• Keine Ergebnisse gefunden

The automation of tasks depends on elevating thestatus of the web from machine-readable to somethingwe might call machine-understandable

N/A
N/A
Protected

Academic year: 2022

Aktie "The automation of tasks depends on elevating thestatus of the web from machine-readable to somethingwe might call machine-understandable"

Copied!
13
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

Semantic Web

SeanBechhofer, IanHorrocks,CaroleGoble, andRobertStevens

InformationManagementGroup,DepartmentofComputerScience,

UniversityofManchester,

OxfordRoad,ManchesterM139PL,UK

{seanb,horrocks,carole,stev ensr} @cs. man. ac.uk ,

http://img.cs.man.ac.uk

Abstract. Ontologies will play a pivotalr^oleinthe \SemanticWeb",

where they will provide a source of precisely dened terms that can

becommunicatedacrosspeopleand applications.OilEd,isanontology

editorthathasaneasytouseframeinterface,yetatthesametimeallows

users to exploit the full powerof anexpressive webontology language

(OIL).OilEdusesreasoningtosupportontologydesign,facilitating the

developmentofontologiesthatarebothmoredetailedandmoreaccurate.

1 Introduction

Ontologieshavebecomeanincreasinglyimportantresearchtopic.Thisisaresult

bothoftheirusefulnessinarangeofapplicationdomains[1{3],andofthepivotal

r^olethattheyaresetto playinthedevelopmentoftheSemanticWeb

TheSemanticWebvision,asarticulatedbyTimBerners-Lee[4],isofaWeb

in which resources are accessible not only to humans, but also to automated

processes, e.g., automated \agents" roaming the web performing useful tasks

suchasimprovedsearch(intermsofprecision)andresourcediscovery,informa-

tion brokering and information ltering. The automation of tasks depends on

elevating thestatus of the web from machine-readable to somethingwe might

call machine-understandable. The key ideais to havedata on theweb dened

andlinkedinsuchawaythatitsmeaningisexplicitlyinterpretablebysoftware

processesratherthanjust beingimplicitlyinterpretablebyhumans.

To realise this vision, it will be necessaryto annotate web resources with

metadata(i.e.,datadescribingtheircontent/functionality).Standardisationpro-

posalsforannotationlanguageshavealreadybeensubmittedtotheWorldWide

WebConsortium(W3C),inparticularRDF(ResourceDescriptionFramework)

andRDFSchema(see[5]foradiscussionofther^olesoftheselanguagesandof

XML/XMLSchema).However,suchannotationswillbeoflimitedvaluetoauto-

matedprocessesunlesstheyshareacommonunderstandingastotheirmeaning.

Ontologies,can helptomeetthisrequirementbyprovidinga\representationof

asharedconceptualisation of aparticular domain"that canbecommunicated

(2)

representationlanguage:ittalksaboutclassesandproperties(binaryrelations),

range and domain constraints (on properties), and subclass and subproperty

(subsumption)relations.However,RDFSisarelativelyprimitivelanguage(the

aboveisanalmostcompletedescriptionofitsfunctionality),andmoreexpressive

power would clearly be necessary/desirable in order to describe resources in

suÆcientdetail.Moreover,such descriptionsshould beamenableto automated

reasoning iftheyaretobeused eectivelybyautomatedprocesses.

These considerations have led to the development of OIL [7], an ontology

languagethat extendsRDFS withamuchrichersetof modellingprimitives.A

similar RDFS based web ontologylanguagecalled DAML wasbeendeveloped

aspartoftheDARPADAMLproject[8]andthetwolanguageshavenowbeen

mergedunder the nameDAML+OIL 1

. OIL hasaframe-like syntax, which fa-

cilitates tool building, yet canbemapped ontoan expressivedescription logic

(DL),whichfacilitatestheprovisionofreasoningservices.OilEdisanontology

editingtoolforOIL(andDAML+OIL)thatexploitsboththesefeaturesinorder

toprovideafamiliar andintuitivestyleofuserinterfacewiththeaddedbenet

ofreasoningsupport.Its main noveltyliesin the extensionof theframeeditor

paradigmto dealwithaveryexpressivelanguage,and theuseofahighlyopti-

mised DLreasoningengineto providesound andcomplete yet still empirically

tractablereasoningservices.

Reasoning with terms from deployed ontologies will be important for the

SemanticWeb, but reasoning support is also extremely valuableat the ontol-

ogy design phase, where it canbe used to detect logically inconsistent classes

and to discoverimplicit subclass relations.This encouragesa moredescriptive

approach to ontologydesign, with thereasonerbeingused to inferpartof the

subsumption lattice (see the case study presented in Section 4); the resulting

ontologiescontainfewererrors,yet providemoredetaileddescriptionsthatcan

beexploitedbyautomatedprocessesintheSemanticWeb.Finally,reasoningis

of particular benet when ontologies are large and/or multiply authored, and

alsofacilitatesontologysharing,mergingandintegration[9];considerationsthat

willbeparticularly importantin thedistributed webenvironment.

2 Oil and DAML+OIL

The development of OIL resulted from eortsto combine the best features of

frameandDLbasedknowledgerepresentationsystems,whileatthesametime

maximising compatibility with emerging web standards. The intention was to

designalanguagethatwasintuitivetohumanusers,andyetprovidedadequate

expressivepowerforrealisticapplications(manyearlyDLsfailedonthissecond

count|see[10]).

The resulting language combines a familiar frame like syntax (derived in

partfrom theOKBC-liteknowledgemodel[11]), withthepowerand exibility

1

(3)

andinverseslots,generalaxioms,etc.).Thelanguageisdenedasanextension

ofRDFS, therebymaking OIL ontologies (partially)accessibleto any \RDFS-

aware"application.

Theframesyntaxislessdauntingtoontologists/domainexpertsthanaDL

stylesyntax,anditfacilitatesamodellingstyleinwhichontologiesstartoutsim-

ple(in terms oftheir descriptivecontent)and are graduallyextended,bothas

thedesignitselfisrenedandasusersbecomemorefamiliarwiththelanguage's

advancedfeatures (seeSection4). Theframeparadigmalsofacilitatesthecon-

structionandadaption oftools,e.g.,theOntoEditand Protegeeditorsandthe

Chimaera integration tool are all beingadapted to use OIL/DAML+OIL [12,

13,9].

Ontheotherhand,basingthelanguageonanunderlyingmappingtoavery

expressiveDL(SH Q)providesawelldened semanticsandaclearunderstand-

ingofitsformalproperties,inparticularthattheclasssubsumption/satisability

problemisdecidableandhasworstcaseExpTimecomplexity[14].Themapping

also provides a mechanism for the provision of practicalreasoning services by

exploitingimplementedDLsystems,e.g.,theFaCTsystem[15].

OILextendsstandardframelanguagesinanumberofdirections.Oneofthe

keyideasisthatananonymousclassdescription,orevenbooleancombinations

of class descriptions, can occur anywhere that a class name would ordinarily

beused,e.g.,in slotconstraintsand inthelist ofsuperclasses.Forexample,in

Figure 1(which usesOIL's\humanreadable"presentationsyntaxrather than

themoreverboseRDFSserialisation),aherbivoreisdescribedasananimalthat

eatsonly plants orpart-ofplants.Pointsto noteare that universally quantied

(value-type) and existentiallyquantied (has-value) slot constraintsareclearly

dierentiated, and that the constraint on theeats slot is adisjunction, one of

whosecomponentsisananonymousclassdescription(inthiscase,just asingle

slotconstraint).Inaddition,itisassertedthat thepart-ofslotistransitive,and

thatitsinverseistheslothas-part.Furtherdetailsofthelanguagewillbegiven

inSection 3,andacompletespecicationcanbefoundin[7].

slot-def part-of

subslot-of structural-relation

inversehas-part

properties transitive

class-def denedherbivore

subclass-of animal

slot-constraint eats

value-typeplantOR

slot-constraint part-ofhas-value plant

Fig.1.OILlanguageexample

(4)

OilEdis asimpleontologyeditor that supports the constructionof OIL-based

ontologies. The basicdesign has beenheavily inuencedby similar tools such

as Protege [13] and OntoEdit [12], but OilEd extends these approaches in a

numberofways,notablythroughanextensionofexpressivepowerandtheuse

ofareasoner.

However, OilEd is not intended as a replacement for such tools|the cur-

rentimplementation of OilEdis intendedprimarily asa prototypeto test and

demonstrate novel ideas,and compromiseshave been made in the design and

implementation. For example, the tool does not provide key functionality for

collaborativeontologydevelopmentsuchasversioning,integrationandmerging

ofontologies.Similarly,thepowerfultailorabilityandknowledgeacquisitionas-

pectsoftoolssuchasProtegehavebeenignoredcompletely.Rather,thedesign

hasconcentratedondemonstratinghowtheframeparadigmcanbeextendedto

dealwithamoreexpressivemodellinglanguage,andhowreasoningcanbeused

tosupportthedesignandmaintenanceofontologies.

3.1 OilEdFunctionality

Basicfunctionalityallowsthedenitionanddescriptionofclasses,slots,individ-

ualsand axiomswithin an ontology.Ingeneral, editingfunctions are provided

through graphical means|mousedriven drop down menus, toolbars and but-

tons.We will notprovide adetailed descriptionof thegraphical userinterface

here,asit isrelativelystandard(see Figure2, which providesascreen shotof

theeditorsclassdenitionpanel).Instead,wewilldiscussthenovelfunctionality

oeredbythetool.

Frame Descriptions The central component used throughout OilEd is the

notionofaframedescription. Thisconsistsofacollectionofsuperclassesalong

with a list of slot constraints. This is similar to other frame systems. Where

OilEddiers,however,isthat whereveraclassname canappear,arecursively

dened,anonymousframedescriptioncanbeused.Inaddition,arbitraryboolean

combinationsofframesorclasses(usingand,orandnot)canalsoappear.This

isin contrastto conventional framesystems,where in general,slot constraints

andsuperclassesmustbeclassnames.

As well as being able to assert individuals as slot llers, several types of

constraintson slot llers can be asserted (these kinds of constraint are some-

timescalled facets).These include value-typerestrictions(all llersmust be of

aparticular class), has-valuerestrictions(there must be at least one llerof a

particularclass),andexplicitcardinalityrestrictions(e.g.,atmostthreellersof

agivenclass).Eachconstrainthasaclearlydenedmeaning,removingthecon-

fusionpresentin someframesystems,where, forexample,itisnotalwaysclear

whether thesemanticsof aslot-constraintshould beinterpreted asauniversal

orexistentialquantication.

Class DenitionsA class denition species the class name, along with an

(5)

isdenedorprimitive.Ifdened,theclassistakentobeequivalenttothegiven

description(necessaryandsuÆcientconditions).Ifprimitive,theclass istaken

tobeanexplicitsubclassofthegivendescription(necessaryconditions).Inthe

specicationoftheOILlanguage,classescanhavemultipledenitions.InOilEd,

thisisdisallowedforimplementationreasons.Insteadclassesmusthaveasingle

denition, but the sameeect can be achieved through theuse of equivalence

axiomsasdiscussedbelow.Ontologiesusingmultipledenitionscanbereadby

the tool. The rst denition encountered will be used as the class denition,

withanysubsequentdenitionsbeingtranslatedtotheappropriateaxioms.

Slot DenitionsAslot denition givesthenameof theslot andallowsaddi-

tional propertiesof theslot tobeasserted,e.g.,thenamesof anysuperslotsor

inverses.Ifrisasuperslotof s,thenanytwoobjectsrelatedviasmustalsobe

relatedviar(i.e.,s(a;b)!r(a;b));ifrisaninverseofs,thenaisrelatedtobvia

sibisrelatedtoaviar(i.e.,s(a;b)$r (b;a)).Domainandrangerestrictions

ona slot canalsobespecied.Forexample,wecanconstrain therelationship

parentto have bothdomain and range person, asserting that only personscan

have,andbe,parents.Aswithclass descriptions,thedomainandrangerestric-

tionscan bearbitrary class expressions such as anonymous frames orboolean

(6)

global, andapplytoeveryoccurrenceoftheslot,whether explicitorimplicit.

Aslotrcanalsobeassertedtobetransitive(i.e.,r (a;b)andr(b;c)!r (a;c)),

functional(i.e.,r (a;b)andr(a;c)!b=c)orsymmetric(i.e., r(a;b)!r (b;a)).

All assertions made aboutslots are used by the reasoner, and may induce

hierarchicalrelationshipsbetweenclasses,e.g.,asaresultofdomain andrange

restrictions.

AxiomsAnother areawhere theexpressive powerof OIL/OilEdexceedsthat

oftraditionalframelanguages/editorsisin thekindsofaxiom thatcanbeused

to assert facts aboutclasses and their relationships.As well asstandard class

denitions (which are really arestricted form of subsumption/equivalence ax-

iom),OilEdaxiomscanalsobeusedtoassertthedisjointnessorequivalenceof

classes(with the expected semantics) along with coverings. A coveringasserts

thateveryinstanceofthecoveredclassmustalsobeaninstanceofatleastone

ofthecoveringclasses.Inaddition,coveringscanbesaidtobedisjoint,inwhich

case everyinstance of the coveredclass mustbe an instance of exactlyone of

thecoveringclasses.

Again,these axiomsarenotrestrictedto classnames,but caninvolvearbi-

traryclass expressions(anonymous framesorboolean combinations).This isa

verypowerfulfeature,andisoneofthemainreasonsforthehighcomplexityof

theunderlyingdecisionproblem.

IndividualsLimitedfunctionalityisprovidedtosupporttheintroduction and

descriptionof individuals|the intention within OilEd is that such individuals

are foruse within class descriptions,rather thansupportingthe production of

largeexistentialknowledgebases (itissupposed that RDF/RDFSwill beused

directly for this purpose). As an example, we may wish to dene the class of

Italians as being all those Persons who were born in Italy, where Italy is not a

classbutanindividual.

As theFaCTsystem doesnotsupport reasoningwith individuals,they are

treated(forreasoningpurposes)asdisjointprimitiveclasses.Thisisnotanideal

solutionasitdoesleadtosomeinferencesbeinglost,inparticularthoseresulting

fromtheinteractionbetweenindividualsandmaximumcardinalityconstraints.

E.g., it would not be possible to infer that Persons who are citizens of Italy,

andof nootherCountry, arecitizensof atmostoneCountry. Workis currently

underwayto extendtheFaCTreasonertodealexplicitlywithsuch individuals,

sothat completeinferencecanbeprovided.

ConcreteDatatypesConcretedatatypes(stringandintegers),alongwithex-

pressionsconcerningconcrete datatypes(suchasmin, maxorranges)canalso

beusedwithinclassdescriptions.However,theFaCTreasonerdoesnotsupport

reasoningoverconcretedatatypes,andatpresentOilEdsimplyignoresconcrete

datatype restrictions when reasoningaboutontologies. The theory underyling

concretedatatypesis,however,wellunderstood[16],andworkisalsoinprogress

(7)

for the denition of data types. These are not fully supported in our current

versionofOilEd.

3.2 Reasoning

Inadditiontotheextendedexpressivitydiscussedabove,OilEd'sprincipalnov-

elty isin its useof reasoningto check classconsistency andinfersubsumption

relationships. Reasoning services are currently provided by the FaCT system,

but in principal any reasonerwith the appropriate functionality/connectivity

couldbeused.

FaCTisaDLclassierthatoerssoundandcompletereasoning(satisabil-

ity,subsumptionandclassication)fortwoDLs:SH FandSH Q. FaCT'smost

interestingfeaturesareitsexpressivelogic(inparticulartheSH Qreasoner),its

optimisedtableauximplementation(whichhasnowbecomethestandardforDL

systems),anditsCORBAbasedclient-serverarchitecture[15].

TheSH QlanguagecancompletelycaptureOIL ontologies, withtheexcep-

tionoftworecentlyadded features:concrete datatypes(strings, numbers,etc.)

and named individuals in class descriptions. As mentioned above, individuals

canbedealtwithbytreatingthemaspairwisedisjointatomicclasses(although

with somelossofinferentialpower), whileextendingFaCT todeal withOIL's

concretedatatypesshould berelativelystraightforward.

FaCT's optimisations arespecically aimedat improving thesystem'sper-

formancewhenclassifyingrealisticontologies. Theseoptimisations leadtoper-

formance improvements of several orders of magnitude when compared with

older DL and modal logic reasoners, and make the use of reasoning support

feasableinspiteofthediscouragingworstcasecomplexityoftheunderlyingde-

cisionproblem(ExpTime).Theperformanceimprovementisoftensogreatthat

itis impossibleto measureprecisely asunoptimisedsystemsarevirtually non-

terminatingwith ontologies that FaCT is easily ableto deal with [15].Taking

alargemedicalterminologyontologyasanexample[17],FaCTisabletocheck

theconsistencyof all 2,740classes anddetermine thecomplete classhierarchy

in about45seconds of(700MHzPentiumIII) CPU time;unoptimisedsystems

have been run for several weeks without their completing even a single class

consistencytest.

In the current version of OilEd, reasoning is performed on a \single-shot"

basis,i.e.,atsomesuitablepointtheuserconnectstothereasonerandrequests

vericationoftheontology.ConnectionisviaFaCT'sCORBAbasedclient-server

interface,which hastheadvantagethat FaCT servers(s)canberunningeither

locally or remotely,and can provideaservice to manyOilEdusers. Moreover,

the FaCT system has reasoning engines for both SH Q and SH F knowledge

bases,and if both servicesare available the usercan choose to connect to the

faster SH F reasonerto verify anontologythat doesnotinclude either inverse

slotsorcardinalityconstraints.Thecurrentimplementation simplyinformsthe

userifthisisappropriate;futureenhancementswillinclude automaticselection

(8)

Whenvericationisrequested,theontologyistranslatedintoanequivalent

SH Q(orSH F)knowledgebase andsentto thereasonerforclassication[18].

OilEdthenqueriestheclassiedknowledgebase,checkingforinconsistentclasses

andimplicitsubsumptionrelationships.Theresultsarereportedtotheuserby

highlightinginconsistentclasses andrearranging the class hierarchy displayto

reect any changes discovered.FaCT/OilEddoesnotprovide any explanation

ofitsinferences,althoughthiswould clearlybeusefulinontologydesign[19].

Figures 3and 4showthe eects ofclassicationon (part of)thehierarchy

derived from the TAMBIS ontology (see Section 4). When verifying the on-

tology, anumber of newsubsumption relationships are discovered(due to the

classdenitions in themodel). Inparticularwecansee that,afterverication,

holoenzymeisnotonlyanenzyme,butalsoaholoprotein,andthatmetal-ionand

small-moleculearebothsubclassesofcofactor.

During subsequent editing, changes to the ontologyare notcommunicated

tothereasonerinstantaneously,butonlywhenexplicitlyrequestedbytheuser.

(9)

thesimpleinteractionmodel describedherewasconsideredappropriateforthe

initialprototype.

3.3 Export

AlthoughOilEd isprimarily intendedasan editorfor OIL ontologies, thetool

willexporttoanumberofformats.These includeOIL Standard(the \human-

readable"presentation format forOIL that wasused in Figure 1), OIL-RDFS

(OIL'sstandardRDFSserialisation)andDAML+OIL(alsoRDFS).Inaddition,

ontologiescanbeexportedasHTML,facilitatingviewingoftheontologywithout

the tool and class hierarchies generated by the classier can be exported as

graphsforviewingwithAT&T's Dotty 2

application.

ByexportingontologiesasRDFS, itisenvisagedthat \RDFS-aware"appli-

cations will be able to read and interpret OIL ontologies even if they are not

fully\OIL-aware".Ofcourse,suchapplicationswouldbeunabletomakeuseof

alloftheinformationinthemodel,butmaybeabletouse,forexample,thesub-

classhierarchieswithintheontology.Inordertofacilitatethis,OilEdallowsthe

2

(10)

tologybeforeexport, thusmakingthisinformationavailableto non-OILRDFS

applications,orevenOIL-awareapplicationsthat donotemployreasoning.

4 Case Study: the TAMBIS Ontology

The r^ole of ontologies in bioinformatics (the discipline of applying computing

to molecular biology) hasbecome prominent in the last few years. Ontologies

are usedas amechanismfor expressing and sharingcommunityknowledge,to

dene common vocabularies (e.g., for database annotations), and to support

intelligentqueryingovermultiple databases[20].TAMBIS(TransparentAccess

toMultipleBioinformaticInformationSources)isamediationsystemthatuses

anontologytoenablebiologiststoaskquestionsovermultipleexternaldatabases

usingacommonqueryinterface.TheontologyiscentraltotheTAMBISsystem:

it provides a model that queries can be formed against, it drives the query

formulation interface, it indexes the middleware wrappers of the component

sources,anditsupportsthequeryrewritingprocess[21].TheTAMBISontology

(TaO) covers the principal concepts of molecular biology and bioinformatics:

macromolecules;theirmotifs,theirstructure,function,cellularlocationandthe

processes in which they act. It is an ontologyintended for retrieval purposes

ratherthan hypothesis generation,soit isbroadand shallowratherthan deep

andnarrow[20].

TheTaOwasoriginallymodelledintheDLGrail[17].Itwassubsequently

migratedtoOILinorderto(a)exploitOIL'shighexpressivitysoastomaintain

abetter delitywith biological knowledge asit is currently perceived; (b) use

reasoning support when building and evolving complex ontologies where the

knowledge is dynamic and shifting; and (c) be able to deliver the TaO as a

conventionalframeontology(withallsubsumptionsmadeexplicit),thusmaking

itaccessibleto awiderrangeof(legacy)applicationsandcollaborators.

Theapproachtodevelopingtheontologywasdirectlyinuencedbytherange

ofexpressivitythatOILaords,andthecapabilitiesofOilEditself,particularly

itsreasoning facilities.Themodellingphilosophy wasto bedescriptive,i.e., to

modelpropertiesandallowasmuchaspossibleofthesubsumptionlatticetobe

inferredbythereasoner.Thedesignmethodologywastorstconstructabasic

framework of primitive foundation classes and slots, working both top down

and bottom up, mainly using explicitly stated superclasses.The ontologywas

thenincrementallyextendedandrenedbyaddingnewclasses,elaboratingslot

llersandconstraints,and \upgrading"todened classeswhereverpossible,so

thatclassspecicationsbecamesteadilymoredetailed andmoreaccurate.This

process wasguided by subsumption reasoning|whenelaborating or changing

classes,thereasonercouldbeusedtocheckconsistencyandtoshowtheimpact

(11)

sentation syntax) that we will use to illustrate this methodology. 3

Originally,

holoprotein,enzyme andholoenzymewereallprimitiveclasses,withnoslotcon-

straints,andanexplicitlyassertedclasshierarchy:holoproteinandenzymewere

subclassesofprotein, andholoenzyme wasasubclassofenzyme. Duringtheex-

tensionandrenementphase,thepropertiesofthevariousclassesweredescribed

in moredetail: it was assertedthat aholoproteinbindsa prosthetic-group, that

anenzyme catalyses areaction, and that aholoenzyme bindsaprosthetic-group.

Severaloftheclasseswerealsoupgradedtobeingdenedwhentheirdescription

constitutedbothnecessaryandsuÆcientconditionsforclass membership, e.g.,

a protein is a holoprotein if and only if it binds a prosthetic-group. This allows

thereasonerto inferadditionalsubclassrelationshipsw.r.t.holoprotein, andin

particular that holoenzymeis asubclass of holoprotein. This latterrelationship

probablywouldhavebeenmissediftheontologyhadbeenhandcrafted.

class-def protein

class-def dened holoprotein

subclass-of protein

slot-constraintbindshas-value prosthetic-group

class-def dened enzyme

subclass-of protein

slot-constraintcatalyseshas-value reaction

class-def dened holoenzyme

subclass-of enzyme

slot-constraintbindshas-value prosthetic-group

class-def dened cofactor

subclass-of (metal-ionor small-molecule)

disjointmetal-ionsmall-molecule

Fig.5.SimpliedfragmentofTAMBISontology

The extension and renement phase also included the addition of axioms

asserting disjointness, equality and covering, further enhancing the accuracy

of the model. Referringagain to Figure 5, ourbiologist initially assertedthat

cofactorwasasubclass of bothmetal-ionand small-molecule(a commonconfu-

sionoverthesemanticsof'and'and'or')ratherthanbeingeitherametal-ionora

small-molecule.Subsequently,whenitwasassertedthatmetal-ionandsmall-molecule

are disjoint, thereasoner inferred that cofactor was logicallyinconsistent, and

themistake was rectied.Modelling mistakessuchas these litter bioontologies

craftedbyhand.

Other advantagesderivedfrom theuseofOilEdincluded:

3

The complete ontology can be found at http://img.cs.man.ac.uk/stev ens/

(12)

language, made ontology developmentmuch less daunting to our biologist

thanwritingSH Qlogicexpressionswouldhavebeen.

{ ClipboardfacilitiesprovidedbyOilEdallowed(partsof)framestobecopied

andpasted,makingiteasytoexperimentwithnewdenitionsandtomain-

tain a consistent modelling style. E.g., coenzymeA-requiring-oxidoreductase

wasbuiltbycopyingnad-requiring-oxidoreductaseandchangingtheconstraint

on thebindsslot from nadto coenzymeA. Thereasonerthenautomatically

migrated theclassfrom beingasubclass ofholoenzymeto beingasubclass

ofcoenzyme-requiring-enzyme.

{ Class denitions canbe assimple aspossible yet ascomplexasnecessary.

PartsoftheTaOaresimplyprimitiveframesandslots;otherpartsarevery

elaborate andexploitthefullexpressivepoweroftheOILlanguage.

{ In TAMBIS, the ontology is managed by an ontology server that makes

fulluseoftheclassdenitions,e.g.,toclassifyusergeneratedqueryclasses.

However,beingabletodeliverastatic\snapshot"oftheontologyintheform

ofanRDFStaxonomyhasprovedextremelyconvenientwhenworkingwith

collaboratorswhoarebuildingontologiesthatareinfactsimpletaxonomies,

suchastheGeneOntology [22].

5 Conclusion

Ontologiesareusefulinarangeofapplications,andwillplayapivotalr^oleinthe

SemanticWeb, wheretheywillprovideasourceofpreciselydenedtermsthat

canbecommunicatedacrosspeopleandapplications.Reasoningwithrespectto

suchtermswillbeimportantforboththedesignanddeploymentofontologies.

WehavepresentedOilEd,anontologyeditorthat hasaneasyto useframe

interface,yetatthesametimeallowsuserstoexploitthefullpowerofanexpres-

sivewebontologylanguage(OIL/DAML+OIL).WehavealsoshownhowOilEd

uses reasoning to support ontology design and maintenance, and presented a

casestudy illustratinghowthis facilitycan beused to developontologiesthat

describetheirdomainsin moredetailandwithgreateraccuracy.

OilEd is aprototype,designed to test and demonstrate novelideas, and it

stilllacksmanyfeaturesthatwouldberequiredofafully-edgedontologydevel-

opmentenvironment,e.g.,itprovidesnosupportforversioning,orforworking

withmultipleontologies.Moreover,thereasoningsupportprovidedbytheFaCT

systemisincompleteforOILextendedwithconcretedatatypesandindividuals,

anddoesnotinclude additionalservicessuchas explanation. However,in spite

ofthese shortcomings,OilEdisalreadysuÆcientlywelldeveloped tobeavery

usefultool,andto demonstratetheutilityofOIL'sintegrationoffeaturesfrom

frame,DLandweblanguages.

References

1. G. van Heijst, A. Schreiber, and B. Wielinga. Using explicit ontologies inKBS

(13)

FOIS-98,1998.

3. M.UscholdandM.Gruninger. Ontologies:Principles,methodsand applications.

K.Eng.Review,11(2):93{136,1996.

4. T.Berners-Lee. WeavingtheWeb. OrionBusinessBooks,1999.

5. S.Deckeretal. Thesemanticweb|ontherespectiverolesof XMLandRDF.

IEEEInternet Computing,2000.

6. T.R.Gruber. Towardsprinciplesforthedesignofontologiesusedforknowledge

sharing. InProc.of Int.WorkshoponFormalOntology,1993.

7. D.Fenseletal. OILinanutshell. InProc.ofEKAW-2000,LNAI,2000.

8. J.Hendler and D. L. McGuinness. TheDARPA agent markuplanguage. IEEE

IntelligentSystems,jan2001.

9. D.L.McGuinness,R.Fikes,J.Rice,andS.Wilder. Anenvironmentformerging

andtestinglargeontologies. InProc.of KR-00,2000.

10. J.DoyleandR.Patil. Twothesesofknowledgerepresentation. ArticialIntelli-

gence,48:261{297,1991.

11. V. K. Chaudhri et al. OKBC: A programmaticfoundation for knowledgebase

interoperability. InProc.ofAAAI-98,1998.

12. S.StaabandA.Maedche.Ontologyengineeringbeyondthemodelingofconcepts

andrelations. InProc.oftheECAI'2000 WorkshoponApplicationofOntologies

andProblem-Solving Methods, 2000.

13. W.E.Grossoetal. Knowledgemodelingatthemillennium(thedesignandevo-

lutionofprotege-2000). InProc.ofKAW99,1999.

14. I.Horrocks,U.Sattler,andS.Tobies.Practicalreasoningforexpressivedescription

logics. InProc.ofLPAR'99,pages161{180,1999.

15. I. Horrocks. Benchmark analysis with fact. In Proc. TABLEAUX 2000, pages

62{66,2000.

16. F.BaaderandP.Hanschke.Aschemeforintegratingconcretedomainsintoconcept

languages. InProc.ofIJCAI-91,pages452{457,1991.

17. A.Rectoretal. TheGrailconceptmodellinglanguageformedicalterminology.

ArticialIntelligenceinMedicine,9:139{171, 1997.

18. S.Deckeretal. Knowledgerepresentationontheweb.InProc.ofDL2000,pages

89{98,2000.

19. D.McGuinnessandA.Borgida. Explainingsubsumptionindescriptionlogics. In

Proc.of IJCAI-95,pages816{821,1995.

20. P.GBaker,C.A Goble,S.Bechhofer,N.WPaton,R. Stevens,andA. Brass. An

OntologyforBioinformatics Applications. Bioinformatics,15(6):510{520,1999.

21. C.Goble,R.Stevens,G.Ng,S.Bechhofer,N.W.Paton,P.G.Baker,M.Peim,and

A. Brass. Transparent Access to Multiple Bioinformatics Information Sources.

IBMSystemsJournal, 40(2),2001.

22. M. Ashburneret al. Geneontology: Toolfor the unicationof biology. Nature

Genetics,25:25{29, 2000.

Referenzen

ÄHNLICHE DOKUMENTE

Abstract: As advanced analytics has become more mainstream in enterprises, usability and system-managed performance optimizations are critical for its wide adoption. As a

Setting up the environment can be facilitated by Web service technology because the corresponding IT resources are seen as being resource within a Grid run by the service

Note that Definition 1 explains why (Horn-SHOIQ) ontologies captured by any of the normative profiles contain only safe roles: in the case of EL, roles can be existentially

In Proceedings of the International Workshop on Using Linguistic Information for Hybrid Machine Translation (LIHMT 2011) and of the Shared Task on Applying Machine Learning

In Proceedings of the International Workshop on Using Linguistic Information for Hybrid Machine Transla- tion (LIHMT 2011) and of the Shared Task on Ap- plying Machine

If there are no similar concepts found in step 1 or the predicted SSA values are lower than the threshold, the system does not make a recommendation which might mean that the

We may thus conclude that both the viscosity and ro- tation suppress the instability of the superposed grav- itating streams when the streams rotate about an axis in the

1’028 experimental animals with plastic or electronic ear tags were examined both before scalding and after dehairing of the carcase at five different abattoirs.. Results