Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
0Activity
0 of .
Results for:
No results containing your search query
P. 1
null

null

Ratings: (0)|Views: 3 |Likes:
Published by Pearltrees2

More info:

Published by: Pearltrees2 on Mar 02, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

03/02/2011

pdf

text

original

 
OntheMeta-LogicofArguments
MichaelWooldridge*
*
DeptofComputerSciencePeterMcBurney*SimonParsons'tDeptofComputerandInformationScienceUniversityofLiverpoolLiverpoolL697ZF,U.K.BrooklynCollege,CUNYBrooklyn,11210NY,USA
ABSTRACT
1.
INTRODIJCTION
Argumentationhasreceivedsteadilyincreasingattentioninthemulti-agentsystemscommunityoverthepastdecade,withparticularinterestintheuseofargumentmodelsfromtheinformallogiccommunity.The
formalisation
ofsuchar-gumentsystemsisanecessarystepiftheyaretobesuccess-fullydeployed,andtheirpropertiesrigorouslyunderstood.However,thereisasyetnowidelyacceptedapproachtotheformalisationofargumentsystems.Inthispaper,wetakeasourstartingpointtheviewthatargumentsanddialoguesareinherently
meta-logical,
andthatanyproperformalisa-tionofargumentmustembracethisaspectoftheirnature.Forexample,astatementthatservesasajustificationofanargumentisisstatement
about
anargument:theargumentforwhichthejustificationservesmustitselfbereferredtointhejustification.Fromthisstartingposition,wedevelopaformalisationofargumentsusingahierarchicalfirst-ordermeta-logic,inwhichstatementsinsuccessivelyhighertiersoftheargumentationhierarchyrefertostatementsfurtherdownthehierarchy.Thisenablesustogiveacleanfor-malseparationbetweenobject-levelstatements,argumentsmadeabouttheseobjectlevelstatements,andstatementsaboutarguments.Argumentationhasreceivedsteadilyincreasingattentioninthemulti-agentsystemscommunityoverthepastdecade,withparticularinterestintheuseofargumentmodelsfromtheinformallogiccommunitysuchasthatofWaltonandKrabbe[19,24].The
formalisation
ofsuchargumentsys-temsisanecessarystepiftheyaretobesuccessfullyde-ployed,andtheirpropertiesrigorouslyunderstood.Mostargumentsystemscanbeclassifiedaccordingtowhethertheargumentstheyconsiderarestructured,typicallylogi-calentities(e.g.,[2,11,12,13]),oratomic,abstractentities(inthesenseofDung'sabstractargumentmodel[7,1]).However,althoughsomeresearchhasconsideredthelinksbetweenthesedifferenttypesofsystems[3],noonemodelisuniversallyaccepted,andboththeabstractandlogicalargumentationparadigmshavewell-knownproblemsasamodelofrationalargument[18].Inthispaper,wefocusonalogic-basedviewofargu-ments[13].Wetakeasourstartingpointtheviewthatar-gumentsanddialoguesareinherently
meta-logical
processes.Bythis,wemeanthattheargumentsmadebyprotagonistsinadebatemust
refer
toeachother.Thisisbecausear-gumentsarenotjustaboutwhichstatesofaffairsexistin
CategoriesandSUbjectDescriptors
1.2.11[Distributedartificialintelligence]:multiagentsystems;1.2.4[KnowledgeRepresentationFormalismsandMethods]:predicatelogic
GeneralTerms
theory,languages
Keywords
theworld,orhowobjectsintheworldstandinrelationtoone-another.
If
thiswerethecase,thendialogueswouldbeimpoverishedindeed,essentiallyrestrictedtoassertingthetruthorfalsityofstatements.Webelievethatrationalar-gumentationalsoinvolvesputtingforward
argumentsaboutarguments,
anditisinthissensethattheyare
meta-logical.
Forexample,astatementthatservesasajustificationofanargumentisastatement
about
anargument:theargumentforwhichthejustificationservesmustitselfbe
referredto
inthejustification.Oneofourmainaimsinthispaperistoputthisideaofmeta-argumentonthemapofargumentationresearch.Butmulti-agentsystems,argumentation,knowledgerepresenta-tion,meta-levelreasoning
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforprofitorcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthefirstpage.Tocopyotherwise.torepublish,topostonserversortoredistributetolists,requirespriorspecificpermissionand/orafee.
AAMAS'05,
July25-29,2005,Utrecht,Netherlands.Copyright2005ACM1-59593-094-9/05/0007...$5.00.
wealsohopetoshowhowameta-logicaltreatmentofargu-mentcanclarifysomeapparentlydifficultissuesinthefor-malisationofargument.Ourbasicapproachinvolvesdevel-opinga
hierarchical
formalisationoflogic-basedarguments.Thatis,weconstructa(well-founded)tower
60,61,...
ofarguments,wherearguments,statements,andpositionsatalevel
n
inthehierarchymayrefertoargumentsandstate-mentsatlevelsm,for0::;m
<
n.
Inthebottomtier
60
ofthehierarchyare
objectlevel
statementsaboutthedomainofdiscourse.Theapparatusweuseforformalisingsuchanar-gumentsystemisa
hierarchicalfirst-ordermeta-logic,
atypeoffirst-orderlogicinwhichindividualtermsinthelogiccanrefertotermsinanotherlanguage
(cf.
Konolige'sfirst-order
 
formalisationofknowledgeandaction[10]).Thisformali-paper,wesetoutandworkwitha3-tierhierarchy,asillus-tratedinFigure1.Throughouttheremainderofthepaper,sationenablesustogiveacleanformalseparationbetweenobject-levelstatements,argumentsmadeabouttheseobjectlevelstatements,andstatementsaboutarguments.wewilldenotetheselevelsofthehierarchyby
~o,
~1,
etc.,with
~o
alwaysbeingthe
lowest
levelofthehierarchy.TheTheremainderofthepaperisstructuredasfollows.First,inthefollowingsection,wegiveamotivationandinformalintroductiontotheframework.
In
section3,wepresentaproof-of-conceptformalisationofourapproachusinghierar-chicalmeta-logic,andinsection4,wepresentsomeconclu-sions.Ourworkmakestwokeycontributionstothetheoryofargumentation.First,andperhapsmostimportantly,wemotivateandestablishthenotionofmeta-argumentationasanissueinitsownright,andpresentafirstformalisationofthisprocess.Although
meta-languages
havebeenusedintheformalisationofdialecticalsystems[20],tothebestofourknowledgewearethefirsttousea
meta-logic
inthisway.Oursecondcontributionistoshowhowanumberofdiffer-tiersofthehierarchyareasfollows:
~o
TheObjectLevel:
Thistierofthehierarchydoesnotactuallycontainargumentsatall.
It
consistsofstate-mentsaboutthedomainofdiscourse,andinparticu-lardefinestheinterrelationshipsbetweentheentitiesinthedomainofdiscourse.Inalegalsetting(whichisperhapstheparadigmexampleofadomainforfor-malargumentanddiscourse),wecanthinkof
~o
asconsistingoftheestablishedfactsofthecase,(suchasevidencethatmaybeintroduced),aswellasnon-logicalaxiomsaboutthedomain.
~1
GroundArguments:
Argumentsexistforthefirsttimeasfirstclassentitiesinthistierofthehierarchy.
~1
entapproachestoargumentationmaybeuniformlycom-binedwithinthemeta-logicframework:inparticular,thelogic-basedapproachesof[2,13],theabstractargumenta-tionframeworkofDung[7],andBench-Capon'svalue-basedargumentationframework[1].Notethattheintegrationofabstractargumentframeworksandlogic-basedframeworksdefineswhatconstitutesanargument:inthemodelofargumentthatweuse,anargumentconsistsofacon-elusionandsomesupportingstatements,withanotionoflogicalconsequencebetweenthem[2,13].Bycon-trast,inToulmin'sschemeanargumentismorecom-plex,consistingofaclaim(e.g.,"Johnisold"),awar-ispossibleonly
because
weadoptameta-logicalperspec-tive:theintegrationinvolvesstatingandreasoningaboutrelationsoverlogicalformulae,whichcannotbeachievedwithoutsomemeta-logicalapparatus.rant(e.g.,"over70isold")withassociatedbacking(e.g.,somedemographicdata),andsomedata(e.g.,"Johnis78")[22].Notethatthehierarchicalmeta-
2.AHIERARCHICALSYSTEMOFARGIJ-
logicapproachitselfisconsistentwithbothsuchmod-elsofargument,andindeedmanyothers;butwefind
MENTS
itconvenienttoworkwiththelogicalmodel.Sincewecanrefertoargumentsinthistierinthehierar-Beforeproceedingtotheformaldetailsofourapproach,wepresentsomemoredetailedmotivationforit.Asnotedintheintroduction,ourkeymotivationisthefollowingob-servation:Argumentationandformaldialogueisnecessarilyameta-logicalprocess.chy,wecanalsocapturerelationships
between
argu-mentshere.Forexample,thecanonicalnotionofoneargument
attacking
anotherisarelationbetweenar-guments[7],andcannotthereforebepresentatanylowertierofthehierarchy.Although"attack"isonerelationthatmayexistbetweenarguments,itisofcoursenottheonlyone:sincetheobjectlevel
~o
willThisseemsincontrovertible:eventhemostsuperficialstudyofargumentationandformaldialogueindicatesthat,notonlyareargumentsmadeaboutobject-levelstatements,theyarealsomade
about
arguments.Insuchcases,anargumentismadewhich
refers
toanotherargument.Moreover,thereareclearlyalsocaseswherethelevelofreferralgoesevendeeper:whereargumentsrefertoargumentsthatrefertoarguments.Perhapstheparadigmexamplesofsuchmeta-argumentationwouldbeinacourtroomsetting,whereanadvocateobjectstoanargumentoftheopposingadvocate,orwhereajudgerulesanargumentinadmissible.Here,theoftencontaininconsistencies,thenotionofattackwilloftennotbeenoughtoobtainausefulcoherentview.WethereforeuseBench-Capon'snotionof
value-based
argument,whichoverlaysattackwith
values
thattheargumentappealsto,andhencemakesitpossibletochoosebetweenargumentsonthebasisofthevaluestheyrepresent[1].
~2
Meta-Arguments:
Noticethatatthe
~1
tierofthehi-erarchy,wecanmakestatementsthatareaboutobject-levelstatements,(e.g.,wecanassertthataparticularstructurerepresentsanacceptableargument)butweargumentsbeingputforwardrefertoargumentsmadeaboutthedomainofdiscourse,butareclearlynotactuallyaboutthedomainofdiscourseitself.cannotdirectlyrefertothe
process
bywhichanargu-mentisestablished.Thatis,in
~1
wecannotsaythat"wecanestablishthat
a
isanargumentusingaxioms
If
oneacceptsthevalidityof
(*),
thenitisnaturaltoviewargumentastakingplaceatanumberoflevels.Atthelow-
T".
Hencepropertiesofargumentsthatinvolverefer-ringtotheaxiomsorproceduresviawhichweinfactestlevel,wedonotreallyhaveargumentsatall-wehavestatementsaboutthedomainofdiscourse.Atthenextlevelintheargumentationhierarchy,wehaveargumentsthem-selves:thesearestatementsabouttheobject-levelstate-ments,andsoon.Ofcourse,inanyattempttoformalisesuchamodelofarguments,wemustdefinethecompositionofeachlevelofthehierarchy.Therearemanychoicestobeestablishthattheyareargumentscannotbecapturedin
~1.
However,suchproperties
can
becapturedin
~2.
Inparticular,themainconstructionusedin
~2
isthatofanargumentreferringtoanargument.Toillustratethevalueofthis,wewillshowhowwecandistinguishin
~2
between"classical"
121
arguments(inwhichthefulltechnicalapparatusofclassicallogicproofcanbeusedtoestablishaconclusion),and
intu-itionistic
121
arguments,whereamorerestrained(andmadehere-particularlyathigherlevelsofthehierarchy-andweareinnopositiontogiveacanonicalview.Inthis
 
refersto
Meta-arguments:
argumentsaboutarguments&sentencesaboutarumentconstruction
Groundarguments:
statementsaboutstatementsabouttheworldreferstorefersto
Objectlevel:
statementsaboutthedomainofdiscourseTheDomainofDiscourse
Figure
1:A
hierarchyofarguments.somewouldarguemorerealistic)notionofproofisused[6].afirst-ordermeta-logicissimplyafirst-orderlogicwhosedomain(thesetofentitiesthatmaybereferredtointhelanguage)includessentencesofanotherlanguage(the
object
language).Animportantdistinctionismadebetweenmeta-languagesthatcanrefertothemselves(i.e.,languageswhosedomaincontainsthesetofsentencesofthelanguageitself),whichareusuallycalled
self-referential,
andthosewherethisisnotpossible.Self-referentiallanguagestendtoberathercomplexandintricatesystemstodealwith:firstbecausewhenoneassumesevenseeminglyinnocentandinnocuousaxiomstheytendtobecomeinconsistent,andsecondbe-causetheyallowonetoexpressparadoxicalstatementssuchasthe"liar"paradox
[15,
23].
First-orderhierarchicalmeta-languages
provideasomewhatmorestablelogicalfounda-tion[23].Thebasicideaofsuchlanguagesisthatwedefinea(well-founded)toweroflanguages
L
o
-L
1-···,
suchthatthedomainof
Lo
isthesetofentitiesinthedomainofdiscourse,andthedomainofeachlanguage
La
for
u
>
0containsthesetofformulaeoflanguage
E;
for0::;
v
<
u,
butcontainsnosentencesfromlanguages
Lw
for
w:;:,
u.
Inthisway,wehavetheabilitytomakestatementsaboutstatementsaboutstatements...tosomearbitrarylevelofdepth,butbecauseourlanguagesarestrictlyhierarchical(canonlyre-fertosentencesoflanguagesfurtherdownthehierarchy),self-reference(andallthelogicalproblemsitentails)isnotpossible.Hierarchicalmeta-languageshavebeenusedasthebasisofseveralformalismsforreasoningaboutaction(see,e.g.,
[10])
andrecentlytheapproachofusingmeta-languagepredicatesinplaceofmodaloperatorsforreferringto(forexample)whatanagentknowsorbelieveshasundergonesomethingofarevival(see,e.g.,[9]).Wewillpresentourformalisationofeachtierinthehierarchyinturn,startingwith
60.
Notethatwedonotgiveasyntaxandsemanticsforeachlanguage
Li,
astheseareavailableelsewhereinthelitera-ture(e.g.,
[10,23]).
Wewillassumethatthelanguagescon-taintheconventionallogicalconnectivesofnegation(","),disjunction("V"),conjunction
("1\"),
implication
("--+"),
andbi-conditional("-"),theusualapparatusoffirst-orderOfcourse,thereisnoreasonwhythishierarchyshouldquantification(
"'7'''
"3"),functionalterms,equality,andnotbecontinued:thesamelogicalapparatusweusecanessentiallybecopiedintolayersfurtherupthehierarchy,permittingargumentsaboutargumentsaboutargumentsasdesired.Whereargumentationisusedinhumansettings,thisisexactlywhatseemstohappen:consideranargumentthattakesplacebetweenadvocatesinacourtoflaw,andthen(furtherupthehierarchy),argumentsmadebythejudgeaboutthesearguments,andthenpotentiallyargumentsmadeinasupremecourtabouttheargumentsmadebythejudgeinthelowercourt.Tocleanly(andprop-erly)capturethiskindofsetting,itseemstousthatourhierarchicalapproachisnotonlyappropriate,butperhapsessential.However,forthepurposesofthispaper,wewillrestrictourattentiontothethreelayersindicatedhere.
3.THEFORMALFRAMEWORK
Meta-levelreasoning(reasoningaboutreasoning)hasavenerablehistoryinartificialintelligence,and
logical
ap-proachestometa-levelreasoninghavebeenwidelystudied,logicalconstantsfortruth("true")andfalsity("false").Moreover,foreachlanguage
E,
weassumealogicalcon-sequencerelation
F
L,'
Technically,eachlevel
6i
inourhierarchywillconstitutea
theory
inthelanguage
Li.
3.1TheObject/DomainLevel:
60
Wecanunderstand
60
asstatingthebasic"facts"oftheargumentationdomain,andthenon-logicalaxiomsassoci-atedwith
it.".
Weoftenreferto
60
astheobject-level,ordomaintheory.Thus,inthedomaintheory
60,
wedefineallthepropertiesabouttheargumentationdomainthatmaybeadmittedintothediscourse.Forsimplicityofexpositionhere,wewillassumethattheseareexpressedusingproposi-tionallogic,althoughofcoursethereisnoreasoninprinciplewhyoneshouldnotusearicherlanguage.Formally,
60
willbeasetofformulaexpressedinpropositionallogic.1By"non-logical"axioms,wemeanaxiomsorruleswhichreferspecificallytothedomainathand,andwhicharenotwitharangeofapproachesdevelopedandevaluated(see[8,validaccordingtothesemanticsofthelogic.

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->