontology.Thisapproachiscalledontology-drivensoftwareengineering(Tetlowetal.,2005).Thesameapproachcanbefollowedindatabasede-sign.ThepropositionofOntology-BasedDatabaseapproach(Pierraetal.,2005)istouseanontologyasafirstlevelofdatabaseconceptsspecification.Thisontologyisthenspecializedtodefineaconceptualmodel.Becauseallparticularsystemshavepartic-ularrequirements,differentconceptualmodelsmaybebuiltonthesameconsensualontology.Thelinkbetweenontologies,conceptualandlogicalmodelsiskeptinsideadatabase.Thisarchitectureenablestheevolutionofboththeconceptualmodelandoftheon-tologyandprovidesacommonaccesstoinformationthroughtheontology.Theadvantageofthisapproachistomakeclearwhatiscommonbetweentwosys-tems,andwhatisdifferent.

3.2Dataexchange

Aconsensualdomainconceptualizationthatcanbereferencedmayeasilybeusedasaninterchangefor-matfordataoverthisdomain(ISO13584-42,1998;Chawatheetal.,1994).Unlikeusualexchangeformatthatspecifythecompletestructureoftheexchangeddataandwherethemeaningofeachpieceofdatare-sultsfromitsplaceintheglobalstructure,ontology-basedexchangeareveryflexible.Insuchanexchange,themeaningofeachpieceofdatamaybedefinedlo-cally,byreferencingontologyidentifiers.Thisallow

quitedifferentexchangestructurestobesoundlyin-terpretedbythesamereceivingsystem.

3.3Dataintegration

Domainontologiesistheonlyartefactthatallowstoreconcile,atthesemanticslevel,heterogenousdatasourcemodels.Whendomainontoogiesareexplicitlyrepresentedindatabases,theintegrationmaybefullyautomatedevenwheneachsourcespecializeslocallythesharedontology(Bellatrecheetal.,2004).IntheSemanticWebapproaches,thelinkbetweensourceandontologyisusuallysupportedbymetadata.Theintegrationisoftenautomaticbecausetheontologiesusedinthisprocesscaptureandidentifyconceptsinaformalanduniqueway.

Innaturallanguageprocessing,thelinkbetweensourcesandanontologyconsistsofthewordscon-tainedinthedocuments.Mostofthewordshavingcontext-dependentmeaning,theintegrationprocessisoftenuser-assistedtoprovidemeaningfulresults.

3.4Dataaccessandsearch

Anontologyprovidesanaccesstodatathatreferencetheconceptsitdefines.Dependingontheexpressivepoweroftheontologymodel,datamaybebrowsedusingtheis-aconcepthierarchy,queriedusingkey-wordorwithmoresophisticatedquerylanguages.Ontologiesarealsousedtoquerydatabases.Theapproachconsistsinenrichingthequeriesonthelog-icalmodelbyexpressionsinvolvingontology-definedconceptsandexpressions(Dasetal.,2004).

Tosumup,ontologyapplicationsarewidespread.Foracompletesurveydescribingvarioususagesofontology,theinterestedreadercanreferto(UscholdandJasper,1999).Howeverallontologiesarenotsimilar.Nextsectionproposesataxonomyofon-tologiestohighlighttheirdifferencesandtheconse-quencesontheirusageforthevariousapplicationdo-mainsseenpreviously.

4AtaxonomyofdomainOntologies

Severalorthogonalcriteriahavebeenusedtoclas-sifyontologies.Thefirstmajorcriterionistheman-nerofconceptualizingadomain.Indeed,adomaincanbeconceptualizedasasetofwordsorasasetofconcepts.Thisconceptualizationwayleadstothedistinctionbetweenlinguistic(taxonomic)ontologiesandconceptual(descriptive)ontologies(Cullotetal.,2003;Pierra,2003).Following(Pierra,2003),wecallLinguisticOntologies(LO)thoseontologieswhosescopeistherepresentationofthemeaningofthewords

usedinaparticularUniverseofDiscourse,inapartic-ularlanguage.Ontheotherhand,ConceptualOntolo-gies(CO)arethosewhosegoalistherepresentationofthecategoriesofobjectsandofthepropertiesofobjectsavailableinsomepartoftheworld.

Asthesetwokindsofontologyaddressquitedif-ferentproblemsandfields,itisfundamentaltoclar-ifywhichkindofontologyissuitedineachparticularbusinesscontext.Beforepresentingourtaxonomy,letusreviewsomefundamentalsofontologies.

4.1Fundamentalsofontologies

Conceptsdefinedinaconceptualontologycanbeclas-sifiedintwocategories.

Primitiveconceptsarethoseconcepts”forwhichwearenotabletogiveacompleteaxiomaticdef-inition”(Gruber,1993).Here,thedefinitionre-liesonatextualdocumentationandaknowledgebackgroundsharedbetweenthereaders.Thesetofprimitiveconceptsdefinetheborderofthedomainconceptualizedbyanontology.Primitiveconceptsarethegroundonwhichallotherontologycon-ceptswillbebuilt.Thedefinitionofprimitivecon-ceptsbeingalways,atleastpartially,informal,theonlyqualitycriteria,onehasforsuchdefinitions,isthattheyrepresentaconsensusoversomecom-munity.Withoutsuchaconsensusonecannotassettheusabilityofanontology.

Besidesprimitiveconcepts,anumberofontologymodelsfocusonthecapabilitytocreateconserva-tivedefinitions(Gruber,1993),i.e,toassociateanewtermoranewconcepttosomethingthatisalreadydefinedbyanothermeanintheontologyunderdesign.Thischaracteristicisthebasisofin-ferencemechanismslikeautomaticclassification.Definedconceptsarethoseconceptsforwhichtheontologyprovidesacompleteaxiomaticdefinitionbymeansofnecessaryandsufficientconditionsex-pressedintermsofotherconcepts(eitherprimitiveconceptsorotherdefinedconcepts).

Whendefinedconceptsareintroduced,conceptequiv-alencerelationneedstobedefinedinordertobeabletocompare,classifyorrelatedefinedand/orprimitiveconcepts.

Conceptequivalencecanbedefinedattheclasslevel.ThisistheapproachfollowedbymodelsbasedonDescriptionLogics(DL)likeOWL(Bechhoferetal.,2004)orontheCarinlanguageusedinthePICSELproject(Roussetetal.,2002).Forexample,inPIC-SELtheconceptofHotelisdefinedasaspecializa-tionoftheprimitiveconceptHousingPlace.AHous-ingPlaceisdefinedasaplacehavingassociatedbuild-ings,roomsandmealservices.AhotelisthenfullydefinedasthoseHousingPlaceswhichhavemorethanfiveroomsandhaveonlyCollectiveBuilding.

Conceptequivalencecanalsobeexpressedatthepropertylevel.Thisisthecaseinmodelswherederiva-tionfunctionscanbedefined.F-Logic(Kiferetal.,1995)isoneofthemodelssupportingthiscapabil-ity.Forexample,theproperty”boss”relatinganem-ployeetoanotheremployeecanbederivedfromtheproperties”belongto”and”chair”.

Thankstothedistinctionbetweenprimitivecon-ceptsanddefinedconcepts,wearenowabletopro-poseourdatabase-orientedtaxonomyofontologies.

4.2

CanonicalConceptualOntology(CCO)

Indatabasedesign,whenconceptualmodelisdefined,ambiguityorpossiblemulti-representationofthesamefactisforbidden.Thesameapproachisfollowedforperformingexchangesbetweentwodifferentdatabases.Itconsistsindefiningacanonicalvocabularyinwhicheachinformationinthetargetdomainiscapturedinanuniquewaywithoutdefininganysynonymouscon-structs.Forexample,intheSTEPproject,exchangemodelsaredefinedintheEXPRESSlanguageasaSTEPApplicationProtocol(AP).Thesecanonicex-changemodelsareusedbyindustrialuserstoexchangeproductdescriptionsbetweendifferentorganizations.Ontologieswhosedefinitionsfollowthisapproach

color arecalledCanonicalConceptualOntologies.InCCOs,A_WINE B_WINE String eachdomainconceptisdescribedinasingleway,us-(A) ingasingledescriptionthatmayincludenecessary

is_a B_REDWINE B_WHITEWINE condition.Asaconsequence,CCOsincludeprimi-property tiveconceptsonly.(B) DefiningCCOsisthemaingoaloftheontology

modeldefinedinthePLIB(PartLibrary)standard(ISO13584-Figure1:LocalWineCCO42,1998)withafirstfocusonrepresentingandex-changingformalontologiesoftechnicaldomains.Such

CCOsuseaproperty-basedcharacterizationofthein-ApplyingsomeNCCOoperators,issuedfromaDL

volvedconcepts.Thismeansthataclassisonlycre-syntax,wecanwritethat:

atedwhenitprovesnecessaryfordefiningthedomainBREDWINE≡AWINE󰀃color:redofsomeproperties.Therefore,inPLIB,thegener-ThisaxiomstatesthattheconceptofBREDWINEalization/specializationconcepthierarchiesareratherdefinedinCCO(B)isequivalenttotheconceptAWINE”flat”.AnexampleofsuchanontologyforelectronicdefinedinCCO(A)restrictedbythevalueofthecolorcomponentscanbefoundin(IEC61360-4,1999).attribute.TheseprimitiveshaveformalsemanticswhichTheseexamplesshowthatusageofCCOsintheenablesapowerfulreasoningmechanismimplemen-contextofdataexchangeisfruitful.Moreargumentstationintoolsnamedreasoners(e.g,RACER(HaarslevaregivenlatterinthispaperbystudyinganexchangeandM¨oller,2001)).Thesetoolssupportmechanismsscenario.forconceptsandinstancesclassification.Asaresult,

thetwolocalCCOscanbeautomaticallymergedintotheNCCOoffigure2.ThisNCCOprovidesaglobal4.3NonCanonicalConceptual

accesstodataofthetwoCCOs.

Ontology(NCCO)Inthisexample,thereasonerhasinferredthatallin-stancesofBREDWINEareinstancesofAWINEIndatabasedesign,conceptequivalenceplaysanim-havingthevalueredforthepropertycolor.When

portant(butsecond-order)role.Eachdatabasead-thesefactsarematerialized,asitisproposedinOn-dressesaparticulardomainanddefinesacanonicwaytoMerge(Douetal.,2003),itbecomepossibletosplitofrepresentinganyfactaboutthisdomain.Then,inthemergedontologyintotheCCO(A)andtheCCO

ordertoachieveadegreeofindependencewithre-specttothechoiceoftheconceptsofferedtousers,a

databasedesignerprovidesviews.Thesedefinedcon-ceptsarespecifiedusingtheCREATEVIEWoperatoronthebasisoftheprimitiveconceptsthatconstitutethedatabaseschema.Indeductivedatabasethisfunc-tionalityalsoexists,providedwithderivedentities.So,whatevertheconstructofferedtoexpresscon-ceptequivalence,wecallNonCanonicalConceptualOntologiesthoseontologieswhichcontainnotonlyprimitivesconceptsbutalsodefinedconcepts.

NCCOsareparticularyusefulwhentheyareusedasglobalqueryschemas.Forexample,inthePIC-SELprojectmentionedearlier,primitiveconceptsoflocalCCOsareexpressedasdefinedconceptsfortheprimitiveconceptsofaglobalontologyusedasglobalqueryschema.Intheglobalontology,conceptex-pressionsandrulesexpressedonbasicconceptsofthetourismdomain(HousingPlace,Flight...)defineawiderangeofterms(Hotel,Bed&Breakfast...)use-fulforuserstoformulateaqueryondatareferencedbylocalCCOs.

NCCOconstructorsarealsoveryusefultodefinemappingsbetweendifferentontologies.Forexample,figure1presentstwoCCOsofthedomainofwines.TheCCO(A)ispropertyorientedwhiletheCCO(B)isentityoriented.

String color A_WINE B_WINE is_a color B_REDWINE B_WHITEWINE color property = red = white restriction Figure2:IntegratedWineNCCO

(B)withthenewinferredinstances.ThisprocessshowsthatNCCOconstructorsareusefulforintegra-tiontasks.

4.4LinguisticOntologies

Inotherapplicationdomainslikeinformationretrievalornaturallanguageprocessing,humanlanguagesplayakeyrole.Evenindatabase,naturallanguagesareusedinvariousplaces.Indeed,tableandattributenamesarechosentoreflecttheirmeaning.Moreover,conceptualmodeldocumentationislargely,andinanumberofcasescompletely,expressedinnaturallan-guage.

WecallLinguisticOntology(LO)thoseontologiesthatdefinewordsorcontextualusagesofwords.Inthiskindofontologyonlywordsrelationships(syn-onym,hyponym,...)areavailable.Wordsrelation-shipsbeinghighlycontextual,machineinferenceingeneralneedsexpertsupervision.Moreover,approxi-materelationshipsmaybedefined.

Wordnet(http://www.cogsci.princeton.edu/wn)isthemostwell-knownrepresentativeofthiscategoryofontologies.Itprovideswithtextualdefinitions,syn-onymoustermsandrepresentationsforthevariouscon-ceptsthatcanbedenotedbyaterm.Theyareintendedtobeusedasasophisticatedthesauri.

LOshelptorecognizeconceptualsimilaritiesbe-tweensentencesevenifdifferenttermsareused.Sincewordmeaningsarecontextualandtheirrelationshipsareapproximative,wrongsimilaritiesmaybeproducedandresultscanneverbeconsideredasreliable.Anex-ampleofLOsusageisgivenin(Dasetal.,2004).ALOontypesofcookingisusedtoretrievemoremean-ingfulresultsonthefoodservedinarestaurant.Thus,queryingthe’latinamerican’servedfoodretrievestu-pleshaving’american’or’mexican’asvalueoftypeofcooking.Anumberofsemi-automaticdatabasein-tegrationapproaches,e.g.,(Visseretal.,1999),usesuchlinguisticontologies.

4.5

Disciplinespecificviewofdomainontology

Currently,thethreecategoriesofontologymainlycor-respondtothreedifferentdisciplinesofcomputersci-

enceandhavefewconnexions.Ontologymodelsfo-cuseitheronCCO,NCCOorLOdesign.

CCOaremainlyconsideredinthedataprocessingcommunity.InCCOs,ontologydescriptionsfocusonprimitivesconceptscharacterizationandidentifi-cation.

Theyincludepreciseandcomplexdescriptionsoftheprimitiveconcepts.Thesedescriptionsarepro-videdusingCCOorientedmodelconstructs.Forexample,theDOGMAformalism(JarrarandMeers-man,2002),adatabaseinspiredontologymodel,providescontextualidentificationofconcepts.InthePLIBmodel,primitiveconceptscanbeassoci-atedwithreferencetorealdocuments,pictures,us-agerestrictions....Thismodelalsodistinguishestherigidproperties(GuarinoandWelty,2002),i.e.,propertiesessentialforanyinstanceofaclassfromthosethatmayornotholdorexistaccordingtotheroleinwhichanentityisinvolved.

Theydon’tcontainmodelmappings.Consequently,theencodingofsuchconversionsisachievedinanapplication.

NCCOmodelsweredeveloppedbyartificialintelli-gencecommunity.Therefore,theyfocusoninferenceandconceptequivalence.

NCCOscontainsconservativedefinitionofdefinedconceptsusingusefuloperatorslikebooleanoper-ators(unionofclasses...).

Asarule,theyincludelessprecisedescriptionsfortheprimitiveconcepts.Forexample,inOWL,aprimitiveconceptdescriptionislimitedtoalabel,acommentandtheproperties(roles)thatcanapplytoitsinstances.

LOsweredesignedforcomputationallinguistic.InLOs(e.gWordnet),eachwordisassociatedwithsev-eralsynsets(setsofsynonymous)thatreflectsitsvar-iousmeanings.Theimprecisionoftheconceptualiza-tionisduetothefollowingfacts:wordsmeaningdependsuponacontext;

wordsrelationships(e.gsynonymy)havenoformaldefinitionwhereasconceptsrelationships(e.gsub-sumption)have.

5

Relationshipbetweenontologycategoriesandproposalforalayeredmodel

ThepreviousobservationsleadustoidentifysomerelationshipsbetweenCCOs,NCCOsandLOs.MappingsbetweenCCOmightalsobedefinedofequivalenceoperatorsofsomeNCCO;

NCCOmodelscanusepowerfulCCO-orientedmodelconstructstodefinetheirownprimitiveconcepts;LOsmightdefinethevariousmeaningofeachwordofaparticularlanguagebyreferencetoaNCCO.Thisreferencewouldprovideabasisforformalandexactreasoningandautomatictranslationofcontext-specificterms.

Asafurthersteptowardsthisobservation,wefirstproposealayeredmodelfordomainontologydesign.Then,wepresentanexampleofusageofthislayeredmodel.

5.1

Alayeredmodelforontologydesign

Anoftenusedguide(NoyandMcGuinness,2001)proposesasevenstepsapproachforNCCOdevelop-ment.

1.Determinethedomainandscopeoftheontologytobedeveloped.2.Considerreusingexistingavailableontologiesthatsomeoneelsehasdeveloped.3.Listtheimportanttermsintheontologywithoutconsideringthepossibleconceptsoverlapstheymayleadto.4.Definetheclassesandtheassociatedclasshierar-chy.Fromthelistcreatedinstep3,selectthetermsthatdescribeobjectshavingindependentexistence.Thesetermswillbeclassesintheontologyandwillbecomeanchorsintheclasshierarchy.5.Definethepropertiesassociatedtoclasses.Indeed,mostoftheremainingtermsarelikelytobeprop-ertiesoftheseclasses.6.Definetheconstraints(cardinality,domainandrangerestrictions)thatholdonproperties.7.Createinstancesofclassesinthehierarchy.ThisapproachexploitstheNCCOcapabilitytode-fineequivalentconceptsandthustointegrateseveralontologiesaddressingthesamedomain.SinceweclaimthatNCCOscanbenefitfrombeingarticulatedwithCCOs,weproposeanalternativeapproachforthedevelopmentofaNCCOstartingfromaCCO.1.ThefirststepofthedesignofanontologyshouldbetoagreewithinacommunityonaCCO.Toreachthisagreement,itisrequiredto:

defineclearlywhatisthedomaincoveredbytheontology;

chooseapowerfulmodeltodefinepreciselytheprimitiveconceptsexistinginthedomain;

provideacommonunderstandingofacanonicsetofconceptscoveringthedomain.Thiscon-ceptualizationmustaccommodateawideanddi-verserangeoftechnicalandbusinessrequire-mentssharedbythemembersofthecommunity.Itmustreachawiderecognitionandacceptance.2.Withinthecommunityofusersand/ordevelopers,onthebasisofthedefinedCCO,aNCCOmaybebuiltforusebymembersofthiscommunityeithertobuildtheirownviewofthedomainortomodelformallyalltheconceptsexistinginthetargetdo-mainthatareassociatedwithausuallinguisticde-notation(wordorsequenceofwords).Proceedingthiswayensuresthepreservationofthecapabil-itytoshareandexchangeinformationexpressedintermsoftheCCO.3.InordertoallowtheuseofthedefinedNCCOforlinguisticinferenceand/ortoprovideanend-userfriendlyuserinterfaceinvariouslanguages,alistoflanguage-specifictermsneedstobedefinedandassociatedtoeachconceptintheNCCO.Accordingtothisalternativeapproach,figure3il-lustratesalayeredmodel,calledtheonionmodel,oftheresultingdomainontology.ACCOprovidesafor-malbasistomodelandtoexchangeefficientlytheknowledgeofadomain.ANCCOprovidesmecha-nismstorelatedifferentconceptualizationsmadeonthisdomain.Finally,LOprovidesnaturallanguagerepresentationoftheconceptsofthisdomain,possi-blyinthevariouslanguageswheretheseconceptsaremeaningful.

LO Class Expression : Description Logic NCCO Property Other Expression … CCO F-Logic Property Expression : Derivation function Figure3:Theonionmodelofdomainontology

5.2

Anexchangescenariobasedonlayereddomainontology

Indatabaseuniverse,eachdatabaseusesacanonicalvocabulary.Usually,eachdatabaseusesadifferentcanonicalvocabulary.InsteadofdefiningaNCCO

coveringallthetermsofallthesources(seefig4(A)),theonionmodelsuggeststhatallexchangesareperformedusingaconsensualCCO.Eachsourcejustcontainsthedescriptionsofitsownconceptsintermsofthe(primitive)conceptsofthecanonicalconcep-tualontology(seefig4(B)).Thisapproachhasbeenputintopracticefordatabasesintegrationin(Bella-trecheetal.,2004).Noticethatifeachconceptisrepresenteddifferentlyinthenparticipatingsources,andifthereexistsineachsourceameanvalueofpconcepts,solution(A)needstoimplementn∗pmappingsineachsource,whensolution(B)requiresonlypmappingsineachsource.Moreover,thisap-proachalsoappliestovirtualexchange,i.e,mediator(Wiederhold,1991).

CCO CCO CCO CCO S1 S2 S1 S2 CCO NCCO CCO CCO S3 (A) CCO CCO CCO S4 S3 (B) S4 NCCO Operators Figure4:UseofontologiesforcanonicdataexchangeThisclause,andtheaboveshowtheinterestofar-ticulatingthethreecategoriesofontologiesaccordingtoouronionmodelacrossthewholelifecycleofdo-mainontologies.

CCOsprovidecanonicalandaccuratedescriptionsofeachconceptofagivendomain.Itprovidessoundbasisforexchangebetweendifferentsources.NCCOoperatorsareusedtointeractwithotherap-plicationsorsourcesthathavealreadytheirownparticularontologies.

LOsprovidelinguisticcapabilitiesoverprimitiveanddefinedconcepts.

Nextsectiondiscussesthevarioustoolsthatwouldbeusefultofacilitatetheuseofontologiesindatamanagementactivityandcomparewiththosecurrentlyexisting.

6

OntologyBasedDataManagementSystem(OBDMS)

AnOBDMSisasuiteoftoolsprovidingsupportforusingontologiesandontologyinstancesdata.Indatamanagement,manydifferentfunctionalitiesmaybeexpectedfromanOBDMS.Inthissectionwelista

setofsuchfunctionalitiesandrelatethemtoourtax-onomy.

6.1OBDMSfunctionalities

AnOBDMSmayprovidethefollowingfunctionali-ties.

F1.Respectofstandard.Ifaformalsemanticsisde-finedforsomesupportedontologymodelstandard,thissemanticsshallberespectedbythedefinedOB-DMS.

F2.Handlingofexchangeformat.Whenanexchange

formatisprovidedwithanontologymodel,im-port/exportontologiesandontologiesinstancedatainthisformatareneeded.

F3.Datamanipulation.AnOBDMSshouldprovide

supporttoinsert,updateordeleteontologycon-ceptsandinstancedata(supportofCCO).

F4.Linguisticsupport.AnOBDMSshouldsupport

andexploitlinguistic-orientednaminganddescrip-tionofconceptsinvariouslanguages(supportofLO).

F5.Dataquerying.Anefficientwayshouldbeavail-ableforqueryingbothontologiesandontologiesdata.

F6.OntologyMapping.AnOBDMSshouldprovide

mappingfunctionalitiestointegratedifferenton-tologies(supportofNCCOoperators).

F7.Ontologyasaspecification.AnOBDMSshould

providethepossibilitytoextractfromanontologyaspecificationofasoftwareorofadatabaseschema.

6.2

ComparisonofsomeOBDMSimplementations

Thedescriptionofvarioususefulcapabilitiesneededtofacilitatetheuseofontologiesindatamanagementbeingcompleted,thissectionreviewsthreeOBDMSaccordingtothefunctionalitiestheyoffer.6.2.1

RDFTools

RDFSuiteandSesamearetwosuitesoftoolsforRDFSstorageandquerying.Adatabaseisusedforthepersistenceofthedata.Thesetwotoolsaddcon-straintsontheRDFSstandard.Theyproposeimport/-exportmoduletomanagedatadescribedinanRDFsyntax.Third-partytoolscanbeassociatedtothissuitebyusingthismodule.Thus,anontologyeditor(e.g.Protege)canbeassociatedtothesetools.How-ever,thiseditorwillnotbesynchronizedwiththestor-agesystem.QueryingdatainSesameandRDFSuiteisperformedthroughtheRQLlanguage.Thereisnosupportprovidedbythesetoolstointegratedifferentontologiesortousethemasaspecification.

6.2.2PLIBSuite

PLIBSuiteisasetoftoolsforPLIBontologiesstor-age,edition,queryingandintegration.TheexchangeformatofPLIBontologiesisEXPRESSphysicalfile.PLIBSuiteincludefullsupportofthisformat.Cur-rently,ontologiesstoragerelyontheOntoDBpro-totype.ThisprototypeisanimplementationofthenotionofOntologyBasedDatabasepresentedinsec-tion3.Thus,itstoresontologies,dataandlogicalmodelsofdatadefinedbyextractionand/orspecial-izationfromontologies.ItrespectsfulldefinitionofthePLIBstandardanditsarchitectureisflexibleenoughtomanagetheevolutionofthisstandard.Thispro-totypeisdirectlylinkedtothePLIBEditorsoftwarethatcanbeusedtovisualizeandmanageontologiesanddata.Eachconceptisassociatedtoanameindifferentnaturallanguageswithsynonymousnames.PLIBEditorexploitsthisnaturallanguagecapabilitytoprovideamultilingualinterface.However,thereisnolinguisticinferenceusingsynonymousnames.PLIBEditorprovidesaquerymodulethatenablestobuildvisuallyaqueryondatafromtheontologies.ThismodulereliesontheOntoQLquerylanguageal-lowingtomanageandtoquerybothontologiesanddata.IntegrationofPLIBontologiesisenabledwhensubsumptionlinksaredefinedbetweendifferenton-tologies.6.2.3

RacerPro

RacerProisanin-memoryOWLreasoningsystem.RacerProsupportsthefullOWLstandardandincludesagraphicaluserinterfacetomanipulateontologiesanddata.ThesedataaredirectlypersistedinanOWLfile.QueryingthesedataispossiblethroughthequerylanguagenRQL.However,memorydemands,concur-rencycontrolandscalabilityofthisimplementationarestillopenissues.Asapayoff,RacerProoffersafullreasoningsystemthatenablesautomaticclassifi-cationofconceptsandofdata.Aswehaveseeninsection4.3,thisfunctionalityisveryusefulforinte-gration.

Table1summarizesthepreviousstudy.

RDFToolsRacerProPLIBSuiteF1partialyesyesF2yesyesyesF3partialyesCCOF4nonopartialF5yespartialyesF6noNCCOpartialF7nonoyesTable1:FulfilledfunctionalitiesbyexistingOBDMS

Itappearsthatnoneoftheexistingtoolscoversthecompletesetoffunctionalitiesrequiredtoimplementthescenarioproposedinsection5.2.Therefore,nextsectionproposesanarchitectureusingexistingOB-DMStosolvethisproblem.

6.3

Anintegratedarchitecturetoimplementourdataexchangescenario

OurexchangescenariorequiresoneconsensualCCOmanagedinanOBDMSprovidingthepossibilitytomapNCCOs(F6).SinceRaceristheonlyexistingOBDMSprovidingthisfunctionalityweproposetouseitforthispurpose.

EachsourcedefinesitsownCCOandmanagesitsinstances.WeproposetousePLIBSuiteforeachpar-ticularsource.TheuseofPLIBSuiteforthispur-poseensurethataprecisedescriptionoftheconceptswillbeavailable(F3).Moreoveritprovidesascal-ablerepositorytostoreallthedata(F5).Forlinguis-ticsupport(F4),itprovidesthenecessaryresourcestoreferenceaLOsuchasWordnet.

Theexchangeofdatarequiresthefollowingsteps:1.AsourceexportsitsprimitiveconceptstoRacer(F2).TheybecometheconsensualCCO.NoticethatthissteprequiresthatPLIBSuitecanexportdatainOWLformat.

2.OthersourcesmayuseRacereditortodefinemap-pingbetweentheirconceptsandthedefinedcon-sensualCCO.

3.UsingtheRacerreasoningsystem,aclassificationofalltheconceptsisdone.Thus,theconsensualCCObecomeaNCCO.

4.Datainstancestobeexchangedbetweenthede-finedsourcesareimportedintoRacer.

5.UsingtheRacerreasoningsystem,aclassificationoftheseinstancesisperformed.

6.Forconceptsofagivensourceunderwhichinstanceshavebeenclassified,anexportoftheseinstancestothissourceisachieved.Thissteprequirestokeeptrackoftheoriginofeachconcept.

Otherpropositionslike(PanandHeflin,2003)aremadetoassociateareasonerwithadatabase.How-evertheaimsofthesepropositionsistoprovideafullreasoningsystemusingadatabase.Yet,OWLreason-ingisnotfullysupportedbythesesystems.

7Conclusion

Thispaperhasinvestigatedtheconceptofdomainontologyinadatabaseperspective.Ontologybecom-ingabuzzword,oftenusedasanewtermforalready

existingmodels,wehavefirstproposedthreecriteriatocharacterizeontology.Adomainontologymustbeformal,i.e,allowingsomeautomaticreasoningandconsistencycheckingcapability,consensualinsomecommunityandabletobereferencedfromanyen-vironment.Thesethreecriteriacharacterizedomainontologyasanewkindofmodelincomputerscienceandleadustoproposeanewdefinitionofdomainon-tologyasaformalandconsensualdictionaryofcate-goriesandpropertiesofentitiesofadomainandtherelationshipsthatholdamongthem.

Domainontologymodelsbeingmainlydevelopedbythreecommunities,wehaveproposedataxonomyofdomainontologyintoCCO,NCCOandLO.Afterreviewingthepartialdomaincoverageofthesevari-ousmodels,wehaveproposedalayeredmodel,calledtheonionmodelofdomainontology,allowingtode-signandtousethecapabilitiesofeachcategoryofontologyinanintegratedenvironment.WehavealsodiscussedunderthenameofOBDMSwhatkindsofservicesshouldbeprovidedtoallowapowerfulusageofontologyindatamanagement.

CurrentlythereexistneitherexchangeformatnorOBDMSabletorepresentandtomanagedomainon-tologiescorrespondingtothecompleteonionmodel.First,wearedevelopinganXMLschemathatwillin-tegratebothOWLandPLIBcapabilities.Second,weareextendingthePLIBSuitetosupportOWLclassexpressionconstructsusingbothaconnexionwithanOWLreasonerandarepresentationbySQLviews.Finally,weareworkingonaquerylanguage,calledOntoQL,allowingtoquerythethreelayersoftheonionmodel.

REFERENCES

Bechhofer,S.,vanHarmelen,F.,Hendler,J.,Horrocks,I.,

McGuinness,D.L.,Patel-Schneider,P.F.,andStein,L.A.(2004).OWLWebOntologyLanguageRefer-ence.WorldWideWebConsortium.Bellatreche,L.,Pierra,G.,Xuan,D.N.,Hondjack,D.,and

Ait-Ameur,Y.(2004).Anaprioriapproachforau-tomaticintegrationofheterogeneousandautonomousdatabases.InDEXA,pages475–485.Bozsak,E.,Ehrig,M.,Handschuh,S.,Hotho,A.,Maed-che,A.,Motik,B.,Oberle,D.,Schmitz,C.,Staab,S.,Stojanovic,L.,Stojanovic,N.,Studer,R.,Stumme,G.,Sure,Y.,Tane,J.,Volz,R.,andZacharias,V.(2002).Kaon-towardsalargescalesemanticweb.InEC-WEB’02:ProceedingsoftheThirdInterna-tionalConferenceonE-CommerceandWebTechnolo-gies,pages304–313,London,UK.Springer-Verlag.Brickley,D.andGuha,R.(2004).RDFVocabularyDe-scriptionLanguage1.0:RDFSchema.WorldWideWebConsortium.

Chawathe,S.S.,Garcia-Molina,H.,Hammer,J.,Ireland,

K.,Papakonstantinou,Y.,Ullman,J.D.,andWidom,J.(1994).Thetsimmisproject:Integrationofhetero-geneousinformationsources.InIPSJ,pages7–18.Cullot,N.,Parent,C.,Spaccapietra,S.,andVangenot,C.

(2003).Ontologies:Acontributiontothedl/dbde-bate.InSWDB,pages109–129.Das,S.,Chong,E.I.,Eadon,G.,andSrinivasan,J.

(2004).Supportingontology-basedsemanticmatch-inginrdbms.InVLDB,pages1054–1065.Dou,D.,McDermott,D.,andQi,P.(2003).Ontol-ogytranslationonthesemanticweb.InProceed-ingofthe2ndInternationalConferenceonOn-tologies,DatabasesandApplicationsofSemantics,(ODBASE’2003),pages952–969.Gangemi,A.,Guarino,N.,Masolo,C.,andOltramari,A.

(2003).Sweeteningwordnetwithdolce.AIMagazine,24(3):13–24.Gruber,T.R.(1993).Atranslationapproachtoportableon-tologyspecifications.Knowl.Acquis.,5(2):199–220.Guarino,N.andWelty,C.(2002).Evaluatingontological

decisionswithontoclean.Commun.ACM,45(2):61–65.Haarslev,V.andM¨oller,R.(2001).Racersystemdescrip-tion.InIJCAR’01:ProceedingsoftheFirstInter-nationalJointConferenceonAutomatedReasoning,

pages701–706.Springer-Verlag.IEC61360-4(1999).Standarddataelementtypeswithas-sociatedclassificationschemeforelectriccomponents-part4:Iecreferencecollectionofstandarddatael-ementtypes,componentclassesandterms.Technicalreport,InternationalStandardsOrganization.ISO10303(1994).Initialreleaseofinternationalstan-dard(is)10303.Technicalreportis10303,Interna-tionalStandardsOrganization.ISO13584-42(1998).Industrialautomationsystemsand

integrationpartslibrarypart42:Descriptionmethod-ology:Methodologyforstructuringpartsfamilies.Technicalreport,InternationalStandardsOrganiza-tion.Jarrar,M.andMeersman,R.(2002).Formalontologyen-gineeringinthedogmaapproach.InOntheMovetoMeaningfulInternetSystems,2002-DOA/Coop-IS/ODBASE2002ConfederatedInternationalConfer-encesDOA,CoopISandODBASE2002,pages1238–1254.Springer-Verlag.Kifer,M.,Lausen,G.,andWu,J.(1995).Logicalfounda-tionsofobject-orientedandframe-basedlanguages.J.ACM,42(4):741–843.Meersman,R.(2001).Ontologiesanddatabases:Morethan

afleetingresemblance.InOES/SEOWorkshopRome.(2001).Niles,I.andPease,A.(2001).Towardsastandardupperon-tology.InProceedingsofthe2ndInternationalCon-ferenceonFormalOntologyinInformationSystems(FOIS-2001),pages2–9.

Noy,N.F.andMcGuinness,D.L.(2001).Ontologyde-velopment101:Aguidetocreatingyourfirstontol-ogy.Technicalreportksl-01-05andstanfordmedicalinformaticstechnicalreportsmi-2001-0880,StanfordKnowledgeSystemsLaboratory.Pan,Z.andHeflin,J.(2003).Dldb:Extendingrelational

databasestosupportsemanticwebqueries.InPSSS.Parent,C.,Spaccapietra,S.,andZimanyi,E.(1999).

Spatio-temporalconceptualmodels:datastructures+space+time.InGIS’99:Proceedingsofthe7thACMinternationalsymposiumonAdvancesingeographicinformationsystems,pages26–33.ACMPress.Pierra,G.(2003).Context-explicationinconceptualon-tologies:Theplibapproach.InJardim-Gonalves,R.,Cha,J.,andSteiger-Garao,A.,editors,Proceedingsofthe10thISPEInternationalConferenceonConcur-rentEngineering(CE2003),pages243–254.Pierra,G.,Dehainsala,H.,Ait-Ameur,Y.,andBellatreche,

`baseontologique:L.(2005).Basededonn´eesa

principesetmiseenœuvre.InIng´enieriedesSyst`emesd’Information(toappear).Rousset,M.-C.,Bidault,A.,Froidevaux,C.,Gagliardi,

H.,Goasdou,F.,Reynaud,C.,andSafar,B.(2002).Constructiondem´ediateurspourint´egrerdessourcesd’informationmultipleseth´et´erog`enes:Picsel.revueI3,2(1):9–59.Tetlow,P.,Pan,J.,Oberle,D.,Wallace,E.,Uschold,M.,and

Kendall,E.(2005).OntologyDrivenArchitecturesandPotentialUsesoftheSemanticWebinSystemsandSoftwareEngineering.WorldWideWebConsor-tium.Uschold,M.andJasper,R.(1999).Aframeworkfor

understandingandclassifyingontologyapplications.InProceedingsoftheIJCAI99WorkshoponOn-tologiesandProblem-SolvingMethods(KRR5),Stock-holm,Sweden,(August1999).Visser,P.R.S.,Beer,M.D.,Bench-Capon,T.J.M.,Diaz,

B.M.,andShave,M.J.R.(1999).Resolvingonto-logicalheterogeneityinthekraftproject.InDatabaseandExpertSystemsApplications,10thInternationalConference,DEXA’99,pages668–677.Wiederhold,G.(1991).Obtaininginformationfromhet-erogeneoussystems.InProceedingsoftheFirstWorkshoponInformationTechnologiesandSystems(WITS’91),pages1–8.CambridgeMA,MITSloanSchoolofManagement.