SeanBechhofer 1
,LeslieCarr 2
,CaroleGoble 1
,SimonKampa 2
,andTimothy
Miles-Board 2
1
InformationManagementGroup
DepartmentofComputerScience,KilburnBuilding
UniversityofManchester,OxfordRoad,ManchesterM139PL
http://img.cs.man.ac.uk
2
Intelligence,Agents,MultimediaGroup
DepartmentofElectronics&ComputerScience
UniversityofSouthampton,Higheld,SouthamptonSO171BJ
http://www.iam.ecs.soton.ac.uk
Abstract. Semantic metadatawill playa signicant roleinthe
provi-sionoftheSemanticWeb.Agentswillneedmetadatathatdescribesthe
content of resources inorder to perform operations, suchas retrieval,
overthoseresources.Inaddition,ifrichsemanticmetadatais supplied,
thoseagents canthen employreasoning over themetadata, enhancing
theirprocessingpower.Keytothisapproachistheprovisionof
annota-tions,boththroughautomaticandhumanmeans.Thesemanticsofthese
annotations, however, intermsof the mechanismsthroughwhichthey
areinterpretedandpresentedtotheuser,aresometimesunclear.Inthis
paper, weidentifyanumberofcandidateinterpretationsofannotation,
anddiscusstheimpacttheseinterpretationsmayhaveonSemanticWeb
applications.
1 Introduction
The SemanticWeb(SW) vision, asarticulatedbyTim Berners-Lee[2],is ofa
Web in which resources are accessible not only to humans, but also to
auto-mated processes,e.g., automated \agents" roaming theweb performing useful
taskssuchasimprovedsearch(intermsofprecision)andresourcediscovery,
in-formationbrokeringandinformationltering.Theautomationoftasksdepends
onelevatingthestatusofthewebfrommachine-readabletosomethingwemight
call machine-understandable. Thekeyideais to havedata on theweb dened
andlinkedinsuchawaythatitsmeaningisexplicitlyinterpretablebysoftware
processesratherthanjustbeingimplicitlyinterpretablebyhumans.
To realise this vision, it will benecessary to associate metadata (i.e., data
describingcontent/functionality)withwebresources.Onemechanismfor
asso-ciating such metadata is annotation. In particular, we may wish to annotate
resourceswith semanticmetadatathatprovidessomeindicationofthecontent
annota-requirelanguageswhichwill supporttherepresentationofsemanticmetadata.
Standardisation proposals for metadata languages have already been
sub-mittedto theWorldWideWebConsortium(W3C), inparticular theResource
DescriptionFramework (RDF)and RDFSchema(RDF(S)) {see [8]for a
dis-cussionoftherolesoftheselanguagesandofXML/XMLSchema.However,such
annotations will be of limited value to automated processes unless they share
a commonunderstanding as to their meaning.Ontologies (which havealready
provedtheirusefulnessinarangeofapplicationdomains[28,23,26]canhelpto
meet this requirement by providing a\representationof a shared
conceptuali-sationof aparticulardomain"andashared,controlledvocabularythatcanbe
communicatedacrosspeopleand applications[11,12].
Inadditiontotherequirementforrepresentationlanguagesthatsupportthe
sharing and exchange of semantic information between applications, we must
alsohaveacommonunderstanding oftheannotationprocess.Schemasto
sup-port annotation have been developed[19], but these do not explicitly provide
support for this understanding. What does it mean when we make an
anno-tation, and what are the implicit tasks that are beingperformed? Within the
SWcontext,confusionreignsastotheinterpretationoftheannotationtask.In
order to support theuseof automatedagents(a central tenetto the Semantic
Web vision),wemustbeexplicit abouttheassumptionsthatwemakeandthe
contextwithinwhichsuchannotationsshouldbeinterpreted.Notethatouruse
ofthephrase\the semantics ofsemanticannotation"refersto theprovisionof
aconsistent interpretationof the task,but wedo notintend to presentherea
formalsemantics,suchasthat providedforlanguageslikeDAML+OIL [27].
Thepaperis structured asfollows.Werstgiveabriefintroduction tothe
COHSE projectandtheapproachbeingadoptedthere. Thisgivesanoverview
of ourmotivationand somecontextfor thefollowingdiscussions. Wethen
dis-cussannotationandprovideaclassicationofannotationtasksalongwiththeir
intended semantics. We discuss the related issue of identication and nally
concludewithsomeremarksconcerningfuturedirectionsandrecommendations.
Referencestoexisting andrelatedworkaremadethroughoutthepaper.
2 The COHSE Project
OurinterestinannotationhereiswithinthecontextoftheCOHSE(Conceptual
OpenHypermediaService)project[3].COHSEaimsto bringtogetheranopen
hypermediaarchitecture(inparticulartheDistributedLinksService[4]orDLS)
with ontological services in order to provide an architecture for the Semantic
Web[10].
Detailed descriptions of the COHSE system can be found in [3,10]. Put
briey, theCOHSEapproachconsists ofaCOHSEagent (alongwith
tememployseither aspecialistbrowser(basedonMozilla 3
)oraproxythrough
which all http requests are routed. The words and terms that appeared in a
documentareusedas entrypointsto anontology.Therelevantconceptsinthe
ontologycanthenbeusedtodetermineappropriatetargetsforlinks outofthe
givenresource.KeytothenoveltyoftheCOHSEapproachistheprovisionofan
editorial component within theagent.Thiscomponentusesinformationwithin
theontology(suchashierarchicalclassication)in order todetermine whether
the links are suitable or to perhaps expand orcull the set of possible targets.
Figure1showsasimplied viewofthebasicarchitectureofthesystem.
Figure2showsapagetakenfromSun'sJavaTutorialwebsite.InFigure3,
weseethesamepageaugmentedbytheCOHSEagent.Anumberoflinkanchors
(signiedbythesmall\L"icon)havebeenaddedtothepage.Oneofthesehas
been opened up, and we see a collection of possible targets which have been
annotated as being \about" the particular concept selected { in this case the
conceptofbyte.
Inaddition to using the wordsand phrasesthat appear in the documents,
theCOHSEagentcanalsouseexplicitmetadataappliedtotheresources(rather
thanrelyingsolelyonmappingsfromwordsandterms).Thisapproachrelieson
theabilityto annotate resourceswith semanticmetadata{where bysemantic
metadata we mean the explicit binding of concepts to resources rather than
the use of terms and words as simple proxies for the concepts. The explicit
annotationscanthenhelpguidetheeditorial componentinitslinkingstrategy.
For example, if a passage in a web page has been annotated as being about
aparticularsubject,sayprogrammingdatatypes,theeditorialcomponentmay
knowthattherearecertaintermsthat shouldbefocusedonwithinthecontext
of that annotation(saythe termsint oroat){an exampleofan agentusing
semanticinformationtomakedecisionsasto itsbehaviour.
2.1 Produce and Consume
Thesituationhasparallelswiththeunderlyingmotivationfortheuseofrich
lan-guagesforrepresentingcontentontheSemanticWeb.LanguageslikeDAML+OIL
[6]arebeingproposed as mechanismswhich provide\machine-processable"
se-manticinformation.Theyprovideanexplicitrepresentationoftherelationships
between terms and concepts which canthen be used by reasoners orsoftware
agents to interpret those terms and concepts. The vision is one of providing
sharedconceptualizations,whichthenallowcommunitiestoshareandexchange
informationunambiguously.
WithinCOHSE(orindeedotherSWsystems),therearetwocomplementary
strands,withannotationprovidersenrichingcontentandannotationconsumers
using those annotations to process, organise and present information to end
users.Theconsumercouldbeasophisticatedontologicalsearchprocessorportal,
or alternativelydocumentenrichmentthrough theadditionof links asusedby
provided bythe use of concept models or ontologies, part of it is provided by
sharedassumptionsaboutthewaythesetermsare tobeused. Ifwebuildrich
models,and thenusethem in ahaphazardfashion, weare in somewayselling
ourselvesshort.
2.2 Linking as Annotation
The provision of dynamic linking as used by the COHSE projectcan beseen
as akindofannotation{in thiscasehypertextlinksarebeingprovidedrather
thansometextualannotation.Thiswillbediscussedinfurtherdetailbelow(as
anexampleof Link Annotation), but weintroducethenotionhereasithas
relevancetothedescriptionofCOHSE.
Koivunenet al. [18] discuss approaches to Web annotationsand categorise
systems as, in the main, either proxy-based or browser-based. In a
proxy-basedapproach,the annotationsanddocument aremergedbytheproxy, with
the browser seeing only the merged documents. In abrowser-basedapproach,
a specialist browser application will merge the annotations with the original
documentswhile browsing.Annotations canbestoredseparately andprovided
viasomeannotationservice(orkeptwithintheproxyitself).
COHSEhastwoimplementations,aseitherproxyorbrowser{theproxy
edge base forretrieval,and to provideanchorsfor links astheannotationsare
used to derive outward links from resources.At its simplest, then, annotation
within COHSEcanbeseenas amechanismthatallowstheusertospecify
pos-sible link anchors within a document, with the anchor being associated with
a conceptual description. This descriptionwill then be used to determine
ap-propriatelinks atread-time.Simultaneously,theannotationsarebeingused to
providelink targets(asisthecaseinother, resource-discoverybased,systems).
2.3 Extending Simple Annotation
COHSE'scurrentimplementationadoptsabasicapproachtotheinterpretation
of annotation{an annotationsimplyassociatesaresourcewith aconceptand
no attempt ismade to disambiguatethe relationshipbetweentheconcept and
resource.This simpleapproach hasserveduswelland allowsusto extendand
enrichthehypertext.AnexperimentbasedonSun'sJavaTutorialsite 4
hasbeen
conducted and an evaluation ofthe resultinghypertextstructure produced by
theCOHSEagentshowspromisingresults 5
.
Apossibleextensiontothis situation istoprovidefurtherinformation that
describesinmoredetailtherelationshipbetweentheresourceandtheannotation
concept.Thisthenhasanimpactin twoways:
{ itcanaectthewaythat theagentpresentsthelink anchor;
{ itcanaectthewaythat possiblelinktargetsarefoundordisplayed.
Thisleadsustoadesiretoclassifyandcategorisethedierentwaysinwhich
this association between resource and concept could be made. The remainder
of the paper proposes a number of dierent interpretations of the annotation
process and discusses how those interpretations could aect the behaviour of
systemssuchasCOHSE.
3 Annotation
annotationnoun.A notebywayofexplanationorcommentaddedto
atextordiagram. NewOxfordDictionary ofEnglish
Annotationtakesmanyformsandthereareanumberofwhatwecouldterm
\popular"ideasofannotation.Marshall[21]writesthat \[annotation]hasbeen
construed in manyways: aslink making,as path building, ascommentary, as
markinginoraroundexistingtext,asadecenteringofauthority,asarecordof
reading andinterpretation, orascommunitymemory".Herewebrieypresent
aroughclassicationofannotationtypes.Wewill returntothisin moredetail
in thelatersectiononsemantics.
4
http://java.sun.com/tutorial
added toresources.Annotationsof thiskindhavebeenusedfor manyyears in
communitiessuchasbiology.ForexampletheSWISS-PROTdatabase[24]
con-tainsproteinsequenceinformationalongwithannotationsdescribingfunctions,
structure, domains,sites and soon.WithinadatabaselikeSWISS-PROT,the
annotationsare rst-class citizens, andare, in eect,the data. Although some
use is made of controlled vocabularies such as GO [25], the hand-crafted and
hand-curatedannotationsareprimarilyaimedathumanreaders.
This is the kind of activity supported by Annotea [17,29,18]. Extensions
to the basic schema allow the use of richer annotation types [5] (for example
commentariescanbemarked asreplies orgatheredinto threaded discussions),
but aprinciple characteristic of this approach is that it is primarily aimed at
humanreaders(andauthors).
Systemssuch asthe DistributedLinks Service (DLS)[4] or3rd Voice [20],
allowtheadditionoflinkstoarbitrarydocuments(includingthoseincontrolof
athird party).ThisLink Annotationextends thetextualannotationnotion,
where herethe contentof theannotation isgiven,not bysome text, but bya
linkdestination(andpossiblyassociatedbehaviour).Again,linkannotationcan
beseentobeanactivityprimarilytargetedathumanreaders.
Finally,wecanconsiderwhatmightbecalledSemanticAnnotation,where
thecontentoftheannotationconsistsofsomerichsemanticinformation 6
.This
idea of semantic annotationhas been pursuedin boththe Ontobroker[7] and
SHOE [16] projectsandmorerecentlyin COHSE [3]. InbothOntobrokerand
SHOE,specialisedmarkupwasinsertedintowebpages{thismarkupcontained
semanticinformation drawn from an ontologyproviding richer descriptions of
resourcecontent. InCOHSE,amoreopen annotationframework followingthe
DLSphilosophyisinuse,allowingthedecorationofarbitraryresourceswithout
thenecessitytocontroltheoriginaldocument.SemanticAnnotationistargeted
notonlyat humanreadersof resources,but alsoat softwareagents{thisdoes
bring withittherequirementthatrelationshipsareexplicitlyrepresented.The
useofsemanticinformationtakenfromwelldenedontologieswillallowagents
tomakedecisionsbasedonthoseresourcedescriptions(forexampletheCOHSE
editorialcomponentasdescribedabove).
Returning to Marshall [21], a number of dierent axes or dimensions are
identied that reect the forms of annotation. Included in these are anotion
of formal vs. informal. Informalincludes personal noteswritten in themargin
while readinganarticle.Formalisdeemedto bemetadatafollowingstructural
standardsand assignedvaluesusing conventional naming authorities. The use
of semanticannotation, drawing onconceptual models represented using
well-dened knowledge representation languages canbe seen to sit at the extreme
end of theformalspectrum,perhapsevenmoreso thanMarshall's original
in-tension of formality. Also of interest is the identication of explicit vs. tacit
annotation.Accordingto Marshall,manypersonalannotationsaretacit{they
6
terpretation. Forexample,abookmark, highlightedsentence ortheannotation
\No!"areexamplesof tacitannotationasweneedextrainformationaboutthe
annotator, or thehistoryof theannotationprocess in orderto interpret them.
AnexplicitannotationwillcarrysuÆcientinformationforitsinterpretation.As
Marshallsays,thedimensionof explicitvs.tacitiscruciallyrelatedto
intelligi-bility {in thecontextof theprovisionofmarkup intended for software agents
orprocessestherequirementofexplicitnessisparticularlystrongassuchagents
will notpossessthereal worldknowledge,reading history,culturalbackground
andsoon,ofhumanreaders.
4 Semantics of Annotation
Herewepresentaclassicationofpossibleusesofannotation.Thiscanbeseenas
a classicationof thepossiblesemantics of theannotation relationship(where
here we use the term semantics in a loose fashion). For the purposes of this
discussion, we consider the following situation. A web page with the URL U
is being viewed and a region of the document corresponding to an XPointer
expressionXhasbeenselected.Thisistobeannotatedwithaconceptexpression
C.
Whatdoesit nowmean to annotate resourceU#X withconcept C? Table1
lists anumberof what wemight callusecases regardingthis action. For each
class described in the table, we discuss the ideas in moredetail using simple
concreteexamplestoillustratethedierences.
Notethatthedistinctionbetweenthesedierentannotationtypesintroduced
inSection3canbecomeblurred.Forexample,semanticannotation(e.g.the
as-sociationofaresourcefragmentwithamachine-processableconceptdescription
as discussedhere) may resultintheaddition ofalink iftheresourceisviewed
using the COHSE agent. The Type column of Table 1 gives an idea of the
annotationtypein termsofSection 3.
DecorationistheAnnoteaviewoftheworld,whereannotationsareseenas
commentariesonresources.Inthesimpleannotationschemeusedby Annotea,
thebodyofaresourceisachunkofHTML,whichsimplyprovidesthe(textual)
contentofthe annotation.Other approaches(such asCOHSE[1])mayextend
thisschema,however,toprovideannotationsofothertypes.
Linking(orpossiblyTransclusion,toborrowTedNelson'sterm)providesa
simple COHSEview {annotationsare simply amechanismthat provideslink
anchors.Ifthecontentofthat annotationhappensto beacomplexconceptual
descriptionthatthenenablesaclientagenttosupport\better"linking,thenall
to thegood.
Instance Identication makes a strong assertion about the resource U#
X, i.e. that it is an instance of a particular class. For example, the resource
http://www.w3.org/TR/xptrisaCandidateRecommendationoftheW3C.The
Decoration Whenthe userviewsU,theconcept Cwill
decorate theresource fragment referred to
byU#X.
link/textual
Linking When theuserviews U,links about Cwill
appear with the source anchor being the
fragmentU#X.
link
Instance IdenticationWe are making an assertion that there is
someindividualxintheworld,suchthatx
isaninstanceoftheconceptC,andtheurl
U#Xidentiesx.
kbpopulation
Instance Reference We are making an assertion that there is
someindividualxintheworld,suchthatx
isaninstanceoftheconceptC,andtheurl
U#Xinsomewayreferstox.
kbpopulation
Aboutness TheresourcefragmentU#Xis\about"C. textual
Pertinence ForanyxsuchthatxisaninstanceofC,the
informationintheresourcefragmentU#Xis
pertinenttox.
textual
Table 1.PossibleUsesofAnnotation.UisaURL,XisanXPointerexpressionandC
isaconcept
by thegivenURI. Dereferencing theURIprovidesexactlythe objectthat the
assertionisabout.
ForannotationsofclassInstanceReference,thesituationislessclear.The
resource http://www.mcfc.co.uk/player.asp?PLAYER=1191 is about Shaun
Goater the Manchester City football (soccer) player. We could annotate this
resource with the concept Footballer, but the intended interpretation here
is that there is an object in the world (Shaun Goater) that is an instance of
FootballerandwhichisreferredtoorreferencedbythegivenURIratherthan
astatementthat theURIis aninstance oftheconcept Footballer.Ahuman
readerseeingsuchanannotationwouldimplicitlyassumethattheassertionwas
beingmadeaboutthesubjectof thepage(e.g. ShaunGoater),astheideaofa
webpagebeingaFootballerisnonsensical {tomakethisinference,however,
requiresbackgroundandworldknowledge.
This distinction between Instance Identication and Instance
Refer-enceandthemechanismsthatmaybeusedtosupportthedierenceisdiscussed
in thelatersectiononIdentication.
Aboutness gives a rather loose notion of annotation. In contrast to
In-stance Identication andInstance Reference,there is noassertion ofthe
existence of a specic instance of the concept C. Instead there is a loose
as-sociation of the resource with the concept. As an example of this, the page
http://www.nczooeletrack.org/ is about Elephants. It does not discuss a
<rdf:type rdf:resource=
"http://cohse.semanticweb.or g/on tolog ies/ docs #W3C_ Cand idate _Rec omme ndati on"/ >
</rdf:Description>
Fig.6.XPointerspecisaCandidateRecommendation
Pertinencegiveswhatwemightcallakindofweakontologicalextension.It
allowsustomakeassertionsabouttheclassesandconceptswithintheontology
withoutactuallyexplicitlyenocodingorembeddingthat informationwithinthe
ontology.Ofcoursethismeansthattheinformationmaynotbereadilyavailable
toreasoningagents,butitmaybethattheinformationisnotappropriatefora
reasoner.Forexample,theEPSRC(EngineeringandPhysicalSciencesResearch
Council)websitemayhavepageswhichcontainuseful informationrelating to
Researcherssuchasemploymentopportunities,payscalesandthelike.Thisis
notnecessarilyinformationwewouldwishtomodeldirectlywithintheontology,
but is,in themain,ofinteresttoresearchers.An annotationofsucharesource
couldbeconsidered tobein thePertinenceclass.
AboutnessandPertinencecouldbeconsideredasexamplesoftextual
an-notationasintroducedabove{althoughthecontentoftheannotationmayhave
somericherstructure,theannotationisessentiallyanoteorcommentaryonthe
resource. We canconsider Decoration and Linking as enabling mechanisms
for the construction of hypertexts, in other words link annotation (although
Decoration is also akindof commentarymechanism).Incontrast, Instance
Identicationand Instance Reference areaboutknowledgebase
construc-tion, i.e. the population of an ontology or conceptual schema with instances
and do notcorrespond directly with link or textual annotation (although the
informationcouldultimatelybeusedto generatelinks).
TheOntoMattool[14]supportsannotationcorrespondingtoInstance
Iden-tication.Instancesofconceptsareintroducedandhavellersfortheir
relation-shipsharvestedfrom information appearing onthe web pagebeingannotated.
Inthecurrentversion,annotationsarenotanchoredtoparticularresource
frag-ments,but areinsteadstoredasmarkupwithin thewebpagebeingannotated.
ThenewinstanceshavegeneratedidentierswhicharebasedontheURIofthe
page being annotated. This could be extended to use external storage of the
annotations(forexampleusinganannotationserviceorRDFrepository)along
withanXPointermechanism.
5 Instance vs. Aboutness and Identication
A key questionto address when we consider annotation is that of instance-of
vs. aboutness. RDF hasa built in property rdf:type that allows us to make
theclassW3C_CandidateRecommendation.RDFiswellsetuptodealwithsuch
assertions. However,there may often be situations where we wantto makean
assertionthataparticularresourceisaboutaparticularconcept(intermsofits
content),rather thansaying itis aninstance of it. Thisrelates toannotations
ofkindInstance Referenceasdiscussedabove.
Theinstance vs. aboutness issueis closely relatedto the problem of
iden-tication of objects within the Semantic Web. In a fully-edged
implemen-tation of the SW, we would expect to be able to make assertions not only
about web resources, but also about objects, for example being able to
as-sert information about Sean Bechhofer the person, not just about the URIs
http://www.cs.man.ac.uk/~seanbormailto:[email protected]
to do this we needmechanismsthat allowus to referto objectsthat maynot
directlyhaveanexplicitlydereferenceableURI.Mechanismssuchastdb[22]or
existential quantication overDAML+OIL uniqueProperties[15] have been
proposed which allow us to refer to \the thing described by x" or \the thing
withpropertyx".
Thesemechanismswillthen allowus tosupport interpretationssuch as
In-stance Referenceasdiscussed above.Forinstance, in ourexampleof Shaun
Goater,wecannowsaythattdb:20011030:http://www.mcfc.co.uk/player.
asp?PLAYER=1191hasrdf:typeFootballer,inotherwordsthethingdescribed
bythegivenURI(i.e.ShaunGoater)isafootballer.
Note that weshould not confuse RDF's rdf:about attribute with
\about-ness"asdiscussedhere.WithinRDF,rdf:aboutisreallyasyntacticmechanism
thatrelatesaresourcetoRDFstatementsconcerningit,ratherthandescribing
thecontentofsomeresource.
TheCREAMframework[13]distinguishesdierentrolesthatcorrespondto
thetreatmentof an annotation.Quotationcopies anexcerpt from aresource
(such as thestring\ShaunGoater").Thisis aratherlooseassociation,similar
to Instance Reference asdescribed above{the copiedstring isreferring to
some object in theworld. Referenceallowsthe metadatato use apointer to
a resource fragment { the example given uses a pointer to a particular place
athttp://www.whitehouse.govinordertorefertothecurrentU.S.president.
If the actual president changes, the metadata will continue to refer to \the
president". Inthisexample,this isagainanInstance Reference astheURL
isnotthepresident,but isareferencetothepresident.
TheAnnotea[19]schema 7
containspropertieswhich linktheAnnotation to
theresourcewhichitisannotating{annotatesreferstotheenclosingURIand
contextprovidesthepreciselocation,sayusing XPointer [9].Theschemaalso
containsabodypropertywhichprovidesalink to thebodyof theannotation.
Theschema,however,remainsagnosticastotheexactsemanticsofthe
annota-tion(intermsofourclassicationabove).Allthattheannotationassertsisthat
theselected resourcehas anannotationwhichconsists of theselectedconcept.
Thereisnodirectinstance-ofassertionanditisupto theapplicationusingthe
annotationstodecideontheappropriateinterpretation.
This is weak, and we suggest that extensions to theschemaare needed in
ordertorecordandrepresentwhattheintendedsemanticsoftheannotationare.
Asanexampleofthisapproach,theCOHSEannotationsemployedanextension
of the W3C Annotea schema, with a property http://cohse.semanticweb.
org/annotation-ns#concept being used to indicate that the content of the
annotation is a concept. This property is aspecialisation of the http://www.
w3.org/2000/10/annotation-ns#bodypropertyfromtheAnnotea schema.In
addition, approaches such as tdb [15] giveus the machinery to represent the
diering interpretations of annotation, The tdb namespace provides [22] \...a
space which is useful for describing entities, concepts, abstractions, and other
items which are notthemselvesnetwork accessibleresources,but havebeen at
some point described by network accessible resources. The "tdb" namespace
designates the "thing described by" a resource at a given URI at the given
time."
Wemustensure,however,thattoolsprovideadequatesupportforusers
dur-ingtheprocessofannotation.Forexample,wemayexpecttobeoereddierent
optionscorresponding to theclass orcategoryofannotationbeingmade. This
can be seen asarequirement forexplicitness in the process. We cannot make
assumptionsaboutwhattheintendedsemanticsoftheannotationshouldbe.
6 Application Behaviour
Whatmighttheeectsofthedierentusesbeonthebehaviourofapplications?
WeuseCOHSE asanexampleofanapplication makinguseofsemantic
anno-tationsinthefollowingdiscussion,althoughthistopicisrelevanttomanyother
SWapplications.
ConsiderthedistinctionbetweenInstanceIdentication,InstanceReference
and Aboutness as introduced above. The rst states that a resource R is an
instance ofconceptC,thesecondstatesthat Rrefersto someinstance ofCand
thethird statesthattheresourceRis abouttheconcept C(whereaboutness is
targetsfor theconceptC,when Chasbeenidentied astheconceptassociated
with asourceanchor.Theclassicationofthe annotationsmayhelp the
appli-cation in organisingand presenting thelinks to the user. Forexample, rather
thansimplydisplayingalistoftargets, thetargetscouldbegroupedaccording
to whether they are deemed to be actual instances of the concept, or simply
\about" theconcept. Figure 7givesan exampleof what the popup link menu
might look like in theCOHSE applicationwith theleft hand side showingthe
barelist andtherighthand thereorganisedlist.
InourJavatutorialexample 8
,anexampleofwherethisbehaviourcouldhelp
theuserwouldbeifwearedealingwithaconceptsuchasJavaServletEngine.
TheusermaywanttondoutmoreinformationaboutJavaServletEngines(for
exampleAPIdocumentationoranoverviewofwhataServletEngineis).Inthis
case pagesdescribedas being about theconcept maybe useful.Alternatively,
theusermayactuallywanttogoandgetaJavaServletEngine,inwhichcasean
InstanceReference or Identication annotationwill be ofmore relevance. The
issue hereis verymuch concernedwith how information canbe organisedand
presentedtotheuser.
The scenario described above could, of course, hold true of any resource
discovery agent { for example this extra information could be of benet for
searchenginesinrankingandpresentinginformation.
Iftheannotationhasbeenusedtoderiveasourceanchorforalink,thismay
then aect theway that possiblelink targets are found. If the user is looking
at a resource R which has been annotated as being about some concept C, a
sensible option forthe agentwould be to presentlinks with targets which are
instances of C. Alternatively,if the resourcebeingviewed is described a being
aninstanceof C,thenitmaybemoreappropriateto displayresourcesaboutC
rst(providingmewithsomemorecontext)ratherthanother instancesofthe
the concept. Of course, such behaviour is stronglyapplication dependent, and
mayalso depend on factorssuch asuserpreferences. However,thepresence of
theextrainformation associatedwith theannotationallowsthe agentto make
moreinformedchoicesaboutthewaythat resultsarepresentedto theuser.
7 Concluding Remarks
Semantic metadata is set to play a major part in the implementation of the
SemanticWeb and annotation will be aprimary mechanismforsupplying the
metadatawhichwillthenbeusedbyagentsastheyretrieveinformation.Inthis
paperwehavepresentedanumberofdierentinterpretationsfortheprocessof
semanticannotation.Current annotationmechanisms donot support this
dis-tinction,oriftheydoitisinanimplicitratherthanexplicitfashion.Extensions
to existing annotation schemas (such as Annotea) can provide somesupport,
but mustbedoneinanagreedfashionto ensureasharedunderstanding.
notation information can be of use to applications, asis demonstrated by the
currentCOHSE system.Withoutanagreementontheunderlying assumptions
behindtheuseofsemanticannotation,however,softwareagentswithintheSW
will beunable to perform theirtasks in atrulyconsistentfashion. The
consis-tentinterpretationofnotionssuchasInstanceIdenticationandAboutnesswill
help SW applications to presentand use informationin waysthat will further
benetusers.Keytotheprovisionofworkablesemanticannotationisaneedfor
explicitness. We requireexplicitness of context to allow us to determine how
to interpret theconceptual contentof the annotations.Inaddition, we require
that the intendedsemanticsof theannotationbemade explicitin order that
agentswhichusetheannotationscanprocessandinterpretthemconsistently.
8 Acknowledgements
ThisworkwassupportedbyEPSRCGrantGR/M75426.Theauthorswouldlike
to thankBernardHoran ofSunMicrosystemsforhiscommentsonthepaper.
References
1. S.Bechhofer,I.Horrocks,C.Goble,andR.Stevens.OilEd:aReason-ableOntology
Editorforthe SemanticWeb. InProceedings of KI2001,JointGerman/Austrian
conference onArticialIntelligence,Vienna,September2001.
2. T.Berners-Lee. WeavingtheWeb. OrionBusinessBooks,1999.
3. L.Carr,S.Bechhofer,C.A.Goble,,andW.Hall.ConceptualLinking:
Ontology-based Open Hypermedia. In Proceedings of WWW10, Tenth World Wide Web
Conference,HongKong,May2001.
4. L. Carr, D. DeRoure, W.Hall, ,and G.Hill. The DistributedLinkService: A
ToolforPublishers,AuthorsandReaders.WorldWideWebJournal,1(1):647{656,
1995.
5. P.Cross,L.Miller,andS.Palmer.UsingRDFtoAnnotatethe(Semantic)Web.In
K-CapWorkshoponKnowledgeMarkupandSemanticAnnotation,Victoria,B.C.,
Canada,October2001.
6. DAML+OIL.http://www.daml.org/language.
7. S.Decker,M.Erdmann,D.Fensel,,andR.Studer. Ontobroker:OntologyBased
AccesstoDistributedandSemi-StructuredInformation. InR.Meersman,Z.Tari,
,and S.Stevens,editors, Semantic Issuesin Multimedia Systems.Proceedings of
DS-8,pages351{369.KluwerAcademicPublishers, 1999.
8. S.Decker,F.vanHarmelen,J.Broekstra,M.Erdmann,D.Fensel,M.KleinI.
Hor-rocks,,andS.Melnik.TheSemanticWeb|ontheRespectiveRolesofXMLand
RDF. IEEEInternet Computing,2000.
9. S.DeRose,E.Maler,andR.DanielJr.XMLPointerLanguage(XPointer)Version
1.0.W3CCandidateRecommendation.http://www.w3.org/TR/xptr/,September
2001.
10. C.A.Goble,S.Bechhofer,L.Carr,D.DeRoure,,andW.Hall.ConceptualOpen
sharing. InProceedingsofInternationalWorkshoponFormalOntology,1993.
12. N.Guarino. FormalOntologyandInformation Systems. InProceedingsof
FOIS-98.,1998.
13. S.HandschuhandS.Staab. AuthoringandAnnotationofWebPagesinCREAM.
In Proceedings of WWW2002, Eleventh World WideWeb Conference, Honolulu,
Hawaii,May2002.
14. S.Handschuh,S.Staab,andA.Maedche.CREAM-Creatingrelationalmetadata
with a component-based, ontology-drivenannotation framework. InProceedings
of K-CAP2001, First InternationalConferenceon KnowledgeCapture,Victoria,
B.C.Canada,October2001.
15. S.Hawke.HowWeIdentifyThings(ontheSemanticWeb)?http://www.w3.org/
2001/03/identification-problem /.
16. J.Hein,J.Hendler,,andS.Luke.SHOE:AKnowledgeRepresentationLanguage
for Internet Applications. Technical Report CS-TR-4078(UMIACS TR-99-71),
DepartmentofComputerScience,UniversityofMaryland,1999.
17. J.Kahan,M.-R.Koivunen,E.Prud'Hommeaux,andR.Swick.Annotea:AnOpen
RDF Infrastructure for Shared Web Annotations. In Proceedings of the Tenth
InternationalWorldWideWebConference,HongKong,May2001.
18. M.-R. Koivunen,D.Brickley,J.Kahan,E. Prud'Hommeaux,,and R.R. Swick.
The W3C Collaborative Web Annotation Project ... or how to have fun while
building an RDF infrastructure. http://www.w3.org/2000/02/coll abor ation /
annotation/papers/annotationinf %ras truc ture, May2000.
19. M.-R. Koivunen and Ralph Swick. Metadata Based Annotation Infrastructure
oersFlexibilityandExtensibilityforCollaborativeApplicationsandBeyond..In
K-CapWorkshoponKnowledgeMarkupandSemanticAnnotation,Victoria,B.C.,
Canada,October2001.
20. M. Margolis and D. Resnick. ThirdVoice:Vox PopuliVox Dei? First Monday,
4(10),1999. http://www.firstmonday.dk/i ssue s/iss ue4_ 10/m argol is/.
21. C.C. Marshall. Towards an ecology of hypertextannotation. In Proceedings of
HT98,ACMConferenceonHypertext, pages40{49,PittsburghPA,USA,1998.
22. L. Masinter. \duri" and \tdb": URN Namespaces basedon datedURIs. IETF
Internet-Draft,http://larry.masinter.net/duri. html , April2002.
23. D.L.McGuinness. Ontologicalissuesforknowledge-enhancedsearch. In
Proceed-ingsof FOIS-98,1998.
24. SWISS-PROTAnnotatedProteinSequenceDatabase. http://www.expasy.org.
25. The Gene Ontology Consortium. Gene Ontology: a tool for the unication of
biology. NatureGenetics, 25:25{29, 2000.
26. M. UscholdandM.Gruninger. Ontologies:Principles,methodsandapplications.
KnowledgeEngineeringReview,11(2):93{136, 1996.
27. F. van Harmelen, P.F. Patel-Schneider, and I. Horrocks. A
Model-Theoretic Semantics for DAML+OIL. http://www.daml.org/2001/03/
model-theoretic-semantics.htm l.
28. G.vanHeijst, A.Schreiber,,and B.Wielinga. UsingexplicitontologiesinKBS
development.InternationalJournalofHuman-ComputerStudies,46(2/3):183{292,
1997.
29. World Wide Web Consortium. Annotea Project. http://www.w3.org/2001/