Here - arcoe

26 Pages·2012·1.75 MB·English

Workshop Notes
The ECAI-10 Workshop on
Automated Reasoning about Context and Ontology Evolution
ARCOE-10
August 16-17, 2010
Lisbon, Portugal
held at the
European Conference on Artificial Intelligence
Co-Chairs
Alan Bundy
Jos Lehmann
Guilin Qi
Ivan José Varzinczak
http://www.arcoe.org

Introduction to the Notes of the ECAI-10 Workshop ARCOE-10
Automated Reasoning about Context and Ontology Evolution
Alan Bundya, Jos Lehmanna, Guilin Qib, Ivan Jos´e Varzinczakc
a School of Informatics, University of Edinburgh
b School of Computer Science and Engineering, Southeast University
c Meraka Institute, Pretoria, South Africa
Methods of automated reasoning have solved a large number of problems in Computer Science by using formal
ontologies expressed in logic. Over the years, though, each problem or class of problems has required a different on-
tology, andsometimesadifferentversionoflogic. Moreover, theprocessesofconceiving, controllingandmaintaining
an ontology and its versions have turned out to be inherently complex. All this has motivated much investigation
in a wide range of disparate disciplines - from logic-based Knowledge Representation and Reasoning to Software
Engineering, from Databases to Multimedia - about how to relate ontologies to one another.
Just like the previous edition, ARCOE-10 aims at bringing together researchers and practitioners from core areas of
Artificial Intelligence (Knowledge Representation and Reasoning, Contexts, and Ontologies) to discuss these kinds
of problems and relevant results.
Historically, there have been at least three different, yet interdependent motivations behind this type of research:
defining the relationship between an ontology and its context, providing support to ontology engineers, enhancing
problem solving and communication for software agents.
ARCOECallforAbstractshasbeenformulatedagainstsuchhistoricalbackground. SubmissionstoARCOE-10have
been reviewed by two to three Chairs or PC members and ranked on relevance and quality. Approximately eighty
percent of the submissions have been selected for presentation at the workshop and for inclusion in these Workshop
Notes.
Thanks to the invaluable and much appreciated contributions of its Program Committee, its Invited Speakers
and its authors, ARCOE-10 provides participants with an opportunity to position various approaches with respect
to one another. Hopefully, though the workshop and these Notes will also start a process of cross-pollination and
set out the constitution of a truly interdisciplinary research-community dedicated to automated reasoning about
contexts and ontology evolution.
(Edinburgh, Nanjing, Pretoria – June 2010)
ARCOE-10 Co-Chairs
Alan Bundy (University of Edinburgh, UK)
Jos Lehmann (University of Edinburgh, UK)
Guilin Qi (Southeast University, China)
Ivan Jos´e Varzinczak (Meraka Institute, South Africa)
ARCOE-10 Invited Speakers
Tommie Meyer (Meraka Institute, South Africa)
Fausto Giunchiglia (University of Trento, Italy)
ARCOE-10 Program Committee
Grigoris Antoniou (FORTH, Greece)
Franz Baader (TU Dresden, Germany)
Richard Booth (Ma´ha˘a W´ıtta´yaalai Mahasarakham, Thailand)
Paolo Bouquet (Universit`a di Trento, Italy)
Jerome Euzenat (INRIA Grenoble Rhone-Alpes, France)
Giorgos Flouris (FORTH, Greece)
Chiara Ghidini (FBK Fondazione Bruno Kessler, Italy)
Deborah McGuinness (Rensselaer Polytechnic Institute, USA)
Thomas Meyer (Meraka Institute, South Africa)
Maurice Pagnucco (The University of New South Wales, Australia)
Valeria de Paiva (Palo Alto Research Center (PARC), USA)
Jeff Pan (University of Aberdeen, UK)
Dimitris Plexousakis FORTH, Greece)
Luciano Serafini (FBK Fondazione Bruno Kessler, Italy)
Renata Wassermann (Universidade de S˜ao Paulo, Brazil)
Table of Contents

Marcio Ribeiro and Renata Wassermann
More about AGM Revision in Description Logic……………………….…………………….…....….7-8

Richard Wallace and Tabbasum Naz
Context and Intention in Ontologies………….……………………………………………….……...9-10

Christoph Benzmuller and Adam Pease
Reasoning with Embedded Formulas and Modalities in SUMO…..……………….……………….11-12

Hai Nguyen, Natasha Alechina and Brian Logan
Ontology Debugging with Truth Maintenance Systems……………………………………...……..13-14

Jos Lehmann, Alan Bundy and Michael Chan
Qualitative Causal Analysis of Empirical Knowledge for Ontology Evolution in Physi………...…15-16

Claudia d'Amato and Nicola Fanizzi
Uncertainty Reasoning through Similarity in Context……………………………………..……….17-18

Adam Pease, Chris Benzmueller
Ontology Archaeology: Mining a Decade of Effort on the Suggested Upper Merged Ontology...…19-20

Thomas Meyer, Kodylan Moodley and Ivan Varzinczak
First Steps in the Computation of Root Justifications…………………………….…………..…….21-22

Michael Chan, Jos Lehmann and Alan Bundy
A Contextual Approach to Detection of Conflicting Ontologies……………………….……..…….23-24

ARCOE-10 Workshop Notes
More About AGM Revision in Description Logics (cid:14)
Ma´rcioMorettoRibeiroandRenataWassermann
UniversityofSa˜oPaulo
{marciomr,renata}@ime.usp.br
1 Introduction (uniformity) If for all K(cid:48) ⊆ K, K(cid:48) ∪ {α} is inconsistent iff
K(cid:48)∪{β}isinconsistentthenK∩K∗α=K∩K∗β
Beliefrevisionstudiesthedynamicsofbeliefsdefiningsomeoper-
ationsinlogicallyclosedsets(beliefsets):expansion,revisionand The set of rationality postulates we considered is: closure, suc-
contraction.Revision,inparticulardealswiththeproblemofaccom- cess,inclusion,consistency,relevanceanduniformity.Thefollowing
modatingconsistentlyanewlyreceivedpieceofinformation. proposition is an evidence that this is a good choice of rationality
Mostoftheworksonbeliefrevisionfollowingtheseminalpaper postulates:
[1]assumethattheunderlyinglogicoftheagentsatisfiessomeas-
Proposition3 ForlogicsthatsatisfytheAGMassumptions,closure,
sumptions.In[5]weshowedhowtoapplyrevisionofbeliefsetsto
success,inclusion,consistency,relevanceanduniformityareequiv-
logicsthatarenotclosedundernegation.Wehave,however,assumed
alenttotheoriginalAGMpostulatesforrevision:closure,success,
thatthelogicsatisfiesapropertycalleddistributivity.Inthepresent
consistency,vacuityandextentionality.
workweshowalistofdescriptionlogicsthatarenotclosedunder
negationandstudywhichofthemaredistributive.
Weproposedalsoaconstructioninspiredinsomeideasfrom[4]:
1.1 AGMparadigm Definition4(Maximallyconsistentsetw.r.tα) [4]X ∈K ↓αiff
X ⊆K,X∪{α}isconsistentandifX ⊂X(cid:48) ⊆KthenX(cid:48)∪{α}
Themostinfluentialworkinbeliefrevisionis[1].Inthisworkthe isinconsistent.
authors defined a number of rationality postulates for contraction
andrevision,nowknownastheAGMpostulates.Theauthorsthen Definition5(Selectionfunction) [1]AselectionfunctionforK is
showedconstructionsfortheseoperationsandprovedthatthecon- afunctionγsuchthatifK ↓α(cid:54)=∅,then∅(cid:54)=γ(K ↓α)⊆K ↓α.
structionsareequivalenttothepostulates(representationtheorem) Otherwise,γ(K ↓α)={K}.
Mostworksinbeliefrevisionassumesomepropertiesontheun-
derlyinglogic:compactness,Tarskianicity,deductionandsupraclas- TheconstructionofarevisionwithoutnegationisdefinedasK∗γ
(cid:84)
sicality, which we will refer to as the AGM assumptions. The last α= γ(K ↓α)+α.
twotogetherareequivalenttothefollowingtwopropertiestogether We proved that, for distributive logics, this construction is com-
forTarskianlogics: pletelycharacterizedbythesetofrationalitypostulateswearecon-
sideringi.e.weprovedtherepresentationtheoremrelatingthecon-
Definition1(distributivity) Alogic(cid:104)L,Cn(cid:105) isdistributiveifffor structiontothesetofpostulates[5].
allsetsofformulasX,Y,W ∈2L,wehavethatCn(X∪(Cn(Y)∩
Cn(W)))=Cn(X∪Y)∩Cn(X∪W).
1.2 DescriptionLogics
Definition2(closureundernegation) Alogic(cid:104)L,Cn(cid:105) isclosed Descriptionlogics(DLs)formsafamilyofformalismstorepresent
under negation iff for all A ∈ 2L there is a B ∈ 2L such that terminologicalknowledge.Thesignatureofadescriptionlogicisa
Cn(A∪B) = L andCn(A)∩Cn(B) = Cn(∅).ThesetB is tuple(cid:104)N ,N ,N (cid:105) ofconceptnames,rolesnamesandindividual
C R I
thencalledanegationofA. namesofthelanguage[2].Fromasignatureitispossibletodefine
complexconceptsviaadescriptionlanguage.EachDLhasitsown
AGMrevisioninnon-classicallogics: In[5]wearguedthatsome descriptionlanguagethatadmitsacertainsetofconstructors.
The semantic of a DL is defined using an interpretation I =
descriptionlogicsarenotclosedundernegationand,hence,donot
(cid:104).I,∆I(cid:105) such that ∆I is a non-empty set called domain and .I is
satisfytheAGMassumptions.Furthermore,themostcommonway
todefinerevisionisviaLeviidentity(K∗α=K−¬α+α),which aninterpretationfunction.Foreachconceptnametheinterpretation
assumestheexistenceofthenegationofα.Weproposedthenanew associatesasubsetofthedomain,foreachrolenameabinaryrela-
tioninthedomainandforeachindividualanelementofthedomain.
constructionandasetofpostulatesforrevisionforlogicsthatarenot
Theinterpretationisthenextendedtocomplexconcepts.
closedundernegation.
AsentenceinaDLisarestrictiontotheinterpretation.ATBoxisa
Weusedtwopostulates,borrowedfromthebeliefbaseliterature:
setofsentencesoftheformC (cid:118)C thatrestrictstheinterpretation
1 2
(relevance) If β ∈ K \K ∗α then there is K(cid:48) such that K ∩ ofconcepts1,anABoxisasetofsentencesoftheformC(a),R(a,b),
(K∗α)⊆K(cid:48) ⊆KandK(cid:48)∪{α}isconsistent,butK(cid:48)∪{α,β}is
inconsistent. 1AssumingthatthelogicadmitsGCIaxioms
(cid:14)(cid:4)8LMW(cid:4)[SVO(cid:4)[EW(cid:4)WTSRWSVIH(cid:4)F](cid:4)*%4)74(cid:4)ERH(cid:4)’24U
7
ARCOE-10 Workshop Notes
a=banda(cid:54)=bthatrestrictstheinterpretationofindividuals.Some TheexampleabovedependsontheexistenceoftheABox.Infact,
DLs,likeALCH,admitsalsoanRBoxwhichisasetofsentencesof ALCwithemptyABoxisdistributive:
theformR(cid:118)Sthatrestrictstheinterpretationsofroles.
LetΣ=(cid:104)T,A,R(cid:105)beatuplewhereT,AandRareaTbox,an Proposition11 Consider a DL (cid:104)L,Cn(cid:105) such that for every sen-
ABoxandanRBoxrespectively.Asentenceαisaconsequenceof tence α ∈ L there is a sentence α(cid:48) ∈ L such that Cn(α) =
Σ(Σ(cid:15)αorα∈Cn(Σ))iffforallinterpretationsIifIsatisfiesΣ Cn(α(cid:48)) and α(cid:48) has the form (cid:62) (cid:118) C for some concept C. Then
thenIsatisfiesα. (cid:104)L,Cn(cid:105)isdistributive.
TwocharacteristicstheDLsweareconsideringthatwillbeimpor-
SinceinALCO theABoxcanbewrittenintermsoftheTBox,
tantinthisworkare:inALCeverysentenceintheTBoxisequivalent
ALCOisdistributiveeveninthepresenceoftheABox.
toasentenceoftheform(cid:62)(cid:118)C foraconceptC [3]andinALCO
Finally,ifweconsideralogic(cid:104)L,Cn(cid:105) thatadmitsrolehierarchy,
everysentenceintheABoxisequivalenttoasentenceoftheform
butdoesnotadmitroleconstructors,then(cid:104)L,Cn(cid:105) isnotdistribu-
(cid:62)(cid:118)C.
tive.Considerthefollowingexample:
2 PropertiesofDescriptionLogics Example12: LetX = {R (cid:118) S ,R (cid:118) S },Y = {S (cid:118)
1 2 1
S }andZ = {S (cid:118) S }.WehavethatCn(Y)∩Cn(Z) =
3 2 3
The main contribution of this work is to show a set of description Cn(∅).HenceR (cid:118) S ∈/ Cn(X ∪(Cn(Y)∩Cn(Z))),but
3
logicsthatarenotclosedundernegationandwhichofthemaredis- R(cid:118)S ∈Cn(X∪Y)∩Cn(X∪Z).
3
tributive i.e. we show a set of logics such that representation theo-
remforrevisionwithoutnegationisapplicable.Itturnsoutthatmost
BesidesALCH,thelogicsbehindOWL1(SHOIN forOWL-
DLsthatadmitsGCIaxioms(GCIaxiomsallowcomplexconcepts
DL and SHIF for OWL-lite), OWL-2 (SROIQ) and the OWL
inbothsidesofthesentence)arenotclosedundernegation,butmany
profilesOWL-RLandOWL-QLadmitrolehierarchy,butdonotad-
ofthemarealsonotdistributive.
mitroleconstructors.Noneoftheselogicsaredistributive.
Thefollowingtablesumsuptheresultsofthissection:
Classic negation in DLs: We will say that two roles R and S
are unrelated iff neither R (cid:118) S ∈ Cn(∅) nor S (cid:118) R ∈ Cn(∅). DescriptionLogic Negation Distributivity
ThemainresultofthissectionprovesthatifthesignatureofaDL ALC no no
(cid:104)L,Cn(cid:105) hasinfinitelymanyunrelatedrolesandadmits∀,(cid:116),¬and ALCwithoutABox no yes
GCIaxiomsthen(cid:104)L,Cn(cid:105) isnotclosedundernegation. ALCO no yes
ALCH,OWL-lite,OWL-DL no no
OWL-QL,OWL-RLandOWL2 ? no
Theorem6 ConsideraDL(cid:104)L,Cn(cid:105) thatadmitstheconstructors
¬,∀,(cid:116)andgeneralconceptinclusionaxiomsintheTBox.Ifthere
isaninfinitenumberofunrelatedroles,then(cid:104)L,Cn(cid:105) isnotclosed
3 Conclusionandfuturework
undernegation
In this work we continued the work started in [5] by showing for
Theproofofthistheoremcomesfromthefactthatif(cid:104)L,Cn(cid:105) ad- whichDLstheAGMrevisionwithoutnegationcanbeapplied.We
mits(cid:116),¬theneverysentencecanbewrittenas(cid:62)(cid:118)Candfromthe showedthatmostDLsthatadmitsGCIsarenotclosedundernega-
followinglemmas: tion,butmostofthemarealsonotdistributive.WeshowedthatALC
withemptyTBoxandALCO aretwoexceptions.Theselogicsare
Lemma7 LetAandB beconceptssuchthat(cid:62) (cid:118) Aand(cid:62) (cid:118) B
distributive and not closed under negation. Hence, the representa-
arenottautologiesandletRbearolenamethatisunrelatedwithany
tiontheorempresentedin[5]holdsforALC withemptyABoxand
rolethatappearsinAorB.ThenCn(∅)⊂Cn((cid:62)(cid:118)A(cid:117)∀R.B)⊆
ALCO.
Cn((cid:62)(cid:118)A)∩Cn((cid:62)(cid:118)B).
Inadditiontothat,weshowedthatthepostulatesusedin[5]are
equivalenttotheAGMpostulatesiftheunderlyinglogicsatisfiesthe
Lemma8 IfCn((cid:62) (cid:118) A) = Cn(∅)and(cid:62) (cid:118) B isanegationof
AGMassumptions.Thisisagoodevidencethatwechoseagoodset
(cid:62)(cid:118)AthenCn((cid:62)(cid:118)B)=L
ofrationalitypostulates.
Asfutureworkweshouldlookforaconstructionthatcanbechar-
As a corollary of this result we have that many well known de-
acterizedbythissetofpostulates(orasimilarone)notonlyindis-
scription logics are not closed under negation. Hence, for all these
tributive,butinanyTarskiancompactlogic.
logicstheAGMresultsarenotapplicable:
Corollary9 The following DLs are not closed under negation: REFERENCES
ALC,ALCO,ALCH,OWL-liteandOWL-DL.
[1] C.Alchourro´n,P.Ga¨rdenfors,andD.Makinson,‘Onthelogicoftheory
change’,JournalofSymbolicLogic,50(2),510–530,(1985).
[2] The Description Logic Handbook, eds., F. Baader, D. Calvanese,
DistributivityinDLs: Inthissectionweshowalistofdistributive
D.McGuinness,D.Nardi,andP.Patel-Schneider,CambridgeUniver-
andnon-distributiveDLs.Westartwithanexampleshowingthatthe
sityPress,2003.
logicALCisnotdistributiveingeneral. [3] F.BaaderandU.Sattler,‘Anoverviewoftableaualgorithmsfordescrip-
tionlogics’,StudiaLogica,69(1),5–40,(2001).
Example 10: Let X = {a = b}, Y = {C(a)} and [4] J.P.Delgrande,‘Hornclausebeliefchange:Contractionfunctions’,in
ProceedingsofKR,eds.,G.BrewkaandJ.Lang,pp.156–165,(2008).
Z = {C(b)}, then Cn(Y) ∩ Cn(Z) = Cn(∅). Hence
[5] M.M.RibeiroandR.Wassermann,‘AGMrevisionindescriptionlog-
C(a) ∈/ Cn(X∪(Cn(Y)∩Cn(Z))),butC(a) ∈ Cn(X∪ ics’,inProceedingsofARCOE,(2009).
Y)∩Cn(X∪Z).
8
ARCOE-10 Workshop Notes
Context and Intention in Ontologies
RichardJ.Wallace and TabbasumNaz1
1 EXTENDEDABSTRACT
1.1 (Meta-)Context
Ontologies are only useful within a given context. Sometimes this
contextisquitespecific;sometimesitisbroad-basedorgeneric.
In many cases, the context of an ontology can be described as
theintentionofthatontology.Infact,mostpracticalontologiesare
usedforaspecificpurpose,andthisintentionalityisusuallyreflected
throughouttheirorganisation.Thisincludestheconceptsdefined,the
division of superordinate concepts into subordinate categories, and
thepropertiesthatarespecified.
However,moreoftenthannotthispurposeremainsimplicit.Itmay
notevenbeexpressedinanyoftheconceptsintheontology.More
significantly,therelevanceofagivenconceptwithrespecttothepur-
poseoftheontologyisneverspecifiedinaclear,unambiguousfash-
ion.Instead,conceptsareapparentlyincluded(orexcluded)onthe
basisofintuitionandtrial-and-error.
Moregenerally,theimplicationsofintentionalityforontologyor- Figure1. Portionsoftwoontologiesdevelopedindependentlyforthe
ganisationarenotclearandhaveneverbeenspelledout. traveldomain.
1.2 Anexample <Event>,and<ContactData>areassociatedwithspecificrolesin
theactivitythattheontologyismeanttosupport.
The first example is taken from the travel domain. Portions of
Ingeneral,whenperusingtheseontologies,onehasthesenseofan
two independently created ontologies are shown in Figure 1. The
overalllackofcoherence,althoughatpresentitisdifficulttospecify
”Tourism”ontology(leftpanelinfigure)wasdevelopedforSeman-
whatformthiscoherenceshouldtake.This,infact,istheproblem
ticWebsitesrelatedtotourism[4].The”e-tourism”ontology,also
thatthepresentresearchwillattempttoaddress.
knownas”OnTour”(right-handpanel)ispartofaWebassistantto
aiduserssearchingforvacationpackages[7].Despitethesimilarity
1.4 Existingandotherpossibleapproaches
ofdomainandintention,theontologiesarestrikinglydifferent.The
concepthierarchiesarequitedissimiliar,asarethepropertiesdefined (cid:15) Meta-ontologyofintentions:givenontologyasanindividualbe-
(notshowninfigure). longingtosomeclassofintentions.
Example: In the Ontology Metadata Vocabulary (OMV)[5],
1.3 Issuesraisedbythisexample the concept <OntologyTask> contains information
about the task the ontology was intended to be used
The tourism example shows how extensive differences in ontology
for. <OntologyTask> has further pre-defined tasks i.e.
organisationcanbewhentheintentionis(apparently)thesame.
<AnnotationTask>, <MatchingTask>, <IntegrationTask>,
Itshouldalsobenotedthatinneithercaseisthepurpose,orinten-
<QueryFormulationTask>,etc.
tion,oftheontologyexplicitlyrepresented.Thus,thereisnoconcept
(Inourview,thereissomethingawkwardandineptaboutthisap-
of a trip or of planning a trip. There is also an issue of the level
proach,asifthemachineryofontologieswasbeingtakenoffthe
orgeneralityofaconcept.Forexamplethee-tourismontologyhas
shelfandputtouseinarote,unthinkingfashion.)
<Ticket>,<Location>,<DateTime>,and<ContactData>atthe
(cid:15) Intentionality might be specified by means of top-level con-
same level in the ontology, but these concepts differ greatly in ab-
cepts.Thismightinvolveadatabase-viewapproach,basedonthe
straction.
class/subclassrelationsinthefullontology.
Moreimportantly,theseconceptshaveverydifferentrelationsto
the activity of aiding tourists or searching for vacation packages.
Some, like <Location> and <DateTime>, are general concepts 1.5 Presentapproach
associated with basic ‘stage setting’, while others like <Ticket>,
Webeginwiththequestion:Isintentionawell-definedconcept?So
1CorkConstraintComputationCentre,UniversityCollegeCork,Cork,Ire- thatdealingwithitisawell-definedproblem?Infact,thereisaliter-
land,email:[email protected],[email protected] atureofsomeproportionsinphilosophythatdealswiththisquestion.
9
ARCOE-10 Workshop Notes
Inparticular,theworkofBratman[1]includesanextensivediscus- (cid:15) Characterisinganontologywithrespecttointentions.Here,apos-
sionofthistopic.Anditishisdefinitionthatwewilluse.Therefore, sibleapproachistocreateapartialorderbasedonintentions,i.e.
weconsidertheideaofintentionasboundupwiththecreationofa latticestructure,inordertolocateanontologywithinalatticeof
plan. ontologies.Thesupremummightbealltheintentionsforthatlat-
Thus,weapproachontologyconstructionasifwewerebuilding tice.
aplan.Thismeansthatwemuststartbydefiningthegoal.Thegoal (cid:15) Historical aspects of ontologies, i.e. developing/emerging inten-
willbeoneoftheconceptsintheontology. tions.
It seems most natural to use the HTN style of planning [9], in (cid:15) Intentionsandagents(esp.BDIagents).Explicitintentionsmay
whichamajoractionisplandecomposition.Anexampleisshown enhancetheaccessibilityofanontologytosoftwareagents.
below for the tourism domain. Here, the goal is to support trip-
planning.Thisgoalisthendecomposedintosubgoals;alternatively,
1.7 UnrelatedWork.
wecanthinkofabasicactiontrip-planning,decomposedintocom-
ponentactions.(Notethat<trip>isnotincludedineitheroftheon- Therearemanypapersunrelatedtothepresentwork.Here,ourcon-
tologiescitedabove,althoughitisakeyconceptinthisapplication cerniswiththosecaseswherethismightnotberecognised.Anob-
domain.) vious example is the development of ontologies that are meant to
beusedinconnectionwithplanning.Thistopichasreceivedafair
PlanTrip amountofattentionduringthepastfewyears.Infact,arecentICAPS
(cid:8)H workshopwasdevotedtothistopic[2].Morerecently,Jinggeetal.
(cid:8) H
(cid:8) H havediscussedhowtocombineontologieswithHTNplanning[3].
(cid:8) H
(cid:8)(cid:8) HH Itshouldbeobviousthatthisworkhaslittleornothingincommon
(cid:8) H
(cid:8) H withthepresentwork,whosepurposeistoguideontologybuilding
(cid:8) H
(cid:8)(cid:8) H ratherthantoprovideaplannerwithdomainknowledge.
Asecondtypeofunrelatedworkisconcernedwithdevelopingon-
BookFlight FindAccomodation ChooseActivity
tologiestocharacterisethegeneralplanningprocess.Here,theonly
examplethatweareawareofisapaperbyRajpathakandMotta[6].
Although this seems less distantly related to our concerns than the
workscitedabove,itshouldbeclearthatthepurposeofthisontol-
Figure2.Toplevelofplanfortourismontology.
ogy,whichistoaidplanningatagenericlevel,isnotwhatwehave
describedinthispaper.Nonetheless,althoughtheintentioninquite
differentinthetwocases,anontologyofplansmaybeusefulinthe
Ineachcase,weassumeamappingbetweenanactionandacon-
presentcontext.
cept.Wealsomapinthesamewaybetweenpreconditionsaswellas
fromeffectstoconcepts.Thisisbasicallyhowwebuildourontology.
This may also give us a way of evaluating the completeness of ACKNOWLEDGEMENTS
theontology.Thatis,anontologyiscompleteonlyifitisassociated
ThisworkwassupportedbyScienceFoundationIrelandunderGrant
withasetofactions,etc.thatcanaccomplishthebasicgoal,which
05/IN/I886. and by the Irish Marine Institute, Project Ref. No.
representstheintentionoftheontology.
isPBA/KI/07/001.
Manydetailsremaintobeworkedout,e.g.howtoguidetheuserin
theplanningprocess,howtoupdateandrevisethegrowingontology,
therelation,ifany,oftheplanstructuretothestructureoftheontol- REFERENCES
ogy, and how to set up properties appropriately (the precondition-
[1] M.E.Bratman,Intention,Plans,andPracticalReason,Harvard,1987.
action-effect relations may help to guide this aspect of ontology [2] ICAPS05,ICAPS2005WorkshopontheRoleofOntologiesinAIPlan-
building).Wemustalsoexplorethevariousformsofabstractionin ningandScheduling,2005.
planning[8]toseeiftheyhaveanybearinginthiscontext. [3] S.Jingge,C.Jianzhong,andL.Yiping,‘Combiningontologyengineer-
ingwithhtnplanning’,inProc.2006ConferenceonLeadingtheWebin
Aspartoftheprocessofelaborating,andchecking,ourideas,we
ConcurrentEngineering:NextGenerationConcurrentEngineering,pp.
arebuildingasystemforontologyplanning,tentativelycalledOn-
214–221.IOS,(2006).
toPlanner. Currently, the basic scheme of operation that we envis- [4] T.Kalinka,Tourism.owl,http://protegewiki.stanford.edu/index.php/Pro-
ageisto(i)parsethepredicatesusedinplanconstructiontoobtain tegeOntologyLibrary#OWLontologies.
nouns and verbs which indicate relevant concepts, (ii) characterise [5] R.Palma,J.Hartmann,andP.Haase,OMV–OntologyMetadataVocab-
ularyfortheSemanticWeb,TechnicalReportVersion2.4.1,2009.
or‘locate’theconceptsviaatop-levelontology,(iii)introducethese
[6] D.RajpathakandE.Motta,‘Anontologicalformalizationoftheplanning
conceptsandpossiblyrelatedconceptsandpropertiesintothedevel- task’,inProc.ThirdInternationalConferenceonFormalOntologyin
opingontology. InformationSystems-FOIS2004,eds.,A.VarziandL.Vieu,pp.305–
316.IOS,(2004).
[7] K. Siorpass, K. Prantner, and D. Bachlechner, e-tourism On-
1.6 Furtherissues
tology, Digital Enterprise Research Institute,Galway, http://e-
tourism.deri.at/ont/index.html,2005.
(cid:15) Mostontologiesincorporatemorethanoneintention-usuallynec-
[8] J.D.Tenenberg,‘Abstractioninplanning’,inReasoningAboutPlans,
essary(justasanautomobileoramobilephoneincorporatesmore eds.,J.F.Allen,H.A.Kautz,R.N.Pelavin,andJ.D.Tenenberg,213–
thanoneintention). 284,MorganKaufmann,(1991).
(cid:15) Intentionality has implications for merging ontologies, again [9] D.Wilkins,PracticalPlanning,MorganKaufmann,1988.
which have not been worked out. Merging and matching might
well be facilitated if the intentionality of each ontology had an
explicitrepresentation.
10