25.10.2012 Views

Obligationes

Obligationes

Obligationes

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

TableofContents<br />

Preface 7<br />

LiborBěhounek<br />

FuzzyLogicsInterpretedasLogicsofResources 9<br />

FrancescoBerto<br />

StrongParaconsistencyandExclusionNegation 23<br />

CatarinaDutilhNovaes<br />

Medieval<strong>Obligationes</strong>asaRegimentationof<br />

‘theGameofGivingandAskingforReasons’ 35<br />

ChristianG.Fermüller<br />

TruthValueIntervals,Bets,andDialogueGames 51<br />

BjørnJespersen&MarieDuˇzí<br />

ProceduralSemanticsforMathematicalConstants 65<br />

KoheiKishida<br />

NeighborhoodIncompatibilitySemanticsforModalLogic 79<br />

VojtěchKolman<br />

WhatdoGödelTheoremsTellusabout<br />

Hilbert’sSolvabilityThesis? 91<br />

TimmLampert<br />

WittgensteinonPseudo-Irrationals,<br />

DiagonalNumbersandDecidability 103<br />

RosenLutskanov<br />

WhatistheDefinitionof‘LogicalConstant’? 119<br />

OndrejMajer&MichalPeliˇs<br />

EpistemicLogicwithRelevantAgents 131<br />

PeterMilne<br />

BettingonFuzzyandMany-valuedPropositions 145<br />

JaroslavPeregrin<br />

InferentializingConsequence 155


MartinPleitz<br />

MeaningandCompatibility:<br />

BrandomandCarnaponPropositions 169<br />

DagPrawitz<br />

InferenceandKnowledge 183<br />

CarolineSemmling&HeinrichWansing<br />

ASoundandCompleteAxiomaticSystemof<br />

bdi–stitLogic 201<br />

SebastianSequoiah-Grayson<br />

AProceduralInterpretationofSplitNegation 219<br />

StewartShapiro<br />

ReferencetoIndiscernibleObjects 231<br />

PeterSchroeder-Heister<br />

SequentCalculiandBidirectionalNaturalDeduction:<br />

OntheProperBasisofProof-theoreticSemantics 245<br />

Vítězslav ˇ Svejdar<br />

RelativesofRobinsonArithmetic 261<br />

LucaTranchini<br />

TheRoleofNegationinProof-theoreticSemantics:<br />

aProposal 273<br />

MichaelvonBoguslawski<br />

OivaKetonen’sLogicalDiscovery 289


Preface<br />

TheinternationalsymposiumLogicaorganizedbytheInstituteofPhilosophyoftheAcademyofSciencesoftheCzechRepublichasarelativelylong<br />

history. Itbeganin1987whenthefirstconferenceofserieswasheldin<br />

LibliceChateau.Inthebeginningtheconferenceswereofmostlylocalimportance,butovertheyearstheyhaveacquiredthestatusofaprestigious<br />

internationalconferencewithamultidisciplinaryflavour.<br />

Theannualsymposiacoverabroadfieldoflogicaltopicsandaimto<br />

promotedialoguebetweenvariousbranchesoflogic. Logicahostedmany<br />

presentationsbytopspecialistsinmathematicalandphilosophicallogicas<br />

wellasanalyticalphilosophyandlinguistics. Theprofessionalorientation<br />

oftheconferencecanbeillustratedbymentioningseveralnamesofscholars<br />

thattheconferencewelcomedastheinvitedspeakers:NuelBelnap,Simon<br />

Blackburn,RobertBrandom,MelvinFitting,YuriGurevich,PetrHájek,<br />

RomHarré,JaakkoHintikka,WilfridHodges,DavidLewis,PerMartin-Löf,<br />

BarbaraPartee,GrahamPriest,GregRestall,GabrielSandu,andStewart<br />

Shapiro.<br />

Amongthecentralpointsofthe‘publicationpolicy’behindtheproceedingsoftheconferenceistherulethatthevolumehastoappearbeforebeginningofthenextyear’sconference.This‘rush’,however,doesnotaffect,webelieve,thequalityofthepreparationofthevolumes.<br />

Untillast<br />

yeartheywerepublishedbytheInstitute’spublishinghouseFilosofia;the<br />

presentvolumeisthefirstpreparedforCollegePublications.Itcontainsa<br />

majorityofthepaperspresentedatthesymposiumLogica2008,whichtook<br />

placefromJune16to20intheformerFranciscanmonasteryofHejnice,the<br />

CzechRepublic,wheretheparticipantsspentfivedaysnotonlyinthelecturinghallbutalsoinmanyinformaldiscussionsduringthebreaks,lunches<br />

andsocialevents.AsyoucanseeTheLogicaYearbook2008bringstogether<br />

varioustextsfrommathematicalandphilosophicallogic,historyandphilosophyoflogic,andnaturallanguageanalysis.<br />

Ashasbecomeusualfor<br />

theYearbookseries,thearticleshavenotbeensortedbysubject—they<br />

areorderedalphabeticallybyauthoranditisuptothereadertopickand<br />

choose.<br />

Editor’sacknowledgements<br />

BoththeLogicasymposiumandthisbookseriesaretheresultofajoint<br />

effortofmanypeople,whodeservemydeepthanks.Amongthemarethe<br />

mainorganizersVladimírSvobodaandTimothyChildersfromtheDepartmentofLogicoftheInstituteofPhilosophyandPavelBaran,thedirector<br />

oftheInstituteofPhilosophy. Theconferencewaspromotedalsobythe<br />

GrantAgencyoftheCzechRepublic,whichprovidedsignificantsupportby<br />

financingthegrantprojectno.401/07/0904. Wearefurtherindebtedto<br />

MarieVučková,HeadoftheForeignRelationsDepartmentoftheInstitute,


8 TableofContents<br />

fororganizationalsupport.Theorganizationwouldbeimpossiblewithout<br />

thehelpofPetraIvaničováduringandbeforetheconference.Ourstayin<br />

HejniceMonasterywasmadepleasantthroughtheeffortsofFatherMiloˇs<br />

Rabanandthestaffofthemonastery.SpecialthanksalsogototheBernard<br />

FamilyBreweryofHumpolec,traditionalsponsorofthesocialprogramme<br />

ofthesymposium.IwouldalsoliketothanktoMarieBenediktováforthe<br />

layoutofthisvolume.ManythanksgotoCollegePublicationsanditsmanagingdirectorJaneSpurr.Lastbutnotleastwewouldliketothankallthe<br />

conferenceparticipantsandtoauthorsofthearticlesfortheiroutstanding<br />

cooperationduringtheeditorialprocess.<br />

Prague,May2009 MichalPeliˇs


Fuzzy Logics Interpreted as Logics of Resources<br />

Libor Běhounek ∗<br />

Girard’slinearlogic(1987)isofteninterpretedasthelogicofresources,<br />

whileformalfuzzylogics(seeesp.Hájek,1998)areusuallyunderstoodas<br />

logicsofpartialtruth. Iwillarguethatdeductivefuzzylogicscanbeinterpretedintermsofresourcesaswell,andthatundermostcircumstances<br />

theyactuallycaptureresource-awarereasoningmoreaccuratelythanlinear<br />

logic.Theresource-basedinterpretationthenprovidesanalternativemotivationforformalfuzzylogics,andgivesanexplanationofthemeaningof<br />

theirintermediarytruthvaluesthatcanbejustifiedmoreeasilythantheir<br />

traditionalmotivationbasedonpartialtruth.<br />

1 Linearandsubstructurallogics<br />

Recallthatlinearlogicanditsvariantsarerepresentativesofbasicsubstructurallogics(see,e.g.,Restall,2000,Paoli,2002,Ono,2003),i.e.,logicsthat<br />

resultfromdiscardingsomeofthestructuralrulesfromtheGentzen-style<br />

calculiLKandLJforclassicalandintuitionisticlogic.Inparticular,linear<br />

logic LLdiscardstherulesofcontraction (C)<br />

andweakening (W)<br />

Γ,A,A,∆ =⇒ Σ<br />

Γ,A,∆ =⇒ Σ<br />

Γ =⇒ Σ<br />

A,Γ =⇒ Σ<br />

Γ =⇒ Σ,A,A,Π<br />

Γ =⇒ Σ,A,Π<br />

Γ =⇒ Σ<br />

Γ =⇒ Σ,A<br />

fromthecalculusLKforclassicallogic.Intuitionisticlinearlogic ILLdiscardsthesamerules<br />

(C,W)fromthecalculusLJforintuitionisticlogic.<br />

Affinelinearlogic ALLandintuitionisticaffinelinearlogic IALLdiscard<br />

onlytheruleofcontraction (C)fromthecalculiLKandLJ,respectively,<br />

butretaintheruleofweakening (W).<br />

∗ TheworkwassupportedbyGrantNo.IAA900090703“Dynamicformalsystems”ofthe<br />

GrantAgencyoftheAcademyofSciencesoftheCzechRepublicandbyInstitutional<br />

ResearchPlanAV0Z10300504.


10 LiborBěhounek<br />

Recallfurtherthatsubstructurallogicsworkingeneralwithtwoconjunctions:<br />

thelatticeconjunction ∧(alsocalledweak,additive,orextensionalconjunction)andfusion<br />

&(alsocalledgroup,strong,multiplicative,<br />

orintensionalconjunction).Similarlythereareingeneraltwodisjunctions<br />

(lattice ∨andstrong)aswellastwoimplications,twonegations,etc.,but<br />

thelattersplitconnectiveswillnotplayasignificantroleinouraccount,<br />

asweshallmainlydealwithintuitionisticsubstructurallogics(whichlack<br />

strongdisjunction)andcommutativefusion(thenbothimplicationscoincide).Sinceinsuchsubstructurallogicsimplicationinternalizesthesequent<br />

sign =⇒and &thecommaontheleft-handsideofsequents(cf.Ono,2003),<br />

thevalidityofthesequent A1,... ,An =⇒ Bisequivalenttothevalidity<br />

oftheformula A1 & ... & An → B. Consequently,theruleofcontraction<br />

correspondstothevalidityof A → A & Aandtheruleofweakeningtothe<br />

validityof A & B → A.<br />

Thealgebraicsemanticsofsubstructurallogicsisthatofresiduatedlattices(see,e.g.,Jipsen&Tsinakis,2002;Ono,2003;Galatos,Jipsen,Kowalski,&Ono,2007),i.e.,latticesendowedwithanadditionalmonoidaloperation<br />

∗(representing &)monotonew.r.t.thelatticeorder ≤,anditstwo<br />

residuals /, \(representingimplications)thatsatisfytheresiduationlaw<br />

x ∗ y ≤ z iff y ≤ x\z iff x ≤ z/y.<br />

If ∗iscommutative,thetworesiduals /, \coincideandareusuallydenoted<br />

by ⇒.Thesetofdesignatedelementsis {x | x ≥ 1},where 1istheneutral<br />

elementofthemonoidaloperation ∗.Ifconvenient,residuatedlatticesmay<br />

beexpanded(toOno’s FL-algebras)byaconstant 0forfalsity,whichmakes<br />

itpossibletodefinenegationas x ⇒ 0.<br />

Thetermsubstructurallogicswillinthispaperdenotelogicsofclasses<br />

ofresiduatedlattices,followingthestipulativedefinitionbyOno(2003).<br />

Inparticular,(affine)intuitionisticlinearlogicisthelogicofall(bounded<br />

integral)commutativeresiduatedlattices, 1 and(affine)linearlogicisthe<br />

logicofthosethatfurthermoresatisfythelawofdoublenegation.<br />

2 Linearlogicasthelogicofresources<br />

Thereasonwhylinearlogichasbeenregardedasthelogicofresourcesis<br />

illustratedbyGirard’s(1995)well-known‘Marlboro–Camels’example:<br />

1 Aresiduatedlatticeiscalledcommutativeifitsmonoidaloperation ∗iscommutative;<br />

itiscalledboundedintegralif 0 ≤ x ≤ 1forallelements x.Weshallusuallyworkwith<br />

commutativeresiduatedlatticesonly.


FuzzyLogicsInterpretedasLogicsofResources 11<br />

Considerthepropositions<br />

Thenthesequent<br />

expressingtheinference<br />

D =“Ipay$1.”,<br />

M =“IgetapackofMarlboro.”,<br />

C =“IgetapackofCamels.”<br />

D → M,D → C =⇒ D → M & C<br />

IfIpay$1,IgetapackofMarlboro<br />

IfIpay$1,IgetapackofCamels<br />

∴IfIpay$1,IgetapackofMarlboroandIgetapackofCamels<br />

isderivablebytherulesofclassicalaswellasintuitionisticlogic.Theinferenceis,however,viewedascounter-intuitive,iftheconclusionisstraightforwardlyunderstoodasgettingbothpacks.Thedisputablesequentisnot<br />

derivableinlinearlogic,though:linearlogiconlyderivesthesequent<br />

D → M,D → C =⇒ D & D → M & C<br />

whichunderasimilarinterpretationcapturesthefactthatIneedtopay<br />

twodollarstogetbothpacksofcigarettes.<br />

Inthissense,linearlogicissaidtoregardformulaeas‘resources’,which<br />

are‘spent’whenusedaspremisesofimplications(intheMarlboro–Camels<br />

example,thepremise Disspentbybeingdetachedfrom D → Mtoobtain<br />

M,andcannotbeusedagainfor D → Ctoobtain M & C).Moreformally,<br />

sincepremisescannotinlinearlogicbecontracted(duetothelackoftherule<br />

(C)),theyactastokensfor‘resources’neededtosupporttheconclusion:a<br />

sequentisvalidinlinearlogiconlyifithastheneededamountsofpremises<br />

requiredforarrivingattheconclusion. 2 Inotherwords,linearlogic‘counts’<br />

premisesofsequentsasiftheyrepresentedresourcesneededfor‘buying’the<br />

conclusion(wheredifferentpropositionalletterswouldrepresentdifferent<br />

typesofresources,whiletheiroccurrencesinthesequentwouldrepresent<br />

tokensorunitsofthattype).<br />

Nevertheless,thisfeatureoflinearlogicisduesolelytotheabsenceof<br />

therule (C)ofcontraction,andthereforeiscommontoallcontractionfreesubstructurallogics.<br />

Itisnotclearwhyexactlylinearlogicshould<br />

2 Exactlytheneededamountsin LLor ILL;atleasttheneededamountsintheiraffine<br />

versions(itbeinganeffectofweakeningthatweneednotspendallpremises).Inlogics<br />

withboth(C)and(W),e.g.,classicalorintuitionisticlogic,eachpremiserequiredfor<br />

arrivingattheconclusiononlyneedstobepresentatleastonce.


12 LiborBěhounek<br />

bemoreadequateasalogicofresourcesthananyothercontraction-free<br />

logic. Rather,itistobeexpectedthatdifferentcontraction-freelogics<br />

willcorrespondtodifferentassumptionsonthestructureofresources. In<br />

thefollowingsectionsIwillarguethatlinearlogicsareinfactadequate<br />

onlyforverygeneralstructuresofresources,whileundermostcommon<br />

circumstances,strongerlogicsareappropriate.<br />

3 Thestructureofresources<br />

Asafirsttask,weneedtorefineourconceptionofresources.Sinceweaim<br />

ataninformalsemanticexplanationofcertainlogics,insteadofgivinga<br />

formaldefinitionweshalljustlistafewexamplesindicatingwhatkindof<br />

resourceswehaveinmind,andspecifythemathematicalpropertiesthey<br />

areassumedtosatisfy.<br />

Ournotionofaresourcewillberatherbroad:itcanincludeanykind<br />

ofthingsthatcanbecountedormeasured,thatcanbeacquiredandexpended,orusedforanypurpose.<br />

Amongtheresourcesweconsiderare,<br />

e.g.:money(costs,prices,debts,etc.);goods(packsofcigarettes,clothes,<br />

cars,etc.);industrialmaterials(chemicals,naturalrawmaterials,machine<br />

components,etc.);cookingingredients(flour,salt,potatoes,etc.);computer<br />

resources(diskspace,computationtime,etc.);penalties(whichcanberegardedasakindofcostsincurred);sets,multisets,orsequences(tuplesor<br />

vectors)oftheabove;etc.<br />

Itcanbeobservedthatallofthese(aswellasmanyother)kindsof<br />

resourcesexhibitthestructureofaresiduatedlattice. Inparticular,there<br />

is:<br />

•Apartialorder �comparingtheamountsoftheresources. Forinstance,300gofflourismorethan200gofflour;twopensandthree<br />

pencilsaremorethanonepenandthreepencils;etc.Forthesakeof<br />

compatibilitywithfurtherdefinitions,weshallunderstand x � yas<br />

“theresource xislargerthanorequalto y.”Theorderneednotbe<br />

linear,asforinstancetwopensarenotcomparablewiththreepencils<br />

(ifdifferentitemsarecountedseparately).However,itcanbeassumed<br />

that �isalatticeorder,asthisistrueforallprototypicalcases:by<br />

definition,itamountstosupposingthatforanytworesources x,y(for<br />

instance: x=2pensand3pencils; y=1penand4pencils),there<br />

istheleastresourcethatisatleastaslargeasboth(inthiscase,<br />

2pensand4pencils)andthelargestresourcethatisatmostaslarge<br />

asboth(here,1penand3pencils). Eventhoughtheremayexist<br />

resourcesthatdonotsatisfythisassumption,weleavethemasidein<br />

ourconsiderations.


FuzzyLogicsInterpretedasLogicsofResources 13<br />

•Amonoidaloperation ∗ofcomposition(orfusion)ofresources. For<br />

example,300gofflourand200gofflouris500gofflour;2pensand<br />

3pencilsplus1penand3pencilsare3pensand6pencils;etc.Putting<br />

theresourcestogethercanbeassumedtobeassociative(i.e.,wepresumethatthetotalsumdoesnotdependontheorderofsummation).<br />

Thekindsofresourcesweconsideralwayshaveaneutralelement e,<br />

theemptyresource,whichdoesnotchangetheamountwhenadded<br />

toanotherresource:e.g.,0gofflour;0pensand0pencils;etc.Even<br />

thoughcompositionofresourcesneednotbecommutative(consider,<br />

e.g.,theorderofaddingingredientswhencooking),forthesakeof<br />

simplicityofexpositionweshallonlyconsidercommutative ∗here<br />

(generalizationtonon-commutative ∗isalwaysstraightforward).<br />

•Finally,resourcesofalltypicalkindscanbe‘subtracted’or‘evened<br />

up’,i.e.,theircompositionhastheresidualoperation ⇒expressingthe<br />

remainder,orthedifferenceofamounts: x ⇒ yistheleastresource<br />

tobeaddedto xinordertogetaresourceatleastaslargeas y. 3 For<br />

example,if x=200gofflourand y=300gofflour,then x ⇒ yis<br />

100gofflour,asoneneedstoadd100gofflourto200gofflourto<br />

getatleast300g;whileif x=2pensand3pencils,and y=1pen<br />

and3pencils,then x ⇒ yis0pensand0pencils(i.e.theempty<br />

resource e),asweneednotaddanythingto xtogetatleast y.<br />

Allkindsofresourcesweconsiderthushavethestructureofa(commutative)residuatedlattice<br />

L = (L, ∧, ∨, ∗, ⇒,e). Particularkindsofresources<br />

canhaveadditionalproperties:forexample,mostusualkindsofresources<br />

satisfytheso-calleddivisibilitycondition x ∗ (x ⇒ y) = x ∧ y.<br />

Sinceweaimatasimpleresource-basedinterpretationofexistinglogical<br />

calculiratherthandevelopmentofanexpressivelyrichlogicofresources<br />

forcomputerscience,wedonotconsidersuchphenomenaas,e.g.,resource<br />

dynamicsorpossiblenon-totalityof ∗(whicharemodeledbysuchsystemsas<br />

thelogicofbunchedimplications,computationlogics,orsynchronousand<br />

asynchronouscalculi—see,e.g.,Pym&Tofts,2006forreferences),but<br />

onlyreconstructandrefinetheassumptionsonresourcesthatareadopted<br />

bylinearlogic.<br />

4 Formulaeasresources<br />

Thereareatleasttwopossiblerepresentationsofresource-basedsemantics<br />

ofsubstructurallogics. Oneofthemtakesresources(i.e.elementsofthe<br />

3 I.e. x ⇒ y = sup{z | z ∗ x � y},whichisanequivalentformulationoftheresiduation<br />

lawincompletelattices. Forincompletelattices,amorecautiousformulationbasedon<br />

Dedekind–MacNeillecutsisdue,namely {z | z ∗ x � y} = {z | z � x ⇒ y},whichisa<br />

generalequivalentoftheresiduationlaw.


14 LiborBěhounek<br />

residuatedlattice LdescribedinSection3)directlyassemanticvaluesassignedtopropositionalformulae.<br />

Recallthatalogicalcalculuscanhave<br />

interpretationsotherthanpropositional:cf.,e.g.,theinterpretationofthe<br />

Lambekcalculusasthecategorialgrammar(wherethesemanticvaluesof<br />

formulaearegrammaticalcategories),ortheCurry–Howardcombinatorial<br />

interpretationoftheimplicationalfragmentofintuitionisticlogic(where<br />

formulaeareinterpretedastypesandproofsasprograms). Inasimilar<br />

vein,wecaninterpretthealgebraicsemanticsofsubstructurallogicsunder<br />

the“formulae-as-resources”paradigmasfollows:<br />

•Thesemanticvalueofaformula ϕisaresource �ϕ� ∈ L.<br />

•TheTarskicondition �1� = eofthealgebraicsemanticsinterpretsthe<br />

formula 1astheemptyresource(or‘beingforfree’).<br />

•Similarly,theclause �ϕ & ψ� = �ϕ� ∗ �ψ�saysthatconjunctionrepresentsthefusionofresources.<br />

•Thevalueofimplication, �ϕ → ψ� = �ϕ� ⇒ �ψ�,istheresource<br />

neededtogetatleast �ψ�,giventheresource �ϕ�.<br />

•Finally,thelatticeconnectives ∧, ∨representthemeetandjoinof<br />

resources(withrespecttothesizeorder �ofresources).<br />

Theformula ϕisregardedasvalidunderagivenevaluationiff e � �ϕ�,i.e.,<br />

iffitrepresentsaresourcethatisforfreeorevencheaper.<br />

5 Resourcesaspossibleworlds<br />

Anotherwayhowtointerpretsubstructurallogicsintermsofresources<br />

(cf.Pym&Tofts,2006)istoregardthestructure Lofresourcesasa<br />

Kripkeframe (L, �)endowedwithamonoidalstructure (∗,e).Unlikeinthe<br />

“formulae-as-resources”paradigm,formulaearehereinterpretedaspropositions,andresourcesonlyserveasindicesthatmay(ormaynot)validate<br />

them. Theforcingrelation r � ϕ,“theresource r ∈ Lsupportstheformula<br />

ϕ,”isrequiredtosatisfythefollowingconditions:<br />

• e � 1,<br />

• r � ϕ & ψiff ∃s,t ∈ L: r � s ∗ tand s � ϕand t � ψ,<br />

• r � ϕ → ψiff ∀s ∈ L: if s � ϕ,then r ∗ s � ψ,<br />

• r � ϕ ∧ ψiff r � ϕand r � ψ(“sharedresources”—contrastthe<br />

clausefor &),


FuzzyLogicsInterpretedasLogicsofResources 15<br />

• r � ϕ ∨ ψiff ∃s,t ∈ L: r � s ∨ tand (s � ϕor s � ψ)and (t �<br />

ϕor t � ψ),<br />

andtheconditionofpersistence(if r � sand s � ϕ,then r � ϕ),expressing<br />

that“largerresourcessufficeaswell.”Theformula ϕisdefinedtobevalid<br />

under �iff e � ϕ,i.e.,iffsupportedevenbytheemptyresource. 4<br />

6 Theroleoftautologies<br />

Intheabovesemantics,tautologiesw.r.t.aclass Kof(commutative)residuatedlatticesaredefinedastheformulae<br />

ϕthatgetavalue �ϕ� � eunder<br />

allevaluationsofpropositionallettersinanyresiduatedlattice L ∈ K(resp.<br />

aresupportedby eunderall �ineveryKripkeframe L ∈ K).Thetautologiesofsubstructurallogicsthusrepresentcombinationsofresourcesthat<br />

arealways“forfreeorcheaper”.<br />

Moreimportantly,sinceallresiduatedlatticesvalidate<br />

e � r ⇒ s iff r � s,<br />

tautologiesoftheform ϕ → ψinternalizesoundrulesofresourcetransformationsthat“preserveexpenses”(inthesenseof�).Inferenceinsubstructurallogicscanthusbeunderstoodasinferencesalvisexpensis,inasimilar<br />

mannerasinferencesalvaveritateinclassicallogic. 5<br />

Classesofresiduatedlatticesadmittedaspossiblestructuresofresources<br />

thendetermineparticularlogicsofresourcesintheabovesense.Inparticular,bytheknowncompletenesstheorem,<br />

ILListhelogicofallcommutative<br />

residuatedlattices,andsoitisanadequatelogicifjustthegeneralstructure<br />

ofacommutativeresiduatedlatticeisassumedforadmissiblekindsofresources.Itsvariants<br />

IALL, ALL,and LLrestrictthestructureofresources<br />

tonarrowerclassesofcommutativeresiduatedlattices,andothersubstructurallogicscorrespondtofurtherspecificclassesofresiduatedlatticesof<br />

resources. 6<br />

InthefollowingsectionsIwillarguethatmosttypicalkindsofresources<br />

satisfytheso-calledprelinearitycondition,andsoareinfactgovernedby<br />

deductivefuzzylogicsratherthanlinearlogics.<br />

4 Asthisisnottheaimofthispaper,weomitthedetailsonthecorrespondencebetween<br />

theKripke-styleandalgebraicsemanticsofsubstructurallogics. Formoreinformation<br />

see(Ono&Komori,1985).<br />

5 Notethatthegeneralvalidityof �ϕ� � �ψ�definesthelocal consequencerelation<br />

(expressed,i.a.,bysequentsinSection1),whileHilbert-stylecalculiforsubstructural<br />

logicsusuallycapturetheglobalconsequencerelation“e � �ψ�whenever e � �ϕ�”.<br />

6 Forexample,classicallogiccanbeinterpretedasthelogicdistinguishingjusttwosizes<br />

ofresources:empty e = �1�andnon-empty f = �0� ≺ e.


16 LiborBěhounek<br />

7 Deductivefuzzylogics<br />

Deductivefuzzylogics canbedelimitedaslogicsof(classesof)linearly<br />

orderedresiduatedlattices(Běhounek&Cintula,2006;Běhounek,2008).<br />

Amongtheextensionsof ILLtheycanbecharacterizedasthosethatsatisfy<br />

theaxiomofprelinearity (Pre): ((A → B) ∧ 1) ∨ ((B → A) ∧ 1),orinthe<br />

presenceofweakening,equivalently (A → B) ∨ (B → A).<br />

Letuscallresiduatedlatticesforwhichasubstructurallogic Lissound,<br />

L-algebras. Theprelinearityaxiomensuresthatadeductivefuzzylogic L<br />

issoundandcomplete,notonlyw.r.t.theclassofall L-algebras(thegeneralcompletenesstheorem),butalsow.r.t.theclassofalllinear<br />

L-algebras<br />

(thelinearcompletenesstheorem). Thelinearcompletenesstheoremcharacterizesdeductivefuzzylogicsamongsubstructurallogics;thefinitaryones<br />

aremoreovercharacterizedbythelinearsubdirectdecompositionproperty,<br />

whichsaysthateach L-algebraisasubdirectproduct 7 oflinear L-algebras.<br />

(SeeCintula,2006fordetails.)<br />

Besidesthegeneralandlinearcompletenesstheorems,mostimportant<br />

deductivefuzzylogicsfurthermoreenjoythestandardcompletenesstheorem,<br />

i.e.thecompletenessw.r.t.asetof(selected) L-algebrasontheunitinterval<br />

[0,1]ofreals(withtheusualordering ≤),calledthestandard L-algebras.<br />

Since L-algebrason [0,1]arefullydeterminedbythemonoidaloperation<br />

∗,standard-completedeductivefuzzylogicscanbedefinedaslogicsof(sets<br />

of)suchmonoidaloperations ∗on [0,1].Forexample,<br />

•ŁukasiewiczlogicŁisthelogicoftheŁukasiewiczt-norm x ∗ y =<br />

(x + y − 1) ∨ 0,<br />

•Gödel–Dummettlogic Gisthelogicoftheminimum,i.e.of x∗y = x∧y,<br />

•Productfuzzylogic Πisthelogicoftheordinaryproductofreals,<br />

x ∗ y = x · y,<br />

•Hájek’sbasicfuzzylogic BListhelogicofallcontinuoust-norms, 8<br />

•Monoidalt-normlogic MTListhelogicofallleft-continuoust-norms,<br />

•Uninormlogic UListhelogicofallleft-continuousuninorms,etc.<br />

Formoreinformationontheselogicssee(Hájek,1998;Esteva&Godo,2001;<br />

Metcalfe&Montagna,2007).<br />

7 I.e.asubalgebraofthedirectproductwithallprojectionstotal.<br />

8 Acommutativeassociativemonotonebinaryoperation on [0, 1]withaneutralelement<br />

e ∈ [0, 1]iscalledauninorm.At-normisauninormwith e = 1.


FuzzyLogicsInterpretedasLogicsofResources 17<br />

Theweakestdeductivefuzzylogicextendingasubstructurallogic Lis<br />

often 9 obtainedbyaddingtheprelinearityaxiom (Pre)to L:forinstance,<br />

ILL + (Pre) = UL<br />

IALL + (Pre) = MTL<br />

aretheweakestdeductivefuzzylogicsextendingintuitionisticlinearlogics,<br />

orthelogicsoflinearcommutative(boundedintegral)residuatedlattices.<br />

(For LLand ILL,thedoublenegationlawistobeaddedto ULresp.<br />

MTL.)<br />

8 Fuzzylogicsaslogicsofcosts<br />

Sincedeductivefuzzylogicsarelogicsof(specialclassesof)residuatedlattices,theycanbeinterpretedaslogicsofresourcesinthesamewayasother<br />

substructurallogics. Specifically,bythelinearcompletenesstheorem(see<br />

Section7),deductivefuzzylogicsaresoundandcompletew.r.t.particular<br />

classesoflinearresiduatedlattices,andsotheyareadequateforresources<br />

thatarelinearlyorderedby �. Inotherwords,deductivefuzzylogicsare<br />

thoselogicsofresourcesinwhichwecanassumethatallresourcesarecomparable.Prototypicallinearlyorderedresourcesarecosts,thatis,resourcesconvertedtomoney.Eventhoughresourcesingeneralneednotbecomparable(cf.theexamplesinSection3),theircosts(ifspecified)canalwaysbecompared,asmoney(ofasinglecurrency)formsalinearscale.<br />

10 Besidesmoney,<br />

therearemanyotherkindsofresourcesthatarelinearlyordered,e.g.,gallonsoffuel,computationtime,operationalmemory,etc.<br />

Irrespectiveof<br />

theirnature,weshallcallalllinearlyorderedresourcescosts,todistinguishthemfromresourcesthatarenotlinearlyordered.<br />

Forconvenience,<br />

costswithvaluesintheinterval [0,+∞],e.g.,monetaryprices(where 0is<br />

“gratis”and +∞mayrepresentthepriceofunattainablegoods),willbe<br />

calledprices.<br />

Deductivefuzzylogicscanthusberegardedaslogicsofcosts,inthesame<br />

senseaslinearlogicsareregardedaslogicsofresources.Differentwaysof<br />

addingupcosts—givenbythefusionoperation—yielddifferentdeductive<br />

fuzzylogics.Themosttypicalexamplesaregivenbelow:<br />

•Ifpricesaresummedbyordinaryaddition,weobtaintheproduct<br />

logic Π,sincetheresiduatedlattice [0,+∞]withthefusion +andthe<br />

latticeorder ≥isisomorphic(viathefunction p ↦→ 2 −p )tothestandardproductalgebra<br />

[0,1]withthefusion ·andthelatticeorder ≤.<br />

9 Alwaysifmodusponensistheonlyderivationruleof L(Cintula,2006).<br />

10 ThisideaisduetoPetrCintula(pers.comm.).


18 LiborBěhounek<br />

Notethatinthestandardproductalgebra, 0representstheinfinite<br />

costand 1thenullcost. Iftheinfinitecostisnotconsidered,the<br />

standardproductalgebrawithout 0(calledthestandardcancellative<br />

hoop)anditslogic CHL(cancellativehooplogic,seeEsteva,Godo,<br />

Hájek,&Montagna,2003)areobtained. 11<br />

•Ifpricesareboundedbyavalue a ∈ (0,+∞)andsummedbybounded<br />

additiontruncatedat a,weobtaintheŁukasiewiczlogicŁ,sincethe<br />

residuatedlattice [0,a]withboundedadditionand ≥isisomorphic<br />

via p ↦→ (a − p)/atothestandard[0,1]algebraforŁukasiewiczlogic.<br />

Thebound a(correspondingto 0inthestandardalgebra)appears<br />

naturallyif,e.g.,afixedmaximumpriceisset,ifthereisamaximal<br />

possiblecostinthegivensetting,oriftheprice aisinthegiven<br />

contextunaffordable.<br />

•Ifpricesarecombinedbythemaximum,Gödellogic G(oritshoop<br />

variant)isobtained(bythesameisomorphism p ↦→ 2 −p asinthecase<br />

ofaddition).Themaximummayseemastrangeoperationforsummationofprices,butitoccursnaturallywheneverthecostscanbeshared<br />

bythesummands. Forexample,iftemporaryresultscanbeerased<br />

beforethecomputationproceeds,thememoryneededfortemporary<br />

resultsisonlythemaximum(ratherthansum)oftheirsizes.<br />

Logicsofotherparticulart-normsareobtainedbyusingvariouslydistorted‘addition’ofprices.Forinstance,thelogicofanordinalsumofthe<br />

threebasict-normscorrespondstousingdifferentsummationrules(ofthe<br />

threedescribedabove)indifferentintervalsofprices. Thelogic MTLis<br />

obtainedifallmonotonecommutativeassociativeleft-continuousoperations<br />

withthezeropriceactingastheneutralelementareadmittedas‘addition’<br />

ofprices;similarlyfor BLandcontinuoussuchoperations,etc. Thelogic<br />

ULandotheruninormlogicsonlydifferbypermittingalsonegativeprices,<br />

whichexpressgainsratherthancosts.<br />

9 Fuzzylogicsaslogicsofresources<br />

Inspiteofthelinearcompletenesstheorem,whichmakesitpossibleto<br />

regarddeductivefuzzylogicsaslogicsoflinearlyorderedcosts,algebras<br />

fordeductivefuzzylogicsneednotbelinear(consider,e.g.,theirdirect<br />

11 Ifthecostscomeinpackages(e.g.,ifonehastobuyawholepackofcigarettesevenif<br />

oneneedsonlyafew),thealgebraisingeneraljustaΠMTL-chaininsteadofaproduct<br />

algebra,andtheresultinglogicingeneralonlyextendsthelogic ΠMTL(Esteva&Godo,<br />

2001)oritshoopvariant. Asimilareffectofpackaging,whichdestroysthedivisibility<br />

ofthealgebra(seeSection3),canbeobservedinotheralgebrasofcostsaswell. (This<br />

observationisbasedonremarksbyRostislavHorčíkandPetrCintula.)


FuzzyLogicsInterpretedasLogicsofResources 19<br />

products).Bythegeneralcompletenesstheorem(seeSection7),adeductive<br />

fuzzylogic Lisalsosoundandcompletew.r.t.theclassofall L-algebras:<br />

thus Lcanalsobeinterpretedasthethelogicofallkindsofresourcesthat<br />

formthestructureofa(possiblynon-linear) L-algebra.<br />

Letusrestrictourattentiontofinitarydeductivefuzzylogicsonly,as<br />

theyincludeallprototypicalcases;forthesakeofbrevity,letuscallthem<br />

justfuzzylogicsfurtheron.Bythelinearsubdirectdecompositiontheorem<br />

(seeSection7),any L-algebraforafuzzylogic Lcanbedecomposedinto<br />

asubdirectproductoflinear L-algebras.Fuzzylogicscanthusbecharacterizedaslogicsofsuchresourcesthateitherarelinearlyordered,orcan<br />

atleastbedecomposedintolinearlyorderedcomponents.Inotherwords,<br />

asoundandcompleteresource-basedsemanticsoffuzzylogicsneednotbe<br />

justthatofcosts,butalsothatofresourcesrepresentableastuples(possibly<br />

infinitary)ofcosts.<br />

Itcanbeobservedthatmanykindsofnon-linearresourcescanactually<br />

berepresentedastuplesoflinearlyorderedvalues.Forexample,ingredients<br />

formakingpizzaandthoseformakingspaghettiarenotsubsetsofeach<br />

other,thuscookingingredientsdonotformalinearlyorderedresiduated<br />

lattice. 12 Nevertheless,theycanbedecomposedinto(potentiallyinfinitely<br />

many)linearlyorderedcomponents,astheamountsofeachindividualitem<br />

onaningredientlistarealwayscomparable;andindeeditcanbechecked<br />

thattheprelinearityaxiomisvalidinthisresiduatedlattice. 13<br />

Infact,mosttypicalresources(includingthosementionedinSection3)<br />

areindeeddecomposableinthiswayintolinearcomponents. Evenmany<br />

resourcesforwhichsuchadecompositionisnotknown(e.g.,humanintelligence)canatleastbebelievedtobelinearlydecomposable(intosome<br />

unknownandveryfinelinearcomponents). Itisactuallyratherhardto<br />

findakindofresourcesthatdemonstrablycannotbesodecomposed.<br />

Thuswecanconcludethatalltypicalkindsofresourcesarelinearlydecomposable,andthereforetheysatisfytheaxiomofprelinearity,whichisnotvalidinlinearlogicnorinitsaffineorintuitionisticvariants;consequently,theyareactuallygovernedbydeductivefuzzylogicsratherthanlinearlogics.Linearlogicsarethusonlyadequateforaverygeneralstructureofresources,whichadmitseventherarekindsofresourcesthatarenot<br />

decomposableintolinearlyorderedcomponents.Asregardsmostusualkind<br />

12 Theelementsoftheresiduatedlatticeofallpossibleingredientlists(suchascanbe<br />

foundinrecipebooks)aretuplesofquantitiesofparticularingredienttypes(e.g.,[300g<br />

offlour,2tomatoes,2ltofoil],zeroamountsomitted).Thetuplesarenaturallyordered<br />

byinclusion(i.e.pointwisebycomponentsizes),andfusionrepresentsaddingupamounts<br />

ofeachingredient.<br />

13 Sincethefusionofamountsis(unbounded)additionineachcomponentandinfinite<br />

amountsdonotoccur,byextendingtheconsiderationsofSection8theresiduatedlattice<br />

canactuallybeidentifiedasacancellativehoop,andthelogicofcookingingredientsas<br />

thecancellativehooplogic CHL.


20 LiborBěhounek<br />

ofresources,linearlogicistooweakforthem,asitdoesnotvalidatethelaw<br />

ofprelinearitytheyobey. Assumingcommutativityoffusion,theweakest<br />

logicadequatefortypicalresourcesistheuninormlogic UL(or MTLif<br />

weakeningisassumed,i.e.,iftheemptyresourceisthesmallest). Specific<br />

structuresoftypicalresourcesaregovernedbyevenstrongerfuzzylogics<br />

—inparticular,productlogic Πifresourcesarecombinedbyadditionin<br />

eachlinearcomponent,ŁukasiewiczlogicŁiftheadditionisbounded,and<br />

Gödellogic Ginthecaseofsharedresources(i.e.,iftheycomponentwise<br />

combinebythemaximum).<br />

Thusitturnsoutthatdespitethecommonopinion,itisactuallyfuzzy<br />

logics,ratherthanlinearlogics,thatcouldbecategorizedastypicallogics<br />

ofresources. 14 Theinterpretationintermsofresourcesandcostsmoreover<br />

providesanalternativemotivationfordeductivefuzzylogicsandanexplanationofthemeaningoftheirintermediarytruth-valuesthatcaninsome<br />

respectsbemoreeasilyjustifiedthanthestandardaccountbasedondegrees<br />

ofpartialtruth.<br />

LiborBěhounek<br />

InstituteofComputerScience<br />

AcademyofSciencesoftheCzechRepublic<br />

PodVodárenskouvěˇzí2,18207Prague8,CzechRepublic<br />

behounek@cs.cas.cz<br />

References<br />

Běhounek,L.(2008).Onthedifferencebetweentraditionalanddeductivefuzzy<br />

logic.FuzzySetsandSystems,159(10),1153–1164.<br />

Běhounek,L.,&Cintula,P.(2006).Fuzzylogicsasthelogicsofchains.Fuzzy<br />

SetsandSystems,157(5),604–610.<br />

Cintula,P.(2006).Weaklyimplicative(fuzzy)logicsI:Basicproperties.Archive<br />

forMathematicalLogic,45(6),673–704.<br />

Esteva,F.,&Godo,L.(2001).Monoidalt-normbasedlogic:Towardsalogicfor<br />

left-continuoust-norms.FuzzySetsandSystems,124(3),271–288.<br />

Esteva,F.,Godo,L.,Hájek,P.,&Montagna,F.(2003).Hoopsandfuzzylogic.<br />

JournalofLogicandComputation,13(4),531–555.<br />

Galatos,N.,Jipsen,P.,Kowalski,T.,&Ono,H.(2007).Residuatedlattices:An<br />

algebraicglimpseatsubstructurallogics.Amsterdam:Elsevier.<br />

Girard,J.-Y.(1987).Linearlogic.TheoreticalComputerScience,50(1),1–102.<br />

14 Thepricepaidforthemoreaccurateaccountisamorecomplexprooftheory,asprelinearitydestroysthegoodproof-theoreticalpropertiesoflinearlogics.


FuzzyLogicsInterpretedasLogicsofResources 21<br />

Girard,J.-Y. (1995). Linearlogic: Itssyntaxandsemantics. InJ.-Y.Girard,<br />

Y.Lafont,&L.Regnier(Eds.),Advancesinlinearlogic:ProceedingsoftheWorkshoponLinearLogic,CornellUniversity,June1993<br />

(pp.1–42). Cambridge<br />

UniversityPress.<br />

Hájek,P.(1998).Metamathematicsoffuzzylogic.Dordercht:Kluwer.<br />

Jipsen,P.,&Tsinakis,C.(2002).Asurveyofresiduatedlattices.InJ.Martinez<br />

(Ed.),Orderedalgebraicstructures(pp.19–56).Dordrecht:Kluwer.<br />

Metcalfe,G.,&Montagna,F. (2007). Substructuralfuzzylogics. Journalof<br />

SymbolicLogic,72(3),834–864.<br />

Ono,H.(2003).Substructurallogicsandresiduatedlattices—anintroduction.In<br />

V.F.Hendricks&J.Malinowski(Eds.),50yearsofStudiaLogica(pp.193–228).<br />

Dordrecht:Kluwer.<br />

Ono,H.,&Komori,Y.(1985).Logicswithoutthecontractionrule.Journalof<br />

SymbolicLogic,50(1),169–201.<br />

Paoli,F.(2002).Substructurallogics:Aprimer.Kluwer.<br />

Pym,D.,&Tofts,C. (2006). Acalculusandlogicofresourcesandprocesses.<br />

FormalAspectsofComputing,18(4),495–517.<br />

Restall,G.(2000).Anintroductiontosubstructurallogics.NewYork:Routledge.


Strong Paraconsistency and Exclusion Negation<br />

1 Trueornot?<br />

Francesco Berto ∗<br />

Strongparaconsistency,alsocalleddialetheism,istheviewaccordingto<br />

whichtherearedialetheias,thatis,sentences Asuchthatboth Aand ¬A<br />

aretrue, 1 anditisrationaltoacceptandassertthem(aneminentcasebeing<br />

allegedlyprovidedbythevariousversionsoftheLiarparadox).Onecould<br />

thereforepicturedialetheismasdisputingtheLawofNon-Contradiction<br />

(LNC).Asamatteroffact,though,allthemainformulationsoftheLNC<br />

arenotdisputedbyatypicaldialetheist,inthesensethatsheiscommitted<br />

toacceptthem. Thedialetheicattitudeofthedialetheistisexpressedby<br />

typicallyaccepting,andasserting,boththeusualversionsoftheLNCand<br />

sentencesinconsistentwiththem.<br />

Ofcourse,thiscallsforadrasticrevisionofourstandardnotionsoftruth<br />

andnegation.Philosophersoftendisagreeonthecontentofbasiclogicaland<br />

metaphysicalconcepts(suchasidentity,existence,necessity,etc.),oronthe<br />

validityofsomeverybasicprinciplesofinference(suchasContraposition<br />

ortheDisjunctiveSyllogism).Itiswellknownthatthiskindofdiscussion<br />

oftenfacesimpasses,orturnsintoahardconflictofintuitions. Itisvery<br />

difficulttoestablishwhensomepartyorotherbeginstobegthequestion.<br />

Onewonderswhetheranon-standardexplanationofabasiclogicalnotion<br />

involvesarealdisagreementwithaclassicalaccountofthatnotion,orits<br />

principlessimplydescribeadifferentthingusingthesamenameorsymbol<br />

(thefamous“changeofsubject”Quineanmotto).<br />

∗ TheideasontheNOT-operatordevelopedinthispaperhavebeenhintedatin(Berto<br />

2007,Ch.14),andexposedextensivelyin(Berto2008)—IamgratefultotheEditorsof<br />

theAustralasianJournalofPhilosophyforthepermissiontoreusesomeofthatmaterial.<br />

ThankstoGrahamPriest,FrancescoPaoli,RossBrady,MaxCarrara,VeroTarca,Luca<br />

Illetterati,andDiegoMarconi,forhelpfulcomments,andtotheparticipantstoLogica<br />

2008forthelivelydiscussionofthetalkgiventhere.<br />

1 See(Berto&Priest,2008).


24 LiborBěhounek<br />

Thisseemstobedecidedlythecasewithstrongparaconsistency.When<br />

someoneclaimsthatboth Aandnot-Aaretrue,onewonderswhatismeant<br />

by“true”;and,ofcourse,by“not”:<br />

Thefactthatalogicalsystemtolerates Aand ∼Aisonlysignificant<br />

ifthereisreasontothinkthatthetildemeans‘not’. Don’twesay<br />

‘InAustralia,thewinterisinthesummer’,‘InAustralia,peoplewho<br />

standuprighthavetheirheadspointingdownwards’,‘InAustralia,<br />

mammalslayeggs’,‘InAustralia,swansareblack’?If‘InAustralia’<br />

canthusbehavelike‘not’(...),perhapsthetildemeans‘InAustralia’?<br />

2<br />

IsparaconsistentnegationjustanIn-Australiaoperator?Inathoroughly<br />

arguedessay,CatarinaDutilhNovaeshasrecentlysuggestedthat,critics<br />

notwithstanding,therealphilosophicalchallengeforparaconsistentlogics<br />

doesnotconsistinprovidingaplausibleaccountfornegation,butforthe<br />

notionofcontradiction. 3 Attackstoparaconsistencydeliveredbyclaiming<br />

thatparaconsistentnegationisnotnegation,accordingtoDutilhNovaes,<br />

“canbeneutralizedifitisshownthattheconflationbetweencontradiction<br />

andnegationisnotlegitimate,”andthat“paraconsistentnegationisin<br />

principleasrealanegationasanyother;” 4 for,asthenicesurveyofthe<br />

historyoflogicalnegationprovidedinherpapershows,thereisnounique<br />

realnegationaround. 5<br />

Onthecontrary,itisthenotionofcontradictionwhichspellstroublefor<br />

(strong)paraconsistentists.Theconceptofcontradiction“canbe[defined]<br />

withoutusingthenegation: Aand Barecontradictorypropositionsiff<br />

A ∨ Bholdsand A ∧ Bdoesnothold,regardlessoftheformof Aand B.”<br />

Therefore“contradictionisthepropertyofapairofpropositionswhich<br />

cannotbothbetrueandcannotbothbefalseatthesametime;”sincetwo<br />

propositionsthatarecontradictoriesaccordingtoclassicallogiccanboth<br />

betrueaccordingtothe(strong)paraconsistentist,oneconcludesthatthe<br />

lattersimplyrejectstheclassicalnotionofcontradiction.So“paraconsistent<br />

logiciansmustgiveanaccountofwhatcontradictionamountstowithina<br />

paraconsistentsystem.” 6<br />

Astrongparaconsistentistmayobjecttothecharacterizationofthenotionofcontradictionjustgiven,forthedefinitionusesanegationinthe<br />

definiens:contradictories“cannotbothbetrueandcannotbothbefalse.”<br />

Nowisthat“not”aclassicaloraparaconsistentnegation?(Strong)paraconsistentlogicianssuchasGrahamPriestprefertoassertthatnegation<br />

2 (Smiley,1993,p.17).<br />

3 See(DutilhNovaes,2007).<br />

4 (DutilhNovaes,2007,pp.479and482).<br />

5 Onthis,seealso(Wansing,2001).<br />

6 (DutilhNovaes,2007,pp.479and483).


FuzzyLogicsInterpretedasLogicsofResources 25<br />

isacontradictory-formingoperator,butdefinecontradictorinesswithout<br />

adoptinganegationinthedefiniens: Aand Barecontradictoriesiff,if A<br />

istruethen Bisfalse;andif Aisfalsethen Bistrue.Manyparaconsistent<br />

negations,then,turnouttobecontradictory-formingoperators;forinsuch<br />

logicsasLP(Priest’sLogicofParadox) 7 andFDE(BelnapandDunn’sFirst<br />

DegreeEntailment),negationactuallyisanoperatorthattruth-functionally<br />

switchestruthandfalsity:if Aistrue,then ¬Aisfalse;if Aisfalse,then<br />

¬Aistrue;if Aisbothtrueandfalse,then ¬Ais,too(andifthesemanticsadmitstruth-valuegaps,wemayalsohavethat<br />

Aisneithertruenor<br />

false;then, ¬Aisneithertruenorfalse,too). Nowthedebatehasbeen<br />

movedbacktothenotionsoftruthandfalsity:dotheyoverlap?Cansome<br />

truth-bearerbearboth?<br />

Ifwewanttohaveanon-question-beggingdebateondialetheiasand<br />

theLNC,insteadofconcentratingontruthandfalsitywemaygobackto<br />

negation.Or,atleast,thisisthewaypursuedinthispaper.Musttherebe<br />

auniquegoodaccountofnegation? Perhaps,asDutilhNovaesforcefully<br />

argues,not. Wemayhavedistinctintuitionsondifferentsententialand<br />

predicatenegations,whichmaybecharacterizedbydifferenttheories.This<br />

doesnotentail,though,thatnonon-question-beggingdebateisfeasible.On<br />

thecontrary,Ithinkitispossibletocharacterizeanegation(Ishalllabelit<br />

“NOT”)withthefollowingpleasantfeatures:<br />

1.itsdefinitiondoesnotrefertothecontentiousconcepttruth;<br />

2.ithasastrongpre-theoreticalmotivation,becauseofitsindispensable<br />

expressivefunctioninlanguageandcommunication;and<br />

3.itisfullyacceptedalsobydialetheists,becauseitisbasedonadeep<br />

metaphysicalintuitiontheyshowtofullyshare: theintuitionofexclusion.<br />

IfthecharacterizationofNOTproposedinthefollowingissufficientto<br />

conferadeterminatemeaningtothenegationinquestion,wecanconvenientlyphraseaformulationoftheLNCviasuchanegation.<br />

ThisLNC<br />

mightbeindisputablealsofromthedialetheist’spointofview. “Indisputable”shouldbeunderstoodinthefollowingsense:<br />

thedialetheistis<br />

forcedtoacceptit,withoutalsoacceptingsomethinginconsistentwithit.<br />

ItmightbeaversionoftheLNConwhichboththeorthodoxfriendand<br />

thedialetheicfoeofconsistencycanagreeinthissense.<br />

7 See(Priest,1979),(Priest,1987).


26 LiborBěhounek<br />

2 TheExclusionProblemandPriest’sPragmaticWayOut<br />

Iwillstartwithaproblemfacingstrongparaconsistency,whichhasbeen<br />

variouslyrecognisedintheliterature. Ibelieveittobethemaintheoreticaltroublefordialetheism,andIhaveelsewhereproposedtocallitthe<br />

ExclusionProblem. 8 Itgoesasfollows.<br />

Whenyousay:“A”,andadialetheistreplies:“¬A”,shemightnothave<br />

managedtoruleoutwhatyouhavesaid,preciselybecauseofthefeatures<br />

ofherparaconsistentnegation. Inthedialetheicframework, ¬Adoesnot<br />

ruleout Aonlogicalgrounds: itmaybethecaseboththat Aandthat<br />

¬A,sothedialetheistmayacceptthemboth.Alsosaying“Aisfalse,”and<br />

even“Aisnottrue,”neednotruleout Aonthedialetheist’sside.Inmany<br />

paraconsistentlogics,beginningwithLP,givenanysetofsentences S,it<br />

islogicallypossiblethateverysentenceof Sistrue. Thishappensinthe<br />

so-calledtrivialmodelofLP:ifallatomicsentencesarebothtrueandfalse,<br />

thenallsentences(truth-functionally)are. Inanutshell:nothingisruled<br />

outonlogicalgroundsonlyinthedialetheicframework.Manyauthorshave<br />

inferredthatdialetheismfacestheriskofendingupinexpressible. 9<br />

AccordingtoPriest,though,thesetroubleswithrulingoutthingscan<br />

besolvedbyturningintotherealmofpragmatics. Inordertohelpthe<br />

dialetheistruleoutsomething,hehasprovidedaninterestingtreatmentof<br />

thenotionofrejection. Letuscallacceptanceandrejectiontwomental<br />

statesasubject xhastowards(thepropositionexpressedby)asentence.<br />

Acceptanceandrejectionarepolaropposites:torejectsomethingistopositivelyrefusetobelieveit.<br />

Assertionanddenial,ontheotherhand,are<br />

(typically)linguisticactsor,equivalently,illocutionaryforcesattachedto<br />

utterances.Roughly,assertionanddenialarethelinguisticcounterpartsof<br />

acceptanceandrejection.Acceptanceandassertion,and,respectively,rejectionanddenial,areoftenconflatedbyphilosophers,andanywayformost<br />

ofourpurposeswecanrunlinguisticactsandthecorrespondingmental<br />

statestogether.Let’shavetwosententialoperators,“⊢x”and“⊣x”,whose<br />

readingis,respectively,“rationalagent xaccepts/asserts(that)”and“rationalagent<br />

xrejects/denies(that).” Thestandardtreatmenthasitthat<br />

rejection/denialisequivalenttotheacceptance/assertionofnegation:<br />

⊣x A ↔⊢x ¬A. (1)<br />

Ifweunderstanditintermsoflinguisticacts,(1)istheclaim,famously<br />

heldbyFregeandPeterGeach,accordingtowhichtodenysomethingjust<br />

istoassertitsnegation. ButPriestsaysthataccepting ¬Aisdifferent<br />

fromrejecting A: adialetheistcandotheformerandnotthelatter—<br />

8 See(Berto,2006),(Berto,2007,Ch.14).<br />

9 See,e.g.,(Parsons,1990),(Batens,1990),(Shapiro,2004).


FuzzyLogicsInterpretedasLogicsofResources 27<br />

exactlywhenshethinksthat Aisparadoxical.When Aisadialetheia,the<br />

naturalassumption(1)breaksdown,andnegationanddenialcomeapart.<br />

Adenial/rejectionof Abecomesanon-derivativementalorlinguisticact,in<br />

thatitisdirectlyaimedat A(oratthecontentof A,orattheproposition<br />

expressedby A,etc). 10<br />

Giventhat(1)canfail,thedialetheistcanacceptboth Aand ¬A,butshe<br />

doesnotneedtoacceptandreject A.Actually,accordingtoPriestshecannotevendothat:Priestconsidersacceptanceandrejectionasreciprocally<br />

incompatible,eventhough Aand ¬Aarenot:<br />

Someonewhorejects Acannotsimultaneouslyacceptitanymore<br />

thanapersoncansimultaneouslycatchabusandmissit,orwina<br />

gameofchessandloseit.Ifapersonisaskedwhetherornot A,he<br />

canofcoursesay‘Yesandno’.Howeverthisdoesnotshowthathe<br />

bothacceptsandrejects A. Itmeansthatheacceptsboth Aand<br />

itsnegation.Moreoverapersoncanalternatebetweenacceptingand<br />

rejectingaclaim. Hecanalsobeundecidedastowhichtodo. But<br />

dobothhecannot. 11<br />

Andthisishowthedialetheistcanmanagetoruleoutsomething,and<br />

toexpressthis.Althoughtheshecannotruleout Abysimplysaying“¬A”,<br />

shecanreject A.Sothepragmaticincompatibilityofacceptance/assertion<br />

andrejection/denialplaysapivotalroleinPriest’sreplytotheExclusion<br />

Problem.<br />

3 NOT<br />

Thisshowsthatevendialetheistshaveanintuitionofexclusion,orincompatibilitybetweensomethingandsomethingelse.SoIproposetosearchfor<br />

anoperator(arguably,anegation)thatallowsustocaptureandexpressthe<br />

intuition.Weneedtostartfromthisverynotionpreciselybecausewewant<br />

toavoidexplicitlyemployingtheconceptsoftruthandfalsitytocharacterizesuchanoperator.Thedialetheistcastsdoubtsontheirbeingexclusive,<br />

bypointingoutthatsometruth-bearers,notably,theLiars,fallunderboth<br />

conceptssimultaneously.Truthtablesortruthconditionsfornegationcan<br />

giveusnosenseoftheconnectionbetweennegationandexclusionunless<br />

wealreadysharetheintuitionthattruthandfalsityruleouteachother.<br />

Thisbringsusbacktotheissueofthenotionofcontradiction,raisedby<br />

DutilhNovaes.Specifically,weshouldrefrainfromexpressingexclusionvia<br />

thetraditionalconceptofcontrariness: defining Aand Bascontrariesiff<br />

“A∧B”islogicallyfalsewon’thelpwhendiscussingwiththedialetheist.But<br />

10 See(Priest,2006,p.104).<br />

11 (Priest,1989,p.618).


28 LiborBěhounek<br />

onemaytrywiththeintuitivenotionofexclusionitself,takenasprimitive.<br />

Animalsandinfantsperceiveincompatibilitiesintheworldlongbeforethey<br />

havedevelopedormasteredanarticulatedlanguagetoexpressthem.One<br />

oftheusesofalinguisticitemthatcountsasanegationcantherefore<br />

beexplained,asHuwPricehasclaimed,as“initiallyameansofregistering<br />

(publiclyorprivately)aperceivedincompatibility.”Andif“incompatibility<br />

[is]averybasicfeatureofaspeaker’s(orproto-speaker’s)experienceofthe<br />

world,” 12 thenonecanexplainthenegationwearelookingforinterms<br />

ofincompatibility. Weonlyneedtoassumethatordinaryspeakersand<br />

rationalagentshavesomeacquaintancewithexclusions—thingsofthe<br />

worldrulingouteachother:theycanrecognizethemintheworld,andin<br />

theircommercewiththeworld.<br />

Ishalltalkofmaterialexclusionor,equivalently,ofmaterialincompatibility.<br />

Onemaycharacterizeitintermsofconcepts,properties,states<br />

ofaffairs,propositions,orworlds,dependingonone’smetaphysicalpreferences.<br />

13 Materialexclusionbearsthisnametostressthefactthatitisnot<br />

amerelylogical,inthesenseofformal,notion: itisbasedonthematerialcontentoftheinvolvedconcepts,orproperties,etc.<br />

Someexamples:<br />

phenomenologicalcolourincompatibilities,suchasbeing(solidly)Redand<br />

being(solidly)Green;conceptsthatexpressourcategorizationofphysical<br />

objectsinspaceandtime,suchas xbeinghererightnowand xbeingway<br />

overthererightnow,forasuitablysmall x. 14 Or xbeinglessthantwo<br />

incheslongand xbeingmorethanthreefeetlong. 15 ButalsoPriest’sabove<br />

x’scatchingthebusand x’smissingthebuswilldo.<br />

Ok,thiswastheintuition. Howdoweformalizeit? Afeasibleformal<br />

accountmayadapttheideadevelopedbyMichaelDunnthat“onecan<br />

definenegationintermsofoneprimitiverelationofincompatibility(...)in<br />

ametaphysicalframework.” 16 Soletustalkintermsofpropositions(that<br />

whichisexpressedbyasentence)andbuildasmallalgebra. Thinkofa<br />

structure 〈U, ⊂,V, •〉where Uisasetofpropositions; ⊂and •arebinary<br />

relationsdefinedon U;and Visaunaryoperationonsubsetsof U. ⊂isto<br />

bethoughtofasapre-order,and“p ⊂ q”canbereadas“Theproposition<br />

pentailstheproposition q”.Givenasetofpropositions P ⊆ U,VPisthe<br />

12 (Price,1990,pp.226–228).<br />

13 Forinstance,wemayviewitastherelationthatholdsbetweenacoupleofproperties P1<br />

and P2iff,byhaving P1,anobjecthasdismissedanychanceofsimultaneouslyhaving P2.<br />

Orwemayalsoclaimthatmaterialincompatibilityholdsbetweentwoconcepts C1and<br />

C2ifftheveryinstantiating C1byaputsabaronthepossibilitythataalsoinstantiates<br />

C2.Orwemaysaythatitholdsbetweentwostatesofaffairs S1and S2ifftheholding<br />

of S1precludesthepossibilitythat S2alsoholds(inworld w,attime t,etc.).<br />

14 See(Tennant,2004,p.362).<br />

15 See(Grim,2004,p.63).<br />

16 (Dunn,1996,p.9).


FuzzyLogicsInterpretedasLogicsofResources 29<br />

(possiblyinfinitary:moreonthissoon)disjunctionofallthepropositions<br />

in P.And •ispreciselyourprimitiverelationofmaterialexclusion.<br />

Apropositionmayhaveoneormoreincompatiblepeers: itmayrule<br />

outawholeassortmentofalternatives. PatrickGrim,forinstance,talks<br />

abouttheexclusionaryclassofagivenproperty.Theexclusionaryclassof<br />

aproposition p,then,istheset<br />

E = {x;x • p}.<br />

Then NOT-pisnothingbutVE. If Ehasfinitecardinality,then NOT-p<br />

isjustanordinarydisjunction: q1 ∨ · · · ∨ qnwhere q1,... ,qnareallthe<br />

membersof E.<br />

Supposethereareinfinitelymanypropositionsincompatiblewith p.This<br />

isaheavymetaphysicalassumption,butletusgrantit.Then, NOT-pturns<br />

outtobeaninfinitarydisjunction. Ifonehas(understandable)problems<br />

withinfinitarydisjunctions,wecannotavoidquantifyingonpropositions:<br />

NOT-p =df ∃x(x ∧ x • p). (2)<br />

Bothinthefinitaryandinfinitarycase,itisclearinwhichsense NOT-p<br />

isthelogicallyweakestamongthe nincompatibles:itisentailedbyany qi,<br />

1 ≤ i ≤ n,suchthat qi • p. Onemayexpressthepointviathefollowing<br />

equivalence:<br />

x ⊂ NOT-p iff x • p. (3)<br />

Putting NOT-pfor x,andbydetachment,weget:<br />

NOTp • p, (4)<br />

NOT-pisincompatiblewith p. Theright-to-leftdirectionof(3),then,<br />

tellsusthat NOT-pistheweakestincompatible,i.e.,itisentailedbyany<br />

incompatibleproposition. 17<br />

Whichlogicshouldbereadoffthealgebradependsonwhichalgebraic<br />

postulateswewanttoadd.Dependingonthechoiceswemake, NOTwillbecomepalatableforsomelogicians,eventhoughotherswillbedisappointed.<br />

Onemayassume,reasonablyenough,that •issymmetric. Butifinthe<br />

algebraicframework NOTisstipulatedasanoperationofperiodtwo,i.e.<br />

NOT-NOT-p = p, (5)<br />

thisislikelytoberejectedbyanintuitionist,thoughnotbymanyparaconsistentlogicians.TheintuitionistmayalsoobjecttothefactthatNOThas<br />

17 Variationsonthethemeofthecharacterizationofnegationviaincompatibility,and<br />

onnegationastheminimalincompatible,canbefoundin(Brandom,1994,pp.381ff.);<br />

(Harman,1986,pp.118-20);(Peacocke,1987).


30 LiborBěhounek<br />

beendefinedusingotheroperators,whichgoesagainsttheindependenceof<br />

logicalconstantsinaconstructivistframework(aremarkIowetoFrancesco<br />

Paoli).Or,ifwemaketheprimafacienaturalassumptionthat:<br />

If p ⊂ qand x • q, then x • p, (6)<br />

wecaneasilygetcontraposition. Butsucharesultwouldberejectedby<br />

thosewhowanttodismisscontrapositiononthebasisofconsiderationson<br />

theconditional,andalsobysomeparaconsistentists.Differentphilosophicalparties(classicists,intuitionists,paraconsistentists,etc.)<br />

haveopposed<br />

viewsonwhatnegationis,whereastheaimhereistoprovideanintuitive<br />

depictiononwhichallpartiescanagree;thisiswhyIfindformalization<br />

usefulonlytoacertainextent.<br />

4 MinimalLNC<br />

Independentlyofthepossibleadditionalcharacterization, NOThassome<br />

nicefeatures.First,isnotexplicitlydefinedviatheconcepttruth.AsGrahamPriesthaspointedouttome(incommunication),thismaynotprevent<br />

truthfromjumpinginagain.Ihavebeenforcedtoadmitthat,givensome<br />

(albeitdebatable)metaphysicalassumptions,wemayneedpropositional<br />

quantificationtospelloutthedetailsof NOT. Andsuchquantificationis<br />

inter-definablewithtruth. Butwhat NOTisexplicitlyreferredtoisthe<br />

conceptexclusion,whoseprimitivenesshasbeenarguedforabove.<br />

Secondly, NOThasastrongpre-theoreticalappealasanexclusion-expressingtool:<br />

itallowsustoruleoutthingsbyclaimingthatsomething<br />

incompatiblewiththemisgoingon.Thisiswhatatleastsomeoftheitems<br />

wequalifyasnegationshouldhelpustodo.<br />

Finally,dialetheistsgraspthenotionofexclusion. Theyaskustostop<br />

using“not”or“true”asexclusion-expressingdevices,because“not-A”is<br />

insufficientbyitselftoruleout A,and“Aistrue”isinsufficientbyitselfto<br />

ruleoutthat Aisalsofalse.ButPriest’saccountofacceptanceandrejection<br />

showsthatthedialetheistbelievesintheimpossibilityofsomecouplesof<br />

facts’,orstatesofaffairs’,simultaneouslyobtaining;or,equivalently,that<br />

sheassumesthatsomethingsmateriallyexcludesomeothers: x’ssimultaneouslycatchingandmissingthebus,forinstance;and,ofcourse,<br />

x’s<br />

simultaneouslyacceptingandrejectingthesame A,thisbeing,aswehave<br />

seen,abasicstepinPriest’sanswertotheExclusionProblem. NOTis<br />

supposedtoworkeveninaframeworkinwhichnothingisruledouton<br />

logicalgroundsalone,becauseitisnotmerelylogically,i.e.formally,but<br />

metaphysically(“materially”)founded. Thedialetheistmayhaveavacuousnotionoflogical,formalincompatibility.Butshedoeshaveanotionof<br />

materialincompatibility.


FuzzyLogicsInterpretedasLogicsofResources 31<br />

Nowforthefinalstep: expresstheLNCvia NOT. TakeAristotle’s<br />

traditionalformulationoftheLNC,inBook ΓoftheMetaphysics,andjust<br />

putinitour NOT. Theformulationcanbetakenasadefinitionof“the<br />

impossible”:<br />

ForthesamethingtoholdgoodandNOTholdgood<br />

simultaneouslyofthesamethingandinthesamerespect<br />

isimpossible. 18<br />

“P1does NOTholdgoodof x”shouldbeashortformfor“to xbelongs<br />

someproperty P2,whichismateriallyincompatiblewith P1.” Thisdoes<br />

notseemtobequestionablebythedialetheistanymore,providedshehas<br />

understood NOT—andtounderstand NOTistounderstandexclusion.<br />

Ifthedialetheistrefusestosubscribetothecharacterizationof NOTvia<br />

theintuitivenotionofexclusion,sheseemstoactuallyendupasunableto<br />

expresstheexclusionofanyposition(isshetryingtoexcludeexclusion?).<br />

AndadialetheismwithouttheLNCstatedintermsof NOTlooksvery<br />

muchlikeatrivialism(ItotallyagreewithDutilhNovaes,whopressesa<br />

pointverysimilartothisoneinheressay). 19 SuchaLNC,touseAristotle’s<br />

words,is“aprinciplewhicheveryonemusthavewhoknowsanythingabout<br />

being.” 20<br />

DoesthisreplytoDutilhNovaes’challengefor(strong)paraconsistency,<br />

namely,thatofdefining“P-contradictions,thatis,contradictionsthatare<br />

sothreateningtoatheorythattheyreallycompromiserationalinferencemakingwithinit”?<br />

21 Tosomeextent,yes—ifthestepsoftheargumentationproposedabovework.<br />

ButthesuccessfortheversionoftheLNC<br />

phrasedintermsofNOTisverylimited:forthatLNCsimplyrulesoutthe<br />

simultaneousobtainingofreciprocallyexclusionarystatesofaffairs. The<br />

questionremainsopenofwhicharetheexclusionarystatesofaffairs(or<br />

properties,etc.). Andnowthediscussionbetweendialetheistsandantidialetheistscandevelopwithasignificantdecreaseinissuesofquestionbeggingandclashesofintuitions.Whatisincompatiblewithwhat?Given<br />

twoproperties P1and P2,thequestionwhethertheyareexclusivecaninvolvebroadlyempiricalmatters,difficultanalysesofourconceptualtoolkit<br />

and/orofouruseofordinarylanguageexpressions.Somecasesmaybeeasy<br />

toresolve;butothersmayproducebattlesofintuitions:areyoungandold<br />

actuallyexclusive?Blueandgreen?Trueandfalse?Circularandsquare?<br />

Ihaveclaimedthatmaterialexclusionisbasedonthecontentoffacts,concepts,orproperties;buthowdoweknowwhatthecontentofaconcept<br />

18 SeeAristotleMet.1005b18–21(Aristotle,1984).<br />

19 See(DutilhNovaes,2007,p.487)).<br />

20 Arist.Met.1005b14–15(Aristotle,1984).<br />

21 (DutilhNovaes,2007,p.489).<br />

(7)


32 LiborBěhounek<br />

is,orwhicharetheactualfieldsofapplicationsofaproperty?Theformal<br />

characterizationof NOT,ofcourse,doesnotentailspecialcommitmentson<br />

whicharethespecificproperties,orconcepts,orstatesofaffairs,between<br />

whichitholds.Suchcommitmentsarefallible.Wecancometobelievethat<br />

someproperties,orconcepts,orstatesofaffairs,areincompatible,andthen<br />

findoutthattheyarenot. Wouldthisentailexplosion,thatis,anything<br />

beingderivable,andtrivialism? Well,not: thestandardstrategyinthis<br />

caseissimplytoretractourpreviousassumptionthattheywere.<br />

Sothedialetheistwhohasnotroubleswithourminimal(7)canstill<br />

objecttootherformulationsoftheLNC,e.g.,becausetheyarephrased<br />

intermsoftruthandfalsity: thosewhoruleoutthatanysentencecould<br />

bebothtrueandfalsetaketruthandfalsityasexclusionaryconcepts;the<br />

dialetheisthasqualmsonthis,andperhapscounterexamplestooffer(say,<br />

theLiarsentences). Buttheissueaddressedhereiswhetherallconcepts<br />

(orproperties,etc.) arelikethat;andthedialetheistagreesthatsome<br />

concepts(orproperties,etc.) doruleouteachother. Thisistheshared,<br />

basicintuition NOTappealsto.<br />

FrancescoBerto<br />

DepartmentofPhilosophyandTheoryofSciences,UniversityofVenice-<br />

Ca’Foscari<br />

Dorsoduro3484D,30123Venice,Italy<br />

bertofra@unive.it<br />

References<br />

Aristotle. (1984). Metaphysics. InJ.Barnes(Ed.),TheCompleteWorksof<br />

Aristotle(Vol.2).Princeton,N.J.:PrincetonUniversityPress.<br />

Batens,D. (1990). Againstglobalparaconsistency. StudiesinSovietThought,<br />

39,209–229.<br />

Berto,F.(2006).Meaning,metaphysics,andcontradiction.AmericanPhilosophicalQuarterly,43,283–297.<br />

Berto,F. (2007). HowtoSellaContradiction.TheLogicandPhilosophyof<br />

Inconsistency.London:CollegePublications.<br />

Berto,F. (2008). Adynatonandmaterialexclusion. AustralasianJournalof<br />

Philosophy,86,165–190.<br />

Berto, F., & Priest, G. (2008). Dialetheism. In The Stanford<br />

Encyclopedia of Philosophy. Stanford, CA: CSLI. Available from<br />

http://plato.stanford.edu/entries/dialetheism/<br />

Brandom,R. (1994). MakingitExplicit. Cambridge,MA:HarvardUniversity<br />

Press.


FuzzyLogicsInterpretedasLogicsofResources 33<br />

Dunn,J.(1996).Generalizedorthonegation.InH.Wansing(Ed.),(pp.3–26).<br />

Berlin–NewYork:DeGruyter.<br />

DutilhNovaes,C. (2007). Contradiction: Therealphilosophicalchallengefor<br />

paraconsistentlogic.InJ.Béziau,W.Carnielli,&D.Gabbay(Eds.),Handbook<br />

ofParaconsistency(pp.477–492).London:CollegePublications.<br />

Grim,P.(2004).Whatisacontradiction? InG.Priest,J.Beall,&B.Armour-<br />

Garb(Eds.),(pp.49–72).Oxford:ClarendonPress.<br />

Harman,G.(1986).Changeinview.Cambridge:MITPress.<br />

Parsons,T. (1990). Truecontradictions. CanadianJournalofPhilosophy,20,<br />

335–354.<br />

Peacocke,C.(1987).Understandinglogicalconstants:arealist’saccount.ProceedingsoftheBritishAcademy,73,153–200.<br />

Price,H.(1990).Why’Not’?Mind,99,221–238.<br />

Priest,G. (1979). Thelogicofparadox. JournalofPhilosophicalLogic,8,<br />

219–241.<br />

Priest,G. (1987). Incontradiction: astudyofthetransconsistent. Dordrecht:<br />

MartinusNijhoff.(2nd,extended,edn,Oxford:OxfordUniversityPress,2006.)<br />

Priest,G.(1989).Reductioadabsurdumetmodustollendoponens.InG.Priest,<br />

R.Routley,&N.J.(Eds.),ParaconsistentLogic.EssaysontheInconsistent(pp.<br />

613–626).München:PhilosophiaVerlag.<br />

Priest,G.(2006).DoubtTruthtobeaLiar.Oxford:OxfordUniversityPress.<br />

Priest,G.,Beall,J.,&Armour-Garb,B.(Eds.). (2004). TheLawofNon-<br />

Contradiction.NewPhilosophicalEssays.Oxford:ClarendonPress.<br />

Shapiro,S. (2004). Simpletruth,contradiction,andconsistency. InG.Priest,<br />

J.Beall,&B.Armour-Garb(Eds.),(pp.336–354).Oxford:ClarendonPress.<br />

Smiley,T.(1993).Cancontradictionsbetrue?I.ProceedingsoftheAristotelian<br />

Society,67,17–34.<br />

Tennant,N.(2004).Ananti-realistcritiqueofdialetheism.InG.Priest,J.Beall,<br />

&B.Armour-Garb(Eds.),(pp.355–384).Oxford:ClarendonPress.<br />

Wansing,H.(Ed.).(1996).Negation.ANotioninFocus.Berlin–NewYork:De<br />

Gruyter.<br />

Wansing,H. (2001). Negation. InL.Goble(Ed.),TheBlackwellGuideto<br />

PhilosophicalLogic(pp.415–436).Oxford:Blackwell.


Medieval<strong>Obligationes</strong> as a Regimentation of<br />

‘the Game of Giving and Asking for Reasons’<br />

1 Introduction<br />

Catarina Dutilh Novaes ∗<br />

Medievalobligationesdisputationswereahighlyregimentedformoforaldisputationopposingtwoparticipants,respondentandopponent,andwhere<br />

inferentialrelationsbetweensentencestookprecedenceovertheirtruthor<br />

falsity.In(DutilhNovaes,2005),(DutilhNovaes,2006)and(DutilhNovaes,<br />

2007,Ch.3)Ipresentedaninterpretationofobligationesaslogicalgamesof<br />

consistencymaintenance;thisinterpretationhadmanyadvantages,inparticularthatofcapturingthegoal-oriented,rule-governednatureofthiskind<br />

ofdisputationbymeansofthegameanalogy.Italsoexplainedseveralofits<br />

featuresthatremainedotherwisemysteriousinalternativeinterpretations,<br />

suchastheroleofimpertinentsentencesandwhy,whilethereisalwaysa<br />

winningstrategyforrespondent,thegameremainshardtoplay.However,<br />

thelogicalgameinterpretationdidnotprovideafullaccountofthedeontic<br />

aspectofobligationes—ofwhatbeingobligedtoacertainstatementreallyconsistsin—beyondthegeneral(andsuperficial)commitmenttowards<br />

playing(andwinning)agame.Afterall,theverynameinvokesnormativity,<br />

soaninterpretationofobligationesthatdoesnotfullyaccountforthedeonticcomponentseemstobemissingacrucialaspectofthegeneralspiritof<br />

theenterprise.Inordertoamendthisshortcominginmypreviousanalysis<br />

Iherepresentanextensionofthegame-interpretationbasedonthenotion<br />

of‘thegameofgivingandaskingforreasons’—henceforth,GOGAR 1 —<br />

presentedinChapter3ofR.Brandom’sMakingitExplicit(Brandom,1994)<br />

asconstitutingtheultimatebasisforsociallinguisticpractices.Thebasic<br />

∗ ThankstoEdgarAndrade-LoteroandOleThomassenHjortlandforcommentsonan<br />

earlierdraftofthepaper.<br />

1 FollowingJ.MacFarlane’sterminology,cf.http://johnmacfarlane.net/gogar.html.


36 CatarinaDutilhNovaes<br />

ideaisthatobligationescanbeseenasaregimentationofsomeofthecore<br />

aspectsofGOGAR.<br />

What is to be gained from a comparison between obligationes and<br />

GOGAR?Fromthepointofviewofthelatter,thecomparisoncanshed<br />

lightonitsgenerallogicalstructure:ifobligationesreallyarearegimentationofGOGAR,thentheycancertainlycontributetomakingitsstructure<br />

explicit(whichisofcourseanothercrucialelementofBrandom’sgeneral<br />

enterprise).Indeed,anobligatioissomethingofaSprachspielforGOGAR,<br />

asimplifiedmodelwherebysomeofGOGAR’spropertiescanbemademanifest.Asforobligationes,whatcanbegainedfromthecomparison,besidestheemphasisonitsfundamentallydeonticnature,isabetterunderstandingofitsgeneralpurpose.<br />

Atfirstsight,thishighlyregimentedformof<br />

disputation,wheretruthdoesnotseemtohaveanymajorroletoplay,may<br />

seemlikesterilescholasticlogicalgymnastics. ButifitisputinthecontextofGOGAR—which(presumably)capturestheessenceofoursocial,<br />

linguisticandrationalbehaviors—thenitssignificancewouldappeartogo<br />

wellbeyondthe(mere)developmentoftheabilitytorecognizeinferential<br />

relationsandtomaintainconsistency.<br />

2 GOGAR<br />

AcrucialelementofthephilosophicalsystempresentedbyBrandominMakingitExplicit(andfurtherexpandedinseveralofhissubsequentwritings)is<br />

themodeloflanguageusethathereferstoas‘thegameofgivingandasking<br />

forreasons’.Brandominsiststhatlanguageuseandlanguagemeaningfulnesscanonlybeunderstoodinthecontextofsocialpracticesarticulating<br />

informationexchangeandactions—linguisticspeech-acts(typically,the<br />

makingofaclaim)aswellasnon-linguisticactions.<br />

Infact,GOGARshouldaccountforwhatmakesussocial,linguisticand<br />

rationalanimals. AsBrandomconstruesit,GOGARisfundamentallya<br />

normativegameinthattheproprietyofthemovestobeundertakenby<br />

theparticipantsisatthecentralstage.Itis,however,notatranscendental<br />

kindofnormativity,requiringanalmightyjudgeoutsidethegametokeep<br />

trackofthecorrectnessofthemovesundertaken;rather,theparticipants<br />

themselvesareinchargeofevaluatingwhetherthemovesundertakenare<br />

appropriate. Itisa“deonticscorekeepingmodelofdiscursivepractice”.<br />

InGOGAR,weareallplayers(speakers)andscorekeepersconcomitantly;<br />

weundertakemovesandkeeptrackofeverybody’smoves(includingour<br />

own)atthesametime. Thefocuson(givingandaskingfor)reasonsis<br />

animportantaspectofhowthemodelcapturestheconceptofrationality:<br />

weareresponsiblefortheclaimswemake,andthusmustbepreparedto


<strong>Obligationes</strong>asGOGAR 37<br />

providereasons 2 forthemwhenchallenged. Underlyingthisfactisthe<br />

ideaofalogicalarticulationofcontentssuchthatsomecontentscountas<br />

appropriatereasonsforothercontents.<br />

Inprinciple,asageneralmodeloflanguageuse,GOGARshouldencompassalldifferentkindsofspeech-acts:<br />

assertions,questions,butalso<br />

promises,orders,expressionsofdoubtetc.However,forBrandomthereis<br />

onefundamentalkindofspeech-actinthegame,namelythatofmaking<br />

anassertion. 3 Anassertionisbothsomethingthatcancountasareason<br />

(ajustification)foranotherassertionandsomethingthatmayconstitute<br />

achallenge—typically,whenaspeakerSmakesanassertionincompatiblewithsomethingpreviouslysaidby<br />

T—andthusprovoketheneedfor<br />

furtherreasons(Tmustdefendtheoriginalassertion): hence,givingand<br />

askingforreasons.<br />

Theneedtodefendone’sassertionsthreatenedbychallengesthrough<br />

furtherreasonsindicatesthatoneissomehowresponsibleforone’sassertions.ThisisindeedthecaseaccordingtotheGOGARmodel,andthisfact<br />

isaccountedforbytheabsolutelycrucialconceptofdoxasticcommitment.<br />

Justasapromisecreatesthecommitmenttofulfillwhathasbeenpromised,<br />

themakingofanassertioncreatesthecommitmenttodefendit,i.e.,tohave<br />

hadgoodreasonstomakeit.Thisisbecauseoneoftenreliesontheinformationconveyedbyanassertionmadebyanotherpersoninordertoassessaparticularsituationandthenactupontheassessment;butiffalseinformationistransmitted,thentheassessmentwillprobablybemistaken,and<br />

theactioninquestionwillprobablynothavethedesiredoutcome;itmay<br />

evenhavedeleteriousconsequencesfortheagent.Insuchcases,itisfairto<br />

saythatthepersonhavingconveyedtheincorrectinformationisresponsible<br />

fortheinfelicitousoutcome,justasarecklessdriverisresponsibleforthe<br />

accidentshe/she(directlyorindirectly)causes.Ifsomebodyshouts‘fire!’as<br />

aprankinacompletelyfullstadium,forexample,thiswillprobablycause<br />

considerablemayhem,andtheinfelicitousjokerwillbeheldaccountablefor<br />

allthedamagecaused.Sogiventhepotentialpracticalconsequencesofan<br />

assertion,itisnotsurprisingatallthatliabilityshouldbeinvolvedinthe<br />

makingofanassertion.<br />

ForBrandom,thecommitmenttothecontent 4 ofanassertioninfactgoes<br />

beyondtheassertionitself:oneisalsocommittedtoeverythingthatfollows<br />

fromtheoriginalassertion,i.e.,everythingthatcanbeinferredfromit.The<br />

2 Etymologically,rationalitycomesfromratio,‘reason’inLatin.<br />

3 “Thefundamentalsortofmoveinthegameofgivingandaskingforreasonsismaking<br />

aclaim—producingaperformancethatispropositionallycontentfulinthatitcanbe<br />

theofferingofareason,andreasonscanbedemandedforit.”(Brandom,1994,p.141).<br />

4 ItisnotentirelycleartomethoughwhetherBrandomseescommitmentsashaving<br />

contentsorsentencesorclaimsastheirobjects,butitseemstomethatcontentswould<br />

bethemostappropriateobjectsofcommitments.


38 CatarinaDutilhNovaes<br />

inferentialrelationsbetweenassertionsareaprimitiveelementofBrandom’s<br />

system(codifiedintermsofmaterialinferences,notformalones);ashe<br />

sometimessays,theyare“unexplainedexplainers”(Brandom,1994,p.133).<br />

Materialinferencesarepainstakinglydiscussedin(Brandom,1994,Ch.2),<br />

butforourpurposeswhatisimportantistorealizethatcommitmenttoa<br />

contenttransfersovertoothercontentsbymeansofinferentialrelations.<br />

Butbesidesbeingcommittedtocontents,thereisanotherprimitivedeonticstatusthataspeakermayormaynotenjoywithrespecttocontents:<br />

entitlement.Fromthepointofviewofascorekeeper, 5 foraspeaker Stobe<br />

entitledtoassertingagivencontentamountsto Sbeinginthepositionto<br />

offergroundsthatjustifybeliefinthecontent,andthusthemakingofthe<br />

correspondingassertion;thisdeonticstatusisattributedwhenthespeaker<br />

hasgood(enough)reasonstobelievethecontenttobethecase.Brandom<br />

remarksthat“commitmentandentitlementcorrespondtothetraditional<br />

deonticprimitivesofobligationandpermission”(Brandom,1994,p.160);<br />

herejectsthisterminologybecausehewishestoavoidthestigmataofnorms<br />

associatedwithhierarchyandcommands(asnotedabove,thescorekeeping<br />

isdonehorizontallybyallparticipants). Butultimately,acommitmentis<br />

indeedanobligation,andanentitlementisindeedapermission,andthus<br />

beingcommittedtoacontentamountstobeingobligedtoitinexactlythe<br />

samesenseofbeingobligedduringanobligatiodisputation(asweshall<br />

see):onehasadutytowardsacertaincontent,whichtransfersovertoall<br />

thecontentsthatfollowfromit.<br />

Fromthetwoprimitiveconceptsofcommitmentandentitlement,Brandomderivestheequallyimportantconceptofincompatibility:<br />

content p<br />

beingincompatiblewithcontent qamountstocommitmentto pprecluding<br />

entitlementto q.Itisnotsomuchthatitisfactuallyimpossibleforoneto<br />

becommittedto pwhilebelievingoneselftobeentitledto q;thiscanoccur,<br />

justasonecanmakeconflictingpromisesandholdinconsistentbeliefs.But<br />

again,thisisamatterofdeonticscorekeeping: fromthepointofviewof<br />

thescorekeepers,ifaspeakeriscommittedto pthereisawholeseriesof<br />

contents q, t,etc.towhichthespeakerinquestionissimplynotentitledas<br />

longashemaintainshiscommitmentto p.Butifheneverthelessinsistsin<br />

beingcommittedto pandentitledto qatthesametime,thenheissimply<br />

makingabadmovewithinGOGAR.<br />

Brandomcorrectlynoticesthatincompatibility,asmuchasentailment,<br />

isessentiallyarelationbetweensetsofcontents,notbetweencontentsthem-<br />

5 Thedeonticstatusesofcommitmentandentitlementarealwaysperspectival,i.e.definedbythedeonticattitudesof(self-)attributingcommitmentsandentitlementsofeach<br />

scorekeeper.“Suchstatusesarecreaturesofthepracticalattitudesofthemembersofa<br />

linguisticcommunity—theyareinstitutedbypracticesgoverningthetakingandtreating<br />

ofindividualsascommitted.”(Brandom,1994,p.142).


<strong>Obligationes</strong>asGOGAR 39<br />

selves. 6 Takeasetofthreecontents,e.g.,thoseexpressedbythesentences<br />

‘Everymanisrunning’,‘Socratesisaman’and‘Socratesisnotrunning’.<br />

Commitmenttoeitheroneofthetwofirstcontentsalonedoesnorprecludeentitlementtothethirdcontent,butcommitmenttobothofthem<br />

doesprecludeentitlementtothethird,justascommitmenttothefirsttwo<br />

contentssimultaneouslyentailscommitmenttothecontent‘Socratesisrunning’.<br />

Thisaspectwillbesignificantforthecomparisonwithobligationes<br />

lateron,asithintsatthefundamentallydynamicnatureoftheGOGAR<br />

model:everynewassertionmaderequirestherecalibrationofeverybody’s<br />

deonticstatusesbythescorekeepers—oftheasserter,inparticular,butin<br />

factofeverybodyelseaswell,asGOGARalsoaccountsforinter-personal<br />

transmissionofentitlementbytestimony.Inotherwords,aspeaker’sdeonticstatus—hercommitmentsandentitlements—ismodifiedeverytimean<br />

assertionismade,moresalientlybutnotexclusivelybythespeakerherself.<br />

Indeed,thereseemtobefourmainsourcesofentitlementaccordingto<br />

theGOGARmodel.<br />

1.Interpersonal,intracontentdeferentialentitlement: Speaker 1isentitledto(asserting)content<br />

pbecausespeaker 2,areliablesource,<br />

asserted p.<br />

2.Intrapersonal,intercontentinferentialentitlement: Speaker 1isentitledto(asserting)<br />

qbecausesheisentitledto(asserting) pand p<br />

entails q.<br />

3.Perception:Speaker 1isentitledto(asserting) pbecauseshehashad<br />

a(reliable)perceptualexperiencecorrespondingto p.<br />

4.Defaultentitlement: ‘freemoves’,thecontentsentitlementtowhich<br />

issharedbyallspeakersinsofarasthesecontentsconstitutecommon<br />

knowledge—everybodyknowsit,andeverybodyknowsthateverybodyknowsit.<br />

AfinalpointIwishtoaddressinmybriefpresentationofGOGAR<br />

isthenotionofinference,morespecificallymaterialinference. Brandom<br />

criticizestheformalistviewofinference,accordingtowhicheveryvalid<br />

inferenceisaninstanceofaformallyvalidschema;rather,theinferential<br />

relationsthataretheprimitiveelementsofhisinferentialsemanticsareofa<br />

conceptualnature,whilealsofirmlyembeddedinpractices:“Inferringisa<br />

kindofdoing.”(Brandom,1994,p.91)Thefocusonthenotionofmaterial<br />

inferencealsoechoesimportantfeaturesofobligationes,asinthelatter<br />

6 (Brandom,2008,Lect.5). Thecasesofrelationsinvolvingsinglecontentscanbeseen<br />

aslimit-cases,relatingsingletonsets.


40 CatarinaDutilhNovaes<br />

frameworktherelationof‘following’(sequitur)inquestionisnotrestricted<br />

toformallyvalidschemata. 7<br />

3 Medievalobligationes<br />

Anobligatiodisputationhastwoparticipants,OpponentandRespondent.<br />

Inthecaseofpositio,themostcommonandwidelydiscussedformofobligationes,thegamestartswithOpponentputtingforwardasentence,usuallycalledthepositum,whichRespondentmustacceptforthesakeofthedisputation,unlessitiscontradictoryinitself.<br />

Opponentthenputsforward<br />

othersentences(theproposita), oneatatime,whichRespondentmust<br />

eithergrant,denyordoubtonthebasisofinferentialrelationswiththe<br />

previouslyacceptedordeniedsentences—or,incasetherearenone(and<br />

thesearecalledimpertinent 8 sentences)onthebasisofthecommonknowledgesharedbythosewhoarepresent.Inotherwords,ifRespondentfails<br />

torecognizeinferentialrelationsorifhedoesnotrespondtoanimpertinent<br />

sentenceaccordingtoitstruth-valuewithincommonknowledge,thenhe<br />

respondsbadly.Respondent‘losesthegame’ifheconcedesacontradictory<br />

setofpropositions.ThedisputationendsifandwhenRespondentgrantsa<br />

contradiction,orelsewhenOpponentsays‘cedattempus’,‘timeisup’.Opponentandpossiblyalargerpanelofmasterspresentatthedisputationare<br />

inchargeofkeepingtrackofRespondent’srepliesandofevaluatingthem<br />

oncethedisputationisover.<br />

Anobligatiodisputationcanberepresentedbythefollowingtuple:<br />

Ob = 〈KC,Φ,Γ,R(φn)〉<br />

KCisthestateofcommonknowledgeofthosepresentatthedisputation.<br />

Φisanorderedsetofsentences,namelytheverysentencesputforward<br />

duringthedisputation. Γisanorderedsetofsetsofsentences,whichare<br />

formedbyRespondent’sresponsestothevarious φn. Finally, R(φn)isa<br />

functionfromsentencestothevalues 1, 0,and ?,correspondingtotherules<br />

Respondentmustapplytoreplytoeach φn.<br />

Therulesforthepositumare<br />

• R(φ0) = 0iff φ0 � ⊥,<br />

• R(φ0) = 1iff φ0 � ⊥.<br />

7 Indeed,theterminologyofformalvs.materialconsequences,fromwhichtheterminology<br />

usedbyBrandom(directlyborrowedfromSellars)ultimatelyderives,wasconsolidated<br />

inthe 14 th century;see(DutilhNovaes,2008).<br />

8 Throughoutthetext,Iwillusetheterms‘pertinent’and‘impertinent’,theliteraltranslationsoftheLatinterms‘pertinens’and‘impertinens’.<br />

Butnoticethattheyareoften<br />

translatedas‘relevant’and‘irrelevant’,forexampleinthetranslationofBurley’streatise<br />

(Burley,1988).


<strong>Obligationes</strong>asGOGAR 41<br />

Therulesforthepropositaare<br />

•Pertinentpropositions: Γn−1 � φnor Γn−1 � ¬φn;<br />

–If Γn−1 � φnthen R(φn) = 1;<br />

–If Γn−1 � ¬φnthen R(φn) = 0;<br />

•Impertinentpropositions: Γn−1 � φnand Γn−1 � ¬φn;<br />

–If KC � φnthen R(φn) = 1;<br />

–If KC � ¬φnthen R(φn) = 0;<br />

–If KC � φnand KC � ¬φnthen R(φn) =?.<br />

Asthedisputationprogresses,differentsetsofsentencesareformedat<br />

eachround,namelythesetsformedbythesentencesthatRespondenthas<br />

grantedandthecontradictoriesofthesentenceshehasdenied.Thesesets<br />

Γncanbeseenasmodelsofthesuccessivestagesofdeonticstatusesof<br />

Respondentwithrespecttothecommitmentsundertakenbyhimateach<br />

reply.Thesets Γnaredefinedasfollows:<br />

•If R(φn) = 1then Γn = Γn−1 ∪ {φn};<br />

•If R(φn) = 0then Γn = Γn−1 ∪ {¬φn};<br />

•If R(φn) =?then Γn = Γn−1.<br />

Forreasonsofspace,Ishallkeepmypresentationofobligationesvery<br />

brief. Theinterestedreaderisurgedtoconsultthevastprimaryandsecondaryliteratureonthetopic,<br />

9 butfurtheraspectsoftheframeworkwill<br />

bediscussedinthenextcomparativesectionsaswell.<br />

4 Comparison<br />

InthissectionIundertakeasystematiccomparisonofthetwoframeworks.<br />

Theemphasiswillbelaidonsimilarities,butIwillalsomentionsomeimportantdissimilarities.Essentially,whatisatstakeduringanobligatiodisputationistheabilitytoappreciatethe(logicalandpractical)consequences<br />

ofthecommitmentsundertakenbyRespondent. Respondediscommitted<br />

(i.e.obligated)tothesentenceshegrantsaswellastothecontradictories<br />

ofthesentenceshedenies. Thedeonticstatusofentitlementplaysaless<br />

prominentrolewithinobligationes,asthepointreallyistoexplorewhat<br />

oneisobligatedtoonceoneobligatesoneselftothepositum. Besidesthis<br />

9 Myownpreviouswork(DutilhNovaes,2005),(DutilhNovaes,2006),(DutilhNovaes,<br />

2007)canserveasastartingpoint.


42 CatarinaDutilhNovaes<br />

generalandfundamentalpointofsimilarity,thereareseveralspecificsimilaritiesbetweenGOGAR’sandobligationalconcepts:<br />

10<br />

Thekeyroleofinferentialrelations Inbothmodels,(intra-personal,<br />

inter-content)transferofcommitmenttakesplacethroughinferentialrelations,notrestrictedtoformallyvalidinferences.Bymeansofthetransferof<br />

commitment,Respondentisobligatedtoeverythingthatfollowsfromwhat<br />

hehasgranted/deniedsofar,aswellastothecontradictoriesofwhatis<br />

incompatible(repugnans)withwhathehasgranted/deniedsofar.Indeed,<br />

thenotionof‘repugnant’sentencescorrespondspreciselytoBrandom’snotionofincompatibility.<br />

Therelationofinferencerelatessetsofsentences/contents Both<br />

frameworkscorrectlytreattherelationofinference(andthecorresponding<br />

transferofcommitment)asrelatingsetsofsentencestosetsofsentences(althoughusuallytheconsequentsetisasingleton).<br />

Indeed,withinrational<br />

discursivepractices,whatcountsarenotsomuchtheinferentialrelations<br />

betweenindividualsentences/contents; asamatteroffact,weareusuallycommittedtoawiderangeofsentences/contents.Itistheinteractionbetweenthesedifferentcommitmentsthatcountstodefineourfurthercommitments:whatveryoftenhappensisthatcommitmentto<br />

paloneorto q<br />

alonedoesnotcommitthespeakerto t,butjointcommitmentto pandto<br />

qdoes.Intheobligationalframework,everypropositumthatisgrantedor<br />

deniedmodifiesRespondent’scommitments.<br />

Thedynamicnatureofbothmodels Acorollaryofthepreviouspoint<br />

isthatbothmodelsaredynamic,i.e.,temporalityisanimportantfactor.<br />

In(DutilhNovaes,2005),Ihaveexploredindetailthedynamicnatureof<br />

obligationes,andGOGARisdynamicinverymuchthesameway. Both<br />

modelsdealwithphenomenathattakeplaceinsuccessivesteps,andeach<br />

stepistosomeextentdeterminedbytheprevioussteps(afeaturethat<br />

isaccuratelycapturedbythegamemetaphor). Inbothcases,theorder<br />

ofoccurrenceofthesestepsiscrucial. Forexample,ifthepositumofan<br />

obligatiois‘Everymanisrunning’,andthenextstepis‘Youarerunning’,<br />

thispropositum mustbedeniedasimpertinentandfalse(sincenothing<br />

hasbeensaidaboutRespondentbeingamansofar).However,ifafterthe<br />

samepositum‘Youareaman’isproposedandaccepted(asimpertinentand<br />

true),andafterward‘Youarerunning’isproposed,thenthelattershould<br />

10 Forreasonsofspace,Iheretreatonlythemostsalientpointsofsimilarity. Notice<br />

thoughthatthereareothers,forexampletheroleplayedbypragmaticelementsinboth<br />

cases.


<strong>Obligationes</strong>asGOGAR 43<br />

beacceptedasfollowingfromwhathasbeengrantedsofar,contrarytothe<br />

firstscenario.<br />

Impertinentpropositionsanddefaultentitlement Eventhoughobligationesdealessentiallywithcommitmentsandlesssowithentitlements,<br />

onespecifickindofentitlementisneverthelesspresentintheframework.<br />

WhileRespondent’srepliestopertinentsentencesarefullydeterminedby<br />

hispreviouscommitments,therearenocommitmentsconcerningimpertinentsentences(asthisisexactlywhattheyare:thusfaruncommitted-to<br />

contents). Whatmustdeterminehisrepliestoimpertinentsentencesare<br />

exactlytheuncontroversialentitlementssharedbyallthosewhoarepresent<br />

atthedisputation.Theseincludecircumstantialinformation(suchasbeing<br />

inParisorbeinginRome),aswellasverygeneralcommonknowledge,for<br />

examplethatthePopeisaman.Inotherwords,Respondentisentitledto<br />

accepting,denyingordoubtingasentenceonthebasisofhisfactualknowledgeconcerningthem;theseareBrandom’s‘freemoves’,withthesame<br />

socialdimensioninsofarasitconcernscommonknowledge.<br />

Scorekeeping WithinGOGAR,scorekeepingissomethingofametaphor<br />

ratherthanareality—nobodyexplicitlywritesdownthecommitments<br />

andentitlementsofotherspeakers.Scorekeepingisrathersomethingdone<br />

tacitly,andusuallyoneisnotevenreallyawareofdoingit. Butwithin<br />

obligationes,scorekeepingisforreal. ThisisexactlywhatImeanwhenI<br />

saythatthelatterisaregimentedmodeloftheformer:someimplicit,tacit<br />

elementsofGOGARaremadeexplicitandtangiblewithinobligationes.<br />

Indeed,thosepresentatthedisputation(inparticularOpponent)explicitlykeepscoreofRespondent’ssuccessivedeonticstatusesofcommitments<br />

duringadisputation;whenhethenfailstorecognizeapreviouslytaken<br />

commitment,herespondsbadlyandlosesthegame. Moreover,oncethe<br />

disputationisover,Respondent’sperformanceisexplicitlyevaluatedbya<br />

panelofMasterspresentattheoccasion.<br />

Caveats Whiletheresemblancebetweenthetwoframeworksisoverwhelming,thereareofcourseimportantpointsofdissimilarity.Morespecifically,andasnotedbefore,obligationesisalessencompassingmodel,treatingonlyasubclassofthephenomenacapturedbyGOGAR.<br />

•<strong>Obligationes</strong>onlyaccountforthecommitmentsandentitlementsof<br />

onespeaker,namelyRespondent.<br />

•Askingforreasonsisnotpartofanobligatio:OpponentcannotchallengeRespondent,<br />

exceptbysaying‘cedat tempus’ifRespondent<br />

grantsacontradiction.


44 CatarinaDutilhNovaes<br />

•<strong>Obligationes</strong>offernoextensivetreatmentofthedifferentkindsofentitlementsandofthemechanismsoftransferofentitlement.<br />

•GOGARismeanttobeamodeloftheverymeaningfulnessoflanguage—i.e.,therelationsofcommitment-preservingentailmentand<br />

entitlement-preservingentailmentdefinethemeaningofutterances—<br />

whereasobligationesoperatewithalanguagethatismeaningfulfrom<br />

thestart.<br />

Forthesereasons,anobligatioisbestseenasasimplifiedmodelofhow<br />

aspeakermustbehavetowardsassertions.Thissimplificationmayonthe<br />

onehandentaillossofgenerality,butontheotherhanditmayoffera<br />

viewpointfromwhichsomepropertiesofoursocialdiscursivepracticesare<br />

mademanifestandcanthusmoreeasilybestudied.<br />

5 Whatisgainedthroughthecomparison?<br />

Forobligationes<br />

Thedeonticnatureofobligationes Eversincescholarsofmedieval<br />

philosophybecameinterestedinobligationeshalfwaythe 20 th century,the<br />

verynameofthisformofdisputationwasasourceofpuzzlement.Inwhat<br />

senseexactlydidsuchadisputationconsistinanobligation? Whowas<br />

obliged,andwhatwasheobligedto?Althoughsomemodernanalysesdid<br />

emphasizeitsdeonticnature(see(Knuuttila&Yrjonsuuri,1988)),itisfair<br />

tosaythatthedeonticcomponentwasessentiallyoverlookedinmostof<br />

them(includingmyowngameinterpretation). Onapersonalnote,Ican<br />

saythatIonlyfullyunderstoodhowthoroughlydeontictheobligationes<br />

frameworkreallywasagainstthebackgroundofGOGAR,andinparticular<br />

bymeansoftheconceptofcommitment.<br />

RecallthatIhaveaccountedforthenotionofcommitmenttoastatement/contentintermsofthepracticalconsequencesthattherelianceonitstruthcanhaveforotherpeople’slives,insofarastheyassumethestatementtobetrueunlesstheyhavegoodreasonsnotto(Brandom’s‘default<br />

entitlement’andLewis’‘conventionoftruthfulnessandtrust’)andinsofar<br />

astheymakepracticaldecisionsonthebasisontheirrelianceonitstruth.<br />

Ofcourse,giventhesomewhat‘artificial’settingofanobligatiodisputation,nopracticalconsequencesaretobeexpected.Nevertheless,thebasic<br />

ideaseemstobethatcommitment—obligation—transfersoverbymeans<br />

ofinferentialrelations: ifrespondentiscommittedto φnand φnimplies<br />

φm,thenrespondentisalsocommittedto φm. Now,sincerespondentis<br />

alwayscommittedtoatleastonestatement,thepositum,thisfirstcommitmentsetsthewholewheelofcommitmentsinmotion.Soanobligatioisnot<br />

onlyaboutlogicalrelationsbetweensentencesandconsistencymaintenance;


<strong>Obligationes</strong>asGOGAR 45<br />

moreimportantly,itisaboutthedeonticstatusesofcommitmentsandentitlementsandthe(intrapersonal,inter-content)mechanismsofinheritance<br />

ofthesestatuses.<br />

Thegeneralpurposeofobligationes Morepervasiveandsignificant<br />

thanthepuzzlementcausedbythetermobligationesitselfisthestillwidespreadperplexityofscholarsconcerningtheverypurposeofsuchdisputations:afterall,what’sthepoint?Whatareobligationesabout?Theyare<br />

notabouttruth,asmorestraightforwardformsofdisputationare,giventhat<br />

thepositumisgenerally,andconspicuously,apossiblebutfalsesentence.<br />

Manyofthemoderninterpretationshavesoughttoestablisharationalefor<br />

obligationes—alogicofconditionals,aframeworkforbeliefrevision—,<br />

buttheshortcomingsofeachoftheseinterpretationsonlycontributedto<br />

thegrowingfrustrationrelatedtotheapparentelusivenessofthe‘point’<br />

ofobligationes. Itcouldn’tpossiblybeamereformoftestingastudent’s<br />

skills,i.e.“schoolboy’sexercise”,assuggestedintheearlysecondaryliteratureofthe1960’s.<br />

Ifthereisnorealpurposetoitbeyondtheintricate<br />

logicalstructureoftheframework,thenitmightbemerelysterilescholastic<br />

logicalgymnasticsafterall,justasmostofthetechniquesofscholasticism<br />

accordingtothestandardpost-scholastic(i.e.Renaissance)criticism.<br />

ButwhenputinthecontextofGOGAR,obligationesseemtoprovidea<br />

modelofwhatitmeanstoactandtalkrationally,i.e.totakepartin(mainly,<br />

butnotexclusively)discursivesocialpractices. Thusviewed,obligationes<br />

couldalsomostcertainlyfulfillanimportantpedagogicaltask,namelythat<br />

ofteachingastudenthowtoarguerationally—i.e.,howtoarguemindfulofone’sentitlementsandcommitments,ofthereasons(grounds)forendorsingorrejectingstatements,andoftheneedtodefendone’sowncommitments—butitsimportanceclearlygoesbeyondmerelypedagogical<br />

purposes. Interestingly,throughoutthelaterMiddleAgestheformatof<br />

obligationeswasextensivelyadoptedforscientificinvestigations,precisely<br />

becauseitprovidesagoodmodelforrationalargumentation. Inawide<br />

varietyofcontexts(rangingfromlogictotheology,fromethicstophysics),<br />

oneencountersextensiveuseoftheobligationesvocabularyandconceptsin<br />

thepresentationofarguments.Thusseen,theframeworkisfarfrombeing<br />

afutilelogicalexercise:rather,itpresentsaregimentationofsomecrucial<br />

aspectsofwhatitistoargueandactrationally,ofwhichGOGARisalsoa<br />

(moreencompassing)model.<br />

Forthegameofgivingandaskingforreasons<br />

Underlyinglogicalstructure Whileintermsofthe‘biggerpicture’,it<br />

ismostlyobligationesthatcanbenefitfromthecomparison,onthelevel<br />

of(logical)detailGOGARhasmuchtolearnfromobligationes.Eversince


46 CatarinaDutilhNovaes<br />

thepublicationof(Brandom,1994),Brandomhasbeenrefiningthelogical<br />

structureunderlyingGOGARinparticularandhisinferentialistsemantics<br />

ingeneral,especiallythroughthedevelopmentofwhathecalls‘incompatibilitysemantics’.Nevertheless,anddespitethepowerfulnessofthemodern<br />

logicaltechniquesoftenemployedbyBrandomandhiscollaborators,these<br />

fairlyrecentdevelopmentsarestillsomewhatovershadowedbythecenturies<br />

ofresearch(involvingaverylargenumberoflogicians)onthelogicofobligationes.<br />

11 Indeed,the(primaryandsecondary)literatureonthetopic<br />

containssophisticatedanalysesofthelogicalandpragmaticpropertiesof<br />

theframework,whichare(presumably)applicabletoGOGARsothatthe<br />

comparisoncancontributetomakingGOGAR’slogicalstructureexplicit.<br />

Forexample,Ihaveprovedelsewhere(DutilhNovaes,2005)thattheclass<br />

ofmodelssatisfying Γnbecomessmallerinthenextstepofthegameonlyif<br />

φn+1isimpertinent;if φn+1ispertinent,thentheclassofmodelssatisfying<br />

Γnisthesameastheclassofmodelssatisfying Γn+1,eventhough Γnand<br />

Γn+1arenotthesame. 12 ThisresultcanbeinterpretedintermsofGOGAR<br />

inthefollowingmanner:whenaspeakermakesanassertion pwhichactually<br />

followsfromanysentenceorsetofsentencespreviouslyassertedbyhim,<br />

thenhissetofcommitmentsistherebynotaugmented.Inotherwords,his<br />

deonticstatusremainsthesame,ashewasdefactoalreadycommittedto p.<br />

Mixingthetwovocabularies,onecansaythataspeaker’sdeonticstatusis<br />

modifiedonlyifheassertsanimpertinentsentence;assertionsofpertinent<br />

sentenceshavenoeffectwhatsoeverinthissense. Now,thisisjustone<br />

exampleofhow,giventhattheobligationesframeworkisamoreregimented<br />

formoftherational,discursivepracticesalsomodeledbyGOGAR,such<br />

logicalpropertiesaremoreeasilyinvestigatedagainstthebackgroundof<br />

theformerratherthanthelatter.<br />

Strategicperspective Whenspeakingof‘thegameofgivingandasking<br />

forreasons’,Brandomseemstobetakingseriouslytheanalogybetweenthe<br />

rationaldiscursivepracticespresumablycapturedbyGOGARandactual<br />

games.ItisundoubtedlyalsoareferencetoWittgenstein’slanguage-games,<br />

butthequestionimmediatelyarises:howmuchofagameisGOGAR,actually?Tothebestofmyknowledge,Brandomdoesnotfurtherexplorethe<br />

comparisontogames,justashedoesnotdiscussspecificgame-theoretic<br />

propertiesofGOGAR;thisseemstome,however,tobeapromisingline<br />

ofinvestigation. Twoimportantgame-theoreticalpropertiesthatcometo<br />

mindarethegoal(s)tobeattainedwithinacertaingame,i.e.theexpected<br />

11 <strong>Obligationes</strong>wereoneofthemaintopicsofinvestigationinthelatemedievalLatin<br />

tradition,asattestedbytheverylargenumberofsurvivingtextsrangingfromthe 12 th<br />

tothe 15 th Century.<br />

12 Assuming,ofcourse,thatRespondenthasrepliedaccordingtotherules.


<strong>Obligationes</strong>asGOGAR 47<br />

outcome,andthepossiblestrategiestoplaythegame(usually,oneisinterestedinmaximizingthepayoff,i.e.intheratioofbestpossibleoutcome<br />

vs.themosteconomicalstrategy).Basedonthesetwoconcepts,itwould<br />

seemthatGOGARisinfactafamilyofgames,notasinglegame,aseach<br />

particulargameoftheGOGARfamilyhasitsowngoals.Mostofthemare<br />

cooperativegames,whereparticipantshaveacommongoalratherthanthat<br />

ofbeatingtheopponent,e.g.,dialogueswherepeopleexchangeinformation<br />

andcoordinatetheiractions. Nevertheless,thereareofcoursenumerous<br />

situationsofdiscursivepracticeswherethepointreallyistobeattheopponent,suchas,e.g.,inacourtoflaw.<br />

Ineachcase,differentstrategies<br />

mustbeemployed:inthecaseofcooperativegames,Griceanmaximsmay<br />

beseenasagoodaccountofstrategiestomaximizeunderstandingbetween<br />

theparties;inthecaseofcompetitivegames,however,acompletelydifferentstrategymustbeused,onewheredeceit,forinstance,mayevenhave<br />

someroletoplay.<br />

<strong>Obligationes</strong>isobviouslyacompetitivegame: ifRespondentgrantsa<br />

contradiction,helosesthegame;butifheisabletomaintainconsistency,<br />

hebeatsOpponent.Andeventhoughthemedievalauthorsthemselvesdid<br />

notaccountforobligationesintermsofgames(nordidtheyhaveknowledge<br />

ofthespecificgame-theoreticalconceptsjustdiscussed),medievaltreatises<br />

onobligationesarefilledwithstrategicadviceforRespondentonhowtoperformwellduringanobligatio.Thesetreatisespresentnotonlyrulesdefiningthelegitimatemoveswithinthedisputationbutalsopractical,strategicadvice.<br />

13 SomeofthestrategicrulespresentedinBurley’streatiseare:“One<br />

mustpayparticularattentiontotheorder[ofthepropositions]”(Burley,<br />

1988,p.385);“Whenapossiblepropositionhasbeenposited,itisnotabsurdtograntsomethingimpossibleperaccidens”(Burley,1988,p.389);<br />

“Whenafalsecontingentpropositionisposited,onecanproveanyfalse<br />

propositionthatiscompossiblewithit”(Burley,1988,p.391).<br />

ThepointhereisthatthestrategicperspectivepresentintheseobligationestreatisescanverylikelybetransposedtotheGOGARframework<br />

toproduceinterestingresults. InthecaseofGOGARgameswherethe<br />

differentspeakersaretrulyopposedtooneanotherandthepointisreally<br />

tobeattheopponent,thenthestrategictipsfromtheobligationestreatisescanbeusedstraightforwardly.<br />

Buteveninthecaseofcooperative<br />

gamesofGOGAR,theobligationalstrategiesmaystillbequitehelpful,as<br />

theyessentiallydescribeproceduresthatmayenableonetomaintainconsistency—certainlyadesirableoutcomeinthecontextofrationaldiscursive<br />

practices. TheheartofthematteristhatGOGARdoesnotemphasize<br />

theplayer-perspective:rather,Brandom’sdescriptionofGOGARisthatof<br />

13 “Itisimportanttoknowthattherearesomerulesthatconstitutethepracticeofthis<br />

artandothersthatpertaintoitsbeingpracticedwell.”(Burley,1988,p.379)


48 CatarinaDutilhNovaes<br />

thetheoriststandingoutsidethegameandofferingamodeltoexplainthe<br />

use(s)andmeaningfulnessoflanguage.Inthissense,theplayer-perspective<br />

offeredbyobligationesmaycomeasaninterestingcomplement.<br />

Theroleofdoubting BrandompresentsGOGARashavingonlyone<br />

quintessentialkindofmove,i.e.makingaclaim. Wehaveseenthatchallengingisalsoanimportantmove,butachallengeismadebymeansof<br />

theassertionofanincompatiblecontent.Incontrast,obligationesfeature<br />

threemainkindsofmovesforRespondent:granting,denyinganddoubting.<br />

Grantingobviouslycorrespondstoasserting,andinasense,denyingisalso<br />

akindofassertionwithintheobligationesframework,namelytheassertion<br />

ofthecontradictoryofthedeniedsentence. Ihavealsopointedoutthat<br />

challengingisnotalegitimatemoveforeitherRespondentorOpponent,a<br />

factthatisrelatedtotheregimentedandsimplifiednatureofobligationes<br />

asamodelofrationaldiscursivepractices.ButGOGARclaimstobemuch<br />

moreencompassingthanobligationesdoes,sowhileitseemsreasonablefor<br />

obligationestoleavesomeimportantelementsout,thesamedoesnothold<br />

ofGOGAR.Now,GOGARhasnoresourcestodealwiththephenomenon<br />

ofnotbeingsure,ofrecognizingthatonedoesnotdisposeofsufficient<br />

groundstoassertacontentoritscontradictory(knowingthatyoudon’t<br />

know),whereasthisseemstobeaveryimportantelementofourrational<br />

discursivepractices. Incontrast,byhavingdoubtingasoneofthelegitimatemovesforRespondent,theobligationalframeworkfaresbatterinthis<br />

respect.<br />

ItmightbearguedthatdoubtingisnotrelevantforGOGARinsofarasit<br />

hasnoimpactonaspeaker’sdeonticstatus,asitissimplythelackofcommitmentorentitlement;notso.AparticularrulepresentedinKilvington’s<br />

treatmentofobligationesinhisSophismata(Kilvington,1990,sophism48)<br />

showsthatdoubtingcanindeedalteraspeaker’sdeonticstatus. Therule<br />

isthefollowing:if‘pimplies q’isagoodconsequence,andifRespondent<br />

hasdoubted p,thenhemustnotdeny q,i.e.,heisnotentitledto ¬q.<br />

Thisisbecause,inavalidconsequence,iftheconsequentis(knowntobe)<br />

false,thentheantecedentwillalsobe(knowntobe)false,soifRespondent<br />

hasdoubtedtheantecedent,hemusteitherdoubtorgranttheconsequent.<br />

Thisisjustanexampleoftheintricaciesofthelogicofdoubtingandof<br />

howdoubtingcanindeedhaveanimpactonone’sdeonticstatus.Theobligationalliteratureisfilledwithmanymoreofsuchexamples,inparticular<br />

inthetreatmentsofdubitatio, 14 oneoftheformsofanobligationaldis-<br />

14 Inadubitatio,thefirstsentence(theobligatum)isnotapositum,itisadubium,a<br />

sentencewhichRespondentmustdoubtforthesakeofthedisputationjustasheaccepts<br />

theposituminapositio;hemustthenseewhatfollows(intermsofhiscommitmentsand<br />

entitlements)fromhavingdoubtedthefirstsentence.


<strong>Obligationes</strong>asGOGAR 49<br />

putationalongwithpositio(whichisinsomesensethe‘standard’formof<br />

obligationes,andtheonediscussedinthistextsofar). Thus,Isuggest<br />

thatGOGARshouldpaymoreattentiontospeech-actsotherthanassertionsasalsohavinganimpactonaspeaker’sdeonticstatus—doubtingin<br />

particular,asshownwithintheobligationalframework.<br />

6 Conclusion<br />

InthisbriefcomparativeanalysisofGOGARandmedievalobligationesI<br />

hopetohaveindicatedhowfruitfulasystematiccomparisonbetweenthe<br />

twoframeworkscanbe. ForreasonsofspaceIhaveheremerelysketched<br />

suchacomparison,andamorethoroughanalysisshallremainatopicfor<br />

futurework.<br />

CatarinaDutilhNovaes<br />

DepartmentofPhilosophyandILLC,UniversityofAmsterdam<br />

NieuweDoelenstraat15,1012CPAmsterdam,TheNetherlands<br />

c.dutilhnovaes@uva.nl<br />

http://staff.science.uva.nl/∼dutilh/<br />

References<br />

Brandom,R. (1994). MakingitExplicit. Cambridge,MA:HarvardUniversity<br />

Press.<br />

Brandom,R. (2008). Betweensayinganddoing. Oxford: OxfordUniversity<br />

Press.<br />

Burley,W.(1988).Obligations(selection).InN.Kretzmann&E.Stump(Eds.),<br />

TheCambridgeTranslationsofMedievalPhilosophicalTexts:LogicandthePhilosophyofLanguage(pp.369–412).Cambridge:CambridgeUniversityPress.<br />

DutilhNovaes,C.(2005).Medievalobligationesaslogicalgamesofconsistency<br />

maintenance.Synthese,145(3),371–395.<br />

DutilhNovaes,C. (2006). RogerSwyneshed’sobligationes: Alogicalgameof<br />

inferencerecognition?Synthese,151(1),125–153.<br />

DutilhNovaes,C.(2007).Formalizingmedievallogicaltheories.Berlin:Springer.<br />

DutilhNovaes,C.(2008).Logicinthe 14 th CenturyafterOckham.InD.Gabbay<br />

&J.Woods(Eds.),Handbookofthehistoryoflogic(Vol.2: Mediaevaland<br />

RenaissanceLogic,pp.433–504).Amsterdam:Elsevier.<br />

Kilvington,R. (1990). Sophismata. Cambridge: CambridgeUniversityPress.<br />

(Englishtranslation,historicalintroductionandphilosophicalcommentarybyN.<br />

KretzmannandB.E.Kretzmann.)


50 CatarinaDutilhNovaes<br />

Knuuttila,S.,&Yrjonsuuri,M. (1988). Normsandactionsinobligationaldisputations.<br />

InO.Pluta(Ed.),DiePhilosophieim14.und15.Jahrhundert(pp.<br />

191–202).Amsterdam:B.R.Güner.


Truth Value Intervals, Bets, and Dialogue Games<br />

Christian G. Fermüller ∗<br />

FuzzylogicsinZadeh’s‘narrowsense’(Zadeh,1988),i.e.,truthfunctional<br />

logicsreferringtotherealclosedunitinterval [0,1]assetoftruthvalues,areoftenmotivatedaslogicsfor‘reasoningwithimprecisenotionsand<br />

propositions’(see,e.g.,(Hájek,1998)).Howevertherelationbetweenthese<br />

logicsandtheoriesofvagueness,asdiscussedinaprolificdiscourseinanalyticphilosophy(Keefe&Rosanna,2000),(Keefe&Smith,1987),(Shapiro,<br />

2006)ishighlycontentious.Wewillnotdirectlyengageinthisdebatehere<br />

butratherpickoutso-calledintervalbasedfuzzylogicsasaninstructive<br />

exampletostudy<br />

1.howsuchlogicsareusuallymotivatedinformally,<br />

2.whatproblemsmayarisefromthesemotivations,and<br />

3.howbettinganddialoguegamesmaybeusedtoanalyzetheselogics<br />

withrespecttomoregeneralprinciplesandmodelsofreasoning.<br />

Themaintechnicalresult 1 ofthisworkconsistsinacharacterizationofan<br />

importantintervalbasedlogic,considered,e.g.,in(Esteva,Garcia-Calvés,<br />

&Godo,1994),intermsofadialoguecumbettinggame,thatgeneralizes<br />

RobinGiles’sgamebasedcharacterizationofŁukasiewiczlogic(Giles,1974),<br />

(Giles,1977). However,ouraimistoaddressfoundationalproblemswith<br />

certainmodelsofreasoningwithimpreciseinformation.Wehopetoshow<br />

thatthetraditionalparadigmofdialoguegamesasapossiblefoundationof<br />

logic(goingback,atleast,to(Lorenzen,1960))combinedwithbetsas‘test<br />

cases’forrationalityinthefaceofuncertaintymighthelptosortoutsome<br />

oftherelevantconceptualissues.Thisisintendedtohighlightaparticular<br />

meetingplaceoflogic,games,anddecisiontheoryatthefoundationofa<br />

fieldoftencalled‘approximatereasoning’(see,e.g.,(Zadeh,1975)).<br />

∗ ThisworkissupportedbyFWFprojectI143–G15.<br />

1 Duetolimitedspace,westatepropositionswithoutproofs.


52 ChristianG.Fermüller<br />

1 T-normbasedfuzzylogicsandbilattices<br />

PetrHájek,intheprefaceofhisinfluentialmonographonmathematical<br />

fuzzylogic(Hájek,1998)asserts:<br />

Theaimistoshowthatfuzzylogicasalogicofimprecise(vague)<br />

propositionsdoeshavewelldevelopedformalfoundationsandthat<br />

mostthingsusuallynamed‘fuzzyinference’canbenaturallyunderstoodaslogicaldeduction.(Hájek,1998,p.viii)<br />

Asthequalification‘vague’,addedinparenthesisto‘imprecise’,betrays,<br />

someterminologicaland,arguably,alsoconceptionalproblemsmaybelocatedalreadyinthisintroductorystatement.<br />

Theseproblemsrelateto<br />

thefactthatfuzzylogicisoftensubsumedunderthegeneralheadingsof<br />

‘uncertainty’and‘approximatereasoning’. Inanycase,Hajekgoesonto<br />

introduceafamilyofformallogics,basedonthefollowingdesignchoices<br />

(comparealso(Hájek,2002)):<br />

1.Thesetofdegreesoftruth(truthvalues)isrepresentedbythereal<br />

unitinterval [0,1].Theusualorderrelation ≤modelscomparisonof<br />

truthdegrees; 0representsabsolutefalsity,and 1representsabsolute<br />

truth.<br />

2.Thetruthvalueofacompoundstatementshallonlydependonthe<br />

truthvaluesofitssubformulas. Inotherwords:thelogicsaretruth<br />

functional.<br />

3.Thetruthfunctionfor(strong)conjunction(&)shouldbeacontinuous,commutative,associative,andmonotonicallyincreasingfunction<br />

∗ : [0,1] 2 → [0,1],where 0 ∗ x = 0and 1 ∗ x = x.Inotherwords: ∗<br />

isacontinuous t-norm.<br />

4.Theresiduum ⇒∗ofthe t-norm ∗—i.e.,theuniquefunction ⇒∗:<br />

[0,1] 2 → [0,1]satisfying x ⇒∗ y = sup{z | x ∗ z ≤ y}—servesasthe<br />

truthfunctionforimplication.<br />

5.Thetruthfunctionfornegationis λx[x ⇒∗ 0].<br />

ProbablythebestknownlogicarisinginthiswayisŁukasiewiczlogicL<br />

(Łukasiewicz,1920),wherethe t-norm ∗Lthatservesastruthfunctionfor<br />

&isdefinedas x ∗L y = max(0,x + y − 1). Itsresiduum ⇒Lisgivenby<br />

x ⇒L y = min(1,1 − x + y).Apopularalternativechoiceforconjunction<br />

takestheminimumasitstruthfunction.Besides‘strongconjunction’(&),<br />

alsothislatter‘weak(min)conjunction’(∧)canbedefinedinall t-norm<br />

basedlogicsby A ∧ B def<br />

= A&(A → B). Maximumastruthfunctionfor<br />

disjunction(∨)isalwaysdefinablefrom ∗and ⇒∗,too.


TruthValueIntervals 53<br />

Otherimportantlogics,likeGödellogicG,andProductlogicP,can<br />

beobtainedinthesameway,butwewillconfineattentiontoL,here.At<br />

thispointweliketomentionthatarich,deep,andstillgrowingsubfield<br />

ofmathematicallogic,documentedinhundredsofpapersandanumberof<br />

books(beyond(Hájek,1998))istriggeredbythisapproach.Consequently<br />

itbecameevidentthatdegreebasedfuzzylogicsareneithera‘poorman’s<br />

substituteforprobabilisticreasoning’noratrivialgeneralizationoffinitevaluedlogics.<br />

Anumberofresearchershavepointedoutthat,whilemodellingdegrees<br />

oftruthbyvaluesin [0,1]mightbeajustifiablechoiceinprinciple,itis<br />

hardlyrealistictoassumethatthereareproceduresthatallowustoassign<br />

concretevaluestoconcrete(interpreted)atomicpropositionsinacoherent<br />

andprincipledmanner. Whilethisproblemmightbeignoredaslongas<br />

weareonlyinterestedinanabstractcharacterizationoflogicalconsequence<br />

incontextsofgradedtruth,itisdeemeddesirabletorefinethemodelby<br />

incorporating‘imprecisionduetopossibleincompletenessoftheavailable<br />

information’(Estevaetal.,1994)abouttruthvalues. Accordingly,itis<br />

suggestedtoreplacesinglevalues x ∈ [0,1]bywholeintervals [a,b] ⊆ [0,1]<br />

oftruthvaluesasthebasicsemanticunitassignedtopropositions. The<br />

‘naturaltruthordering’ ≤canbegeneralizedtointervalsindifferentways.<br />

Following(Estevaetal.,1994)wearriveatthesedefinitions:<br />

Weaktruthordering: [a1,b1] ≤ ∗ [a2,b2]iff a1 ≤ a2and b1 ≤ b2<br />

Strongtruthordering: [a1,b1] ≺ [a2,b2]iff b1 ≤ a2or [a1,b1] = [a2,b2]<br />

Ontheotherhand,setinclusion(⊆)iscalledimprecisionorderinginthis<br />

context.Thesetofclosedsubintervals Int [0,1]of [0,1]isaugmentedbythe<br />

emptyinterval ∅toyieldso-calledenrichedbilatticestructures 〈Int [0,1], ≤ ∗ ,<br />

0,1, ∅,L,N ∗ 〉aswellas 〈Int [0,1], ≺,0,1, ∅,L,N ∗ 〉,where Listhestandard<br />

latticeon [0,1],withminimumandmaximumasoperators,and N ∗ isthe<br />

extensionofthenegationoperator Ntointervals;inourparticularcase<br />

N ∗ ([a,b]) = [1 − b,1 − a]and N ∗ (∅) = ∅.<br />

Quiteanumberofpapershavebeendevotedtothestudyoflogics<br />

basedonsuchintervalgeneratedbilattices. Letusjustmentionthatthe<br />

GhentschoolofKerre,Deschrijver,Cornelis,andcolleagueshasproduced<br />

animpressiveamountofworkonintervalbilatticebasedlogics(see,e.g.,<br />

(Cornelis,Deschrijver,&Kerre,2006)).<br />

Whileitisstraightforwardtogeneralizebothtypesofconjunction(tnormandminimum)aswellasdisjunction(maximum)from<br />

[0,1]to Int [0,1]<br />

byapplyingtheoperatorspoint-wise,itseemslessclearhowthe‘right’<br />

generalizationofthetruthfunctionforimplicationshouldlooklike. In<br />

(Cornelis,Arieli,Deschrijver,&Kerre,2007), (Cornelis,Deschrijver,&<br />

Kerre,2004) [a,b] ⇒∗ C [c,d]def = [min(a ⇒ c,b ⇒ d),b ⇒ d]isstudied,butin


54 ChristianG.Fermüller<br />

(Estevaetal.,1994)theauthorssuggest [a,b] ⇒∗ E [c,d]def = [b ⇒ c,a ⇒ d].<br />

Ashasbeenpointedoutin(Hájek,n.d.)thereseemstobeakindoftradeoff<br />

involvedhere.While ⇒∗ Cpreservesalotofalgebraicstructure—inpartic ularityieldsaresiduatedlatticewhichcontainstheunderlyinglatticeover<br />

[0,1]asasubstructure—thefunction ⇒∗ Eisnotaresiduum,butleads tothefollowingdesirablepreservationpropertythatismissingfor ⇒∗ C .If<br />

M2isaprecisiationof M1(meaning:foreachpropositionalvariable p, M2<br />

assignsasubintervaloftheintervalassignedto pby M1),thananyformula<br />

satisfiedby M1isalsosatisfiedby M2. 2Below,wewilltrytoshowthata gamebasedapproachmightjustifythepreferenceof ⇒∗ Eover ⇒∗C against<br />

abackgroundthattakesthechallengeofderivingformalsemanticsfrom<br />

firstprinciplesaboutlogicalreasoningmoreseriouslythanthementioned<br />

literatureon‘intervallogics’.<br />

2 Worriesabouttruthfunctionality<br />

Itisinterestingtonotethatboth,(Estevaetal.,1994)and(Cornelisetal.,<br />

2007),(Cornelisetal.,2004),refertoGinsberg(Ginsberg,1988),whoexplicitlyintroducedbilatticesfollowingideasof(Belnap,1977).MostprominentlyGinsbergconsiders<br />

B = 〈{0, ⊤, ⊥,1}, ≤t, ≤k, ¬〉asendowedwiththe<br />

followingintendedmeaning:<br />

• 0and 1represent(classical)falsityandtruth,respectively, ⊤represents‘inconsistentinformation’and<br />

⊥represents‘noinformation’.<br />

Theideahereisthattruthvaluesareassignedafterreceivingrelevant<br />

informationfromdifferentsources. Accordingly ⊤isidentifiedwith<br />

theinformationset {0,1}, ⊥with ∅andtheclassicaltruthvalueswith<br />

theirsingletonsets.<br />

• ≤t,definedby 0 ≤t ⊤/⊥ ≤ 1,isthe‘truthordering’.<br />

• ≤k,definedby ⊥ ≤t 0/1 ≤ 1,isthe‘knowledgeordering’.<br />

•Negationisdefinedby ¬(0) = 1, ¬(1) = 0, ¬(⊤) = ⊤, ¬(⊥) = ⊥.<br />

Whilethefour‘truthvalues’of Bmayjustifiablybeunderstoodtorepresent<br />

differentstatesofknowledgeaboutpropositions,itisveryquestionableto<br />

trytodefinecorresponding‘truthfunctions’forconnectivesotherthannegation.<br />

Indeed,itissurprisingtoseehowmanyauthors 3 followed(Belnap,<br />

1977)indefendingafourvalued,truthfunctionallogicbasedon B.Itshould<br />

beclearthat,intheunderlyingclassicalsettingthatistakenforgranted<br />

byBelnap,theformula A ∧ ¬Acanonlybefalse (0),independentlyofthe<br />

2 Here,aformulaisdefinedtobesatisfiedifitevaluatestothedegenerateinterval [1, 1].<br />

3 DozensofpapershavebeenwrittenaboutBelnap’s4-valuedlogic.


TruthValueIntervals 55<br />

kindofinformation,ifany,wehaveaboutthetruthof A. Ontheother<br />

hand,ifweneitherhaveinformationabout Anorabout B,then B ∧ ¬A<br />

couldbetrueaswellasfalse,andtherefore ⊥shouldbeassignednotonly<br />

to A, B,and ¬A,butalsoto B ∧ ¬A(incontrastto A ∧ ¬A).Thissimple<br />

argumentillustratesthatknowledgedoesnotpropagatecompositionally—<br />

awellknownfactthat,however,hasbeenignoredrepeatedlyintheliterature.(Forarecent,forcefulreminderontheincoherencyoftheintended<br />

semanticsforBelnap’slogicwereferto(Dubois,n.d.).)<br />

Inourcontextthiswarningaboutthelimitsoftruthfunctionalityis<br />

relevantattwoseparatelevels. First,itimpliesthat‘degreesoftruth’for<br />

compoundstatementscannotbeinterpretedepistemicallywhileupholding<br />

truthfunctionality. Indeed,mostfuzzylogicianscorrectlyemphasizethat<br />

theconceptofdegreesoftruthisorthogonaltotheconceptofdegreesof<br />

belief. Whiletruthfunctionsfordegreesoftruthcanbemotivatedand<br />

justifiedinvariousways—belowwewillreviewagamebasedapproach<br />

—degreesofbeliefsimplydon’tpropagatecompositionallyandcallfor<br />

othertypesoflogicalmodels(e.g.,‘possibleworlds’). Second,concerning<br />

theconceptofintervalsofdegreesoftruth,oneshouldrecognizethatitis<br />

incoherenttoinsistonbothatthesametime:<br />

1.truthfunctionsforallconnectives,liftedfrom [0,1]to Int [0,1],and<br />

2.theinterpretationofaninterval [a,b] ⊆ [0,1]assignedtoa(compound)<br />

proposition F asrepresentingasituationwhereourbestknowledge<br />

aboutthe(definite)degreeoftruth d ∈ [0,1]of Fisthat a ≤ d ≤ b.<br />

Giventhemathematicaleleganceof1,thatresults,amongotherdesirable<br />

properties,inalowcomputationalcomplexityoftheinvolvedlogics, 4 one<br />

shouldlookforalternativesto2.GodoandEsteva 5 havepointedoutthat,<br />

ifweinsiston2justforatomicpropositions,thenatleastwecanassert<br />

thatthecorresponding‘real’,butunknowntruthdegreeofanycomposite<br />

proposition F cannotlieoutsidetheintervalassignedto F accordingto<br />

thetruthfunctionsconsideredin(Estevaetal.,1994)(describedabove).<br />

However,theseboundsarenotoptimal,ingeneral. AswewillseeinSection5,takingcluesfromGiles’sgamebasedsemanticforL(Giles,1974),<br />

(Giles,1977),atightercharacterizationemergesifwedismisstheideathat<br />

intervalsrepresentsetsofunknown,butdefinitetruthdegrees.<br />

4 ItiseasytoseethatcoNP-completenessoftestingvalidityforL(andmanyother t-norm<br />

basedlogics)carriesovertotheintervalbasedlogicsdescribedabove.<br />

5 Privatecommunication.


56 ChristianG.Fermüller<br />

3 RevisitingGiles’sgameforL<br />

Giles’sanalysis(Giles,1974),(Giles,1977)ofapproximatereasoningoriginallyreferredtothephenomenonof‘dispersion’inthecontextofphysicaltheories.Later(Giles,1976)explicitlyappliedthesameconcepttotheproblemofproviding‘tangiblemeanings’to(composite)fuzzypropositions.<br />

6 For<br />

thispurposeheintroducesagamethatconsistsoftwoindependentcomponents:<br />

Bettingforpositiveresultsofexperiments.<br />

Twoplayers—say:meandyou—agreetopay1€totheopponentplayer<br />

foreveryfalsestatementtheyassert.By [p1,... ,pm�q1,...,qn]wedenote<br />

anelementarystateofthegame,whereIasserteachofthe qiinthemultiset<br />

{q1,... ,qn}ofatomicstatements(representedbypropositionalvariables),<br />

andyouasserteachatomicstatement pi ∈ {p1,... ,pm}.<br />

Eachpropositionalvariable qreferstoanexperiment Eqwithbinary<br />

(yes/no)result.Thestatement qcanbereadas‘Eqyieldsapositiveresult’.<br />

Thingsgetinterestingastheexperimentsmayshowdispersion;i.e.,thesame<br />

experimentmayyielddifferentresultswhenrepeated.However,theresults<br />

arenotcompletelyarbitrary:foreveryrunofthegame,afixedriskvalue<br />

〈q〉 r ∈ [0,1]isassociatedwith q,denotingtheprobabilitythat Eqyieldsa<br />

negativeresult.Forthespecialatomicformula ⊥(falsum)wedefine 〈⊥〉 r =<br />

1. Theriskassociatedwithamultiset {p1,... ,pm}ofatomicformulas<br />

isdefinedas 〈p1,...,pm〉 r = m�<br />

〈pi〉 r .Itspecifiestheexpectedamountof<br />

i=1<br />

money(in €)thathastobepaidaccordingtotheaboveagreement. The<br />

risk 〈〉 r associatedwiththeemptymultisetis 0. Theriskassociatedwith<br />

anelementarystate [p1,... ,pm�q1,...,qn]iscalculatedfrommypointof<br />

view. Thereforethecondition 〈p1,...,pm〉 r ≥ 〈q1,... ,qn〉 r expressesthat<br />

Idonotexpect(intheprobabilitytheoreticsense)anyloss(butpossibly<br />

somegain)whenwebetonthetruthoftheinvolvedatomicstatementsas<br />

stipulatedabove.<br />

6 E.g.,Gilessuggeststospecifythesemanticsofthefuzzypredicate’breakable’byassigninganexperimentlike’droppingtherelevantobjectfromacertainheighttoseeifit<br />

breaks’. Theexpecteddispersivenessofsuchanexperimentrepresentthe’fuzziness’of<br />

thecorrespondingpredicate.Anarguablyevenbetterexampleofadispersiveexperiment<br />

intheintendedcontextmightconsistinaskinganarbitrarilychosencompetentspeaker<br />

forayes/noanswertoquestionslike’IsChristall?’ or‘IsShakirafamous?’ forwhich<br />

truthmaycogentlybetakenasamatterofdegree.


TruthValueIntervals 57<br />

Adialoguegameforthereductionofcompositeformulas.<br />

GilesfollowsideasofPaulLorenzenthatdatebackalreadytothe1950s(see,<br />

e.g.,(Lorenzen,1960))andconstrainsthemeaningoflogicalconnectives<br />

byreferencetorulesofadialoguegamethatproceedsbysystematically<br />

reducingargumentsaboutcompositeformulastoargumentsabouttheir<br />

subformulas.<br />

Thedialogueruleforimplicationcanbestatedasfollows:<br />

R → IfIassert A → Bthen,wheneveryouchoosetoattackthisstatement<br />

byasserting A,Ihavetoassertalso B.(Andviceversa,i.e.,forthe<br />

rolesofmeandyouswitched.)<br />

Thisrulereflectstheideathatthemeaningofimplicationisspecifiedby<br />

theprinciplethatanassertionof‘if A,then B’(A → B)obligesoneto<br />

assert B,if Aisgranted. 7<br />

Inthefollowingweonlystatetherulesfor‘me’;therulesfor‘you’are<br />

perfectlysymmetric.Fordisjunctionwestipulate:<br />

R∨IfIassert A1 ∨ A2thenIhavetoassertalso Aiforsome i ∈ {1,2}<br />

thatImyselfmaychoose.<br />

Therulefor(weak)conjunctionisdual:<br />

R∧IfIassert A1 ∧ A2thenIhavetoassertalso Aiforany i ∈ {1,2}that<br />

youmaychoose.<br />

Onemightaskwhetherassertingaconjunctionshouldn’tobligeonetoassert<br />

bothdisjuncts.Indeed,forstrongconjunction 8 wehave<br />

R&IfIassert A1&A2thenIhavetoasserteitherboth A1and A2,or<br />

just ⊥.<br />

Thepossibilityofasserting ⊥insteadoftheattackedconjunctionreflects<br />

Giles’s‘principleofhedgedloss’: oneneverhastoriskmorethan1€for<br />

eachassertion.Asserting ⊥isequivalentto(certainly)paying1€.<br />

Incontrasttodialoguegamesforintuitionisticlogic(Lorenzen,1960),<br />

(Felscher,1985)orfragmentsoflinearlogic,nospecialregulationonthe<br />

successionofmovesinadialogueisrequiredhere. Moreover,weassume<br />

thateachassertionisattackedatmostonce:thisisreflectedbytheremoval<br />

oftheoccurrenceoftheformula Ffromthemultisetofformulasasserted<br />

byaplayer,assoonasithasbeenattacked,orwhenevertheotherplayer<br />

hasindicatedthatshewillnotattackthisoccurrenceof Fduringthewhole<br />

7 Notethat,since ¬Fisdefinedas F → ⊥,accordingto R →andtheabovedefinitionof<br />

risk,theriskinvolvedinasserting ¬pis 1 − 〈p〉 r .<br />

8 Gilesdidnotconsiderstrongconjunction.Theruleisfrom(Fermüller&Kosik,2006).


58 ChristianG.Fermüller<br />

runofthedialoguegame. Everyrunthusendsinfinitelymanystepsin<br />

anelementarystate [p1,...,pm�q1,...,qn]. Givenanassignment 〈·〉 r of<br />

riskvaluestoall piand qiwesaythatIwinthecorrespondingrunofthe<br />

gameifIdonothavetoexpectanylossinaverage,i.e.,if 〈p1,...,pm〉 r ≥<br />

〈q1,... ,qn〉 r .<br />

AsanalmosttrivialexampleconsiderthegamewhereIinitiallyassert<br />

p → qforsomeatomicformulas pand q;i.e.,theinitialstateis [�p → q].In<br />

response,youcaneitherassert pinordertoforcemetoassert q,orexplicitly<br />

refusetoattack p → q.Inthefirstcase,thegameendsintheelementary<br />

state [p�q];inthesecondcaseitendsinstate [�].Ifanassignment 〈·〉 r of<br />

riskvaluesgives 〈p〉 r ≥ 〈q〉 r ,thenIwin,whatevermoveyouchoosetomake.<br />

Inotherwords:Ihaveawinningstrategyfor p → qinallassignmentsof<br />

riskvalueswhere 〈p〉 r ≥ 〈q〉 r .<br />

Notethatwinning,asdefinedhere,doesnotguaranteethatIdonot<br />

loosemoney.Ihaveawinningstrategyfor p → p,resultingeitherinstate<br />

[�]orinstate [p�p]dependingonyour(theopponents)choice.Inthesecond<br />

case,althoughthewinningconditionisclearlysatisfied,Iwillactuallyloose<br />

1€,iftheexecutionoftheexperiment Epassociatedwithyourassertion<br />

of phappenstoyieldapositiveresult,buttheexecutionofthesameexperimentassociatedwithmyassertionof<br />

pyieldsanegativeresult. Itis<br />

onlyguaranteedthatmyexpectedlossisnon-positive.(‘Expectation’,here,<br />

referstostandardprobabilitytheory. Underafrequentistinterpretation<br />

ofprobabilitywemaythinkofitasaverageloss,resultingfromunlimited<br />

repetitionsofthecorrespondingexperiments.)<br />

TostateGiles’smainresult,recallthatavaluation vforŁukasiewicz<br />

logicLisafunctionassigningvalues ∈ [0,1]tothepropositionalvariables<br />

and 0to ⊥,extendedtocompositeformulasusingthetruthfunctions ∗L,<br />

max, min,and ⇒L,forstrongandweakconjunction,disjunctionandimplication,respectively.<br />

Theorem1((Giles,1974),(Fermüller&Kosik,2006)).Eachassignment<br />

〈·〉 r ofriskvaluestoatomicformulasoccurringinaformula Finducesa<br />

valuation vforLsuchthat v(F) = xifandonlyifmyoptimalstrategy<br />

for Fresultsinanexpectedlossof (1 − x)€.<br />

Corollary1. FisvalidinLifandonlyifforallassignmentsofriskvalues<br />

toatomicformulasoccurringin FIhaveawinningstrategyfor F.<br />

4 Playingunderpartialknowledge<br />

ItisimportanttorealizethatGiles’sgamemodelforreasoningaboutvague<br />

(i.e.,here,unstable)propositionsimpliesthateachoccurrenceofthesame<br />

atomicpropositioninacompositestatementmaybeevaluateddifferently


TruthValueIntervals 59<br />

atthelevelofresultsofassociatedexecutionsofbinaryexperiments.This<br />

featureinducestruthfunctionality:thevaluefor p∨¬pisnottheprobability<br />

thatexperiment Epeitheryieldsapositiveoranegativeresult,whichis 1<br />

bydefinition;itratheris 1−x,where x = min(〈p〉 r ,1−〈p〉 r )ismyexpected<br />

loss(in €)afterhavingdecidedtobeteitherforapositiveorforanegative<br />

resultofanexecutionof Ep(whatevercarrieslessriskforme).<br />

Theplayersonlyknowtheindividualsuccessprobabilities 9oftherelevant experiments.Alternatively,onemaydisregardindividualresultsofbinary<br />

experimentsaltogetherandsimplyidentifytheassignedprobabilitieswith<br />

‘degreesoftruth’. Inthisvariantthe‘pay-off’justcorrespondstothese<br />

truthvalues,andGiles’sgameturnsintoakindofHintikkastyleevaluation<br />

gameforL.<br />

Howdoesallthisbearonthementionedproblemsofinterpretationfor<br />

intervalbasedfuzzylogics?Rememberthatboth,(Estevaetal.,1994)and<br />

(Cornelisetal.,2007,2006,2004)seemtosuggestthatanintervaloftruth<br />

values [a,b]represents‘impreciseknowledge’aboutthe‘realtruthvalue’ c,<br />

inthesensethatonly c ∈ [a,b]isknown. Forthebettinganddialogue<br />

gamesemanticthissuggeststhattheplayers(oratleastplayer‘I’)now<br />

havetochoosetheirmovesinlightofcorresponding‘imprecise’(partial)<br />

knowledgeaboutthesuccessprobabilitiesoftheassociatedexperiments.<br />

However,whilethismayresultinaninterestingvariantoftheGiles’sgame,<br />

itsrelationtothetruthfunctionalsemanticssuggestedforlogicsbasedon<br />

Int [0,1]andL-connectivesisdubious.<br />

Thefollowingsimpleexampleillustratesthisissue.Supposetheinterval<br />

v ∗ (p) = [v ∗ 1 (p),v∗ 2<br />

(p)]assignedtothepropositionalvariable pis [0,1],re-<br />

flectingthatwehavenoknowledgeatallaboutthe‘realtruthvalue’ofthe<br />

propositionrepresentedby p. Accordingtothetruthfunctionspresented<br />

inSection1,theformula p ∨ ¬pevaluatesalsoto [0,1],since v∗ (¬p) = [1 −<br />

v∗ 2 (p),1−v∗ 1 (p)] = [0,1]andhence v(p∨¬p) = [max(0,0),max(1,1)] = [0,1].<br />

Stickingwiththe‘impreciseknowledge’interpretation,theresultinginterval<br />

shouldreflectmyknowledgeaboutmyexpectedlossifIplayaccordingto<br />

anoptimalstrategy.However,while 1 − v∗ 2 (p ∨ ¬p) = 0isthecorrectlower<br />

boundonmyexpectedlossafterperformingtherelevantinstanceof Ep,to<br />

requirethat 1 − v∗ 1 (p ∨ ¬p) = 1isthebestupperboundforthelossthat<br />

Ihavetoexpectwhenplayingthegameisproblematic. Whenplayinga<br />

mixedstrategythatresultsinmyassertionofeither porof ¬pwithequal<br />

probability,thenmyresultingexpectedlossis0.5€,not1€.<br />

Weintroducesomenotationtoassistprecisestatementsabouttherelationbetweentheintervalbasedsemanticsof(Estevaetal.,1994)andGiles’s<br />

9 Thesemightwellbepurelysubjectiveprobabilitiesthatmaydifferforthetwoplayers.<br />

ToproveTheorem1oneonlyhastoassumethatIcanactonassignedprobabilitiesthat<br />

determine‘myexpectation’ofloss.


60 ChristianG.Fermüller<br />

game.Let v ∗ beanintervalassignment,i.e.anassignmentofclosedintervals<br />

⊆ [0,1]tothepropositionalvariablesPV.Then v ∗ L denotestheextensionof<br />

v ∗ fromPVtoarbitraryformulasviathetruthfunctions ⇒ ∗ E forimplicationandthepoint-wisegeneralizationsof<br />

max, min,and ∗Lfordisjunction,<br />

weakconjunction,andstrongconjunction,respectively. Callanyassignment<br />

vofreals ∈ [0,1]compatiblewith v ∗ if v(p) ∈ v ∗ (p)forall p ∈PV.<br />

Thecorrespondingriskvalueassignment 〈·〉 r v,definedby 〈p〉 r v = 1 − v(p),is<br />

alsocalledcompatiblewith v ∗ .<br />

Proposition1.If,givenanintervalassignment v∗ ,theformula Fevaluatesto<br />

v∗ L (F) = [a,b]thenthefollowingholds:<br />

∗ForthegameinSection3,playedon F:All(pure)strategiesforme<br />

thatareoptimalwithrespecttosomefixedriskvalueassignment 〈·〉 r v<br />

compatiblewith v ∗ resultinanexpectedlossofatmost (1 − a)€,but<br />

atleast (1 − b)€.<br />

Notethatintheabovestatementmyexpectedlossreferstoariskvalue<br />

assignment 〈·〉 r vthatisfixedbeforethedialoguegamebegins. Iwillplay<br />

optimallywithrespecttothisassignment.Sincethecorrespondingexpected<br />

pay-offisallthatmattershere,wetechnicallystillhaveagameofperfect<br />

informationandthereforenogeneralityislostbyrestrictingattentionto<br />

purestrategies. Theboundsgivenby v∗ Lformyexpectedlossarenot optimalingeneral. Inotherwords,theinversedirectionofProposition1<br />

doesnothold.Toseethis,consideragaintheintervalassignment v∗ (p) =<br />

[0,1]resultingin v∗ L (p ∨ ¬p) = [0,1]. Obviously,Icannotloosemorethan<br />

1€,evenifIplaybadly,butmyexpectedlossunderanyfixedriskvalue<br />

isnevergreaterthan 0.5€ifIplayoptimallywithrespect<br />

assignment 〈·〉 r v<br />

to 〈·〉 r v .<br />

Ontheotherhand,stickingwithourexample‘p ∨ ¬p’,onecanobserve<br />

thatthebestupperboundformylossisindeed1€ifIdonotknowtherelevantriskvaluesandIstillhavetostickwithsomepurestrategy.Thisisbecausethechosenstrategymightsuggesttoassert<br />

pevenif,unknowntome,<br />

theexperiment Epalwayshasanegativeresult.Inotherwords,thebounds 1<br />

and 0areoptimalnowandcoincidewiththelimitsof v∗ L (p ∨ ¬p).However,<br />

ingeneral,thisscenario—playingapurestrategyreferringtoriskvalues<br />

thatneednotcoincidewiththeriskvaluesusedtocalculatetheexpected<br />

pay-off—mayleadtoanexpectedlossoutsidetheintervalcorresponding<br />

to v∗ L .Forasimpleexampleconsider p ∨q,where v∗ (p) = [0.4,0.4],i.e.,the<br />

playersknowthattheexpectedlossassociatedwithanassertionof pis 0.6€,<br />

and v∗ (q) = [0,1],i.e.,theriskassociatedwithasserting qcanbeanyvalue<br />

between 1and 0.Wehave v∗ L (p ∨ q) = [max(0,0.4),max(0.4,1)] = [0.4,1].<br />

Undertheassumptionthat 〈q〉 r v = 0,whichiscompatiblewith v ∗ (q),my<br />

beststrategycallsforasserting qinconsequenceofasserting p ∨ q. But


TruthValueIntervals 61<br />

ifthestate [�q]isevaluatedusingtheriskvalue 〈q〉 r v = 1,whichisalso<br />

compatiblewith v∗ (q),thenIhavetoexpectasurelossof 1€,although<br />

1 − 1 = 0isoutside [0.4,1].<br />

5 Cautiousandboldbettingonunstablepropositions<br />

Wesuggestthatamoreconvincingjustificationoftheformalsemantics<br />

of(Estevaetal.,1994)arisesfromthefollowingalternativegamebased<br />

modelofreasoningunderimpreciseknowledge.Likeabove,let v∗beanas signmentofintervals ⊆ [0,1]tothepropositionalvariables.Again,weleave<br />

thedialoguepartofGiles’sgameunchanged.Butinreferencetothepartial<br />

informationrepresentedby v∗ ,weassigntwodifferentsuccessprobabilities<br />

toeachexperiment Eqcorrespondingtoapropositionalvariable q,reflecting<br />

whether qisassertedbymeorbyyouandconsiderbestcaseandworstcase<br />

scenarios(frommypointofview)concerningtheresultingexpectedpay-off.<br />

Moreprecisely,myexpectedlossforthefinalstate [p1,... ,pm�q1,... ,qn]<br />

whenevaluated v∗-cautiouslyisgivenby n�<br />

〈qi〉<br />

i=1<br />

r m�<br />

h − 〈pi〉<br />

i=1<br />

r l ,butwhenevaluated<br />

v∗-boldlyitisgivenby n� m�<br />

−<br />

〈qi〉<br />

i=1<br />

r l 〈pi〉<br />

i=1<br />

r h ,wheretheriskvalues 〈q〉r h<br />

and 〈q〉 r laredeterminedbythelimitsoftheinterval v∗ (q) = [a,b]asfollows:<br />

〈q〉 r h = 1 − aand 〈q〉r l = 1 − b.<br />

Proposition2.Givenanintervalassignment v ∗ ,thefollowingstatements<br />

areequivalent:<br />

(i)Formula Fevaluatesto v∗ L (F) = [a,b].<br />

(ii)ForthedialoguegameinSection3,playedof F:ifelementarystates<br />

areevaluated v ∗ -cautiouslythentheminimalexpectedlossIcanachievebyanoptimalstrategyis<br />

(1−b)€;ifelementarystatesareevaluated<br />

v ∗ -boldlythenmyoptimalexpectedlossis (1 − a)€.<br />

6 Conclusion<br />

Wehavebeenmotivatedbyvariousproblemsthatarisefrominsistingon<br />

truthfunctionalityforaparticulartypeoffuzzylogicintendedtocapture<br />

reasoningunder‘impreciseknowledge’. Mostimportantlyforthecurrent<br />

purpose,wehaveemployedadialoguecumbettinggameapproachtomodel<br />

logicalinferenceinacontextof‘dispersiveexperiments’fortestingthetruth<br />

ofatomicassertions.Thisanalysisnotonlyleadstodifferentcharacterizationsofanimportantintervalbasedfuzzylogic,butrelatesconcernsabout<br />

propertiesoffuzzylogicstoreflectionsonrationalityquaplayingoptimally<br />

inadequategamesfor‘approximatereasoning’.


62 ChristianG.Fermüller<br />

ChristianG.Fermüller<br />

InstitutfürComputersprachen,TUWien<br />

Favoritenstraße9–11,A–1040Vienna,Austria<br />

chrisf@logic.at<br />

http://www.logic.at/staff/chrisf/<br />

References<br />

Belnap,N.D.(1977).Ausefulfour–valuedlogic.InG.Epstein&J.M.Dunn<br />

(Eds.),Modernusesofmultiple–valuedlogic(pp.8–37).<br />

Ciabattoni,A.,Fermüller,C.G.,&Metcalfe,G. (2005). Uniformrulesand<br />

dialoguegamesforfuzzylogics.In(pp.496–510).SpringerVerlag.<br />

Cornelis,C.,Arieli,O.,Deschrijver,G.,&Kerre,E.E. (2007). Uncertainty<br />

modelingbybilattice–basedsquaresandtriangles. IEEETransactionsonfuzzy<br />

Systems,15(2),161–175.<br />

Cornelis,C.,Deschrijver,G.,&Kerre,E.E.(2004).Implicationinintuitionistic<br />

andinterval–valuedfuzzysettheory:construction,classification,application.Intl.<br />

J.ofApproximateReasoning,35,55–95.<br />

Cornelis,C.,Deschrijver,G.,&Kerre,E.E.(2006).Advancesandchallengesin<br />

interval–valuedfuzzylogic.FuzzySetsandSystems,157(5),622–627.<br />

Dubois,D.(n.d.).Onignoranceandcontradictionconsideredastruth-values.(To<br />

appearintheLogicJournaloftheIGPL.)<br />

Esteva,F.,Garcia-Calvés,P.,&Godo,L. (1994,March). Enrichedinterval<br />

bilattices:Anapproachtodealwithuncertaintyandimprecision.International<br />

JournalofUncertainty,FuzzinessandKnowledge-BasedSystems,1,37–54.<br />

Felscher,W.(1985).Dialogues,strategies,andintuitionisticprovability.Annals<br />

ofPureandAppliedLogic,28,217–254.<br />

Fermüller,C.G.(2003).Theoriesofvaguenessversusfuzzylogic:Canlogicians<br />

learnfromphilosophers?NeuralNetworkWorldJournal,13(5),455–466.<br />

Fermüller,C.G.(2004,October).RevisitingGiles’sgame:Reconcilingfuzzylogic<br />

andsupervaluation. (ToappearinLogic,GamesandPhilosophy: Foundational<br />

Perspectives,PragueColloquiumOctober2004.)<br />

Fermüller,C.G.(2007).Exploringdialoguegamesasfoundationoffuzzylogic.In<br />

M. ˇ Stěpničkaetal.(Eds.),Newdimensionsinfuzzylogicandrelatedtechnologies.<br />

Proceedingsofthe 5 th EUSFLATconference(Vol.I,pp.437–444).<br />

Fermüller,C.G.,&Kosik,R.(2006).Combiningsupervaluationanddegreebased<br />

reasoningundervagueness(No.4246).SpringerVerlag.<br />

Giles,R.(1974).Anon-classicallogicforphysics.StudiaLogica,33(4),399–417.<br />

Giles,R.(1976).Łukasiewiczlogicandfuzzysettheory.InternationalJournalof<br />

Man-MachineStudies,8(3),313–327.


TruthValueIntervals 63<br />

Giles,R. (1977). Anon-classicallogicforphysics. InR.Wojcicki&G.Malinkowski(Eds.),SelectedpapersonŁukasiewiczsententialcalculi(pp.13–51).<br />

Ossolineum:PolishAcademyofSciences.<br />

Ginsberg,M.L.(1988).Multivaluedlogics:auniformapproachtoreasoningin<br />

artificialintelligence.ComputationalIntelligence,4(3),265–316.<br />

Hájek,P. (n.d.). Onafuzzylogicwithimprecisetruthvalue. (Unpublished<br />

manuscript.)<br />

Hájek,P.(1998).Metamathematicsoffuzzylogic.Dordrecht:Kluwer.<br />

Hájek,P. (2002). Whyfuzzylogic? InD.Jacquette(Ed.),Acompanionto<br />

philosophicallogic(pp.595–606).Oxford:Blackwell.<br />

Keefe,R.,&Rosanna. (2000). Theoriesofvagueness. Cambridge:Cambridge<br />

UniversityPress.<br />

Keefe,R.,&Smith,P.(Eds.). (1987).Vagueness:Areader. Cambridge,MA–<br />

London:MITPress.<br />

Lorenzen,P.(1960).LogikundAgon.InAtticongr.internat.difilosofia(Vol.4,<br />

pp.187–194).Sansoni,Firenze.<br />

Łukasiewicz,J.(1920).Ologicetròjwartościowej.RuchFilozoficzny,5,169–171.<br />

Shapiro,S.(2006).Vaguenessincontext.Oxford:OxfordUniversityPress.<br />

Zadeh,L.A.(1975).Fuzzylogicandapproximatereasoning.Synthese,30(3–4),<br />

407–428.<br />

Zadeh,L.A.(1988).Fuzzylogic.IEEE:Computer,21(4),83–93.


Procedural Semantics for Mathematical Constants<br />

1 Introduction<br />

Bjørn Jespersen Marie Duˇzí ∗<br />

Considernumericalconstantslike‘1’and‘π’.Whatistheirsemantics?We<br />

aregoingtoargueinfavourofarealistproceduralsemantics,accordingto<br />

whichsenseanddenotationarecorrelatedasprocedureandproduct.Soit<br />

isobviousthatourproceduralsemanticsbearssimilaritiestoMoschovakis’s<br />

asbasedonalgorithmandvalue. WeareinoppositiontoKripke’sunrealisticrealistcontentionthatthesemanticsof‘π’consistsinnothingother<br />

than‘π’rigidlydenoting π.Yes,‘π’doesdenote π—indeed,‘π’qualifies<br />

asastronglyrigiddesignatorof π,cf.(Kripke,1980,p.48)—butthere<br />

issubstantiallymoretothesemanticsof‘π’thanmerelythedenotation<br />

relation.Inthispaperwefocuson‘π’,sinceourgeneraltop-downstrategy<br />

istodevelopasemanticsforthehardest(oraveryhard)caseandthen<br />

generalisedownwardstoincreasinglylesshardcasesfromthere.<br />

Inoutline,ourproceduralsemanticssaysthat‘π’expressesasitssensea<br />

procedurewhoseproductis π.Theprocedureis,asamatterofmathematicalconvention,adefinitionofπandtheproductis,asamatterofmathematicalfact,the(transcendental)numbersodefined.Forcomparison,‘1’<br />

expressesasitssensetheprocedureconsistinginapplyingthesuccessor<br />

functiontozeroonceanddenoteswhatever(natural)numberemergesas<br />

theproductofthisprocedure.<br />

Theupsideofaproceduralsemanticsfor‘π’isthattounderstand,asa<br />

readerorhearer,andexerciselinguisticcompetence,asawriterorspeaker,<br />

onemustmerelyunderstandaparticularnumericaldefinitionandneednot<br />

knowwhichnumberitdefines.Proceduralsemantics,whetherrealistoridealist,construessenseasanitinerariomentisabstractingfromtheitinerary’s<br />

destination. Makingthedenotationofanumericalconstantirrelevantto<br />

∗ ThisworkissupportedbyGrantNo.401/07/0451,SemantisationofPragmatics,ofthe<br />

GrantAgencyoftheCzechRepublic.


66 BjørnJespersen&MarieDuˇzí<br />

understandingandlinguisticcompetenceisnotpressinginthecaseof‘1’,<br />

butitissointhecaseof‘π’.Thedownside,however,isthatatleasttwo<br />

equivalent,butobviouslydistinct,definitionsof πarevyingfortheroleas<br />

thesenseof‘π’.Oneistheratioofacircle’sareaanditsradiussquared;the<br />

otheristheratioofacircle’scircumferencetoitsdiameter.Theyareequivalent,becausethesamenumberisharpoonedbybothdefinitions.Butthe<br />

proceduresareconceptuallydifferent,sotheyshouldnotbothbeassigned<br />

to‘π’asitssenseonpainofinstallinghomonymy.Thiskindofpredicament<br />

hasbecomehistoricallyfamous.SaysFrege,<br />

SolangenurdieBedeutungdieselbebleibt,lassensichdieseSchwankungendesSinnesertragen,wiewohlauchsieindemLehrgebäudeeinerbeweisendenWissenschaftzuvermeidensindundineinervollkommenenSprachenichtvorkommendürften.<br />

(Frege,1892,n.2,<br />

p.42)<br />

Weshallsuggestasolutiontothispredicament.Thecrustofthesolution<br />

istorelegateeachdefinitionof πtoindividualconceptualsystems. Since<br />

aninterpretedsignsuchas‘π’isapairwhoseelementsareacharacter(in<br />

thiscasetheGreekletter‘π’)andasense,therewillbeasmanysuchpairs<br />

asthereareconceptualsystemsdefining π.Disambiguationof‘π’-involving<br />

discoursewillconsistinmakingexplicitwhichparticular π-definingsystem<br />

shouldsupplythesenseofatokenofthecharacter‘π’.<br />

Arelatedpredicament,whichweshallalsoaddress,iswhether‘π’isbest<br />

construedasanamefor πorasashorthandforadefinitedescription.Ifa<br />

name,thesenseof‘π’will,inoursemantics,beaprimitiveprocedureconsistingintheinstructiontoobtain,oraccess,<br />

πinonestep.Theprocedure<br />

willnottellushowtoobtain π,butonlythat πistobeobtained. This<br />

doesnotsitwellwith πbeingsomethingascomplicatedasatranscendental<br />

number. Butitdoessitwellwith‘π’beingitselfaprimitive,orsimple,<br />

characternotdisclosinganyinformationaboutitsdenotation. Soatleast<br />

onaliteralanalysis,accordingtowhichsyntacticandsemanticstructures<br />

arebyandlargeisomorphic,‘π’shouldbepairedoffwithanon-compound<br />

sense.If‘π’isadefinitedescription(indisguise),thesenseof‘π’will,inour<br />

semantics,beacompoundprocedureconsistingintheinstructiontomanipulatevariousmathematicaloperationsandconceptsinordertodefinea<br />

number. Onlytheproblem,aswejustpointedout,is,whichprocedure?<br />

Isittheinstructiontocalculatetheratioofacircle’sareaanditsradius<br />

squared,orisittheinstructiontocalculatetheratioofacircle’scircumferenceanditsdiameter,orisitsomeyetotherinstruction?Whicheverit<br />

maybe,though,thegrammaticalconstant‘π’willbesynonymouswiththe<br />

definitedescription‘theratio...’chosen.Theproblemofhomonymydoes<br />

notrearitsheadincasethesenseof‘π’isaprimitiveprocedure,forthen


ProceduralSemanticsforMathematicalConstants 67<br />

‘π’isonlyequivalent(co-denoting)withaparticulardefinition. Infact,<br />

sinceallthevariantsofdefinitionsco-denotethesamenumber,‘π’willbe<br />

equivalentwithallsuchdescriptions.<br />

Ourunderlyingsemanticschemaisdepictedinthefollowingfigure.<br />

procedure<br />

(sense)<br />

expresses<br />

constant<br />

produces<br />

Semanticschema<br />

denotes(names)<br />

denotedentity<br />

(ifany)<br />

Therelationaprioriofexpressingasobtainingbetweenconstantandsense<br />

exhauststhepuresemanticsoftheconstant. Assoonasaprocedureis<br />

explicitlygiven,itsproduct(ifany)isimplicitlygiven,fortherelationfrom<br />

proceduretoproductisaninternalone: aprocedurecanhaveatmost<br />

oneproduct,andthatproductisinvariant.Theprocedurewillproduceits<br />

productindependentlyofanyalgorithm;thisiswhytherelationbetween<br />

procedureandproductisaninternalone.Butforepistemologicalreasons<br />

wewillneedsomewayorotherofcalculatingitsproducttolearnwhatit<br />

is,soweneedaπ-calculatingalgorithmtoshowuswhatnumbersatisfies<br />

whatever π-definingcondition.Suchanalgorithmwill,ipsofacto,revealto<br />

uswhatthedenotationof‘π’is.Thenumber 3.14159...whichis πisitself<br />

noplayerinthepuresemanticsof‘π’. πisjustwhatevernumberrollsout<br />

asthevalueofthegivenprocedure.Thenumber 3.14159...isitselfoflittle<br />

mathematicalinterestandofnosemanticimport.Thepropertiesof π,by<br />

contrast,areofgreatinterest;e.g.,whether πisnormalinsomebase;and<br />

establishingthat πistranscendental(andnotjustrational)wasamajor<br />

mathematicalachievement.<br />

Analgorithmmayappearinoneoftwocapacities. Eitheritisanintermediarybetweenthedefinitionandthenumbersodefined:<br />

thenthe<br />

algorithm(whicheveritis)isnoplayerinthepuresemanticsof‘π’. Or<br />

analgorithmistheverysenseof‘π’:thenthealgorithmisaplayerinthe<br />

puresemanticsof‘π’.Ourproceduralsemanticsallowsthataπ-calculating<br />

algorithmmayitselfbeelevatedtoplayingtheroleofsenseof‘π’.Insucha<br />

case‘π’willhaveasitssenseoneparticularwayofcalculating π.Analgorithmisaparticularkindofprocedureandcanassuchfigureasalinguistic<br />

senserelativetoaproceduralsemantics.<br />

Intheformercase,ifthedefinitionisaconditionthenthealgorithm<br />

willcalculatethesatisfierofthecondition.Fullcompetencewithrespectto


68 BjørnJespersen&MarieDuˇzí<br />

thedefinitiontheratio...willyieldknowledgeofaconditiontobesatisfied<br />

byarealnumber,butwillnotyieldknowledgeofwhichnumbersatisfies<br />

it. Sothedefinitionis,strictlyspeaking,adefinitionofsomethingfora<br />

numbertobe;namely,theratiooftwogeometricproportions.Ifthesense<br />

defines πastheratiobetweentheareaofacircleanditsradiussquared,<br />

amatchingalgorithmmustcalculatethisratio.Fulllinguisticcompetence<br />

withrespectto‘π’neitherpresupposes,norneedinvolve, knowledgeof<br />

howtocalculate π.Whatcompetenceconsistsindependsonwhetherthe<br />

senseof‘π’isaprimitiveorcompoundprocedure.Ifprimitive,competence<br />

requiresknowingwhichtranscendentalreal 3.14159... is π.Ifcompound,<br />

competencerequiresunderstandingtheconcepttheratioof,aswellaseither<br />

theconceptstheareaof,theradiusof,thesquareof,ortheconceptsthe<br />

circumferenceof andthediameterof,togetherwithknowledgeofhowto<br />

mathematicallymanipulatethem. Aschoolchildwillunderstandsucha<br />

complexprocedure;ittakesaprofessionalmathematiciantodevelopand<br />

comprehendaπ-calculatingalgorithm.Thetaskfacingthemathematician<br />

istocomeupwithanalgorithmequivalentwiththedefinitiondefiningthe<br />

givenratio.<br />

Inthelattercase,whereanalgorithmisthesenseof‘π’,fulllinguisticcompetencewithrespectto‘π’istounderstandadefinitionof<br />

πand,<br />

again,notofthenumbersodefined.Butsincethealgorithmisnownotan<br />

intermediarybetweendefinitionandnumber,linguisticcompetencewillbe<br />

hardertocomeby,sincethesenseof‘π’isnowlikelytoinvolvemuchmore<br />

complicatedmathematicalnotionsthanjust,say,thoseofratio,area,and<br />

circumference,suchasthelimitofaninfiniteseries.<br />

2 BeyondBenacerraf<br />

Assumethatthetruth-conditionof“...π...”requires πtoexistasanindependent,abstractentity.Assume,further,thatwecanhavenoepistemic<br />

accesstoentitiesthatwecanhavenocausalinteractionwith. Thennext<br />

stopisBenacerraf’sdilemmaasformulatedfor π: wedonotknowwhat<br />

numberis π;yetwewanttodub π‘π’inordertotalkabout πin“...π...”.<br />

Sohowis‘π’tobeintroducedintomathematese?Moreover,nowthat‘π’<br />

hasactuallybeenintroducedintostandardmathematicalvocabularyand<br />

beeninuseforthreehundredyears,whatwouldarealist(asopposedto<br />

constructivistorotherwiseidealist)construalofitssemanticslooklike?<br />

WeproposeplacingourproceduralsemanticswithinthegeneralFregean<br />

programmeofexplicatingsense(Sinn)asthemodeofpresentation(Artdes<br />

Gegebenseins)oftheentity(Bedeutung)thatasensedetermines.Muskens<br />

correctlypointsoutthat“TheideawasprovidedwithextensivephilosophicaljustificationinTich´y[(1988)]”andthat“[Tich´y’s]notionofsensesas


ProceduralSemanticsforMathematicalConstants 69<br />

constructions essentiallycapturesthesameidea.” (Tich´y,2004,p.474)<br />

GoingwiththisFregeanprogramme,however,raisesabatchofquestions<br />

deservinganddemandingtobeanswered.Justhowfinelyaresensessliced?<br />

Whatistheontologicalstatusofasense?Whatdoesasense‘looklike’;in<br />

particular,whatisitsstructure? Andhowdoesasensedeterminesomething?<br />

WeagreewithMoschovakis’conceptionofsense(‘referentialintension’,<br />

inhisvernacular)as‘an(abstract,idealized,notnecessarilyimplementable)<br />

algorithmwhichcomputesthedenotationof[aterm]’(Moschovakis,2006,<br />

p.27);seealso(Moschovakis,1994). 1<br />

Moschovakisoutlineshisconceptionthus:<br />

Thestartingpoint... [is]theinsightthatacorrectunderstanding<br />

ofprogramminglanguagesshouldexplaintherelationbetweenaprogramandthealgorithmitexpresses,sothatthebasicinterpretation<br />

schemeforaprogramminglanguageisoftheform<br />

program P →algorithm(P) →den(P).<br />

Itisnothardtoworkoutthemathematicaltheoryofasuitably<br />

abstractnotionofalgorithmwhichmakesthiswork;andoncethis<br />

isdone,thenitishardtomissthesimilaritywiththebasicFregean<br />

schemefortheinterpretationofanaturallanguage,<br />

term A →meaning(A) →den(A).<br />

Thissuggestedatleastaformalanalogybetweenalgorithmsand<br />

meaningswhichseemedworthinvestigating,andprovedaftersome<br />

worktobemorethanformal:whenweviewnaturallanguagewitha<br />

programmer’seye,itseemsalmostobviousthatwecanrepresentthe<br />

meaningofaterm AbythealgorithmwhichisexpressedbyAand<br />

whichcomputesitsdenotation.(Moschovakis,2006,p.42)<br />

Inmodernjargon,TILbelongstotheparadigmofstructuredmeaning.<br />

However,Tich´ydoesnotreducestructuretoset-theoreticsequences,as<br />

1 Moschovakis’notionofalgorithmbordersonbeingtoopermissive,sincealgorithmsare<br />

normallyunderstoodtobeeffective.(See(Cleland,2002)fordiscussion.)Tich´yseparates<br />

algorithmsfromconstructions:“Thenotionofconstructionis... correlativenotwiththe<br />

notionofalgorithmitselfbutwithwhatisknownasaparticularalgorithmiccomputation,<br />

thesequenceofstepsprescribedbythealgorithmwhenitisappliedtoaparticularinput.<br />

Butnoteveryconstructionisanalgorithmiccomputation.Analgorithmiccomputation<br />

isasequenceofeffectivesteps,stepswhichconsistinsubjectingamanageableobject...<br />

toafeasibleoperation. Aconstruction,ontheotherhand,mayinvolvestepswhichare<br />

notofthissort.”(Moschovakis,1994,p.526),(Moschovakis,2006,p.613)


70 BjørnJespersen&MarieDuˇzí<br />

doKaplanandCresswell. 2 NordoesTich´yfailtoexplainhowthesense<br />

ofamoleculartermisdeterminedbythesensesofitsatomsandtheir<br />

syntacticarrangement(asMoschovakisobjectsto‘structural’approaches<br />

in(Moschovakis,2006,p.27)).<br />

Ingeneral,aprocedureisastructureencompassingoneormoresteps<br />

thatindividuallydetailhowtodetermineaproductandjointlydetailhow<br />

todeterminetheproductoftheprocedurethattheyaresub-proceduresof.<br />

(Thisholdsevenforone-stepprocedures.)Structuresareneededasmolecularunitsinwhichtoorganiseatomicsub-proceduresinaparticularorder.Acompoundstructureconstitutesahierarchyofsub-procedures.Thephilosophicalideainformingourproceduralsemanticsisthatsincesensesareprocedures,anytwosensesareidenticaljustwhentheyare,roughlyspeaking,procedurallyindistinguishable.<br />

(Weshallindividuatesensesinterms<br />

ofproceduralisomorphism;seebelow.)Intuitively,anytwoproceduresare<br />

identicaljustwhentheyareinstructionstodothesametothesamethings<br />

inthesameorder.<br />

3 Logicalfoundations<br />

TILconstructionsareprocedures. Constructionsdivideintoatomicand<br />

compound,accordingastheyencompassoneormoresteps. Theatomic<br />

onesareVariableandTrivialization;thecompoundones,Compositionand<br />

Closure. 3 Avariable xconstructsanobjectrelativetoavaluationfunction<br />

pairingvariablesandentitiesoff,suchthat xconstructsthevalueassigned<br />

toit.TheTrivialization 0 Xconstructstheentity X(whichmaybewhatever<br />

sortofentityfoundintheontologyofTIL).ACompositionistheprocedure<br />

ofapplyingafunctionatoneofitsargumentstoobtainthevalue(ifany)<br />

atthatargument;thefunctionalvalueistheproductofthatprocedure.A<br />

Closureistheprocedureofarrangingobjects x1,...,xnand yasfunctional<br />

argumentsandvalues,respectively;theresultingfunctionistheproductof<br />

thatprocedure. Ifthesenseof‘π’issimple,itssenseistheTrivialization<br />

of π: 0 π.Ifcomplex,itisaComposition.Ineithercasetheproductofthe<br />

respectiveprocedureisthesametranscendentalnumber.<br />

2 Kaplanmaywellhavebeentheonetoreintroducethenotionofstructuredmeaning<br />

intomainstreamanalyticphilosophyoflanguage. See(Kaplan,1978),writtenin1970;<br />

butseealso(Lewis,1972).(Cresswell,1985)hasbecomethestandardpointofreference.<br />

Allthreeagreethatstructure,especiallyastructuredproposition,is(orcanbemodelled<br />

as)anordered n-tuple.Thiswon’tdo,though,sincesequencesunderdeterminestructure<br />

andsocannotsolveRussell’soldproblemofpropositionalunity.<br />

3 Andfourothers—Execution,DoubleExecution,Tuple,Projection—thatwedonot<br />

needhere.


ProceduralSemanticsforMathematicalConstants 71<br />

Herefollowsanoutlineofthelogicalbackboneofourproceduralsemanticsfor‘π’.<br />

TILconstructions,aswellastheentitiestheyconstruct,all<br />

receivealogical(asopposedtolinguistic)type.<br />

Definition1(Typeoforder1).<br />

Let Bbeabase,whereabaseisacollectionofpair-wisedisjoint,non-empty<br />

sets.Then:<br />

(i)Everymemberof Bisanelementarytypeoforder 1over B.<br />

(ii)Let α, β1,...,βm(m > 0)betypesoforder 1over B. Thenthe<br />

collection (αβ1 ...βm)ofall m-arypartialmappingsfrom β1×· · ·×βm<br />

into αisafunctionaltypeoforder 1over B.<br />

(iii)Nothingisatypeoforder 1over Bunlessitsofollowsfrom1and2.<br />

Definition2(Construction).<br />

(i)TheVariable xisaconstructionthatconstructsanobject Oofthe<br />

respectivetypedependentlyonavaluation v;it v-constructs O.<br />

(ii)Trivialization: Where Xisanobjectwhatsoever(anextension,an<br />

intensionoraconstruction), 0 XistheconstructionTrivialization.It<br />

constructs Xwithoutanychange.<br />

(iii)TheComposition [XY1 ...Ym]isthefollowingconstruction.<br />

If X v-constructsafunction fofatype (αβ1 ...βm),and Y1 ... Ym<br />

v-constructentities B1,...,Bmoftypes β1,... ,βm,respectively,then<br />

theComposition [XY1 ...Ym] v-constructsthevalue(anentity,ifany,<br />

oftype α)of fonthetuple-argument 〈B1,... ,Bm〉. Otherwisethe<br />

Composition [XY1 ...Ym]doesnot v-constructanythingandsois vimproper.<br />

(iv)TheClosure [λx1 ...xmY ]isthefollowingconstruction.<br />

Let x1,x2,... ,xmbepairwisedistinctvariables v-constructingentitiesoftypes<br />

β1,...,βmand Y aconstruction v-constructingan αentity.Then<br />

[λx1 ...xmY ]istheconstruction λ-Closure(orClosure).<br />

It v-constructs thefollowingfunction f oftype (αβ1 ... βm). Let<br />

v(B1/x1,...,Bm/xm)beavaluationidenticalwith vatleastupto<br />

assigningobjects B1,... ,Bmoftypes β1,... ,βm,respectively,tovariables<br />

x1,... ,xm. If Y is v(B1/x1,... ,Bm/xm)-improper(see(iii)),<br />

then fisundefinedon 〈B1,... ,Bm〉. Otherwisethevalueof fon<br />

〈B1,...,Bm〉istheentityoftype α v(B1/x1,... ,Bm/xm)-constructed<br />

by Y.<br />

(v)Nothingisaconstruction,unlessitsofollowsfrom(i)through(iv).


72 BjørnJespersen&MarieDuˇzí<br />

Definition3(Ramifiedhierarchyoftypes).<br />

Let Bbeabase.Then:<br />

T1(typesoforder 1):definedbyDefinition1.<br />

Cn(constructionsoforder n)<br />

(i)Let xbeavariablerangingoveratypeoforder n. Then xisa<br />

constructionoforder nover B.<br />

(ii)Let X beamemberofatypeoforder n. Then 0 X, 1 X, 2 X are<br />

constructionsoforder nover B.<br />

(iii)Let X,X1,...,Xm(m > 0)beconstructionsoforder nover B.Then<br />

[XX1 ... Xm]isaconstructionoforder nover B.<br />

(iv)Let x1,...,xm,X(m > 0)beconstructionsoforder nover B.Then<br />

[λx1 ...xmX]isaconstructionoforder nover B.<br />

(v)Nothingisaconstructionoforder nover Bunlessitsofollowsfrom<br />

Cn(i)–(iv).<br />

Tn+1(typesoforder n + 1)<br />

Let ∗nbethecollectionofallconstructionsoforder nover B.<br />

(i) ∗nandeverytypeoforder naretypesoforder n + 1.<br />

(ii)If 0 < mand α,β1,... ,βmaretypesoforder n + 1over B,then<br />

(αβ1 ...βm)(seeT12)isatypeoforder n + 1over B.<br />

(iii)Nothingisatypeoforder n + 1over Bunlessitsofollowsfrom(i)<br />

and(ii).<br />

Theontologicalstatusofaconstructionisanobjective,abstract,structuredprocedureresidinginaPlatonicrealm.Constructionsarenotinherentlylinguisticsenses,fortheyexistpriortoandindependentlyoflanguage.<br />

Buttheymaybemade,vialinguisticconvention,toserveaslinguistic<br />

senses. Thatis,intruerealistfashion,TILconsiderslanguageacode. 4<br />

Programmaticallystated,oursemanticsfor‘π’complementstheontology<br />

for πputforwardin(Brown,1990).<br />

Aconstructiondetermineswhatitconstructsbyconstructingit.Sothe<br />

logicofdeterminationconsistsintheconstructionaldescentfromaproceduretoitsproduct,asspecifiedforeachparticularkindofconstruction<br />

inDefinition2. Constructionsaretoofinelyindividuatedtofigureaslinguisticsenses,sincesomeoftheproceduraldifferencestheyembodyare<br />

logicallyinsignificantandarenotencodedlinguistically. Mostobviously,<br />

4 See(Tich´y,1988,pp.228ff.).


ProceduralSemanticsforMathematicalConstants 73<br />

two α-equivalentconstructionslike λx[ 0 > x 0 0]and λy[ 0 > y 0 0]arejustthat<br />

—twoconstructionsoftheclassofpositivenumbersandnotone;yetthe<br />

differencebetweenthe λ-boundvariables xand yisprocedurallyirrelevant.<br />

Thesolutiontothegranularityproblemconsistsinformingequivalence<br />

classesofprocedurallyisomorphicconstructionsandprivilegingamember<br />

ofeachsuchclassastheproceduralsenseofagivenunambiguoustermor<br />

expression.Technically,thequestisforasuitabledegreeofextensionality<br />

inthe λ-calculus. Needlesstosay,itremainsanopenresearchquestion<br />

exactlywhatthedesirablecalibrationoflinguisticsensesshouldbe,butour<br />

currentthesisisthatprocedures,andhencesenses,shouldbeidentifiedup<br />

to α-and η-equivalence. 5<br />

4 Kripke’s‘π’andours<br />

CentraltoKripke’sdenotationalsemanticsisthedistinctionbetweenfixing<br />

thereferenceandgivingthemeaning/asynonym. 6 OneofKripke’sillustrationsisthis:<br />

[‘π’]isnotbeingusedasshortforthephrase‘theratioofthecircumferenceofacircletoitsdiameter’[...]Itisusedasanameforareal<br />

number,whichinthiscaseisnecessarilytheratioofthecircumference<br />

ofacircletoitsdiameter.(Kripke,1980,p.60)<br />

Kripke’ssemanticsfor‘π’issimple(simplistic,asitturnsout):<br />

‘π’<br />

rigidlydesignates<br />

Thedescription‘theratio...’ servestosingleouttheuniqueratioshared<br />

byallcircles,afterwhichthatnumberisbaptised‘π’. Thedescriptionis<br />

subsequentlykickedoffandsodoesnotformpartofthesemanticsproper<br />

ofKripke’s‘π’.Thisisproblematic.Nobodyknowsofsomeoneparticular<br />

realthatitis π.Sonobodyknowsofsomeoneparticularrealthatitisthe<br />

referenceof‘π’.Soitisobscurewhatlinguisticcompetencewithrespectto<br />

‘π’wouldconsistin.Notethatitisnotanoptiontosaythat‘π’designates<br />

whateverrealistheratioofacircle’scircumferencetoitsdiameter,forthis<br />

uniquenessconditionformsnopartofKripke’ssemanticsfor‘π’. 7 Kripke’s<br />

introductionof‘π’isimpeccable,andhis‘π’doesdenote π.Butwecannot<br />

5 See(Duˇzí,Jespersen,&Materna,ms.1, §2.2)or(Jespersen,ms). Fordiscussionof<br />

Frege’squestfortherightcalibrationofSinn,see(Sundholm,1994)and(Penco,2003).<br />

6 —adistinctionanticipatedatleastby(Geach,1969).<br />

7 TheKripkeancanhaverecoursetosomecausaltheoryofreferenceinthecaseofwords<br />

forempiricalentitiesliketigers,lemonsandgold. ButBenacerraf’sfirsthornblocks<br />

thisavenue.WesurmisethatKripkeanrigiddesignationcannotpossiblybeextendedto<br />

numericalconstantsandothertermsdenotingabstractentities.<br />

π


74 BjørnJespersen&MarieDuˇzí<br />

usehis‘π’todenote π,norcanweunderstandanyoneelse’suseof‘π’,since<br />

wecannotknowwhichparticulartranscendentalnumberis π. Inshort,<br />

Kripke’s‘π’hasbeenseveredradicallyfromanyhumanlypossiblelinguistic<br />

practice,soitisinoperative.<br />

Intheidiomofproceduralsemantics,Kripkefocusesentirelyonthe<br />

productattheexpenseoftheprocedure.Asamatterofmathematicalfact,<br />

3.14159...is π,butwhyintroduceanon-descriptivenamewhenthatname<br />

seversthelinkbetweencondition/procedureandsatisfier/product?Itseems<br />

thatonKripke’ssemanticsitwillbeadiscovery,andnotaconvention,that<br />

πistheratioofacircle’scircumferencetoitsdiameter.Ifso,italsoseems<br />

thatKripke’s‘π’misconstruesmathematicalpractice.<br />

Some π-producingproceduremustfigureinthesemanticsof‘π’;only<br />

how?TILfacesadilemmaofitsown,aswesawabove.Ontheonehand,a<br />

literalanalysisof‘π’woulddictatethatthesenseof‘π’be 0 π,yieldingthe<br />

schema<br />

‘π’ 0 π π<br />

expresses constructs<br />

Theadvantageofthisconstrualisthatwhatlookslikeaconstantisa<br />

constant(andnotadefinitedescriptionmasqueradingasone). However,<br />

thisistooclosetoKripke’s‘π’forcomfort. Wewouldbereinstatingthe<br />

problemthatthesemanticsof‘π’pairsnomathematicalconditionoffwith<br />

‘π’.Tomaster‘π’, 0 πwouldsuffice.TheTrivializationmerelyinstructsus<br />

toconstruct πandnotalsohowtoconstructit.<br />

Ontheotherhand,notleastepistemicconcernsdictatethatthesense<br />

of‘π’oughttobeanontologicaldefinitionof π,yieldingtheschema<br />

‘π’ [ιx[∀y[x = [ 0 Ratio [... y ...][... y ...]]]]] π<br />

expresses constructs<br />

(By‘ontologicaldefinition,wemeanacompoundconstruction(here,aComposition)that,inthiscase,constructsthenumber<br />

π,therebylayingdown<br />

what πis. Anontologicaldefinitioncontrastswithalinguisticdefinition,<br />

whichintroducesanewtermassynonymouswithanexistingterm.)This<br />

makes‘π’ashorthandtermsynonymouswith‘theratio...’,anditssense<br />

isanontologicaldefinitionof π.Theadvantageofthisconstrualisthatit<br />

pairsamathematicalconditionoffwith‘π’;butagain,which?Thereisno<br />

criteriontohelpdecidewhichofthepossibleontologicaldefinitionsshould<br />

bethesenseof‘π’.Itwouldbearbitrarytoselectoneandassignitassense;<br />

butassigningthemallintroduceshomonymy.<br />

Itwouldseemevidentthatalanguage-userneedstoknowatleastone<br />

definitionof πinordertouseandunderstand‘π’. Ifwegowiththe<br />

Trivialization-basedanalysisof‘π’,thefirststeptowardenhancingitis


ProceduralSemanticsforMathematicalConstants 75<br />

tomakethelogico-semanticfactthat 0 πisequivalent with [ιx[∀y[x =<br />

[ 0 Ratio [...y ... ][...y ...]]]]]partofthesemanticsof‘π’. 0 πisindifferent<br />

tohow πisconstructedbythisorthatcompoundconstruction,soasfaras<br />

theequivalencerelationgoes,anycompound π-constructionisasgoodas<br />

anyother.‘π’maybeintroducedasequivalentwith<br />

or<br />

[ιx∀y[x = [ 0 Ratio [ 0 Area y][ 0 Square[ 0 Radiusy]]]]] (1)<br />

[ιx∀y[x = [ 0 Ratio [ 0 Circumferencey][ 0 Diameter y]]]], (2)<br />

oranyothercompound π-constructingconstruction.Understandingisanothermatter.Onethingistounderstand(1);anotherthingistounderstand<br />

(2).Onemaywellknowthat‘π’isequivalenttothisCompositionwithout<br />

knowing,ipsofacto,thatitisequivalenttothatComposition.<br />

5 Realisticrealism?<br />

Bothcausaltheoryofreferenceanddenotationalsemanticsareneitherhere<br />

northereasatheoryoftermsforabstractentitiessuchasnumbers. We<br />

areputtingforwardaproceduralsemanticsasarivaltheoryinordernot<br />

togetgoredbyBenacerraf’shornsorturninglinguisticcompetencewith<br />

mathematicalconstantsintoanenigma. Wesuggest,inthefinalanalysis,<br />

thatthesemanticsof‘π’oughttobethatitisshorthandfor,andthereforesynonymouswith,adefinitedescriptionexpressingadefinitionof<br />

π<br />

anddenotingthenumbersodefined. Butforeachdefinitionnof πthere<br />

isgoingtobeapair 〈‘π’,definitionn(π)〉.Sohowdowehandletheresultinghomonymy?SchwankungendesSinnesareneitherherenorthereina<br />

regimentedlanguagesuchasmathematese. Oursolutionrevolvesaround<br />

conceptualsystems.<br />

By‘conceptualsystem’wemeanasetofconstructionsthatisfullydeterminedbythechosensetofsimpleconcepts.SimpleconceptsareTrivialisationsofnonconstructionalentitiesoforder<br />

1.Thecompoundconceptsofa<br />

conceptualsystemarethenallthecompoundconstructionsthatareformed<br />

accordingtotherulesofDefinition2(plusperhapsinvolvingadditional<br />

constructions)usingsimpleconceptsandvariables.Theexactdefinitionof<br />

conceptualsystemcanbefoundin(Materna,2004).<br />

Relativetoaparticularconceptualsystem,apair 〈‘π’,definitionn(π)〉is<br />

anunambiguousassignmentofexactlyonedefinitionof πto‘π’,provided<br />

theconceptualsystemisindependent,i.e.,itssetofsimpleconceptsis<br />

minimal.Consequently,‘π’isnotambiguous,forthischaractermustalways<br />

begiventogetherwithaparticulardefinitionof πculledfromaparticular<br />

conceptualsystem. Theappearanceofambiguityarisesonlywhentwoor


76 BjørnJespersen&MarieDuˇzí<br />

moreconceptualsystemsareinvokedinthecourseofadiscourseinwhich<br />

tokensof‘π’occur.<br />

Theupshotofoursolutionisthatthereareseveral π-denotingconstants<br />

sharingthesamefirstelement,‘π’.Sowhentwomathematiciansareboth<br />

deployingtokensof‘π’,thereisariskofthemtalkingatcrosspurposes,<br />

untilandunlesstheycomparenotesand,incaseofinvokingdifferentconceptualsystems,cometoagreeonthesamedefinitionof<br />

πintheinterestof<br />

synonymy.Yetthemathematicalresultstheymayhave πindividuallyobtainedwithrespectto<br />

πareboundtobeequivalent,foranytwodefinitions<br />

of πareboundtoconvergeinthesamenumber.Afterall,theproblemwas<br />

alwaystodowithSchwankungendesSinnesandneverSchwankungender<br />

Bedeutung. 8<br />

BjørnJespersen<br />

SectionofPhilosophy,DelftUniversityofTechnology<br />

TheNetherlands<br />

b.t.f.jespersen@tudelft.nl<br />

MarieDuˇzí<br />

DepartmentofComputerScience,VSB–TechnicalUniversityOstrava<br />

17.listopadu15,70833Ostrava,CzechRepublic<br />

marie.duzi@vsb.cz<br />

http://www.cs.vsb.cz/duzi/<br />

References<br />

Benacerraf,P.(1973).Mathematicaltruth.JournalofPhilosophy,70,661–679.<br />

Brown,J.R.(1990). πinthesky.InA.D.Irvine(Ed.),Physicalisminmathematics(pp.95–120).Dordrecht:Kluwer.<br />

Cleland,C.E.(2002).Oneffectiveprocedures.MindsandMachines,12,159–<br />

179.<br />

Cresswell,M.J.(1975).Hyperintensionallogic.StudiaLogica,34,25–38.<br />

Cresswell,M.J.(1985).Structuredmeanings.Cambridge,MA:MITPress.<br />

Duˇzí,M.,Jespersen,B.,&Materna,P.(ms.1).Proceduralsemanticsforhyperintensionallogic:FoundationsandapplicationsofTIL.(Insubmission.)<br />

8 VersionsofthispaperwerereadbyMarieattheJointParis/ArchéWorkshopPhilosophy<br />

ofMathematicsandAbstractEntities,ENS,Paris,February28 th –March1 st ,2008,and<br />

byBjørnatLOGICA’08,Hejnice,June16 th –20 th ,2008,atDepartmentofFuzzyModelling,TUOstrava,June25<br />

th ,2003,atDepartmentofPhilosophy,UniversityofMilan,<br />

April22 nd ,2008,andatDepartmentofPhilosophy,UniversityofPadua,May5 th ,2008.<br />

Portionsofthepresentpaperhavebeenliftedfrom(Duˇzíetal.,ms.1, §3.2.1).Thispaper<br />

isanabridgedversionof(Duˇzí,Jespersen,&Materna,ms.2).


ProceduralSemanticsforMathematicalConstants 77<br />

Duˇzí,M.,Jespersen,B.,&Materna,P. (ms.2). ‘π’inthesky. (ForthcominginActsofknowledge:history,philosophyandlogic.AFestschriftforGöran<br />

Sundholm,London:CollegePublicationsTributeSeries,2009.)<br />

Frege,G.(1892). ÜberSinnundBedeutung(G.Patzig,Ed.).Göttingen:VandenhoeckundRuprecht.(Reprint1986.)<br />

Geach,P.T.(1969).TheperilsofPauline.ReviewofMetaphysics,23,287–300.<br />

Jespersen,B. (ms). Hyperintensionsandproceduralisomorphism: Alternative<br />

(1/2).(K.Kijania-Placek(ed.).London:CollegePublications.)<br />

Kaplan,D.(1978).Dthat.InP.Cole(Ed.),Syntaxandsemantics.NewYork:<br />

AcademicPress.<br />

Kripke,S.(1980).Namingandnecessity.Oxford:Blackwell.<br />

Lewis,D. (1972). Generalsemantics. InD.Davidson&G.Harman(Eds.),<br />

Semanticsofnaturallanguage(pp.169–218).Dordrecht:Reidel.<br />

Materna,P.(2004).Conceptualsystems.Berlin:Logos–Verlag.<br />

Moschovakis,Y. (1994). Senseanddenotationasalgorithmandvalue. Berlin:<br />

Springer–Verlag.<br />

Moschovakis,Y.(2006).Alogicalcalculusofmeaningandsynonymy.Linguistics<br />

andPhilosophy,29,27–89.<br />

Muskens,R. (2005). Senseandthecomputationofreference. Linguisticsand<br />

Philosophy,28,473–504.<br />

Penco,G.(2003).Frege:twotheses,twosenses.HistoryandPhilosophyofLogic,<br />

24,87–109.<br />

Sundholm,B.G.(1994).Proof-theoreticalsemanticsandFregeanidentitycriteria<br />

forpropositions.TheMonist,77,294–314.<br />

Sundholm,B.G.(2000).Virtuesandvicesofinterpretedclassicalformalisms.In<br />

T.Childers&J.Palomäki(Eds.),Betweenwordsandworlds:AFestschriftfor<br />

PavelMaterna.Prague:Filosofia,CzechAcademyofSciences.<br />

Tich´y,P.(1986).Constructions.PhilosophyofScience,53,514–534.(Reprinted<br />

in(Tich´y,2004).)<br />

Tich´y,P.(1988).ThefoundationsofFrege’slogic.Berlin:deGruyter.<br />

Tich´y,P.(2004).Collectedpapersinlogicandphilosophy(V.Svoboda,B.Jespersen,&C.Cheyne,Eds.).PragueandDunedin:Filosofia,CzechAcademyof<br />

SciencesandUniversityofOtagoPress.


Neighborhood Incompatibility Semantics for<br />

Modal Logic<br />

Kohei Kishida ∗<br />

Thispaperintroducesneighborhoodsemanticsforpropositionalmodallogic<br />

intotheframeworkofBrandom’s(2008)incompatibilitysemantics.Neighborhoodsemanticsformodallogic,asitisconventionallystudied,canbe<br />

consideredtobeakindofpossible-worldsemantics,inthesensethata<br />

systemofneighborhoodscodifiesageneralizedaccessibilityrelationamong<br />

pointsofthespace,orworlds,atwhichthetruthvaluesofpropositionsare<br />

evaluated.Suchasemanticsfeaturestherepresentationalnotionsoftruth<br />

andpossibleworldasitsbasicprimitiveconstituents. Brandom’sincompatibilitysemantics,incontrast,isfoundedupontheinferentialnotionsof<br />

incoherenceandincompatibilityofsentences.Thechiefgoalofthispaperis<br />

toshowthatthisinferentialistframeworkofincompatibilitysemanticscan<br />

alsoadoptthenotionofneighborhoodtointerpretmodality,asthecoreidea<br />

ofneighborhoodsemanticsworksindependentlyoftherepresentationalnotions.<br />

1 IncompatibilitySemantics:AQuickReview<br />

Herewequicklyreviewthebasicdefinitionsandfactsinincompatibility<br />

semanticsthatarerelevanttothispaper;see(Brandom,2008)forafull<br />

expositionofthesemantics.<br />

Wewrite Lbothforagivensententiallanguageandforthesetofits<br />

sentences. Let Incbeanysubsetof PL,thepowersetof L,thatisclosed<br />

upwardintermsof ⊆,i.e.,if X ∈ Incand X ⊆ Y then Y ∈ Inc. Wesay<br />

Xisincoherentif X ∈ Inc;then Incbeing ⊆-upwardclosedmeansthat<br />

addingmoresentencestoanincoherentset Xofsentencesnevercuresthe<br />

incoherence. Wealsosay Y isincompatiblewith Xif X ∪ Y ∈ Inc,and<br />

∗ TheauthorwouldliketothankAlpAker,RobertBrandom,JaroslavPeregrin,José<br />

MartínezFernández,andespeciallyNuelBelnap,forinsightfulcommentsandhelpful<br />

suggestions.


80 KoheiKishida<br />

write I(X)forthecollectionof Y ⊆ PLincompatiblewith X,i.e.<br />

I(X) = {Y ⊆ L | X ∪ Y ∈ Inc}.<br />

When p ∈ L,wewrite I(p)for I({p}).<br />

Theentailmentrelation, p � q,isdefinedby I(q) ⊆ I(p),i.e.,thatif X<br />

isincompatiblewith qthenitisincompatiblewith p. Ingeneral, X � Y,<br />

i.e.,theconjunctionof Xentailingthedisjunctionof Y,isdefinedby<br />

�<br />

X � Y ⇐⇒ I(p) ⊆ I(X),<br />

thatis,anythingthatrulesoutall p ∈ Yrulesout X.Applyingthistothe<br />

case Y = ∅inparticular,with �<br />

I(p) = PL,wehavethefollowing,which<br />

p∈∅<br />

p∈Y<br />

agreeswithwhatisusuallymeantby X � ∅:<br />

X � ∅ ⇐⇒ PL ⊆ I(X) ⇐⇒ X ∈ Inc.<br />

When Lhasanegationoperator ¬,weassumethat Isatisfies<br />

X ∈ I(¬p) ⇐⇒ X � p forevery p ∈ L;<br />

i.e., ¬pisincompatiblewithallandonly Xthatentail p. Thenwehave<br />

I(¬¬p) = I(p),andhence X ∈ I(p) ⇐⇒ X � ¬p. Also,when Lhasa<br />

disjunctionoperator ∨, Iisassumedtosatisfy I(p ∨ q) = I(p) ∩ I(q);i.e.,<br />

Ximpliesneither pnor qisthecaseifandonlyifitdeniesboth pand q. 1<br />

Apair (L, I)ofsuch Land Iiscalledanincompatibilityframe.<br />

2 NeighborhoodSemanticsinthePossible-WorldFramework<br />

Tointroduceneighborhoodincompatibilitysemantics,itishelpfultofirst<br />

reviewtheneighborhoodsemanticsasconventionallystudiedinthepossibleworldframeworkandtothendrawaformalanalogy.<br />

Letusrecallthatpossible-worldsemanticsinterpretsamodallanguage L<br />

withasetÏofpossibleworldsbyassigningtoeachsentence p ∈ Lasubset<br />

�p�ofÏ,sometimescalledaproposition.Thenthataworld w ∈Ïliesin<br />

theinterpretation �p�of pmeans pistrueat w,andhence psemantically<br />

entails qifandonlyif �p� ⊆ �q�.So,forexample,theTaxiom ✷p ⊢ p ⊢ ✸p<br />

ofmodallogiccorrespondsto �✷p� ⊆ �p� ⊆ �✸p�.Thissuggests,fromthe<br />

pointofviewoftopology,thatthe ✷operatorcorrespondstoageneralized<br />

interioroperation,while ✸correspondstoageneralizedclosure,definedon<br />

asystemofneighborhoodsonÏasfollows.<br />

1 Theofficialdefinitionin(Brandom,2008)firstdefines I(p ∧ q) = I({p, q})andthen<br />

defines p ∨ qas ¬(¬p ∧ ¬q),whichstillentails I(p ∨ q) = I(p) ∩ I(q).


NeighborhoodIncompatibilitySemantics 81<br />

Eachworld w,orpointofthespaceÏ,isassignedacollectionofsubsets<br />

ofthespace,calledtheneighborhoodsof w;wewrite Nwforthiscollection.<br />

Then,given A ⊆ X,apoint wisintheinterior(inthegeneralizedsense)<br />

of Aifandonlyifithasaneighborhood U containedin Atowitness<br />

that wis“wellinside” A. And wisintheclosureof Aifandonlyifall<br />

ofitsneighborhoodsintersect A,orinotherwords,ifandonlyif whas<br />

noneighborhood Udisjointfrom Atowitnessthat wis“welloutside” A.<br />

Now,givenaninterpretation �p� ⊆ Xof p,itsinteriorandclosureinterpret<br />

✷pand ✸p,respectively. Moreformally,wehavethefollowing(✷pw)and<br />

(✸pw)(thesubscript“pw”isjusttoconnote“thepossible-worldcase”): 2<br />

w ∈ �✷p� ⇐⇒ ∃U ∈ Nw · U ⊆ �p�, (✷pw)<br />

w ∈ �✸p� ⇐⇒ ∀U ∈ Nw · U ∩ �p� �= ∅. (✸pw)<br />

Letusnotetwothingshere.First,thisneighborhoodsettingsubsumes<br />

Kripkesemantics.ThatisbecauseKripkesemanticsisobtainedfromthis<br />

neighborhoodsemanticsbyfurtherassumingthateachworld whasexactly<br />

oneneighborhood Rw,calledtheworlds“accessiblefrom w”.Then(✷pw)<br />

and(✸pw)boildowntothefollowingconditions,whichareclearlyequivalent<br />

totheusualtruthconditionsfor ✷pand ✸p:<br />

w ∈ �✷p� ⇐⇒ Rw ⊆ �p�,<br />

w ∈ �✸p� ⇐⇒ Rw ∩ �p� �= ∅.<br />

Second,thisneighborhoodsemanticshas ✷and ✸dualtoeachother,i.e.,<br />

✸isjust ¬✷¬,while ✷isjust ¬✸¬,becausethecondition �¬p� =Ï\�p�<br />

(i.e.,thatnegationisinterpretedbycomplementinÏ)implies<br />

w ∈ �✸p� ⇐⇒ ∀U ∈ Nw · U ∩ �p� �= ∅ by(✸pw)<br />

⇐⇒ ¬∃U ∈ Nw · U ⊆Ï\�p� = �¬p�<br />

⇐⇒ w /∈ �✷¬p� by(✷pw)<br />

⇐⇒ w ∈Ï\�✷¬p� = �¬✷¬p�.<br />

2 Thisdefinitionofinteriorandclosureislessgeneralthanthestandardversioninneighborhoodsemantics(see,e.g.,(Chellas,1980)).<br />

Indeed,theformerisequivalenttothe<br />

latterwiththeassumptionthatallfamiliesofneighborhoodsare ⊆-upwardclosed.The<br />

reasonIadoptthelessgeneraldefinitioninthispaperisaphilosophicalonethat,when<br />

importedtotheincompatibilityframework,itrendersavailabletousthecounterfactualrobustnessinterpretationofneighborhoods,whichwewillseeinSection3.<br />

Thereisno<br />

technicalreasonwecouldnotadoptandimportthestandard,fullygeneralformulation<br />

totheincompatibilityframework.


82 KoheiKishida<br />

3 NeighborhoodSemanticsintheIncompatibilityFramework<br />

Toimporttheideafromthepossible-worldframeworktotheincompatibility<br />

framework,letuscomparethetwoframeworksattheground(i.e.nonmodal)level.<br />

First,whilethepossible-worldsemanticsinterpretssentences pwithsubsets<br />

�p�ofÏcontainingworlds w,theincompatibilitysemanticsinterprets<br />

themwithsubsets I(p)of PLcontainingsets Xofsentences. Thiscomparisonsuggeststhatweconsider<br />

PLratherthanÏtobethespace,and<br />

X ∈ PLratherthan w ∈Ïtobepointsinthisspace.<br />

Thenrecallthat,becausethatapoint Xofthisspace PLliesinasemanticinterpretant<br />

I(p)meansincompatibility,entailment p � qcorrespondsto<br />

thereverseinclusion I(p) ⊇ I(q)intheincompatibilityframework,rather<br />

thantheinclusion �p� ⊆ �q�inthepossible-worldcase.Forexample,theT<br />

axiom ✷p ⊢ p ⊢ ✸pcorrespondsto I(✸p) ⊆ I(p) ⊆ I(✷p).Thissuggests<br />

that,intheincompatibilityframework, ✸shouldbeinterpretedbyinterior<br />

ratherthanclosure,and ✷byclosureratherthaninterior.<br />

Therefore,theneighborhoodincompatibilitysemanticsshouldsimplyreplaceÏwith<br />

PL,andswitch ✸and ✷intheinterpretation.Thisideacan<br />

beputasfollows.Aneighborhoodincompatibilityframeisatriple (L, I, N)<br />

consistingof:<br />

•A(sentential)language Lwithmodaloperators;<br />

•Amap Isuchthat (L, I)isanincompatibilityframeon L,treating<br />

thenon-modaloperatorsof Lproperly(inthemannerreviewedin<br />

Section1);<br />

•Aneighborhoodfunction N : PL → PPPLwhoseinteriorandclosure<br />

operationsinterpret ✸and ✷,i.e.,thatsatisfiesthefollowing:<br />

X ∈ I(✸p) ⇐⇒ ∃U ∈ NX · U ⊆ I(p), (✸)<br />

X ∈ I(✷p) ⇐⇒ ∀U ∈ NX · U ∩ I(p) �= ∅. (✷)<br />

Wedefine<br />

�<br />

�asbefore: (L, I, N)has X � Y ifandonlyif (L, I)has<br />

I(p) ⊆ I(X).<br />

p∈Y<br />

Asthisdefinitionispurelyformal,weneedtoexplainwhat,conceptually,<br />

isgoingonhere.First,fixapoint X ∈ PL.Then Xisasetofsentences,<br />

e.g., p /∈ X, q ∈ X,... When U ⊆ PLisaneighborhoodof X,namely<br />

U ∈ NX,itcontainsotherpoints Y, Z,... eachofwhichisasetofsentences,e.g.,<br />

p ∈ Y, q /∈ Y,... Then Ymightbeobtainedbyadding pto X,<br />

dropping qfrom X,andsoon,andsimilarlyfor Z.Hencewecanconsider<br />

Y or Ztobemodifying Xwithcounterfactualhypotheses;forexample,


NeighborhoodIncompatibilitySemantics 83<br />

whenweareattheinformationstate X, Y isjust Xwiththecounterfactualsupposition“If<br />

pwereknowntrue,but qwereunknown,andsoon”.<br />

Theneachneighborhood U ⊆ PLof Xisan“admissible”wayofgrouping<br />

togethersuchcounterfactualhypotheseson X,wherethe“admissibility”is<br />

formallyexpressedby Ulyingin NX.<br />

Now,underthisinterpretationofneighborhoods U ∈ NX, U ⊆ I(p)(i.e.,<br />

∀Y ∈ U[Y ∈ I(p)])means“Whatevercounterfactualhypothesis Y (within<br />

therangeof U)wemaymakeon X,itwouldstillbeincompatiblewith<br />

p”. Inshort, U ∈ NXsuchthat U ⊆ I(p)witnessestheincompatibility<br />

of X with piscounterfactuallyrobust. Accordingly,(✸)statesthat X<br />

isincompatiblewithpossibly-pifandonlyifsomeneighborhood Uof X<br />

witnessesthecounterfactualrobustnessof Xbeingincompatiblewith p.On<br />

theotherhand, U ∈ NXsuchthat U ∩ I(p) = ∅(i.e., ∀Y ∈ U[Y /∈ I(p)])<br />

witnessesthatthecompatibilityof Xwith piscounterfactuallyrobust,and<br />

hence(✷)statesthat Xisincompatiblewithnecessarily-pifandonlyifthe<br />

counterfactualrobustnessof Xbeingcompatiblewith pisneverwitnessed.<br />

Or,rewriting(✷)intermsofentailmentwith X � ¬p ⇐⇒ X ∈ I(p),we<br />

have<br />

X � ¬✷p ⇐⇒ X ∈ I(✷p) ⇐⇒ ∀U ∈ NX · U ∩ I(p) �= ∅<br />

⇐⇒ ∀U ∈ NX∃Y ∈ U · Y ∈ I(p)<br />

⇐⇒ ∀U ∈ NX∃Y ∈ U · Y � ¬p;<br />

thatis, Xentailsnot-necessarily-pifandonlyifeveryneighborhood Uof<br />

Xcontainsacounterfactualhypothesis Ythatentailsnot-p.<br />

4 LogicforNeighborhoodIncompatibilitySemantics<br />

Thissectionreviewswhatrulesandaxiomsarevalidorinvalidinneighborhoodincompatibilitysemantics.Bysayingthatanaxiom(scheme)<br />

X ⊢ Y<br />

isvalidinaneighborhoodincompatibilityframe (L, I, N),wemeanthat<br />

(L, I, N)has X � Y,i.e. �<br />

I(p) ⊆ I(X)(forallinstancesofthescheme). 3<br />

p∈Y<br />

Also,bysayingthatarule(scheme)<br />

X0 ⊢ Y0<br />

X1 ⊢ Y1<br />

isvalidin (L, I, N),wemeanthatif (L, I, N)has X0 � Y0thenithas<br />

X1 � Y1(forallinstancesofthescheme).<br />

3 Weuse �and ⊢differentlyasfollows: X � Yisastatementthat Xentails Y(inagiven<br />

frame);incontrast, X ⊢ Y isasequentratherthanastatement.


84 KoheiKishida<br />

First,the ✸operatorpreservestheorderofentailment �;thatis,the<br />

ruleMbelowisvalidinallneighborhoodincompatibilityframes. 4<br />

p ⊢ q<br />

✸p ⊢ ✸q<br />

Misvalidbecause p � q(i.e., I(q) ⊆ I(p))implies ✸p � ✸q(i.e., I(✸q) ⊆<br />

I(✸p))inanyframe. Toshowthis,assume I(q) ⊆ I(p)and X ∈ I(✸q).<br />

Then,by(✸), Xhassome U ∈ NXsuchthat U ⊆ I(q) ⊆ I(p),which<br />

means,againby(✸),that X ∈ I(✸p).Asimilarargumentshowsthat ✷also<br />

preservesentailment,becauseanyneighborhoodintersecting I(q) ⊆ I(p)<br />

intersects I(p)aswell.<br />

Apartfromthesepreservationrules,neighborhoodsemanticsissogeneral<br />

astoprovidecounterexamplestomanyrulesandaxiomsthatarevalidin<br />

other(mostnotablyKripke’srelational)semantics.Forexample, ✸maynot<br />

preserveincoherence;thatis,evenwhen pisincoherent(viz., I(p) = PL),<br />

✸pmaybecoherent(viz., I(✸p) �= PL),therebyfailingtherule<br />

M<br />

p ⊢<br />

. N✸<br />

✸p ⊢<br />

Thisfailsinapathologicalframeofneighborhoodswheresome X ∈ PL<br />

has NX = ∅(andhence,by(✸), X /∈ I(✸q)forany q).Also,therule<br />

p ⊢<br />

✷p ⊢<br />

canfailinanotherkindofpathologicalframewheresome X ∈ PLhas<br />

∅ ∈ NX(andhence,by(✷), X /∈ I(✷q)forany q).Infact,N✸andN✷are<br />

validintheframeswithoutthesepathologies.<br />

Thefactsdescribedsofarapplytoneighborhoodsemanticsingeneral,<br />

notonlyintheincompatibilityframeworkbutalsointheconventional<br />

possible-worldframework. Onemajordivergenceoftheformerfromthe<br />

latteristhefollowingpointregardingcompleteness.Notethateveryframe<br />

satisfiesoneofthefollowing(1)–(3):<br />

N✷<br />

N∅ �= ∅but ∅ /∈ N∅, (1)<br />

∅ ∈ N∅, (2)<br />

N∅ = ∅. (3)<br />

Themodallogic MNthatisobtainedbyaddingM,N✸,N✷toclassical<br />

logicissoundandcompletewithrespecttotheframessatisfying(1),inthe<br />

4 Thisrulecanbeavoidedifweadoptthestandard,moregeneraldefinitionofinterior;<br />

seeFootnote2.TherulethatisvalidinsteadofMinthemoregeneralformulationisto<br />

infer ✸p ⊢ ✸qfromboth p ⊢ qand q ⊢ p.


NeighborhoodIncompatibilitySemantics 85<br />

sensethatasequentoraninferencefromsequentstoanotherisvalidinall<br />

thoseframesif(soundness)andonlyif(completeness)itisatheoremora<br />

derivableruleof MN. Also,let MI✸and MI✷bethelogicsobtainedby<br />

addingMaswellasthefollowingaxiomsI✸andI✷,respectively,toclassical<br />

logic:<br />

✸p ⊢, I✸<br />

✷p ⊢ . I✷<br />

Then MI✸and MI✷aresoundandcompletewithrespecttotheframes<br />

satisfying(2)and(3),respectively. Moreover,weneedtoformulatethe<br />

completenessofanaxiomorruleinamannerthatdependsonthesethree<br />

classesofframes,asfollows. WesayanaxiomorruleAiscompletewith<br />

respecttoasemanticconditionCifthelogicsobtainedbyaddingAto MN,<br />

MI✸, MI✷,respectively,arecompletewithrespecttotheclassesofframes<br />

satisfyingCaswellas(1),(2),(3),respectively.Ontheotherhand,wecan<br />

definethesoundnessofAwithrespecttoCintheusualmanner.<br />

Themostnotabledivergenceoftheneighborhoodincompatibilitysemanticsfromtheconventionalneighborhoodsemanticsisthatthedualityof<br />

✸<br />

and ✷mayfailintheformer,eventhoughthenon-modalbaseofthelogic<br />

isclassical. Recallthattheproof(attheendofSection2)oftheduality<br />

inthepossible-worldframeworkwasbasedontheinterpretationofnegation,<br />

¬,intermsofcomplementinthespaceÏ,sothat �p� ∩ �¬p� = ∅<br />

and �p� ∪ �¬p� =Ï. Theincompatibilityframeworkinterprets ¬differently,whichiswhythedualityfailsintheframework.Eventhoughafull<br />

constructionofcounterexamplesrequirestoomanydetailstocoverhere,<br />

aroughbutheuristicdescriptioncanbegivenasfollows. Intheincompatibilityframework,<br />

I(p)and I(¬p)normallyintersectwitheachother;<br />

indeed, I(p) ∩ I(¬p)equalstheset Incofincoherentsetsofsentences.So,a<br />

coherentpoint Xmayhaveanonempty U ⊆ Inc = I(p) ∩ I(¬p)asitsonly<br />

neighborhood.Then U ∩ I(p) �= ∅andhence X ∈ I(✷p)by(✷)(because<br />

NX = {U}),butatthesametime U ⊆ I(¬p);thismeans X ∈ I(✸¬p)<br />

by(✸),whichthenimplies X /∈ I(¬✸¬p)since Xiscoherent,thatis,<br />

X /∈ Inc = I(✸¬p) ∩ I(¬✸¬p).Sothis Xwitnesses I(✷p) �⊆ I(¬✸¬p),i.e.,<br />

¬✸¬p � ✷p.Theupshotisthatthisentailmentfailswhenneighborhoods<br />

aretoostrong(i.e.,whenpointsinthemlieinboth I(p)and I(¬p)),which<br />

cannothappenintheclassicalpossible-worldframework(i.e.,nopointever<br />

liesinboth �p�and �¬p�).Quiteexpectably,theotherdirection ✷p � ¬✸¬p<br />

ofthedualityfailswhenneighborhoodsaretooweak(i.e.,whenpointsin<br />

themlieinneither I(p)nor I(¬p)),whichcannothappenintheclassical<br />

possible-worldframework(i.e.,everypointhastolieineither �p�or �¬p�).<br />

Whiletherearecertainsemanticconditionstotheeffectthatneighborhoodsarenottoostrong,ornottooweak,withrespecttowhicheitherof


86 KoheiKishida<br />

¬✸¬p ⊢ ✷pand ✷p ⊢ ¬✸¬pissoundandcomplete,themoreimportant<br />

pointisthatneighborhoodsemanticsismoreexpressiveintheincompatibilityframeworkthanintheclassicalpossible-worldframework.Thisisnot<br />

merelyinthesensethatthedualityaxiomsareinvalid,butthesemantics<br />

separatesthe ✸and ✷versionsofmanyaxioms;forexample,<br />

p ⊢ ✸p, T✸<br />

✷p ⊢ p. T✷<br />

Eventhoughtheseaxiomsaretreatedjustasequivalentinpossible-world<br />

semantics,theycorrespondtotwodifferentsemanticconditionsonneighborhoodincompatibilityframes.Tolayouttheseconditions,weneedtointroduceapreorder(i.e.areflexiveandtransitiverelation)<br />

�on PLdefinedasfollows:<br />

X � Y ⇐⇒ I(X) ⊆ I(Y ).<br />

So, X � Y roughlymeans Xisweakerthan Y,orthat Y conjunctively<br />

entails Xconjunctively. Then �generalizes ⊆inthesensethat X ⊆ Y<br />

entails X � Y, 5 andmoreovereverysemanticinterpretant I(p)isclosed<br />

upwardintermsof �.Then,givenanyset U ⊆ PLofpoints,wewrite ↑U<br />

and ↓Uforthe �-upwardand �-downwardclosuresof U ⊆ PL,respectively,<br />

i.e.,<br />

↑U = {Y ∈ PL | X � Yforsome X ∈ U},<br />

↓U = {X ∈ PL | X � Yforsome Y ∈ U}.<br />

Inthepossible-worldframework,theTaxiom p ⊢ ✸p(or ✷p ⊢ p)correspondstothesemanticconditionthat<br />

w ∈ Uforall U ∈ Nw.Incontrast,<br />

intheincompatibilityframework,usingthenotionsdefinedabovewecan<br />

modifythisconditiontoobtainthefollowingtwoversions:<br />

X ∈ ↑U forall U ∈ NX, (4)<br />

X ∈ ↓U forall U ∈ NX. (5)<br />

Then(4)and(5)havetheaxiomsT✸andT✷,respectively,soundand<br />

complete.<br />

Hereweonlyshowthesoundness. ThatT✸issoundwithrespectto<br />

(4)meansthat(4)entails I(✸p) ⊆ I(p). Toshowthis,letusassume<br />

X ∈ I(✸p). Itmeanssome U ∈ NXisincludedin I(p). Then,because<br />

I(p)is �-upwardclosed, ↑Uisstillincludedin I(p). Hence(4)implies<br />

X ∈ ↑U ⊆ I(p),therebyestablishingthesoundness. Wecansimilarly<br />

5 Therearemoresensesinwhichwecansay �generalizes ⊆.See(Kishida,n.d.-b).


NeighborhoodIncompatibilitySemantics 87<br />

showthesoundnessofT✷withrespectto(5),i.e.,(5)entailing I(p) ⊆<br />

I(✷p),byassuming X /∈ I(✷p).Thismeanssome U ∈ NXisdisjointfrom<br />

I(p).Then,againbecause I(p)is �-upwardclosed, ↓Uisstilldisjointfrom<br />

I(p).Hence(5)implies X ∈ ↓Uandtherefore X /∈ I(p),establishingthe<br />

soundness.<br />

Upwardanddownwardclosuresdifferentiatethe ✸and ✷versionsinthe<br />

caseoftheTaxioms,andtheydosointhefollowingcaseaswell.Consider<br />

thecondition:<br />

If U ∈ NX,thereis V ∈ NXsuchthat<br />

every Y ∈ Vhassome UY ∈ NYwith UY ⊆ U. (6)<br />

Thismeansthateveryneighborhood Uof Xhasanother Vof Xsuchthat<br />

Uis(asupersetof)aneighborhoodofeverypointin V;inshort,(6)says<br />

thateachneighborhoodisaneighborhoodofaneighborhood.Nowreplace<br />

thelast Uin(6)with ↑U,toobtain:<br />

If U ∈ NX,thereis V ∈ NXsuchthat<br />

every Y ∈ Vhassome UY ∈ NYwith UY ⊆ ↑U. (7)<br />

Thenthefollowing ✸versionoftheS4axiomissoundandcompletewith<br />

respectto(7):<br />

✸✸p ⊢ ✸p. S4✸<br />

Toshowthesoundness,i.e.,that(7)entails I(✸p) ⊆ I(✸✸p),assume<br />

X ∈ I(✸p). Thismeans I(p)includessome U ∈ NXand ↑U. Then(7)<br />

yields V ∈ NXsuchthatevery Y ∈ V hassome UY ⊆ ↑U ⊆ I(p),i.e. Y ∈<br />

I(✸p),whichmeans V ⊆ I(✸p).Therefore Vwitnesses X ∈ I(✸✸p).<br />

Replacing ↑Uwith ↓Uin(7),wecanshowbyessentiallythesameidea<br />

thatS4✷issound:<br />

✷p ⊢ ✷✷p. S4✷<br />

Hereis,however,someasymmetrybetween ✸and ✷.EventhoughS4✷is<br />

soundwithrespecttothecondition(7)with ↓Uinplaceof ↑U,itisnot<br />

complete.Toachievethecompleteness,weneedtoweakentheconditiona<br />

littlebit,toobtain:<br />

If N∅ �= ∅,thenthefollowingholdsforevery X ∈ PL:<br />

if U ∈ NX,thereis V ∈ NXsuchthat<br />

every Y ∈ Vhassome UY ∈ NYwith UY ⊆ ↓U.


88 KoheiKishida<br />

Inthefollowingcasetheasymmetrybetween ✸and ✷isevenbigger.<br />

Considerthefollowingaxioms(whicharedualtoeachotherintheconventionalframework):<br />

C✸issoundandcompletewithrespectto:<br />

✸(p ∨ q) ⊢ ✸p ∨ ✸q, C✸<br />

✷p ∧ ✷q ⊢ ✷(p ∧ q). C✷<br />

U0,U1 ∈ NX =⇒thereis U2 ∈ NXsuchthat U2 ⊆ ↑U0and U2 ⊆ ↑U1.<br />

(8)<br />

Forthesoundness(i.e.,(8)entailing I(✸p∨✸q) ⊆ I(✸(p∨q))),assume X ∈<br />

I(✸p ∨✸q) = I(✸p) ∩ I(✸q).Thismeans Xhassomeneighborhoods U0 ⊆<br />

I(p)and U1 ⊆ I(q),which,asalways,entails ↑U0 ⊆ I(p)and ↑U1 ⊆ I(q).<br />

Now,(8)yields U2 ∈ NXsuchthat U2 ⊆ ↑U0 ⊆ I(p)and U2 ⊆ ↑U1 ⊆ I(q),<br />

i.e. U2 ⊆ I(p)∩I(q) = I(p∨q).Hence U2witnesses X ∈ I(✸(p∨q)).Inthis<br />

way,C✸issoundwithrespectto(8),andinfactcomplete. Nevertheless,<br />

theannouncedasymmetrybetween ✸and ✷isthatitisanopenproblem<br />

evenwhatconditionhasC✷soundandcomplete,becauseitdoesnotseem<br />

toworktoreplace ↑Uwith ↓U. 6 Itisalsointerestingtoseehow ✸and ✷<br />

interactwitheachother.Theaxiom<br />

✷p ⊢ ✸p D<br />

issoundandcompletewithrespecttotheconditionthat U ∩ V �= ∅forall<br />

U,V ∈ NX.ItiseasytoshowDtobesoundbecauseif X ∈ I(✸p),i.e.,if<br />

a U ∈ NXhas U ⊆ I(p),thentheconditionsaysevery V ∈ NXintersects<br />

U,therebyintersecting I(p),i.e. X ∈ I(✷p).<br />

Itisanopenproblemwithrespecttowhatconditionthefollowingaxiom<br />

Biscomplete:<br />

p ⊢ ✷✸p; B<br />

butthereisaconditionwithrespecttowhichBissound:<br />

Givenacollection {Xi ∈ PL | i ∈ I }ofanysizeand Y ∈ PL,<br />

ifeach Xihassome Ui ∈ NXiwith Y /∈ Ui,<br />

thenaV ∈ NYhas Xi /∈ Vforall Xi. (9)<br />

Toshowthesoundness(i.e.,(9)entailing I(✷✸p) ⊆ I(p)),suppose Y /∈<br />

I(p). Write I(✸p) = {Xi ∈ PL | i ∈ I};theneach Xiliesin I(✸p),i.e.,<br />

Xihassome Ui ∈ NXi suchthat Ui ⊆ I(p)andhence Y /∈ Ui. Then(9)<br />

yields V ∈ NY suchthat Xi /∈ V forall Xi,i.e. V ∩ I(✸p) = ∅;therefore<br />

Y /∈ I(✷✸p).<br />

6 Thedifficultyarises partlyfromthelackofunderstandingofwhat I(✷p ∧ ✷q) =<br />

I({✷p, ✷q})lookslike,incontrastto I(✸p ∨ ✸q)understoodsimplyas I(✸p) ∩ I(✸q).


NeighborhoodIncompatibilitySemantics 89<br />

5 Conclusion<br />

Wehaveshownthattheideaofneighborhoodsemanticstointerpretmodal<br />

operatorswithinteriorandclosureoperationscanbestraightforwardlyimported—quiteindependentlyofthenotionsoftruthandpossibleworlds—totheframeworkofincompatibilitysemantics.Thephilosophicaladvantageofputtingthenotionofneighborhoodinthisframeworkisthattheconnectionbetweenneighborhoodsandmodalitycanbedirectlyandnaturallyinterpretedintermsoftheideaofcounterfactualrobustness.Wehave<br />

alsoshownthetechnicalmeritofthesemanticsthatwecanseparatethe<br />

behaviorsof ✸and ✷whilekeepingthenon-modalbaseofthelogicclassical,which,combinedwiththecounterfactual-robustnessinterpretation,will<br />

enableustoapplymodallogictoanevenwiderrangeofcases.<br />

KoheiKishida<br />

DepartmentofPhilosophy,UniversityofPittsburgh<br />

1001CathedralorLearning,Pittsburgh,PA15260U.S.A.<br />

kok6@pitt.edu<br />

References<br />

Brandom,R. (2008). Incompatibility,modalsemantics,andintrinsiclogic. In<br />

Betweensayinganddoing:Towardsananalyticpragmatism(chap.5). Oxford:<br />

OxfordUniversityPress.<br />

Chellas,B.(1980).Modallogic:Anintroduction.Cambridge-NewYork:CambridgeUniversityPress.<br />

Kishida,K. (n.d.-a). Neighborhoodincompatibilitysemanticsandcompleteness.<br />

(Draft.)<br />

Kishida,K.(n.d.-b).Possible–worldrepresentationofincompatibility.(Draft.)


What do Gödel Theorems Tell us about<br />

Hilbert’s Solvability Thesis?<br />

Vojtěch Kolman ∗<br />

Whendealingwiththefoundationalquestionsofelementaryarithmetic,<br />

wefindourselvesstandingintheshadowofGödel,justasourpredecessors<br />

stoodintheshadowofKant,totheextentthatwetendtoseeGödel’s<br />

famousincompletenesstheoremsasanewCritiqueofPureReason.Inits<br />

mostexuberantform(commonparticularlyamongtheso-calledworking<br />

mathematicians)thisamountstoclaimingthat<br />

humanreasonhasencountereditslimitsbyprovingthatthereare<br />

truthswhicharehumanlyunprovable(“inaccessible”)andthatitis<br />

impossibleforourmindtoproveitsownconsistency. 1<br />

Thisattitudeisnotonlyatvariancewiththe(Kantian)doubtsaboutthe<br />

possibilityofprovingtheunprovabilityinanabsolutesense,but,more<br />

specificallyandfamously,withtheso-calledHilbertprogramofsolvingeverymathematicalproblembyaxiomaticmeans.<br />

InhisParisianaddress, 2<br />

Hilbertnotonlyphrasedtheconjecturethatallquestionswhichhuman<br />

mindasksmustbeanswerable(theso-calledaxiomofsolvability) 3 but<br />

supplementedit,asakindofchallenge, withalistoftenandlaterof<br />

twenty-threeproblemsofprimeinterest,includingtheSecondProblemof<br />

theconsistency(andcompleteness)ofarithmeticalaxioms.<br />

InHilbert’slaterwritings,particularlyinhisKönigsbergaddress, 4 the<br />

solvabilityargumenttakesamoresubtleform.Introducingthefinitemode<br />

∗ WorkonthispaperhasbeensupportedinpartbygrantNo.401/06/0387oftheGrant<br />

AgencyoftheCzechRepublicandinpartbytheresearchprojectMSM0021620839of<br />

theMinistryofEducationoftheCzechRepublic.<br />

1 (Gödel,1995,p.310)himselfphraseditlikethis: “thereexistabsolutelyunsolvable<br />

diophantineproblems[...],wheretheepithet‘absolutely’meansthattheywouldbe<br />

undecidable,notjustwithinsomeparticularaxiomaticsystem,butbyanymathematical<br />

proofthehumanmindcanconceive.”<br />

2 See(Hilbert,1900).<br />

3 See(Hilbert,1900,p.297).<br />

4 See(Hilbert,1930).


92 VojtěchKolman<br />

ofthought(finiteEinstellung) 5 asanewkindofKantianintuition,Hilbert<br />

arguesthattheharmonybetweennature(experience)andthought(theory)<br />

mustlieexactlyinthetranscendentalfacttheyarebothfinite. 6 Asa<br />

consequence,theseeminginfinityofhumanknowledge(particularlyinthe<br />

realmofmathematics)musthavefiniterootswhicharetobeidentified<br />

withafinite(orfinitelydescribable)systemofrulesandaxioms,andfinite<br />

deductionsfromthem. 7 Hence,“wemustknow,weshallknow.” 8 Obviously,<br />

thisisatranscendentaldeductionofitsownkind,namelyofinferentialismor<br />

broaderaxiomatismfromfinitism,startingwiththewords:inthebeginning<br />

wasasign. 9<br />

Gödel(1931),soweareusuallytold,putanendtoHilbert’soptimism<br />

byprovingthattheSecondProblemisessentiallyunsolvable.Thisverdict<br />

issometimessupportedbytheseeminglyanalogouscaseofHilbert’sFirst<br />

Problem,theContinuumHypothesis,which,partiallyalsoduetoGödel,<br />

wasprovedtobeundecidableonthebasisofcurrentlyacceptedaxioms.In<br />

thispaperIwouldliketopresentGödel’stheoremsnotasadirectrefutation<br />

ofHilbert’saxiombutonlyasanimpulsetophraseitwithmorecaution,<br />

insuchawaythattheContinuumHypothesisisnolongerregardedasa<br />

realproblem. Iwilldrawontworatherdifferentsources,both,however,<br />

connectedtoHilbert’sphilosophy,namely<br />

•thelatemetamathematicalviewsofZermeloand<br />

•Lorenzen’spost-Hilbertianprogramofoperativemathematics.<br />

Thiswillleadmetoacloseranalysisofthedistinctionbetweenproofand<br />

truthwhichdoesnotendorseoneofthemattheexpenseoftheother,as<br />

Lorenzen,theconstructivist,andZermelo,thePlatonist,stilltendtodo.<br />

1<br />

First,letusdiscussthepossibilityofprovingtheunsolvabilityofsomething.<br />

Thereisageneralpattern:ifsomeonecomesalongwithapositivesolution<br />

toagivenproblem,onecanchecktoseethatitdoestherequiredwork.<br />

Butifitistobeshownthattheproblemisunsolvable,onehastogivea<br />

precisedelimitationofthemethodsthatcanbeemployed.Thisbringsusto<br />

thedifferencebetweenmethodinthebroader(general)andinthenarrower<br />

(limited)sense.<br />

5 See(Hilbert,1930,p.385)andalso(Hilbert,1926,p.161).<br />

6 See(Hilbert,1930,pp.380–381).<br />

7 See(Hilbert,1930,p.379)andalso(Hilbert,1918).<br />

8 (Hilbert,1930,p.387).<br />

9 See(Hilbert,1922,p.163).


WhatdoGödelTheoremsTellusaboutHilbert’sSolvabilityThesis? 93<br />

Toillustratethepointletustakesomefamousgeometricalproblemslike<br />

thetrisectionofanangleorthequadratureofthecircle.Duetothemethods<br />

ofmodernalgebrawepositivelyknowthattheseproblemsareunsolvable<br />

bystraightedgeandcompass. However,wealsoknowthattheancient<br />

mathematicians(Hippias,Archimedes)alreadysolvedthembyextended<br />

—so-calledmechanical—means(quadratix,spiral) 10 wheretheepithet<br />

“mechanical”meansmainlythattheyweredevisedadhoc. Similarly,ifI<br />

giveyou—meaningsomebodysufficientlyeducatedinpredicatelogic—a<br />

formula,Iamquitesureyouwillbeabletodecide,inafinitenumberof<br />

steps,whetheritisatautologyornot.Whatyoumightnotbeabletodo,<br />

however,istosolvetheproblemwithpre-chosenschematicmethodssuch<br />

aswithoneparticularTuringmachine.<br />

Now,asmaybeexpected,asimilarobservationappliestoGödel’stheorems,onlythistimeitistheprovabilityitselfthelimitsofwhichbegthe<br />

question.Gödelshowedthatforanyaxiomaticsystemofarithmeticthere<br />

willalwaysbeanindividualsentencethatisundecidablebyit. Thegist<br />

ofhisargumentliesinthefactthatthisunprovablesentenceofarithmetic<br />

(informallysaying“Iamunprovable”)isunprovablebecauseitistrue(it<br />

isunprovable),itstruthbeingprovenasapartoftheargument. So,the<br />

wholeargumentworksonlybecauseitemploystwodifferentconceptsof<br />

proof,thefirstbeingthatofPrincipiaMathematica(orPeanoarithmetic)<br />

andthesecondbeingthebroaderoneinwhichtheargumentisclinched.<br />

Zermelo,inhisunjustlyinfamouscorrespondencewithGödel,wasprobablythefirstpersontomakethisobservation.Settinghimselfthenatural<br />

question,“Whatdoesoneunderstandbyaproof?”,hisanswerwentlike<br />

this:<br />

Ingeneral,aproofisunderstoodasasystemofpropositionsthat,<br />

whenacceptingthepremises,yieldsthevalidityoftheassertionas<br />

beingreasonable. Andthereremainsonlythequestionofwhatmay<br />

be“reasonable”. Inanycase—asyouareshowingyourself—not<br />

onlythepropositionsofsomefinitaryschemethat,alsoinyourcase,<br />

mayalwaysbeextended. So,inthisrespect,weareofthesame<br />

opinion,however,Iaprioriacceptamoregeneralschemethatdoes<br />

notneedtobeextended.Andinthissystem,reallyallpropositions<br />

aredecidable. 11<br />

Whatneedstobeexplainednowisthenatureofthedifferencebetween<br />

proofinthenarrowerandbroadersense,orbetweentheproof andtruth,<br />

andthesenseinwhichthesecondoneis“decidable”,orbetter:complete<br />

andunextendable,asZermeloclaims.<br />

10 See,e.g.,(Heath,1931).<br />

11 See(Gödel,2003,p.431).


94 VojtěchKolman<br />

2<br />

Theanalogousdifferencesbetweenthegeneralandnarrowerconstruability<br />

ordecidabilityislessproblematicsincetheadhocconstructiveordecision<br />

methods(likequadratixorspiral)arestillboundtosomehumanlyfeasible<br />

means,andsoquitenaturallycountedasconstructionsandalgorithms.The<br />

traditionalproblemofarithmeticisitsveryrelationshiptotheempirical<br />

world,as(alreadybeforeKant)expressedintheclaimitisascienceof<br />

analyticalnature. Hence,thewholeissueofthedifferencebetweenthe<br />

truthandproofcanbeboileddowntoasinglequestion:<br />

whatisarithmeticaltruthoutsideofaspecificaxiomaticsystem?<br />

Itisexactlythelackofanyexplicitanswertothisquestionthatleadsto<br />

thePlatonistaccountofarithmeticaltruth. Theusualmodel-theoretical<br />

expositionoperatingwithanunexplainedconceptofstandardmodel(“2 +<br />

2 = 4”istrueifandonlyif 2 + 2 = 4)confirmsthisimage,particularly<br />

whenitstartstoinvokeour“intuitions”.<br />

However, tounderstandsentenceslike“2 + 2 = 4”and“23 + 4 <<br />

(6 × 3) + 2”youneednomoremathematicsthanthatprovidedbyagood<br />

secondaryeducation.Thisistosaythattheyarenottrueorfalse,atleast<br />

notinthefirstplace,becausetheyarededucibleinPeanoarithmetic,or<br />

happentoinexplicablyholdinthestandardmodel,butbecausetheyare<br />

transformableintothesimplerformsof“4 = 4”and“27 < 20”whereonly<br />

knowledgeofthesequence 1,2,3,4,... andtheabilitytocomparesymbolsisneeded.Thisisthebasisoftheoperativistaccountofarithmetical<br />

truthasdevelopedbyLorenzeninhisEinführungindieoperativeLogikund<br />

Arithmetik(Lorenzen,1955),inoppositiontotheusualstandardsofFrege<br />

thatconsidersuchjustificationsprescientific.AccordingtoLorenzen, 12 the<br />

ultimatefoundationofarithmetic(includinghigheranalysis)liesexactlyin<br />

theseprescientificpracticesofcountingandoperatingwithsymbols.They<br />

canbemadeexplicitinsynthetic(recursive)definitionslike<br />

⇒ |, ⇒ x + | = x|<br />

x ⇒ x|, x + y = z ⇒ x + y| = z|<br />

⇒ | × x = x ⇒ | < x|<br />

x × y = p, p + y = q ⇒ x| × y = q x < y ⇒ x < y|<br />

introducing(inunaryform)thenumberseries,theoperations +, ×andthe<br />

relation


WhatdoGödelTheoremsTellusaboutHilbert’sSolvabilityThesis? 95<br />

theconsequencesofthesedefinitions,prospectivelywithinthebroaderframe<br />

ofgame-orproof-theoreticalsemantics(Lorenzen’sdialogicalgames). 13<br />

AsforGödel’sresults,Lorenzen 14 claimsthatinsteadofbeingabout<br />

arithmetic,ascompletelygivenbyitsoperativedefinition,theymerelytell<br />

ussomethingaboutPeano’sformalisminitsparticularshapeofafirst-order<br />

schemewithinthelanguagecontaining 0, s, +and ×. So,comingfrom<br />

theotherside,LorenzenarrivedatthesamebasicdifferenceasZermelo.<br />

ItisalsoinaccordbothwithLorenzen’slaterviews,asdevelopedinhis<br />

Metamathematik(1962),andwithZermelo’slateprojectofinfinitistlogic, 15<br />

torephrasethisdifferenceininferentialisttermsasthedistinctionbetween<br />

twodifferentkindsofconsequence:stronglyeffectiveorfull-formal ⊢and<br />

themoreliberalorsemi-formal |=. 16 Now,simplifyingheavily:<br />

Full-formalarithmetic,likethearithmeticofPeano,isarithmeticinthe<br />

narrowersense,anddealswithschematicallyormechanicallygivenand<br />

controllableaxiomsandrules. Semi-formalarithmeticorthearithmetic<br />

properemploys—inaccordwiththeinfinitenatureofthenumbersequence<br />

1,2,3,...—ruleswithinfinitelymanypremises,particularlythe (ω)-rule<br />

A(1),A(2),A(3),etc. ⇒ (∀x)(Ax). (ω)<br />

Asanarithmeticalruleitistransparentandsoundenough(or“reasonable”,<br />

asZermelowouldsay),aslongasoneinterpretsthe“etc.” correctly. In<br />

fact,Tarski’sideaofsemantics 17 employsthiskindofrulessystematically,<br />

withthe (ω)-ruleasaspecialcaseofthemoregeneral<br />

A(N)forallsubstituents N ⇒ (∀x)A(x). (∀)<br />

Thisruleisthennothingelsethanthewell-knownpartoftheso-called<br />

semanticdefinitionoftruth.Hence,thesignificanceofsemi-formalismisto<br />

makeusthinkofsemanticdefinitionsasspecial(moregenerouslyconceived)<br />

systemsofrules(proofsystems)which—startingwithsomeelementary<br />

sentences—evaluatethecomplexonesbyexactlyoneoftwotruthvalues.<br />

Themostimportantpointtonoticeisthatthesemi-formalrulesarecalled<br />

semanticnotbecausetheyareinfinitebutbecausethey,unlikePeano’s<br />

formalism,workwithauniquelydeterminedrangeofquantification.<br />

Asaconsequence,arithmeticaltruthneednotbeguaranteedbyGod<br />

orbyintuition,but,as(Zermelo,1932,p.87)putit,simplybythefact<br />

thatthebroaderconceptof“mathematicalproofisnothingotherthana<br />

systemofpropositionswhichiswell-foundedbyquantification.”Zermelo’s<br />

13 See(Lorenzen&Lorenz,1978).<br />

14 See(Lorenzen,1974,p.21–22).<br />

15 See(Zermelo,1932).<br />

16 Bothdistinctionsaredueto(Schütte,1960).<br />

17 Seeespecially(Tarski,1936).


96 VojtěchKolman<br />

claimthatallthesentencesaredecidedbyhis“moregeneralscheme”,i.e.,<br />

completelyandcorrectlyevaluatedbyarithmeticalsemi-formalism,canbe<br />

“proved”byaneasymeta-inductionlikethis:<br />

1.Elementaryarithmeticalsentences(M = P, M < N)areevaluated<br />

unambiguouslyastrueorfalseonlyonthebasisofcalculationswith<br />

numerals.<br />

2.Tarski’sdefinitionprovidesfortheevaluationofmorecomplexsentences,particularlybecause:eitherforeveryterm<br />

Nfrom 1,2,3,... ,<br />

thesentence A(N)istrueandhence (∀x)A(x)istrue,orthereis N<br />

from 1,2,3,...suchthat A(N)isfalse,and (∀x)A(x)isfalse,tertium<br />

nondatur.<br />

Itisaknownfactthattheintuitionistsandsomeconstructivists(including<br />

Lorenzen, 18 butnot,e.g.,Weyl 19 )questionthecompletenessofthisevaluation,arguingthattheexistenceofconcretestrategiesforprovingorrefuting<br />

every A(N)doesn’tentailtheexistenceofageneralstrategyfor A(x).To<br />

giveafamiliarexample:thereisnoproblemindemonstratingwhether,for<br />

anygivenevennumber M,itisthesumoftwoprimes.However,thetruth<br />

valueofthegeneraljudgmentthateveryevennumberisthesumoftwo<br />

primes(GoldbachConjecture)isstillunknown, 250yearsaftertheproblem<br />

wasfirstposed.Hence,itispossiblethatwehaveproofsforallthesentences<br />

A(N)withoutknowingit,i.e.,withouthavingthegeneralstrategyofhow<br />

toproveapropositionconcerningthemall.<br />

Consequently,adecisionmustbemadewhethertheinfinitevehiclesof<br />

truthandjudgmentsuchas(∀)or(ω)shouldbereferredtoasrules<br />

3<br />

1.onlyinthecasewhenwepositivelyknowthatalltheirpremisesare<br />

true,i.e.,whenwehaveatourdisposalsomegeneralstrategyfor<br />

provingallofthematonce,or<br />

2.moreliberally,ifweknowsomehowthatalltheirpremisesarepositivelytrueorfalse.Thegeneraldistinctionbetweentheconstructive<br />

andclassicalmethodsinarithmeticisbasedonthis.<br />

Now,ifoneleaves,like,e.g.,LorenzenandBishop,theconceptofeffective<br />

procedureorprooftoalargeextentopenanddoesnottieit,like,e.g.,<br />

GoodsteinandMarkov,totheconceptoftheTuringmachine, 20 thereisstill<br />

18 See,e.g.,(Lorenzen,1968,p.83).<br />

19 See(Weyl,1921,p.156).<br />

20 Forfurtherdiscussionofthesedifferencessee,e.g.,(Bridges&Richman,1987).


WhatdoGödelTheoremsTellusaboutHilbert’sSolvabilityThesis? 97<br />

roomforaneffective,yetliberalenoughsemantics(semi-formalsystem)<br />

andastronglyeffectiveor‘mechanical’syntaxoraxiomatics(full-formal<br />

system). Hence,theconstructivistreadingdoesnotnecessarilywipeout<br />

thedifferencesbetweentheproofandtruth,as,e.g.,Brouwer’smentalism<br />

orWittgensteins’sverificationismseemto. Asaresult,onecanofficially<br />

differentiatenotonlybetweenfull-formal ⊢andsemi-formal |=consequence,<br />

butalsobetweensemi-formalconsequenceinastricter(constructive)sense<br />

andinthemoreliberal(classical)sense. Allthesedifferencesstemfrom<br />

(Gödel,1931)forthefollowingreason:<br />

Gödel’stheoremaffectsonlythefull-formalsystems,becausetheirschematicnaturemakesitpossibletodeviseageneralmeta-strategyforconstructingtruearithmeticalsentencesnotprovableinthem.Theunprovable<br />

sentenceofGödelisoftheso-calledGoldbachtype,i.e.,itisoftheform<br />

(∀x)A(x)where A(x)isadecidablepropertyofnumbers.Now,Gödel’sargumentshowsthatthisdecisionisdonealreadybyPeanoaxiomsinthesense<br />

thatalltheinstances A(N)arededucibleand,hence,setastrue.So,with<br />

Gödel’sproofwehaveageneralstrategyforprovingallthepremises A(N)<br />

atonce,whichmakesthecriticalunprovablesentence (∀x)A(x)constructivelytrue,i.e.provablebymeansofthe<br />

(ω)-ruleinterpretedconstructively.<br />

Lorenzen(1974,p.222)putitlikethis: 21<br />

ω-incompleteness[...]demonstratesthatnotallconstructivelytrue<br />

propositionsarelogicallydeduciblefromtheaxioms. Thisshould<br />

comeasnosurprise. Auniversalproposition (∀x)A(x)isconstructivelytruewhen<br />

A(N)forall Nistrue. Butinorderlogicallyto<br />

deducetheuniversalproposition (∀x)A(x),wemustfirstdeduce A(x)<br />

withafreevariable x.Soweshouldhaveexpected ω-incompleteness.<br />

ButPeanoarithmeticis ω-completeifwerestrictourselvestoaddition.<br />

ThepointofGödel’sproofwastodemonstratethatPeanoarithmetic<br />

withonlyadditionandmultiplication(withoutthehigherformsof<br />

inductivedefinition)alreadyshowsthe ω-incompletenessthatwasto<br />

beexpectedingeneral.<br />

Itisofrealsignificanceherethatitwasnoneotherthan(Hilbert,1931)<br />

who—probablystillunawareofGödel’sresult 22 —employedthe (ω)-rule<br />

asameansofimprovinghisoldprojectoffoundingarithmeticonaxiomatic<br />

grounds. So,ourclaimthatGödel’stheoremsdidnotdestroybutrefine<br />

Hilbert’soptimisminthesuggestedsemi-formalwayissoundalsofroma<br />

historicalperspective.Andusingtheconceptofsemi-formalismagain,we<br />

canextendthisoptimismyetfurtherbyclaimingthatfull-formalsystems<br />

21 TranslationbyK.R.Pavlovicin(Lorenzen,1987,p.240–241).<br />

22 SeeBernays’remarksin(Hilbert,1935,p.215)butalsoFeferman’scommentaryin<br />

(Gödel,1986,pp.209–210).


98 VojtěchKolman<br />

suchasPeanoandRobinsonarithmeticareconsistentsimplybecausetheir<br />

axiomsareprovableinthearithmeticalsemi-formalismand,moreover,even<br />

initsconstructivevariant.This,infact,istheusualmodel-theoreticargument:<br />

ifatheoryisinconsistent,thenitdoesnothaveamodel,<br />

inarelativesetting:<br />

ifPeanoarithmeticisinconsistent,thensoisthearithmeticalsemiformalism.<br />

Inthefirstcasetheconsequentisprecluded“byfiat”. Inthesecondcase<br />

onedoesnotneedtousesuchtricks,becauseitwasactuallyprovedthatthe<br />

rulesofsemi-formalismdonotevaluatearithmeticalsentencesincorrectly.<br />

4<br />

Now,shouldweperhapsfollowZermelofurtheranddiscardthenarrower<br />

conceptofprooftotallybysayingthateverythingtrueisprovable?While<br />

thedangerofthefirstextremeliesinthefactthatthenarrower,limited<br />

methodscanandeventuallywillfailbecauseoftheirlimitedness,theshortcomingofZermelo’salternativeisthatitissafetothepointofbecoming<br />

totallyidle. Theproblemsofsettheoryareaparticularlygoodexample<br />

ofsuchasituation. Letmeillustrateitverybrieflywiththehelpofthe<br />

conceptofcontinuum. 23<br />

Continuumhashadanintricatehistoricaldevelopment,fromthePythagoreandefinitionofproportionbymeansofareciprocalsubtraction,throughtheEuclidiantheoryofpointsconstructiblebymeansofarulerandcompass,totheCartesianideaofnumbersasrootsofpolynomials.Bygrasping<br />

realnumbersasarbitrary(Cauchy)sequences,ratherthanassequencesthat<br />

areinsomesenselaw-like,Cantorbelievedhimselftohavewonthewhole<br />

gamebysimple“fiat”. Butthiswasnomoresubstantiatedthanitwould<br />

havebeenfortheGreekstodefinerealnumbersaspointsconstructibleby<br />

whatevermeans,orforusnowtosaythateverythingtrueisprovable.Obviously,thiswoulddisposeofproblemslikethequadratureofthecircle,the<br />

axiomatizabilityofarithmetic,orthe“Entscheidungsproblem”,butitwould<br />

alsodisposeofthewholeofmathematics—insofarasitisunderstoodas<br />

anenterpriseofsolvingproblemssomehowrelatedtohumanlivesrather<br />

thanasapurescienceindulgedinforitsownsake. Hence,thereasonfor<br />

retaininganddevelopingthedifferencebetweenthebroader(andvaguer)<br />

andthenarrower(morelimited)sphereofmethodsliesinthefactthatit<br />

23 Foradetailedaccountsee(Kolman,n.d.).


WhatdoGödelTheoremsTellusaboutHilbert’sSolvabilityThesis? 99<br />

mirrorsthegeneralprocessofexplainingsomethingcomplicatedthrough<br />

somethinglesscomplicated.<br />

Settheoryrunsintoproblemsbecauseofitsfailuretokeepthesedifferencesapart.<br />

Settheoristsbelieve,ontheonehand,thattheContinuum<br />

Hypothesisiseithertrueorfalsewhetherweknowitornot,but,onthe<br />

otherhand,theonlyspecificideatheycangiveusaboutitsstandardmodel<br />

isonelooselyconnectedtoZermelo’sfull-formalism,bywhichitis,however,undecidable,i.e.neithertruenorfalse.So,becausetheonlycriterion<br />

oftruthistheincompleteandpossiblyinconsistentfull-formalism,wemust<br />

facethepossibilitythatthestatusofquestionslike“howbigisthecontinuum?”<br />

maybesimilartothatofquestionslike“howmanyhairsdoes<br />

Othellohave?”,notbecausewedonotyetknowtheanswer,butbecause<br />

noanswerisavailable. Thisdeficitdoesnotmakesuchquestionshumanindependent,butonlydeeplyfictitious,thereasonforwhich,again,isnot<br />

thattheyarestillundecided(suchadecisionisnotdifficulttomake,e.g.,<br />

byendorsing V = L)butbecausenothingreallyimportanthingesonthem.<br />

Myconclusionmayresemblethepositionof(Feferman,1998,p.7),accordingtowhomtheContinuumHypothesis,unlikeHilbert’sSecondProblem,“doesnotconstituteagenuinedefinitemathematicalproblem,”becauseitisan“inherentlyvagueorindefiniteone,asarepropositionsof<br />

highersettheorymoregenerally.”Ihaveattempted,however,tobemore<br />

specificaboutwherethedifferencebetweensettheoryandarithmeticcomes<br />

from.Theso-callediterativehierarchy,describedinapseudo-constructive<br />

mannerbyZermelo’saxioms,isnotamodelinthesamesenseinwhichthe<br />

standardmodelofarithmeticis,becausetheconceptofsubsetisleftunexplained,alongwiththerangeofquantificationandtherespective<br />

(∀)-rule. 24<br />

Tosumup:Hilbert’ssolvabilitythesisisnotrefutedbyGödel’sincompletenesstheorems,norbytheContinuumHypothesis;however,theyoblige<br />

ustorephraseitasfollows:everyproblemis(potentially)solvableifitis<br />

endowedwithwell-definedtruth-conditions,or,asZermelowouldputit,<br />

witha“reasonable”conceptoftruth.<br />

VojtěchKolman<br />

DepartmentofLogics,FacultyPhilosophy&Arts,CharlesUniversity<br />

nám.JanaPalacha2,11638Praha1,CzechRepublic<br />

vojtech.kolman@ff.cuni.cz<br />

24 OnecanpossiblysaythatsettheoryhasfailedbothofFrege’scriteriaforreference,as<br />

describedsoinfluentiallybyQuine,namely“tobeistobeavalueofaboundvariable”<br />

and“noentitywithoutidentity”with“|P(N)| =?”takenasevidence.


100 VojtěchKolman<br />

References<br />

Bridges,D.,&Richman,F. (1987). VarietiesofConstructiveMathematics.<br />

Cambridge:CambridgeUniversityPress.<br />

Ebbinghaus,H.-D.(2007).ErnstZermelo.AnApproachtoHisLifeandWork.<br />

Berlin:Springer.<br />

Ewald,W.(Ed.).(1996).FromKanttoHilbert.ASourceBookintheFoundations<br />

ofMathematicsI–II.Oxford:ClarendonPress.<br />

Feferman,S.(1998).IntheLightofLogic.Oxford:OxfordUniversityPress.<br />

Gödel,K.(1931). ÜberformalunentscheidbareSätzeder‘PrincipiaMathematica’<br />

undverwandterSystemeI.MonatsheftefürMathematikundPhysik,37,349–360.<br />

Gödel,K.(1986).CollectedworksI(S.Feferman,J.Dawson,S.Kleene,G.Moore,<br />

R.Solovay,&J.vanHeijenoort,Eds.).Oxford:OxfordUniversityPress.<br />

Gödel,K.(1995).Somebasictheoremsonthefoundationsofmathematicsand<br />

theirimplications.InS.Feferman,J.Dawson,W.Goldfarb,C.Parsons,&R.Solovay(Eds.),CollectedWorksIII.Oxford:OxfordUniversityPress.<br />

Gödel,K.(2003).CollectedWorksV.CorrespondenceH–Z(S.Feferman,J.Dawson,W.Goldfarb,C.Parsons,&W.Sieg,Eds.).Oxford:OxfordUniversityPress.<br />

Heath,L.,SirThomas.(1931).AManualofGreekMathematics.Oxford:ClarendonPress.<br />

Hilbert,D.(1900).MathematischeProbleme.InVortraggehaltenaufdeminternationalenMathematiker–KongresszuParis1900(pp.253–297).Nachrichtenvon<br />

derKöniglichenGesellschaftderWissenschaftenzuGöttingen.(Pagereferences<br />

aretothereprintin(Hilbert,1935).)<br />

Hilbert,D. (1918). AxiomatischesDenken. MathematischeAnnalen,78,405–<br />

415.<br />

Hilbert,D. (1922). NeubegründungderMathematik.ErsteMitteilung. AbhandlungenausdemmathematischenSeminarderHamburgischenUniversität,<br />

1,157–177.(Pagereferencesaretothereprintin(Hilbert,1935).)<br />

Hilbert,D.(1926). ÜberdasUnendliche.MathematischeAnnalen,95,161–190.<br />

Hilbert,D. (1930). NaturerkennenundLogik. DieNaturwissenschaften,18,<br />

959–963.(Pagereferencesaretothereprintin(Hilbert,1935).)<br />

Hilbert,D.(1931).DieGrundlegungderelementarenZahlentheorie.MathematischeAnnalen,104,485–495.<br />

Hilbert,D.(1935).GesammelteAbhandlungen.DritterBand:Analysis,GrundlagenderMathematik,Physik,Verschiedenes.Berlin:Springer.<br />

Kolman,V. (n.d.). Iscontinuumdenumerable? InM.Peliˇs(Ed.),TheLogica<br />

Yearbook2007(pp.77–86).Praha:Filosofia.<br />

Lorenzen,P.(1955).EinführungindieoperativeLogikundMathematik.Berlin:<br />

Springer.


WhatdoGödelTheoremsTellusaboutHilbert’sSolvabilityThesis? 101<br />

Lorenzen,P.(1962).Metamathematik.Mannheim:BibliographischesInstitut.<br />

Lorenzen,P.(1968).MethodischesDenken.FrankfurtamMain:Suhrkamp.<br />

Lorenzen,P. (1974). KonstruktiveWissenschaftstheorie. FrankfurtamMain:<br />

Suhrkamp.<br />

Lorenzen,P. (1987). ConstructivePhilosophy. Amherst: TheUniversityof<br />

MassachusettsPress.(EditedandtranslatedbyK.R.Pavlovic.)<br />

Lorenzen, P.,&Lorenz, K. (1978). Dialogische Logik. Darmstadt: WissenschaftlicheBuchgesellschaft.<br />

Schütte,K.(1960).Beweistheorie.Berlin:Springer.<br />

Tarski,A. (1936). Opojęciuwynikanialogicznego. PrzeglądFilozoficzny,39,<br />

56–68.<br />

Weyl,H.(1921). ÜberdieneueGrundlagenkrisederMathematik.Mathematische<br />

Zeitschrift,10,39–79.(Pagereferencesaretothereprintin(Weyl,1968,vol.II.).)<br />

Weyl,H.(1968).GesammelteAbhandlungenI–IV.Berlin:Springer.<br />

Zermelo,E. (1932). ÜberStufenderQuantifikationunddieLogikdesUnendlichen.JahresberichtderDeutschenMathematiker–Vereinigung,41,85–88.


Wittgenstein on Pseudo-Irrationals,<br />

Diagonal Numbers and Decidability<br />

Timm Lampert ∗<br />

Inhisearlyphilosophyaswellasinhismiddleperiod,Wittgensteinholds<br />

apurelysyntacticviewoflogicandmathematics. However,hissyntactic<br />

foundationoflogicandmathematicsisopposedtotheaxiomaticapproachof<br />

modernmathematicallogic. TheobjectofWittgenstein’sapproachisnot<br />

therepresentationofmathematicalpropertieswithinalogicalaxiomatic<br />

system,buttheirrepresentationbyasymbolismthatidentifiesthepropertiesinquestionbyitssyntacticfeatures.<br />

Itrestsonhisdistinctionof<br />

descriptionsandoperations;itsaimistoreducemathematicstooperations.<br />

ThispaperillustratesWittgenstein’sapproachbyexamininghisdiscussion<br />

ofirrationalnumbers.<br />

1 Tractarianheritage<br />

IntheTractatus,TLPforshort,Wittgensteindistinguishesbetweenoperationsandfunctions.AsdoRussellandWhiteheadinthePrincipiaMathematica,PMforshort,heuses“functions”inthesenseof“propositional<br />

functions”,whicharerepresentablebysymbolsoftheform ϕxwithina<br />

logicalformalism. Incontrast,theconceptofoperationisWittgenstein’s<br />

owncreation.AccordingtoWittgenstein,the“basicmistake”ofthesymbolismofPMisthefailuretodistinguishbetweenpropositionalfunctions<br />

andoperations(WVCp.217,andTLP4.126).Inthisrespect,thesyntax<br />

ofPMsuffersfromthesamedeficiencyasthesyntaxofordinarylanguage.<br />

Wittgensteindistinguishesbetweenfunctionsandoperationsbythecriterionofthepossibilityofiterativeapplication,TLP5.25f.:<br />

(Operationsandfunctionsmustnotbeconfusedwitheachother.)<br />

Afunctioncannotbeitsownargument,whereasanoperationcan<br />

takeoneofitsownresultsasitsbase.<br />

∗ IamgratefultoVictorRodychfordiscussionsandcomments.


104 TimmLampert<br />

Duetoitspossibleiterativeapplication,anoperationgeneratesaseries<br />

ofinternallyrelatedelements.Thisseriesisdefinedbyaninitialmember,<br />

η,andanoperation, Ω(ξ),thatmustbeappliedtogenerateanewmember<br />

fromapreviousone ξ.Theformofsuchadefinitionis [η,ξ,Ω(ξ)].Thisseries<br />

isnotdefinedasan“infiniteextension”butbytheiterativeapplicationof<br />

anoperationthatdeterminesforms. Thenaturalnumbers,forexample,<br />

aredefinedbytheoperation +1. Startingwith 0asinitialmember,this<br />

yieldstheseries 0, 0 + 1, 0 + 1 + 1etc.,whichisdenotedby [0,ξ,ξ + 1],<br />

cf.TLP6.03.AccordingtoWittgenstein’spointofviewnumbersareforms<br />

definedbyoperations(cf.WVC,p.223).Theyareneitherobjectsdenoted<br />

bynamesnorclassesorclassesofclassesdescribedbyfunctions. While<br />

functionsdeterminetheextensionofapropertyindependentofitssymbolic<br />

representation,operationsdeterminethesyntaxofsymbols.Operationsdo<br />

notrefertoanythingoutsidethesymbols;theydetermineformal(internal)<br />

propertiesratherthanmaterial(external)properties. Operationsdonot<br />

stateanything,butdeterminehowtovarytheformoftheirbases(inputs)<br />

withoutcontributinganycontent.Incontrast,functions,e.g.,“xishuman,”<br />

statethattheirargumentshavesomeproperty,whichisnotdeterminedby<br />

thesymbolofthearguments.Afunctiondeterminesanextensionofobjects,<br />

namelythe“totality”orclassofobjectsthatsatisfythefunction.<br />

Operationsareinternallyrelated,theycan“counteracttheeffectofanother”and“canceloutanother”(TLP5.253);theyformasystem.InTLPWittgensteinreconstructssocalled“truthfunctions”suchasnegation,conjunction,disjunctionandimplicationas“truthoperations”.Theyformthe<br />

systemoflogicaloperations. Likewise,heunderstandsaddition,multiplication,subtractionanddivisionasasystemof“arithmeticoperations”.In<br />

bothcases,thisforcessignificantchangesinthetraditionalsymbolismof<br />

logicandarithmetic.Inlogic,heinventshisab-notation,inwhichthetruth<br />

operatorsarenotrepresentedby ¬, ∧, ∨or →butbyab-operations,which<br />

assigna-andb-polestoa-andb-poles(cf.,e.g.,CL,letters28,32,NL,<br />

pp.94–96,102,MN,pp.114–116,andTLP6.1203). Bythisheintendsto<br />

overcomewithinpropositionallogicthe“basicmistake”ofPMinfailingto<br />

distinguishsymbolicallybetweenoperationsandfunctions. Inarithmetic<br />

hedefinesnaturalnumbersbyoperations,cf.TLP6.02–6.04,andindicates<br />

asymbolismofprimitivearithmeticwhollyrestingonoperations(cf.TLP<br />

6.24f.). HeexplicitlyopposesthistotheFrege’sandRussell’sprogramto<br />

reducemathematicstoa“atheoryofclasses”(TLP6.031),theseclasses<br />

beingdefinedbypropositionalfunctions.<br />

WittgensteincalledforasymbolismbasedonoperationsasacounterprogramtoFrege’sandRussell’slogicism.<br />

Thisstillholdsforhismiddle<br />

period. Insteadofhispeculiarterm“operation,”hefrequentlyusesthe<br />

commonexpression“law,”andinsteadofthetechnicalterm“propositional


WittgensteinonPseudo-Irrationals 105<br />

function,”heusesthelessspecificexpressionof“description”. Yet,he<br />

stillclaimsthatmathematicsisdealingwithsystems,operationsorlaws<br />

andnotwithtotalities,functionsordescriptions(cf.,e.g.,WVC,p.216f.,<br />

orMS107,p.116). Likewise,heclaimsthat“thefalsitiesinphilosophy<br />

ofmathematics”arebasedonaconfusionofthe“internalpropertiesof<br />

aform”,whicharedeterminedbyoperations,and“properties”interms<br />

ofmaterialpropertiesofdailylife,whichareidentifiedbypropositional<br />

functions,cf. PGII, §42. Healsocallstheviewthatbasesmathematics<br />

onfunctionsthe“extensionalview”whereasheprofessesan“intensional<br />

view”thatidentifiesmathematicalpropertiesbysyntacticpropertiesofan<br />

adequatesymbolicrepresentation(PGII,VII, §41,RFMV, §34–40).<br />

InthefollowingwegoontoillustrateWittgenstein’sintensionalviewin<br />

hisintermediate(1929–1934)discussionofirrationalnumbers.Finally,we<br />

willapplythisdiscussiontodiagonalnumbers,aswellastothenotionsof<br />

enumerability,decidabilityandprovability.Weherebywanttoaddresstwo<br />

challengesfacedbyWittgenstein’sprogram:<br />

(i)Howtoapplyittootherpartsofmathematicsbesidesprimitivearithmetic?<br />

(ii)Howtorelateittothebasicnotionsandimpossibilityresultsofmodernmathematicallogic?<br />

2 Irrationals<br />

Cauchysequences<br />

IrrationalsarecustomarilydefinedasequivalenceclassesofidenticalCauchy<br />

sequences.ACauchysequenceisaninfinitesequenceofrationalnumbers<br />

a1,a2,...suchthattheabsolutedifference |am − an|canbemadelessthan<br />

anygivenvalue ǫ > 0whenevertheindices m,naretakentobegreaterthan<br />

somenaturalnumber k. TwoCauchysequences a1,a2,... and a ′ 1 ,a′ 2 ,...<br />

areidenticalifandonlyifforanygiven ǫ > 0thereissomenaturalnumber<br />

ksuchthat |an − a ′ n| < ǫforall ngreaterthan k. Theideabehindthis<br />

definitionisthatallmethodsapproximatingthe“trueexpansion”ofanirrationalnumbermustonceresultinthesameexpansionuptoacertaindigit.<br />

Forexample,themethodsillustratedinTables1and2bothapproximate<br />

thetruedecimalexpansionof √ 2inaplainmanner.<br />

a1 a2 a3 a4 a5 a6 a7 a8 a9<br />

x 2 < 2 1 1.25 1.375 1.40625 1.4140525<br />

x 2 > 2 2 1.5 1.4375 1.421875<br />

Table1.Method1


106 TimmLampert<br />

a1 a2 a3 a4 a5 a6 a7 a8 a9 a10<br />

x 2 < 2 1 1.4 1.41 1.414 1.4141<br />

x 2 > 2 2 1.5 1.42 1.415 1.4142<br />

Table2.Method2<br />

Atsomepointthemethodscomeupwithidenticaldecimalexpansions<br />

uptoacertaindigit. Forexample,from a9onbothsequencesbeginwith<br />

1.41. Thus,goingfurtherandfurtheroneapproximatesmoreandmore<br />

“the”expansionoftheirrationalnumber.However,nofinitesequencewill<br />

everrepresentthe“trueexpansion”,asitisthelimitofallsequencesapproximatingit;the“trueexpansion”isbeyondallfinitesequences—itis<br />

infinite.<br />

WithrespecttoWittgenstein’spointofview,itisimportanttonotethat<br />

thesemethodsofapproximationdonotgeneratethenextdigitsbyiteration.<br />

Instead,atanystepitmustbecheckedwhetherthesquareoftheresultis<br />

< 2or > 2.<br />

Wittgenstein’scritique<br />

Wittgenstein’smaincritiqueofthedefinitionofirrationalnumbersinterms<br />

ofCauchysequencesisthatthisdefinitiondoesnotprovideanidentity<br />

criterion,whichdecidestheidentityoftworealnumbers(PR §§186,187,<br />

191,195). Theproblemisthat,onthestandardconceptionofirrational<br />

numbersasinfinitesequencesofrationalnumbers,foranyinfinitesequence<br />

sthereareinfinitemanysequencesthatareidenticalwith suptoacertain<br />

digit k.However,thedefinitiondoesnotprovideamethodtospecifysome<br />

upperboundfor kincomparingtwoarbitraryrealnumbers. Thus,no<br />

finitecomparisonissufficienttodecidewhethertwoarbitrarysequences<br />

areidentical. Thedefinitionhasitthatthe“trueexpansion”liesbeyond<br />

allfinitesequences. Therefore,itprovidesonlyasufficientcriterionfor<br />

anegativeanswerbutnosufficientcriterionforapositiveanswertothe<br />

questionofidentifyingarbitraryrealnumbers.Inthisrespect,wehavethe<br />

samesituationasinthecaseofdeterminingwithinatraditionallogical<br />

calculuswhethersomeformulaoffirstorderlogicisnotatheorem.<br />

Onemightreplytothiscritiquethatonecannotclaimthedecidability<br />

ofthingsthatsimplyarenotdecidable;thenatureoftherealnumbersas<br />

infinitesequencesimpliesthatonecannotdecideupontheidentityoftwo<br />

realnumbers. However,infactitisfromthepurporteddefinitionthat<br />

theproblemarises,anditisnotcarvedinstonethatthisindeedcaptures<br />

the“nature”ofrealnumbers. AccordingtoWittgenstein’sanalysisthe<br />

definitionisnothingbutaconsequenceoftheextensionalviewofmodern<br />

mathematics. Thisspuriouslytakesthedesignationsofrealnumbersby


WittgensteinonPseudo-Irrationals 107<br />

ordinarylanguageasdescriptionsofeverydayproperties,whichdetermine<br />

acertainextension. Forexample,inthecaseof √ 2onewronglyanalysestheordinaryexplanationintermsof“thenumberthatwhenmultiplied<br />

byitselfisidenticalwith 2”asadescriptionofamaterial,non-symbolic<br />

property. Thispropertyisthenconceivedasbeingsatisfiedbythe“true<br />

infiniteexpansion”,whichisapproximatedbymultiplyingfinitesequences<br />

withthemselvesandcomparingtheresultto 2.InordertocometounderstandWittgenstein’spointofview,itiscrucialtorecognizethatthereisan<br />

alternativetothisconceptionthatreferstoknownmathematics.According<br />

tothispointofview,realnumbersarenotdefinedbyextensions,butby<br />

lawsinthesenseofWittgenstein’soperations.<br />

Wittgenstein’salternative<br />

InordertocometounderstandWittgenstein’spositiononemustrecognizethatherejectsmethodsofapproximationsuchastheaboveillustrated<br />

methods1and2.Althoughthesekindsofmethodsofapproximationmight<br />

becalled“laws,”theyarenot“laws”intermsofoperations.Theyarenot<br />

operationsbecausetheydonotgenerateasequencebyiteration. Howto<br />

goondoesnotsimplydependonthepreviousmembersbutonacomparisonbetweenthelastmemberandsomecondition.<br />

Forexample,ateach<br />

stageinthedevelopmentofthedecimalexpansionof √ 2,onemustconsiderwhethersquaringthelastmemberisgreaterorsmallerthan<br />

2.This<br />

methodisincompatiblewithWittgenstein’spurelysyntacticfoundationof<br />

mathematicalproperties. Inhisprogram,anysequencemustbedefinable<br />

byanoperationthatdeterminesnothingbutthesyntaxofthemembersof<br />

thesequence.Onlyinthiswayisthepropertyconstitutingthesequencereducedtoaninternalpropertyofformsthatcanbeidentifiedbythesymbolic<br />

featuresofthemembersoftheseries.<br />

Wittgenstein’s wellknownrejection of“arithmetical experiments”is<br />

basedonhisrequirementtodefinesequencesbysyntacticmeansalone,<br />

PR §190:<br />

Inthiscontextwekeepcomingupagainstsomethingthatcouldbe<br />

calledan“arithmeticalexperiment”.Admittedlythedatadetermine<br />

theresult,butIcan’tseeinwhatwaytheydetermineit(cf.,e.g.,the<br />

occurrencesof 7in π.)Theprimeslikewisecomeoutfromthemethod<br />

forlookingforthem,astheresultsofanexperiment. Tobesure,I<br />

canconvincemyselfthat 7isaprime,butIcan’tseetheconnection<br />

betweenitandtheconditionitsatisfies. —Ihaveonlyfoundthe<br />

number,notgeneratedit.<br />

Ilookforit,butIdon’tgenerateit.Icancertainlyseealawinthe<br />

rulewhichtellsmehowtofindtheprimes,butnotinthenumbers


108 TimmLampert<br />

thatresult. Andsoitisunlikethecase + 1<br />

1!<br />

canseealawinthenumbers.<br />

1 1 , −3! , + 5! etc.,whereI<br />

Imustbeabletowritedownapartoftheseries,insuchawaythat<br />

youcanrecognizethelaw.<br />

Thatistosay,nodescriptionistooccurinwhatiswrittendown,<br />

everythingmustberepresented.<br />

Theapproximationsmustthemselvesformwhatismanifestlyaseries.<br />

Thatis,theapproximationsthemselvesmustobeyalaw.<br />

TheseriesofprimesisWittgenstein’sparadigmofaseriesthatcannotbe<br />

generatedbyanoperation. Althoughoperationsareavailabletogenerate<br />

aninfiniteseriesofprimes,nooperationisknowntogeneratetheprimesin<br />

acertainorderthatensuresthatallprimesareenumerated.Inhisdetailed<br />

discussionsofprimesinotherplaces,Wittgensteindrawstheconsequence<br />

thatwestilllackofaclearconceptof“the”primes. Allwehaveisa<br />

conceptofwhat“a”primeis,whichallowsustodecidewhetheragiven<br />

numberisprimeornot(PR §159,161,cf.(Lampert,2008)).Forthesame<br />

reason,herejectsthedefinitionofarealnumber Pasthedualfractionwith<br />

an = 1if nisprimeand an = 0otherwise(cf.PGII, §42).Thisdefinition<br />

doessatisfythedefinitionofrealnumbersbyCauchysequences,butitdoes<br />

notsatisfyWittgenstein’scriterionofbeingdefinablebyanoperation. In<br />

thequotedpassage,Wittgensteinemphasizesthatwedohaveamethod<br />

tolookforthenextprime: wegothroughtheseriesofnaturalnumbers<br />

anddecideonebyonewhethereachmembersatisfiestheconditiontobe<br />

divisibleonlyby 1anditself. However,thismethoddoesnotsatisfyhis<br />

standardsofadefinitionbyoperation.Aslongaswearenotabletoreduce<br />

thepropertyofbeingaprimetosomeoperationgeneratingtheseriesof<br />

primesbyiteration,“wecan’tseetheconnection”betweenthemembersof<br />

theseriesandtheconditiontheysatisfy:wecannot“recognizethelaw”in<br />

theseries.Theproblemisthesameaswiththeaboveillustratedmethods<br />

ofapproximating √ 2.Insteadofgeneratingthenextmemberbyiteration,<br />

wemustdecidewhethersomeconditionissatisfiedornotinordertofind<br />

thenextmember.<br />

Wittgenstein’sreferencetotheseriesofprimesasanillustrationofarithmeticalexperimentsdemonstratesthathisconceptofoperationisnotequivalenttothatofprimitiverecursivefunction.Primesaredefinablebyaprimitiverecursivefunction,butnotbyanoperation.<br />

1 Iterationinthecaseof<br />

operationsmeansthattheoutputofthe n th applicationofanoperationis<br />

1 ThequestioninwhatsenseWittgensteincharacterizesrealnumbersas“laws”isthoroughlydiscussedintheliterature(cf.<br />

(DaSilva,1993), (Frascolla, 1994,pp.85–92),<br />

(Marion,1998),(Rodych,1999)and(Redecker,2006,ch.5.2)). However,themainreasonwhytheidentificationoflawswithWittgenstein’snotionofoperationsseemedto


WittgensteinonPseudo-Irrationals 109<br />

itselftheinputofthe n + 1 th applicationoftheverysameoperation. In<br />

contrast,recursioninthecaseofprimitiverecursivefunctionsmeansthat<br />

thevalueofaprimitiverecursivefunction fforthesuccessorof n, S(n),is<br />

definedbyreferringtothevalueoftheverysamefunction ffor n. This<br />

doesnotimplythatthevaluesof farethemselvestheirarguments.Thisis<br />

onlytrueincaseofthesuccessorfunction,whichitselfisprimitiverecursive.However,theidentityfunction,e.g.,<br />

I(x) = x,andthezerofunction,<br />

Z(x) = 0,onwhichthedefinitionofprimitiverecursivefunctionsarebased,<br />

arefunctionsanddonotdefineaseriesbyiterativeapplication.Thesame<br />

holdsforprimitiverecursivecharacteristicfunctions. Theyhavetheform<br />

“f(x) = 0if ϕ(x)and f(x) = 1otherwise.”InWittgenstein’sterms,characteristicfunctionsareaparadigmof“descriptions”andnotofoperations.<br />

Incontrast,anyiterationbyapplyingoperationshastheform an = Ω ′ �ai<br />

where �aistandsformemberspreviousto an. Forexample,theseriesof<br />

Fibonaccinumbersisdefinedby an = an−2 + an−1.Recursioninthecase<br />

ofprimitiverecursivefunctionsispartofastrategyofdefiningprimitive<br />

recursivefunctions,whereasoperationsarenotdefinedbyiterationbutappliediteratively.Theyaredefinedbysomepurelysyntacticvariationthatgeneratesaformalseriesofsystematicallyvariedmembersifiterativelyapplied.<br />

InthecaseofFibonaccinumbers,thisoperationconsistsofadding<br />

thelasttwomembers.Startingfrom 0and 1,thisgeneratestheseries 0, 1,<br />

0+1, 1+(0+1), (0+1)+(1+(0+1)), (1+(0+1))+((0+1)+(1+(0+1)))<br />

etc. 2<br />

IfnotevenprimitiverecursivefunctionssatisfyWittgenstein’sstandards<br />

ofapurelysyntacticfoundationofmathematics,thiscausesdoubtswhether<br />

hisprogrammeisrealizableatall.Likewise,hisrejectionofarithmeticalexperimentsandhisclaimto“recognizethelaw”intheserieshascaused<br />

trouble.ThedecimalsequencesofirrationalsdonotsatisfyWittgenstein’s<br />

demandforsequencesthatmanifestlyobeyalaw.DonotirrationalscontradictWittgenstein’sclaimfromtheirverynature?Thus,itseemsunclear<br />

howWittgenstein’spointofviewcanevendojusticetosuchbasicirrational<br />

numbersas πand √ 2(cf.,e.g.,(Redecker,2006,p.212)).<br />

However,theseproblemsonlyariseifoneoverlooksthefactthatthe<br />

possibilityofdefinitionsbyoperationsdependsonthemodeofrepresentation.Incaseofirrationals,thesyntacticfeaturesofthedecimalsystemareresponsiblefortheir“lawless”representation.However,thiskindofrepresentationisnotessential;itobscurestheirlawfulnatureinsteadofrevealing<br />

it.InMS107p.91,Wittgensteinwrites(translatedbyT.L.):<br />

beinsufficienttomostcommentatorsisthatoperationsinWittgenstein’ssensewerenot<br />

distinguishedsharplyfromthenotionofprimitiverecursivefunctions.<br />

2 Bracketsaremerelyintroducedtoidentify an−2and an−1.


110 TimmLampert<br />

Theprocedureofextracting √ 2inthedecimalsystem,e.g.,isan<br />

arithmeticalexperiment,too. However,thisonlymeansthatthis<br />

procedureisnotcompletelyessentialto √ 2andarepresentationmust<br />

existthatmakesthelawrecognizable.<br />

Toseetheconnectionbetweenthemembersofasequencerepresentinga<br />

realnumberandtheconditionorpropertythatthesememberssatisfy,one<br />

mustrefertoanequivalencetransformationthatreducesthispropertyto<br />

aninternalpropertyofforms. Thereisnoequivalencetransformationbetween<br />

√ 2andadecimalnumber.Thisalreadyshowsthatitisimpossible<br />

torepresent √ 2bythedecimalsystem;whateverdecimalnumberonegenerates,itcannotbeidenticalwith<br />

√ 2—referringto“infiniteextensions”is<br />

justanotherexpressionofthisdeficiency.However,usingtherepresentation<br />

bycontinuedfractions,itispossibletorepresent √ 2byanoperation,cf.<br />

MS107,p.126(translatedbyT.L.,cf.MS107,p.99):<br />

[...] in 1<br />

2 , 1<br />

2+ 1<br />

2<br />

1 ,<br />

2+ 1<br />

2+ 1<br />

2<br />

recognizeinthedecimaldevelopment.<br />

etc. onecanrecognizethelawonecannot<br />

Theconnectionbetweenthepropertyof √ 2as“thenumberthatmultipliedwithitselfisidenticalwith2”anditsdefinitionbyitscontinued<br />

fractionisduetoequivalencetransformation:<br />

x 2 = 2 | √<br />

x = √ 2 | a = 1 + (a − 1)<br />

x = 1 + ( √ 2 − 1) | a = 1 1<br />

a<br />

x = 1 + 1 1<br />

√2−1<br />

x = 1 + 1 √<br />

2+1<br />

√ 2<br />

2 −12 x = 1 + 1 √<br />

2+1<br />

x = 1 + 1<br />

1+x<br />

x − 1 = 1<br />

1+x<br />

1 x − 1 = 2+(x−1)<br />

| 1<br />

a−b<br />

= a+b<br />

a 2 −b 2<br />

| a = a<br />

√ 2 2 −1 2<br />

| x = √ 2, a + b = b + a<br />

| −1<br />

| 1 + x = 2 + (x − 1)<br />

Thus, √ 1<br />

2−1isrepresentablebytheoperation 2+(x−1) .Startingwith 1−1<br />

1<br />

for x−1,theiterativeapplicationofthisoperationyieldstheseries 2+(1−1) ,<br />

2+<br />

1<br />

1<br />

2+(1−1)<br />

, 1<br />

1<br />

2+ 1<br />

2+<br />

2+(1−1)<br />

etc. ThisisidenticaltotheseriesWittgenstein<br />

mentionsifoneeliminates +(1 − 1)byanequivalencetransformation. In


WittgensteinonPseudo-Irrationals 111<br />

theshortnotationofregularperiodiccontinuedfractions, √ 2isdefinableby<br />

[1;2].Acontinuedfractionofarealnumberisperiodicifandonlyifthereal<br />

numberisaquadraticirrational(theoremofLagrange). Thenotationof<br />

continuedfractionidentifiesacommonpropertyofquadraticirrationalsby<br />

acommonsyntacticfeature,andthusshowsthatthispropertyisaninternal<br />

property.Otherirrationalnumbersarerepresentablebyregularcontinued<br />

fractionsthatarenotperiodicbutstilldefinablebyoperations,suchasthe<br />

Eulernumber e : [2;1,2,1,1,4,1,1, 6, 1,1,8,...].Anothertypeofirrational<br />

numbersarenotdefinablebyoperationswithinregularcontinuedfractions<br />

butwithinirregularcontinuedfractionssuchas 4<br />

π = 1+ 12 .Further-<br />

2+ 52<br />

. ..<br />

more,thecontinuedfractionrepresentationforanumberisfiniteifandonly<br />

ifthenumberisrational. Thisshowsthatthismodeofrepresentationrevealsbyitssyntacticpropertiesinternalpropertiesofnumbersthatarenot<br />

identifiedbythedecimalnumbersystem. Welearnmoreabout“thelaws<br />

ofnumbers”,theirinternalstructure,byrepresentingtheminthenotation<br />

ofcontinuedfractions.<br />

Mathematicalproofsrevealthisinternalstructurebyequivalencetransformations.<br />

Consider,forexample,thegoldenratio. Itsrepresentationas<br />

adecimalnumberdoesnotshowitsexceptionalnature. However,byan<br />

equivalencetransformationresultinginanoperationdefiningacontinued<br />

fraction,internalpropertiesofthegoldenratioareidentifiedbythesyntacticfeaturesofthisadequaterepresentation.Thisprocedurereducesthe<br />

propertythattheratiooftwoquantities aand bisidenticaltotheratioof<br />

thesumofthemtothelargerquantity atoanoperation:<br />

φ = a<br />

b<br />

= a + b<br />

a<br />

= 1 + b<br />

a<br />

2+<br />

3 2<br />

1<br />

= 1 + . (1)<br />

φ<br />

Bytheoperation 1 + 1<br />

φ ,theperiodic,regularcontinuedfraction [1;1]is<br />

defined. Bythisrepresentationitisproventhatthegoldenratiois“the<br />

mostirrationalandthemostnoblenumber,”becausethesepropertiesare<br />

identifiedbythelowestpossiblenumbersinaninfiniteregularcontinued<br />

fraction.Furthermore,bythisrepresentationitisproventhattheratioof<br />

twoneighbouredFibonaccinumbersconvergestothegoldenratio.Forthe<br />

Fibonaccinumbersaredefinedby an+1 = an+an−1.Thus,with a = anand<br />

b = an−1weyieldequation(1).Thesyntaxofcontinuedfractionsprovides<br />

symbolicconnectionsthatprovecertaininternalrelationsbetweennumbers.<br />

Thecontinuedfractionrepresentationofanyirrationalnumberisunique.<br />

Thus,anydefinitionofarealnumberbyanoperation(or“induction”)definingacontinuedfractionsatisfiesWittgenstein’scriterionforrepresentinga<br />

realnumber,MS107p.89(translationT.L.):


112 TimmLampert<br />

Iwantarepresentationoftherealnumberthatrevealsthenumber<br />

inaninductionsuchthatIhaveherewiththeonlyproper,unique<br />

symbol.<br />

Itisbythispropertyofuniquenessthatthesymbolicrepresentationof<br />

irrationalsbycontinuedfractionsservesasanidentitycriterion,whichallows<br />

onetocompareirrationalsandrationalnumbers.Theprincipleisthesame<br />

asinthecaseofcomparingfractionsbyconvertingthemtofractionswith<br />

identicaldenominators. Theproblemofdecidingtheidentityofnumbers<br />

resultsfromadeficiencyintheirrepresentation,allowingforambiguity.<br />

Thisdoesnotmeanthattheremustbeoneandonlyonepropernotation<br />

fornumbers. Nordoesitmeanthatcontinuedfractionsare“the”proper<br />

notationofrealnumbers.Differentinternalpropertiesofnumbers,andherewithdifferenttypesofnumbers,maybeidentifiedbydifferentsystemsof<br />

representation.Anddifferenttypesofnumbersmaybecomparablewithin<br />

differentmodesofrepresentation(cf.MS107,p.123).Naturalnumberscan<br />

becomparedaccordingtotheconventionsofthedecimalsystem,fractions<br />

arecomparablebyconvertingthemtofractionswithidenticaldenominator,rationalnumbersandquadraticirrationalsarecomparablebyregular<br />

continuedfractionsetc. Furthermore,newproofsconsistofmakingnew<br />

symbolicconnections.Theyinventnewpossibilitiesofcomparingnumbers<br />

andofrevealingtheirinternalrelations.Notallinternalrelationsofanumbertoothernumbersmustberevealedwithinonlyonenotationalsystem.<br />

Forexample,insteadofrepresenting πbyanirregularcontinuedfraction<br />

√ 2<br />

√ 2+ √ 2<br />

� √ √<br />

2+ 2+ 2<br />

(asquotedabove), 2<br />

πcanalsoberepresentedby 2 · 2 · 2 ·· · ·<br />

or π 2 2 4 4 6 6 8 8<br />

2by 1 · 3 · 3 · 5 · 5 · 7 · 7 · 9 · · · ·.Theinternalpropertiesofdifferentnumbersmaycallforoperationsreferringtodifferentmodesofrepresentation.<br />

Thereneednotbea“systemofirrationalnumbers”inthesenseasthereis<br />

a“systemofnaturalnumbers”ora“systemofrationalnumbers”(cf.PG<br />

II, §42,RFM,app.3, §33).Aswehaveseen,onlyquadraticirrationalsare<br />

definablebyperiodic,regularcontinuedfractions,andanothertypeofirrationalsisnotevendefinablebyregularcontinuedfractions.Differenttypes<br />

ofirrationalsaredefinablebydifferentkindsofoperationswithindifferent<br />

modesofrepresentation.<br />

AccordingtoWittgenstein’sintensionalpointofview,ourmathematicalcomprehensionandknowledgedependsonthesyntaxofmathematical<br />

representation. Thisisnotduetopsychologicalreasons. Instead,thisis<br />

becausemathematicalproofsmakesymbolicconnectionsbetweendifferent<br />

modesofrepresentation,andbecausethesolvabilityofmathematicalproblemsdependsonimposingadequatenotations.Insteadofconcludingfrom<br />

aspecific,deficientmodeofrepresentationthelawlessnatureofirrational<br />

numbers,whichmakesitimpossibletodecideupontheiridentityandwhich


WittgensteinonPseudo-Irrationals 113<br />

invokesmisconceptionssuchas“infiniteextensions”,oneshouldlookforadequaterepresentationsthatrevealtheirlawfulnatureandmakeitpossible<br />

todecideupontheiridentity.Thisisdonebyreducingtheirpropertiesto<br />

operationsinsteadofconceptualizingthemintermsoffunctions.Ifsucha<br />

reductionisnotavailable,thismeansthatonedoesnothaveafullunderstandingofthepropertiesinquestion.<br />

Wecanthenonlyrefertoavague<br />

understandingexpressedwithinadeficient,descriptivesymbolism.Onlyby<br />

imposinganadequateexpressionthatdepictsthosepropertiesbyitssyntacticfeatures,canwebesurethatthosepropertiesareproperlydefined.<br />

Thisapproachisinconflictwithbasicimpossibilityresultsofmodern<br />

mathematicallogic,suchasthenon-enumerabilityoftheirrationals,the<br />

undecidabilityoffirst-orderlogicortheincompletenessoflogicalaxiomatizationsofarithmetics.<br />

ThisdoesnotmeanthatWittgenstein’spointof<br />

viewimpliesthattheseresultsarefalseinthesensethattheirnegationis<br />

true. Instead,hisintensionalviewimpliesthatitdoesnotmakesenseto<br />

speakof“theirrationals”unlessanoperationisknownthatallowsusto<br />

generatethembyiteration(andthustoenumerate“theirrationals”).This,<br />

ofcourse,doesnotmeanthatheclaimsthatsuchanoperationisormust<br />

beavailable. Likewise,hisintensionalviewimpliesthatonecannotspeak<br />

ofdecidabilityorprovabilityinanabsolutesense,suchthatonecansayin<br />

advancethatcertainpropertiesofformulaeofacertainsyntaxarenotdecidableorprovable,independentofthesyntacticmanipulationsthatmight<br />

beinventedtoidentifythoseproperties. AccordingtoWittgenstein“beingatautology”(“beingtrueinallinterpretations”)or“beingatheorem”<br />

offirstorderlogicisnotdefinedproperlyunlesssomesortofequivalence<br />

procedureisinventedthatconvertsfirstorderformulaetoanadequaterepresentationthatidentifiestheirlogicalpropertiesbyitssyntacticproperties.<br />

Fromthispointofview,itcannotbesaidthatitisimpossibletodefinesuch<br />

procedures,becausethepropertiesinquestionthataresaidtobeundecidableorunprovablearenotrepresentedproperlyunlesssuchproceduresare<br />

available. Likewise,fromWittgenstein’spointofviewtheincompleteness<br />

ofaxiomaticsystemsofarithmeticmeansinthefirstplacethatthosesystemsdonotproperlyrepresentthepropertiesinquestion.Itdoesnotmean<br />

thatweknowthatacertainpropertyholds,butitsformalrepresentationis<br />

notderivable.Instead,itmeansthatwehaveadeficientunderstandingof<br />

thatpropertyexpressedbyaninadequaterepresentation.Inthefollowing,<br />

wewillshowthatthisconflictbetweenWittgenstein’spointofviewand<br />

theimpossibilityresultscanallbetracedbacktohisrejectionof“descriptions”intermsofcharacteristicfunctionsasadequateformstorepresent<br />

realnumbers.


114 TimmLampert<br />

3 Pseudo-Irrationals<br />

Wittgensteinillustrateshispointofviewbyprovidingseveraldefinitionsof<br />

pseudo-irrationals. ThesearedefinitionsofirrationalsintermsofCauchy<br />

sequences. However,contraryto √ 2or πnoreductionstooperationsof<br />

thesedefinitionsareavailable.Thus,accordingtoWittgensteinthereareno<br />

irrationalscorrespondingtothosedefinitions.Besidestheabovementioned<br />

definitionof Pasthedualfraction 0.a1a2 ...with an = 1if nisprimeand<br />

an = 0otherwise,Wittgensteindiscussesthefollowingdefinitions(cf. PG<br />

II, §42):<br />

π ′ :Thedecimalnumber a1.a2a3 ...with anan+1an+2 = 000<br />

if anan+1an+2 = 777in π;otherwise an = anof π.<br />

F:Thedualfraction 0.a1a2a3 ...with an = 1if x n + y n = z n issolvable<br />

for n(1 ≥ x,y,z ≥ 100);otherwise an = 0.<br />

Allthesedefinitionsareintendedtodefineanirrationalnumberbya<br />

characteristicfunction.Inthiscase,thedots“...”refertoan“infiniteextension”.Thus,theyareill-definedaccordingtoWittgenstein’sstandards.<br />

Theydonotidentifyanumberbutdescribeanarithmeticalexperiment.<br />

Wittgensteinemphasizesthatevenifthecharacteristicfunctionsbecome<br />

reducibletooperations,thisdoesnotmeanthatthisshowsthatthedefinitionsinfactdefineirrationalnumbers.<br />

Instead,itmeansthatvague<br />

definitionsthatdonotidentifynumbersarereplacedwithexactdefinitions<br />

thatareabletoidentifynumbers.He,forexample,considersthesituation<br />

whenFermat’stheoremisproven. Duetohisrejectionofdescriptions,he<br />

doesnotanalysethissituationintermsofcomingtoknowthenumber F<br />

thatbeforewasonlydescribed.Instead,theproofallowsonetoreplacethe<br />

pseudo-definitionof F,whichdoesnotidentifyanumber(neitherarational<br />

noranirrationalone),with F = 0.11,whichisarationalnumber(PGII,<br />

§42).Before,itwasnotdecidablewhether“F”denotesanumbersuchthat<br />

F = 0.11ornot;thedefinitionbydescriptionsimplydidnotdefinerules<br />

todothis.Thisdemonstratesthelackofmeaningthatisgivento“F”by<br />

thepreviousdefinition.Theproof,ifitisvalid,makesconnectionstoother<br />

partsofmathematicsthatwerenotrecognizedbeforeandthusgives“F”a<br />

clearmeaning.<br />

Cantor’sproofofthenon-enumerabilityofirrationalnumbersisbasedon<br />

definingadiagonalnumberbyacharacteristicfunction. Givensomeenumerationofdualfractionsbetween<br />

0and 1,theproofofthenon-enumerabilityof“all”ofthemisbaseduponthefollowingdiagonalnumber<br />

D:<br />

D:Thedualfraction 0.a1a2 ...with an = 0ifthe n ′th digitofthe n ′th dual<br />

fractionis 1;otherwise an = 1.


WittgensteinonPseudo-Irrationals 115<br />

Tothisdefinition,thesameobjectionsapplyastothedefinitionsof P, π ′<br />

or F:Itisadefinitionbydescriptionintermsofacharacteristicfunction.<br />

Itdescribesanarithmeticalexperimentanddoesnotidentifyanumber,<br />

whichcanonlybedonebyanoperation. However,suchanoperationis<br />

notavailable. Thus,itisnotmeaningfultosaythat Disan“irrational<br />

number”notoccurringintheassumedenumerationofirrationals. This,<br />

ofcourse,doesnotmeanthatWittgensteinclaimsthat“theirrationals”<br />

areenumerable. Instead,heobjectstoidentifyingirrationalnumbersby<br />

non-periodic,infinitedecimalordualfractions.Thiscriteriondoesnotsay<br />

anythingaboutacertaintypeofnumbers;itonlysayssomethingaboutthe<br />

deficiencyofthedecimalnotation(PGII, §41).Thisnotationcannotserve<br />

astheuniquenotationforrealnumbers,asitdoesnotmakeitpossible<br />

todecideupontheidentityofnumbers.Likewise,Wittgensteinobjectsto<br />

thepictureofarealnumberasa“point”onthe“line”ofrealnumbers.<br />

Theseitemsareelementsoftheextensionalview.Theyarisefromtreating<br />

“isanirrationalnumber”aswellas“isarationalnumber”or“isanatural<br />

number”asconcepts(propositionalfunctions)identifyingcertainsetsof<br />

numbers. Thismakesitpossibletoaskaboutthe“cardinality”ofthose<br />

sets.This,inturn,allowsone<br />

(i)touse“infinite”asanumberwordandspeakof“theinfinitenumber”<br />

ofobjectssatisfyingsomeconcept,and<br />

(ii)tocomparethecardinalityofsetsbycoordinatingtheirelements.<br />

Finally,fromthisandthemethodofdiagonalizationonecomestospeakof<br />

setswithacardinalitygreaterthanthatofthesetofnaturalnumbers.First<br />

andforemost,Wittgenstein’scriticismisthatthisconceptualmachineryis<br />

ratheranexpressionoftheextensionalviewthanadescriptionofthenature<br />

ofnumbers(RFM,app.3, §19).Hecutstherootsof(transfinite)settheory<br />

byconceptualizing“typesofnumbers”intermsof“systems”insteadof<br />

“sets”.Accordingtohisintensionalpointofview,thecriteriontoidentify<br />

atypeofnumberisthepossibilitytogeneratethembyanoperation. As<br />

thisimpliestheirenumerabilityintermsoftheiterativeapplicationofan<br />

operation,itdoesnotmakesensetospeakoftypesofnumbersthatarenot<br />

enumerable.<br />

AccordingtoChurch’sthesis,theconceptofdecidabilityisrepresentable<br />

byaprimitiverecursivecharacteristicfunction. Thus,onthebasisofan<br />

enumerationoffirst-orderlogicformulaebytheirGödelnumbers,thepropertyofbeingatheorem(oratautology)isrepresentablebythefollowing<br />

number:<br />

T:Thedualfraction 0.a1a2 ...with an = 0if ⊢ ϕn(or |= ϕn);and an = 1<br />

otherwise.


116 TimmLampert<br />

Onthebasisofdiagonalization,undecidabilityproofsdemonstratethat<br />

characteristicfunctionssuchastheonedefining Tcannotbeprimitiverecursive.<br />

FromWittgenstein’spointofview,theseproofsarebasedupon<br />

aconfusionofmaterialandformalproperties. Asaformalproperty,theoremhood(orbeingatautology)isnotrepresentablebyacharacteristic<br />

function. Instead,thesepropertiesareonlyrepresentedadequatelybya<br />

sharedsyntacticpropertyinanidealnotation. Thisisillustratedbythe<br />

representationoftautologiesviatruthtablesordisjunctivenormalformsof<br />

propositionallogicaswellasbymeansofVenndiagramsinmonadicfirst<br />

orderlogic.Wittgenstein’sconceptioncallsforequivalencetransformations<br />

toidentifythetruthconditionsoflogicalformulaebymeansofsyntactic<br />

propertiesoftheirproperrepresentation.Thisconceptiondiffersfromthe<br />

traditionalsemanticsoffirst-orderlogic.Presuminganendlessenumeration<br />

ofinterpretations ℑ1, ℑ2,...,eachbeingeitheramodeloracounter-model<br />

ofaformula A,onemightrepresentthetruthconditionof Aaccordingto<br />

theseinterpretationsbythefollowingnumber:<br />

θ(A):Thedualfraction 0.a1a2 ...with an = 0if ℑn |= Aand an = 1<br />

otherwise.<br />

Onthecontrary,Wittgenstein’sapproachcallsforarepresentationof<br />

thetruthconditionsofaformula Athatallowsonetoidentifythetruth<br />

conditionsof Awithoutdecidingwhethersingleinterpretationsaremodels<br />

orcounter-modelsof A.Furthermore,theproperrepresentationoffirstorderformulaeshouldrevealtheinternalrelationsofnon-equivalentlogical<br />

formulaebymakingitpossibletogeneratethesystemoftruthconditions<br />

byoperations.TohaveanideaofwhatWittgensteinenvisages,onemight<br />

thinkofasystematicgenerationofreduceddisjunctivenormalformsofthe<br />

Quine–McCluskeyalgorithm, 3 thatrepresentallpossibletruthfunctions<br />

ofpropositionallogic. Likewise,thetaskoffirstorderlogicistodefine<br />

analogousdisjunctivenormalformsandproceduresfortheiruniquereductionwithinfirstorderlogic.Toclaimthatthisisimpossiblepresumesthe<br />

extensionalviewthatisrejectedbyWittgenstein’sendeavour.<br />

Likewise,Gödelrepresents“xisaproofof y”byaprimitiverecursive<br />

function xByindefinition45ofhisincompletenessproof(cf.(Gödel,1931,<br />

p.358)).Onthisbasis,heexpresses“xisprovable”by ∃yyBxindefinition<br />

46. ThisisincompatiblewithWittgenstein’sclaimthattheinternalrelationofbeingprovable(derivable)shouldbedefinedbyoperationsinsteadof<br />

propositionalfunctions.This,inturn,presumesaproofprocedureinterm<br />

3 NotethatthereduceddisjunctivenormalformsoftheQuine–McCluskeyalgorithm<br />

areunique; anyequivalentpropositionalformulaisrepresentedbythesamereduced<br />

disjunctivenormalform.AmbiguityonlycomesintoplayinthesecondstepoftheQuine–<br />

McCluskeyalgorithmthatintendstominimizereduceddisjunctivenormalforms.


WittgensteinonPseudo-Irrationals 117<br />

ofequivalencetransformationstoanadequatesymbolismthatmakessuch<br />

adefinitionpossible,insteadofaproofprocedureintermsoflogicalderivationsfromaxioms.Thelackofsuchadefinitionmeansadeficiencyinthesyntacticrepresentationoftheformulaeinquestion.AccordingtoWittgenstein’spointofview,theconclusionthatmustbedrawnfromGödel’sincompletenessproofistolookforaformalrepresentationofarithmeticthat<br />

isnotbasedupontheconceptofpropositionalfunction,whichisatthe<br />

heartofanylogicalformalization.<br />

Wittgenstein’sintensionalreconstructionofmathematicsisnotmeant<br />

tobea“refutation”oftheextensionalviewofmodernmathematicallogic.<br />

Instead,firstandforemostitintendstoproposeadecisivealternativeconceptualizationofmathematicsthatradicallydiffersinitsfoundations.Accordingtohim,thefruitofthisendeavourshouldbeaclarificationofthephilosophicalproblemsofmodernmathematicsthatwillhavethesameinfluenceontheincreaseofmathematicsassunshinehasonthegrowthof<br />

potatoshoots(PG,II, §25).<br />

TimmLampert<br />

UniversityofBerne<br />

timm.lampert@philo.unibe.ch<br />

Abbreviations<br />

CL Wittgenstein,L.(1997).CambridgeLetters,Oxford:Blackwell.<br />

MN Wittgenstein,L.(1979). NotesdictatedtoG.E.MooreinNorway,in:<br />

Notebooks1914–1916,pp.108–119.Oxford:Blackwell.<br />

MS107 Wittgenstein,L.(2000)Manuscript107accordingtovonWright’scatalogue.<br />

PublishedinWittgenstein’sNachlass: theBergenElectronic<br />

Edition.London:OxfordUniversityPress.<br />

NL Wittgenstein,L.(1979). NotesonLogic,in: Notebooks1914–1916,<br />

pp.93–107.Oxford:Blackwell.<br />

PG Wittgenstein,L.(1974). PhilosophicalGrammar,Blackwell: London<br />

1974.<br />

PR Wittgenstein,L.(1975)PhilosophicalRemarks,Oxford:Blackwell.<br />

RFM Wittgenstein,L.(1956)RemarksontheFoundationsofMathematics,<br />

Oxford:Blackwell.<br />

TLP Wittgenstein, L. (1994). Tractatus Logico-Philosophicus. London:<br />

Routledge.<br />

WVC Waismann, F. (1979) Wittgenstein and the Vienna Circle, Oxford:<br />

Blackwell.


118 TimmLampert<br />

References<br />

DaSilva,J.J. (1993). Wittgensteinonirrationalnumbers. InK.Puhl(Ed.),<br />

Wittgenstein’sphilosophyofmathematics(pp.93–99).Vienna:Hölder–Pichler–<br />

Tempsky.<br />

Frascolla,P.(1994).Wittgenstein’sphilosophyofmathematics.Routledge:London.<br />

Gödel,K.(1931). ÜberformalunentscheidbareSätzederPrincipiaMathematica<br />

undverwandterSysteme.MonatsheftefürMathematikundPhysik,38,173–198.<br />

Lampert,T. (2008). Wittgensteinontheinfinityofprimes. Journalforthe<br />

HistoryandPhilosophyofLogic,29,63–81.<br />

Marion,M.(1998).Wittgensteinandfinitism.Synthese,105,141–176.<br />

Redecker,C.(2006).Wittgensteinsphilosophiedermathematik.Frankfurt:Ontos.<br />

Rodych,V. (1999). Wittgensteinonirrationalsandalgorithmicdecidability.<br />

Synthese,118,279–304.


What is the Definition of ‘Logical Constant’?<br />

Rosen Lutskanov ∗<br />

Thedesignofthispaperistomotivateandintroduceinformallyadefinitionofthenotionof‘logicalconstant’whichdoesnotpresupposethe<br />

analytic/syntheticdistinction.Tothisend,I’mgoingto<br />

1<br />

1.exploretheoriginofthisnotion;<br />

2.showwhyitisimportanttodefineit;<br />

3.reviewsomeparadigmatic(butostensiblyunsatisfactory)allegeddefinitions;<br />

4.hintatthetruerelationbetweenthenotionsof‘logicalconstant’and<br />

‘analyticity’;<br />

5.makemanifesttheimplicitrenderingofanalyticitywhichisnestedin<br />

theclassicaldefinitionsoflogicalconstants;<br />

6.discussthestrictlyalternativeconstrualofanalyticityprominentin<br />

present-dayphilosophyoflogic;<br />

7.providesketchydefinitionoflogicalconstantsthatdeviatesfromthe<br />

firstbutremainstruetothesecond.<br />

ItwasBolzano,whointhedistant1837wasprobablythefirsttosuggest<br />

thatthereareconceptsbelongingtologicalone:accordingtohisownexample,thefactthatthequestion“whethercorianderimprovesone’smemory”obviouslydoesnotconcernlogicatall,suggeststhatitisnotaboutcorianderbutstudiessomethingdifferent(Hodges,2006,p.42).Inhisownview,<br />

thelogic’ssubjectmatterisexhaustedbythe‘logicalideas’whichaffectthe<br />

∗ IwouldliketoexpressmydeepgratitutetotheorganizersofLOGICA2008forthe<br />

granttheyhaveawardedmeforparticipationintheconference.


120 RosenLutskanov<br />

logicalformofpropositions(or‘sentencesinthemselves’)andaretobesingledoutbytheconditionthattheirvariancemodifiesthetruth-valueofany<br />

expressioncontainingthem(Siebel,2002,p.590).ThencameFregewhoin<br />

his“Begriffsschrift”(1879)hadaclear-cutdivisionofsymbols,employedin<br />

formallanguages,intotwokinds: ‘thosethatcanbetakentomeanvariousthings’(variablearguments)and‘thosethathaveafullydeterminate<br />

sense’(constantfunctions). Asfaraslogicisconcerned,thesecondkind<br />

ofsymbolscorrespondstoBolzano’s‘logicalideas’becauseitrepresents<br />

thosepartsoftheformalexpressionthathavetoremaininvariantunder<br />

replacement(Frege,1960,p.13).FinallyourstoryreachesRussell,whoin<br />

1903introducedthenowfamiliarterm‘logicalconstant’assubstitutefor<br />

Bolzano’s‘logicalideas’andFrege’s‘logicalfunctions’. Accordingtohim<br />

thesearethenotionsaccountableforthetruthofallpropositionswhichwe<br />

viewasapriorijustified.Buthedidnotprovideaformalcharacterization,<br />

onlythefollowingdeliberatelyconfusingexplanation:“logicalconstantsare<br />

allnotionsdefinableintermsofthefollowing:Implication,therelationofa<br />

termtoaclassofwhichitisamember,thenotionofsuchthat,thenotion<br />

ofrelation,andsuchfurthernotionsasmaybeinvolvedinthegeneralnotionofpropositionsoftheaboveform”(Russell,1903,p.3).Thereasonfor<br />

suchstrikingobscurityisthefactthathethoughtthat“logicalconstants<br />

themselvesaretobedefinedonlybyenumeration,fortheyaresofundamentalthatallthepropertiesbywhichtheclassofthemmightbedefined<br />

presupposesometermsoftheclass”(Russell,1903,pp.8–9).<br />

2<br />

LaterRussell’sinventionsurvivedthedemiseofhislogicism,althoughits<br />

introductionwasinitiallymotivatedaspartoftheattempttoshowthat<br />

allmathematicalnotionsarereducibletothenotionsoflogic(exemplified<br />

bythe‘logicalconstants’). Todaywearegenerallyinclinedtoclaimthat<br />

“logicalconceptiswhatcanbeexpressedbyalogicalconstantinalanguage”hencethequestion“Whatislogic?”istobeansweredbyanswering<br />

thequestion“Whatisalogicalconstant?” (Hodes,2004,p.134). Onthe<br />

otherhand,wejustcannotaffordtreatingthenotionoflogicalconstant<br />

asindefinableasRusselldid,sincepresentlywehaveatourdisposalalternativelistsoflogicalconstantsimposingonusdifferentconceptionsabout<br />

thesubjectmatteroflogic.Famously,QuinedidhisbesttoexpelRussell’s<br />

“relationofatermtoaclassofwhichitisamember”fromthelistoflogicalconstants,claimingthatthetheoryofthe‘∈’-relationisnotlogicbut<br />

“settheoryinsheep’sclothing”(Quine,2006,p.66).Thisexcommunication<br />

ofset-membershipfromtheprovinceoflogicisthesoledifferencebetween<br />

Quine’snominalisticpreferenceforfirst-orderlogicandRussell’sontologi-


WhatistheDefinitionof‘LogicalConstant’? 121<br />

callyexuberanttypetheory. Inthefaceofthismanifestdiscrepancy,we<br />

havetoadmitthattheonlywaytoprovideamotivatedchoiceoflogical<br />

frameworkistoexhibitjustifieddefinitionoftheterm‘logicalconstant’and<br />

toshowhowourconceptionoflogicstemsoutofit. Theage-old‘laundrylist’comprisingthevenerablemembersofthefamilyoflogicalnotions<br />

isnotenough. So,whichdefinitionsof‘logicalconstant’arecurrentlyin<br />

circulation?<br />

3<br />

Luckily,wehaveplentyofanswersofthistoilsomequestion;regrettably,<br />

noneofthemfaredverywell.ThefirstattempttoproviderigorousdefinitionwasprovidedbyCarnapinhis“LogicalSyntaxofLanguage”whichwas<br />

subsequentlysimplifiedbyTarski. Hisdefinitionwasfoundedontheconceptsof‘premiss-class’and‘range’(Spielraum):<br />

twopremiss-classeswere<br />

saidtobe‘equipollent’ifeachofthemisconsequenceoftheotherandthe<br />

rangewasdefinedasclassofpremiss-classes Mwiththepropertythateach<br />

premiss-classwhichisequipollenttoapremiss-classbelongingto Malso<br />

belongsto M.ThenCarnapexplainedthattherange Mofaproposition p<br />

represents“theclassofallpossiblecasesinwhich pistrue”or“thedomain<br />

ofallpossibilitiesleftopenby p”(Carnap,1959,p.199).Inthissetting,it<br />

seemsnaturaltodefinethe‘logicaljunctions’assimpleset-theoreticaloperationsonranges:by’supplementary’rangeofagivenrange<br />

M1wemean<br />

arange M2comprisingthosepremiss-classesthatdon’tbelongto M1;then<br />

foraproposition p1withrange M1wecandefineits‘negation’ p2asthe<br />

propositionwhoserangecoincideswiththesupplementaryrangeof p1.In<br />

thesamevein,wecandefinethe‘disjunction’oftwopropositions p1and p2<br />

asanotherproposition p3whoserangeistheunionoftherangesof p1and<br />

p2(Carnap,1959,p.200). AyearlaterTarskishowedthatthedefinition<br />

canbesimplifiedbysubstituting‘content’for‘range’(thecontentof pis<br />

theclassofallnon-analyticconsequencesof p):then p2isnegationof p1iff<br />

theyhaveexclusivecontentsand p3isdisjunctionof p1and p2iffitscontentisproductofthecontentsof<br />

p1and p2(Carnap,1959,p.204).These<br />

attempteddefinitionsofCarnapandTarskiwerenotconceivedassatisfactory,probablybecausetheyfoundedtheconceptualapparatusoflogicon<br />

theconceptualapparatusofsettheory. Thisisnotanepistemologically<br />

flawlessmove:theoperationofsentencenegationseemsmorefamiliarthan<br />

theintricateoperationofclasscomplementation;thatiswhythefirstisnot<br />

tobedefinedbymeansofthesecond.<br />

MaybethisisthereasonwhylaterTarskitookanothercourse. In<br />

hisfamouslecture“Whatarelogicalnotions”(1966)heproposedthenow<br />

classicaldefinition:logicalarejustthesenotionswhichareinvariantunder


122 RosenLutskanov<br />

allpermutationsoftheuniverseofindividualsontoitself. Thisdefinition<br />

provokedseverecriticismsbecauseittreatsaslogicalpropertiesallcardinalityfeaturesofthedomainofdiscourse.<br />

Anotherpainfuldefectwas<br />

exposedbyMcGeewhodefinedanoperationof‘wombatdisjunction’(∪W)<br />

suchthat‘p ∪W q’istrueif‘p ∨ q’istrueandtherearewombats(there<br />

isanelementofthedomainofthemodelwhichsatisfiesthepredicate‘is<br />

wombat’)andfalseotherwise(Feferman,1997,p.9–10). Clearly,wombat<br />

disjunctionisinvariantunderarbitrarypermutations,butitishardtoadmitthatitislogicalnotion—inordertoestablishthetruthorfalsityof<br />

anypropositioncontainingessentialoccurrencesofwombatdisjunctionwe<br />

needtocorroborateaspecificempiricalassumptionconcerningtheexistence<br />

(ornon-existence)ofwombats. Thereareseveralwell-knownattemptsto<br />

rectifyTarski’sdefinitionbyreplacing‘invarianceunderarbitrarypermutations’with‘invarianceunderarbitrarybijections’(Mostowski,1957),‘rigidinvarianceunderarbitrarybijections’(McCarthy,1981),and‘invarianceunderarbitraryhomomorphisms’(Feferman,1997).Asfarasweknow,noone<br />

oftheseattemptsisabletodiscriminateproperlybetweenthelogicaland<br />

theempirical(Mostowski’scriterionqualifies‘unicorn’aslogicalnotion)or<br />

thelogicalandthemathematical(Feferman’scriterionrenders‘thereexist<br />

infinitelymany’asbelongingtologic). Thatiswhy,wecanrecapitulate<br />

thispartofthediscussionbynoticingthat“itseemsinevitabletoconclude<br />

thattheseproposalsinspiredbyTarski... donotevenmeettheminimal<br />

requirementofextensionaladequacy”(Gomez-Torrente,2002,p.20).<br />

Athirdvariantfordefinitionofthenotionoflogicalconstantstemsfrom<br />

theworksofGentzen. Hisfollowerswereinclinedtoclaimthatlogical<br />

constantsaretobeidentifiedsolelybytheintroductionandelimination<br />

rulesgoverningtheirinferentialuses.Conjunction,forexample,isnothing<br />

butthispartofourlexiconthatfeaturesininferenceslike<br />

A, B<br />

A ∧ B<br />

and<br />

A ∧ B<br />

A,B .<br />

ThisbrightideawasshatteredbyPrior,whoprovidedhisinfamous tonkcounterexampledealingwithanewparticle‘tonk’governedbythefollowing<br />

rules:<br />

A<br />

Atonk B<br />

( tonk-Int) and<br />

Atonk B<br />

B<br />

( tonk-Elim).<br />

Theintroductionof‘tonk’allowsshowingtheformallanguageinquestion<br />

tobeinconsistent:justsubstitute‘¬A’for‘B’andapplysuccessively(tonk-<br />

Int)and(tonk-Elim). Thiswasintendedtomeanthatnotanysetofintroductionandeliminationrulesdefinesalogicalconstant:somethingmore<br />

hadtobeadded. Themysteriousadditionalingredientwaslateridentifiedas‘conservativity’(Belnap)or‘harmony’(Dummett).<br />

InDummett’s


WhatistheDefinitionof‘LogicalConstant’? 123<br />

ownexplanation,“Letuscallanypartofadeductiveinferencewhere,for<br />

somelogicalconstant c,ac-introductionruleisfollowedimmediatelybya<br />

c-eliminationrulea‘localpeakfor c’. Thenitisarequirement,forharmonytoobtainbetweentheintroductionrulesandeliminationrulesfor<br />

c,<br />

thatthelocalpeakfor cbecapableofbeingleveled,thatis,thatthereisa<br />

deductivepathfromthepremisesoftheintroductionruletotheconclusion<br />

oftheeliminationrulewithoutinvokingtherulesgoverningtheconstant<br />

c”(Dummett,1991,p.248). AsDummetthimselfreadilyacknowledged,<br />

“Theconservativeextensioncriterionisnot,however,tobeappliedtomore<br />

thanasinglelogicalconstantatatime.Ifwesoapplyit,weallowforthe<br />

priorexistence,inthepracticeofusingthelanguage,ofdeductiveinference,<br />

sincethereareanumberoflogicalconstants”but“theadditionofjustone<br />

logicalconstanttoalanguagedevoidofthem... cannotyieldaconservativeextension”since“ifdeductiveinferenceisevertobesaidtobeableto<br />

increaseourknowledge,thenitmustsometimesenabletorecognizeastrue<br />

astatementthatweshouldnot,withoutitsuse,beenablesotorecognize”<br />

(Dummett,1991,p.220).Thisdifficultyseemsinsurmountable:wecanuse<br />

thelevelingoflocalpeakstechniquetoidentifyasingleparticleaslogical<br />

constant,butitisnotpossibletorelyonthesamestrategytodelineatethe<br />

realmoflogicalnotions.<br />

4<br />

Uptothispointwehavereviewedthreeparadigmaticattemptstoprovide<br />

definitionofthenotionoflogicalconstant.Itappearsthatnoneofthemis<br />

materiallyadequate:<br />

(i)Carnap’sset-theoreticapproachconstruedlogicalnotionsusingprecisemathematicalmethodsbutdidnotevenposethequestionwhich<br />

operationsonpremiss-classesaretobeviewedasbelongingtologic;<br />

(ii)Tarski’smodel-theoreticapproachcouldnotsingleouttheclassof<br />

logicalconstantsandexperiencedseriousdifficultieswithborderline<br />

casessuchasnon-existentobjectsandmathematicalentities;<br />

(iii)Dummett’sproof-theoreticapproachprovidedjustifiedcriteriaforlogicalityofsingleconnectives(‘intrinsicharmony’)butcouldn’tachieve<br />

generallyapplicablestandard(for‘totalharmony’).<br />

Butthematerialinadequacyisnotthesoleoreventhegravestshortcoming<br />

ofthesepurporteddefinitions. Theyallweredevisedwithaneyeonthe<br />

notionscurrentlyrecognizedas‘logical’butwerenotcouchedinabroad<br />

theoreticalframework,clarifyingtheirinterplaywithsomeparticularrenderingofthenotionofanalyticity.<br />

Ifweturnbackweshallseethatthe


124 RosenLutskanov<br />

adventoflogicalconstantswasnecessitatedbythefactthatthe‘analytic<br />

program’(theattempttoidentify‘logicaltruth’and‘analyticaltruth’)was<br />

essentialpartofthe‘logicistprogram’:atruthisanalyticifitcanbereducedtogenerallogicallawsanddefinitions.<br />

Inanutshellthisreduction<br />

establishesthatonlylogicalconstantsoccuressentiallyinitandthelogicalconstantsweredrivenoutonstagesimplytoprovideatouchstonefor<br />

terminationofthisreductiveprocedure.Thatiswhy,“Thequestion‘What<br />

isalogicalconstant?’ wouldbeunimportantwereitnotfortheanalytic<br />

program”(Hacking,1994,p.3). Nowweareabletoperceivewherethe<br />

realproblemlies:ontheonehand,whenwetrytodefinelogicalconstants<br />

anddologic,wesilentlypresupposethatitispossibletodiscriminaterigorouslybetweenanalyticallytrue(truebyvirtueoflinguisticconventions)<br />

andsyntheticallytrue(truebyvirtueofbrutemattersoffact);ontheother<br />

hand,whenwetrytomakesenseofwhatwearedoinganddophilosophy<br />

oflogic,weovertlyblurtheanalytic/syntheticdistinction.Inthefollowing<br />

twoparagraphsI’lldomybesttoexplainwhythisdouble-mindednessisso<br />

crucialinthepresentcontext.<br />

5<br />

Whenwedomathematicallogic,weinvariablyandunwittinglysticktothe<br />

‘Viennese’orthodoxy.Thewayformallanguagesarepresentedandlogical<br />

symbolsareemployedwasmodeledupontheparadigmofWittgenstein’s<br />

Tractatus. Letusrememberthathis“fundamentalidea”wasthatwhile<br />

allotherwordsstandforobjects,“thelogicalconstantsarenotrepresentatives”(Wittgenstein,1963,prop.4.0312).<br />

Thisconceptionwasthesole<br />

basisoftheideathatthe‘real’propositionsareempiricallycontentful‘picturesofreality’,whilethepropositionsoflogicarerepresentationallyidle<br />

‘tautologies’(Wittgenstein,1963,prop.4.462).Carnaprehearsedthesame<br />

lineofthoughtinhisworksonformalsemantics:hestartedwiththesuggestionthat“wemustdistinguishbetweendescriptivesignsandlogicalsigns<br />

whichdonotthemselvesrefertoanythingintheworldofobjects,butserve<br />

insentencesaboutempiricalobjects”(Carnap,1958,p.6)andconcluded<br />

thatitispossibletoclassifyanysentenceas‘L-sentence’(thatis,‘logical’<br />

=‘analytic’=‘trueorfalseonlogicalgrounds’)or‘F-sentence’(‘factual’<br />

=‘synthetic’=‘trueorfalsebyvirtueoffactsoftheworld’). Although<br />

developedindifferentsetting,Tarski’smodel-theoreticapproachtoformal<br />

semanticsreiteratesthesamestepswhicharemirroredinthetwotypesof<br />

clausesinhisrecursivetruthdefinition: ontheoneside,wehaveabase<br />

clauseintroducingavaluationfunctionthatassignstruth-valuestoatomic<br />

sentencesinthemodel(heresentencesreceivetruth-valuesonextra-logical<br />

reasons;ifwehaveinmindsomeparticularinterpretationofthelanguage


WhatistheDefinitionof‘LogicalConstant’? 125<br />

wecansaythattheyare‘trueorfalsebyvirtueoffactsoftheworld’);<br />

ontheotherside,wehaverecursiveclausewhichdeterminesinwhatway<br />

thetruth-valuesofcomplexsentencesbuiltfromatomiconesandlogical<br />

constantsdependonthetruth-valuesalreadyassignedtoatomicsentences<br />

(heresentencesreceivetruth-valuesonintra-logicalreasons,thedefinitional<br />

sub-clausesfortheparticularlogicalconnectivesareanalyticallytruelinguisticconventionsfixingthemeaningoflogicalvocabulary).Finally,ifwe<br />

takealookattherivalproof-theoreticapproachchampionedbyDummett,<br />

wewouldseethesamepattern.Theinsistenceon‘conservativity’indealingwithintroductionandeliminationrulesforlogicalconstantscouldbe<br />

motivatedonlybytheideaofthepurelytautologouscharacteroflogically<br />

validinferences.Thelocalpeaksshouldbeinprinciple‘levelable’,precisely<br />

becausethemanipulationwithlogicalvocabularyaddsnosubstantivenew<br />

informationabouttheworld—inshort,becauselogicisanalyticandhas<br />

nothingtodowithsentences,truebyvirtueoffactsoftheworld.<br />

6<br />

Whenwedophilosophyoflogic,weareoftensaidcompletelydifferent<br />

things,incompatiblewiththeideathatlogicaltruth(conceivedasaparadigmaticcaseofanalyticity)istobedemarcatedfromfactualtruth.Starting<br />

withWittgensteinagain,weseethatallhislaterdevelopment—from<br />

“SomeRemarksonLogicalForm”whereheadmitsthat“wecanonlyarriveatacorrectanalysisbywhatmightbecalled,thelogicalinvestigationofthephenomenathemselves”(Wittgenstein,1993b,p.30)to“OnCertainty”wherehedeniedthepossibilitytodistinguishfromtheoutsetlogical<br />

fromempiricalpropositionsbecause“theriver-bedofthoughtsmayshift”<br />

(Wittgenstein,1993a,p.15)—canbeseenasrejectionoftheprevious<br />

sharpdivisionofalllocutionsintovacuouslytrue‘tautologies’andmeaningful‘picturesofreality’.Tarskihimself,asearlyas1930,wascommitted<br />

tothesamelineofthought:inanoteofCarnap’sdiary,datedFebruary22,<br />

1930weread:“8–11withTarskiataCafe.Aboutmonomorphism,tautology,hewillnotgrantthatitsaysnothingabouttheworld;heclaimsthatbetweentautologicalandempiricalstatementsthereisonlyameregradualandsubjectivedistinction”(Mancosu,2005,pp.328–329).Severalyearslater,in“Ontheconceptoffollowinglogically”,weread:“Atthefoundationofourwholeconstructionliesthedivisionofalltermsofalanguageinto<br />

logicalandextra-logical.Iknownoobjectivereasonswhichwouldallowone<br />

todrawaprecisedividinglinebetweenthetwocategoriesofterms... the<br />

divisionoftermsintologicalandextra-logicalexertsanessentialinfluence<br />

onthedefinitionalsoofsuchtermsas‘analytic’and‘contradictory’;yetthe<br />

conceptofananalyticsentence... tomepersonallyseemsrathermurky


126 RosenLutskanov<br />

(Tarski,2002,pp.188–189).Stilllater,in1944Tarskiconfessedinaletter<br />

toMortonWhitethatheisinclinedtothinkthat“logicalandmathematical<br />

truthsdon’tdifferintheiroriginfromempiricaltruths—bothareresults<br />

ofaccumulatedexperience... [andwehavetobepreparedto]rejectcertain<br />

logicalpremises(axioms)ofourscienceinexactlythesamecircumstances<br />

inwhichIamreadytorejectempiricalpremises(e.g.,physicalhypotheses)”<br />

(White,1987,p.31).Itwouldnotbestrange,ifthesewordssoundfamiliar:<br />

thesamecritiqueswereformulatedbyQuine,whometCarnapinPrague<br />

in1933andforcedhimtoadmittheuntenabilityoftheanalytic/synthetic<br />

distinction: “Isthereadifferenceinprinciplebetweenlogicalaxiomsand<br />

empiricalsentences? He[Quine]thinksnot. PerhapsI[Carnap]seeka<br />

distinctionjustforitsutility,butitseemsheisright: gradualdifference:<br />

theyaresentenceswewanttoholdfast”(Quine,2004,p.55).In“Truthby<br />

Convention”(Quine,1936)stressedthatsomeanalyticallytruestatements<br />

—definitionalconventions—canbeoverthrownforempiricalreasons,and<br />

in“Twodogmasofempiricism”(1951)introducedthefieldmetaphorthat<br />

obliteratescompletelytheanalytic/syntheticdistinction,makingevident<br />

thatitis“follytoseekaboundarybetweensyntheticstatements,which<br />

holdcontingentlyonexperience,andanalyticstatements,whichholdcome<br />

whatmay”(Quine,1961,p.50).Generally,thedestructionofthisdistinctionwaseffectedinHarvard:<br />

fromthe1940disputesofCarnap,Tarski<br />

andQuine,toWhite’searly“Theanalyticandthesynthetic:anuntenable<br />

dualism”(1950),Quine’sground-braking“Twodogmasofempiricism”and<br />

Goodman’sreflectiveequilibriumtheorydevelopedin“Thenewriddleof<br />

induction”(1954).<br />

7<br />

Thedefinitionsoflogicalconstantswehavediscussedwereshowntobemotivatedbytheuntenableassumptionthatwearecapableofdiscriminating<br />

rigorouslybetweenanalytic(truebyvirtueoflinguisticconventions)and<br />

synthetic(truebyvirtueofmattersoffact)propositions. Itseemstome<br />

thatitisjustifiedtosearchforadefinitionoflogicalconstantsthatconforms<br />

tothemainstreamphilosophyoflogic,afortioriadefinitionwhichdoesnot<br />

presupposetheanalytic/syntheticdistinction. Needlesstosay,everything<br />

Icansuggestonthistopicuptothepresentmomentissketchyandinconclusive.Firstofall,Iadmitthatlogicisconcernedwiththecodificationof<br />

inferentialpracticeswhicharegenerally‘outthere’beforewetrytoimpose<br />

normativerestrictionsonthem. ThesepracticesproducewhatBrandom<br />

calls‘materialinferences’—inferencesthatarenotjustifiedwithrecourse<br />

tothefeaturesoflogicalvocabularybutseemasimmediatelyacceptable.<br />

Anychainofmaterialinferencescanbecalledan‘argument’—thissug-


WhatistheDefinitionof‘LogicalConstant’? 127<br />

geststhatingeneralthematerialinferencesareseriallyorderedandaimat<br />

something—theclaimthatneedstobeestablishedastrueorfalse. Any<br />

argumentcanbemodelednaturallyinaslightmodificationoftheframeworkdevelopedinGuptaandBelnap’s“RevisionTheoryofTruth”(Gupta<br />

&Belnap,1993).Letusconsideraformallanguage Landamodel M0that<br />

assignstosomesentencesin Lthevalue‘true’:thesearethe‘axioms’(in<br />

theirancientinterpretationas‘sentencesproposedforconsideration’)that<br />

wetemporarilyacceptastrue.Thenthesetofpossiblematerialinferences<br />

withpremisestruein M0definesajump-operatorcorrelatingwithitanothermodelof<br />

L(letusdesignateitas‘M1’)containingallthosesentences<br />

thathavetobeacceptedastrueonthebasisofthebootstrapmodel M0<br />

(ingeneral,wedonotsupposethatthejumpoperatorismonotone:some<br />

previouslyacceptedsentencescanberefutedatlaterstages). Thesame<br />

procedurecanbeappliedagainandagainwhichgivesrisetoindefinitely<br />

extendibleseriesofmodels M0, M1, M2,etc. whichweshallcall‘anargument’(whosepremisesaretheaxioms,definedby<br />

M0). Inthecourse<br />

ofanytypicalargument Athereshallbesentencesthatatsomestageof<br />

itsdevelopment(say Mn)receiveconstantinterpretation(thesearethefixpointsofthejumpoperator);forthosethatareevaluatedastrueinall<br />

successivestages(Mn+1, Mn+2,etc.) weshallsaythattheyare‘rendered<br />

stablytrue’(bytheargument A). Now,insteadofasingleargument,let<br />

usconsiderabunchofarguments A1, A2,etc.andsupposethatthereisa<br />

classofsentences,classifiedasstablytruebyanyoneofthem;Ipropose<br />

thesesentencestobecalled‘renderedvalid’(bythesetofarguments A1,<br />

A2,etc.).Mysuggestionistoequatelogicalitywiththejustdefinedconcept<br />

of‘validity’;inthiswayweremainfairtosomeoftraditionallyrecognized<br />

distinctivefeaturesoflogicaltruth:<br />

(i)itistopic-neutral(becauseitisnotrelativetoaparticularargumentativesetting);<br />

(ii)itisnecessary(becauseitinevitablyshowsupinanytrainofreasoning<br />

belongingtothegeneralargumentativesetting);<br />

(iii)itisanalytic(becausegivenavalidsentenceandasetofargumentative<br />

premiseswecandemonstratebymeansofanalysisoftheaccepted<br />

materialinferencesthatitisgenuinepropositionoflogic).<br />

Moreover,thisdivisionofsentencesintoanalytic(renderedvalid)andsynthetic(notrenderedvalid)cannotbedrawnfromtheoutsetbecauseingeneralthequestion“isthesentencesrenderedvalidbythesetofarguments<br />

A1, A2,etc.?”isnotdecidable.<br />

Afterwehavesecuredaworkablenotionoflogicality,wecanaskourselvesagain:whatisalogicalconstant?<br />

Theansweristhatalexicalunit


128 RosenLutskanov<br />

istobetreatedaspieceoflogicalvocabularywhenitisinvariablyinterpretablecomponentofsomesetofvalidsentences.Thenwecanhuntdown<br />

thelogicalconstantsusingthe‘inverse’logicalapproachdevelopedbyvan<br />

Benthemwhosuggestedthatinsteadofchoosingsomepredefinedsetoflogicalconstantsandaskingwhattypesofinferencesarevalidatedbythem,wecantakesomeintuitivelyconvincingsetof(material)inferencesthatvalidateparticularpropositionsandsearchforthespecificconstantsthatare<br />

accountableforthem.Thismethodologicalshiftfrompredefinednormative<br />

accountsoflogicalitytopurelydescriptiveexplorationsofinferentialpracticeswasnamed“Copernicanrevolutioninlogic”(Benthem,1984,p.451).<br />

WhatI’vetriedtodohere,wastoshowthattherevolutionmustgoon...<br />

RosenLutskanov<br />

InstituteforPhilosophicalResearch,BulgarianAcademyofSciences<br />

6PatriarchEvtimiiBlvd.,Sofia1000,Bulgaria<br />

rosen.lutskanov@gmail.com<br />

http://www.philosophybulgaria.org/en/Sekcii/Logika/Sastav.php<br />

References<br />

Benthem,J.v.(1984).Questionsaboutquantifiers.JournalofSymbolicLogic,<br />

49(2),443–466.<br />

Carnap,R.(1958).Introductiontosymboliclogicanditsapplications.NewYork:<br />

DoverPublications.<br />

Carnap,R. (1959). Thelogicalsyntaxoflanguage. Paterson,NJ:Littlefield,<br />

AdamsandCo.<br />

Dummett,M.(1991).Thelogicalbasisofmetaphysics.Cambridge,MA:Harvard<br />

UniversityPress.<br />

Feferman, S. (1997). Logic, logics, and logicism. (Retrieved from:<br />

http://math.stanford.edu/∼feferman/papers.)<br />

Frege,G.(1960).Begriffsschrift.InP.Geach&M.Black(Eds.),Translationsfrom<br />

thephilosophicalwritingsofGottlobFrege(pp.1–20).Oxford:BasilBlackwell.<br />

Gomez-Torrente,M.(2002).Theproblemoflogicalconstants.BulletinofSymbolicLogic,8(1),1–37.<br />

Gupta,A.,&Belnap,N.(1993).Therevisiontheoryoftruth.Cambridge,MA:<br />

MITPress.<br />

Hacking,I.(1994).Whatislogic?InD.Gabbay(Ed.),Whatisalogicalsystem?<br />

(pp.1–34).Oxford:ClarendonPress.<br />

Hodes,H.(2004).Onthesenseandreferenceofalogicalconstant.ThePhilosophicalQuarterly,54(214),134–165.


WhatistheDefinitionof‘LogicalConstant’? 129<br />

Hodges,W.(2006).Thescopeandlimitsoflogic.InD.Jacquette(Ed.),Handbook<br />

ofthephilosophyofscience.Philosophyoflogic(pp.41–64).Dordrecht:North–<br />

Holland.<br />

Mancosu,P.(2005).Harvard1940–1941:Tarski,CarnapandQuineonafinitistic<br />

languageofmathematicsforscience.HistoryandPhilosophyofLogic,26,327–<br />

357.<br />

McCarthy,T.(1981).Theideaofalogicalconstant.JournalofPhilosophy,78,<br />

499–523.<br />

Mostowski,A. (1957). Onageneralizationofquantifiers. FundamentaMathematicae,44,12–36.<br />

Quine,W.(1936).Truthbyconvention.InO.Lee(Ed.),Philosophicalessaysfor<br />

A.N.Whitehead(pp.90–124).NewYork:Longmans.<br />

Quine,W. (1961). Twodogmasofempiricism. InFromalogicalpointofview<br />

(pp.39–52).Cambridge,MA:HarvardUniversityPress.<br />

Quine,W.(2004).Twodogmasinretrospect.InR.Gibson(Ed.),Quintessence:<br />

ReadingsfromthephilosophyofW.V.Quine(p.54-63).Cambridge,MA:Harvard<br />

UniversityPress.<br />

Quine,W. (2006). Philosophyoflogic(seconded.). Cambridge,MA:Harvard<br />

UniversityPress.<br />

Russell,B.(1903).Theprinciplesofmathematics(Vol.I).Cambridge:Cambridge<br />

UniversityPress.<br />

Siebel,M. (2002). Bolzano’sconceptofconsequence. TheMonist,85(4),580–<br />

599.<br />

Tarski,A.(2002).Ontheconceptoffollowinglogically.HistoryandPhilosophy<br />

ofLogic,23,155–196.<br />

White,M. (1987). AphilosophicalletterofAlfredTarski. TheJournalof<br />

Philosophy,84(1),28–32.<br />

Wittgenstein,L.(1963).Tractatuslogico–philosophicus.London:Routledgeand<br />

KeganPaul.<br />

Wittgenstein,L.(1993a).Oncertainty(G.Anscombe&G.Wright,Eds.).Oxford:<br />

BasilBlackwell.<br />

Wittgenstein, L. (1993b). Someremarksonlogicalform. InJ.Klagge&<br />

A.Nordmann(Eds.),Philosophicaloccasions,1912–1951(pp.29–36).Indianapolis:HackettPublishingCompany.


1 Introduction<br />

Epistemic Logic with Relevant Agents<br />

Ondrej Majer Michal Peliˇs ∗<br />

Theaimofepistemiclogicsistoformalizeepistemicstatesandactionsof<br />

(possiblyhuman)rationalagents.Atraditionalmeansforrepresentingthese<br />

statesandactionsemploystheframeworkofmodallogics,whereknowledge<br />

correspondstosomenecessityoperator.Modalaxioms(K,T,4,5,...)then<br />

correspondtostructuralpropertiesoftheagent’sknowledge. Employing<br />

strongmodalsystemssuchas S5leadstorepresentationsofagentswho<br />

aretooidealinmanyrespects—theyarelogicallyomniscient,theyhave<br />

aperfectreflectionoftheirbothpositiveandnegativeknowledge(positive<br />

andnegativeintrospection)etc.Sometimestheserepresentationsarecalled<br />

epistemiclogicsofpotentialratherthanactualknowledge.<br />

Frameworksrepresentingonlyperfectagentshavebeenfrequentlycriticized,see(Fagin,Halpern,Moses,&Vardi,2003)and(Duc,2001),and<br />

somestepstowardsmorerealisticrepresentationshavebeenmade(e.g.,<br />

(Duc,2001)).Wealsoattempttorepresentagentsinanenvironmentmore<br />

realistically.Ourmotivationisepistemic,weshallconcentrateonanagent<br />

workingwithexperimentalscientificdata.<br />

Arealisticagent<br />

Ouragentisascientistundertakingexperimentsorobservations.Hertypicalenvironmentisanexperimentalsetupandherknowledgeisusually<br />

experimentaldata(inputsandoutputsofanexperiment/observation)and<br />

somegeneralizationsextractedfromtheexperimentaldata.<br />

∗ Workonthistextwassupportedinpartbygrantno.401/07/0904oftheGrantAgency<br />

oftheCzechRepublicandinpartbygrantno.IAA900090703(Dynamicformalsystems)<br />

oftheGrantAgencyoftheAcademyofSciencesoftheCzechRepublic.Wewishtothank<br />

toTimothyChildersforvaluablecomments.


132 OndrejMajer&MichalPeliˇs<br />

Weassumetheobservations(‘facts’)aretypicallyrepresentedbyatoms<br />

andtheirconjunctionsanddisjunctions,whilegeneralizations(‘regularities’)<br />

arerepresentedbyconditionals(andtheircombinations). Aconditional<br />

issupposedtorecordaregularlyobservedconnectionbetweenthefacts<br />

representedbytheantecedentandthefactsrepresentedbytheconsequent.<br />

Itseemstobeclearthatformanyreasonsthematerialimplicationisnot<br />

anappropriaterepresentationofsuchaconditional.Oneofthemainreasons<br />

isthatthematerialimplicationmayconnectanytwoarbitraryformulas α,<br />

β.Forexample,<br />

1. α → (β → α),<br />

2.(α ∧ ¬α) → β,<br />

3. α → (β ∨ ¬β),<br />

aretautologiesofclassicallogic.Inourepistemicinterpretationthematerial<br />

implicationwouldmakea‘law’fromeverytwo‘facts’,whichwouldobviously<br />

maketherepresentationuseless. Ithasotherundesirableproperties. It<br />

cannotdealwitherrorsinthedata,whichresulttocontradictoryfacts<br />

(asituationwhichmayverywellhappeninthescientificpracticedueto<br />

equipmenterrors).Onesucherrorcorruptsalltheremainingdata(froma<br />

contradictioneverythingfollows—see2). Italsoadmits’laws’whichare<br />

ofnouseastheirconsequentisatautology(asin3—atautologyfollows<br />

fromanything)<br />

Thetautologies1–3arejustexamplesoftheparadoxesofmaterialimplication.Asthese‘paradoxes’werecompletelysolvedonlyinthesystemsof<br />

relevantlogics,theobviouschoiceforaconditionalforourscientificagent<br />

isrelevantimplication.<br />

2 Relationalsemanticsforrelevantlogics<br />

Ourpointofdeparturewillbethedistributiverelevantlogic RofAnderson<br />

andBelnap(1975). Themostnaturalwaytointroducerelevantlogicsis<br />

certainlyprooftheoretical(see,e.g.,(Paoli,2002)).Howeverwewouldlike<br />

tofollowthemodaltraditioninrepresentinganagent’sepistemicstatesas<br />

asetofformulasandmaketheagent’sknowledgedependentnotonlyon<br />

thecurrentepistemicstate,butalsoonthestatesepistemicalternatives.<br />

Technicallyspeakingwewanttousearelationalsemantics.Thiscannotbe<br />

astandardKripkesemanticswithpossibleworldsandabinaryaccessibility<br />

relation,butamoregeneralrelationalstructure.<br />

FormallyourframeworkwillbebasedontheRoutley–Meyersemantics,<br />

asdevelopedbyMares(Mares,2004),Restall(Restall,1999),Paoli(Paoli,


EpistemicLogicwithRelevantAgents 133<br />

2002),andothers,towhichweshalladdepistemicmodalities.Thissemanticshasbeenunderconstantattackforitsseemingunintuitivness,butwe<br />

believeitfitsverywellourmotivations.<br />

Wegiveaninformalexpositionofstructuresintherelevantframeand<br />

definitionofconnectives(forformaldefinitionsseetheappendixA).<br />

Relevantframe<br />

Arelevantframeisastructure F = 〈S,L,C,✂,R〉,where Sisanon-empty<br />

setofsituations(states), L ⊆ Sisanon-emptysetofdesignatedlogical<br />

situations, C ⊆ S 2 isacompatibility relation, ✂ ⊆ S 2 isarelationof<br />

involvement, R ⊆ S 3 isanrelevancerelation.<br />

Amodel Misarelevantframewiththerelation �,where s � ϕhasthe<br />

samemeaningasinKripkeframes—that scarriestheinformationthat<br />

theformula ϕistrue(ϕ ∈ sifweconsiderstatestobesetsofformulas).<br />

Situations Situationsorinformationstatesplaythesameroleaspossible<br />

worldsinKripkeframes. Weassume,theyconsistofdataimmediately<br />

availabletotheagent. Likepossibleworlds,wecanseesituationsassets<br />

offormulas,but,unlikepossibleworlds,situationsmightbeincomplete<br />

(neither ϕnor ¬ϕistruein s)orinconsistent(both ϕand ¬ϕaretrue<br />

in s).<br />

Conjunctionanddisjunction Classical(weak)conjunctionanddisjunction<br />

correspondtothesituationwhentheagentcombinesdataimmediately<br />

availabletoher,i.e.datafromhercurrentsituation. Theybehaveinthe<br />

samewayasinthecaseofclassicalKripkeframes—theirvalidityisgiven<br />

locally:<br />

s � ψ ∧ ϕiff s � ψand s � ϕ<br />

s � ψ ∨ ϕiff s � ψor s � ϕ<br />

Weakconnectivesaretheonlyoneswhicharedefinedlocally.Thetruth<br />

ofnegationandimplicationdependsalsoonthedatainsituations,related<br />

totheactualones,sotheyaremodalbynature. Itispossibletodefine<br />

strongconjunctionanddisjunctionaswell(seeappendixA).<br />

Implication Implicationisamodalconnectiveinthesensethatitstruth<br />

dependsnotonlyonthecurrentsituation,butalsoonitsneighborhood.It<br />

canbeagainunderstoodinanalogywiththestandardmodalreading.We<br />

saythatanimplication (ϕ → ψ)holdsnecessarilyinaKripkeframeiff<br />

inallworldswheretheantecedentholds,theconsequentholdsaswell. In<br />

otherwords,theimplication (ϕ → ψ)holdsthroughalltheneighborhood


134 OndrejMajer&MichalPeliˇs<br />

oftheactualworld. Intherelevantcasetheneighborhoodofasituation<br />

sisgivenbypairsofsituations y, zsuchthat s, y, zarerelatedbythe<br />

ternaryrelation R.Weshallcall y, zantecedentandconsequentsituations,<br />

respectively.Wesaythattheimplication (ϕ → ψ)holdsatthesituation s<br />

iffitisthecasethatforeveryantecedentsituation ywhere ϕ(theantecedent<br />

oftheimplication)holds, ψ(theconsequentoftheimplication)holdsatthe<br />

correspondingconsequentsituation z.<br />

s � (ϕ → ψ) iff (∀y,z)(Rsyzimplies (y � ϕimplies z � ψ))<br />

Therelation Rreflectsinourinterpretationactualexperimentalsetups.<br />

Antecedentsituationscorrespondtosomeinitialdata(outcomeofmeasurementsorobservations)ofsomeexperiment,whiletherelatedconsequent<br />

situationscorrespondtothecorrespondingresultingdataoftheexperiment.<br />

Implicationthencorrespondstosome(simple)kindofarule:ifIobserve<br />

inmycurrentsituation,thatateveryexperiment(representedbyacouple<br />

antecedent—consequentsituation)eachobservationof ϕisfollowedbyan<br />

observationof ψ,thenIaccept‘ψfollows ϕ’asarule.<br />

Logicalsituations Theframeworkwepresentedsofarisveryweak:there<br />

arejustfewtautologiesvalidinallsituationsandsomeoftheimportant<br />

ones—thosebeingusuallyconsideredasbasiclogicallawsaremissing.<br />

Forexamplethewidelyacceptedidentityaxiom (α → α)andtheModus<br />

Ponensrulefailtoholdineverysituation.<br />

Thisisconnectedtothequestionoftruthinarelevanceframe(model).<br />

IfwetakeahintfromKripkeframes,weshouldequatetruthinaframe<br />

withtruthineverysituation. Butthiswouldgivesusanextremelyweak<br />

systemwithsomeveryunpleasantproperties(cf.(Restall,1999)).Designers<br />

ofrelevantlogicstookadifferentroute—insteadofrequiringtruthinall<br />

situations,theyidentifythetruthinaframejustwiththetruthinall<br />

logicallywellbehavedsituations. Thesesituationsarecalledlogical. In<br />

ordertosatisfythe‘goodbehavior’ofasituation litisenoughtorequire<br />

thatalltheinformationinanyantecedentsituationrelatedto liscontained<br />

inthecorrespondingconsequentsituationaswell:foreach x,y ∈ S, Rlxy<br />

implies |x| ⊆ |y|,where |s|isthesetofallformulas,whicharetrueinthe<br />

situation s.<br />

Itiseasytoseethatsituationsconstrainedinthiswayvalidateboththe<br />

identityaxiomand(implicative)ModusPonens.<br />

Involvement Involvementisarelationresemblingthepersistencerelation<br />

inintuitionisticlogic—wecanseeitasarelationofinformationgrowth.<br />

Howevernoteverytwosituationswhichareininclusionwithrespecttothe<br />

validatedformulasareintheinvolvementrelation.Werequirethatsuchan


EpistemicLogicwithRelevantAgents 135<br />

inclusionisobservedorwitnessed.Noteverysituationcanplaytheroleof<br />

thewitness—onlythelogicalsituationscan.<br />

x ✂ y iff (∃l ∈ L)(Rlxy)<br />

Negation InKripkemodelsthenegationofaformula ϕistrueataworld<br />

iff ϕisnottruethere.Assituationscanbeincompleteand/orinconsistent,<br />

thisisnotanoptionanymore. Negationbecomesamodalconnective<br />

anditsmeaningdependsontheworldsrelatedtothegivenworldbya<br />

binarymodalrelation Cknownascompatibility. Informallywecansee<br />

thecompatiblesituationsasinformationsourcesourscientistwantstobe<br />

consistentwith. (Imaginethedataofresearchgroupsworkingonrelated<br />

subjects.)<br />

Theformula ¬ϕholdsat s ∈ Siffitisnot‘possible’(inthestandard<br />

modalsensewithrespecttotherelation C)that ϕ: atnosituation s ′ ,<br />

compatiblewith(‘accessiblefrom’)thesituation s,itisthecasethat ϕ<br />

(either s ′ isincompletewithrespectto ϕor ¬ϕholdsthere).<br />

s � ¬ϕ iff (∀s ′ ∈ S)(sCs ′ implies s ′ �� ϕ)<br />

Informallyspeaking,theagentcanexplicitlydenysomehypothesis(a<br />

pieceofdata)onlyifnoresearchgroupinherneighborhoodclaimsitis<br />

true. Thisconditionalsohasanormativeside:shehastobeskepticalin<br />

thesensethatshedenieseverythingnotpositivelysupportedbyanyofher<br />

colleagues(inthesituationsrelatedtoheractualsituation).<br />

Ifwewanttograntnegativefactsthesamebasiclevelaspositivefacts,we<br />

canreadtheclauseforthedefinitionofcompatibilityintheotherdirection:<br />

theagentcanrelateheractualsituationjusttothesituationswhichdonot<br />

contradicthernegativefacts.<br />

Dependingonthepropertiesofthecompatibilityrelationweobtaindifferentkindsofnegations.Weshallshortlycommentonthem.Thecompatibilityrelationisingeneralnotreflexive:inconsistentsituationsarenotself-compatibleandsoreflexivityholdsonlyforconsistent<br />

situations.Itisclearthatforaninconsistentself-compatiblesituationthe<br />

clausefornegationwouldnotwork.Ontheotherhand,inconsistentsituationscanbecompatiblewithsomeincompletesituations.<br />

Noris Ctransitive. Letushavesituations x, y, zsuchthat x � ϕ,<br />

z � ¬ϕ,and ydoesnotincludeeither ϕor ¬ϕ.Assumethat xCyand yCz.<br />

Thenaccordingtothedefinitionofnegationitcannotbethat xCz.<br />

Itisquitereasonabletoassumethat Cissymmetric. Thiscondition<br />

impliesthatwegetonlyonenegation(otherwisewewouldgetleftandright<br />

negation)andwegetthe’unproblematic’halfofthelawofdoublenegation<br />

(if x � ϕ,then x � ¬¬ϕ).


136 OndrejMajer&MichalPeliˇs<br />

Wealsoassume Cisdirectedandconvergent.Directednessmeansthat<br />

thereisatleastonecompatiblesituationforeach x ∈ S.Convergencesays<br />

thatthereisamaximalcompatiblesituation x ⋆ .(SeeappendixA.)<br />

Maximalcompatiblesituations(withrespectto x)canbeinconsistent<br />

abouteverythingnotconsideredin x.Fromthesymmetryof Cweobtain<br />

x✂x ⋆⋆ .Ifweassume,moreover, x☎x ⋆⋆ ,thenwegettheoperation ⋆withthe<br />

property x = x ⋆⋆ ,i.e.theRoutleystar.Thedefinitionofnegation-validity<br />

isthenwrittenintheform:<br />

x � ¬ϕ iff x ⋆ �� ϕ<br />

TheRoutleystarhasbeenoneofthecontroversialpointsoftheRoutley–<br />

Meyersemantics,butinourmotivationithasaquitenaturalexplanation:if<br />

compatiblesituationsrepresentcolleaguesfromdifferentresearchgroupsour<br />

agentcollaborateswith,thenthemaximalcompatiblesituationcorrespond<br />

toacolleague(‘boss’)whohasalltheinformationtheothercolleaguesfrom<br />

thegrouphave.Theniftheagentwantstoacceptsomenegativeclauseshe<br />

doesnothavetospeaktoeachofthecolleaguesandaskhis/heropinion,<br />

shejustasksthe‘boss’directlyandknowsthatbossesopinionrepresents<br />

theopinionsoftheentirecompatibleresearchgroup.<br />

Thiscompletesourexpositionofrelationalsemanticsforrelevantlogics.<br />

Wenowmovetoepistemicmodalities.<br />

3 Knowledgeinrelevantframework<br />

Therehavebeensomeattemptstocombineanepistemicandrelevantframework(see(Cheng,2000)and(Wansing,2002)),buttheyhaveadifferent<br />

aimthenourapproach.<br />

Fromapurelytechnicalpointofviewthereareanumberofwaysto<br />

introducemodalitiesintherelevantframework—GregRestallin(Restall,<br />

2000)providesanicegeneraloverview. Aswementioned,therelevant<br />

frameworkalreadycontainsmodalnotions. Wethereforedecidedtouse<br />

thesenotionstointroduceepistemicmodalitiesratherthantointroduce<br />

newones.<br />

Intheclassicalepistemicframewhatanagentknowsinaworld wis<br />

definedaswhatistrueinallepistemicalternativesof w,whicharegivenby<br />

thecorrespondingaccessibilityrelation.Ourideaoftheagentasascientist<br />

processingsomekindofdatarequiresadifferentapproach.<br />

Weassumeouragentinhercurrentsituation sobserves(hasadirect<br />

approachto)somedata,representedbyformulaswhicharetrueat s.Sheis<br />

awareofthefactthatthesedatamightbeunreliable(oreveninconsistent).<br />

Inordertoacceptsomeofthecurrentdataasknowledgetheagentrequires<br />

aconfirmationfromsome‘independent’resources.


EpistemicLogicwithRelevantAgents 137<br />

Inourapproachresourcesaresituationsdealingwiththesamekindof<br />

dataavailableinthecurrentsituation.Aresourceshallbemoreelementary<br />

thanthecurrentsituation,i.e.,itshouldnotcontainmoredata(aresource<br />

isbelow sinthe ✂-relation). Alsothedatafromtheresourceshouldnot<br />

contradictthedatainthecurrentsituation(aresourceiscompatiblewith s).<br />

Definition4(Knowledge).<br />

s � Kϕ iff (∃x)(sC ✁ xand x � ϕ),<br />

where sC ✁ xiff sCxand x ✂ sand x �= s.<br />

Inshort, ϕisknowniffthereisanresource(‘lower’compatiblesituation<br />

differentfromtheactualone)validating ϕ.<br />

Weallowedouragenttodealwithinconsistentdatainordertogetamore<br />

realisticpicture.However,theagentshouldbeabletoseparateinconsistent<br />

data. Themodalityweintroducedprovidesuswithsuchanappropriate<br />

filter. Letusassumeboth ϕand ¬ϕarein s(e.g.,ouragentmightreceivesuchinconsistentinformationfromtwodifferentsources).Theagent<br />

considersboth ϕand ¬ϕtobepossible,butneitherofthemisconfirmed<br />

informationasaccordingtothedefinition,nosituationcompatibleto scan<br />

containeither ϕor ¬ϕ.<br />

Basicproperties<br />

Itistobeexpectedthatoursystemblocksalltheundesirablepropertiesof<br />

bothmaterialandstrictimplication. Moreover,weruledoutthevalidity<br />

ofsomeofthepropertiesof‘classical’epistemiclogicsthatwehavecriticized,inparticular,bothpositiveandnegativeintrospection,aswellas<br />

someclosureproperties.<br />

Letushavearelevantframe F = 〈S,L,C,✂,R〉.Recallthatthetruthin<br />

theframe Fcorrespondstothetruthinthelogicalsituationsof F(underany<br />

valuation). Wewillalsousethestrongernotionoftruthinallsituations<br />

of F(underanyvaluation). Fromtheviewpointofourmotivationthe<br />

latternotionismoreinterestingasouragentmighthappentobeinother<br />

situationsthanthelogicalones.<br />

Ourapproachmakesthe‘truthaxiom’Tvalid.Foranysituation s ∈ S,<br />

if ϕisknownat s(s � Kϕ),thenthereisa✂-lowercompatiblewitness<br />

with ϕtrue,whichmakes ϕtobetrueat saswell.Thus,formula<br />

Kα → α<br />

isvalid.<br />

TheaxiomKandthenecessityrule,commontoallnormalepistemic<br />

logics,fail.First,letusassumethat ϕisvalidformula.Thenecessityrule


138 OndrejMajer&MichalPeliˇs<br />

( ϕ<br />

)wouldimplythevalidityof Kϕ. |= ϕmeansthat ϕistrueinevery<br />

Kϕ<br />

logicalsituation l. However,for l � Kϕaconfirmationfromadifferent<br />

resourceisrequired,theremustbeasituation xsuchthat x � ϕand lC✁x, which,ingeneral,doesnotneedtobethecase.<br />

Second,inourinterpretationthevalidityofaxiomKisnotwellmotivatedanddoesnothold.<br />

Kisinfacta‘distributionofconfirmation’: If<br />

animplicationisconfirmedthentheconfirmationoftheantecedentimplies<br />

theconfirmationoftheconsequent.<br />

�|= K(α → β) → (Kα → Kβ)<br />

Introspection Aswedefinedknowledgeasindependentlyconfirmeddata,<br />

theepistemicaxioms4and5correspondinourframeworktoa‘second<br />

orderconfirmation’ratherthantointrospection.Itiseasytoseethatboth<br />

axiomsfail.<br />

�|=Kα → KKα,<br />

�|=¬Kα → K¬Kα<br />

Necessityandpossibility Wedonotintroducepossibilityusingthestandarddefinition<br />

Mϕ def<br />

≡ ¬K¬ϕ.Ourideaofepistemicpossibilityisthatour<br />

agentconsidersallthedataavailableatthecurrentsituationaspossible.If<br />

weintroduceformally s � Mϕas s � ϕ,thenitfollowsfromtheTaxiom<br />

thatinallsituationsnecessityimpliespossibility:<br />

(∀s ∈ S)(s � Kϕ → Mϕ)<br />

Howeverforthestandarddualpossibilitythisisnottrue.<br />

�|= Kϕ → ¬K¬ϕ<br />

Letuscommentontherelationofnegationandnecessityinourframework.<br />

Thereisadifferencebetween s �� Kϕand s � ¬Kϕ. Theformer<br />

simplysaysthat ϕisnotconfirmedatthecurrentsituation s,whilethelattersaysthat<br />

ϕisnotconfirmedinthesituationscompatiblewith s.From<br />

thispointofviewitisuncontroversialthatboth Kϕ(confirmationinthe<br />

currentsituation)and ¬Kϕ(thelackofconfirmationinthecompatiblesituations)mightbetrueinsomesituation<br />

s(thenecessaryconditionisthat<br />

sisnotcompatiblewithitself).<br />

Closureproperties ItiseasytoseethatthemodalModusPonens<br />

Kα K(α → β)<br />


EpistemicLogicwithRelevantAgents 139<br />

doesnothold(forthereasonsgiveninthesectiononKaxiom).However,<br />

itsweakerversion<br />

Kα K(α → β)<br />

β<br />

holdsnotonlyinlogicalsituations,butinallsituations.If Kαand K(α →<br />

β)aretrueinany s ∈ S,then s � β.AxiomTandtheassumption Rsss<br />

arecrucialhere.<br />

Contradictioninoursystemisnon-explosive: ϕand ¬ϕmightholdina<br />

contradictorysituation,whichneednotbeconnectedtoanysituationwhere<br />

ψholds.<br />

�|= (ϕ ∧ ¬ϕ) → ψ<br />

Ontheotherhand,theknowledgeofcontradictionimpliesanything(asa<br />

contradictionisneverconfirmed):<br />

|= K(ϕ ∧ ¬ϕ) → ψ<br />

Modaladjunctionalsodoesnothold—if Kαand Kβaretruein s,then<br />

obviously (α∧β)istruetherebecauseofthetruthaxiombut K(α∧β)does<br />

notneedtobetruein s.(Ifeachof αand βisconfirmedbysomeresource,<br />

therestillmightbenoresourceconfirmingtheirconjunction.)<br />

4 Conclusion<br />

Weintroducedasystemofepistemiclogicbasedontheframeworkofrelevantlogic.Wegaveanepistemicinterpretationoftherelationalsemanticsforrelevantlogicsanddefinedepistemicmodalitiesmotivatedbythisinterpretation.<br />

Insteadofintroducingadditionalrelationsintotheframework,<br />

wearguedinfavorofusingmodalitiesbasedontherelationsalreadycontainedintheframe.<br />

Thewholeprojectisataninitialstage:thereismuchtobedoneboth<br />

technicallyandintheareaofinterpretation.Inparticularweshalldevelop<br />

inamoredetailtheepistemicinterpretationofourframework,givean<br />

axiomatizationofoursystem,andcharacterizeitsformalproperties.<br />

A Relevantlogic R<br />

Therearemoreformalsystemsthatcanbecalledrelevantlogic.Fromthe<br />

proof-theoreticalviewpoint,allofthemareconsideredtobesubstructural<br />

logics(see(Restall,2000)and(Paoli,2002)). Herewepresenttheaxiom<br />

systemand(Routley–Meyer)semanticsfrom(Mares,2004)withsomeelementsfrom(Restall,1999).


140 OndrejMajer&MichalPeliˇs<br />

Syntax<br />

Weusethelanguageofclassicalpropositionallogicwithsignsforatomic<br />

formulas P = {p,q,... },formulasbeingdefinedintheusualway:<br />

Axiomschemes<br />

1. A → A<br />

ϕ ::= p | ¬ψ | ψ1 ∨ ψ2 | ψ1 ∧ ψ2 | ψ1 → ψ2<br />

2. (A → B) → ((B → C) → (A → C))<br />

3. A → ((A → B) → B)<br />

4. (A → (A → B)) → (A → B)<br />

5. (A ∧ B) → A<br />

6. (A ∧ B) → B<br />

7. A → (A ∨ B)<br />

8. B → (A ∨ B)<br />

9. ((A → B) ∧ (A → C)) → (A → (B ∧ C))<br />

10. (A ∧ (B ∨ C)) → ((A ∧ B) ∨ (A ∧ C))<br />

11. ¬¬A → A<br />

12. (A → ¬B) → (B → ¬A)<br />

Stronglogicalconstants ⊗(groupconjunction,fusion)and ⊕(groupdisjunction)aredefinablebyimplicationandnegation:<br />

Rules<br />

• (A ⊕ B) def<br />

≡ ¬(¬A → B)<br />

• (A ⊗ B) def<br />

≡ ¬(¬A ⊕ ¬B)<br />

AdjunctionFrom Aand Binfer A ∧ B.<br />

ModusPonensFrom Aand A → Binfer B.


EpistemicLogicwithRelevantAgents 141<br />

Routley–Meyersemantics<br />

An R-frameisaquintuple F = 〈S,L,C,✂,R〉,where Sisanon-emptysetof<br />

situationsand L ⊆ Sisanon-emptysetoflogicalsituations.Therelations<br />

C ⊆ S 2 , ✂ ⊆ S 2 ,and R ⊆ S 3 wereintroducedinsection2,herewesumup<br />

theirproperties.<br />

Propertiesoftherelation R Thebasicpropertyof R:<br />

if Rxyz, x ′ ✂ x, y ′ ✂ y,and z ✂ z ′ , then Rx ′ y ′ z ′ .<br />

Thismeansthattherelation Rismonotonicwithrespecttotheinvolvement<br />

relation.<br />

Moreoveritisrequiredthat:<br />

(r1) Rxyzimplies Ryxz;<br />

(r2) R 2 (xy)zwimplies R 2 (xz)yw,where R 2 xyzwiff<br />

(∃s)(Rxysand Rszw);<br />

(r3) Rxxx;<br />

(r4) Rxyzimplies Rxz ⋆ y ⋆ .<br />

Propertiesoftherelation C Compatibilitybetweentwostatesisinherited<br />

bythestatesinvolvedinthem(’lessinformativestates’):<br />

If xCy, x1 ✂ x,and y1 ✂ y,then x1Cy1.<br />

Moreover,werequirethefollowingproperties:<br />

(c1)(symmetricity) xCyimplies yCx;<br />

(c2)(directedness) (∀x)(∃y)(xCy);<br />

(c3)(convergence) (∀x)(∃y(xCy)implies (∃x ⋆ )(xCx ⋆ and<br />

∀z(xCzimplies z ✂ x ⋆ )));<br />

(c4) x ✂ yimplies y ⋆ ✂ x ⋆ ;<br />

(c5) x ⋆⋆ ✂ x.<br />

Model R-model MisaR-frame Fwithavaluationfunction v: P → 2 S .<br />

Thetruthofaformulaatasituationisdefinedinthefollowingway:<br />

• s � piff s ∈ v(p),<br />

• s � ¬ϕiff s ⋆ �� ϕ,


142 OndrejMajer&MichalPeliˇs<br />

• s � ψ ∧ ϕiff s � ψand s � ϕ,<br />

• s � ψ ∨ ϕiff s � ψor s � ϕ,<br />

• s � (ϕ → ψ)iff (∀y,z)(Rsyzimplies (y � ϕimplies z � ψ)).<br />

Aswealreadysaid,thetruthofaformulainamodelandinaframe,<br />

respectively,isdefinedastruthinalllogicalsituationsofthismodel/frame.<br />

Asusual, R-tautologiesareformulastrueinallrelevantframes.Whenever<br />

ϕisaR-tautology,wewrite |= ϕandsaythat ϕisavalidformula.<br />

Thecondition(r1)validatestheimplicativeversionofModusPonens<br />

(axiomschema3).Itdoesnotvalidatetheconjunctiveversion (A ∧ (A →<br />

B)) → B,whichrequires(r3).<br />

(r2)correspondstothe‘exchangerule’ (A → (B → C)) → (B →<br />

(A → C)),whichisderivablefromtheaxiomsgivenabove.<br />

(r4)validatescontraposition(axiomschema12).Ifweworkwithoutthe<br />

Routleystar,thiscanberewrittenas:<br />

Rxyzimplies (∀z ′ Cz)(∃y ′ Cy)(Rxy ′ z ′ ).<br />

Directednessandconvergenceconditionsarenecessaryforthedefinition<br />

oftheRoutleystar.From(c1)weobtainthevalidityof (A → ¬¬A)and<br />

fromthelastcondition(c5)wegettheaxiomschema11.<br />

OndrejMajer<br />

InstituteofPhilosophy,AcademyofSciencesoftheCzechRepublic<br />

Jilská1,11000Praha1<br />

majer@site.cas.cz<br />

http://logika.flu.cas.cz<br />

MichalPeliˇs<br />

InstituteofPhilosophy,AcademyofSciencesoftheCzechRepublic<br />

Jilská1,11000Praha1<br />

pelis@ff.cuni.cz<br />

http://logika.flu.cas.cz<br />

References<br />

Cheng,J.(2000).Astrongrelevantlogicmodelofepistemicprocessesinscientific<br />

discovery. InE.Kawaguchi,H.Kangassalo,H.Jaakkola,&I.Hamid(Eds.),<br />

InformationmodellingandknowledgebasesXI(pp.136–159).Amsterdam:IOS<br />

Press.<br />

Duc,H.N. (2001). Resource-boundedreasoningaboutknowledge. Unpublished<br />

doctoraldissertation, FacultyofMathematicsandInformatics, Universityof<br />

Leipzig.


EpistemicLogicwithRelevantAgents 143<br />

Fagin,R.,Halpern,J.,Moses,Y.,&Vardi,M.(2003).Reasoningaboutknowledge.<br />

Cambridge,MA:MITPress.<br />

Mares,E.(2004).Relevantlogic.Cambridge:CambridgeUniversityPress.<br />

Mares,E.,&Meyer,R. (1993). Thesemanticsof r4. JournalofPhilosophical<br />

Logic,22,95–110.<br />

Paoli,F.(2002).Substructurallogics:Aprimer.Dordrecht:Kluwer.<br />

Restall,G. (1993). Simplifeidsemanticsforrelevantlogics(andsomeoftheir<br />

rivals).JournalofPhilosophicalLogic,22,481–511.<br />

Restall,G.(1995).Four-valuedsemanticsforrelevantlogics(andsomeoftheir<br />

rivals).JournalofPhilosophicalLogic,24,139–160.<br />

Restall,G.(1996).Informationflowandrelevantlogics.InLogic,languageand<br />

computation: The1994Moragaproceedings.CSLILectureNotes(Vol.58,pp.<br />

463–477).Stanford,CA:CSLI.<br />

Restall,G. (1999). Negationinrelevantlogics: HowIstoppedworryingand<br />

learnedtolovetheRoutleystar. InD.Gabbay&H.Wansing(Eds.),Whatis<br />

negation?(Vol.13,pp.53–76).Dordrecht:Kluwer.<br />

Restall,G. (2000). Anintroductiontosubstructurallogics. London–NewYork:<br />

Routledge.<br />

Wansing,H. (2002). Diamondsareaphilosopher’sbestfriends. Journalof<br />

PhilosophicalLogic,31,591–612.


Betting on Fuzzy and Many-valued Propositions<br />

1 Introduction<br />

Peter Milne<br />

Ina1968article,‘ProbabilityMeasuresofFuzzyEvents’,LotfiZadehproposedaccountsofabsoluteandconditionalprobabilityforfuzzysets(Zadeh,<br />

1968).Where Pisanordinary(“classical”)probabilitymeasuredefinedon<br />

a σ-fieldofBorelsubsetsofaspace X,and µAisafuzzymembershipfunctiondefinedon<br />

X,i.e.afunctiontakingvaluesintheinterval [0,1],the<br />

probabilityofthefuzzyset Aisgivenby<br />

�<br />

P(A) = µA(x)dP.<br />

X<br />

Thethingtonoticeaboutthisexpressionisthat,inaway,there’snothing<br />

“fuzzy”aboutit.Tobewelldefined,wemustassumethatthe“levelsets”<br />

{x ∈ X : µA(x) ≤ α}, α ∈ [0,1],<br />

are P-measurable. Theseareordinary,“crisp”,subsetsof X. Andthen<br />

P(A)isjusttheexpectationoftherandomvariable µA.—Thisisentirely<br />

classical.Ofcourse,youmayinterpret µAasafuzzymembershipfunction<br />

butreallywehave,ifyou’llpardonthepun,inlargemeasurelostsightof<br />

thefuzziness.<br />

Soyoumightask:<br />

•isthistheonlywaytodefinefuzzyprobabilities?<br />

Theanswer,Ishallargue,isyes.<br />

DefiningconditionalprobabilityZadehoffered<br />

P(A|B) = P(AB)<br />

, when P(B) > 0,<br />

P(B)


146 PeterMilne<br />

where<br />

Onemightwonder:<br />

∀x ∈ X µAB(x) = µA(x) × µB(x).<br />

•isthistheonlywaytodefineconditionalprobabilities?<br />

Theanswer,Ishallsuggest,isno,itisnottheonlywaybutitistheonly<br />

sensibleway.<br />

Zadehassignsprobabilitiestosets.WhatIofferhere,usingDutchBook<br />

Arguments,isavindicationofZadeh’sspecificationswhenprobabilityis<br />

assignedtopropositionsratherthansets.(Buttranslationbetweenpropositiontalkandsetandeventtalkisstraightforward.It’sjustthatproposition<br />

talkfitsbetterwithbettingtalk.)<br />

2 Betsandmany-valuedlogics<br />

Iapply“theDutchBookmethod”,asJeffPariscallsit(Paris,2001),to<br />

fuzzyandmany-valuedlogicsthatmeetasimplelinearitycondition.Ishall<br />

callsuchlogicsadditive.<br />

Additivity<br />

Foranyvaluation vandforanysentences Aand B<br />

v(A ∧ B) + v(A ∨ B) = v(A) + v(B)<br />

where‘∧’and‘∨’theconjunctionanddisjunctionofthelogicinquestion.<br />

Additivityiscommon:theGödel,Łukasiewicz,andproductfuzzylogics<br />

arealladditive,asareGödelandŁukasiewicz n-valuedlogics.<br />

InordertoemployDutchBookarguments,weneedabettingscheme<br />

suitablysensitivetotruth-valuesintermediatebetweentheextremevalues<br />

0and 1.Settingouttheclassicalcasetherightwaymakesonegeneralization<br />

obvious.<br />

Ratherthanbettingodds,whicharealgebraicallylesstractable,weuse,<br />

asisstandard,a“normalized”bettingschemewithfairbettingquotients.<br />

Classically,withabeton Aatbettingquotient pandstake S:<br />

•thebettorgains (1 − p)Sif A;<br />

•thebettorloses pSifnot-A.<br />

Taking 1fortruth, 0forfalsity,and v(A)tobethetruth-valueof A,wecan<br />

summarisethisschemelikethis:<br />

thepay-offtothebettoris (v(A) − p)S.


BettingonFuzzyandMany-valuedPropositions 147<br />

Andnowweseehowtoextendbetstothemanyvaluedcase:weadoptthe<br />

sameschemebutallow v(A)tohavemorethantwovalues.Thesloganis:<br />

thepay-offisthelargerthemoretrue Ais. 1<br />

Usingthisbettingscheme,weobtainDutchBookargumentsforcertainseeminglyfamiliarprinciplesofprobability,seeminglyfamiliarinthat<br />

formallytheyrecapitulateclassicalprinciples.<br />

• 0 ≤ Pr(A) ≤ 1;<br />

• Pr(A) = 1when |= A;<br />

• Pr(A) = 0when A |=;<br />

• Pr(A ∧ B) + Pr(A ∨ B) = Pr(A) + Pr(B).<br />

Here ∧and ∨aretheconjunctionanddisjunction,respectively,ofanadditivefuzzyormany-valuedlogic.<br />

Otherprinciplesthatmayormaynotbeindependent,dependingonthe<br />

logic:<br />

• Pr(A) + Pr(¬A) = 1when v(¬A) = 1 − v(A);<br />

• Pr(A) ≥ xwhen,underallvaluations, v(A) ≥ x;<br />

• Pr(A) ≤ xwhen,underallvaluations, v(A) ≤ x;<br />

• Pr(A) ≤ Pr(B)when A |= B.<br />

I’llshowhowtwooftheargumentsgoasthere’saninterestingconnection<br />

withthestandardDutchBookargumentsusedintheclassical,two-valued<br />

case.<br />

Welet xrangeoverthepossibletruth-values(whichalllieintheinterval<br />

[0,1]).Clearly,forgiven p,wecanchooseavalueforthestake Sthatmakes<br />

Gx = (x − p)S<br />

negative,forallvaluesof xintheinterval [0,1],if,andonlyif, pislessthan<br />

0orgreaterthan 1.Hence<br />

0 ≤ Pr(A) ≤ 1.<br />

1 Thesuggestedpay-offschemeis,ofcourse,onlythemoststraightforwardwaytoimplementtheslogan.<br />

Onecoulddistorttruthvalues: takeastrictlyincreasingfunction<br />

f : [0, 1] 2 → [0, 1]with f(0) = 0, f(1) = 1,andtakepay-offstobegivenby (f(v(A))−p)S.<br />

Analogously,Zadehcouldhavetaken �<br />

X f(µA(x))dPtodefinedistortedprobabilities.—<br />

Andthepointisthatsuch“probabilities”aredistortedforwhen fisnottheidentity<br />

functionitmaybethat P(A) < ceventhough µA(x) > c,forall x ∈ X.


148 PeterMilne<br />

Sofarsogood,buthere’sthecutebit:<br />

Gx = xG1 + (1 − x)G0,<br />

so Gxisnegativeforallvaluesof x ∈ [0,1]if,andonlyif, G1and G0areboth<br />

negative.Fromtheclassicalcase,weknowthatthenecessaryandsufficient<br />

conditionforthelatteristhat plieoutsidetheinterval [0,1].Itsufficesto<br />

lookattheclassicalextremestofixwhatholdsgoodforalltruth-valuesin<br />

theinterval [0,1].<br />

Nextweconsiderfourbets:<br />

1.abeton A,atbettingquotient pwithstake S1;<br />

2.abeton B,atbettingquotient qwithstake S2;<br />

3.abeton A ∧ B,atbettingquotient rwithstake S3;<br />

4.abeton A ∨ B,atbettingquotient swithstake S4.<br />

Weassumethatforallallowedvaluesof v(A)and v(B),<br />

v(A ∧ B) + v(A ∨ B) = v(A) + v(B) and v(A ∧ B) ≤ min{v(A),v(B)}.<br />

Then,where x, y,and zarethetruth-valuesof A, Band A ∧ Brespectively,thepay-offis<br />

Gx,y = (x − p)S1 + (y − q)S2 + (z − r)S3 + ((x + y − z) − s)S4.<br />

Thiscanberewrittenas<br />

Gx,y = zG1,1 + (x − z)G1,0 + (y − z)G0,1 + (1 − x − y + z)G0,0.<br />

Theco-efficientsareallnon-negativeandcannotallbezero.Thus Gx,yis<br />

negative,forallallowable x, y,and z,justincase G1,1, G1,0, G0,1,and<br />

G0,0areallnegative.FromthestandardDutchBookargumentforthetwovalued,classicalcase,weknowthistobepossibleif,andonlyif,<br />

p+q �= r+s.<br />

Hence<br />

Pr(A ∧ B) + Pr(A ∨ B) = Pr(A) + Pr(B).<br />

3 Theclassicalexpectationthesisforfinitely-many-valued<br />

Łukasiewiczlogics<br />

AsaninitialvindicationofZadeh’saccount,wefindthatinthecontext<br />

ofafinitely-many-valuedŁukasiewiczlogic,allprobabilitiesareclassical<br />

expectations. Thatis,theprobabilityofamany-valuedpropositionisthe


BettingonFuzzyandMany-valuedPropositions 149<br />

expectationofitstruth-valueandthatapropositionhasaparticulartruthvalueisexpressibleusingatwo-valuedproposition.<br />

Sointhissetting,in<br />

analogywithZadeh’sassignmentofabsoluteprobabilitiestofuzzysets,all<br />

probabilitiesareexpectationsdefinedoveraclassicaldomain.<br />

InallŁukasiewiczlogics,conjunctionanddisjuctionareevaluatedbythe<br />

functions max{0,x + y − 1}and min{1,x + y},respectively.<br />

EmployingŁukasiewicznegationandoneormoreofŁukasiewiczconjunction,disjunction,andimplication,onecandefineasequenceof<br />

n + 1<br />

formulasofasinglevariable, Jn,0(p),Jn,1(p),... ,Jn,n(p),whichhavethis<br />

property(Rosser&Turquette,1945):inthesemanticframeworkof (n+1)valuedŁukasiewiczlogicitisthecasethatforeveryformula<br />

A,forall k,<br />

0 ≤ k ≤ n,andforeveryvaluation v,<br />

v(Jn,k(A)) = 1,if v(A) = k<br />

n ;<br />

v(Jn,k(A)) = 0,if v(A) �= k<br />

n .<br />

Inthesemanticframeworkof (n + 1)-valuedŁukasiewiczlogic,forallsentences<br />

A,<br />

|= Jn,0(A) ∨Ł Jn,1(A) ∨Ł · · · ∨Ł Jn,n(A) and<br />

Jn,i(A) ∧Ł Jn,j(A) |=, 0 ≤ i < j ≤ n. (*)<br />

Fromtheprobabilityaxioms,wehave,forallsentences A,that<br />

�<br />

Pr(Jn,i(A)) = 1.<br />

0≤i≤n<br />

Thepropositionsoftheform Jn,i(A)aretwo-valued,so, (n + 1)-valued<br />

Łukasiewiczlogicreducingtoclassicallogiconthevalues 0and 1,thelogic<br />

ofthesepropositionsisclassical. Thus,whenrestrictedtothesepropositionsandtheirlogicalcompounds,theprobabilityaxiomsgiveusaclassical,<br />

finitelyadditive,probabilitydistribution. Whatweshownextisthatthis<br />

classicalprobabilitydistributiondeterminestheprobabilitiesofallpropositionsinthelanguage.<br />

Theorem2(ClassicalExpectationThesis).Intheframeworkof (n + 1)valuedŁukasiewiczlogic,<br />

Pr(A) = 1<br />

n<br />

�<br />

0≤i≤n<br />

iPr(Jn,i(A)).<br />

Proof.From(*)andthetwo-valuednessofthe Jn,i(A)’swehave<br />

A =�= (A ∧Ł Jn,0(A)) ∨Ł (A ∧Ł Jn,1(A)) ∨Ł · · · ∨Ł (A ∧Ł Jn,n(A)).


150 PeterMilne<br />

Fromourprobabilityaxiomsitfollowsthatlogicallyequivalentpropositions<br />

mustreceivethesameprobability,so<br />

Pr(A) = �<br />

Pr(A ∧Ł Jn,i(A)). (†)<br />

0≤i≤n<br />

Weconsidertwobets,oneon A ∧Ł Jn,k(A)atbettingquotient pand<br />

stake S1,theotheron Jn,k(A)atbettingquotient qwithstake S2. The<br />

pay-offsare:<br />

G = k<br />

n<br />

G �= k<br />

n<br />

� �<br />

k<br />

= − p S1 + ((1 − q)S2) when Ahastruth-value<br />

n k<br />

n ,<br />

= −pS1 − qS2 when Ahastruth-valueotherthan k<br />

n .<br />

Setting S2 = − k<br />

nS1givesapay-off,independentofthetruth-valueof �<br />

A,of − p S1,whichcanbemadenegativebychoiceof S1provided<br />

� qk<br />

n<br />

p �= qk<br />

n .Ontheotherhand,forarbitrary S1and S2,when p = qk<br />

n thetwo<br />

pay-offsare<br />

�<br />

k<br />

G k<br />

= = (1 − q)<br />

n n S1<br />

�<br />

+ S2 when Ahastruth-value k<br />

, and<br />

n<br />

�<br />

k<br />

G k<br />

�= = −q<br />

n n S1<br />

�<br />

+ S2 when Ahastruth-valueotherthan k<br />

n .<br />

Thesecannotbothbenegative.Hence<br />

Substitutingin(†),weobtain:<br />

Twocomments<br />

Pr(A ∧Ł Jn,k(A)) = k<br />

n Pr(Jn,k(A)).<br />

Pr(A) = 1<br />

n<br />

�<br />

0≤i≤n<br />

iPr(Jn,i(A)).<br />

Firstly,havingbeenobtainedbyanindependentDutchBookargument,the<br />

ClassicalExpectationThesismayseemtobeanadditionalprinciple.Infact<br />

itisnot;itisderivablefromouraxiomsforprobability.Toshowthiswehave<br />

tointroduceapropositionalconstant,introducedintoŁukasiewiczlogicby<br />

Słupeckiinordertoobtainexpressivecompleteness(Słupecki,1936).


BettingonFuzzyandMany-valuedPropositions 151<br />

Inthesemanticsof (n + 1)-valuedŁukasiewiczlogic,inwhichallformulasareassignedvaluesintheset<br />

� 0, 1 2 n−1<br />

n , n ,..., n ,1� ,thepropositional<br />

constant thasthisinterpretation:<br />

underallvaluations v, v(t) =<br />

n − 1<br />

n .<br />

Let t1bethe (n − 2)-fold ∧Ł-conjunctionof twithitself. For 1 < k ≤ n,<br />

let tkbethe (k − 1)-fold ∨Ł-disjunctionof t1withitself. v(t1) = 1<br />

n and<br />

v(tk) = k<br />

n .Sincewehave<br />

tk ∧Ł t1 |=, 1 ≤ k < n, and<br />

|= tn,<br />

fromourprobabilityaxiomsweobtain:<br />

Pr(tk) = k Pr(t1), 1 ≤ k ≤ n, and<br />

Pr(tn) = 1,<br />

hence<br />

Pr(tk) = k<br />

, 1 ≤ k ≤ n.<br />

n<br />

Usingthe ti’swecanderivetheClassicalExpectationThesis.(I’llskipthe<br />

detailshere.)<br />

Secondly,theDutchBookargumentfortheClassicalExpectationThesis<br />

goesthroughwithanynotionofconjunctionforwhich v(A&B) = v(A)<br />

when v(B) = 1and v(A&B) = 0when v(B) = 0. Also,the Jn,i(A)’s<br />

beingtruth-functional,theClassicalExpectationThesisholdsgoodofevery<br />

propositioninthesemanticframework,notjustthoseexpressibleusingthe<br />

Łukasiewiczconnectives.<br />

4 Theextensiontoinfinitelymanytruth-values(asketch)<br />

Foranyrationalnumber xintheinterval [0,1],thereisaformula φ(p)ofa<br />

singlepropositional-variable p,constructedusingŁukasiewicznegationand<br />

anyoneormoreofŁukasiewiczconjunction,disjunction,orimplication,<br />

suchthat,underanyvaluationtakingvaluesin [0,1], v (φ(A/p)) = 0if<br />

v(A) ≤ xand v (φ(A/p)) > 0otherwise(McNaughton,1951).<br />

EmployingtheGödelnegation, 2 then,wehave,<br />

2 TheGödelnegationis,tobesure,notusuallytakentobepartofthevocabulary<br />

oftheŁukasiewiczlogics. Semantically,however,itcanbedefinedintheŁukasiewicz<br />

fuzzy/many-valuedframeworksastheexternalnegationthatmaps 0to 1andallother<br />

valuesto 0.


152 PeterMilne<br />

•foreachinterval [0,x]with xrational,aformula J [0,x](A)thattakes<br />

thevalue 1underanyvaluation vforwhich v(A) ≤ xandotherwise<br />

takesthevalue 0;<br />

•foreachhalf-openinterval (x,y]withrationalendpoints xand y, x <<br />

y,aformula J (x,y](A)thattakesthevalue 1underavaluation vwhen<br />

v(A) ∈ (x,y]andotherwisetakesthevalue 0.<br />

Givenastrictlyincreasing,finitesequence x0,x1,... ,xn−1ofrational<br />

numbersintheopeninterval (0,1),considerthefamilyof n + 1bets:<br />

�<br />

2≤i≤n<br />

•abeton Aatbettingquotient qwithstake S;<br />

•abeton J [0,x1](A)atbettingquotient p1withstake S1;<br />

•abeton J (xi−1,xi](A)atbettingquotient piwithstake Si, 1 < i < n;<br />

•abeton J (xi,1](A)atbettingquotient pnwithstake Sn.<br />

xi−1 Pr(J (xi−1,xi](A)) ≤ Pr(A) ≤<br />

≤ x1 Pr(J [0,x1](A)) + �<br />

2≤i≤n<br />

xi Pr(J (xi−1,xi](A)),<br />

where xn = 1.Sobytakingfinerandfinerpartitionswecanmoreclosely<br />

approximatetheprobabilityof Afromaboveandbelow.Thismaynotquite<br />

dotofix Pr(A)exactly. Forthatwemayalsoneedtheprobabilitiesofat<br />

mostacountableinfinityof(two-valued)statementsoftheform<br />

v(A) ≤ x<br />

where xisanirrationalnumber. 3<br />

Withtheseinhand,wethenfindthat<br />

Pr(A) =<br />

� 1<br />

0<br />

xdFA(x),<br />

where FAistheordinary,“classical”distributionfunctiondeterminedby<br />

theprobabilitiesofthe J [0,x](A)’s, J (x,y](A)’sandhowevermany v(A) ≤ x’s<br />

with xirrationalwehaveused.<br />

Byintroducingacountablyinfinitefamilyoflogicalconstants,wecan<br />

derivethisclassicalrepresentationfromthepreviouslygivenprinciplesof<br />

probabilitytogetherwiththeprinciple<br />

3 RecallZadeh’sassumptionregardingthe P-measureabilityof“levelsets”.


BettingonFuzzyandMany-valuedPropositions 153<br />

•foranyproposition Alogicallyconstrainedtotakeonlythevalues 0<br />

and 1andforrationalvaluesof xintheinterval [0,1], Pr(tx ∧ A) =<br />

xPr(A),<br />

where txtakesthevalue xunderallvaluations v.<br />

Thereallyneatfeatureofinfinitelymany-valuedŁukasiewiczlogicsis<br />

thatthisprincipleisderivablefromthebasicprinciples<br />

• 0 ≤ Pr(A) ≤ 1;<br />

• Pr(A) = 1when |= A;<br />

• Pr(A) = 0when A |=;<br />

• Pr(A ∧Ł B) + Pr(A ∨Ł B) = Pr(A) + Pr(B).<br />

5 Conditionalprobabilities<br />

Intheclassicalsetting,abeton Aconditionalon Bisabetthatgoesahead<br />

if,andonlyif, Bistrueandisthenwonorlostaccordingastowhether<br />

Aistrueornot. Thepay-offsforsuchaconditionalbetwithstake Sat<br />

bettingquotient pare:<br />

•thebettorgains (1 − p)Sif Aand B;<br />

•thebettorloses pSifnot-Aand B;<br />

•thebettorneithergainsnorlosesifnot-B.<br />

Wecansummarisethisbettingschemelikethis:<br />

v(B)(v(A) − p)S.<br />

Andso,aswithordinarybets,wenowknowonewaytoextendthe<br />

schemeforconditionalbetsonclassical,two-valuedpropositionstomanyvaluedpropositions.<br />

AstraightforwardDutchBookargument,whichagainpiggy-backson<br />

theproofinthetwo-valuedcase,thentellsusthat<br />

where<br />

Pr(A ∧× B) = Pr(A|B) × Pr(B)<br />

v(A ∧× B) = v(A) × v(B).<br />

—Allowingforthechangeofsetting,justwhatZadehsaid.


154 PeterMilne<br />

Youcan,ifyouaresominded,generalizetheclassicalschemeusingany<br />

many-valuedorfuzzyconjunctionthatis“classicalattheextremes”:<br />

(v(A ∧ B) − v(B)p)S.<br />

ADutchBookargument—inallessentials,thesameDutchBookargument<br />

—willthendeliver:<br />

Pr(A ∧ B) = Pr(A|B) × Pr(B).<br />

However, Pr(·|B)satisfiestheaxiomsforanabsoluteprobabilitymeasure<br />

onlywhentheproductconjunction, ∧×isused. 4<br />

PeterMilne<br />

DepartmentofPhilosophy,UniversityofStirling<br />

StirlingFK49LA,UnitedKingdom<br />

peter.milne@stir.ac.uk<br />

References<br />

McNaughton,R.(1951).Atheoremaboutinfinite–valuedsententiallogic.Journal<br />

ofSymbolicLogic,16,1–13.<br />

Paris,J.(2001).Anoteonthedutchbookmethod.InG.DeCooman,T.Fine,<br />

&T.Seidenfeld(Eds.),ISIPTA’01,Proceedingsofthesecondinternationalsymposiumonimpreciseprobabilitiesandtheirapplications,Ithaca,NY,USA(pp.<br />

301–306).Maastricht:ShakerPublishing.(Aslightlyrevisedversionisavailable<br />

on-lineathttp://www.maths.manchester.ac.uk/∼jeff/papers/15.ps)<br />

Rosser,J.,&Turquette,A.(1945).Axiomschemesfor m–valuedpropositional<br />

calculi.JournalofSymbolicLogic,10,61–82.<br />

Słupecki,J.(1936).DervolledreiwertigeAussagenkalkül.Comptesrenduesdes<br />

séancesdelaSociétédesSciencesetLettresdeVarsovie,29,9–11.<br />

Zadeh,L.A.(1968).Probabilitymeasuresoffuzzyevents.JournalofMathematicalAnalysisandApplications,23,421–427.<br />

4 Beyondtheclassical,two-valuedcase,productconjunctionrequiresthattherebean<br />

infinityoftruth-values.


Inferentializing Consequence<br />

Jaroslav Peregrin ∗<br />

Theproofofcorrectnessandcompletenessofalogicalcalculusw.r.t.a<br />

givensemanticscanbereadastellingusthatthetautologies(or,moregenerally,therelationofconsequence)specifiedinamodel-theoreticwaycan<br />

beequallywellspecifiedinaproof-theoreticway,bymeansofthecalculus<br />

(asthetheorems,resp. therelationofinferabilityofthecalculus). Thus<br />

weknowthatbothfortheclassicalpropositionalcalculusandfortheclassicalpredicatecalculustheoremsandtautologiesrepresenttwosidesofthe<br />

samecoin.Wealsoknowthattherelationofinferenceasinstitutedbyany<br />

ofthecommonaxiomsystemsoftheclassicalpropositionalcalculuscoincideswiththerelationofconsequencedefinedintermsofthetruthtables;<br />

whereasthesituationisalittlebitmorecomplicatedw.r.t.theclassical<br />

predicatecalculus(thecoincidenceoccursifwerestrictourselvestoclosed<br />

formulas;otherwise ∀xFxisinferablefrom Fxwithoutbeingitsconsequence).Andofcoursewealsoknowcaseswhereaclassoftautologiesofasemanticsystemdoesnotcoincidewiththeclassoftheoremsofanycalculus.<br />

(Theparadigmaticcaseisthesecond-orderpredicatecalculuswith<br />

standardsemantics.)<br />

Thismaymakeusconsidertheproblemof“inferentializability”.Which<br />

semanticsystemsare“inferentializable”inthesensethattheirtautologies<br />

(theirrelationofconsequence,respectively)coincidewiththeclassoftheorems(therelationofinferability,respectively)ofacalculus?Oneanswer<br />

isready:itisifandonlyifthesetoftautologiesisrecursivelyenumarable.<br />

Butthisanswerisnotveryinformative,indeedsayingthatthesetisrecursivelyenumerableisonlyreiteratingthatitconicideswiththeclassof<br />

theoremsofacalculus. Moreover,payingdueattentiontothetermssuch<br />

as“calculus”and“inference”showsusthatitispossibletorelatethemto<br />

various“levels”,wherebytheproblemofinferentializabilitybecomesquite<br />

nontrivial.<br />

∗ WorkonthispaperwassupportedbythegrantNo.401/07/0904oftheCzechScience<br />

Foundation.


156 JaroslavPeregrin<br />

1 Consequence<br />

Consequence,astheconceptisusuallyunderstood,amountstotruth-preservation,i.e.,<br />

Aisaconsequenceof A1,... ,Aniffthetruthofallof A1,...,An<br />

bringsaboutthetruthof A,i.e.,iffanytruthvaluationmappingallof<br />

A1,... ,Anon 1mapsalso Aon 1. 1 Itisobviousthatthe“any”fromthe<br />

previoussentencecannotmean“anywhatsoever”(ofcoursetheredoesexistafunctionmappingallof<br />

A1,... ,Anon 1and Aon 0!),itmustmean<br />

somethinglike“anyadmissible”.Hencetheremustbesomeconceptofadmissibilityinplay:somemappingsofsentencesof<br />

{0,1}willbeadmissible,<br />

othersnot. But,ofcourse,thatifwetakethesentencestobesentences<br />

ofameaningfullanguage,suchadivisionofvaluationsisforthcoming: if<br />

A1,... ,AnareFidoisadogandEverydogisamammal(hence n = 2), A<br />

isFidoisamammal,thenthevaluationmappingtheformertwosentences<br />

on 1andthelatteroneon 0isnotadmissible—itisnotcompatiblewith<br />

thesemanticsofEnglish.<br />

Henceweassumethatanysemanticsofanylanguageprovidesforthe<br />

divisionofthesentencesofthelanguageintotrueandfalse,therebydividingthespaceofthemappingsofthesentenceson<br />

{0,1}intoadmissible<br />

andinadmissible.(InfactImaintainamuchstrongerthesis,namelythat<br />

anysemanticscanbereduced tosuchadivision,butIamnotgoingto<br />

argueforthisthesishere—Ihavedonesoelsewhere,see(Peregrin,1997).)<br />

Therebyitalsoestablishestherelationofconsequence,astherelationof<br />

truth-preservationforalladmissiblevaluations. Ifweusethesentences<br />

S1,S2,... ofthelanguageinquestiontomarkcolumnsofthefollowing<br />

tableusingallpossibletruth-valuationsasitsrows,wecanlookatthe<br />

delimitationoftheadmissiblevaluationsasstrikingoutrowsofthetable.<br />

1 See(Peregrin,2006).<br />

S1 S2 S3 S4 · · ·<br />

v1 0 0 0 0 · · ·<br />

v2 1 0 0 0 · · ·<br />

v3 0 1 0 0 · · ·<br />

v4 1 1 0 0 · · ·<br />

v5 0 0 1 0 · · ·<br />

v6 1 0 1 0 · · ·<br />

.<br />

. . . . . ..


InferentializingConsequence 157<br />

Amoreexactarticulationofthesenotionsyieldsthefollowingdefinition:<br />

Definition5.Asemanticsystemisanorderedpair 〈S,V 〉,where Sisaset<br />

(theelementsofwhicharecalledsentences)and V ⊆ {0,1} S .Theelements<br />

of {0,1} S arecalledvaluations (of S). (Avaluationwillbesometimes<br />

identifiedwiththesetofallthoseelementsof Sthataremappedon 1by<br />

it.)Theelementsof Varecalledadmissiblevaluationsof 〈S,V 〉,theother<br />

valuations(i.e.theelementsof {0,1} S \ V)arecalledinadmissible. The<br />

relationofconsequenceinducedbythissystemistherelation |=definedas<br />

follows<br />

X |= Aiff v(A) = 1forevery v ∈ Vsuchthat v(B) = 1forevery B ∈ X.<br />

2 VarietiesofInference<br />

Nowconsiderthestipulationofaninference, A1,... ,An ⊢ A(forsome<br />

elements A1,... ,An, Aof S).Suchastipulationcanbeseenasexcluding<br />

certainvaluations:namelyallthosethatmap A1,...,Anon 1and Aon 0.<br />

(Thus,forexample,theexclusionsintheabovetablemightbetheresult<br />

ofstipulating S1 ⊢ S2.)Henceifwecallthepairconstitutedbyafiniteset<br />

ofelementsof Sandanelementof Saninferon,wecansaythatinferons<br />

excludevaluationsandaskwhichsetsofvaluationscanbedemarcatedby<br />

meansofinferons.<br />

Definition6.Aninferon(over S)isanorderedpair 〈X,A〉where Xisa<br />

finitesubsetof Sand Aisanelementof S. Aninferonissaidtoexclude<br />

anelement vof {0,1} S iff v(B) = 1forevery B ∈ Xand v(A) = 0. An<br />

orderedpair 〈S, ⊢〉suchthat Sisasetand ⊢isafinitesetofinferons<br />

(i.e.abinaryrelationbetweenfinitesubsetsof Sandelementsof S)willbe<br />

calledaninferentialstructure.Aninferentialstructureissaidtodetermine<br />

asemanticsystem 〈S,V 〉iff Visthesetofallandonlyelementsof {0,1} S<br />

notexcludedbyanyelementof ⊢.Asemanticsystemiscalledaninferential<br />

systemiffitisdeterminedbyaninferentialstructure.<br />

Nowanobviousquestioniswhichsemanticsystemsareinferential.But<br />

beforeweturnourattentiontoit,wewillconsidervariouspossiblegeneralizationsoftheconceptofinference.<br />

First,letaquasiinferondifferfrom<br />

aninferoninthatitssecondcomponentisnotasinglestatement,buta<br />

finitesetofstatements. Aquasiinferonwillexcludeeveryvaluationthat<br />

mapseveryelementofitsfirstcomponenton 1andeveryelementofits<br />

secondcomponenton 0. (Ofcoursetheconceptofquasiinferondefined<br />

inthiswayiscloselyconnectedwiththeconceptofsequentasintroduced


158 JaroslavPeregrin<br />

by(Gentzen,1934)and(Gentzen,1936). 2 ) Second,letasemiinferondifferfromaninferoninthatitsfirstcomponentisnotnecessarilyfinite.<br />

A<br />

semiquasiinferonwillbeaquasiinferonwithbothitsfirstanditssecond<br />

componentnotnecessarilyfinite. Third,letaprotoinferential structure<br />

beaninferentialstructurewithitssecondcomponentnotnecessarilyfinite<br />

(andthinkoftheconceptsofprotosemiinferential,protoquasiinferentialand<br />

protosemiquasiinferentialstructureanalogously).<br />

Inthefollowingdefinition,weabbreviatetheprefixes,whichhavealready<br />

becomesomewhatmonstrous:<br />

Definition7.Anelementof Pow(S) × Pow(S)iscalledanSQI-onover<br />

S.ItiscalledaQI-onifitisanelementof FPow(S) × FPow(S)(where<br />

FPow(S)isthesetofallfinitesubsetsof S),itiscalledanSI-onifitis<br />

anelementof Pow(S) × SanditiscalledanI-onifitisanelementof<br />

FPow(S) ×S. 3 Theorderedpair 〈S, ⊢〉where ⊢isasetofSQI-ons(QI-ons,<br />

SI-ons,I-ons)willbecalledaPSQI-structure(PQI-structure,PSI-structure,<br />

PI-structure). ItiscalledanSQI-structure(QI-structure,SI-structure,Istructure)iff<br />

⊢isfinite.AnSQI-on 〈X,Y 〉issaidtoexcludeanelement v<br />

of {0,1} S iff v(B) = 1forevery B ∈ Xand v(A) = 0forevery A ∈ Y.A<br />

(P)(S)(Q)I-structure 〈S, ⊢〉issaidtodetermineasemanticsystem 〈S,V 〉iff<br />

Visthesetofallandonlyelementsof {0,1} S notexcludedbyanyelement<br />

of ⊢.Asemanticsystemiscalleda(P)(S)(Q)I-systemiffitisdetermined<br />

bya(P)(S)(Q)I-structure.<br />

Summarizingtheconceptsintroducedinthisdefinition,wehavethefollowingtable:<br />

〈S, ⊢〉isa... iff ⊢isa... ⊢thusbeingasubsetof<br />

I-structure afinitesetofI-ons FPow(S) × S<br />

QI-structure afinitesetofQI-ons FPow(S) × FPow(S)<br />

SI-structure afinitesetofSI-ons Seq(S) × S<br />

PI-structure asetofI-ons FPow(S) × S<br />

SQI-structure afinitesetofSQI-ons Pow(S) × Pow(S)<br />

PQI-structure asetofQI-ons FPow(S) × FPow(S)<br />

PSI-structure asetofSI-ons Seq(S) × S<br />

PSQI-structure asetofSQI-ons Pow(S) × Pow(S)<br />

2 Foranexpositionofsequentcalculusanditsrelationshiptothemorestraightforwardly<br />

inferentialapproachasembodiedinnaturaldeductionsee,e.g.,(Negri&Plato,2001).<br />

3 Throughoutthewholepaperweidentifysingletonswiththeirrespectivesingleelements;<br />

henceweoftenwritesimply vinsteadof {v}.


InferentializingConsequence 159<br />

Ouraimnowistofindcriteriaofthevariouslevelsofinferentializability.<br />

Beforewestateandprovetheoremscrucialinthisrespect,weintroduce<br />

somemoredefinitions.<br />

3 CriteriaofInferentializability<br />

Definition8.Let Ubeasetofvaluationsofasemanticsystem 〈S,V 〉<br />

(i.e.asubsetof {0,1} S ). T(U)(thesetof U-tautologies)willbethesetof<br />

allthoseelementsof Swhicharemappedon 1byallelementsof U;and<br />

analogously C(U)(thesetof U-contradictions)willbethesetofallthose<br />

elementsof Swhicharemappedon 0byallelementsof U.Let Xand Ybe<br />

subsetsof S.Theclustergeneratedby Xand Y, Cl[X,Y ],willbetheset<br />

ofallthevaluationsthatmapallelementsof Xon 1andallelementsof Y<br />

on 0.Generally, Uisaclusteriffitcontains(andhenceisidenticalwith)<br />

Cl[T(U),C(U)].Acluster Uiscalledfinitaryiffboth T(U)and C(U)are<br />

finite,itiscalledinferentialiff C(U)isasingleton.<br />

Nowitisclearthatasemanticsystem 〈S,V 〉isaPSQI-systemiff {0,1} S \<br />

V isaunionofclusters.(HenceeverysemanticsystemisaPSQI-system,<br />

foreverysinglevaluationconstitutesacluster.)Thereasonisthatasystem<br />

isaPSQI-systemifitsinadmissiblevaluationsaredeterminedbyasetof<br />

SQI-onsandwhatanSQI-onexcludesisaclusterofvaluations.Ifweuse<br />

specifickindsofSQI-ons,suchasSI-ons,wewillhaveaspecifickindof<br />

clusters,likeinferentialclusters;andifweallowforonlyafinitenumberof<br />

SQI-ons,wewillhavetocountwithonlyfiniteunions.Thisyieldsusthe<br />

factssummarizedinthefollowingtable:<br />

〈S,V 〉isa... iff {0,1} S \ Visaunionof...<br />

PSQI-system clusters<br />

PSI-system inferentialclusters<br />

PQI-system finitaryclusters<br />

SQI-system afinitenumberofclusters<br />

PI-system finitaryinferentialclusters<br />

SI-system afinitenumberofinferentialclusters<br />

QI-system afinitenumberoffinitaryclusters<br />

I-system afinitenumberoffinitaryinferentialclusters<br />

Theorem3.Asemanticsystem 〈S,V 〉isaPSI-systemiff V contains<br />

every v ∈ {0,1} S suchthatforevery A ∈ C(v)thereisa v ′ ∈ V suchthat<br />

T(v) ⊆ T(v ′ )and A ∈ C(v ′ ).


160 JaroslavPeregrin<br />

Proof.Asemanticsystem 〈S,V 〉isaPSI-systemsystemiff {0,1} S \ V is<br />

aunionofinferentialclusters.ThisistosaythatitisaPSI-systemifffor<br />

every v ∈ {0,1} S \ V thereisaset X ⊆ T(v)andasentence A ∈ C(v)<br />

suchthatnovaluation v ′ suchthat X ⊆ T(v ′ )and A ∈ C(v ′ )isadmissible.<br />

Inotherwords, 〈S,V 〉isaPSI-systemiffforevery v �∈ V thereisaset<br />

X ⊆ T(v)andasentence A ∈ C(v)suchthat V doesnotcontainany v ′<br />

suchthat X ⊆ T(v ′ )and A ∈ C(v ′ ). Bycontraposition, 〈S,V 〉isaPSIsystemiffthefollowingholds:givenavaluation<br />

v,ifforeveryset X ⊆ T(v)<br />

andeverysentence A ∈ C(v)thereisav ′ ∈ V suchthat X ⊆ T(v ′ )and<br />

A ∈ C(v ′ ),then v ∈ V.Thisconditioncanobviouslybesimplifiedto:given<br />

avaluation v,ifforeverysentence A ∈ C(v)thereisav ′ ∈ V suchthat<br />

T(v) ⊆ T(v ′ )and A ∈ C(v ′ ),then v ∈ V.<br />

Theorem4.Asemanticsystem 〈S,V 〉isaPQI-systemiff V contains<br />

every vsuchthatforeveryfinite X ⊆ T(v)andfinite Y ⊆ C(v)thereisa<br />

v ′ ∈ V suchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ).<br />

Proof.Asemanticsystem 〈S,V 〉isaPQI-systemsystemiff {0,1} S \ Visa<br />

unionoffiniteclusters.ThisistosaythatitisaPQI-systemiffforevery<br />

v ∈ {0,1} S \ V therearefinitesets X ⊆ T(v)and Y ⊆ C(v)suchthatno<br />

valuation v ′ suchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ )isadmissible. Inother<br />

words, 〈S,V 〉isaPQI-systemiffforevery v �∈ V therearesets X ⊆ T(v)<br />

and Y ⊆ C(v)suchthat Vdoesnotcontainany v ′ suchthat X ⊆ T(v ′ )and<br />

Y ⊆ C(v ′ ).Bycontraposition, 〈S,V 〉isaPQI-systemiffthefollowingholds:<br />

givenavaluation v,ifforeverysets X ⊆ T(v)and Y ⊆ C(v)thereisa<br />

v ′ ∈ Vsuchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ),then v ∈ V.Thisconditioncan<br />

obviouslybesimplifiedto:givenavaluation v,ifforeveryfinite X ⊆ T(v)<br />

andfinite Y ⊆ C(v)thereisav ′ ∈ Vsuchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ),<br />

then v ∈ V.<br />

Weleaveouttheproofofthefollowingtheorem,asitisstraightforwardly<br />

analogoustotheproofsoftheprevioustwo.<br />

Theorem5.Asemanticsystem 〈S,V 〉isaPI-systemiff Vcontainsevery<br />

vsuchthatforeveryfinite X ⊆ T(v)andevery A ∈ C(v)thereisa v ′ ∈ V<br />

suchthat X ⊆ T(v ′ )and A ∈ C(v ′ ).<br />

Hencewehavenecessaryandsufficientconditionsforasemanticsystem<br />

beingaPSI-,aPQI-,oraPI-system.Unfortunately,wedonothavesuch<br />

conditionsforitsbeinganSQI-,anSI-,aQI-,oranI-system.However,we<br />

areabletoformulateatleastausefulnecessaryconditionforitsbeingan<br />

SQI-system.


InferentializingConsequence 161<br />

Theorem6.Asemanticsystem 〈S,V 〉isanSQI-systemonlyif Vcontains<br />

no vsuchthatforeveryfinite X ⊆ T(v)andfinite Y ⊆ C(v)thereisa<br />

v ′ �∈ V suchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ).<br />

Proof.Asemanticsystem 〈S,V 〉isaPQI-systemiff {0,1} S \ V isafinite<br />

unionofclusters.HenceifitisaPQI-system,theremustexistafiniteset<br />

Iandtwocollections 〈X i 〉i∈I, 〈Y i 〉i∈Iofsubsetsof Ssothat<br />

{0,1} S \ V = �<br />

i∈I<br />

Cl[X i ,Y i ].<br />

Thisisthecaseiff Vequalsthecomplementof �<br />

V = �<br />

i∈I<br />

Cl[X i ,Y i ].<br />

Butas Cl[X i ,Y i ] = {v : X i ⊆ T(v)and Y i ⊆ C(v)},<br />

Cl[X i ,Y i ] = {v : X i �⊆ T(v)or Y i �⊆ C(v)} =<br />

Cl[X<br />

i∈I<br />

i ,Y i ],henceiff<br />

= {v : X i ∩ C(v) �= ∅or Y i ∩ T(v) �= ∅} =<br />

= {v : X i ∩ C(v) �= ∅} ∪ {v : Y i ∩ T(v) �= ∅} =<br />

= �<br />

x∈X i<br />

= �<br />

x∈X i<br />

{v : x ∈ C(v)} ∪ �<br />

{v : y ∈ T(v)} =<br />

Cl[∅, {x}] ∪ �<br />

y∈Y i<br />

y∈Y i<br />

Cl[{y}, ∅].<br />

NowusingthegeneralizeddeMorgan’slawsayingthat<br />

� � � �<br />

=<br />

j∈I j∈J<br />

f∈F j∈f +<br />

Z j<br />

i<br />

f∈F j∈I<br />

Z j<br />

f(j)<br />

where F = IJ ,wecanseethat<br />

V = � �<br />

Cl[f(j), ∅] ∩ �<br />

Cl[∅,f(j)]<br />

j∈f −<br />

where Fisthesetofallfunctionsmappingevery i ∈ Ionanelementof f(i)<br />

of Xi ∪Yi,and f + ,and f − ,respectively,arethesetsofallthoseelementsof<br />

Ithataremappedby fonelementsof Xi,and Yi,respectively.Itfurther<br />

followsthat<br />

V = �<br />

Cl[X f ,Y f ]<br />

f∈F<br />

where X f = {f(j) : j ∈ f + }and Y f = {f(j) : j ∈ f − }.Asboth f + and f −<br />

arefinite,thismeansthat Visaunionoffiniteclusters.Itfollowsthatfor


162 JaroslavPeregrin<br />

every v ∈ V therearefinitesets X ⊆ T(v)and Y ⊆ C(v)suchthatevery<br />

valuation v ′ suchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ )isadmissible. Inother<br />

words,forevery v ∈ Vtherearesets X ⊆ T(v)and Y ⊆ C(v)suchthat V<br />

containsevery v ′ suchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ).Bycontraposition:<br />

givenavaluation v,ifforeveryset X ⊆ T(v)and Y ⊆ C(v)thereisa<br />

v ′ �∈ Vsuchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ),then v �∈ V.Thisconditioncan<br />

obviouslybesimplifiedto:givenavaluation v,ifforeveryfinite X ⊆ T(v)<br />

andfinite Y ⊆C(v)thereisav ′ �∈ Vsuchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ ),<br />

then v �∈ V.<br />

4 AHierarchyofSemanticSystems<br />

Letusintroducesomemoredefinitions.<br />

Definition9.Asemanticsystem 〈S,V 〉iscalled<br />

•saturatediff Vcontainsevery vsuchthatforevery A ∈ C(v)thereis<br />

a v ′ ∈ Vsuchthat T(v) ⊆ T(v ′ )and A ∈ C(v ′ );<br />

•compactiff Vcontainsevery vsuchthatforeveryfinite X ⊆ T(v)and<br />

finite Y ⊆ C(v)thereisav ′ ∈ Vsuchthat X ⊆ T(v ′ )and Y ⊆ C(v ′ );<br />

•co-compactiffVcontainsno vsuchthatforeveryfinite X ⊆ T(v)and<br />

finite Y ⊆ C(v)thereisav ′ �∈ Vsuchthat X ⊆ T(v)and Y ⊆ C(v ′ ).<br />

•compactlysaturatediff V containsevery vsuchthatforeveryfinite<br />

X ⊆ T(v)andevery A ∈ C(v),thereisav ′ ∈ Vsuchthat X ⊆ T(v ′ )<br />

and A ∈ C(v ′ ).<br />

Giventhese,wecanrephrasethetheoremswehaveprovedinthefollowingway:<br />

Theorem7.Asemanticsystem 〈S,V 〉is<br />

•alwaysaPSQI-system;<br />

•aPSI-systemiffitissaturated;<br />

•aPQI-systemiffitiscompact;<br />

•anSQI-systemonlyifitisco-compact;<br />

•aPI-systemiffitiscompactlysaturated;<br />

Moreover,easycorollariesofthetheoremsarethefollowingnecessary<br />

conditionsforasystembeinganSI-,aQI-andanI-system:<br />

Corollary2.Asemanticsystem 〈S,V 〉is<br />

•anSI-systemonlyifitissaturatedandco-compact;


InferentializingConsequence 163<br />

•aQI-systemonlyifitiscompactandco-compact;<br />

•anI-systemonlyifitiscompactlysaturatedandco-compact.<br />

Thekindsofsemanticsystemswehaveintroducedcanbearrangedinto<br />

thefollowingdiagram,wherethearrowsindicatecontainmentinthesense<br />

thatanarrowleadsfromaconcepttoadifferentoneiftheextensionofthe<br />

formerincludesthatofthelatter.<br />

SPQI-system =semanticsystem[Σ1]<br />

PQI-system[Σ2] SPI-system[Σ3] SQI-system[Σ4]<br />

PI-system[Σ5] QI-system[Σ6] SI-system[Σ7]<br />

I-system[Σ8]<br />

Diagram1<br />

Whatwearegoingtoshownowisthatalltheinclusionsareproper.<br />

Thesymbolsinbracketsfollowingeachkindtermisthenameofasemantic<br />

systemwhichwillwitnesstheproperness.Thesystemsarethefollowing(S<br />

issupposedtobeaninfiniteset):<br />

• Σ1 = 〈S, {v ∈ Pow(S) : T(v)isfinite}〉;<br />

• Σ2 = 〈S, {∅}〉;<br />

• Σ3 = 〈S, {v ∈ Pow(S) : C(v)isfinite}〉;<br />

• Σ4 = 〈S,Pow(S) \ {S}〉;<br />

• Σ5 = 〈S, {S}〉;<br />

• Σ6 = 〈{A,B}, {{A}, {B}}〉;<br />

• Σ7 = 〈S, {v ∈ Pow(S) : C(v) = A}〉forafixed A ∈ S;<br />

• Σ8 = 〈{A,B}, {{A,B}, {B}}〉.<br />

ToshowthattheydofitintotheveryslotsofDiagram1wherewehave<br />

putthem,letusfirstgiveonemoredefinition:<br />

Definition10.Avaluationiscalledfullifitmapseverysentenceon 1.<br />

(Inotherwords,thefullvaluationis S.) Avaluationiscalledemptyifit<br />

mapseverysentenceon 0.(Inotherwords,theemptyvaluationis ∅.)


164 JaroslavPeregrin<br />

Σ1isnotsaturated,for Vdoesnotcontainthefullvaluation, f,though<br />

forevery A ∈ C(f)thereisav∈ V suchthat T(f) ⊆ T(v)and A ∈ C(v).<br />

(Asthereisno A ∈ C(v),thisholdstrivially.Itfollowsthatnosystemnot<br />

admittingthefullvaluationissaturated.) HenceitisnotaPSI-system.<br />

Itisnotcompact,because V doesnotcontainthefullvaluation,butfor<br />

everyfinitesubset Xof T(f)itcontainsav ′ suchthat X ⊆ T(v ′ )(whereas<br />

Y ⊆ C(v ′ )foreveryfinitesubset Yof C(f)holdstrivially);henceitisnot<br />

aPQI-system. Moreover,itisnotco-compact,for V containstheempty<br />

valuation,whereasas Vcannotcontainanyvaluationmappingonlyafinite<br />

numberofsentenceson 0,thereis,foreveryfinitesubset Yof S,av ′ �∈ V<br />

suchthat Y = C(v ′ ).HenceitisnotanSQI-system.<br />

Σ2isaPQI-system,foritisdeterminedbytheinfinitesetofQI-ons<br />

{〈{A}, ∅〉 : A ∈ S}.However,itisnotsaturated,for Vdoesnotcontainthe<br />

fullvaluation,henceitisnotaP(S)I-system.Alsoitisnotco-compact,for<br />

Vcontainstheemptyvaluation,whereasforeveryfinitesubset Yof Sthere<br />

isav ′ �∈ V suchthat X ⊆ C(v ′ )(whereasthat Y ⊆ T(v ′ )foreveryfinite<br />

subset Yof T(f)holdstrivially);henceitisnota(S)QI-system.<br />

Σ3isaPSI-system,foritisdeterminedbytheinfinitesetofSI-ons<br />

{〈X,A〉 : X ⊆ Sand Xisinfinite}. However,itisnotcompact,because<br />

V doesnotcontaintheemptyvaluation,butforeveryfinitesubset Yof<br />

Sitcontainsav ′ suchthat Y = C(v);henceitisnotaP(Q)I-system.<br />

Moreover,itisnotco-compact,for V containsthefullvaluation,whereas<br />

foreveryfinitesubset Xof Sthereisav ′ �∈ V suchthat X = T(v ′ ),hence<br />

itisnotanS(Q)I-system.<br />

Σ4isanSQI-system,foritisdeterminedbytheSQI-on 〈S, ∅〉.However,<br />

itisnotsaturated,for V doesnotcontainthefullvaluation,henceitis<br />

nota(P)SI-system. Itisnotcompact,because V doesnotcontainthe<br />

fullvaluation,butforeveryfinitesubset Xof Sitcontainsav ′ suchthat<br />

X = T(v ′ );henceitisnota(P)QI-system.<br />

Σ5isaPI-systemforitisdeterminedbytheinfinitesetofI-ons {〈∅, {A}〉 :<br />

A ∈ S}.Butitisnotco-compact,for Vcontainsthefullvaluation,whereas<br />

foreveryfinitesubset Xof Sthereisav ′ �∈ Vsuchthat X = T(v ′ ),hence<br />

itisnota(S)(Q)I-system.<br />

Σ6 is a QI-system for it is determined by the finite set of QI-ons<br />

{〈∅, {A,B}〉, 〈{A,B}, ∅〉}. Butitisnotsaturated,forthesupervaluation<br />

of Vistheemptyvaluation,henceitisnota(P)(S)I-system.<br />

Σ7isanSI-systemforitisdeterminedbythesingleSI-on 〈S \ {A},A〉.<br />

Butitisnotcompact,for Vcontains,foreveryfinitesubset Yof S \ {A},<br />

a v ′ suchthat T(v ′ ) = Yand C(v ′ ) = A.Henceitisnota(P)(Q)I-system.<br />

Σ8isanI-system,foritisdeterminedbytheI-on 〈∅, {B}〉.


InferentializingConsequence 165<br />

5 Consequencerevisited<br />

Ifwhatweareinterestedinistherelationofconsequence,thenourclassificatoryhierarchybecomesexcessivelyfine-grained.Inparticular,weare<br />

goingtoshowthatforevery(P)(S)QI-systemthereexistsa(P)(S)I-system<br />

withthesamerelationofconsequence. Todothisletusdefineaconcept<br />

introducedby(Hardegree,2006):<br />

Definition11.Let Ubeasetofvaluationsoftheclass Sofsentences.<br />

Thesupervaluationof Uisthevaluationsuchthat T(v) = T(U).<br />

ThenextlemmashowsthatourTheorem3isequivalenttooneofHardegree’sresults:<br />

Lemma1.Asemanticsystem 〈S,V 〉isa(P)(S)QI-systemiff V contains<br />

supervaluationsofallitssubsets.<br />

Proof.Thisfollowsdirectlyfromthefactthat 〈S,V 〉isa(P)(S)QI-system<br />

iffitissaturated,foritcanbeeasilyseenthatitissaturatediff Vcontains<br />

supervaluationsofallitssubsets.<br />

Lemma2.Extendingadmissiblevaluationsofasemanticsystembysupervaluationsdoesnotchangetherelationofconsequence.<br />

Proof.Let 〈S,V 〉beasemanticsystemand |=therelationofconsequence<br />

inducedbyit.Let vbeasupervaluationofasubsetof Vandlet |= ∗ bethe<br />

relationofconsequenceinducedby 〈S,V ∪ {v}〉.Supposethetworelations<br />

donotcoincide;thenthereisasubset Xof Sandanelement Aof Sso<br />

that X |= A,butnot X |= ∗ A. Thismeansthatitmustbethecasethat<br />

v(B) = 1forevery B ∈ Xand v(A) = 0,butthatevery v ′ ∈ V suchthat<br />

v ′ (B) = 1forevery B ∈ Xisboundtobesuchthat v ′ (A) = 1.Butas v ′ is<br />

thesupervaluationofan U ⊆ V,elementsof Umapallelementsof Xon 1,<br />

whereasatleastoneofthemmaps Aon 0;whichisacontradiction.<br />

ThisgivesusthefollowingreducedversionofDiagram1:<br />

PS(Q)I-system =semanticsystem<br />

P(Q)I-system S(Q)I-system<br />

(Q)I-system<br />

Diagram2<br />

Hencefromtheviewpointofconsequence,wehavefourtypesofsemantic<br />

systems:


166 JaroslavPeregrin<br />

•SystemsthatareneitherP(Q)I,norS(Q)I.Thesearesystemsofthe<br />

kindof Σ1and Σ3.<br />

•P(Q)I-systemsthatarenot(Q)I-systems.Examplesare Σ2and Σ5.<br />

•S(Q)I-systemsthatarenot(Q)I-systems.Examplesare Σ4and Σ7.<br />

•(Q)I-systems.Systemsofthekindof Σ6and Σ8.<br />

Consequenceasinducedbythetruthtablesofclassicalpropositionallogic<br />

orbythemodeltheoryoftheclassicalfirst-orderpredicatelogic,ofcourse,<br />

fallintothelastcategory.Indeedanylogicthathasastronglysoundand<br />

completeaxiomatizationmusttriviallybelonghere. Butevenamongthe<br />

semanticsystemsstudiedbylogicianstherearesomethatfalloutsidethis<br />

range((Tarski,1936)madethisintoadeeppoint—consequence,according<br />

tohim,cannotbeingeneralcapturedintermsofinferentialrules).<br />

FromDiagram2wecanseethattherearetwowaystogobeyondthe<br />

boundariesofI-systems:wemayeitheralleviatetherequirementoffinitenessofantecedentsofinferences,oralleviatetherequirementoffiniteness<br />

ofthewholerelationofinference. The ω-rule,whichisoftendiscussedin<br />

connectionwiththeformalizationofarithmetic,isanexampleoftheformer<br />

way;theaxiomschemeofinduction,thatcomprisesaninfinityofconcrete<br />

axioms,istheexampleofthelatter.<br />

Foramorespecificexample,considerthelanguageofPeanoarithmetic<br />

withthesingleadmissiblevaluationdeterminedbytheintendedinterpretationwithinthestandardmodel(letmecallthissystemtruearithmetic,TA).Asitturnsout,thissystemisaPQI-system.Indeed,itcanbedeterminedbythePQI-structurewhoserelationofinferenceconsistsoftheI-ons<br />

oftheform 〈∅,A〉foreverytruesentence AplustheQI-onsoftheform<br />

〈{B}, ∅〉foreveryfalsesentence B.(WeknowthatitisnotanI-system,as<br />

weknowthatthetruthsofTAarenotrecursivelyenumerable.) Callthe<br />

singleadmissiblevaluationofthesystem t.<br />

Ifweextendthe(single-element)setofadmissiblevaluationsofTAby<br />

thefullvaluation,itbecomessaturated(indeedthesupervaluationofevery<br />

subsetofthesetofitsadmissibletruthvaluationswillbeadmissible:the<br />

supervaluationoftheemptysetaswellasthesingletonofthefullvaluation<br />

isthefullvaluation,whereasthesupervaluationofthetworemainingsetsis<br />

thevaluation t).HencethissystemisaPI-system(indeed,itisdetermined<br />

bythePI-structuretherelationofinferenceofwhichconsistsoftheI-onsof<br />

theform 〈∅,A〉foreverysentence Atrueaccordingto tplustheI-onsofthe<br />

form 〈{B},C〉foreverysentence Bfalseaccordingto t,andeverysentence<br />

C)buthasthesamerelationofconsequenceastheprevioussystem.


InferentializingConsequence 167<br />

6 Furthersteps<br />

Ihopetohaveshownhowwecansetupausefulframeworkforasystematic<br />

confrontationofprooftheoryandsemantics,especiallyofinferenceand<br />

consequence;andthatIhavealsoindicatedthatthisframeworkletsusprove<br />

somenontrivialandinterestingresults. However,itshouldbeaddedthat<br />

tobringresultsimmediatelyconcerningtheusualsystemsofformallogic,<br />

ourclassificatoryhierarchywillhavetobemadestillmorefine-grained.<br />

Thepointisthatwhileweonlydistinguishedbetweensystemsthatare<br />

determinedbystructureswithafinitenumberof(S)(Q)I-ons(i.e.(S)(Q)Isystems)andthosewherethefinitenessrequirementisalleviated(theP(S)<br />

(Q)I-systems),wewouldneedtoconsidersystemsinbetweenthesetwo<br />

extremes. Theusualsystemsofformallogiccanbeconsideredasgeneralizingoverinferential(asopposedtopseudoinferential)structuresintwo<br />

steps. First,theyallowforaninfinitenumberof(S)(Q)I-ons,whichare,<br />

however,instancesofafinitenumberofschemata.(Thisis,ofcourse,possibleonlywhenwe,unlikeinthepresentpaper,takeintoaccountsomestructuringofthesetofsentencesandconsequentlyofthesentencesthemselves—ifweconsiderthesentencesasgeneratedfromavocabularybya<br />

setofrules.)ThiscanbeaccountedforintermsofparametricSQI-ons,or<br />

p(S)(Q)I-ons.p(S)(Q)I-systems,then,fallinbetween(S)(Q)I-systemsand<br />

P(S)(Q)I-system.ThusforexamplethesemanticsystemofPAisap(Q)Isystem,fortheinfinityofitsaxiomsistheunionofinstancesofafinite<br />

numberofaxiomschemas. ThesemanticsystemofTAisapSI-system,<br />

forweknowthatwecanhaveitssoundandcompleteaxiomatizationifwe<br />

extendtheaxiomaticsystemwiththeomega-rule,whichis,inourterminology,apSI-on.Secondtheyallowforinfinitesetsof(S)(Q)I-onsthatare<br />

generatedbyafinitenumberofmetainferentialrulesfromsetsofinstances<br />

offinitenumberofschemata.<br />

JaroslavPeregrin<br />

DepartmentofLogics,FacultyPhilosophy&Arts,CharlesUniversity<br />

nám.JanaPalacha2,11638Praha1,CzechRepublic<br />

peregrin@ff.cuni.cz<br />

http://jarda.peregrin.cz<br />

References<br />

Gentzen,G.(1934).Untersuchungenüberdaslogischeschliessen1.MathematischeZeitschrift,39,176-210.<br />

Gentzen,G.(1936).Untersuchungenüberdaslogischeschliessen2.MathematischeZeitschrift,41,405-431.


168 JaroslavPeregrin<br />

Hardegree,D.M.(2006).Completenessandsuper-valuations.JournalofPhilosophicalLogic,34,81-95.<br />

Negri,S.,&Plato,J.von.(2001).Structuralprooftheory.Cambridge:Cambridge<br />

UniversityPress.<br />

Peregrin,J. (1997). Languageanditsmodels. NordicJournalofPhilosophical<br />

Logic,2,1-23.<br />

Peregrin,J.(2006).Meaningasaninferentialrole.Erkenntnis,64,1-36.<br />

Tarski,A. (1936). Überdenbegriffderlogischenfolgerung. ActesduCongrés<br />

InternationaldePhilosophiqueScientifique,7,1-11.


Meaning and Compatibility:<br />

Brandom and Carnap on Propositions<br />

Martin Pleitz<br />

1 Brandom’sandCarnap’ssemantics<br />

RobertrandominhisLockeLectures 1 hasdevelopedanewformalsemanticsthatisbasedentirelyontheoneprimitivenotionoftheincoherenceof<br />

setsofsentences(Brandom,2008,pp.117–175). Alanguage,i.e.asetof<br />

atomicsentences,isformallyinterpretedbyanincoherencepartitionofits<br />

power-set.Theincoherencepartitionmustsatisfyonlyoneconditionthat<br />

Brandomcalls“persistence”,i.e.,allsetscontaininganincoherentsubset<br />

mustthemselvesbeincoherent. Theincompatibilityoftwosentencesis<br />

thendefinedastheincoherenceoftheirunion. Twosentencesarecalled<br />

incompatibility-equivalent(or I-equivalentforshort)ifandonlyiftheyare<br />

incompatiblewiththesamesetsofsentences.Inthatcasethetwosentences<br />

aresaidtobesynonymous.Therefore,thepropositionexpressedbyasentencecanberepresentedbytheincompatibility-setofthesentence,i.e.,by<br />

thesetofsetsofsentencesthatareincompatiblewithit(Brandom,2008,<br />

pp.123ff.).Thisformalframeworkisasemanticsbecausethebasicnotion<br />

ofincoherencesufficestogiveanaccountofthemeaningofsentences.<br />

Morethansixtyyearsearlier,RudolfCarnaphadalreadyproposedto<br />

representpropositionsassetsofsetsofsentences,althoughinadifferentway.<br />

LikeBrandom,Carnapbuildsalanguagefromasetofatomicsentences.<br />

Tothislanguageareaddedtheconnectivesofpropositionallogic. 2 Carnap<br />

thendefinesastate-descriptionasasetsuchthatforeachatomicsentence,<br />

eitherthesentenceoritsnegationisanelementofit(Carnap,1947/1956,<br />

1 BrandomdescribesincompatibilitysemanticsatlengthinthefifthofhisJohnLocke<br />

Lectures,whichheheld2006inOxfordand2007inPragueandthathavebeenpublished<br />

as(Brandom,2008).Thebasicideaissketchedfirstin(Brandom,1985),andmentioned<br />

repeatedlyin(Brandom,1994).<br />

2 Theirmeaningisgivenbytheusualtruth-tables.


170 MartinPleitz<br />

p.9). Twosentencesaresaidtobe L-equivalentifandonlyiftheyare<br />

membersofthesamestate-descriptions.Inthatcasethetwosentencesare<br />

saidtobesynonymous.Thus,thepropositionexpressedbyasentencecan<br />

berepresentedbyitsrange,i.e.,bythesetofstate-descriptionsthatitis<br />

anelementof(Carnap,1947/1956,pp.7–32,p.181). 3 Carnapintendshis<br />

state-descriptionsto“representLeibniz’possibleworldsorWittgenstein’s<br />

possiblestatesofaffairs”(Carnap,1947/1956,p.9).Butstate-description<br />

semanticsdifferscruciallyfromthemainstreamofpossibleworldssemantics.<br />

Ingeneral,possibleworldsareseenasbasicobjects. 4 State-descriptions,by<br />

contrast,areset-theoreticalconstructionsfromlinguisticentities.Therefore<br />

state-descriptionsemanticsisaprecursoroftheoriesthatreducepossible<br />

worldstomaximallycompatiblesetsofsentences. 5<br />

Thefactthatstate-descriptionsarereductionistpossibleworldsbrings<br />

outafirstsimilaritybetweenBrandom’sandCarnap’sformalsemantics:<br />

Botharesemanticswithouttheworld. Meaningisnotmodeledasarelationbetweenlanguageandessentiallynon-linguisticobjects.Rather,bothincompatibilitysemanticsandstate-descriptionsemanticsrepresentmeaningbyset-theoreticalconstructionsbuiltfromlinguisticobjectsalone.<br />

6 It<br />

istheaimofmytalktoshowthatthesimilaritiesbetweenBrandom’sand<br />

Carnap’ssemanticsaremorethansuperficial.Ihopethiscomparisonwill<br />

shedsomelightonboththeories.<br />

2 Modifyingstate-descriptionsemantics<br />

Beforestartingthecomparison,afinaladjustmentmustbemadetoCarnap’stheory.Carnapcannotachievehisownaims,becausehistheorydoes<br />

3 Strictlyspeaking,thisistrueonlyforatomicsentences. Fornon-atomicsentences,we<br />

havetolaydowntherecursivedefinitionofwhatitmeansforasentencetoholdin(i.e.to<br />

betrueat)astate-description.<br />

4 Onthispoint,radicalmodalrealistslikeDavidLewis,whoholdsthatpossibleworlds<br />

areconcreteobjects(Lewis,1986,pp.81ff.),andmoderatemodalrealistslikeSaulKripke,<br />

whotakespossibleworldstobestipulated,havetoagree(Kripke,1972/1980,pp.15ff.).<br />

5 ReductionismisendorsedbyRobertAdams,AlvinPlantinga,RobertStalnaker,AndrewRoper,PhillipBrickerandMaxwellCresswell((Adams,1979),(Plantinga,1974),(Stalnaker,1979),(Roper,1982),(Bricker,1987),(Cresswell,2006)).Thecleareststatementofthetheorywasperhapsgivenbyoneofitsopponents:<br />

DavidLewisdescribes<br />

reductionism—thathecalls“linguisticersatzism”—fromacriticalperspective,butin<br />

greatdetail(Lewis,1986,pp.142–165).<br />

6 Asimilaritythatismorethansuperficialconcernsmodality.BrandomandCarnapshare<br />

aglobalunderstandingofmodality,wherethereisnoequivalentofaKripkeanrelation<br />

ofaccessibility,andnecessityconformstotheaxiomsofS5(Brandom,2008,pp.129ff.,<br />

141ff.),(Carnap,1947/1956,pp.10,174f.,186).Forademonstrationthatincompatibility<br />

semanticscanbemodifiedtoaccommodateotherconceptsofnecessity(asitturnsout,<br />

ofB),see(Peregrin,2007),aswellas(Göcke,Pleitz,&vonWulfen,2008)and(Pleitz&<br />

vonWulfen,2008).


MeaningandCompatibility:BrandomandCarnaponPropositions 171<br />

notruleoutstate-descriptionsthatcontainsentencesthatintuitivelyare<br />

incompatible. ThiscanbeshowntoanexampletakenfromMeaningand<br />

Necessity.<br />

Considertheatomicsentences“Scottishuman”and“Scottisarational<br />

animal”.AccordingtoCarnap’sdefinition,therewillbeastate-description<br />

containingthefirstandthenegationofthesecond.Thereforethetwosentenceswillnotbe<br />

L-equivalentandhencenotsynonymous.Buttheyshould!<br />

CarnapstipulatestheEnglishwords“human”and“rationalanimal”to<br />

meanthesame(Carnap,1947/1956,p.4f.). And,onlyafewpageslater,<br />

heexplicitlystatesthatthepredicates“human”and“rationalanimal”are<br />

coextensionalineverystate-description(Carnap,1947/1956,p.15).Thisis<br />

notasmalltechnicalpoint. AsDavidLewishaspointedout,anytheory<br />

thatreducespossibleworldstosetsofsentencesmustrelyonaprimitive<br />

modalnotion,becausethesentencesthatrepresentpossibleworldsmust<br />

becoherentinasensethatexceedsmerelogicalconsistency. Lewisgives<br />

theexampleof“thepositiveandnegativechargeofpointparticles”(Lewis,<br />

1986,p.154).<br />

Asapreliminarysolution,letusmodifystate-descriptionsemanticsby<br />

rulingoutthosestate-descriptionsthatcontainatomicsentencesthatintuitivelyareincompatible.<br />

Thismodificationwillturnouttoplayacrucial<br />

roleinthecomparisonofstate-descriptionsemanticstoincompatibilitysemantics(Sections5–7).Thereforeitisimportanttonotethatthejustificationofthismodificationofstate-descriptionsemanticsisindependentof<br />

theenterpriseofcomparingittoincompatibilitysemantics:Thecriticism<br />

ofCarnap’stextisimmanent,andLewis’sargumentisperfectlygeneral.<br />

3 Mutualsimulation<br />

Withthismodificationinplace(Section2),thecomparisonofBrandom’s<br />

andCarnap’ssemanticscanstart. Wehaveseenthat,inincompatibilitysemantics,incompatibility-setsexplicate<br />

7 propositions,and I-equivalence<br />

explicatessynonymy.Instate-descriptionsemantics,rangesexplicatepropositions,and<br />

L-equivalenceexplicatessynonymy(Section1). Sothereare<br />

twoquestions:Whatistherelationoftheobjectsrepresentingpropositions,<br />

i.e.ofincompatibility-setsandranges? Whatistherelationofthecriteriaofsynonymy,i.e.of<br />

I-equivalenceand L-equivalence?Iwillanswerthe<br />

firstquestionwithaninformalrecipefortransformingincompatibility-sets<br />

intorangesandviceversaandthesecondwithatheoremwhichsaysthat<br />

I-equivalenceand L-equivalencearecoextensional.<br />

BothBrandomandCarnaprepresentpropositionsbysetsofsetsofsentences.<br />

But,atleastsuperficially,rangesandincompatibility-setsdiffer.<br />

7 Forthisuseof“explicate”,cf.(Peregrin,2007,p.13).


172 MartinPleitz<br />

Intuitively,therangeofasentencewillcontainonlysetsofsentencescompatiblewithit,whiletheincompatibility-setofasentencewillcontainonly<br />

incompatiblesetsofsentences. Butnevertheless,wecanmovebackand<br />

forthfreelybetweenincompatibility-setsandranges.Theincompatibilitysetofasentencecanberepresentedbyitscomplement,i.e.thecompatibility-setofthesentence.<br />

Thecompatibility-setinturncanbestratified<br />

intoabundleofmaximallycoherentsetsofsentencescontainingthesentence,whichisthesameastherangeofthesentence.(Stratificationistheprocedureoftakingoutofthecompatibility-setallsetsofsentencescontainedinothersetsofsentencesofthecompatibility-set.)<br />

Totransforma<br />

rangeintoanincompatibility-set,wejusthavetoretracethosesteps.This<br />

recipeforthetransformationofincompatibility-setsandrangesisonlyinformal,becausethecomparedconceptsstemfromdifferenttheories.<br />

As<br />

yet,incompatibility-setsaredefinedonlyinincompatibilitysemanticsand<br />

rangesaredefinedonlyinstate-descriptionsemantics.<br />

Forthesamereason,thetheoremofcoextensionalitycannotyetbe<br />

statedorprovedineithersemantictheory. 8 WefirsthavetodefinesurrogatesoftheCarnapianconceptsinBrandom’ssemanticsandsurrogatesof<br />

theBrandomianconceptsinCarnap’ssemantics(AppendicesA&B).<br />

InBrandom’sincompatibilitysemantics,state-descriptions ∗ canbedefinedasmaximallycoherentsetsofsentences.<br />

9 Thisgivesusadefinitionof<br />

L-equivalence ∗ asmembershipinthesamemaximallycoherentsets. The<br />

theoremofcoextensionalitycannowbestatedandprovedinincompatibilitysemantics.<br />

Itsaysthattwosentencesare L-equivalent ∗ justincase<br />

theyare I-equivalent,i.e.,thattwosentencesareincludedinthesamemaximallycoherentsetsjustincasetheyareincompatiblewiththesamesets<br />

ofsentences(AppendixA).Thishasalreadybeenshowninasimilarway<br />

byJaroslavPeregrin,inhisCommentsonBrandom’sFifthLockeLecture<br />

(Peregrin,2007,p.17f.).<br />

InCarnap’sstate-descriptionsemantics,wecandefineincoherence ∗ and<br />

incompatibility ∗ onthebasisofmembershipinstate-descriptions. Apair<br />

ofsentencesisdefinedasincompatible ∗10 ifandonlyifthereisnostatedescriptionthatcontainsboth.<br />

(Thisconceptofincompatibility ∗ canalreadybefoundunderthenameof“L-exclusiveness”inCarnap’s1942In-<br />

8 Asyet,therelationof I-equivalenceisdefinedonlyinincompatibilitysemanticsandthe<br />

relationof L-equivalenceonlyinstate-descriptionsemantics.<br />

9 Anasterisk( ∗ )indicatesanotionthatisdefinedinaforeignsetting. —State-descriptionscansafelybetreatedasmaximallycoherentsetsbecauseoftheexclusionof<br />

state-descriptionsthatcontainincompatiblesentences(Section2).<br />

10 Thisformalnotionofincompatibility ∗ inmodifiedstate-descriptionsemanticsislinked<br />

toourintuitivenotionofcompatibility,becausewehaveusedtheintuitivenotionof<br />

compatibilitytoruleoutsomeoftheoriginalstate-descriptions. Inunmodified statedescriptionsemantics,thedefinitionofcompatibility<br />

∗ entailsthatanytwoatomicsentencesarecompatible<br />

∗ .


MeaningandCompatibility:BrandomandCarnaponPropositions 173<br />

troductiontoSemantics(Carnap,1942/1961,pp.70,94).) Onthebasis<br />

ofthedefinedBrandomianconcepts,itispossibletostateandprovethe<br />

theoremofcoextensionalityinstate-descriptionsemantics(AppendixB).<br />

Insum,itcanbeprovedbothinincompatibilitysemanticsandinstatedescriptionsemanticsthattwosentencesare<br />

L-equivalentjustincasethey<br />

are I-equivalent.<br />

4 Towardsagenuinecomparison<br />

ItisimportanttoseethatthisdoesnotshowthatBrandom’sandCarnap’s<br />

semanticsareequivalentsimpliciter.Theresultssofararethatinincompatibilitysemantics,<br />

I-equivalenceandadefinednotionof L-equivalence ∗ are<br />

coextensionalandthatinstate-descriptionsemantics, L-equivalenceanda<br />

definednotionof I-equivalence ∗ arecoextensional.Whyisthisnotenough?<br />

Speakingmetaphorically,foracomparisontobegenuineitmustbeconductedfromtheoutsideofbothsemantictheories.<br />

Internalcomparisonis<br />

alwaysindangerofworkingwithpoorsubstitutes.<br />

Asboththeoriestrytogiveformalmodelsofmeaning,Isuggestthatthe<br />

commongroundforagenuinecomparisonisalanguagethatisintuitively<br />

interpreted.Theintuitiveinterpretationofalanguageisgivenbyatranslationofitstermsinto(afragmentof)naturallanguage.<br />

11 Wewilltherefore<br />

havetoworkwithparticularexamplesoflanguages.Onlyifincompatibilitysemanticsandstate-descriptionsemanticsassignthesamerelationsof<br />

synonymyforeveryintuitivelyinterpretedlanguage,willitbeappropriate<br />

tosaythatthetwosemantictheoriesareequivalentsimpliciter.<br />

Therearethreepreconditionsforafaircomparisonofwhatincompatibilitysemanticsandstate-descriptionsemanticsmakeofaparticularintuitively<br />

interpretedlanguage.Firstly,bothsidesshouldhavesharedintuitionsabout<br />

therelationsbetweenbasicmodalconcepts. 12 Secondly,theyshouldhave<br />

11 “Interpretation”isanumbrellatermforanyprocedurethatassociatesmeaningwith<br />

expressionsofalanguage. Here,inthecontextofformallanguages,wecandistinguish<br />

thefollowingsensesof“interpretation”:<br />

1. Interpretation(oftheexpressionsofanyformallanguage)bytranslationintothe<br />

naturallanguageweuse,<br />

2. interpretation(ofthesentencelettersofpropositionallogic)bytruthtables,<br />

3. interpretation(oftheexpressionsofapredicatelogicallanguage)byrangesof<br />

state-descriptions,i.e.interpretationinstate-descriptionsemantics,and<br />

4. interpretation(ofsentenceletters)byanincoherencepartition,i.e.interpretation<br />

inincompatibilitysemantics.<br />

While2,3and4arethemselvespartofformaltheories,1makesuseofourlanguage,and<br />

hencecanbecalled“intuitive”or“pre-theoretical”interpretation.<br />

12 Forthepre-theoreticalconceptsofnecessity,compatibilityandapossibleworld,there<br />

shouldbeagreementaboutprincipleslikethefollowing: Asentenceisnecessaryifand


174 MartinPleitz<br />

sharedintuitionsaboutmodalityinparticularcases,aswell. 13 Thirdly,we<br />

willneedasharedformalframework,inthefollowingsense:Theintuitive<br />

interpretationofaparticularlanguagemust,ineachtheory,uniquelydetermineaformalinterpretationofthelanguage.ForBrandom,theformal<br />

interpretationofalanguageisgivenbyitsincoherencepartition,andfor<br />

Carnap,bythesetofstate-descriptions. 14<br />

5 Filtersonstate-descriptiontables<br />

Toseehowanincoherencepartitionrelatestoasetofstate-descriptions,<br />

letustakeanotherlookatstate-descriptionsemanticsinitsoriginalform.<br />

Thiswillprovideahelpfulcontrasttounderstandformalinterpretationin<br />

modifiedstate-descriptionsemantics.<br />

AccordingtoCarnap’soriginaldefinition,astate-descriptionisaset<br />

ofsentencesthat,foreveryatomicsentence,containseitherthesentence<br />

oritsnegation. TheideabehindthisdefinitiongoesbacktowhatLudwigWittgensteinwroteabouttruth-tablesintheTractatus.<br />

Hestates<br />

thateachrowofatruth-tablerepresentsapossiblestateofaffairs,and<br />

togethertheyrepresentallpossibilities(e.g.,Tractatus,4.2&4.3). The<br />

analogywithtruth-tablesprovidesasystematictechniqueforgivingacompletelistofstate-descriptionsofagivenlanguage,whichwemaycalla<br />

“state-descriptiontable”. Howmustastate-descriptiontablebechanged<br />

toaccommodatemodifiedstate-descriptionsemantics?ThemodificationintroducedintoCarnap’soriginalframeworkrulesoutthosestate-descriptions<br />

containingintuitivelyincompatiblesentences.Thisamountstocrossingout<br />

rowsinthetruth-table-likelistofallpotentialstate-descriptions. Letus<br />

callalistofallrowsthatarecrossedouta“filteronastate-description<br />

table”. 15<br />

onlyifitistrueineverypossibleworld.Twosentencesarecompatibleifandonlyifthere<br />

isapossibleworldwherebotharetrue.Asetofsentencesrepresentsapossibleworldif<br />

andonlyifitismaximallycompatible.—InthecontextofthecomparisonofBrandom’s<br />

andCarnap’ssemantics,itisimportanttorealizethatthejustificationoftheseprinciples<br />

restsinneithertheory,butinoursharedintuitionsaboutmodalityingeneral.<br />

13 E.g.,thehistoricalCarnapmightwelldisagreewithBrandomwhetheritreallyisimpossiblethatablackberryberedandripe(cf.(Brandom,2008,p.123).<br />

Aswewant<br />

tocomparesemantictheories,notparticularopinionsaboutmodality,wewillhaveto<br />

abstractawayfromsuchdisagreements.<br />

14 Weshouldunderstandourintuitionsaboutaparticularlanguageintermsofanintuitivenotionofcompatibility.Intuitivecompatibilitynaturallyleadstoanincoherence<br />

partitionand,atleastaccordingtothemodificationofCarnap’sframework(Section2),<br />

itdetermineswhatstate-descriptionsthereare.<br />

15 ThisnotionseemstobeexactlythesameaswhatPeregrincalls“inadmissiblevaluations”,cf.JaroslavPeregrin,“Inferentializingsemanticsandconsequence”,talkgivenon<br />

June17,2008,atLogica2008inHejnice.


MeaningandCompatibility:BrandomandCarnaponPropositions 175<br />

Themethodoffiltersonstate-descriptiontablescapturestheformalinterpretationofalanguageasgivenbymodifiedstate-descriptionsemantics.So,howdoesthisrelatetoformalinterpretationinincompatibilitysemantics,i.e.toanincoherencepartition?Asanincoherencepartitionobviously<br />

putsspecificrestrictionsontheadmissibledistributionsoftruth-valuesover<br />

theatomicsentences,ituniquelydeterminesafilteronthesetofpotential<br />

state-descriptions.Butdoesafilteruniquelydetermineanincoherencepartition?Theanswertothisquestionisnotobvious,becauseanincoherencepartitiondeterminesthecoherenceofeverysetofsentences,whilethefilterconcernsonlysetsofmaximallength.<br />

16 Butifaccordingtothefiltera<br />

(potential)state-descriptioniscoherent,thenbypersistenceallitssubsets<br />

arecoherent,too.Inparticular,allsubsetsconsistingofunnegatedatomic<br />

sentenceswillbecoherent. 17 Thereforeafiltergivesusallcoherentsetsof<br />

atomicsentencesandthusuniquelydeterminesanincoherencepartition.<br />

Sowehavefoundagenuinesimilarityofincompatibilitysemanticsand<br />

modifiedstate-descriptionsemantics: Fromanintuitiveinterpretationof<br />

aparticularlanguage,wereachanincoherencepartitioninBrandom’ssemanticsandafilteron(potential)state-descriptionsinCarnap’smodified<br />

semantics.Asthereisabijectionbetweenincoherencepartitionsandfilters,<br />

thereisagenuineequivalencebetweenbothkindsofformalinterpretation.<br />

6 Atomicsentences<br />

Nonetheless,theequivalenceofincoherencepartitionsandfiltersonstatedescriptionshingesonsharedintuitionsabouttheconceptofcompatibility.<br />

HeretherearecontrarytendenciesinBrandomianandCarnapiansemantics.<br />

Thisisobviousintheexampleofasimplelanguagewhereallsetsof<br />

sentencesaresaidtobeintuitivelycompatible.Atleastprimafacie,BrandomandCarnapwillmakeentirelydifferentsenseofthislanguage.Carnapwillunderstandcompatibilityaslogicalindependenceandaccordingly<br />

assignthemaximalnumberofdifferentstate-descriptions. Consequently,<br />

therewillbedifferentrangesforallatomicsentences;notwoatomicsentencesaresynonymous.<br />

Brandom,bycontrast,canmakenotmuchsense<br />

ofthislanguage,becauseinincompatibilitysemantics,thecoherenceofall<br />

sentencestrivializesalanguage: Allatomicsentenceswillhavethesame<br />

incompatibility-set,namelytheemptyset.Soallatomicsentenceswillbe<br />

16<br />

Whatismore, unlikethesetsdealtwithbyanincoherencepartition, moststatedescriptionscontainnegatedsentences.<br />

17<br />

Brandomhasshownthattheadditionofthepropositionalconnectivestoincompatibilitysemanticsisconservative(Brandom,2008,p.127).<br />

Wethereforedonotneedto<br />

computetheincompatibility-setsofconjunctionsandnegationstoanswerthequestion<br />

whetherafilteruniquelydeterminesanincoherencepartition.


176 MartinPleitz<br />

I-equivalentandhencesynonymous. 18 WhyareBrandomandCarnapled<br />

tosodifferentresults?<br />

Toclarifythisissue,letustakeanotherhistoricalstepbackwards,to<br />

whatWittgensteinsaysaboutelementarysentences(Elementarsätze)in<br />

theTractatus. Apartfromtheirsyntacticalatomicity,Wittgenstein’selementarysentenceshavethefollowingproperties:<br />

PropertyofBase: Anydistributionoftruth-valuesoverallelementarysentencesfixesthetruthvaluesofallsentencesofthelanguage.<br />

(Tractatus,4.51,5.3)<br />

Wittgenstein’s elementary sentences are logically independent in the<br />

sensethattheyhavethefollowingtwoproperties:<br />

PropertyofNoEntailment: Noelementarysentenceislogicallyentailedbyanyotherelementarysentence.(Tractatus,5.134)<br />

PropertyofGlobalCompatibility: Notwoelementarysentencesare<br />

logicallyincompatible.(Tractatus,4.211)<br />

Wecanaskofeveryformallanguagewhetheritsatomicsentenceshave<br />

thepropertiesofBase,ofNoEntailmentandofGlobalCompatibility.Inthe<br />

caseofpropositionallogic,allthreequestionsmustofcoursebeanswered<br />

positively(that’swhythemethodoftruth-tablesworks). Thedifferences<br />

ofthethreesemantictheoriesindiscussioncannowbeexpressedinthe<br />

followingway:<br />

ThepropertyofBasedoesnothelptodistinguishbetweenthetheories.<br />

Inallthreesystems,sentencesbuiltonlywiththehelpofpropositional<br />

logicarebasedontheatomicsentences,whilesentencescontainingquantifiersormodaloperatorsarenot.<br />

19 ButthepropertiesofNoEntailment<br />

andGlobalCompatibilityprovidedistinctivecriteria: Accordingtooriginalstate-descriptionsemantics,atomicsentenceshaveboththeproperty<br />

18 Theexampleofthesimplelanguage L = {p, q, r}wheretheset {p, q, r}iscoherent,<br />

illustratesthatthenotionsofrange ∗ and L-equivalence ∗ asdefinedinincompatibility<br />

semanticsinsomecasesarepoorsubstitutesforthenotionsofrangeand L-equivalence<br />

ofstate-descriptionsemantics(Section4).In L,therewillbeonlyonestate-description ∗ ,<br />

i.e.onlyonemaximallycoherentset,namely {p, q, r},andaccordinglyonlyonerange ∗ ,<br />

namely {{p, q, r}},Consequently, p, qand rare L-equivalent ∗ .So I-equivalenceand Lequivalence<br />

∗ areindeedcoextensional(Section3).Butallthesame,theexampleofthe<br />

languageofthreecompatiblesentencesintuitivelycontradictstheequivalenceofincompatibilitysemanticsand(evenmodified)state-descriptionsemantics,becausetothesame<br />

intuitivelyinterpretedlanguageCarnapwouldassigneightdifferentstate-descriptionsand<br />

threedifferentranges.<br />

19 ThepropertyofBaseislostinpredicatelogic(Peregrin,1995,ch.5).Inmodalpropositionallogic,thepropertyofBaseislost,aswell.Thetruth-valuesofallatomicsentences<br />

donotingeneraldeterminethetruth-valueofsentencesoftheform“necessarily α”,<br />

becausethemodaloperatorsarenottruth-functional.


MeaningandCompatibility:BrandomandCarnaponPropositions 177<br />

ofNoEntailmentandofGlobalCompatibility. 20 Accordingtomodified<br />

state-descriptionsemantics,atomicsentencesneednothavethepropertyof<br />

GlobalCompatibility(Section2).Andaccordingtoincompatibilitysemantics,atomicsentencesneedneitherhavethepropertyofGlobalCompatibilitynorthepropertyofNoEntailment.Thusthedifferencebetweenmodified<br />

state-descriptionsemanticsandincompatibilitysemanticsthatemergedin<br />

theexampleofthesimplelanguagehingesonthepropertyofNoEntailment.<br />

21<br />

Buteventhislastdifferencecouldbesmoothedoutifstate-description<br />

semanticswasfullymodifiedbygivingupthepropertyofNoEntailment,as<br />

well.Iseethefollowingreasontodothis.Weallowedatomicsentencesto<br />

beincompatiblebecausewewantedtoexcludestate-descriptionscontaining<br />

particlesbearingnegativeandpositivecharge(Section2).Butforasimilar<br />

reasonwemaywanttoexcludestate-descriptionscontainingwhalesthat<br />

arenotmammals(andthelike). Thisamountstorestrictingtheclassof<br />

state-descriptionsfurther,torespectnotonlyintuitiveincompatibilities,<br />

butintuitiveentailments,aswell.<br />

7 Holisticnegationandthesimulationoflogicalindependence<br />

Nonethelessitmayseemahighpricetogiveupallkindsoflogicalindependencebetweenatomicsentences.Soletusturnagaintoincompatibilitysemantics.WhydoesBrandomaccepttheverylowdegreeoflogicalindependence?IseeareasonthatconcernstheholisticcharacterofBrandom’s<br />

semanticsandcanbeexplainedinthespecialcaseofnegation.<br />

Toillustratetheholisticcharacterofnegationinincompatibilitysemanticsletusreturntothesimplelanguageofthreecompatiblesentences(Section6),thatmaybetranslatedas“Carnapisaphilosopher”,“Scottis<br />

aphilosopher”and“Brandomisaphilosopher”. Inthesimplelanguage,<br />

thethreesentencesarevalidandhence,theirnegationsareincoherent.In<br />

otherwords,wecannotcoherentlysaythatCarnapisnotaphilosopher.<br />

Letusnowenlargethesimplelanguagebyaddingthesentences“Carnap<br />

isaflorist”,“Scottisaflorist”and“Brandomisaflorist”. Letusfurthermorestipulate(somewhatcounterfactually)thatitisincompatibleto<br />

beaphilosopherandaflorist. Nowwehavereachedadifferentlanguage<br />

20 Carnap’soriginaldefinitionisthusinfullagreementwithWittgenstein’sideasaboutelementarysentences.Thisisnotsurprising,ashisstate-descriptionsemanticswasinspired<br />

byWittgenstein’sideasaboutlogicalpossibilities.<br />

21 Inincompatibilitysemanticstherearenotonlyincompatibilitiesbetweenatomicsentences,but,accordingtothedefinitionofincompatibility-entailment,theremayberelationsofentailmentbetweenatomicsentences.<br />

Inmodifiedstate-descriptionsemantics<br />

theremaybeincompatibleatomicsentences,butatleastprimafaciethisisnoreasonto<br />

assumethatthereareentailmentsbetweenatomicsentences.


178 MartinPleitz<br />

wheretherearesixdifferentincompatibility-setsforthesixatomicsentences<br />

andtheirnegationsarecoherent.Inotherwords,onlytheadditiontothe<br />

languageofanincompatiblesentencelike“Carnapisaflorist”makesthe<br />

negationof“Carnapisaphilosopher”coherentlyexpressible. Thisisan<br />

exampleoftheholisticcharacterofnegationasdefinedbyBrandom:The<br />

meaningofnon-pdependsonwhatmaybesaidthatisincompatiblewith p.<br />

Anotherthingwenowhaveachievedisthat,inthesix-sentencelanguage,<br />

theoriginalthreesentencesarelogicallyindependentofeachother. From<br />

ourexamplewecanthusreadoffageneralrecipeforsimulatinglogically<br />

independentatomicsentencesinincompatibilitysemantics.Wejusthaveto<br />

startwithanevennumberofatomicsentences,andlaydownthateachconsecutivepairofsentencesisincompatible,butthesetofalleven-numbered<br />

sentencesiscoherent. 22 Thenalleven-numberedsentenceswillbelogically<br />

independent.So,thoughinincompatibilitysemanticsthereisnolanguage<br />

whereallatomicsentencesarelogicallyindependent,therearelanguages<br />

wherehalfofthemare.Inthiscase,wecanalsosimulatestate-descriptions<br />

thatsatisfyCarnap’soriginaldefinition.Themaximallycoherentsetswill<br />

be I-equivalenttothosesetsofsentencesthat,foreachoneofthelogically<br />

independentsentences,containeitherthesentenceoritsnegation.<br />

Thepossibilityofsimulatinglogicallyindependentsentencesandoriginal<br />

state-descriptionshelpstobringouttheimportantdifferencebetweenincompatibilitysemanticsandoriginalstate-descriptionsemantics:Brandom<br />

canmakesenseofalotmorelanguagesthanCarnaporiginallycould.So,<br />

togiveupthelogicalindependenceofatomicsentencesbroadensthescope<br />

offormalsemantics.<br />

8 ThesimilarityofBrandom’sandCarnap’ssemantics<br />

Insum,therearedeepsimilaritiesbetweenBrandom’sincompatibilitysemanticsandCarnap’smodifiedstate-descriptionsemantics.Notonlydoesthetheoremofcoextensionalityholdinboththeories(Section3),butforeveryintuitivelyinterpretedlanguage,wecanreachequivalentformalinterpretations:anincoherencepartitionorafilteronstate-descriptions(Section5).<br />

Thisequivalenceholdsifstate-descriptionsemanticsisfullymodified,i.e.,if<br />

itabandonstherequirementthattherearenoentailmentsbetweenatomic<br />

sentences(Section6).Andtherearegoodreasonstodothis(Sections6&7).<br />

Inordertobeabletospellouttheresultofmycomparisonintheform<br />

ofaslogan,letmeintroducetheconceptofaminimallyincoherentset.Rememberthatamaximallycoherentsetisacoherentsetsuchthat,forevery<br />

sentencethatisnotamemberofit,thesetplusthatsentenceisincoherent.<br />

Aminimallyincoherentsetisthemirror-imageofthis,becauseitisdefined<br />

22 Thesetofallqueer-numberedsentencesmustofcoursebecoherent,aswell.


MeaningandCompatibility:BrandomandCarnaponPropositions 179<br />

asanincoherentsetsuchthat,foreverysentencethatisamemberofit,<br />

thesetminusthatsentenceiscoherent. Becauseofpersistence,acompletelistofminimallyincoherentsetsisanon-redundantwaytospecifyall<br />

incoherentsets—thatis:anincoherencepartition.<br />

WiththehelpofthisconceptIcansumupmyresultinthefollowingway:<br />

Whenweformallyinterpretalanguage,itisoneandthesamethingwhether<br />

wegiveacompletelistofallmaximallycoherentsetsoracompletelistof<br />

allminimallyincoherentsets. Thus,Brandom’sincompatibilitysemantics<br />

amountstothesameasthereductionistversionofpossibleworldssemantics<br />

turnedinside-out.<br />

MartinPleitz<br />

DepartmentofPhilosophy,UniversityofMünster<br />

Domplatz23,D–48143Münster,Germany<br />

martinpleitz@web.de<br />

A Thetheoremofcoextensionalityinincompatibilitysemantics<br />

Definition12.Aset Sofsentencesof Liscalledastate-description ∗ iff S<br />

ismaximallycoherent,i.e.,iffforeverysentence α,either αisamemberof<br />

Sor αisincompatiblewith S.<br />

Twosentences αand βcalled L-equivalent ∗ ifftheyareincludedinthe<br />

samemaximallycoherentsets.<br />

FirstTheoremofCoextensionality.Thesentences αand βare L-equivalent<br />

∗ iff αand βare I-equivalent.<br />

Proof.“⇐”: Let αand βbe I-equivalent. Then,bythedefinitionof Iequivalence,theyareincompatiblewiththesamesetsofsentences(i.e.<br />

Inc(α) = Inc(β)). Butthentheyarecompatiblewiththesamesentences.<br />

Theythereforearecontainedinthesamemaximallycompatiblesetsof<br />

sentences, i.e., inthesamestate-descriptions ∗ andconsequentlyare Lequivalent<br />

∗ .<br />

“⇒”: Let αand βbenot I-equivalent. Thenthereisaset Xsuchthat<br />

X ∪{α}iscoherentwhile X ∪{β}isincoherent.Asforeverycompatibleset<br />

ofsentencesthereisamaximallycompatiblesetofsentencesthatcontains<br />

it,thereisastate-description ∗ Ssuchthat X ∪ {α} ⊆ S,andtherefore<br />

α ∈ S. As Scontains X,itcannotinclude β,becausetheincoherenceof<br />

X ∪ {β}wouldbypersistencetransferto S. Therefore β �∈ S. As α ∈ S<br />

and β �∈ S, αand βarenot L-equivalent ∗ . 23<br />

23 Forasimilarproof,cf.(Peregrin,2007,p.17f.).


180 MartinPleitz<br />

B Thetheoremofcoextensionalityinstate-description<br />

semantics<br />

Definition13.Asetofsentencesisincoherent ∗ iffthereisnostate-descriptionthatitisasubsetof(notethatincoherence<br />

∗ ispersistent).Apair<br />

ofsentencesisincompatible ∗ iffthereisnostate-descriptionthatcontains<br />

both.<br />

Twosentences αand βare I-equivalent ∗ iff αand βareincompatible ∗<br />

withthesamesetsofsentences.<br />

SecondTheoremofCoextensionality.Thesentences αand βare Iequivalent<br />

∗ ifftheyare L-equivalent.<br />

Proof.“⇐”: Let αand βbe L-equivalent. Thentheyareelementsof<br />

thesamestate-descriptions. Ascompatibility ∗ isdefinedbyrecourseto<br />

membershipinstate-descriptions, αand βthereforearecompatible ∗ with<br />

thesamesetsofsentences. Hencetheyareincompatible ∗ withthesame<br />

setsofsentences.So αand βare I-equivalent ∗ .<br />

“⇒”:Let αand βbenot L-equivalent.Thenthereisastate-description<br />

Dsuchthat αbelongsto Dwhile βdoesnotbelongto D.(Orviceversa.<br />

Forreasonsofsymmetryitsufficestodealwithonecase.)Nowlet δbethe<br />

conjunctionthatcompletelydescribes D.Then,accordingtothedefinition<br />

ofcompatibility ∗ , αand δarecompatible ∗ while βand δareincompatible ∗ .<br />

Consequently, αand βarenot I-equivalent ∗ .<br />

References<br />

Adams,R.M.(1979).Theoriesofactuality.Ithaca&London:CornellUniversity<br />

Press.<br />

Brandom,R. (1985). Varietiesofunderstanding. InN.Rescher(Ed.),Reason<br />

andrationalityinnaturalscience.Lanham:UniversityPressofAmerica.<br />

Brandom,R. (1994). Makingitexplicit. Cambridge,MA–London: Harvard<br />

UniversityPress.<br />

Brandom,R.(2008).Betweensayinganddoing:Towardsananalyticpragmatism.<br />

Oxford–NewYork:OxfordUniversityPress.<br />

Bricker,P.(1987).Reducingpossibleworldstolanguage.PhilosophicalStudies,<br />

52,331–355.<br />

Carnap,R. (1942/1961). Introductiontosemantics. Cambridge,MA:Harvard<br />

UniversityPress.<br />

Carnap,R. (1947/1956). Meaningandnecessity. Chicago: TheUniversityof<br />

ChicagoPress.


MeaningandCompatibility:BrandomandCarnaponPropositions 181<br />

Cresswell,M.J.(2006).Frommodaldiscoursetopossibleworlds.StudiaLogica,<br />

82,307–327.<br />

Göcke,B.,Pleitz,M.,&vonWulfen,H.(2008).HowtoKripkeBrandom’snotion<br />

ofnecessity. InB.Prien&D.Schweikard(Eds.),RobertBrandom: Analytic<br />

pragmatist(pp.135–161).Frankfurt:Ontos.<br />

Kripke,S. (1972/1980).Namingandnecessity. Cambridge,MA:HarvardUniversityPress.<br />

Lewis,D.(1986).Onthepluralityofworlds.Oxford–NewYork:Blackwell.<br />

Loux,M.(1979).Thepossibleandtheactual.Ithaca&London:CornellUniversityPress.<br />

Peregrin,J. (1995). Doingworldswithwords.formalsemanticswithoutformal<br />

metaphysics.Dordrecht–Boston–London:Kluwer.<br />

Peregrin, J. (2007). Brandom’s incompatibility semantics. comments<br />

on Brandom’s Locke lecture V. (Retrieved 22.12.2007 from<br />

http://jarda.peregrin.cz/mybibl/mybibl.php)<br />

Plantinga,A.(1974).Thenatureofnecessity.Oxford:Clarendon.<br />

Pleitz,M.,&vonWulfen,H.(2008).Possibleworldsintermsofincompatibility:<br />

AnalternativetoRobertBrandom’sanalysisofnecessity.InM.Peliˇs(Ed.),The<br />

LogicaYearbook2007(pp.119–131).Prague:Filosofia.<br />

Roper,A. (1982). Towardsaneliminativereductionofpossibleworlds. The<br />

PhilosophicalQuarterly,32,45.<br />

Stalnaker,R.C.(1979).Possibleworlds.Ithaca&London:CornellUniversity<br />

Press.<br />

Wittgenstein,L.(1989).Tractatuslogico–philosophicus.InWerkausgabe(Vol.1).<br />

FrankfurtamMain:Suhrkamp.


Inference and Knowledge<br />

Dag Prawitz ∗<br />

Wesometimesacquirenewknowledgebymakinginferences.Thisfactmay<br />

beseenassoobviousthatitsoundsstrangetoputasaproblemhowandwhy<br />

wegetknowledgeinthatway. Nevertheless,thereisnostandardaccount<br />

oftheepistemicsignificanceofvalidinferences.<br />

Forthisdiscussion,Ishallassumethataperson’sknowledgetakesthe<br />

formofajudgementthatshehasgroundsfor–anassumptionrelatedtothe<br />

ideathatapersonknowsthat p,onlyifshehasgoodgroundsforholding<br />

theproposition ptobetrue. Thequestionhowinferencesgiveknowledge<br />

maythenbeput:howmayonegetinpossessionofgroundsforjudgements<br />

bymakinginferences? Onewouldexpectthatthereisaneasyanswerto<br />

thisquestionbyjustreferringtohowtheconceptsinvolvedareunderstood.<br />

But,asIshallargue,thereislittlehopeofansweringthisquestionwhenthe<br />

notionofvalidinferenceisunderstoodinthetraditionalway. Toaccount<br />

fortheepistemicsignificanceofvalidinferences,weseemtoneedanother<br />

approachtowhatitisforaninferencetobevalidandwhatitistomake<br />

aninference.Ishalldescribeonesuchapproach.<br />

Givenavalidargumentoravalidinferencefromajudgement Atoa<br />

judgement B,itmaybepossibleforanagentwhoisalreadyinpossession<br />

ofagroundfor Atousethisinferencetogetagroundfor B,too. But<br />

theagentisnotensuredagroundfor B,justbecauseoftheinferencefrom<br />

Ato Bbeingvalidandtheagentbeinginpossessionofagroundfor A.<br />

Theagentmaysimplybeignorantoftheexistenceofthisvalidinference,<br />

inwhichcaseitsmereexistencedoesnotmakeherjustifiedinmakingthe<br />

judgement B.Aquestionthathastobeansweredisthereforewhatrelation<br />

theagenthastohavetotheinferencetomakeherjustifiedinmakingthe<br />

judgement B.<br />

∗ ManyoftheideaspresentedherewereworkedoutwhileIwasafellowattheInstitute<br />

ofAdvancedStudiesatUniversitàdiBolognainthespringof2007andwerepresented<br />

inlecturesgivenatthePhilosophyDepartmentofUniversitàdiBologna.


184 DagPrawitz<br />

Ishallrestrictmyselfheretodeductiveinferencesandconclusivegrounds,<br />

andwhenspeakingof“inference”and“ground”,Ishallalwaysmeandeductiveinferenceandconclusiveground.<br />

1 Theproblem<br />

Thequestionunderwhatconditionanagent,callher P,getsaground<br />

fortheconclusionofavalidinferencecanbeformulatedmoreexplicitlyas<br />

follows,where,forbrevity,Irestrictmyselftothecasewhenthereisonly<br />

onepremiss:<br />

Giventhat<br />

thereisavalidinference Jfromajudgement Atoajudgement B, (a)<br />

andthat<br />

theagent Phasagroundfor A, (b)<br />

whatfurtherconditionhastobesatisfiedinorderforittobethecasethat<br />

Phasagroundfor B? (d)<br />

Theproblemisthustostateafurthercondition(c)suchthat(a)–(c)<br />

imply(d).Obviously,asalreadyremarked,(a)and(b)alonedonotimply<br />

(d). Hence,weneedtospecifyacondition(c),inotherwordsarelation<br />

betweenanagent P andaninference J,whichdescribeshowtheagent<br />

arrivesatagroundfortheconclusionof J.<br />

2 Afirstattempttofindacondition(c)<br />

Sincethemereexistenceofavalidinference Jto Bfromajudgement A,<br />

forwhichtheagenthasaground,isnotsufficienttogiveheragroundfor<br />

B,onemaythink 1 thattheextraconditionthathastobesatisfiedisthat<br />

theagent Pknowsthattheinference Jfrom Ato Bisvalid (ck)<br />

(thesubscriptkfor‘knowing’).<br />

Butclearlywedonotnormallyestablishthevalidityofaninference<br />

beforeweuseit.Ifitwereanecessaryconditionalwaystodoso,aregress<br />

wouldresult.Theargumentusedtoestablishthevaliditywouldneedsome<br />

inferences,andifthevalidityofthemhadagaintobeestablishedtogive<br />

theargumentanyforce,therewouldbeaneedofyetafurtherargument<br />

andsoon.Unlessthereweresomeinferenceswhosevaliditycouldbeknown<br />

withoutanyargument,wewouldbeinvolvedinanendlessregressoftrying<br />

1 Thisseemstobetakenforgrantedby,e.g.,John(Etchemendy,1990,p.93).


InferenceandKnowledge 185<br />

toestablishthevalidityofinferencesbeforeanycouldbeusedtogeta<br />

groundforitsconclusion,andhencewewouldneverbeabletoacquire<br />

knowledgebyinferences. Atleastincasethevalidityofaninferenceis<br />

definedinamodel-theoreticalway,analogouslytohowlogicalconsequence<br />

isdefined,itcannotbemaintainedthatknowledgeofvalidityisimmediate<br />

anddoesnotrequireanargumenttobeknown.<br />

Iwanttoremarkinpassingthattheregressnotedaboveisdifferentfrom<br />

thewell-knownBolzano–Lewisregress. 2 Thislatterregresscastdoubtsnot<br />

onlyonthenecessityofthecondition(ck)butevenonwhetherthecondition<br />

issufficienttoguaranteetheagentagroundfortheconclusion.Whyshould<br />

wethink(ck)tobesufficient?Presumablybecausehavingagroundforthe<br />

judgement Aandhenceknowingthatthepropositionoccurringin A(i.e.,<br />

theoneaffirmedby A)istrueandknowingthattheinferencefrom Ato<br />

Bisvalidandhencethatitistruthpreserving(inthesensethatifthe<br />

propositionoccurringin Aistruethensoisthepropositionoccurringin<br />

B),theagentcaninfer Bbysimplyapplyingmodusponens.Thisgivesher<br />

agroundfor B,onemaythink.Ifso,thereasonforsayingthattheagent<br />

hasagroundfor B,whensheknowstheinferencetobevalid,seemstobe<br />

thatthereisanotherinferencethantheoriginalonefrom Ato B,namely,<br />

aninferencefromtwopremisses,oneofwhichistheagent’sknowledgeof<br />

thevalidityoftheoriginalinference.Itisbecauseofthisnewinferencethat<br />

theagentisclaimedtogetagroundfor B. Butbythesamereasoning,<br />

whatreallyguaranteestheagenttohaveagroundfor Bisherknowledgeof<br />

thevalidityofthisnewinference.Inotherwords,thereisathirdinference<br />

withthreepremisses,oneofwhichistheagent’sknowledgeofthevalidity<br />

ofthissecondinference,andsoon.<br />

Alreadyfromthefirstregressdiscussedabovewemustconcludethat(ck)<br />

isnottherightconditionthatweareseekingtodescribehowwegenerally<br />

acquireknowledgeorgroundsbyinferences.Atleast,wemustconcludethis<br />

if“knows”in(ck)meanssomethinglikehavingestablishedbyargument.<br />

Onemaysuggestthatthereisanotherconceptofknowledgethatisrelevanthere,forinstanceknowledgebasedonimmediateevidenceorimplicit<br />

knowledgemanifestedinbehaviour,liketheimplicitknowledgeofmeaning<br />

thatMichaelDummetthascalledattentionto.Idonotwanttodenythat<br />

theremaybeanotionofvalidityofinferenceforwhichonecanclarifysuch<br />

aconceptofknowledgesothat(ck)becomesanappropriatecondition.As<br />

alreadysaid,itwouldcertainlyrequireadeparturefromwhatnowseems<br />

tobethedominantunderstandingofthevalidityofaninferenceinterms<br />

oftruthpreservationforallvariationsofthemeaningofthenon-logicalexpressionsinvolvedintheinference.Anyway,lackingaconceptofknowledge<br />

2 (Bolzano,1837)and(Carroll,1895). Ihavediscussedthisregressmorethoroughlyin<br />

(Prawitz,2009).


186 DagPrawitz<br />

(andofvalidity)thatmakes(ck)appropriate—todeveloponwouldbea<br />

majortask—Ishallnowleavethisfirstattempttofindacondition(c).<br />

3 Asecondattempttofindacondition(c)<br />

Realizingthat(ck)isnottherightcondition,onemayseetheproposalofit<br />

asanoverreactiontothesimpleobservationfirstmade,viz.thatanagent<br />

needtostandinsomerelationtotheinference,ifitistoprovideherwitha<br />

groundforitsconclusion.Ofcourse,themereexistenceofavalidinference<br />

cannotautomaticallyprovidetheagentwithagroundfortheconclusion,<br />

onemaysay.Shehastodosomething.Butshedoesnotneedtoestablish<br />

thevalidityoftheinference.Allthatisneededisthatsheactuallyusesthe<br />

inference.Thenthevalidityoftheinferencedoesprovidetheagentwitha<br />

groundfortheconclusion,giventhatshealreadyhasoneforthepremiss.<br />

Onemaythussuggestthattheconditionsoughtforshouldsimplybe<br />

Pmakestheinference J,thatis,infers Bfrom A. (c)<br />

Thereisclearlysomethingrightinthissuggestion. Oneshoulddistinguishbetweenaninferenceact,andaninferenceinthesenseofanargument<br />

determinedbyanumberofpremissesandaconclusion.Itisfirstwhenan<br />

agentmakesaninference,i.e.,carriesoutaninferenceact,thatthequestionariseswhethersheisjustifiedinmakingtheassertionthatoccursas<br />

conclusionoftheinference.<br />

Butitmustthenbeaskedwhatismeantbymakinganinferenceor<br />

byinferringaconclusion Bfromapremiss A. Weusuallyannouncethe<br />

resultofsuchanactverballybysimplyfirstmakingtheassertion A,then<br />

saying“hence”or“therefore” B,or,inthereverseorder,wefirstmake<br />

theassertion B,andthensay“since”or“because” A. Aninferenceact,<br />

lookeduponasverbalbehaviour,canbeseenasakindofcomplexspeech<br />

actinwhichwedonotonlymakeanassertionbutalsogiveareasonfor<br />

theassertionintheformofanotherassertionorsomeotherassertionsfrom<br />

whichitis(implicitlyorexplicitly)claimedtofollow.<br />

However,ifthisisallthatismeantbyinferringaconclusionfroma<br />

premiss,thenonecannotexpectthatconditions(a),(b),and(c)together<br />

withreasonableexplicationsofthenotionsinvolvedaresufficienttoimply<br />

thatthepersoninquestionhasagroundfortheconclusion B.Toseethis,<br />

onemayconsiderascenariowhereapersonannouncesaninferenceinthe<br />

waydescribed,sayasastepinaproof,butisnotabletodefendtheinference<br />

whenitischallenged.Suchcasesoccuractually,andthepersonmaythen<br />

havetowithdrawtheinference,althoughnocounterexamplemayhavebeen<br />

given.Ifitlaterturnsoutthattheinferenceisinfactvalid,perhapsbya<br />

longandcomplicatedargument,thepersonwillstillnotbeconsideredto


InferenceandKnowledge 187<br />

havehadagroundfortheconclusionatthetimewhensheassertedit,and<br />

theproofthatsheofferedwillstillbeconsideredtohavehadagapatthat<br />

time. Thiswouldbeasituationinwhichconditions(a),(b)and(c)were<br />

allsatisfied,but(d)wouldnotbesaidtohold.<br />

Iftheinferencefrom Ato Bisgenerallyrecognizedasvalid,then,sociologicallyspeakingsotosay,anagentwhohasagroundfor<br />

Aandmakes<br />

theassertion B,giving Aasherreason,willcertainlybeconsideredtohave<br />

agroundforherassertion. Ifinsteadtheinferenceisnotobviouslyvalid<br />

eventoexperts,theagentisnotconsideredtohaveobtainedagroundfor<br />

Bbecauseofmakingtheassertion Bandgiving Aasherreason.Butwe<br />

areofcoursenotsatisfiedwithasociologicaldescriptionofwhenanagentis<br />

consideredtohaveaground(obtained,e.g.,byaddingasaconditionthat<br />

thevalidityoftheinferenceshouldbegenerallyrecognizedbyexpertsin<br />

thefield).<br />

Itthusstillremainstostatetheappropriateconditionunderwhichavalid<br />

inferencegivesanagentagroundfortheconclusionofaninference.Onemay<br />

thinkthatitmustbebasicallyrightthatwegetagroundforajudgement<br />

byinferringitformotherjudgementsforwhichwealreadyhavegrounds,<br />

andthathencecondition(c)isrightlystatedasabove.Butthen“toinfer”<br />

or“tomakeaninference”mustmeansomethingmorethanjuststatinga<br />

conclusionandgivingpremissesasreasons.Thebasicintuitionis,Ithink,<br />

thattoinferisto“see”thatthepropositionoccurringintheconclusionmust<br />

betruegiventhatthepropositionsoccurringinthepremissesaretrue,and<br />

theproblemishowtogetagripofthismetaphoricuseof“see”.<br />

4 Thenatureoftheproblem<br />

Atthispointitmaybegoodtopauseandconsiderinmoredetailthenature<br />

oftheproblemthatIhaveposed.Ihaveusedthetermgroundinconnection<br />

withjudgementstohaveanameonwhatapersonneedstobeinpossession<br />

ofinorderthatherjudgementistobejustifiedorcountasknowledge,<br />

followingthePlatonicideathattrueopinionsdonotcountasknowledge<br />

unlessonehasgroundsforthem.ThegeneralproblemthatIhaveposedis<br />

howinferencesmaygiveussuchgrounds.<br />

AsIusethetermground,aperson’sjudgementisjustifiedorcountsas<br />

knowledgewhensheinpossessionofagroundforthejudgement. Consequently,onedoesnotneedtoshowthatoneisinpossessionofagroundfor<br />

ajudgmentinordertobejustifiedinmakingthejudgement,itisenough<br />

thatinfactoneisinpossessionofsuchaground.Justificationsmustend<br />

somewhere,asWittgensteinputsit.Andthepointwheretheymustendis<br />

exactlywhenonehasgotinpossessionofwhatcountsasajustificationsor<br />

aground;somethingwouldbewronglycalledground,ifitwasnotenough


188 DagPrawitz<br />

thatonetohadgotinpossessionofit,inotherwords,iftherewasyet<br />

somethingtobeshown,inorderthatone’sjudgementwastobeconsidered<br />

justified.<br />

Hence,itisnottheagentPwhohastostateanadequatecondition<br />

(c)andshowthat(d)holds,i.e.,thatshehasagroundfor B,whenthe<br />

conditions(a)–(c)aresatisfied;asjustsaid,sheisjustifiedwhensheisin<br />

possessionofaground,regardlessofwhatshecanshowaboutit.Itisweas<br />

philosopherswhohavetostateanadequatecondition(c)andthenderive<br />

(d)from(a)–(c)togiveanaccountoftheepistemicsignificanceofvalid<br />

inferences. Thepointofmakinginferencesistoacquireknowledge,and<br />

philosophyoflogicwouldnotbeuptoitstask,ifitcouldnotexplainhow<br />

thiscomesabout. Toexplainthisistosayunderwhatconditionsavalid<br />

inferencecansupplyuswithgrounds.<br />

Sincethefactthatweacquireknowledgebymakinginferenceissucha<br />

basicfeatureoflogic,oneshouldexpecttheaccountofthisfacttobequite<br />

simple,oncewehaveunderstoodrightlythekeyconceptsinvolvedhere,in<br />

particularthenotionsvalidinference,inferringormakinganinference,and<br />

ground.Whenthesenotionshavebeenexplicatedappropriately,oneshould<br />

expectittobeasimpleconceptualtruththat(a)–(c)imply(d).<br />

Whatissurprisingisthatthereisnogenerallyacceptedaccountofthe<br />

epistemicsignificanceofinferencesandthatpuzzlingproblemsseemtoarise<br />

whensuchanaccountisattempted. Thisisasignthatourusualunderstandingofthekeyconceptsinvolvedisfaulty.<br />

5 Logicalconsequenceandlogicallyvalidinference<br />

Sinceitisvalidinferencesthatallowepistemicprogress,acrucialingredient<br />

intheaccountmustbetogivethatnotionanadequatemeaning. The<br />

conceptofvalidinferenceistraditionallyconnectedwiththatoflogical<br />

consequenceandwithnecessarytruth-preservation.Oftenonesimplysays<br />

thataninferenceisvalidifandonlyiftheconclusionisalogicalconsequence<br />

ofthepremisses,whichinturnisequatedwithitbeingnecessarilythe<br />

casethattheconclusionistrueifallpremissesare. However,Ihavebeen<br />

followingFregeintakingthepremissesandtheconclusionofanactof<br />

inferencetobespeechactsinwhichapropositionisjudgedtobetrue,hence<br />

takingthepremissesandconclusionofaninferencetobejudgements. 3 The<br />

traditionalideaofinferenceasnecessarilytruthpreservingisthenbetter<br />

formulatedbysayingthataninferenceisvalidifandonlyitnecessarilyholds<br />

3 Inmorerecenttime,thepointthatpremissesandconclusionsarenotpropositionsbut<br />

judgementsorassertionshasespeciallybeenemphasizedbyPerMartin-Löf(Martin-Löf,<br />

1985);seealsoGöran(Sundholm,1998).IncontrasttoFregeandMartin-Löf,however,<br />

Ishallalsoconsiderthecasewhenthepremissesandconclusionarejudgementsmade<br />

underassumptions.


InferenceandKnowledge 189<br />

thatwhenallthepropositionsaffirmedinthepremissesaretrue,thensois<br />

thepropositionaffirmedintheconclusion.<br />

SinceAlfredTarski’s(1936)revivalofBernardBolzano’s(1837)definitionoflogicalconsequence,ithasbeencommontointerpretthemodal<br />

notionofnecessityinthiscontextextensionally,sayingineffect,asweall<br />

know,thataproposition(orsentence) BisalogicalconsequenceofsetaΓ<br />

ofpropositions(orsentences)ifandonlyifforallvariationsofthecontent<br />

ofthenon-logicalnotionsoccurringin Bandintheelementsof Γ,itisin<br />

factthecasethat Bistruewhenalltheelementsof Γare.<br />

Howdoesthevalidityofaninferencecontribute,togetherwiththeother<br />

twoconditions(b)and(c),totheagentbeinginpossessionofagroundfor<br />

theconclusion? Thisisthecrucialquestionthatanyproposednotionof<br />

validityhastoface.Inparticular,whyshouldthefactthataninferenceis<br />

truth-preservingcontributetoourgettingagroundfortheconclusion?As<br />

alreadynoted,anagent’sknowledgethataninferenceistruthpreserving<br />

wouldcontributetohergettingagroundfortheconclusionoftheinference,<br />

butsuchknowledgeshouldnotpresumed,andthefact thatitistruth<br />

preservingisirrelevant.<br />

Now,nobodysuggeststhatthevalidityofaninferenceistobedefinedin<br />

termsofjusttruthpreservation.FollowingBolzanoandTarski,themodeltheoreticaldefinitionsaysthataninferenceisvalidifitistruthpreserving<br />

regardlessofhowthecontentsofnon-logicalexpressionsarevaried. But<br />

thisdoesnotessentiallychangethesituation. Whyshouldtheadditional<br />

factthattheinferenceistruth-preservingalsowhenthecontentofthenonlogicalexpressionsisvariedberelevanttoquestionwhethertheagentcan<br />

seethatthepropositionaffirmedintheactualconclusion(wherethecontent<br />

isnotvaried)tobetrue?Thesamecanbesaidofvaliditydefinedinterms<br />

ofnecessarytruthpreservation,ifthenecessityisunderstoodontologically.<br />

Whyshouldthefactthataninferenceistruthpreservinginotherpossible<br />

worldshelptheagenttoseethatthepropositionaffirmedintheconclusion<br />

istrueintheactualworld?Itisdifficulttoseehowanythingbutknowledge<br />

ofthisfactcouldberelevanthere(andifknowledgeisassumed,itissufficienttoknowthatthetruthofthepropositionsinthepremissesmaterially<br />

impliesthetruthofthepropositionintheconclusion—i.e.,novariation<br />

ofcontentisneeded).Therefore,thereseemstobelittlehopethatonecan<br />

findanappropriatecondition(c)whenvalidityofinferenceisdefinedinthe<br />

traditionalway.<br />

Itisdifferentifthenecessityisunderstoodepistemicallyandthisistaken<br />

tomeanthatthetruthofthepropositionsassertedbythepremisses,which<br />

theagentisassumedtoknow,somehowguaranteesthattheagentcansee<br />

thatthepropositionassertedintheconclusionistrue. Suchanepistemic<br />

necessitycomesclosetoAristotle’sdefinitionofasyllogismasanargument


190 DagPrawitz<br />

where“certainthingsbeinglaiddownsomethingfollowsofnecessityfrom<br />

them,i.e.,becauseofthemwithoutanyfurthertermbeingneededtojustify<br />

theconclusion.” 4 Itisofcourserighttosaythatthereisanepistemic<br />

tiebetweenpremissesandconclusioninavalidinference—somekindof<br />

thoughtnecessity,wecouldsay,thankstowhichtheconclusioncanbecome<br />

justified. Buttosaythisisnottogomuchbeyondourstartingpoint. It<br />

stillremainstosayhowthejustificationcomesabout.<br />

AlthoughtheideaofBolzanoandTarskitovarythecontentofnonlogicaltermsdoesnothelpusindefiningthevalidityofinference,thisideaisstillusefulfordefininglogicalconsequenceandthelogicalvalidityofinference.Oneimportantingredientinourintuitiveideaoflogicalconsequence<br />

andlogicalvalidityofinferenceis,Ithink,thattheyaretopicneutral,<br />

andonenaturalwaytoexpressthisistosaythattheyareinvariantunder<br />

variationsofnon-logicalnotions.<br />

Isuggestthatwedistinguishbetweenlogicalconsequenceanddeductive<br />

(oranalytic)consequenceandsimilarlybetweenaninferencebeinglogically<br />

validanditbeingonly(deductively)valid.Giventhelatterconceptwecan<br />

easilydefinelogicalvalidityinthestyleofBolzanoandTarski:<br />

Aninference Jislogicallyvalid,ifandonlyif,foranyvariationof<br />

thecontentofthenon-logicaltermsoccurringinthepremissesand<br />

conclusionof J,theresultinginferenceisvalid.<br />

Thevariationofcontentmaybeproducedbymakingsubstitutionsforthe<br />

non-logicaltermsinthemannerofBolzanoorbyconsideringassignments<br />

ofvaluestotheminthemannerofTarski;wedonotneedtogointothese<br />

technicaldetailshere.<br />

Thisdefinitionmakesjusticetotheideathatwhetheraninferenceislogicallyvalidisindependentofthemeaningofthenon-logicalterms.However,<br />

thelogicalvalidityofaninferenceisnownotreducedtotruthpreservation<br />

ortothetruthofageneralizedmaterialimplicationbuttothevalidityof<br />

aninferenceundervariationsofthecontentsofnon-logicalterms.<br />

Someinferencesarelogicallyvalid,inadditiontobeingdeductivelyvalid,<br />

andthisisaninterestingfeatureofthem,butitisnotafeatureonwhichthe<br />

conclusivenessoftheinferencehinges.Itthusremainstoanalysedeductive<br />

validityandbringouthowsuchaninferencemaydeliveragroundforits<br />

conclusion.<br />

6 Grounds<br />

Rationaljudgementsandsincereassertionsaresupposedtobemadeon<br />

goodgrounds. Itisnotthatanassertionisusuallyaccompaniedbythe<br />

4 See(Ross,1949,p.287).


InferenceandKnowledge 191<br />

statementofagroundforit;inotherwords,thespeakeroftenkeepsher<br />

groundforherself.Butiftheassertionischallenged,thespeakerisexpected<br />

tobeabletostateagroundforit.Tohaveagroundisthustobeinastate<br />

ofmindthatcanmanifestitselfverbally.<br />

Iamhereinterestedingroundsthatareobtainedbymakinginferences;<br />

allgroundscanofcoursenotbeobtainedinthisway,andIshallsoonreturn<br />

tosomeexamples.Whenanassertionisjustifiedbywayofaninference,itis<br />

commontoindicatethisbysimplystatingtheinferenceinthewaydiscussed<br />

above,andthepremissesoftheinferencesarethenoftencalledtheground<br />

fortheassertion.Thiswayofspeakingmaybeacceptableinaneveryday<br />

context,butitconcealstheproblemthatwearedealingwith,whichis<br />

probablyonereasonwhytheproblemhasbeensoneglected. Itmakesit<br />

seemasifoneautomaticallyhasagroundforaconclusionbyjuststatingan<br />

inferencethatinfacthappenstobevalid—ineffect,itseemsasonemay<br />

getagroundbysimplystatingthatonehasone.Wehavediscussedabove<br />

(Section3)whyagroundforaconclusionisnotforthcomingby“inferring”<br />

itinthissuperficialsense.<br />

Buttherearealsootherreasonswhyitisnotagoodterminologyto<br />

usetheterm“ground”forthepremissesofaninference.Thepremissesare<br />

judgementsorassertionsaffirmingpropositions,andthefactthatonehas<br />

judgedorassertedthemastruecannotconstituteagroundfortheconclusion,norcanthetruthofthepropositionsaffirmedconstitutesuchaground;atleastnotinthesenseofsomethingthatanagentisinpossessionof,therebybecomingjustifiedinmakingtheassertionexpressedinthe<br />

conclusion. Itisratherthefact,ifitisafact,thattheagenthasgrounds<br />

forthepremissesthatisrelevantforherhavingagroundfortheassertion<br />

madeintheconclusion.Butthegroundsforthepremissesaregroundsfor<br />

them,notfortheconclusion. ThequestionthatIhaveposedistherefore<br />

putintheform:giventhegroundsforthepremisses,howdoesonegetfrom<br />

themagroundfortheconclusion?<br />

Weareusedtomeetchallengesofaninferencebybreakingitdowninto<br />

simplersteps,andwhenonesucceedstoreplacetheinferencebyachainof<br />

sufficientlysimpleinferencesthereisinpracticenomorechallenges. But<br />

thephilosophicallyinterestingquestionishowonecanmeetachallengeof<br />

asimpleinferencethatisnotpossibletobreakdownintosimplersteps.It<br />

istemptingtofallbackatthatpointonwhatourexpressionsmeanorin<br />

otherwordsonwhatpropositionsitisthatweaffirmtobetrue.However,<br />

tomymind,itwouldbedubioustosayofalltheseinferencesthatwewant<br />

todefendbutcannotbreakdownintosimplerinferencesthattheirvalidity<br />

isjustconstitutiveforthemeaningofthesentencesinvolved. 5 Thelinethat<br />

5 Insomepreviousworks(e.g.,(Prawitz,1977,1973);cfalsofootnote6)Ihaveidentified<br />

agroundforajudgementwithaproofofthejudgement,orIhavespokenofgroundsfor


192 DagPrawitz<br />

Ishalltakeisinsteadroughlythatthemeaningofasentenceisdetermined<br />

bywhatcountsasagroundforthejudgementexpressedbythesentence.<br />

Orexpressedlesslinguistically:itisconstitutiveforapropositionwhatcan<br />

serveasagroundforjudgingthepropositiontobetrue. Fromthispoint<br />

ofviewIshallspecifyforeachcompoundformofpropositionexpressible<br />

infirstorderlanguageswhatconstitutesagroundforanaffirmationofa<br />

propositionofthatform.Ifonedoesnotlikethislineofapproach,onemay<br />

anywayagreewithmyspecificationofwhatconstitutesagroundforvarious<br />

judgements,whichiswhatmattershere.<br />

Forinstance,aconjunction p&qwillherebeunderstoodasaproposition<br />

suchthatagroundforjudgingittobetrueisformedbybringingtogether<br />

twogroundsforaffirmingthetwopropositions pand q. Wemayputthe<br />

nameconjunctiongrounding,abbreviated &G,onthisoperationofbringing<br />

togethertwosuchgroundssoastogetagroundforaffirmingaconjunction.<br />

Ifwedonotwanttotakethisviewofconjunctions,wemaystillagreefor<br />

otherreasonsthatthereisanoperation &Gsuchthatif βisagroundfor<br />

affirming pand γisagroundforaffirming q,then &G(β,γ)isaground<br />

foraffirming p&q. Similarly,wemaytakeitasafurtherconstitutivefact<br />

aboutconjunctionthatconverselyanygroundforjudgingittobetrueis<br />

formedbytheoperationofconjunctiongroundingorjustagreetothatfor<br />

otherreasons. Whatmattershereisthatthereissuchanoperation &G<br />

suchthatsomethingisagroundforjudging p&qtobetrueifandonlyifit<br />

canbeformedbyapplying &Gtotwogroundsforjudging ptobetrueand<br />

judging qtobetrue,respectively.<br />

Ihavespokenprimarilyofgroundsforjudgementsorassertions. But<br />

forbrevity,wemayalsospeakderivativelyofagroundforaproposition p<br />

meaningagroundforthejudgementorassertionthat pistrue. Wecan<br />

thusstatetheequivalence<br />

αisagroundfortheconjunction p&qifandonlyif α = &G(β,γ)<br />

forsome βand γsuch βisagroundfor pand γisagroundfor q.<br />

Inferencesaremadenotonlyfrompremissesthathavebeenestablishedas<br />

holdingbutalsofromassumptionsandpremissesthatareestablishedunder<br />

assumptions.TocoversuchcasesIshallintroducewhatIshallcallopenor<br />

unsaturatedgroundsbesidesthegroundsthatwehavetalkedaboutsofar<br />

andthatIshallcallclosedgrounds.Bothclosedgroundsandunsaturated<br />

groundswillbesaidtobegrounds.<br />

Anunsaturatedgroundislikeafunctionandisgivenwithanumber<br />

ofopenargumentplacesthathavetobefilledinorsaturatedbyclosed<br />

sentencesandhavetakenthemtobevalidarguments.Iprefernottousethatterminology<br />

now,becauseIwanttotakeproofstobebuiltupbyinferences,andIdonotwanttosay<br />

thataninferenceconstitutesagroundforitsconclusion—thequestionisinsteadhow<br />

aninferencecandeliveragroundfortheconclusion.


InferenceandKnowledge 193<br />

groundssoastobecomeaclosedground. Somethingisagroundforan<br />

assertionof Aundertheassumptions A1,A2,... ,Anifandonlyifitis<br />

an n-aryunsaturatedgroundthatbecomesaclosedgroundforAwhen<br />

saturatedbyclosedgroundsfor A1,A2,... ,An.Writing α(ξ1,ξ2,... ,ξn)for<br />

theunsaturatedgroundand α(β1,β2,... ,βn)fortheresultofsaturatingit<br />

byclosedgrounds βifor Ai,theconditionfor α(ξ1,ξ2,... ,ξn)tobeaground<br />

for Aundertheassumptions A1,A2,... ,Anisthusthat α(β1,β2,... ,βn)is<br />

aclosedgroundfor A.<br />

Groundsarenaturallytypedbythepropositionstheyaregroundsfor.<br />

Theopenplacesinanunsaturatedground,inotherwords,thevariablesused<br />

indisplayingtheunsaturatedground,maythenalsobetypedtoindicate<br />

thetypeofthegroundsthatcansaturatethematthatplace,inotherwords,<br />

thatcanreplacethevariables. Ishallusuallysupplyvariableswithtypes<br />

butshallotherwiseomittypeindications.<br />

Withthesenotionsathand,wecanspecifythataclosedgroundforan<br />

implication p → qissomethingthatisformedbyanoperationthatwecan<br />

callimplicationgrounding, → G,appliedtoa1-aryunsaturatedground<br />

β(ξ p )forjudging qtobetrueundertheassumptionthat pistrue.Theresultofapplyingthisoperationtotheopenground<br />

β(ξ p ),whichIshallwrite<br />

→ Gξ p (β(ξ p )),yieldsthusaclosedgroundfor p → q;itcorrespondsonthe<br />

syntacticalleveltoavariablebindingoperator,andIindicatethisbywriting<br />

thevariable ξ p behindtheoperator.Ifitisappliedtoan n-aryunsaturated<br />

groundfor Aundertheassumptions A1,A2,... ,Anwritten α(ξ1,ξ2,...,ξn),<br />

Ishallwrite → Gξi(α(ξ1,ξ2,...,ξn))toindicatethatitisthe i-thplacein<br />

theunsaturatedgroundthatbecomesbound,whichthendenotesanunsaturatedgroundfor<br />

Aundertheassumptions A1,A2,... ,Ai−1,Ai+1,...,An.<br />

Wehavethustheequivalence<br />

αisagroundfor p → qifandonlyif α = → Gξ p (β(ξ p ))<br />

where β(ξ p )isanunsaturatedgroundforjudgingthat qistrue<br />

undertheassumptionthat pistrue.<br />

Finallywehavetopayattentiontothefactthatthepremissesofaninferencemaybeanopenjudgement<br />

A(x1,x2,... ,xm)(possiblyundersome<br />

openassumptions),bywhichImeanthatitskernelisnotaproposition,but<br />

apropositionalfunction p(x1,x2,...,xm)definedforadomainofindividualssuchthatforany<br />

n-tupleofindividuals a1,a2,... ,am, A(a1,a2,... ,am)<br />

isthejudgementthataffirms p(a1,a2,... ,am).Wemustthereforeconsider<br />

unsaturatedgroundsthatareunsaturatednotonlywithrespecttogrounds<br />

butalsowithrespecttoindividualsthatcanappearasargumentsinpropositionalfunctions.Let<br />

A(x1,x2,...,xm)and Ai(x1,x2,... ,xm)beassertions<br />

whosepropositionalkernelsarepropositionalfunctionsover x1,x2,... ,xm,<br />

andlet A(a1,a2,... ,am)and Ai(a1,a2,... ,am)betheassertionsthatarise


194 DagPrawitz<br />

whenweapplythecorrespondingpropositionalfunctionstotheindividuals<br />

a1,a2,...,am. ThenIshallsaythatsomethingisanunsaturated<br />

groundfortheopenjudgement A(x1,x2,...,xm)undertheassumptions<br />

A1(x1,x2,...,xm), A2(x1,x2,...,xm),... ,An(x1,x2,... ,xm)ifandonlyif<br />

itisanunsaturatedground α(ξ1,ξ2,...,ξn,x1,x2,... ,xm)withrespectto n<br />

closedgroundsand mindividualssuchthatwhensaturatedbytheindividuals<br />

a1,a2,... ,amitbecomesanunsaturatedgroundfor A(a1,a2,... ,am)<br />

undertheassumptions A1(a1,a2,... ,am),A2(a1,a2,...,am),... ,An(a1,a2,<br />

...,am).<br />

Wecanthenspecifythataclosedgroundforageneralizedproposition<br />

∀xp(x)issomethingthatisformedbyanoperationthatIshallcalluniversal<br />

grounding, ∀G,appliedtoanunsaturatedground α(x)forthepropositional<br />

function p(x). Theresultofapplyingthisoperationtotheopenground<br />

α(x),whichIshallwrite ∀Gx(α(x)),againindicatingthat xbecomesbound<br />

bywritingitbehindtheoperator,isthusaclosedgroundfor ∀xp(x).We<br />

havethustheequivalence<br />

αisagroundfor ∀xp(x)ifandonlyif α = ∀Gx(β(x)<br />

where β(x)isanunsaturatedgroundfor p(x).<br />

Ifweidentifynegatedpropositions, ¬p,with (p → ⊥)where ⊥isa<br />

constantforfalsehood,forwhichitisspecifiedthatthereisnogroundfor ⊥,<br />

wehavespecifiedbyrecursionwhatcanbeagroundforsufficientlymany<br />

formsofpropositionsexpressibleinclassicalfirstorderlanguages,except<br />

thatwehavesaidnothingaboutgroundsforatomicpropositions. What<br />

theyarewillofcoursevarywiththecontentoftheatomicpropositions.<br />

InthelanguageoffirstorderPeanoarithmeticwemaytakeagroundfor<br />

anidentitybetweentwonumericalterms t = utobeacalculationofthe<br />

valueof tand ushowingthattheyarethesame.Alternatively,ifwewantto<br />

analyseacalculationasconsistingofstepseachofwhichhasaground,we<br />

needtostartfrommorebasicgrounds.Asalreadysaid,allgroundscannot<br />

beobtainedbyinferences.Theremustinotherwordsbesomepropositions<br />

like t = tor‘0isanaturalnumber’forwhichitisconstitutivethatthereare<br />

specificgroundsforthemthatarenotderivedorbuiltupfromsomething<br />

else.<br />

Outsideofmathematics,wemayconsiderobservationstatements,and<br />

forthem,Isuggest,wetakerelevantverifyingobservationstoconstitute<br />

grounds. Forinstance,agroundforaproposition‘itisraining’istaken<br />

toconsistinseeingthatitrains;taking“seeing”inaveridicalsense,it<br />

constitutesaconclusiveground.Itdoesnotseemunreasonabletosaythat<br />

toknowwhatpropositionisexpressedby“itisraining”istoknow,orat<br />

leastimpliesthatoneknows,howitlookswhenitisraining,andhencethat<br />

oneknowswhatconstitutesagroundforthestatement.


InferenceandKnowledge 195<br />

Inthecaseofintuitionisticpredicatelogicwehavetosayinadditionwhat<br />

countsasagroundfordisjunctionsandexistentialpropositions,whichcan<br />

bedoneinanobviouswayanalogouslytothecaseofconjunction(butwhich<br />

becomestoorestrictivewhendisjunctionsandexistentialpropositionsare<br />

understoodclassically—theseformshavetheninsteadtobedefinedin<br />

termsofotherlogicalconstantsintheusualway).<br />

ThegroundsthatIhavedescribedareabstractentitiesthatcanbeconstructedinthemindandthatwecanbecomeinpossessionofinthatway.Alternatively,wemaythinkofagroundforajudgementasjustarepresentationofthestateofourmindwhenwehavejustifiedajudgement.<br />

Thepossessionofagroundforajudgementcanmanifestitselfinthe<br />

namingofthatobject,andIhaveintroducedanotationfordoingso. An<br />

alternativewayofdefiningthegroundswouldhavebeentolaydownthese<br />

waysofdenotinggroundsasthecanonicalnotationforgrounds,making<br />

adistinctionbetweencanonical andnon-canonical formssincethesame<br />

groundmaybedenotedbydifferentexpressions. Tobeinpossessionofa<br />

groundforajudgementcouldthenbeidentifiedwithhavingconstructeda<br />

termthatdenotesagroundforthatjudgement.<br />

7 Inferences<br />

Asthereaderhasalreadyrealized, theprimitiveoperationsintroduced<br />

abovetospecifythegroundsfortheaffirmationofpropositionsofvarious<br />

formscorrespondtocertaininferencerules,namelyGentzen’sintroduction<br />

rulesinthesystemofnaturaldeductionforfirstorderlanguages. Forinstance,conjunctiongroundingcorrespondstotheschemaforconjunction<br />

introduction. GerhardGentzensawtheintroductionrulesasdetermining<br />

themeaningofthecorrespondinglogicalconstants. Ihavenotfollowed<br />

thatideahere, 6 buthaveinsteadseenthespecificationsofwhatconstitutes<br />

groundsforaffirmingpropositionsofacertainformtobeconstitutivefor<br />

propositionsofthatform.OnecansaythatIhavecarriedoverGentzen’s<br />

ideatothedomainofgrounds,sincethegroundsarebuiltupbyprimitive<br />

groundingoperationsthatcloselycorrespondtohisintroductionrules.More<br />

precisely,itholdsforeverysuchgroundingoperation Φthatifweforman<br />

inferenceaccordingtothecorrespondinginferencerule,thenwecanapply<br />

Φtogroundsforthepremissesofthatinferenceandshallgetasaresulta<br />

groundfortheconclusionoftheinference. Furthermore,havingdefineda<br />

domainofgroundsbypresumingtheseprimitivegroundingoperations,we<br />

candefineotheroperationsonthegroundsofthisdomainthatwillhave<br />

6 Insomeotherworks(see,e.g.,(Prawitz,1973)andcf.footnote5)IhaveusedGentzen’s<br />

ideamoredirectlyinadefinitionofvalidargument,sayingthatanargumentwhoselast<br />

inferenceisanintroductionisvalidifandonlyiftheimmediatesubargumentsarevalid.


196 DagPrawitz<br />

similarpropertywithrespecttootherinferences.Thismakesitpossibleto<br />

givesomesubstancetotheideathataninferenceissomethingmorethan<br />

juststatingaconclusionandreasonsforit,anideathatwedescribedabove<br />

(endofSection3)inmetaphoricaltermsas“seeing”thattheproposition<br />

affirmedintheconclusionistruegiventhatthepropositionsaffirmedinthe<br />

premissesaretrue.Thementalactthatisperformedinaninferencemay<br />

berepresented,Isuggest,asanoperationperformedonthegivengrounds<br />

forthepremissesthatresultsinagroundfortheconclusion,wherebywe<br />

seethatthepropositionaffirmedistrue.<br />

Toillustratetheidealetusconsideraninferencethatisvalidbutnot<br />

logicallyvalid,sayacaseofmathematicalinduction. Howdoweseethat<br />

itsconclusion,theinductionstatement, A(x)say,istrueforanynatural<br />

number n? Isitnotreasonabletosaythatweseethisbyoperatingon<br />

thegivengroundsfortheinductionbaseandtheinductionstep?Westart<br />

withthegivengroundfortheinductionbase A(0)andthensuccessively<br />

applythegroundfortheinductionstep.Intheinductionstepwearriveat<br />

asserting A(n + 1)undertheinductionassumption A(n),anditsgroundis<br />

thusanunsaturatedgroundthatbecomesaclosedgroundfor A(n+1)when<br />

saturatedwith nandaclosedgroundfor A(n).Werealizethatbyapplying<br />

orsaturatingthisground ntimesbythenaturalnumbers 0,1,... ,n − 1,<br />

andthegroundsthatwesuccessivelyobtainfor A(0),A(1),... ,A(n − 1),<br />

wefinallygetinpossessionofagroundfor A(n),whichstatementisthus<br />

seentohold.<br />

Inaccordancewiththisidea,Ishallseeanindividualinferenceactas<br />

individuatedbyatleastthefollowingfiveitems(forbrevityIleaveout<br />

otheradditionalitemsthatmaybeneededtoindividuateaninferencesuch<br />

ashowhypothesesaredischarged):<br />

1.anumberofpremisses A1,A2,...,An,<br />

2.grounds α1,α2,...,αn,<br />

3.anoperation Φapplicabletosuchgrounds,<br />

4.aconclusion B,and<br />

5.anagentperformingtheoperationataspecificoccasion.<br />

Inlogicweareusuallynotinterestedinindividualactsofthiskindand<br />

thereforeabstractawayfromtheagent,whichleavesthefouritems1–4<br />

individuatingwhatIshallrefertoasan(individual)inference. Tomake<br />

orcarryoutsuchaninferenceistoapplytheoperation Φtothegrounds<br />

α1,α2,... ,αn.<br />

Idefineanindividualinferenceindividuatedby1–4tobevalidif α1,<br />

α2,... ,αnaregroundsfor A1,A2,... ,An,respectively,andtheresultof


InferenceandKnowledge 197<br />

applyingtheoperation Φtothegrounds α1,α2,... ,αn,thatis Φ(α1,α2,... ,<br />

αn),isagroundfor B.<br />

Accordingtothisdefinitionanindividualconjunctionintroduction,given<br />

withtwopremissesaffirmingthepropositions p1and p2,grounds α1and α2<br />

forthem,theoperationconjunctiongrounding &G,andtheconclusionaffirmingtheproposition<br />

p1&p2,istriviallyavalidinferencesince &G(α1,α2)<br />

isbydefinitionagroundforaffirming p1&p2,giventhat αiisagroundfor<br />

affirming pi.<br />

Ifweintroducetwooperations &R1and &R2definedforgroundsforaffirmations<br />

of propositions of conjunctive form by the equations<br />

&Ri(&G(α1,α2)) = αi(i = 1or 2),thenanindividualinferenceofthe<br />

typeconjunctionelimination,givenbyapremissaffirmingaconjunction<br />

p1&p2,aground αforit,anoperation &Ri,andaconclusionaffirming pi,<br />

isvalid,sincetheground αforthepremissmustbeoftheform &G(α1,α2)<br />

where αiisagroundfor pi,andsincetheground αiisbydefinitionthe<br />

valueoftheoperation &Riappliedto &G(α1,α2).<br />

Oftenwealsoabstractawayfromthegroundsandfromanyspecific<br />

premissesandconclusionofaninference,preservingonlyacertainformal<br />

relationbetweenthem.Wecanthenspeakofaninferenceformdetermined<br />

onlybythisformalrelationandanoperation Φ.Forinstance,modusponens<br />

maynowbeseenassuchaninferenceform,individuatedbygivingan<br />

operation Φ,namelytheoperation → Rdefinedbelow,andbysayingthat<br />

oneofthepremissesisaffirmingapropositionoftheformofanimplication<br />

p → qwhiletheotherpremissaffirmstheproposition pandtheconclusion<br />

affirmstheproposition q. Ifwealsoabstractawayfromtheoperation Φ,<br />

wegetwhatwemaycallaninferenceschema.<br />

Ishallsaythatsuchaninferenceformisvalidwhenitholdsforany<br />

instanceoftheformwithpremisses A1,A2,... ,An,andconclusion Band<br />

forallgrounds α1,α2,... ,αnfor A1,A2,... ,Anthattheresult Φ(α1,α2,<br />

...,αn)ofapplyingtheoperation Φinquestionto α1,α2,... ,αnisaground<br />

for B. Aninferenceschemaisvalidifitcanbeassignedanoperation Φ<br />

suchthattheresultinginferenceformisvalid.<br />

Forinstance,modusponensasusuallyunderstoodwithoutspecifyingan<br />

operationisaninferenceschema,whichisvalid,becausebyassigningtoit<br />

theoperation → Rdefinedbytheequation<br />

→ R( → Gξ p (β(ξ p ),α) = β(α),<br />

wegetavalidinferenceform. Intheequationabove β(α)istheresultof<br />

saturating β(ξ A )by α.Toseethattheresultinginferenceformisvalid,we<br />

havetoseethattheresultofapplyingtheoperation → Rtothegrounds<br />

forthepremissesofaninferenceofthisformisagroundfortheconclusion<br />

ofthatinference.Supposethat γand αaregroundsforpremissesofthat


198 DagPrawitz<br />

inferenceandthatthepremissesareaffirmingthatthepropositions p → q<br />

and paretrue.Thenbyhowgroundsforimplicationshavebeenspecified,<br />

γisoftheform → Gξ p (β(ξ p )),where β(ξ p )isanunsaturatedgroundfor<br />

affirmingthat qistrueundertheassumptionthat pistrue. Thismeans<br />

thatif β(ξ p )issaturatedbyagroundforaffirmingthat pistrue,theresult<br />

isagroundfortheaffirmationthat qistrue.Now αisagroundforaffirming<br />

that pistrue,hence β(α)isagroundforaffirmingthat qistrue,which<br />

affirmationistheconclusionoftheinference. And β(α)istheresultof<br />

applyingtheoperation → Rtothegivengroundsforthepremissesofthe<br />

inferenceaccordingtothedefinitionof → R.<br />

8 Conclusion<br />

Itshouldnowbeclearthatiftheconceptsofinference,makinganinference,<br />

validityofinference,andgroundareunderstoodinthewaydevelopedhere,<br />

thequestionthatwestartedwithiseasilyanswered.Thegeneralquestion<br />

washowandwhyweacquireknowledgebymakinginferences,andthis<br />

wasmorepreciselyformulatedastheproblemtostatetheconditionsunder<br />

whichanagent Pgetsagroundforajudgementbyinferringitfromother<br />

judgements.Giventhat<br />

andthat<br />

Jisavalidinference<br />

fromjudgements A1,A2,... ,Antoajudgement B, (a)<br />

theagent Phasgrounds α1,α2,...,αnfor A1,A2,... ,An, (b)<br />

theproblemwastostateathirdcondition(c),describingwhatrelation P<br />

hastohavetotheinference Jinorderthatitshouldfollowfrom(8)–(c)<br />

that<br />

Phasorgetsagroundfor B. (d)<br />

Whenanindividualinferenceisindividuatednotonlybyitspremisses<br />

andconclusionbutalsobygroundsforthepremissesandanoperation<br />

applicabletothem,andwhenmakinganinferenceisunderstoodasapplying<br />

thisoperationtothegrounds,inotherwords,astransformingthegiven<br />

groundsforthepremissestoagroundfortheconclusion,itbecomespossible<br />

tostatethethirdconditionthatwehavesoughtforsimplyas<br />

Pmakestheinference J. (c)<br />

Istartedoutfromtheconvictionthatthequestionwhyanagentgetsa<br />

groundforajudgementbyinferringitfrompremissesforwhichshealready


InferenceandKnowledge 199<br />

hasagroundshouldbeeasytoanswer,oncetheconceptsinvolvedare<br />

understoodinanappropriateway.Thisisnowactuallythecase.Whatit<br />

meansforaninference Jtobevalid,asithasnowbeendefined,issimply<br />

thattheoperation Φthatcomeswiththeinference Jyieldsagroundforthe<br />

conclusion Bwhenappliedtothegrounds α1,α2,... ,αnforthepremisses<br />

A1,A2,... ,An—inshort,that Φ(α1,α2,...,αn)isaground B.Therefore,<br />

bymakingtheinference J,thatis,byapplyingtheoperation Φtothegiven<br />

grounds,theagentgetsinpossessionofagroundfortheconclusion.<br />

Itremainstosaysomethingaboutwhatitisforanagenttobeinpossessionofagroundfortheconclusion.<br />

Asalreadysaidabove,itmeans<br />

basicallytohavemadeacertainconstructioninthemindofwhichthe<br />

agentisaware,andwhichshecanmanifestbynamingtheconstruction.<br />

Regardlessofwhethertheconstructionisonlymadeinthemindorisdescribed,itwillbepresenttotheagentundersomedescription,whichwill<br />

normallycontaindescriptionsofanumberofoperations.Itispresupposed<br />

thattheagentknowstheseoperations,whichmeansthatsheisabletocarry<br />

themout,whichinturnmeansthatsheisabletoconvertthetermthat<br />

describesthegroundtocanonicalform. Furthermoretheagentispresupposedtounderstandtheassertionthatshemakesandhencetoknowwhat<br />

kindofgroundsheissupposedtohaveforit.Itfollowsthatwhenanagent<br />

hasgotinpossessionofagroundforanjudgementbymakinganinference,<br />

sheisawareofthefactthatshehasmadeaconstructionthathastheright<br />

canonicalformtobeagroundfortheassertionthatshemakes.<br />

However,itdoesnotmeanthattheagenthasprovedthattheconstructionshehasmadeisreallyagroundforherassertion.Aswehavealready<br />

discussed(Section4),thiscannotbearequirementforherjudgementto<br />

bejustified.Butiftheinferenceshehasmadeisvalid,thensheisinfact<br />

inpossessionofagroundforherjudgement,andthisisexactlywhatis<br />

neededtobejustifiedinmakingthejudgement,ortobesaidtoknowthat<br />

theaffirmedpropositionistrue. Furthermore,althoughitisnotrequired<br />

inorderforthejudgmenttobejustified,byreflectingontheinferenceshe<br />

hasmade,theagentcanprovethattheinferenceisvalid,ashasbeenseen<br />

inexamplesabove.<br />

DagPrawitz<br />

DepartmentofPhilosophy,StockholmUniversity<br />

10691Stockholm,Sweden<br />

dag.prawitz@philosophy.su.se<br />

References<br />

Bolzano,B.(1837).Wissenschaftslehre(Vols.I–IV).Sulzbach:Seidel.<br />

Carroll,L.(1895).WhatthetortoisesaidtoAchilles.Mind,IV,278–280.


200 DagPrawitz<br />

Etchemendy,J. (1990). Theconceptoflogicalconsequence. Cambridge,MA:<br />

HarvardUniversityPress.<br />

Martin-Löf,P. (1985). Onthemeaningofthelogicalconstantsandthejustificationofthelogicallaws.<br />

NordicJournalofPhilosophicalLogic,1,11–60.<br />

(Republishing.)<br />

Prawitz,D.(1973).Towardsafoundationofageneralprooftheory.InP.Suppesetal.(Eds.),Logic,methodologyandphilosophyofscience(pp.225–250).<br />

Amsterdam:North-Holland.<br />

Prawitz,D.(1977).Meaningandproofs.Theoria,XLIII,2–40.<br />

Prawitz,D.(2009).Validityofinference.(ToappearinProceedingsfromthe 2 nd<br />

LaunerSymposiumontheOccasionofthePresentationoftheLaunerPrizeat<br />

Bern2006.)<br />

Ross,W.D. (1949). Aristotle’spriorandposterioranalytics. Oxford:Oxford<br />

UniversityPress.<br />

Sundholm,G. (1998). Inferenceversusconsequence. InTheLogicaYearbook<br />

1998.Prague:CzechAcad.Sc.<br />

Tarski,A.(1936). ÜberdenBegriffderlogischenFolgerung.ActesduCongrès<br />

InternationaldePhilosophieScientifiques,7,1–11. (TranslatedtoEnglishin<br />

A.Tarski,Logic,SemanticsandMetamathematics,Oxford1956.)


A Sound and Complete Axiomatic System of<br />

bdi–stit Logic<br />

1 Introduction<br />

Caroline Semmling Heinrich Wansing<br />

In(Semmling&Wansing,2008),bdi–stitlogichasbeenmotivatedand<br />

introducedsemantically.Thislogiccombinesthebelief,desire,andintention<br />

operatorsfromBDIlogic(Georgeff&Rao,1998;Wooldridge,2000)with<br />

theactionmodalitiesfrom d stitlogic,themodallogicofdeliberatively<br />

seeingtoitthat(Belnap,Perloff,&Xu,2001),(Horty&Belnap,1995).<br />

Themulti-modalbdi–stitlogicisanexpressivelyrichlogic,whichallows<br />

aformalanalysisof,forexample,reasoningaboutdoxasticdecisionsand<br />

beliefrevision,see(Semmling&Wansing,2009),(Wansing,2006a).<br />

In(Semmling&Wansing,2009),wehavepresentedasoundandcomplete<br />

tableaucalculusforbdi–stitlogicbasedonthetableaucalculusfor d stit<br />

logicdefinedin(Wansing,2006b). Inthepresentpaperweintroducea<br />

soundandcompleteaxiomatizationofbdi–stitlogicandprovedecidability<br />

byestablishingthefinitemodelproperty.<br />

2 Syntaxandsemantics<br />

Thesyntaxofbdi–stitlogic<br />

Thelanguageofbdi–stitlogiccomprisesadenumerablesetofsentential<br />

variables (p1,p2,p3,... ),theconstants ⊥, ⊤,theconnectivesofclassical<br />

propositionallogic(¬, ∧, ∨, ⊃, ≡),andthemodalnecessityandpossibility<br />

operators ✷and ✸.Weassumethat ✸isdefinedas ¬✷¬.Thisvocabularyis<br />

supplementedbyactionmodalitiesandoperatorsusedtoexpressthebeliefs,<br />

desiresandintensionsofarbitrary(rational)agents.Additionally,thereis<br />

apossibilityoperator�takenoverfrom(Semmling&Wansing,2008).We<br />

alsoassumeadenumberablesetofagentvariables (α1,α2,...,αn,...).


202 CarolineSemmling&HeinrichWansing<br />

Definition14(bdi–stitsyntax). 1.Everysententialvariable p1,p2,...<br />

andeachconstant ⊥, ⊤isaformula.<br />

2.If α1, α2areagentvariables,then (α1 = α2)isaformula.<br />

3.If ϕ, ψareformulasand αisanagentvariable,then ¬ϕ, (ϕ ∧ψ), ✷ϕ,<br />

�ϕ, αcstit : ϕ, α bel : ϕ, αdes : ϕand α int : ϕareformulas.<br />

4.Nothingelseisaformula.<br />

Aformulaconsistingofonlyonesententialvariableoroneconstantis<br />

calledanatomicformula.Thereadingofaformula α cstit : ϕis“agent α<br />

seestoitthat ϕ”. In(Semmling&Wansing,2008),insteadofthe cstitoperator,anoperatorofdeliberativelyseeingtoitthat,<br />

d stit:,isused.We<br />

introducethe d stit-operatorwiththefollowingequivalence<br />

α d stit : ϕ ≡ (α cstit : ϕ ∧ ¬✷ϕ).<br />

Thisisdonebecauseitmakesthepresentationofthecompletenessproof<br />

easier. Butneverthelessitisalsopossibletouse d stit: asaprimitive<br />

operatorandtochoosetheaxiomsappropriately,cf.(Belnapetal.,2001).<br />

Aformula αbel : ϕisreadas“agent αbelievesthat ϕ”or“agent α<br />

hasthebeliefthat ϕ”.Thereadingsofthedesireoperators α des :andthe<br />

intentionoperators α int :areconceivedinthisvein,too.<br />

Thesemanticsofbdi–stitlogic<br />

Abdi–stitmodelconsistsofaframe F = (Tree, ≤, A,N,C,B,D,I)and<br />

avaluationfunction v. Theframe F isbasedonabranchingtemporal<br />

structureasinStit-Theory,(Belnap&Perloff,1988).Theset Treeisanonemptyset(ofmomentsoftime)and<br />

≤isapartialorder,whichisreflexive,<br />

transitivebutacyclic,suchthateverymoment m ∈ Treehasaunique<br />

predecessor.Thus,thesetofhistories H,definedasthesetofallmaximal<br />

linearlyorderedsubsetsof Tree,andthesetofsituations S = {(m,h)|m ∈<br />

Tree,h ∈ H}oftheframeresultfromtheorderedset (Tree, ≤).Thesetof<br />

historiespassingthroughmoment m ∈ Tree({h | m ∈ h,h ∈ H})isdenoted<br />

by Hm.Thedenumerable,non-emptyset Aisthesetofagents,and Cis<br />

afunctionthatmapseverypairof A × Treetoasetofdisjointsubsets<br />

ofhistoriespassingthrough m,suchthattheunionofallsubsetsis Hm.<br />

Thus, C(α,m) = C α m 1 definesanequivalencerelationon Hm. Histories<br />

hand h ′ aresaidtobechoice-equivalent foragent αatmoment m,if<br />

theybelongtothesamesetin C α m. Theequivalenceclassofanarbitrary<br />

1 Onthisaccountwedonotexplicitlydistinguishbetweenthevariablesandtheagents<br />

anddenotebothby α, α1, . . . , αn, . . ..Thecontextwilldisambiguate.


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 203<br />

history hinmoment misdenotedby Cα m (h) = Cα �<br />

(m,h)<br />

Cα �<br />

(m,h) |h ∈ Hm representsalldistinguishablechoicecellsofagent αat<br />

situation (m,h).<br />

Thefunction N : S → P(P(S))assignsaset Nsofnon-emptysubsetsof<br />

situationstoeverysituation s.Theset Nsiscalledaneighbourhoodsystem<br />

of s.Itselementsarecalledneighbourhoodsof s.Wealsodenoteby Nthe<br />

unionofallneighbourhoodsystems, N = {U|U ∈ Ns,s ∈ S}.Thecontext<br />

willdisambiguate.<br />

Thefunctions Band Daremappingsfrom A × Sto P(N).Aset U ∈<br />

B(α,s) = Bα scanberegardedasaneighbourhoodendorsingcertainbeliefs ofagent αatsituation s.Inthesameway,everyset U ∈ D(α,s) = Dα s is<br />

aneighbourhoodendorsingcertaindesires.<br />

Tointerprettheascriptionofintentionstoanagentinasituation s,we<br />

usethefunction I,whichmapsapair (α,s) ∈ A × Stoaneighbourhood<br />

I(α,s) = Iα s ∈ Nrepresentingallsituationscompatiblewithwhat αintends<br />

at s.<br />

Let F = (Tree, ≤, A,N,C,B,D,I)besuchaframeandlet Selectmbe<br />

thesetofallfunctions σfrom Aintosubsetsof Hm,suchthat σ(α) ∈ Cα m .<br />

Fsatisfiestheindependenceofagentsconditionoftheiractions,ifandonly<br />

ifforevery m ∈ Tree, �<br />

α∈Agent<br />

σ(α) �= ∅<br />

. Thesetofclasses<br />

forevery σ ∈ Selectm.<br />

Apair M = (F,v)isthensaidtobeabdi–stitmodelbasedontheframe<br />

F,where visavaluationfunctionon F,whichmapstheagentvariables<br />

intotheset Aof Fandthesetofatomicformulasintothepowersetof<br />

situations P(S)of Fwiththeconstraintsthat v(⊥) = ∅and v(⊤) = S.<br />

Satisfiabilityofaformulainabdi–stitmodel Misthendefinedasfollows,<br />

where,forabbreviation,wedenoteforanarbitraryformula ϕby �ϕ�the<br />

setofsituationswhichcontaineverysituationof Msatisfyingformula ϕ;<br />

�ϕ� = {s|M,s |= ϕ}.<br />

Definition15(bdi–stitsemantics).Let s = (m,h)beasituationinmodel<br />

M = (F,v),let α, α1, α2beagentvariables,andlet ϕ, ψbeformulas<br />

accordingtoDefinition14.Then:<br />

M,s |= ϕ iff s ∈ v(ϕ),if ϕisanatomicformula.<br />

M,s |= (α1 = α2) iff v(α1) = v(α2).<br />

M,s |= ¬ϕ iff M,s �|= ϕ.<br />

M,s |= (ϕ ∧ ψ) iff M,s |= ϕandM,s |= ψ.<br />

M,s |= ✷ϕ iff M,(m,h ′ ) |= ϕforall h ′ ∈ Hm.


204 CarolineSemmling&HeinrichWansing<br />

M,s |=�ϕ iff thereexists U ∈ Nswith U ⊆ �ϕ�.<br />

M,s |= α cstit : ϕ iff {(m,h ′ )|h ′ ∈ C v(α)<br />

s } ⊆<br />

M,s |= α int : ϕ iff I v(α)<br />

s<br />

⊆ �ϕ�.<br />

M,s |= α des : ϕ iff thereexists U ∈ D v(α)<br />

s<br />

M,s |= α bel : ϕ iff thereexists U ∈ B v(α)<br />

s<br />

{(m,h ′ )|M,(m,h ′ ) |= ϕ,h ′ ∈ Hm}.<br />

with U ⊆ �ϕ�.<br />

with U ⊆ �ϕ�.<br />

Obviously,theoperators�, α des :,and αbel :arenotdefinedbyarelationalsemantics,butbyamonotonicneighbourhood(aliasScott-Montague,<br />

aliasminimalmodels)semantics,cp.(Chellas,1980;Montague,1970;Scott,<br />

1970).Forsuchanoperator op,aformula op ϕ ∧op ¬ϕissatisfiableforany<br />

contingentformula ϕ.Notethatitisnotpossibletosatisfy op ϕforaninconsistent<br />

ϕinabdi–stitmodel,becauseeveryneighbourhoodisnon-empty.<br />

Aneighbourhoodsemanticsofanoperator nopisinuse,if M,s |= nopϕ<br />

iff �ϕ� ∈ Ns. Usually,inthissemantics nopisinterpretedasakindof<br />

necessity-operator. Sinceforsuchanoperatoritisalsopossibletosatisfyformulas<br />

nop ϕ ∧ nop ¬ϕ,we,however,prefertoreadtheoperator�<br />

asakindofpossibility-operator,andcallitneighbourhoodpossibility. A<br />

formula�ϕistrueatasituation,if ϕiscognitivelypossibleatsituation s.<br />

Onemaywonderaboutthemeaningofthedualoperator ¬ op ¬ϕ.The<br />

semanticstellsusthataformula ¬ op ¬ϕissatisfiedatasituation s,if<br />

eachneighbourhoodof snecessarilycontainsasituation s ′ satisfying ϕ.<br />

Butformulassuchas ¬ op ¬ϕ ∧ ¬ opϕarealsosatisfiable,forexample,<br />

ifeveryneighbourhoodcontainsatleasttwosituations,onesatisfying ϕ<br />

andanothersatisfying ¬ϕ,sothatthedualoperatordoesnotexpressa<br />

kindofneighbourhoodnecessity,too. Howcanweexpressnecessityina<br />

neighbourhoodsemantics? Ourproposalis: s |= ⊡ϕiffforevery U ∈ Ns,<br />

U ⊆ �ϕ�(iff ϕisacognitivenecessityatsituation s). Thenobviouslyit<br />

holdsthattheimplications�ϕ ⊃ ¬ ⊡ ¬ϕand ⊡ϕ ⊃ ¬�¬ϕarevalid,but<br />

theimplicationsintheotherdirectionfail. Thus,itispossibletosatisfy<br />

formulasoftheform ¬ ⊡ ¬ϕ ∧ ¬�ϕ.<br />

Inadditiontoneighbourhoodnecessityandpossibility,therearemodal<br />

operatorsnotrelatedtothecognitivepropositionalattitudesofagents: ✷<br />

and ✸. Theseoperatorscanbereadasoperatorsofhistoricalnecessity<br />

andpossibility,respectively,wherehistoricalpossibilityandnecessityare<br />

definedasdualoperators: ✸ϕ ≡ ¬✷¬ϕ.Theyareadoptedfrom(Belnap<br />

&Perloff,1988;Belnapetal.,2001).


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 205<br />

3 Axiomatization<br />

Sincethebdi–stitlogicisconstructedon d stitframes,cf.(Belnap&Perloff,<br />

1988;Belnapetal.,2001),withtheadditionofsomefunctionsasinScott–<br />

Montaguemodels,cf. (Chellas,1980;Montague,1970;Scott,1970),the<br />

axiomatizationisnottoodifficult. Weassumeacompleteaxiomatization<br />

ofthenon-modalpropositionallogicandaddthefollowingaxioms:<br />

(A1) ✷ϕ ⊃ ϕ, ¬✷ϕ ⊃ ✷¬✷ϕ, ✷(ϕ ⊃ ψ) ⊃ (✷ϕ ⊃ ✷ψ).<br />

(A2) αcstit : ϕ ⊃ ϕ, ¬α cstit : ϕ ⊃ α cstit : ¬αcstit : ϕ,<br />

αcstit : (ϕ ⊃ ψ) ⊃ (α cstit : ϕ ⊃ αcstit : ψ).<br />

(A3) ✷ϕ ⊃ α cstit : ϕ.<br />

(A4) α = α, (α = β) ⊃ (β = α), ((α = β) ∧ (β = γ)) ⊃ (α = γ).<br />

(A5) (α = β) ⊃ (ϕ ⊃ ϕ[α/β]). 2<br />

(AIAk) (∆(β0,... ,βk) ∧ ✸β0 cstit : ψ0 ∧ ... ∧ ✸βk cstit : ψk) ⊃<br />

⊃ ✸(β0 cstit : ψ0 ∧ ... ∧ βk cstit : ψk).<br />

Theaxioms(AIAk)representtheindependenceofagentsconditionfor k ∈<br />

Nagents. Theformula ∆(β0,...,βk)statesthat β0,...,βkarepairwise<br />

distinct.Wealsohaveseveralderivationrules,cf.(Belnapetal.,2001):<br />

(RN) ϕ/✷ϕ,<br />

(MP) ϕ,ϕ ⊃ ψ/ψ,,<br />

(APCn) [✸αcstit : ϕ1 ∧ ✸(α cstit : ϕ2 ∧ ¬ϕ1) ∧ ...∧<br />

✸(α cstit : ϕn ∧ ¬ϕ1 ∧ ... ∧ ¬ϕn−1)] ⊃ (ϕ1 ∨ ... ∨ ϕn).<br />

Bytheaxiomof npossiblechoices(APCn),itisassuredthateveryagent<br />

hasatmost ndifferentalternativestoact.Iftheaxiom(APCn)isaccepted,<br />

theresultinglogicisdenotedby Ln.Evidently,itholdsthat Ln+1 ⊆ Ln.<br />

Thenewbdioperatorsareaxiomatizedbythefollowingaxiomsand<br />

derivationrulestakenoverfrom(Chellas,1980).<br />

(Di) α int : ϕ ⊃ ¬αint : ¬ϕ,<br />

(F,Fb,Fd,Fi) ¬�⊥, ¬αbel : ⊥, ¬αdes : ⊥, ¬αint : ⊥,<br />

(RM) (ϕ ⊃ ψ)/(�ϕ ⊃�ψ),<br />

2 Note,thatthesubstitutiondoesnothavetobeuniform.Itispossible,toreplacesome<br />

oralloccurrencesof αwith β.


206 CarolineSemmling&HeinrichWansing<br />

(RMi) (ϕ ⊃ ψ)/(α int : ϕ ⊃ αint : ψ),<br />

(RMb) (ϕ ⊃ ψ)/(α bel : ϕ ⊃ α bel : ψ),<br />

(RMd) (ϕ ⊃ ψ)/(α des : ϕ ⊃ α des : ψ).<br />

Fromtheseaxioms,whichareproventobecompleteincombinationwiththe<br />

axiomsofthe d stitlogicinSection4,wecanderivethefollowingtheorems,<br />

whichstatethemonotonyoftheneighbourhoodoperators,andformthe<br />

typicalaxiomsoftherelationallydefinedones.<br />

(Ni) α int : ⊤,<br />

(Tc) αcstit : (ϕ ∧ ψ) ≡ (α cstit : ϕ ∧ α cstit : ψ),<br />

(Ti) αint : (ϕ ∧ ψ) ≡ (α int : ϕ ∧ α int : ψ), α int : ϕ ∨ α int : ψ) ⊃ α int :<br />

(ϕ ∨ ψ),<br />

(T)�(ϕ ∧ ψ) ⊃ (�ϕ ∧�ψ),<br />

(Tb) α bel : (ϕ ∧ ψ) ⊃ (α bel : ϕ ∧ αbel : ψ),<br />

(Td) α des : (ϕ ∧ ψ) ⊃ (α des : ϕ ∧ α des : ψ)<br />

4 Completenessanddecidability<br />

Sincebdi–stitlogicisbasedon d stitlogic,whichisdecidable,andsinceit<br />

issupplementedwithsomeoperators,whichareinterpretedasindecidable<br />

classicalmodallogicswithaneighbourhoodsemantics,itisnotsurprising<br />

thatalsobdi–stitlogicisdecidable.Wefirstshowthecompletenessofthe<br />

axiomatizationpresentedinSection3,byextendingtheconstructionofa<br />

canonicalBT+AC(agentsandchoicesinbranchingtime)structureof d stit<br />

logic,presented,forexample,in(Belnapetal.,2001),totheconstructionof<br />

aframeofacanonicalbdi–stitmodel.Subsequently,weshowthatbdi–stit<br />

logichasthefinitemodelproperty,i.e.,eachnon-theoremof Lnisfalsifiable<br />

inafinitebdi–stitmodel,bydoingthesameasin(Belnapetal.,2001)for<br />

d stitlogic.Sincethenumberofaxiomschemesandderivationrulesisalso<br />

finite,thedecidabilityofbdi–stitlogicensues.<br />

Completeness<br />

ThesoundnessofthesystemofaxiomsandderivationrulesofSection3is<br />

straightforwardandfor(APCn)and(AIAk)asadducedin(Belnapetal.,<br />

2001). Thus,thissectiondealsonlywiththecompletenessoftheaxiomatization.<br />

Therefore,weintendtocombinetheconstructionofacanonical


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 207<br />

modelfor d stitlogic,representedin(Belnapetal.,2001),andtheconstructionofacanonicalmodelofaclassicalmonotonicmodallogic,cf.(Chellas,<br />

1980).Sincetheconstructionofthecanonical d stitmodelhasamorecomplicatedstructure,thisconstructionconstitutesthebasisandweexpand<br />

itappropriatelytocomprisetheinterpretationofthebelief,desire,and<br />

intentionoperators.<br />

WewillpresenttheconstructionofthecanonicalBT+ACstructure<br />

whichwasdefinedbyMingXu,cf.(Belnapetal.,2001;Xu,1994,1998).But<br />

first,propertiesoftheset WLnofmaximal Ln-consistentsetsofformulas,<br />

relationsonthisset WLnaswellasonsubsets X, Wofitandonsetsof<br />

agentvariablesarestatedinalmostthesamemannerasforthe d stitlogic<br />

Ldmn.Thesubscript nindicatesthattheaxiom(APCn)isincluded.<br />

Foragivensubset W ⊆ WLnwedefinearelation ∼ =Wonasetofagent<br />

variables,bystipulatingthat α1 ∼ =W α2,ifonlyif, α1 = α2 ∈ wforall<br />

w ∈ W.<br />

Lemma3.Therelation ∼ =Wisanequivalencerelationforany W ⊆ WLn.<br />

Proof.Cf. (Belnapetal.,2001). Thepropertyofbeinganequivalence<br />

relationresultsfromtheaxioms(A4),whichcorrespondtoreflexivity,symmetryandtransitivity.<br />

Theotherwayaround,wedefinearelationon X ⊆ WLnbyasetof<br />

agentvariables A: w ∼ =A w ′ ,ifandonlyif, α1 = α2 ∈ wiff α1 = α2 ∈ w ′<br />

forall α1,α2 ∈ A.<br />

Lemma4.Therelation ∼ =Aisanequivalencerelationonanarbitraryset<br />

W ⊆ WLnforanyset Aofagentvariables.<br />

Proof.Itisself-evident.<br />

Forthenexttwolemmaswefixanarbitrarysubset W ⊆ WLn. Then,<br />

therelation R ⊆ W × Wisdefinedby wRw ′ iff {φ|✷φ ∈ w} ⊆ w ′ .<br />

Lemma5.Therelation R ⊆ W × Wisanequivalencerelation.<br />

Proof.Cf. (Belnapetal.,2001). Thepropertyofbeinganequivalence<br />

relationresultsfromaxioms(A1),since ✷φ ⊃ φcorrespondstoreflexivity<br />

and ¬✷φ ⊃ ✷¬✷φtoeuclidity.<br />

Therefore,wecanpartition Wintoequivalenceclasses {Xi}i∈Iwithrespectto<br />

R. Let Xbeanarbitraryelementof {Xi}i∈I. Forsuchasubset<br />

X ⊆ Witholds,thatif (α = β) ∈ wforsome w ∈ X,itfollowsbyRule(RN)<br />

andAxiom(A5)that ✷(α = β) ∈ w,suchthat (α = β) ∈ w ′ forall w ′ ∈ X.<br />

Thatwarrantstheuseoftheequivalenceclasses {βj}j∈J = {[α]X}of ≡ X


208 CarolineSemmling&HeinrichWansing<br />

insteadofagentvariablestodefinethefollowingrelations Rβj ⊆ X × Xfor<br />

all j ∈ J;<br />

wRβjw′ iff {φ|βj cstit : φ ∈ w} ⊆ w ′ . 3<br />

Lemma6.For X ∈ {Xi}i∈Itherelation Rβj ⊆ X × Xisanequivalence<br />

relationforall j ∈ J.<br />

Proof.Cf. (Belnapetal.,2001). Thefirstandthesecondaxiomof(A2)<br />

expressreflexivityandeuclidity,respectively.<br />

Thisdefinitionoftherelations Rβjdependsontheset X.Inthefollowing,this<br />

Xwillbeanarbitrarybutfixedequivalenceclassofrelation R.<br />

Wedenoteby Eβjthesetofallequivalenceclassesofrelation Rβjon X.<br />

Lemma7.Let X, {βj}j∈J, R, Rβj , Eβj forall j ∈ Nbegivenbythe<br />

definitionsabove.Thenitholdsforall w ∈ X, φ:<br />

(i) ✷φ ∈ w iff φ ∈ w ′ forall w ′ ∈ X iff ✷φ ∈ w ′ forall w ′ ∈ X.<br />

(ii) βj cstit : φ ∈ wiff φ ∈ w ′ forall w ′ with wRβj w′ iff βj cstit : φ ∈ w ′<br />

forall w ′ with wRβj w′ .<br />

(iii) βj d stit φ ∈ wiff φ ∈ w ′ forall w ′ with wRβj w′ and ¬φ ∈ w ′′ for<br />

some w ′′ ∈ X.<br />

(iv)Assume ftobeanarbitraryfunctionfrom {βj}j∈Jintotheunionof<br />

Eβj forall j ∈ Jsuchthat f(βj) ∈ Eβj .Thisentails<br />

�<br />

f(βj) �= ∅.<br />

j∈J<br />

(v)Let Lnwith n ≥ 1andlet X, {βj}j∈J, {Rβj }j∈J, {Eβj }j∈Jbedefined<br />

withrespectto Ln. Thenthereareatmost ndifferentequivalence<br />

classes Rβjforevery j ∈ J,i.e.<br />

� �<br />

� ≤ n.<br />

� Eβj<br />

Proof.Cf.(Belnapetal.,2001).For(i),theclaimfollowsbyAxiom(A1)<br />

andRule(RN).For(ii)Axioms(A2)and(A3)andRule(RN)areneeded.<br />

Assertion(iii)resultsfromthedefinitionof d stitand(i),(ii).Clearly,(iv)<br />

isbackedupby(AIAk)and(v)by(APCn)forappropriate k, n.<br />

3 Theabbreviation βj c stit : φmeansthatforsome α ∈ βj, αcstit : φ ∈ w,becauseof<br />

Axiom(A5)itfollowsforall ˜α ∈ βj, ˜αcstit : φ ∈ w.


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 209<br />

Theorem8(completeness).Each Ln-consistentset Φofbdi–stitformulas<br />

issatisfiablebyabdi–stitmodel.<br />

Proof.Let WLnbethesetofallmaximal Ln-consistentsets,let<br />

A = {α|theagentvariable αoccursin Φ}.<br />

Then ∼ =Aisanequivalencerelationon WLn.Wedenoteby Wtheequivalenceclass,suchthatforallagentvariables<br />

α, ˜α ∈ A, β, ˜ β /∈ Aitholdsthat<br />

α = ˜α /∈ Φiff α = ˜α /∈ wand β = ˜ β ∈ wand α = β /∈ wforall w ∈ W.Let<br />

{Xi}i∈Ibethesetofallequivalenceclassesofrelation Ron W.<br />

Thebasisofourbdi–stitmodelisaframe F = (Tree, ≤, A,N,C,B,D,I)<br />

definedonaBranching-timestructure (Tree, ≤).Wedefineitasfollows:<br />

• Tree≔{w|w ∈ W } ∪ {Xi|i ∈ I} ∪ {W };<br />

• ≤≔trcl({(w,w)|w ∈ W }∪{(W,Xi),(Xi,Xi),(Xi,w)|w ∈ Xi,i ∈ I}∪<br />

{(W,W)}); 4<br />

• A≔{α|αbelongstoanarbitrarybutfixedsetofclassrepresentatives<br />

of ∼ =Wonallagentvariables.}; 5<br />

•forall i ∈ Iwedefinethechoiceequivalenceclassesofanyagent α ∈ A<br />

atanymomentin Tree:<br />

C(α,w)≔{{hw}},where hwistheuniquehistorypassingthrough<br />

moment wwith hw = {w,Xi,W },where Xiistheequivalence<br />

classcontaining w.<br />

C(α,W)≔{{hw|w ∈ W }},<br />

AccordingtoLemma6,thereisanequivalenceclass βj with<br />

α ∈ βjandanequivalencerelation R i βj on Xi. Wedenotethe<br />

classesof R i βj on Xiby E i βj andthenwecandefine:<br />

C(α,Xi)≔{H |∃e : e ∈ E i βj and H = {hw|w ∈ e}}.<br />

Sincethereisaone-to-onecorrespondencebetweenall w ∈ W andall<br />

historiesof (Tree, ≤),thefollowingconceptsarewell-defined. Forall m ∈<br />

Tree, w ∈ Wand α ∈ A,wehave:<br />

• |ϕ|≔ �<br />

{(Xi,hw ′)|ϕ ∈ w′ ,Xi ∈ hw ′} ∈ N (m,hw)iff�ϕ ∈ w;<br />

i∈I<br />

4 Here trclstandsforthetransitiveclosureofabinaryrelation.<br />

5 Recallthatweusethesame αforagentvariablesandagents. Sincewenowinterpret<br />

theagentvariablebytheagentvariableitself,thisnamingwasjustakindofforestalling.<br />

Butnotethat A �= Aingeneral.


210 CarolineSemmling&HeinrichWansing<br />

• |ϕ| ∈ Bα (m,hw) iffthereis α bel : ϕ ∈ w;6<br />

• |ϕ| ∈ Dα (m,hw) iffthereis αdes : ϕ ∈ w;<br />

•Wedefinearelation Sα ⊆ W × Wforall α ∈ A,bystipulatingthat<br />

wSαw ′ iff {ϕ|α int : ϕ ∈ w} ⊆ w ′ . Thenwechoosethesets I α s for<br />

everysituation s = (m,hw)in M:<br />

I α (m,hw) = {(w′ ,hw ′)|wSαw ′ }.<br />

Wehavetoshowthatforall w ∈ Wthereisaw ′ ∈ Wwith wSαw ′ ,<br />

whichmeansthat Iα (m,hw) �= ∅forany m ∈ Tree .Fromaxiom(Di),<br />

derivationrule(RMi)andtheorems(Ti),itisevidentthatforany<br />

w ∈ Wtheset S = {ϕ|α int : ϕ ∈ w}isconsistent,thusthereisa<br />

maximalconsistentset w ′ with S ⊆ w ′ .<br />

Inanalogyto(Belnapetal.,2001),weclaimthattheframe Fsatisfies<br />

theindependenceofagentscondition. Foragivenmoment m ∈ Treelet<br />

Selectmbethesetofallfunctionsfrom Aintosubsetsof Hm,thesetof<br />

historiespassingthroughmoment m,whereforall σ ∈ Selectmitholdsthat<br />

σ(α) ∈ Cα m .Then Fsatisfiesthisconditionifandonlyifforeverymoment<br />

mandany σ ∈ Selectm �<br />

σ(α) �= ∅.<br />

α∈A<br />

Let m = wforanarbitrary w ∈ W or m = W,thentheconditionis<br />

evidentlysatisfied. Now,letforanarbitrary i ∈ I, σXibeanyfunction from Ainto P(HXi ),suchthat σXi (α) ∈ Cα forall α ∈ A.Bytheabove<br />

Xi<br />

with σXi (α) =<br />

definitionof C α Xi ,thereisanequivalenceclass ej ∈ E i βj<br />

{hw|w ∈ ej}.Defineafunction fiby fi(α) = ej ∈ Ei,where α ∈ βj<br />

βj<br />

for j ∈ J. Thenforall α, ˜α ∈ βj, fi(α) = fi(˜α),thereisawelldefined<br />

correspondingfunction ˜ fi: {βj}j∈J → �<br />

Eβj .Aswell, w ∈ fi(α)iff hw ∈<br />

j∈J<br />

σXi (α)andbyLemma7(iv),itholdsthat<br />

�<br />

α∈A<br />

fi(α) = �<br />

j∈J<br />

˜fi(βj) �= ∅,suchthat �<br />

α∈A<br />

σXi (α) �= ∅.<br />

Since |Cα w | = |Cα W | = 1forall w ∈ Wand α ∈ A,andforall i ∈ Iitholds<br />

that |Cα Xi | = |Ei |,itobviouslyfollowsbyLemma7(v)thatforany α ∈ A<br />

βj<br />

and m ∈ Tree, Cα m ≤ n,cf. (Belnapetal.,2001). Soanyagent αhasat<br />

most npossiblechoicesintheframe F.<br />

6 BecauseofAxiom(A5)forall β ∈ [α]W itfollows: β op : φ ∈ wiff αop : φ ∈ wfor<br />

op ∈ {c stit,dstit,bel,des,int}andforall w ∈ W.


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 211<br />

Now,wedefineacanonicalmodelonthatframe M = (F,v),where vis<br />

aninterpretationfunction,whichmapseachagentvariable βon v(β) = α ∈<br />

Awith β ∈ [α]Wandeveryatomicformula ponasubsetof Scontaining<br />

allsituations s = (m,hw)with p ∈ wforall m ∈ Tree.Evidently, v(⊤) = S<br />

and v(⊥) = ∅.Foranyagentvariable α, h ∈ H, w ∈ Witholdsthat h ∈<br />

Cα Xi (hw)iffthereis e ∈ Ei and w, w [α]Xi ′ ∈ e,where h = hw ′.Furthermore,<br />

w, w ′ ∈ e ∈ E i [α]X i<br />

iff wRi w [α]Xi ′ ,suchthat h ∈ Cα Xi (hw)iff wRi w [α]Xi ′ .<br />

Weshowbyinductionthat M,(Xi,hw) |= ϕiff ϕ ∈ wforeverybdi–stit<br />

formula ϕand w ∈ Xi,forall i ∈ I.<br />

M,(Xi,hw) |= p ⇔ (Xi,hw) ∈ v(p) bydefinition<br />

⇔ p ∈ w.<br />

M,(Xi,hw) |= (α = β) ⇔ v(α) = v(β) ⇔ α ∼ =W β ⇔ α = β ∈ w.<br />

M,(Xi,hw) |= ¬ϕ ⇔ (Xi,hw) �|= ϕ byinduction<br />

⇔ ϕ /∈ w ⇔ ¬ϕ ∈ w.<br />

M,(Xi,hw) |= ϕ ∧ ψ ⇔ (Xi,hw) |= ϕand (Xi,hw) |= ψ byinduction<br />

⇔<br />

ϕ ∈ wand ψ ∈ w ⇔ (ϕ ∧ ψ) ∈ w.<br />

M,(Xi,hw) |= ✷ϕ ⇔forall h ∈ HXi itholdsthat (Xi,h) |= ϕ<br />

byinduction<br />

⇔ forall h ∈ HXi thereis w′ ∈ Xi<br />

with h = hw ′and ϕ ∈ w′<br />

⇔forall w ′ ∈ Xi,ϕ ∈ w ′byLemma7(i)<br />

⇔ ✷ϕ ∈ w.<br />

M,(Xi,hw) |= β cstit : ϕ ⇔forall h ∈ C α Xi (hw)with β ∈ [α]W<br />

itholdsthat (Xi,h) |= ϕ<br />

⇔forall h ∈ C α Xi (hw)thereis w ′ with h = hw ′<br />

anditholdsthat (Xi,hw ′) |= ϕbyinduction ⇔<br />

forall w ′ ∈ Xi,if wR i w [α]Xi ′ then ϕ ∈ w ′<br />

byLemma7(ii)<br />

⇔ αcstit : ϕ ∈ w<br />

byAxiom(A5)<br />

⇔ β cstit : ϕ ∈ w.<br />

M,(Xi,hw) |=�ϕ ⇔thereis U ∈ N (Xi,hw)∅ �= U ⊆ �ϕ�<br />

⇔thereis ψwith ∅ �= |ψ| ⊆ �ϕ�and�ψ ∈ w<br />

(∗)<br />

⇔�ϕ ∈ w.<br />

Wewanttoshowtheequivalence (∗)<br />

⇔.<br />

⇐:If�ϕ ∈ w,then |ϕ| ∈ N (Xi,hw)with w ∈ Xi.Thatmeansthereis<br />

ψ = ϕwith |ϕ| ⊆ �ϕ�byinduction.


212 CarolineSemmling&HeinrichWansing<br />

⇒:Thereexists ψwith ∅ �= |ψ| ⊆ �ϕ�and�ψ ∈ w.Since<br />

|ψ| = �<br />

{(Xi,hw ′)|ψ ∈ w′ �<br />

,Xi ∈ hw ′}, W = {w|(Xi,hw)isasituation},<br />

i∈I<br />

and |ψ| ⊆ �ϕ�itfollowsforall w ′ ∈ Wand i ∈ I:if M,(Xi,hw ′) |= ψthen<br />

M,(Xi,hw ′) |= ϕ. Byinductionwehaveforall w′ ∈ W: if ψ ∈ w ′ then<br />

ϕ ∈ w ′ .Thusitholdsforall w ′ ∈ Wthat (ψ ⊃ ϕ) ∈ w ′ .Byrule(RM)it<br />

holdsthat (�ψ ⊃�φ) ∈ w ′ forall w ′ ∈ W.Since�ψ ∈ w,wehave�ϕ ∈ w.<br />

i∈I<br />

M,(Xi,hw) |= β bel : ϕ ⇔thereis U ∈ B α (Xi,hw)<br />

∅ �= U ⊆ �ϕ�<br />

with β ∈ [α]Wand<br />

⇔thereis ψwith ∅ �= |ψ| ⊆ �ϕ�and αbel : ψ ∈ w<br />

(∗∗)<br />

⇔ β bel : ϕ ∈ w.<br />

M,(Xi,hw) |= β des : ϕ ⇔thereis U ∈ D α (Xi,hw)<br />

∅ �= U ⊆ �ϕ�<br />

with β ∈ [α]Wand<br />

⇔thereis ψwith ∅ �= |ψ| ⊆ �ϕ�and αdes : ψ ∈ w<br />

⇔ β des : ϕ ∈ w.<br />

Weonlyshow (∗∗),whichissimilartotheargumentforthe�-operator.<br />

Thecorrespondingequivalenceforthedesireoperatorisshownanalogously.<br />

⇐:If β bel : ϕ ∈ w,then,byAxiom(A5), α bel : ϕ ∈ w,suchthat<br />

|ϕ| ∈ B α (Xi,hw) with w ∈ Xi.Thatmeansthereis ψ = ϕwith |ϕ| ⊆ �ϕ�by<br />

induction.<br />

⇒:Thereis ψwith ∅ �= |ψ| ⊆ �ϕ�and αbel : ψ ∈ w.Since<br />

|ψ| = �<br />

{(Xi,hw ′)|ψ ∈ w′ �<br />

,Xi ∈ hw ′}, W = {w|(Xi,hw)isasituation},<br />

i∈I<br />

and |ψ| ⊆ �ϕ�itfollowsforall w ′ ∈ Wand i ∈ I:if M,(Xi,hw ′) |= ψthen<br />

M,(Xi,hw ′) |= ϕ. Byinductionwehaveforall w′ ∈ W: if ψ ∈ w ′ then<br />

ϕ ∈ w ′ .Thusitholdsforall w ′ ∈ Wthat (ψ ⊃ ϕ) ∈ w ′ .Byrule(RMb)it<br />

holdsthat (α bel : ψ ⊃ α bel : φ) ∈ w ′ forall w ′ ∈ W.Since α bel : ψ ∈ w,<br />

wehave α bel : ϕ ∈ w.Againbecauseof(A5),itfollows β bel : ϕ ∈ w.<br />

M,(Xi,hw) |= β int : ϕ ⇔ I α (Xi,hw)<br />

i∈I<br />

⊆ �ϕ�with β ∈ [α]W<br />

⇔forall s = (w ′ ,h ′ w) : if wSαw ′ ,then M,s |= ϕ<br />

byinduction<br />

⇔ forall s = (w ′ ,h ′ w) :if wSαw ′ ,then ϕ ∈ w ′<br />

(∗∗∗)<br />

⇔ β int : ϕ ∈ w.<br />

Atlastwehavetoshowtheequivalence (∗ ∗ ∗).


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 213<br />

⇐:If β int : ϕ ∈ w,then α int : ϕ ∈ wandforall w ′ ∈ Wwith wSαw ′ it<br />

followsthat ϕ ∈ w ′ .<br />

⇒:Forall s = (w ′ ,h ′ w ):if wSαw ′ ,then ϕ ∈ w ′ .Becauseofaxiom(Di)<br />

andmaximalityforany ϕand w ∈ Witholdsthateither α int : ϕ ∈ wor<br />

αint : ¬ϕ ∈ worbothisnotthecase. Theassumption α int : ¬ϕ ∈ wis<br />

contradictory,since ¬ϕ ∈ {ψ|α int : ψ ∈ w} ⊆ w ′ forany w ′ with wSαw ′ .<br />

Assume ¬αint : ϕ ∈ wand ¬αint : ¬ϕ ∈ w. Then ϕ, ¬ϕ /∈ {ψ|α int :<br />

ψ ∈ w}. Since wismaximal,theset {ψ|α int : ψ ∈ w}isclosedunder<br />

implicationby(RMi)and(Ti),suchthatthesets {ϕ} ∪ {ψ|α int : ψ ∈ w}<br />

and {¬ϕ} ∪ {ψ|α int : ψ ∈ w}arebothconsistent. Butthenthereisa<br />

maximalworld w ′′ with {¬ϕ}∪{ψ|α int : ψ ∈ w} ⊆ w ′′ and wSαw ′′ .Butthat<br />

conflictswith ϕ ∈ w ′ forall wSαw ′ .Thus, α int : ϕ ∈ w,resp. β int : ϕ ∈ w.<br />

Atanyrate,thereisone(maybemorethanone,thenchooseone)maximalconsistentset<br />

w0 ∈ Wwith Φ ⊆ w0.This w0belongstoanequivalence<br />

class Xi0of R.Then, M,(Xi0 ,hw0 ) |= ϕforany ϕ ∈ Φ.Thus,anyconsistentset<br />

Φissatisfiable.<br />

Finitemodelproperty<br />

Weconstructtoagivensentence ϕaccordingtoDefinition14afiniteframe<br />

Ffinandaddaspecialinterpretation v,suchthatforeverysubsentence<br />

of ϕitisdecidablewhetherthesubsentenceissatisfiableby (Ffin,v). We<br />

adoptthefiltrationmethodasusedin(Belnapetal.,2001). Wetakethe<br />

canonicalframe F = (Tree, ≤, A,N,C,B,D,I)oftheprevioussectionand<br />

defineafiltrationfirstoverallworldsbythesetofallsubformulasofthe<br />

givenformula ϕincludingallformulasderivedbyAxioms(AIAk)and(Ai)<br />

forall 1 ≤ i ≤ 5,cf. (Belnapetal.,2001). Thenagain,wefiltratethe<br />

equivalenceclassesofrelation Rbyasetofformulasimpliedbysubformulas<br />

of ϕprefixedby d stitor cstitoperators,suchthatwecandefinechoiceequivalenthistories.Tobeginwith,wedefinethesetsofsubformulas:<br />

Σϕ = {ψ|ψisasubsentenceof ϕ} ,<br />

Σi = Σϕ ∪ {¬β int : ¬ψ|β int : ψ ∈ Σϕ} ,<br />

Σd = Σϕ ∪ {β cstit : ψ, ¬✷ψ|β d stit : ψ ∈ Σϕ} ∪ {¬✷¬ψ|✸ψ ∈ Σϕ} ,<br />

Σe = {ψ|ψisasubsentenceofaformulaof Σdorof<br />

{β cstit : ¬β cstit : ψ|β cstit : ψ ∈ Σd}},<br />

Σp = {✸(β0 cstit : φ0 ∧ · · · ∧ βn cstit : φn)|n ≥ 0,β0,... βndiffer<br />

pairwisely,occurin ϕ,andforall 0 ≤ i ≤ n,φi = ψ0 ∧ · · · ∧ ψmi ,<br />

0 ≤ j ≤ mi,thereis βi cstit : ψjin Σe},<br />

Σa = {ψ|ψisasubsentenceofaformulaof Σp ∪ Σe ∪ Σi}.


214 CarolineSemmling&HeinrichWansing<br />

Forall w,w ′ ∈ W,wedefinetheequivalencerelation ≡ Σabysetting w ≡ Σa<br />

w ′ iffforall ψ ∈ Σa: (m,hw) |= ψiff (m,hw ′) |= ψ.By ˜ Wwedenoteachosen<br />

setofrepresentativesofallequivalenceclasses.By ˜ Xiwedenotethesubset<br />

of ˜ Wconsistingofallrepresentatives,whichbelongtotheequivalenceclass<br />

Xiforall i ∈ I. Notethatiftherelation Risfirstappliedtotheset W<br />

andthenrelation Σaisimplementedontheequivalenceclasses,onegetsan<br />

isomorphicBranchingTimeStructureintheend.Since Σaisfinite,soitis<br />

˜Wandtherewithevery ˜ Xi.Thus,wedefineafiniteframe Ffin:<br />

• Tree≔{w|w ∈ ˜ W } ∪ { ˜ Xi|i ∈ I} ∪ { ˜ W }isfinite.<br />

• ≤≔trcl({(w,w)|w ∈ ˜ W } ∪ {( ˜ W, ˜ Xi),( ˜ Xi, ˜ Xi),( ˜ Xi,w)|w ∈ Xi,i ∈ I}<br />

∪ {( ˜ W, ˜ W)}),suchthatthereisagainanone-to-onecorresponding<br />

relationbetweenthehistories ˜ Handtheset ˜ W, ˜ H = {hw|w ∈ ˜ W }.<br />

•thesetofagentsischosenas à = {α|thereis αoccurringin ϕ},thus<br />

Ãisalsofinite. 7<br />

•forall α ∈ Ãwedefinerelations ≡ α Σeoneveryset ˜ Xi,by w ≡ α Σe w′ iff<br />

forall α cstit : ψ ∈ Σeitholdsthat αcstit : ψ ∈ wiff αcstit : ψ ∈ w ′ .<br />

wedenotethesetofallequivalenceclasseson ˜ Xi.Withthis<br />

By Ũi<br />

[α] ˜ Xi<br />

definitionitispossibletodefinethechoiceequivalentfunction ˜ Cin<br />

thefiniteframe:<br />

˜C(α,w)≔{{hw}},where hwistheuniquehistorypassingthrough<br />

moment wwith hw = {w, ˜ Xi, ˜ W },<br />

˜C(α, ˜ W)≔{{hw|w ∈ ˜ W }},<br />

˜C(α, ˜ �<br />

Xi)≔ H|∃e : e ∈ U i<br />

[α] Xi ˜<br />

�<br />

and H = {hw|w ∈ e} .<br />

Sincethereisaone-to-onecorrespondencebetweenall w ∈ ˜ W and<br />

allhistoriesof (Tree, ≤),thefollowingnotionsarewell-defined. Forall<br />

�φ,αbel : φ,αdes : φ ∈ Σϕ, m ∈ Tree, w ∈ ˜ Wand α ∈ Ã,wehave:<br />

• |φ|≔ �<br />

{(w ′ ,hw ′)|φ ∈ w′ } ∈ N (m,hw)iff�φ ∈ w,<br />

i∈I<br />

• |φ| ∈ ˜ Bα (m,hw) iff α bel : φ ∈ w,<br />

• |φ| ∈ ˜ Dα (m,hw) iff α des : φ ∈ w,<br />

7 Weneglecttheproblemofidentitystatementsofagents,sinceitcanbehandledas<br />

above.


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 215<br />

•Forall w ∈ ˜ Wandeach α ∈ Ãset tα w<br />

= {φ|α int : φ ∈ w}. Then,<br />

thereisatleastone ˜w α ∈ ˜ Wwith t α w ∩Σa ⊆ ˜w α ,since t α w isconsistent<br />

by(Ti),(Di).Wedefineforall m ∈ Tree:<br />

Ĩ α (m,hw) = {( ˜wα ,h˜w α)| ˜wα ∈ ˜ W,t α w ∩ Σa ⊆ ˜w α }.<br />

Byconstruction,thesesetsarenotempty.<br />

Lemma8.Let F = (Tree, ≤, A,N,C,B,D,I)bethecanonicalframe,let<br />

ibefixed, Xithecorrespondingequivalenceclass Xi ∈ Tree, ˜Xithecorre-<br />

spondingclassinthefiniteframe Ffin = (Tree, ≤, Ã,Ñ, ˜ C, ˜ B, ˜ D, Ĩ)filtrated<br />

thesetsofequivalenceclassesof<br />

bythesetsofsubformulasof ϕand U i<br />

[α] ˜ Xi<br />

≡ Σeon ˜ Xiforall α ∈ Ã.<br />

(i)If ✷ψ ∈ Σa, w ∈ ˜ Xi,then<br />

(ii)If α cstit : ψ ∈ Σe, w ∈ ˜ Xi,then<br />

✷ψ ∈ w iff ψ ∈ w ′ forall w ′ ∈ ˜ Xi.<br />

α cstit : ψ ∈ w iff ψ ∈ w ′ forall w ′ ∈ ˜ Xiwith w ≡ α Σe w′ .<br />

(iii)Forallequivalenceclasses eα ∈ Ũi<br />

[α] ˜ Xi<br />

�<br />

α∈ Ã<br />

eα �= ∅.<br />

itholdsthat<br />

(iv)Let ϕ ∈ Lnwith n ≥ 1,thenforany j ∈ [0, | Ã|]itholdsthat<br />

Proof.Cf.(Belnapetal.,2001).<br />

�<br />

�<br />

�U i<br />

[α] ˜ Xi<br />

�<br />

�<br />

� ≤ n.<br />

Thisframesatisfiestheindependenceofagentscondition. Forallmoments<br />

m ∈ { ˜ W,w|w ∈ ˜ W }itisevidentthatforanarbitraryfunction<br />

σm : Ã → ˜ Cmtheintersection � {σm(α)|α ∈ Ã}isnotempty.If m = ˜ Xi,<br />

thenforall α ∈ Ãthereis eα ∈ U i<br />

[α] ˜ Xi<br />

with σm(α) = eα,suchthatwiththe<br />

frameproperty8(iii)theset � {σXi ˜ (α)|α ∈ Ã}isalsonotempty.Assuming<br />

Ln,thecorrespondingaxiomofpossiblechoicesisalsofulfilled,sinceforall<br />

α ∈ Ã,<br />

�<br />

�<br />

� ˜ Cα � �<br />

� �<br />

W˜<br />

� = � ˜ Cα � �<br />

� �<br />

w�<br />

= 1and � ˜ Cα � �<br />

� �<br />

Xi<br />

˜ � = �<br />

�Ui �<br />

�<br />

�<br />

� ≤ nbyLemma8(iv).<br />

[α] ˜ Xi<br />

Forany ψ ∈ Σϕwecanshowthatforanyequivalenceclass ˜ Xiitholds<br />

that<br />

Mfin( ˜ Xi,hw) |= ψ iff ψ ∈ w,


216 CarolineSemmling&HeinrichWansing<br />

where Mfin = (Ffin,v)and visthevaluationfunctiondefinedasforthe<br />

canonicalmodel,butrestrictedto ˜ W.Theproofisbyinduction.<br />

Mfin,( ˜ Xi,hw) |= p ⇔ ˜ Xi,hw) ∈ v(p) ⇔ p ∈ w.<br />

Mfin,( ˜ Xi,hw) |= ¬ψ ⇔ ( ˜ Xi,hw) �|= ψ ⇔ ψ /∈ w ⇔ ¬ψ ∈ w.<br />

Mfin,( ˜ Xi,hw) |= φ ∧ ψ ⇔ ( ˜ Xi,hw) |= φand ( ˜ Xi,hw) |= ψ<br />

⇔ φ ∈ wand ψ ∈ w ⇔ (φ ∧ ψ) ∈ w.<br />

Mfin,( ˜ Xi,hw) |= ✷ψ ⇔forall h ∈ H ˜ Xi itholdsthat ( ˜ Xi,h) |= ψ<br />

⇔forall h ∈ H ˜ Xi thereis w′ ∈ ˜ Xiwith<br />

h = hw ′and ψ ∈ w′<br />

⇔forall w ′ ∈ ˜ Xi,ψ ∈ w ′byLemma8(i)<br />

⇔ ✷ψ ∈ w.<br />

Mfin,( ˜ Xi,hw) |= α cstit : ψ ⇔ forall h ∈ ˜ C α ˜ Xi (hw)itholdsthat<br />

( ˜ Xi,h) |= ψ<br />

⇔ eα ∈ U i<br />

[α] ˜ Xi<br />

with w ∈ eα :<br />

⇔forall w ′ ∈ ˜ Xi,if w, w ′ ∈ ejthen ψ ∈ w ′<br />

byLemma8(ii)<br />

⇔ αcstit : ψ ∈ w.<br />

Mfin,( ˜ Xi,hw) |=�ψ ⇔ thereis U ∈ Ñ ( ˜ Xi,hw) ∅ �= U ⊆ �ψ�<br />

⇔ thereis φ ∈ Σϕwith ∅ �= |φ| ⊆ �ψ�and<br />

�φ ∈ w<br />

⇔�ψ ∈ w.<br />

If |φ| ⊆ �ψ�,then |φ| ⊆ |ψ|,i.e.forall w ′ ∈ ˜ W : (ψ ⊃ φ) ∈ w ′ .Assumethere<br />

is w ∈ Wwith (ψ ⊃ φ) /∈ w.Since ˜ Wisacompletesetofrepresentatives<br />

ofall ≡ Σa-equivalenceclasses,thereis ˜w ∈ ˜ Wwith w ≡ Σa ˜w. Forthe<br />

canonicalmodel Mitholdsthat (ψ ⊃ φ) /∈ w.Then M,(Xi,hw) |= ¬(ψ ⊃<br />

φ)and M,(Xi,h˜w) |= (ψ ⊃ φ).Thus, M,(Xi,hw) �|= ψor M,(Xi,hw) |= φ<br />

and M,(Xi,h˜w) |= ψand M,(Xi,h˜w) �|= φ,butthisconflictswith w ≡ Σa<br />

˜w,as φ,ψ ∈ Σa.Therefore,forall w ∈ W : φ ⊃ ψ ∈ w,andsoby(RM)<br />

�φ ⊃�ψ ∈ w.Consequently,�ψ ∈ w.Theotherdirectionisobviouswith<br />

ψ = φ.Similarconsiderationsgive:<br />

Mfin,(Xi,hw) |= αbel : ψ ⇔thereis U ∈ ˜ B α (Xi,hw) ∅ �= U ⊆ �ψ�<br />

⇔thereis φ ∈ Σϕwith ∅ �= |φ| ⊆ �ψ�and<br />

α bel : φ ∈ w ⇔ αbel : ψ ∈ w.


ASoundandCompleteAxiomaticSystemofbdi–stitLogic 217<br />

Mfin,(Xi,hw) |= α des : ψ ⇔thereis U ∈ ˜ D α (Xi,hw) ∅ �= U ⊆ �ψ�<br />

⇔thereis φ ∈ Σϕwith ∅ �= |φ| ⊆ �ψ�and<br />

α des : φ ∈ w ⇔ αdes : ψ ∈ w.<br />

Mfin,(Xi,hw) |= αint : ψ ⇔ Ĩα (Xi,hw) ⊆ �ψ�<br />

⇔forall w ′ ∈ ˜ W :if t α w ∩ Σa ⊆ w ′ ,then ψ ∈ w ′ .<br />

⇔ α int : ψ ∈ w.<br />

Assume αint : ψ /∈ w,then ¬α int : ψ ∈ w.Becauseof(Di)therearetwo<br />

differentcasespossible,(i) αint : ¬ψ ∈ wor(ii) ¬αint : ¬ψ ∈ w.If(i),then<br />

¬ψ ∈ t α wand,since Σi ⊆ Σa, ¬ψ ∈ t α w ∩ Σa ⊆ w ′ .Or(ii) ¬αint : ψ ∈ wand<br />

¬αint : ¬ψ ∈ w;then ψ, ¬ψ /∈ t α w.Butthenthereisaw ′′ with t α w ∩Σa ⊆ w ′′<br />

and ¬ψ ∈ w ′′ .Thesecontradictionsimply α int : ψ ∈ w.<br />

Tosumup,like d stitlogic,bdi–stitlogicisfinitelyaxiomatizableand<br />

hasthefinitemodelproperty.Therefore,itisdecidable.<br />

CarolineSemmling<br />

InstituteofPhilosophy,DresdenUniversityofTechnology<br />

Dresden,Germany<br />

Caroline.Semmling@gmx.de<br />

HeinrichWansing<br />

InstituteofPhilosophy,DresdenUniversityofTechnology<br />

Dresden,Germany<br />

Heinrich.Wansing@tu-dresden.de<br />

References<br />

Belnap,N.D.,&Perloff,M. (1988). Seeingtoitthat: acanonicalformfor<br />

agentives.Theoria,54,175–199.<br />

Belnap,N.D.,Perloff,M.,&Xu,M. (2001). Facingthefuture: Agentsand<br />

choicesinourindeterministworld.NewYork:OxfordUniversityPress.<br />

Chellas,B. (1980). Modallogic:Anintroduction. Cambridge:CambridgeUniversityPress.<br />

Fagin,R.,&Halpern,J.Y. (1988). Belief,awareness,andlimitedreasoning.<br />

ArtificialIntelligence,34,39-76.<br />

Georgeff,M.P.,&Rao,A.S.(1998).DecisionproceduresforBDIlogics.Journal<br />

ofLogicandComputation,8,293–342.<br />

Horty,J.F.,&Belnap,N.D. (1995).Thedeliberativestit:Astudyofaction,<br />

omission,abilityandobligation.JournalofPhilosophicalLogic,24,583–644.<br />

Montague,R.(1970).Universalgrammar.Theoria,36,373–398.


218 CarolineSemmling&HeinrichWansing<br />

Scott,D. (1970). Adviceinmodallogic. InK.Lambert(Ed.),Philosophical<br />

problemsinlogic(pp.143–173).Dordrecht:Reidel.<br />

Semmling,C.,&Wansing,H.(2008).From bdiand stitto bdi-stitlogic.Logic<br />

andLogicalPhilosophy,185–207.<br />

Semmling,C.,&Wansing,H.(2009).Reasoningaboutbeliefrevision.(Toappear<br />

in:E.J.Olsson,S.Rahman,andT.Tulenheimo(eds.),ScienceinFlux,Springer-<br />

Verlag,Berlin.)<br />

Wansing,H.(2006a).Doxasticdecisions,epistemicjustification,andthelogicof<br />

agency.PhilosophicalStudies,128.<br />

Wansing,H.(2006b).Tableauxformulti-agentdeliberative-stitlogic.InG.Governatori,I.Hodkinson,&Y.Venema(Eds.),Advancesinmodallogic(Vol.6,pp.<br />

503–520).London:CollegePublications.<br />

Wooldridge,M.(2000).Reasoningaboutrationalagents.CambridgeMA:MIT<br />

Press.<br />

Xu,M.(1994).Decidabilityofdeliberativestittheorieswithmultipleagents.In<br />

D.M.Gabbay&H.J.Olbach(Eds.),Temporallogic,firstinternationalconference,ICTL’94(pp.332–348).Berlin:Springer-Verlag.<br />

Xu,M. (1998). AxiomsfordeliberativeSTIT. JournalofPhilosophicalLogic,<br />

27,505–552.


A Procedural Interpretation of Split Negation<br />

1 Introduction<br />

Sebastian Sequoiah-Grayson ∗<br />

Takingtheprocedural/dynamicturninthestudyofinformationseriously<br />

meansthatweneedtomakethetransitionfromthestudyofbodiesofinformation,tothestudyofthemanipulationsofsuchbodiesofinformation.<br />

Inthiscase,wewillnotbeabletocarryoutthestudyofinformational<br />

dynamicsbyrestrictingourattentiontobodiesofinformation,orevento<br />

thestructureofthebodiesofinformation,althoughthisisanimportant<br />

component.Wewillalsoneedtopayattentiontotheproceduresviawhich<br />

suchbodiesofinformationarecombinedanddeveloped,andprocessed.<br />

Onerestrictionthatwemightplaceonaparticularstudyofinformational<br />

dynamicsisthatweexamineonlypositiveinformation.Thatis,wemight<br />

restrictourattentiontothepositivefragmentsofvariouslogicsusedto<br />

underpinlogicsofinformationflow. Restrictingourselvestothestudyof<br />

positiveinformationisjustifiableonseveralcounts,nottheleastofwhich<br />

isthatitmakesperfectsensetorestrictourselvestosimplercases,aseven<br />

thesemayturnouttobesurprisinglycomplicated.However,todojustice<br />

tothephenomenaofinformationflow,anyadequatetheoryofinformation<br />

processingwillhavetoallowfortherepresentationofbothpositive,and<br />

negativeinformation. Inthiscase,attentionwillnotberestrictedtothe<br />

positivefragmentofthevariouslogicsusedtounderpinlogicsofinformation<br />

flow.<br />

∗ ManythankstoVladimírSvoboda,MichalPeliˇs,andallbehindLogica2008! This<br />

researchwasmadepossiblebythegeneroussupportoftheHaroldHyamWingateFoundation.ThisresearchwascarriedoutwhilstundertakingaVisitingResearchFellowshipattheTilburgInstituteofLogicandPhilosophyofScience(TiLPS),atTilburgUniversity,TheNetherlands.<br />

IamextremelygratefultoStephanHartmannandeveryone<br />

atTiLPSforthevibrant,research-griendlyatmosphereprovided. Iamalsoextremely<br />

gratefultoEdgarAndrade,JohanvanBenthem,FrancescoBerto,CararinaDutilh,Volker<br />

Halbach,ChristianKissig,GregRestall,HeinrichWansing,andTimWilliamsonformany<br />

invaluablesuggestions.Anyerrorsthatremainaremyown.


220 SebastianSequoiah-Grayson<br />

Thisessayisanargumentforaparticularproceduralinterpretationof<br />

negativeinformation.Inparticular,itisanargumentforaproceduralinterpretationofsplitnegation.Asplitnegationpair<br />

〈∼, ¬〉isdefinableinany<br />

non-commutativelogic.Assuch,aproceduralinterpretationofsplitnegationshouldadoptableinprincipleforanynon-commutativelogic,bethisanon-commutativelinearlogic,oravariantoftheLambekcalculusorwhatever.Accordingly,wewillbeabstractingacrossnon-commutativelogicsin<br />

generalasopposedtolookingindetailatanyonenon-commutativelogicin<br />

particular.However,aninformation–processingapplicationwillbethegeneralmotivation.Fromaphilosophicalstandpoint,theclosestanaloguesare<br />

thenon-commutativelinearlogics,albeitunderproceduralinterpretations.<br />

Linearlogicsweredevelopedinordertotrackresource-use:formulaareunderstoodasresources,andinthiscasenumberoftimestheyoccurbecomes<br />

relevant. Assuch,themarkoflinearlogicsingeneralistherejectionof<br />

contraction. However,iftheformulaaretakentobeconcretedata,then<br />

theaccessibilityoftheseresourcesalsobecomesrelevant.Itisoftenthecase<br />

thatdatahavespatiotemporallocations,suchasinthememoriesofagents<br />

orcomputers,andremotedatawillbelesseasytoaccessthanadjacentdata.<br />

Inamoresensitivelogicofresourcesthen,itisnotonlythemultiplicityof<br />

data,butalsotheirorderthatisimportant.Spatiotemporalobstaclesoften<br />

needtobecircumventedsothatdatamaybeaccessed,hencecommutation<br />

isinappropriatebyvirtueofitsdestroyingtheveryorderingthatwewould<br />

liketopreserve.Insituationswhereactualinformationprocessingisbeing<br />

carriedout,thearrangementofthedataiscrucial(Paoli,2002,28-9).For<br />

recentworkonnon-commutativelinearlogics,see(Abrusci&Ruet,2000),<br />

andforanexplicitlyproceduralexaminationofcommutationinthecontext<br />

ofagent-basedinformationprocessing,see(Sequoiah-Grayson,2009).<br />

Interpretingsplitnegationisaknowndifficulty(Dosen,1993,20). For<br />

anynegationtypetherewillbemorethanonewayofdefiningit.Givena<br />

definition,wethenneedtoprovideaninterpretationoftheresultingnegationintermscommensuratewiththeintendedapplication.Inourcase,the<br />

intendedapplicationistheareaofdynamicinformationprocessing.Given<br />

theproceduralaspect,wewilldefinethenegationof Aintermsof Aimplyingbottom(0).Thisiscommensuratewithinformationprocessingdue<br />

totheimplicationdoingtheworkbeinganalysedinproceduralterms.Sans<br />

theproceduralaspect,aninterpretationofthenegationof Aintermsof A<br />

implying 0goesbackatleastto(Kripke,1965).<br />

Thisessaydevelopsandproposesaparticularinterpretationofthenegationof<br />

Aintermsof Aimplying 0ininformationprocessingterms: In<br />

section2,informationframesandinformationmodelsareintroduced. We<br />

alsointroducethedefinitionofsplitnegation. Theinformationalreading<br />

oftheternaryrelation Rofframesemanticsisintroduced. Insection3,


AProceduralInterpretationofSplitNegation 221<br />

aproceduralinterpretationofsplitnegationunderthedefinitiongivenin<br />

section2isproposed.Uptothispointourexplorationwillhavebeenconductedinpurelymodel-theoreticterms.Itisinsection3thatwetouchontoproof-theoreticalmatters.Thisisessentiallytochecktheproceduralinterpretationagainstaseriesofuniversallyvalidproof-theoreticalproperties<br />

ofsplitnegation.Putsimply,theproposalisthatweinterpretthenegation<br />

of Aintermsoftherulingoutofparticularprocedures,withtheseproceduresbeinganyprocedurethatinvolvescombiningthenegationof<br />

Awith<br />

Aitself.<br />

Thefirststepistointroducethenotionofaninformationframeand<br />

model,sothatwemayspecifyourdefinitionofsplitnegation.<br />

2 InformationFramesandModels<br />

Takeanon-commutativeinformationframe F 〈S, ⊑, •, ⊗, →,←,0〉where S<br />

isasetofinformationstates x,y,...thatmaybeinconsistent,incomplete,<br />

orboth,thebinaryrelation ⊑isapartialorderon Sofinformational<br />

development/inclusion, •isthebinarycompositionoperatoroninformation<br />

statessuchthatduetocommutationfailurewehaveitthat x • y �= y • x,<br />

⊗is(non-commutative)fusion, → and ←arerightandleftimplication<br />

respectively,and 0isbottom. 1 Makingallofthiscleariseasieroncewe<br />

haveamodel.<br />

Amodel M≔〈F,�〉isanorderedpair F 〈S, ⊑, •, ⊗, →,←,0〉and �<br />

suchthat �isanevaluationrelationthatholdsbetweenmembersof Sand<br />

formula.Where Aisapropositionalformula,and x,y,z ∈ F, �obeysthe<br />

hereditycondition:<br />

Forall A,if x � Aand x ⊑ y,then y � A, (1)<br />

Andalsoobeysthefollowingconditionsforeachofourconnectives:<br />

x � A ⊗ Biffforsome y,z, ∈ Fs.t. y • z ⊑ x,y � Aand z � B. (2)<br />

x � A → Biffforall y,z ∈ Fs.t. x • y ⊑ z,if y � Athen z � B. (3)<br />

x � A ← Biffforall y,z ∈ Fs.t. y • x ⊑ z,if y � Athen z � B. (4)<br />

x � 0forno x ∈ F. (5)<br />

1 Anotationalnote: 0iscommonlywrittenas ⊥.Thedifferenceinnotationistoensure<br />

thatnoconfusionismadebetweenbottom,andtheperprelationofincompatibility(Dunn,<br />

1993),(Dunn,1994),(Dunn,1996),writtenas ⊥.Intherecentliteratureonnegation, ⊥<br />

issooftenusedfortheperprelationthatusingitforbottomcreatestoogreatariskfor<br />

misunderstanding.Hence,wefollow(Girard,1987)intheuseof 0forbottom.<br />

Manynon-commutativelogicsarealsonon-associative, suchasthenon-associative<br />

Lambekcalculusamongothers. However,sincenothingthatfollowsdependsoneither<br />

thepresenceorabsenceofassociativity,weshouldbeabletosafelyignorethisissuefor<br />

ourpurposes.


222 SebastianSequoiah-Grayson<br />

Theevaluationrelation �maybeunderstoodindifferentways,dependingonthecontextofapplication.<br />

Forexample,ifweweretobeworking<br />

withlanguageframesandsyntacticallycategorisingparticularalphabetical<br />

strings,wewouldunderstand x � Atomeanstring xisoftype A. We<br />

mightinsteadconsiderascientificresearchprojectwithitsvariousdevelopmentalphases.Inthiscasethedevelopmentrelation<br />

⊑willorderdifferent<br />

statesofaresearchprojectovertime(withtheidealisationthatthereis<br />

noinformation-loss). Herewewouldunderstand x � Atomeanthatthe<br />

proposition Aisknownatstate x,andthatthisparticularstateofdevelopmentintheprojectsupports<br />

A.Wewillinfactreturntothisveryideain<br />

5below.Fornowhowever,weneedsomethingalittlemoregeneral.Along<br />

with(Mares,2009)wewillunderstand x � Atomeanthattheinformation<br />

instate xcarriestheinformationthat A. Hence,wemayalsosaythat x<br />

supportstheinformationthat A. Thisisverysimilartothefamiliarsemanticentailmentrelation<br />

�. Thedifferenceisthatwewanttoallowfor<br />

theinformationatxbeingincompleteand/orinconsistent.Therearemany<br />

applicationswherewemightwanttodothis.Takinginconsistencyasthe<br />

runningexample,considervariousstatesofanagentastheagentreasons<br />

deductively. Inthiscase, xmaysupport Awhere Ais‘pandnot p’,but<br />

thisisdifferentfrom xmaking Atrue,atleastintheusualsenseof“making<br />

true”,asthereisnopossiblewaythattheworldcanbesuchthat xcouldbe<br />

trueofit.Onemightwishtounderstand’supports’as‘makestrue’ifone<br />

holdstoadialethicparaconsistentismwherebyatleastsomecontradictions<br />

aretakentobetrue.However,wewillsidestepthisparticulardebateand<br />

staywiththeinterpretationof’supports’thattakesittobethesubtler<br />

relativeof’makestrue’inthemannerstipulatedabove.<br />

Thereaderfamiliarwiththeternaryrelation Rofframesemanticswill<br />

recognise(2)–(4)astheternaryconditionsfor ⊗, →,and ←respectively,<br />

underanexplicitlyinformationalreading. Rmaybeparsedintermsofthe<br />

twobinaryrelations •and ⊑andsuchthat Rxyzcomesoutas x•y ⊑ z.How<br />

shouldwereadformulascontainingthebinarycompositionrelation •? A<br />

commonandtraditionalwayofunderstandingbinarycompositionissimply<br />

totake x•yas xtogetherwith y.Inthiscase •willbehavemuchthesameas<br />

setunionsuchthat xtogetherwith yisnodifferentto ytogetherwith x,and<br />

xtogetherwithitselfisnodifferentfrom xandsoforth.However,wearenot<br />

restrictedtosuchareadingof •.Thereisingeneralnocanonicalreading<br />

of Rxyz. Thatistosaythatthereisnocanonicalinterpretationofthe<br />

modeltheory.Althoughthisisfrustratingwhenoneencounterstheternary<br />

relationforthefirsttime,itisakeypointwithrespecttotheflexibilityof<br />

ternarysemantics. Inourcase,wehaveitthat •isnon-commutative,so<br />

x•y �= y•x.Internaryterms,non-commutationcomesoutas Rxyz �= Ryxz.<br />

Hence,simplyinterpreting x • yas xtogetherwith ywillblurthevery


AProceduralInterpretationofSplitNegation 223<br />

orderingthatnon-commutationistryingtopreserve. Wecouldstipulate<br />

that“xcomposedwith y”differsfrom“ycomposedwith x”,howeverthis<br />

isslightlystrainedanddoesnotreadstraightoffacasualuseof‘composed’.<br />

Abetterwaytokeepthisdistinctionrobustistoread x•yas xappliedto y.<br />

“Applying”isanorder-sensitivenotion,andonethatfitscomfortablywith<br />

dynamic/proceduraloperations.Sowecanthinkof x•yasthecomposition<br />

of xwith y,wherethiscompositionisorder-sensitive,andwewillmarkthis<br />

order-sensitivitybyspeakingof“application”insteadof“composition”.<br />

Ofcoursewearenotmerelyconcernedwithsyntacticconstructions,as<br />

x,y,z ∈ S,and Sisasetofinformationstates.Weareconcernedwiththe<br />

applicationoftheinformationinonestatetotheinformationinanother.<br />

Onewaythen,ofreading Rxyz,isthat Rxyzholdsifftheresultofapplying<br />

theinformationin xtotheinformationin yiscontainedintheinformation<br />

in z,andthisispreciselywhat x • y ⊑ ztellsus.Anotherwayofputting<br />

thisistosaythattheinformationin zisadevelopmentoftheinformation<br />

resultingfromtheapplicationoftheinformationin xtotheinformationin<br />

y.<br />

Therolethatinformationapplicationplayshereisnotredundant,and<br />

neitherisitmerelytomarkorder-sensitivity.Wearenotsimplyconcerned<br />

withorderedsequencesofinformationstates—somethinglikeanordersensitiveconjunctionwherewewouldhaveonepieceofinformation,then<br />

anotherandthenanotheretc. Weareconcernedwithsomethingmuch<br />

moresubtle. Weareconcernedwiththeinteractionbetweeninformation<br />

states.Thisconcernwithinteraction,orprocess,ispreciselywhyitisthat<br />

weareconcernedwithorder-sensitivityinthefirstplace.Order-sensitivity<br />

isinthissenseameanstoanend,withthisendbeingtheindividuation<br />

ofproceduresofdynamicinformationprocessing. Thissenseof“applied”<br />

carriesoverinanaturalwayfromtheinformationstatesthemselves,tothe<br />

propositionssupportedbytheinformationstates. Itiseasiesttoseethis<br />

withanexample.<br />

Takefusion,anditsframeconditionsgivenin(2).(2)canbeinterpreted<br />

tostatethataninformationstate xcarriestheinformationresultingfrom<br />

theapplicationoftheinformationthat Atotheinformationthat Bifand<br />

onlyif xisitselfadevelopmentoftheapplicationoftheinformationin<br />

state ytotheinformationinstate z,where ycarriestheinformationthat A<br />

and zcarriestheinformationthat B.Thisisalittlelongwinded,andgoing<br />

intherighttolefthanddirectionisalittlemorestraightforward:fortwo<br />

states yand zthatcarrytheinformationthat Aandthat Brespectively,<br />

theapplicationof yto zwillresultinanewinformationstate, x,suchthat x<br />

carriestheinformationthatresultsfromtheapplicationoftheinformation<br />

that Atotheinformationthat B. Theanalogousinterpretationsofthe<br />

frameconditionsforrightandleftimplication((3),and(4),respectively)


224 SebastianSequoiah-Grayson<br />

unpackinasimilarmanner.<br />

Thefusionconnectiveandtheimplicationconnectivesarenotindependent;theyformafamilyofsorts.<br />

Ourfusionandimplicationconnectives<br />

interrelateinthefollowingmanner:<br />

A ⊗ B ⊢ Ciff B ⊢ A → C. (6)<br />

A ⊗ B ⊢ Ciff A ⊢ C ← B. (7)<br />

Indeductiveinformationprocessing,weunderstandthepremisesasdatabasesandtheconsequencerelation‘⊢’astheinformationprocessingmechanism,amorebrutallysyntacticoperationthattheinformationcarrying/supportingof<br />

�. Ininformationalterms,wemayread A ⊢ Basinformationoftype<br />

Bfollowsfrominformationoftype A,ortheinformation<br />

in Bfollowsfromtheinformationin Aetc.Wecanthinkoftypingasencoding,inwhichcasewemightalsoread<br />

A ⊢ Bastheinformationencoded<br />

by Bfollowsfromtheinformationencodedby A.(6)and(7)makesense.<br />

Take(6),startingwiththeleft-to-right-handdirection:Iftheinformation<br />

in C followsfromtheinformationresultingfromtheapplicationofthe<br />

informationin Atotheinformationin B,thenfromtheinformationin<br />

Baloneitfollowsthatwehavetheinformationin Cconditionalonthe<br />

informationin A. Theright-to-left-handdirectionworksoutsimilarly: If<br />

fromtheinformationin Balonewecangettheinformationin Cconditional<br />

ontheinformationin A,thenwecangettheinformationin Cviathe<br />

applicationoftheinformation Atotheinformationin B. Nowtakethe<br />

left-to-right-handdirectionof(7): If,again,theinformationin Cfollows<br />

fromtheinformationresultingfromtheapplicationoftheinformationin A<br />

totheinformationin B,thenfromtheinformationin Aaloneitfollowsthat<br />

wehavetheinformationin C,thistimeconditionalontheinformationin<br />

B.Theright-to-left-handdirectionworksonsimilarlyheretoo:Iffromthe<br />

informationin Aalonewecangettheinformationin Cconditionalonthe<br />

informationin B,thenwecangettheinformationin Cviatheapplicationof<br />

theinformationin Atotheinformationin B.(6)and(7)areinformational<br />

processingversionsofthedeductiontheorem.<br />

Withregardsto(5),noinformation,inanycontextwhatsoever,isof<br />

type 0.Thereisnothingthatwecandotoget 0,and 0isnotsupported<br />

byanyinformationstateinourframe F.<br />

Nowwehavethelogicaltoolsthatweneedinordertobeginlookingat<br />

negativeinformation.Wecandefineasplitnegationpairintermsofdouble<br />

implication:<br />

∼A≔A → 0, (8)<br />

¬A≔0 ← A. (9)


AProceduralInterpretationofSplitNegation 225<br />

Inthiscase,theframeconditionsfor ∼Aand ¬Aarecashedoutinexplicit<br />

informationaltermsasfollows:<br />

x � ∼A[A → 0]iffforeach y,zs.t. x • y ⊑ z,if y � Athen z � 0, (10)<br />

x � ¬A[0 ← A]iffforeach y,zs.t. y • x ⊑ z,if y � Athen z � 0. (11)<br />

Themajorpointssofarhavebeentheinformationaltranslationofthe<br />

ternaryrelation R,suchthat Rxyzcomesoutasas x • y ⊑ z,andthe<br />

definitionofsplitnegationintermsofdoubleimplication,suchthat ∼A≔<br />

A → 0and ¬A≔0 ← A. Thedefinitionalcomponenthereisimportant.<br />

Ourdoubleimplicationconnectives →and ←havetheirconditionsgiven<br />

by R,albeitunderaninformationalreading,in(3)and(4)respectively.<br />

Thismeansthatoursplitnegationconnectives ∼and ¬ultimatelyhave<br />

theirdefinitionsintermsoftheternaryrelationalso.<br />

3 AProceduralInterpretationofSplitNegation<br />

Howshouldweinterpret ∼Aand ¬Agiventheirrespectivedefinitions,<br />

A → 0and 0 ← A? Thetypeofanswerwegiveherewilldependon<br />

thedomain.Forexample,ifwewereworkingwithactions,thenwecould<br />

interpret ∼Aasthetypeofactionthatcannotbefollowedbyanaction<br />

oftype A,andwecouldinterpret ¬Aasthetypeofactionthatcannot<br />

followanactionoftype Aetc.Foranyinterpretationthatwegivetosplit<br />

negation,theinterpretationhastobecompatiblewithcertainproperties<br />

thatholduniversallyforanysplitnegation.Thepurposeofthissectionis<br />

tocheckthepresentlyproposedproceduralinterpretationofsplitnegation<br />

againsttheseproperties,whicharelistedas(12)–(18)below.<br />

Weareworkingwithinformation.Thisisstillfairlygeneralthough,and<br />

variousinformationalapplicationswilllikelyinfluenceourchoiceofinterpretation.Bytakingtheprocedural/dynamicturnandworkingwithinformationflow,theapplicationaspectatworkinbothfusionandthebinary<br />

combinationoperatorgettakenveryseriously.Inthiscase,thesuggestion<br />

isthatweinterpret ∼Aasthebodyofinformationthatcannotbeapplied<br />

tobodiesofinformationoftype A,andthatweinterpret ¬Aasthebody<br />

ofinformationthatcannothavebodiesofinformationoftype Aappliedto<br />

it.Theinterpretationissupportedbythemodeltheory;bytheinformation<br />

statessupporting ∼A, ¬A,and A.If xsupports ∼Aand ysupports A, x<br />

cannotbeappliedto y.Similarly,if xsupports ¬Aand ysupports A,then<br />

ycannotbeappliedto x. Thisisnotbecausesuchanapplicationwould<br />

causeanexplosionofinformation,butbecauseitcouldnevergenerateany<br />

information. Theinterpretationofsplitnegationintermsofrulingout<br />

particularinformationalapplicationsisnotgerrymandered. Itisdirectly<br />

supportedbytheframeconditionsfor ∼Aand ¬A.


226 SebastianSequoiah-Grayson<br />

Toseethis,notethattheframeconditionsininformationaltermsfor<br />

∼Alaidoutin(10)abovetellusthatthereisnoinformationresultingfrom<br />

theapplicationof xto ywhere x � ∼Aand y � A,since x • y ⊑ zand<br />

z � 0(and z � 0nowhere). Supposethoughthatweweretoattemptto<br />

apply ∼Ato A,inotherwordstoattempt ∼A ⊗A.Ininformationalterms,<br />

theframeconditionsforfusion(2)tellusthataninformationstate xwill<br />

support ∼A ⊗ Aiffforsomeinformationstates yand zsuchthat xisan<br />

informationaldevelopmentoftheapplicationoftheinformationin ytothe<br />

informationin z, ysupports ∼Aand zsupports A.However,weknowfrom<br />

ourdefinitionof ∼Aintermsof A → 0,thatthereisnostate xsuchthat<br />

itsupportstheapplicationof ∼Ato A,thisissimplywhat(10)tellsus.<br />

Supportfortherulingoutconditionson ¬Afromtheframeconditions<br />

for ¬Aworkssimilarly,andinvolvesonlyadirectionalchange.Theframe<br />

conditionsfor ¬Alaidoutin(11)abovetellusthatthereisnoinformation<br />

resultingfromtheapplicationoftheinformationstate ytotheinformation<br />

state xwhere y � Aand x � ¬Asince y • x ⊑ zand z � 0(and z �<br />

0nowhere). Ifweweretoattempttoapply Ato ¬A,inotherwords<br />

attempt A ⊗ ¬A,thentherewouldneedtobeaninformationstate xthat<br />

supported A⊗¬A,andthiswouldbethecaseiffthereweresomeinformation<br />

states yand zsuchthat ysupported Aand zsupported ¬Aand xwasan<br />

informationaldevelopmentoftheapplicationof yto z.Fromourdefinition<br />

of ¬Aintermsof 0 ← Ahowever,weknowthatthereisnostate xsuch<br />

thatitsupportstheapplicationof Ato ¬A,thisismarkedoutby(11).<br />

Giventhenon-gerrymanderednatureoftheinterpretationofsplitnegationintermsofproceduralprohibition,weshouldbeabletogiveanaturalinterpretationofgeneralproof-theoretic,henceinformation–processingpropertiesofsplitnegationinsuchterms.Foranysplitnegation,independentlyofwhichstructuralrulesarepresent,thefollowing(12)–(18)hold:<br />

A ⊢ B<br />

∼B ⊢ ∼A<br />

(12)<br />

(12)makessenseintermsoftherulingoutofinformationprocessingprocedures.Givenasplitnegation,andgivenalsothatinformationoftype<br />

B<br />

followsfrominformationoftype A,thenrulingouttheprocedure ∼A ⊗ A<br />

followsfromrulingouttheprocedure ∼B ⊗ B. Thisisjusttosaythat<br />

giventhatwecangetinformationoftype Bfrominformationoftype A,<br />

thenfromthebodyofinformationthatcanneverbeappliedtobodiesof<br />

type B,wecangetthebodyofinformationthatcanneverbeappliedto<br />

bodiesofinformationoftype A.<br />

A ⊢ B<br />

¬B ⊢ ¬A<br />

(13)<br />

Thereasoningwithregardsto(13)isdirectlyanalogoustothatsurrounding<br />

(12):Againgivenasplitnegation,andagaingiventhatinformationoftype


AProceduralInterpretationofSplitNegation 227<br />

Bfollowsfromtheinformationoftype A,thenrulingouttheprocedure<br />

A ⊗ ¬Afollowsfromrulingouttheprocedure B ⊗ ¬B.Thisisjusttosay<br />

thatgiventhatwecangetinformationoftype Bfrominformationoftype<br />

A,thenfromthebodyofinformationthatcanneverhavebodiesoftype<br />

Bappliedtoit,wecangetthebodyofinformationthatcanneverhave<br />

bodiesofinformationoftype Aappliedtoit. Thereasoningsurrounding<br />

(14)and(15)isslightlymoreinvolvedthanin(12)and(13).<br />

A ⊢ ∼B<br />

B ⊢ ¬A<br />

B ⊢ ¬A<br />

A ⊢ ∼B<br />

Wecantake(14)and(15)together,gettingthesplitnegationproperty:<br />

(14)<br />

(15)<br />

A ⊢ ∼Biff B ⊢ ¬A. (16)<br />

Startingwiththeleft-to-right-handdirection: Ifwecan,onthebasisof<br />

informationoftype Aalone,getthebodyofinformationthatcanneverbe<br />

appliedtobodiesofinformationoftype B,thenonthebasisofinformation<br />

oftype Balone,wecangetthebodyofinformationthatcanneverhave<br />

bodiesofinformationoftype Aappliedtoit.Theintermediatestepisthis:<br />

Ifweweretoapply Ato B(i.e. A ⊗ B)thenwewouldgetnothing,viz.<br />

0,since A ⊗ B ⊢ 0,sinceif A ⊢ ∼Bthen A ⊗ B ⊢ 0. Assuch,from<br />

informationoftype Balonewecangetthebodyofinformationthatcan<br />

neverhavebodiesofinformationoftype Aappliedtoit. Theright-toto-left-handdirectionissimilar:<br />

Ifwecan,onthebasisofinformationof<br />

type Balone,getthebodyofinformationthatcanneverhavebodiesof<br />

informationoftype Aappliedtoit,thenweretoapply Bto A(i.e. B ⊗ A)<br />

thenwewouldgetnothing,viz. 0,since B ⊗ A ⊢ 0,sinceif B ⊢ ¬Athen<br />

B ⊗ A ⊢ 0.Assuch,thenfrominformationoftype Aalonewecangetthe<br />

bodyofinformationthatcanneverbeappliedtobodiesofinformationof<br />

type B.(17)and(18)aremorestraightforward:<br />

A ⊢ ¬∼A, (17)<br />

A ⊢ ∼¬A. (18)<br />

Onthebasisofinformationoftype Aalone,wecanruleoutthebodyof<br />

informationthatcanneverbecomposedwithbodiesofinformationoftype<br />

A.Thisisjusttosaythatwecanruleout A → 0.Thisisjustwhat(17)<br />

statesunderaprocedurallyfocusedinformationalinterpretation.Similarly,<br />

onthebasisofinformationoftype Aalone,wecanalsoruleoutthebody<br />

ofinformationthatcanneverhavebodiesofinformationoftype Aapplied<br />

toit.Inthiscasewearerulingout 0 ← A.Inthiscontext,“rulingout”is<br />

aformofproceduralprohibition.


228 SebastianSequoiah-Grayson<br />

Theproposalforaproceduralinterpretationofsplitnegationhasbeen<br />

thatweinterpret ∼A(thatis A → 0)asthebodyofinformationthat<br />

cannotbeappliedtobodiesofinformationoftype A,andthatweinterpret<br />

¬A(thatis 0 ← A)asthebodyofinformationthatcannothavebodiesof<br />

informationoftype Aappliedtoit.Wehaveseenthatthisinterpretationof<br />

splitnegationisentirelynaturaloncewetranslatetheternaryrelation Rinto<br />

itsinformationalform,inwhichcasetheinterpretationisdirectlysupported<br />

bytheframeconditionsfor ∼Aand ¬A,(10)and(11)respectively.Wehave<br />

alsoseenthattheinterpretationiscompatiblewiththeuniversallyvalidsplit<br />

negationproperties(12)–(18).<br />

4 Conclusion<br />

Wehaveseenhowitisthatwemayreconstructtheternaryrelationofframe<br />

semantics, Rxyzinexplicitlydynamicinformationalterms,as x • y ⊑ z.<br />

Thisdynamicinformationalreconstructioncarriesovertoanyconnective<br />

definedintermsoftheternaryrelation,allowingustogiveexplicitlyproceduralaccountsofdoubleimplicationandfusion.Sincewehaveuseddouble<br />

implicationtodefineasplitnegation, ∼A≔A → 0,and ¬A≔0 ← A,we<br />

haveaproceduraldefinitionofsplitnegation.<br />

Giventhedefinitionofsplitnegationinthesedynamicinformational<br />

terms,wehavebeenableto“readoff”anaturalproceduralinterpretation<br />

ofsplitnegation.Thisinterpretationhasbeenshowntobecompatiblewith<br />

theuniversallyvalidpropertiesofasplitnegation.<br />

SebastianSequoiah-Grayson<br />

FormalEpistemologyProject<br />

IEG–ComputingLaboratory–UniversityofOxford<br />

sebsequoiahgrayson@hiw.kuleuven.be<br />

http://users.ox.ac.uk/∼ball1834/index.shtml<br />

References<br />

Abrusci,V.M.,&Ruet,P.(2000).Non-commutativelogic,I:themultiplicative<br />

fragment.Ann.PureAppl.Logic,101(1),29–64.<br />

Dosen, K. (1993). A historical introduction to substructural logics. In<br />

P.Schroeder-Heister&K.Dosen(Eds.),Substructurallogics,studiesinlogicand<br />

computationno.2(pp.1–30).Oxford:ClarendonPress.<br />

Dunn,J.M. (1993). Partialgagglesappliedtologicswithrestrictedstructural<br />

rules.InP.Schroeder-Heister&K.Dosen(Eds.),Substructurallogics,studiesin<br />

logicandcomputationno.2(pp.63–108).Oxford:ClarendonPress.


AProceduralInterpretationofSplitNegation 229<br />

Dunn,J.M.(1994).Starandperp:Twotreatmentsofnegation.InJ.E.Tomberlin(Ed.),<br />

Philosophical perspectives (Vol.7, pp.331–357). Atascadero,CA:<br />

Ridgeview.<br />

Dunn,J.M.(1996).Generalisedorthonegation.InH.Wansing(Ed.),Negation:<br />

Anotioninfocus(pp.3–26).Berlin:deGruyter.<br />

Girard,J.-Y.(1987).Linearlogic.TheoreticalComputerScience,50,1–101.<br />

Kripke,S.A.(1965).SemanticalanalysisofintuitionisticlogicI.InJ.Crossley<br />

&M.Dummett(Eds.),Formalsystemsandrecursivefunctions(pp.92–129).<br />

Amsterdam:North-Holland.<br />

Mares,E. (2009). Generalinformationinrelevantlogic. (ForthcominginSynthese,section:Knowledge,Rationality,andAction,L.FloridiandS.Sequoiah-Grayson(eds.):ThePhilosophyofInformationandLogic,Synthese,KRA,ProceedingsofPIL–07,TheFirstWorkshoponthePhilosophyofInformationand<br />

Logic,UniversityofOxford,November3–4,2007.)<br />

Paoli,F. (2002). Substructurallogics: Aprimer. Berlin–NewYork: Spinger<br />

Verlag.<br />

Sequoiah-Grayson,S.(2009).Apositiveinformationlogicforinferentialinformation.(ForthcominginSynthese,section:Knowledge,Rationality,andAction,L.<br />

FloridiandS.Sequoiah-Grayson(eds.):ThePhilosophyofInformationandLogic,<br />

Synthese,KRA,ProceedingsofPIL–07,TheFirstWorkshoponthePhilosophy<br />

ofInformationandLogic,UniversityofOxford,November3–4,2007.)


1 Theproblem<br />

Reference to Indiscernible Objects<br />

Stewart Shapiro ∗<br />

Somecriticsofmyanteremstructuralism(Shapiro,1997)arguethatI<br />

haveanissuewithstructuresthathaveindiscernibleplaces. 1 Astructure<br />

issaidtobe“rigid”ifitsonlyautomorphismisthetrivialonebasedonthe<br />

identitymapping.Themainexemplarsoftheallegedproblemarenon-rigid<br />

structures.Itisaneasytheoremthatisomorphicstructuresareequivalent:<br />

Let fbeanautomorphismonagivenstructure Mandlet Φ(x1,...,xn)<br />

beanyformulainthelanguageofthestructure. Thenforanyobjects<br />

a1,... ,aninthedomainof M, Msatisfies Φ(a1,... ,an)ifandonlyif M<br />

satisfies Φ(fa1,...,fan).If fisanon-trivialautomorphism,thenthereis<br />

anobject asuchthat fa �= a. Inthiscase, aand faareindiscernible,at<br />

leastconcerningthelanguageofthestructure:anythingtrueofoneofthem<br />

willbetrueoftheother.Sonon-rigidstructureshaveindiscernibleobjects.<br />

Themost-citedexampleisthatofcomplexanalysis.Startwiththelanguageoffields,andconsiderthealgebraicclosureofthereals,whichis<br />

uniqueuptoisomorphism(initssecond-orderformulation).Thecomplex<br />

numbersaretheintendedmodel.Thefunctionthattakesacomplexnumber<br />

a + bitoitsconjugate a − biisanautomorphism. Itfollowsthatfor<br />

anyformula Φ(x),withonly xfree, Φ(a + bi)ifandonlyif Φ(a − bi). In<br />

particular, Φ(i)ifandonlyif Φ(−i). So iand −iareindiscernible;they<br />

∗ Igaveearlyversionsofpartsofthispaperatthephilosophyofmathematicsworkshop<br />

atOxford,theArchéResearchCentreattheUniversityofSt. Andrews,OhioState<br />

UniversityandtheUniversityofMinnesota. Thankstoalloftheaudiencesthere. I<br />

amindebtedtoCraigeRoberts,GabrielUzquiano,OfraMagidor,CathyMüller,Dan<br />

Isaacson,GrahamPriest,KevinScharp,RobertKraut,andJasonStanley.<br />

1 Theearlycriticsinclude(Burgess,1999,pp.287–288),(Hellman,2001,pp.192–196),<br />

and(Keränen,2001,2006). Morerecentparticipantsinthedebateinclude(Ladyman,<br />

2005),(Button,2006),(Ketland,2006),(MacBride,2005, §3),(MacBride,2006b),and<br />

(Leitgeb&Ladyman,2008). Myowncontributionsinclude(Shapiro,2006b),(Shapiro,<br />

2006a),and(Shapiro,2008).


232 StewartShapiro<br />

havethesameproperties,atleastamongthosethatcanbeexpressedin<br />

thelanguage. Anotheroft-citedexampleisEuclideanspace,wherethings<br />

areevenworse.AnytwopointsinEuclideanspacecanbeconnectedwith<br />

alineartranslation,whichisanautomorphism. So,itseems,allofthe<br />

pointsinEuclideanspaceareindiscernible,atleastwithrespecttopropertiesthatcanbeexpressedinthelanguageofgeometry.HannesLeitgeband<br />

JamesLadyman(2008)pointoutthatsincesome(simple)graphshaveno<br />

relations,anybijectiononthemisanisomorphism.Sowiththosegraphs,<br />

everypointisindiscerniblefromeveryother.Thesimplestofthesesimple<br />

graphsareisomorphictothefinitecardinalstructuresintroducedinmy<br />

chapteronepistemologyin(Shapiro,1997,Ch.4).<br />

Whyisthisaproblemforanteremstructuralism? Someill-chosenremarksinmybookatleastsuggestaprincipleoftheidentityofindiscernibles,<br />

which,inlightofexampleslikethese,wouldreducetheviewtoabsurdity.<br />

I’dbecommittedtosayingthat i = −i,andthatthereisonlyonepointin<br />

Euclideanspace. Butthereislittlepointintryingtofigureoutwhatmy<br />

viewwas.Irejecttheidentityofindiscerniblesnow.<br />

Muchofthediscussionofthisissueismetaphysical.JohnBurgess(1999,<br />

p.288)pointsoutthatalthoughthetwocomplexrootsof −1aredistinct,<br />

onmyview“thereseemstobenothingtodistinguishthem.”Thisseemsto<br />

invokesomethingintheneighborhoodofthePrincipleofSufficientReason.<br />

Ifsomethingisso,thentheremustbesomethingthatmakesitso,oratleast<br />

somethingthatexplainswhyitisso. JukkaKeranänen(2001)articulates<br />

ageneralmetaphysicalthesisthatanyonewhoputsforwardatheoryof<br />

atypeofobjectmustprovideanaccountofhowthoseobjectsaretobe<br />

“individuated”.AccordingtoKeränen,foreachobject ainthepurviewof<br />

aproposedtheory,wehavetobetold“thefactofthematterthatmakes a<br />

theobjectitis,distinctfromanyotherobject”ofthetheory,by“providing<br />

auniquecharacterizationthereof.”<br />

Someauthorsenteredthediscussion,onmybehalf,bysuggestingmetaphysicalprinciplesthatareweakerthanKeränen’sindividuationrequirementbutstillmeetBurgess’sdemandthatthetheoristfindsomethingthat<br />

distinguishesdistinctobjects.Theidea,itseems,isthatonecandistinguish<br />

objectswithoutindividuatingthem.Theweakestoftheserequirementsis<br />

athesisthatforany a, b,if a �= bthenthereisanirreflexiverelation R<br />

suchthat Rab(Ladyman,2005).ComplexanalysisandEuclideangeometry<br />

easilypassthistest. Forexample, iand −isatisfytheirreflexiverelation<br />

ofbeingadditiveinversetoeachotheranddistinctfrom 0,andanypairof<br />

distinctpointsinEuclideanspacesatisfytheirreflexiverelationofdeterminingexactlyonestraightline.Nevertheless,thefinitecardinalstructuresand<br />

somegraphsstillfailthetest,unlessnon-identitycountsasanirreflexive<br />

relation(inwhichcase,ofcourse,wedonothaveasubstantialtest).


ReferencetoIndiscernibleObjects 233<br />

Iwishtoputasidethesemetaphysicalmattershere,atleastasfaras<br />

possible.Therearesomerelatedand,Ithink,moreinterestingissuesconcerningthesemanticsandpragmaticsofmathematicallanguages,andperhapslanguagesgenerally.Theseissuesalsobearonlogic,andtheygowellbeyondlocaldisputesconcerninganteremstructuralism.Howdowemanagetotalkabout,andthus,insomesense,refertoindiscernibleobjects?<br />

Idonotintendtoofferadetailedsolutiontothissemanticproblemhere,<br />

justtohighlightitandindicateitsgenerality.Asolution,Ibelieve,would<br />

involveanextendedforayintolinguistics,thephilosophyoflanguage,and<br />

thephilosophyoflogic.<br />

Tobesure,thewholeprojectpresupposesthatthereareindiscernible<br />

objects,andthispresupposesthatthemetaphysicalprinciplesadoptedby<br />

someofmyopponentsarefalse. Ialsodonotintendtoargueforthatin<br />

detailhere(butsee(Shapiro,2008)andrelatedworkcitedthere). The<br />

informallanguageofcomplexanalysishasaterm iwhichissupposedto<br />

denoteoneofthesquarerootsof −1.Atleastgrammatically, iisaconstant,<br />

apropername.Andtheroleofaconstantistodenoteasingleobject—at<br />

leastinasufficientlyregimentedlanguage. Butwhichofthesquareroots<br />

does idenote?Isitnotasifthemathematicalcommunityhasmanagedto<br />

singleoutoneoftheroots,inordertobaptizeitwiththename“i”.They<br />

cannotdoso,asthetworootsareindiscernible.<br />

GottlobFrege(1884)seemstohavenotedourproblem:<br />

Wespeakof‘thenumber 1’,wherethedefinitearticleservestoclass<br />

itasanobject(§57). If,however,wewishedtouse[a]conceptfor<br />

defininganobjectfallingunderit[byadefinitedescription],itwould,<br />

ofcourse,benecessaryfirsttoshowtwodistinctthings:<br />

thatsomeobjectfallsundertheconcept;<br />

thatonlyoneobjectfallsunderit(§74n).<br />

Nothingpreventsusfromusingtheconcept‘squarerootof −1’;but<br />

wearenotentitledtoputthedefinitearticleinfrontofitwithout<br />

moreadoandtaketheexpression‘thesquarerootof −1’ashavinga<br />

sense.(§97)<br />

Complexanalysisisperhapsthemostsalientexampleofthelogicalsemanticphenomenoninquestion,butthereareasomeothers,atleast<br />

ifwegowithastraightforwardreadingofvariousmathematicallanguages<br />

(see(Brandom,1996)). Consider,forexample,theintegers,withaddition<br />

astheonlyoperation. Itis,ofcourse,anAbeliangroup,whoseelements<br />

are:<br />

... , −3, −2, −1,0,1,2,3,...


234 StewartShapiro<br />

Intherelevantlanguage,theoperationthattakesanyinteger ato −ais<br />

anautomorphism. Soanythingintherelevantlanguagethatholdsofan<br />

integer aalsoholdsof −a. Inthisstructure,then, 1isindiscerniblefrom<br />

−1,but,ofcourse, 1isdistinctfrom −1. AnotherexampleistheKlein<br />

group. Ithasfourelements,whichareusuallycalled e, a, b,and c,and<br />

thereisoneoperation,givenbythefollowingtable:<br />

e a b c<br />

a e c b<br />

b c e a<br />

c b a e<br />

Itiseasytoverifythatanyfunction fthatisapermutationonthefour<br />

elementssuchthat fe = eisanautomorphism. Thethreenon-identity<br />

elementsarethusindiscernible,inthelanguageofgroups,andyetthereare<br />

threesuchelementsandnotjustone.Butwhichoneis a?<br />

Categorytheoryisrampantwithexamplesofthephenomenonunder<br />

studyhere. Themainreasonforthis,itseems,isthateverycategorical<br />

notionispreservedunderisomorphism.Totakeoneinstance,anobject O<br />

inacategoryiscalledterminal,ifforanyobject Ainthecategory,there<br />

isexactlyonemapfrom Ato O.Incategorieswithaterminalobject,itis<br />

commontointroduceaterm“1”forsuchanobject. Itistrivialtoshow<br />

thatanytwoterminalobjectsareisomorphic.Indeed,if 1and 1 ′ areboth<br />

terminal,thenthereisexactlyonemapfrom 1to 1 ′ andexactlyonemap<br />

from 1 ′ to 1.Thesetwomapsmustcomposetoanidentitymap—either<br />

theuniquemapfrom 1to 1,ortheuniquemapfrom 1 ′ to 1 ′ ,dependingon<br />

theorderofcomposition. Moreover,anyobjectisomorphictoaterminal<br />

objectisitselfterminal. Soifacategoryhasaterminalobject,itusually<br />

hasmany. Inthecategoryofsets,forexample,anysingletonisterminal.<br />

Whichofthemistheterminalobjectofthecategory,theonepickedoutby<br />

theterm“1”?Theanswer,ofcourse,isthatitdoesnotmatter—justasit<br />

doesnotmatterwhichsquarerootof −1is i.And,heretoo,“1”functions<br />

asasingularterm,atleastasfarassyntaxgoes.<br />

Inacategorywithaterminalobject,itiscommontodefineanelement<br />

ofanobject Atobeamapfrom 1to A.So,itwouldseem,toknowwhich<br />

mapsaretheelementsof A,wehavetoknowwhichobjectis 1.Inasense,<br />

wecan’tknowthis,but,again,itdoesnotmatter.Inlikemanner,aproduct<br />

oftwoobjects A, Bisanobject,usuallywritten A ×B,andapairofmaps,<br />

onefrom A × Bto Aandonefrom A × Bto B,thatsatisfiesacertain<br />

universalproperty. Again,productsarenotusuallyunique: anyobject


ReferencetoIndiscernibleObjects 235<br />

isomorphictoaproductof Aand Bcanitselfbeshowntobeaproduct<br />

of Aand B.Nevertheless,the“×”symbolseemstobeafunctionsymbol,<br />

anditiscommontotalkabout A × Bastheproductof Aand B—just<br />

asitiscommontotalkabout iasthesquarerootof −1.<br />

Thereisarelatedphenomenonconcerningtheuseparametersorfreevariablesinmathematicaldiscourse.Thoseactlikesingulartermsincontext,<br />

butoftenfailuniqueness.Supposethatinthecourseofademonstration,a<br />

geometersays“let ABCDbeanyparallelogram,withtheline ABcongruentandparalleltotheline<br />

CD.” Itfollowsthatthepairofpoints A, B,<br />

(andthelinesegment AB)areindiscerniblefromthepair C, D(andthe<br />

segment CD).Anythingonecansayaboutoneofthepairs(andoneofthe<br />

segments)willbetrueoftheotherpair(andtheothersegment).Sowhich<br />

oneis ABandwhichoneis CD?<br />

2 Indiscernibility,semantics,andexpressiveresources<br />

Tosaythattwoobjectsareindiscernibleistosaythattheycannotbetold<br />

apart. Thisbriefcharacterizationsuggeststhatindiscernibilityisrelative.<br />

Twoballsmaybevisuallyindiscernibletosomeonewhoiscolorblindwhile<br />

beingdiscernibletosomeonewithmorenormalvision.Inthepresentcontext,indiscernibilityisrelativetoexpressiveresources.Twomathematical<br />

objectsmaybeindiscerniblewithsomebatchofresources,butdiscernible<br />

onceexpressiveresourcesareadded. Inthecaseoftheintegers,forexample,<br />

1and −1arediscernibleifonecaninvokemultiplication: 1isthe<br />

multiplicativeidentityand −1isnot.<br />

Attheoutset,Iformulatedtheissueintermsofrigidstructures,with<br />

“rigidity”definedintermsofautomorphisms.Thisregisterstherelativity<br />

toexpressiveresourcesaswell,since“automorphism”isitselfdefinedin<br />

termsofthebackgroundlanguage: alloftheprimitiverelationsmustbe<br />

preserved.<br />

Suppose,then,thatwejustaddaconstant itotheofficiallanguageof<br />

complexanalysis,withtheobviousaxiom i 2 = −1. Then,technically,the<br />

structurebecomesrigid:therearenonon-trivialautomorphisms.Thereasonisthatisomorphismsmustpreserveallofthestructureinthelanguage,<br />

and,inparticular,itmustpreservethedenotationsoftheconstants. If f<br />

isanisomorphismbetween M1and M2,inthelanguageofarithmetic,for<br />

example,theniftheconstant“0”denotes ain M1,then“0”mustdenote fa<br />

in M2.Similarly,let Nbeanymodelofcomplexanalysis,intheenvisioned<br />

languagewithaconstant“i”,andlet fbeanautomorphism.If“i”denotes<br />

ain N,then“i”denotes fain N.Thatis, a = fa.Itfollowsthatforeach<br />

element binthedomain, fb = b.So fistrivial.


236 StewartShapiro<br />

Nevertheless,itseemstomethat,intherelevantintuitivesense,thetwo<br />

squarerootsof −1arestillindiscernible,eveninthislanguage.Let N ′ be<br />

amodelofthetheorythatisjustlike N,exceptthatin N ′ ,“i”denotes −a<br />

(andthus“−i”denotes a).Technically, Nisnotisomorphicto N ′ ,forthe<br />

abovereason.However,itseemstomethatthetwomodelsareequivalent,<br />

intheintuitivesense. Bothhavethesamedomainandtheyagreeonthe<br />

operations.Inparticular,ineachmodel,thesametwoobjectsaretheroots<br />

of −1. Theonlydifferencebetweenthemisthat Ncallsoneofthem“i”:<br />

and N ′ callstheotherone“i”.Itseemstomethatthisisnotasignificant<br />

difference—notunlessweaddsomestructuretothenamingrelation.<br />

Todevelopthispointabit,letusgoupalevel,sotospeak,andthink<br />

ofthesemanticrelationsthemselvesinformal,orstructuralterms. Begin<br />

withasimplegraphthathastwonodesandnoedges. Asnotedabove,<br />

thisstructureiscompletelyhomogeneous.Nowaddtwonewobjects, a, b,<br />

andarelation Rtothestructure. Thenewitem abears Rtooneofthe<br />

nodesintheoriginalgraphand bbears Rtotheothernode. Thisisthe<br />

structureofsomeverysimplesemanticrelationsonthegraph:thinkof a<br />

and basnames,and Rasthereferencerelation.Thismathematical-cumsemanticstructureisnotrigid.Ifwemodifyitbyswitchingthe“referents”<br />

of aand b,wegetanautomorphism. And,intuitively,wehavenotreally<br />

changedtheextendedstructurewiththisswitch.Itisstillthesamesimple<br />

graphwiththesametwonewobjects,thesamerelation R,andthesame<br />

structural–semanticrelations.<br />

Wecandothesamewithourmorestandardmathematicalexample.<br />

Considerastructure Mthatincludestheplacesandrelationsofourmodel<br />

N,ofcomplexanalysis.Inaddition, Mhasnodesrepresentingtheprimitive<br />

terms ofthelanguageofcomplexanalysis(“0”,“i”,“+”,“A”),anda<br />

relation Rrepresentingreference.So,forexample, Rixwouldbeanatomic<br />

formulaintheenvisionedobjectlanguage,sayingthattheconstant“i”refers<br />

to,ordenotes, x.Thetheorywouldincludetheaxiomsofcomplexanalysis<br />

(over N)andtheTarskiansatisfactionclausesbetweenthe“terms”(“0”,<br />

“i”,“+”,“A”)andtherelevantitemsconstructedfrom N.So,forexample,<br />

ourstructurewouldsatisfy ∀x∀y((Rix&Riy) → x = y)and Ria(recalling<br />

that aisoneofthesquarerootsof −1in N).<br />

Thismathematical-cum-semanticstructureisnotrigid. Thefunction<br />

thattakes x+ayto x−ay(within N),takeseach“term”(“0”,“1”,“i”,“+”,<br />

“A”)toitself,andadjuststherelation Raccordingly,isanautomorphism.<br />

Westillhaveamodelofcomplexanalysis,asabove,andalloftheTarskian<br />

satisfactionclausesarestillsatisfied(see(Leitgeb,2007,pp.133–134)).The<br />

problem,here,istosaysomethingaboutthesemanticsandlogicofthe<br />

languagesofmathematics,soconstrued.


ReferencetoIndiscernibleObjects 237<br />

3 Theidentityofindiscernibles<br />

TheissuesherearerelatedtothoseinMaxBlack’s(1952)celebrateddiscussionoftheidentityofindiscernibles.Thepaperisintheformofadialogue.<br />

Onecharacter, A,takestheidentityofindiscerniblestobe“obviouslytrue”,<br />

whiletheopponent, B,takesittobe“obviouslyfalse”.Thelattergivesa<br />

thoughtexperimentmeanttorefutetheprincipleinquestion:<br />

Isn’titlogicallypossiblethattheuniverseshouldhavecontainednothingbuttwoexactlysimilarspheres?Wemightsupposethateachwas<br />

madeofchemicallypureiron,hadadiameterofonemile,thatthey<br />

hadthesametemperature,color,andsoon,andthatnothingelse<br />

existed. Theneveryqualityandrelationalcharacteristicoftheone<br />

wouldalsobeapropertyoftheother.(Black,1952,p.156)<br />

Black’stwospheresareanalogoustothetwosquarerootsof −1.Ofcourse,<br />

Iamnotclaimingthatthereisapossibleworldwhichconsistsofjustthese<br />

twocomplexnumbers. Buttherestoftheanalogyholds,atleastinthe<br />

languageofcomplexanalysis.<br />

Themainthrustof(Black,1952)ismetaphysicaland,asnotedabove,<br />

suchmattersarebeingputasidehereasmuchaspossible.Alongtheway,<br />

however,thearticlebroachesthelogico-semanticissuesofpresentconcern.<br />

Thedefenderoftheindiscernibilityofidenticals, A,firstdeniesthat Bhas<br />

describedacoherentpossibility,andthencontinues,“Butsupposingthat<br />

youhavedescribedapossibleworld,Istilldon’tseethatyouhaverefuted<br />

theprinciple.Consideroneofthespheres, a.”Atthispoint, Binterrupts,<br />

protesting:“HowcanI,sincethereisnowayoftellingthemapart?Which<br />

onedoyouwantmetoconsider?”.Thatis, Brefusestolethisopponentuse<br />

avariable,aparameter,orasingulartermforoneofthespheres.Character<br />

Aresponds:“Thisisveryfoolish.Imeaneitherofthetwospheres,leaving<br />

youtodecidewhichoneyouwishedtoconsider.” Inourcase,itstrikes<br />

measeminentlyreasonabletosay,“let idesignateoneofthesquareroots<br />

of −1. Idon’tcarewhich.” Thepresentproblemistomakesenseofthis<br />

locution.Character Bdeniesthatitcanbemadesenseof.<br />

RobertBrandom(1996,p.298)putsourprobleminsimilarterms:<br />

Nowifweaskamathematician‘Whichsquarerootof −1is i?,’she<br />

willsay‘Itdoesn’tmatter:pickone.’Andfromamathematicalpoint<br />

ofviewthisisexactlyright.Butfromthesemanticpointofview,we<br />

havetherighttoaskhowthistrickisdone—howisitthatIcan<br />

‘pickone’ifIcan’ttellthemapart?WhatmustIdoinordertobe<br />

pickingone,andpickingone? Forwereallycannottellthemapart<br />

—and... notjustbecauseofsomelamentableincapacityofours.


238 StewartShapiro<br />

ThenextexchangeinBlack’sdialogueputssomedetailtothediffering<br />

semanticpresuppositions.Character A,theproponentofthe(obvioustruth<br />

of)theidentityofindiscernibles,continues,“IfIweretosaytoyou‘Take<br />

anybookofftheshelf’itwouldbefoolishonyourparttoreply‘Which?’.”<br />

Bretorts:“It’sapooranalogy.Iknowhowtotakeabookoffashelf,but<br />

Idon’tknowhowtoidentifyoneofthetwospheressupposedtobealone<br />

inspace...” Itseemsthat,forthepurposesofthisargument, Bclaims<br />

thatonecannotuseasingulartermtodesignateanobjectwithoutfirst<br />

“identifying”it,oratleastknowinghowtoidentifyoneoftheobjectsin<br />

question.That,Itakeit,isthematterathandhere,whetheritispossible<br />

tointroducealexicalitemtorefertoanindiscernibleobject.Thecharacter<br />

Atakesthebait:“Can’tyouimaginethatonespherehasbeendesignated<br />

as‘a’?”Thedialoguecontinues:<br />

B. Icanimagineonlywhatislogicallypossible. Nowitislogically<br />

possiblethatsomebodyshouldentertheuniverseIhavedescribed,<br />

seeoneofthespheresonhislefthandandproceedtocallit‘a’...<br />

A.Verywell,nowletmetrytofinishwhatIbegantosayabout a...<br />

[ellipsisinoriginal]<br />

B. Istillcan’tletyou,becauseyou,inyourpresentsituation,have<br />

norighttotalkabout a.AllIhaveconcededisthatifsomethingwere<br />

tohappentointroduceachangeinmyuniverse,sothatanobserver<br />

enteredandcouldseethetwospheres,oneofthemcouldthenhave<br />

aname. ButthiswouldbeadifferentsuppositionfromtheoneI<br />

wantedtoconsider.Myspheresdon’tyethavenames... Youmight<br />

justaswellaskmetoconsiderthefirstdaisyinmylawnthatwould<br />

bepickedbyachild,ifachildweretocomealonganddothepicking.<br />

Thisdoesn’tnowdistinguishanydaisyfromtheothers.Youarejust<br />

pretendingtouseaname.<br />

A.AndIthinkyouarejustpretendingnottounderstandme.<br />

4 Isthisaproblem?<br />

Whatseemstomatterhere,ataminimum,isone’sphilosophyofmathematics,andone’saccountofreference.<br />

Ifsomeonehasaphilosophyof<br />

mathematicsthatacceptsaprincipleoftheidentityofindiscernibles,and<br />

alsoacceptsacertainnaiveaccountofwhatindiscernibilitycomesto,then<br />

hewillnotallowtheforegoing,implicitcharacterizationofthecomplex<br />

numbersasthealgebraicclosureofthereals(orasthestructurecharacterizedbythestandardaxiomatization).<br />

Thatverycharacterizationviolates<br />

theidentityofindiscernibles,sinceitintroducestwodistinctobjectsthat<br />

cannotbetoldapart. Thesamegoesfortheotherexamples,theintegers


ReferencetoIndiscernibleObjects 239<br />

underaddition,theKleingroup,andallofcategorytheory.Thephilosopher<br />

whoacceptsthesestrictureswillnotfacetheforegoingproblemofreference<br />

—sincethereisnosuchproblemonthatview—butshewillneedsome<br />

otheraccountofthevariousstructuresandtheories.<br />

Onewaytoavoidtheproblemistobreakthesymmetry.Whetherone<br />

acceptstheidentityofindiscerniblesornot,onecanthinkofcomplexnumbersaspairsofreals,followinganowcommonmathematicaltechnique.In<br />

thatcase, iisthepair 〈0,1〉and −iis 〈0, −1〉.Sinceonecandistinguish 1<br />

from −1inthereals,andthusinthesecondcoordinate,thereisnoproblem<br />

distinguishingthosetwopairs.<br />

Theproblemwithcomplexanalysis,however,atleastappearstobe<br />

robust—liabletoreappear. Thecustomnowadaysistousepolarcoordinates,inwhichcase<br />

iis � 1, π<br />

� � �<br />

3π<br />

2 and −iis 1, 2 . Soitnowbecomesa<br />

matterofdistinguishingthosepairsfromeachother.This,inturn,becomes<br />

amatterofdistinguishingtheangle π<br />

3π<br />

2fromtheangle 2 . Andhowdoes<br />

onedothat? Well,wecansaythatthefirstispositiveandthesecondis<br />

negative;orthatthefirstgoescounterclockwiseandthesecondclockwise;<br />

orthatthefirstisabovethe x-axisandthesecondbelowit. Butwhen<br />

itcomestoangles,“positive–negative”,“counterclockwise–clockwise”,and<br />

“above–below”allpointtosymmetries—automorphismsoftheplane.So<br />

itseemsthattheindiscernibilityhasreturned.Tobreakthesymmetryin<br />

thecomplexnumbers,weneedtobreakthesymmetryintheplane.<br />

Asnotedabove,onecanbreakthesymmetryintheintegersunderadditionbythinkingofthatasasubstructureoftheintegersunderaddition<br />

andmultiplication. Thereisnoproblemdistinguishing 1from −1inthat<br />

structure.AndperhapsonecandenythatthereissuchathingastheKlein<br />

group.Instead,thereareanumberofKleingroups.Ineachsuchgroup,the<br />

fourelementsareproperlyindividuated.Onewouldhavetogiveasimilar<br />

interpretationofthelanguagesinvokedincategorytheory.Onemightjust<br />

breakthesymmetryglobally—onceandforall—byinsistingthatthe<br />

ontologyofallofmathematicsistheiterativehierarchy,orsomeotherrigid<br />

structure.<br />

Allthiscouldbedone,ofcourse. Thetechnicalresourcesrequiredare<br />

well-known.Notice,however,thattheneedtobreakthesymmetryinvolves<br />

reinterpretingthelanguagesofmathematics. Onequestionwouldconcern<br />

hownaturalthereinterpretationsare,asreadingsoftheoriginallanguages<br />

ofmathematics.<br />

IsuspectthattherewouldnotbeaproblemforFregeconcerningour<br />

issue. Asnotedabove,hedemandedtwothingsbeforeonecouldusethe<br />

definitearticleinaproperlyrigorousmathematicaltreatment.Onehasto<br />

show“thatsomeobjectfallsunder”theconceptinquestion,andtheotheris<br />

“thatonlyoneobjectfallsunderit”(Frege,1884, §74n).Itwouldnotdo,for


240 StewartShapiro<br />

Frege,tosimplydeclarethatthecomplexnumbersarethealgebraicclosure<br />

ofthereals,oreventosaythatweareinterestedinanalgebraicclosureof<br />

thereals.Indoingthis,wewouldfailthefirstrequirement,ofshowingthat<br />

someobjectfallsundertheconcept“squarerootof −1”.Howdoweknow<br />

thatthereareanyalgebraicclosuresoftherealnumbers?Presumably,Frege<br />

wouldhavegivenanexplicitdefinitionofthecomplexnumbers,perhapsas<br />

pairsofreals(which,inturn,wouldbedefinedintermsofcertaincoursesof-values).<br />

2 Thisexplicitdefinitionwouldpresumablybreakthesymmetry<br />

between iand −i. Ifitdidn’t,thentheaccountwouldfailFrege’ssecond<br />

requirement,“thatonlyoneobjectfallsunder”theconcept. Inthatcase,<br />

Fregewouldnotallowtheuseofthedefinitearticle.<br />

Consider,next,anominalist,aphilosopherwhodeniestheexistenceof<br />

mathematicalobjects.Onsuchviews,mathematicshasnodistinctontology.<br />

Thenthedevilisinthedetailsoftheview. Idonotseeanissueherefor<br />

afictionalist,onewholikensmathematicstomake-believe.Onecansurely<br />

tellacoherentstoryaboutobjectsthatareindiscernibleasfarasthedetails<br />

providedbythestorygo(Black’scharacter Bnotwithstanding).Consider,<br />

forexample,thefollowingshortstory: “Oneday,twopeoplemet,fellin<br />

love,andlivedtogether,happilyeverafter.”Whateveritsliterarymerits,<br />

thisissurelyacoherentpieceoffiction.Inbothcases,thereisnothinginthe<br />

storytodistinguishthecharacters. Anythinginthelanguageofthestory<br />

thatholdsofonealsoholdsoftheother.And,ofcourse,wehavenothing<br />

togoonbesidesthedetailsofthestory,pluscommonknowledgeofhuman<br />

psychology,namingconventions,andthelike. Considerthisvariationon<br />

ourstory:“Oneday,ChrismetKelly.Theyfellinlove,andlivedtogether,<br />

happilyeverafter.” Onemightclaimthat,here,Chrisisdistinguished<br />

fromKellybecauseheorsheistheonewhoiscalled“Chris”inthatstory.<br />

But,aswithcomplexanalysis,thisdoesnotseemlikeadistinctionthat<br />

matters. Tobesure,thereareinterestingissuesconcerningthesemantics<br />

and,perhaps,theontologyandmetaphysicsoffiction,butIdonotpropose<br />

toenterthatrealmhere.<br />

Reconstructivenominalistsprovidetranslationsofmathematicallanguagesintovocabularythatdoesnotcommitthemathematiciantothe<br />

existenceofmathematicalobjects. Typically,singulartermsandbound<br />

variablesinmathematicallanguagesarerenderedasboundvariableswithin<br />

thescopeofmodaloperators. Insteadofspeakingofwhatexists,thereconstructivenominalistspeaksofwhatmightexist(asinGeoffreyHellman(1989)),whatcanbeconstructed(alaCharlesChihara(1990)),orwhatfollowsfromaxioms,orwhatever.Whetheranissueanalogoustothepresent<br />

2 ThankstoMichaelDummettforsomekeyinsights.Onemustbespeculativehere,since<br />

Fregeonlygavethebaresthintsathowrealanalysisistofitintohislogicistprogram<br />

(see,forexample,(Simons,1987)).


ReferencetoIndiscernibleObjects 241<br />

onearisesdependsonthedetailsofthetranslation,andIproposetoavoid<br />

thataswell. 3<br />

Suppose,finally,thatsomeonedoesaccepttheexistenceofmathematical<br />

objects—contranominalism—andagreesthatinsomecases,distinct<br />

objectsareindiscernible—contratheprogramofsymmetry-breaking-viareinterpretation.Forpresentpurposes,itdoesnotseemtomatterwhatthe<br />

metaphysicalnatureofthesemathematicalobjectsmaybe.Ourphilosopher<br />

maybeatraditionalplatonist,orshemayholdthatmathematicalobjects<br />

aresomehowmentalconstructions,orthattheyaresocialconstructs,or<br />

whateverelsethephilosopherdreamsup. Ourphilosophermayevenbe<br />

aquietistaboutmathematicalontology,insistingthattheonlythingsto<br />

sayaboutthemarewhatfollowsfromthemathematicaltheories.Allthat<br />

matters,fornow,isthatthelanguagesbeunderstoodliterally,andthat<br />

somenumericallydistinctobjectsareindistinguishable.<br />

Withoutmuchlossofgenerality,wemightaswellkeeponwithour<br />

standardexample:ourphilosopherholdsthatthecomplexnumbersexist,<br />

thatthesquarerootsof −1areindiscernible,andthatthereisaroleplayed<br />

bytheterm“i”.Thenourproblemarises.Hemusteithercomeupsomehow<br />

withareferentfor‘i’,whichwouldbetobreakthesymmetry,orelsehe<br />

mustdescribethelogico-semanticroleofthatterm.<br />

Thisisnottheplacetoattempttosolvethepresentproblem.Thatisa<br />

matterforfutureworkwhich,Ibelieve,involvessubstantialthemesinsemantics,pragmatics,andlogic,bothforthelanguagesofmathematicsand<br />

fornaturallanguagesgenerally. Whatistheroleandfunctionofsingular<br />

terms(orlinguisticitemsthatlookandfunctionlikesingularterms),and<br />

howdosuchtermsgetintroducedintothelanguage? Thepurposeofthe<br />

presentarticleistoarticulatetheissue,andtodelimittherangeofphilosophersofmathematicsforwhomitisasubstantialissue.Attheveryleast,<br />

Ihopetohaveconvincedthegentlereaderthatitisnotaproblemlocalto<br />

anteremstructuralism.<br />

StewartShapiro<br />

DepartmentofPhilosophy,TheOhioStateUniversity<br />

350UniversityHall,230NorthOvalMall,Columbus,Ohio43210,USA<br />

shapiro.4@osu.edu<br />

3 Chihara’s(1990)modalconstructivismisaninterestingcasehere. Accordingly,asingularterm,suchasanumeral,representsthepossibilityofconstructinganopensentence<br />

withcertainsemanticfeatures. Soonecanwonderwhichopensentencewouldcorrespondto“i,asopposedtotheopensentencethatcorrespondswith“−i”.Ipresumethat<br />

Chiharawouldlikencomplexnumberstopairsofreals,asabove.Thiswouldavoidthe<br />

(analogueof)thepresentproblem,bybreakingthesymmetry.


242 StewartShapiro<br />

References<br />

Black,M.(1952).Theidentityofindiscernibles.Mind,61,153–164.<br />

Brandom,R.(1996).ThesignificanceofcomplexnumbersforFrege’sphilosophy<br />

ofmathematics.ProceedingsoftheAristotelianSociety,96,293–315.<br />

Burgess,J. (1999). Reviewof(Shapiro,1997). NotreDameJournalofFormal<br />

Logic,40,283–291.<br />

Button,T. (2006). Realiststructuralism’sidentitycrisis: ahybridsolution.<br />

Analysis,66,216–222.<br />

Chihara,C.(1990).Constructibilityandmathematicalexistence.Oxford:Oxford<br />

UniversityPress.<br />

Frege,G. (1884). DieGrundlagenderArithmetik. Breslau: Koebner. (The<br />

FoundationsofArithmetic,translatedbyJ.Austin,secondedition,NewYork,<br />

Harper,1960.)<br />

Hellman,G.(1989).Mathematicswithoutnumbers.Oxford:OxfordUniversity<br />

Press.<br />

Hellman,G.(2001).Threevarietiesofmathematicalstructuralism.Philosophia<br />

Mathematica,9(III),184–211.<br />

Keränen,J.(2001).Theidentityproblemforrealiststructuralism.Philosophia<br />

Mathematica,3(3),308–330.<br />

Keränen,J.(2006).TheidentityproblemforrealiststructuralismII:areplyto<br />

Shapiro. InF.MacBride(Ed.),Identityandmodality(pp.146–163). Oxford:<br />

OxfordUniversityPress.<br />

Ketland,J. (2006). Structuralismandtheidentityofindiscernibles. Analysis,<br />

66,303–315.<br />

Ladyman,J. (2005). Mathematicalstructuralismandtheidentityofindiscernibles.Analysis,65,218–221.<br />

Leitgeb,H.(2007).StrukturundSymbol.InH.Schmidinger&C.Sedmak(Eds.),<br />

DerMensch: Ein“animalSymbolicum”? (Vol.4,pp.131–147). Darmstadt:<br />

WissenschaftlicheBuchgesellschaft.<br />

Leitgeb,H.,&Ladyman,J.(2008).Criteriaofidentityandstructuralistontology.<br />

PhilosophiaMathematica,16(3),388–396.<br />

MacBride,F. (2005). Structuralismreconsidered. InS.Shapiro(Ed.),Oxford<br />

handbookofphilosophyofmathematicsandlogic(pp.563–589).Oxford:Oxford<br />

UniversityPress.<br />

MacBride,F.(2006a).Identityandmodality.Oxford:OxfordUniversityPress.<br />

MacBride,F.(2006b).Whatconstitutesthenumericaldiversityofmathematical<br />

objects?Analysis,66,63–69.<br />

Shapiro,S. (1997). Philosophyofmathematics: Structureandontology. New<br />

York–Oxford:OxfordUniversityPress.


ReferencetoIndiscernibleObjects 243<br />

Shapiro,S.(2006a).Thegovernanceofidentity.InF.MacBride(Ed.),Identity<br />

andmodality(pp.164–173).Oxford:OxfordUniversityPress.<br />

Shapiro,S.(2006b).Structureandidentity.InF.MacBride(Ed.),Identityand<br />

modality(pp.109–145).Oxford:OxfordUniversityPress.<br />

Shapiro,S.(2008).Identity,indiscernibility,andanteremstructuralism:thetale<br />

of iand −i.PhilosophiaMathematica,16(3),285–309.<br />

Simons,P. (1987). Frege’stheoryofrealnumbers. HistoryandPhilosophyof<br />

Logic,8,25–44.


Sequent Calculi and<br />

Bidirectional Natural Deduction:<br />

On the Proper Basis of Proof-theoretic Semantics<br />

Peter Schroeder-Heister ∗<br />

Philosophicaltheoriesoflogicalreasoningareintrinsicallyrelatedtoformal<br />

models.ThisholdsinparticularofDummett–Prawitz-styleproof-theoretic<br />

semanticsandcalculiofnaturaldeduction.Basicphilosophicalideasofthis<br />

semanticapproachhaveacounterpartinthetheoryofnaturaldeduction.<br />

Forexample,the“fundamentalassumption”inDummett’stheoryofmeaning(Dummett,1991,p.254andCh.12)correspondstoPrawitz’sformal<br />

resultthateveryclosedderivationcanbetransformedintointroduction<br />

form(Prawitz,1965,p.53). Examplesfromotherareasinthephilosophy<br />

oflogicsupportthisclaim.<br />

Ifconceptualconsiderationsaregeneticallydependentonformalones,<br />

wemayaskwhethertheformalmodelchosenisappropriatetotheintended<br />

conceptualapplication,and,ifthisisnotthecase,whetheraninappropriate<br />

choiceofaformalmodelmotivatedthewrongconceptualconclusions.We<br />

willposethisquestionwithrespecttotheparadigmofnaturaldeduction<br />

andproof-theoreticsemantics,andpleadforGentzen’ssequentcalculusasa<br />

moreadequateformalmodelofhypotheticalreasoning.Ourmainargument<br />

isthatthesequentcalculus,whenphilosophicallyre-interpreted,doesmore<br />

justicetothenotionofassumptionthandoesnaturaldeduction. Thisis<br />

particularlyimportantwhenitisextendedtoawiderfieldofreasoningthan<br />

justthatbasedonlogicalconstants.<br />

Toavoidconfusion,aterminologicalcaveatmustbeputinplace:When<br />

wetalkofthesequentcalculusandthereasoningparadigmitrepresents,<br />

wemean,asitscharacteristicfeature,itssymmetryorbidirectionality,i.e.,<br />

∗ ThisworkhasbeensupportedbytheESFEUROCORESprogramme“LogiCCC—ModellingIntelligentInteraction”(DFGgrantSchr275/15–1)andbythejointGerman-French<br />

DFG/ANRproject“HypotheticalReasoning:LogicalandSemanticalPerspectives”(DFG<br />

grantSchr275/16–1).IwouldliketothankLucaTranchiniandBartoszWięckowskifor<br />

helpfulcommentsandsuggestions.


246 PeterSchroeder-Heister<br />

thefactthatitusesintroductionrulesforformulasoccurringindifferent<br />

positions. Wedonotassumethatthesepositionsaresyntacticallyrepresentedbytheleftandrightsidesofasequent,i.e.,wedonotsticktothe<br />

sequentformatwhichgavethecalculusitsname.Inparticular,wepropose<br />

anatural-deductionvariantofthesequentcalculuscalledbidirectionalnaturaldeduction,whichembodiesthebasicconceptualfeaturesofthesequent<br />

calculus. 1 Conversely,thenatural-deductionparadigmtobecriticizedis<br />

thereasoningbasedon(conventional)introductionandeliminationinferences,eventhoughitcanbegivenasequent-calculusformatasinso-called<br />

“sequent-stylenaturaldeduction”. 2 Theconceptualmeaningofnaturaldeductionvs.<br />

sequentcalculus,whichwetrytocapturebythenotionsof<br />

unidirectionalityvs. bidirectionality,istobedistinguishedfromtheparticularsyntaxofthesesystems.<br />

Wehopeitwillalwaysbeclearfromthe<br />

contextwhetheraconceptualmodeloraspecificsyntacticformatismeant.<br />

Wedonotclaimoriginalityforthetranslationofthesequentcalculus<br />

intobidirectionalnaturaldeduction.Thistranslationisspelledoutindetail<br />

in(vonPlato,2001).Thesystemitselfhasbeenknownmuchlonger. 3 Here<br />

wewanttomakeaphilosophicalpointconcerningtheproperconceptof<br />

hypotheticalreasoningthatalsopertainstoapplicationsbeyondlogicand<br />

logicalconstants. Theterm“bidirectionalnaturaldeduction”seemstous<br />

tobeaveryappropriatecharacterizationofthesystemconsidered.Toour<br />

knowledge,ithasnotbeenusedbefore. 4<br />

1 Assumptionsinnaturaldeduction<br />

Inanatural-deductionframework,thereareessentiallytwothingsthatcan<br />

bedonewithassumptions:introducinganddischarging.Ifweintroducean<br />

assumption<br />

A.<br />

1 OthervariantswouldbeSchütte-stylesystemswithmetalinguisticallyspecifiedrightand<br />

leftpartsofformulas(Schütte,1960)orevenFrege-stylesystems,see(Schroeder-Heister,<br />

1999).<br />

2 FirstsuggestedbyGentzenin(Gentzen,1935),thoughnotunderthatname.<br />

3 See,forexample,(Tennant,1992),(Tennant,2002).Forabriefhistorysee(Schroeder-<br />

Heister, 2004, p.33) (footnote). Unfortunately, theearliest proposal ofthissystem<br />

(Dyckhoff,1988)wasaccidentallyomittedthere.<br />

4 VonPlato(2001)simplyspeaksof“naturaldeductionwithgeneraleliminationrules”,<br />

whichcanalsobeunderstoodintheunidirectionalway(dependingonthetreatment<br />

ofmajorpremissesofeliminationinferences). —Theterm“bidirectional”cameupin<br />

personaldiscussionswithLucaTranchinionthepropertreatmentofnegationinprooftheoreticsemantics,atopicwhichiscloselyrelatedtobidirectionalreasoning.<br />

Seehis<br />

contributiontothisvolume(Tranchini,2009).


BidirectionalNaturalDeduction 247<br />

thenwemakethederivationbelow Adependentonthatassumption,and<br />

ifwedischargeitatanapplicationofaninference<br />

(n)<br />

A.<br />

B (n)<br />

C<br />

weretractthisdependency,sothattheconclusionofthatinferenceisnot<br />

longerdependenton A. Asproposedin(Prawitz,1965),thenumeral n<br />

indicatesthelinkbetweenassumptionsandinferencesatwhichtheyare<br />

discharged. OthernotationsareFitch’s(Fitch,1952)explicitnotationof<br />

subproofs,whichgoesbackto(Jaśkowski,1934),wheretheideaofdischargingassumptionswasdevelopedevenbefore(Gentzen,1934/35).<br />

Introducinganddischargingassumptionsisnotverymuchonecando.<br />

Especially,therearenooperationsthatchangetheformofanassumption<br />

andthereforehavetodowithitsmeaning. Inthissense,theyarepurely<br />

structuraloperations. However,itisdefinitelymorethancanbedonein<br />

Hilbert-typecalculi,wherewehaveatbesttheintroductionofassumptions<br />

butnevertheirdischarging.InHilbert-typesystemsassumptionscannever<br />

disappearbymeansofaformalstep. However,wecanmetalinguistically<br />

provethatwecanworkwithoutassumptionsbyusingthemastheleftside<br />

ofaconditionalstatement.Thisisthecontentofthedeductiontheorem:If,<br />

inaHilbert-typesystem,wehavederived Bfrom A,wecaninsteadderive<br />

A →Bbyanappropriatetransformationofthederivationof Bfrom A.<br />

Sinceinnaturaldeductionwehavethedischargingofassumptionsas<br />

aformaloperationattheobjectlevel,wecanexpressthecontentofthe<br />

deductiontheoremasaformalruleofimplicationintroduction:<br />

(n)<br />

A.<br />

B (n)<br />

A →B<br />

AlthoughthisisanimportantstepbeyondHilbert-typecalculi,itisnot<br />

allthatcanpossiblybedoneinextendingtheexpressivepowerofformal<br />

systems.Ourclaimisthatagenuinelysemantictreatmentofassumptions<br />

ismoreappropriatethanapurelystructuraloneasinnaturaldeduction.<br />

Innaturaldeduction,assumptionshaveacloseaffinitytofreevariables:<br />

Assumptionswhicharenotdischargedarecalledopen,whereasdischarged<br />

assumptionsarecalledclosed. Thisterminologyisjustifiedsinceundischargedassumptionsareopenforthesubstitutionofderivationswhoseend<br />

formulaistheassumptioninquestion,whereasclosedassumptionsarenot.


248 PeterSchroeder-Heister<br />

Givenaderivation<br />

A,then<br />

A<br />

Dwiththeopenassumption<br />

Aandaderivation<br />

B<br />

D1<br />

A of<br />

D1<br />

A<br />

D<br />

B<br />

isaderivationof Bwhichmaybeconsideredasubstitutioninstanceofthe<br />

originalderivation.Inthissenseanopenderivationcorrespondstoanopen<br />

term,andaclosedderivation,i.e.aderivationwithoutopenassumptions,<br />

correspondstoaclosedterm. Thisrelationshipbetweenopenandclosed<br />

proofsandopenandclosedtermscanbemadeformallyexplicitbyaCurry–<br />

Howard-styleassociationbetweentermsandproofs,wherethedischarging<br />

ofassumptionsbecomesaformalbindingoperation.<br />

Inourexample,thederivation D1<br />

A canitselfbeopen,justlikeavariable<br />

whichissubstitutedwithanopenterm.Sointheformalconceptofnatural<br />

deductionandthecompositionofderivationsthereisnoprimacyofclosed<br />

derivationsoveropenones. However,thisprimacyenterswiththephilosophicalinterpretationofnaturaldeductioninthetraditionofDummett<br />

andPrawitz. Thereopenassumptionsareinterpretedasplaceholdersfor<br />

closedproofs. 5<br />

2 AssumptionsinDummett-Prawitz-style<br />

proof-theoreticsemantics<br />

Proof-theoretic semantics as advanced by Dummett and Prawitz 6 was<br />

framedbyPrawitzintheformofadefinitionofvalidityofproofs,where<br />

aproofcorrespondstoaderivationinnatural-deductionform. According<br />

tothisdefinition,closedproofsinintroductionformareprimaryasbased<br />

on“self-justifying”steps,whereasthevalidityofclosedproofsnotinintroductionformaswellasthevalidityofopenproofsisreducedtothat<br />

ofclosedproofsusingcertaintransformationproceduresonproofs,called<br />

“justifications”. Givenanotionofvalidityforatomicproofs(i.e.proofs<br />

ofatomicsentences),thedefinitionofvalidityforthecaseofconjunction<br />

andimplicationformulas(totaketwoelementarycases)canbesketchedas<br />

follows:<br />

5 Hereweswitchterminologyfrom“derivation”to“proof”,asinthesemanticalinterpretationwearenolongerdealingwithpurelyformalobjects,forwhichwereservetheterm“derivation”.Prawitzhimselfoftenspeaksof“arguments”toavoidformalisticconnotationsstillpresentwith“proof”.<br />

6 Foranoverviewofthissortofsemanticssee(Schroeder-Heister,2006)andthereferences<br />

therein.


BidirectionalNaturalDeduction 249<br />

•Aclosedproofofanatomicformula Aisvalidifthereisavalidatomic<br />

proofof A.<br />

•Aclosedproofof A∧Bintheintroductionform<br />

D1<br />

D2<br />

A B<br />

A∧B<br />

isvalidifthesubproofs D1and D2arevalidclosedproofsof Aand<br />

B,respectively.<br />

•Aclosedproofof A →Bintheintroductionform<br />

(n)<br />

A<br />

D<br />

B (n)<br />

A →B<br />

isvalidifforeveryclosedproof D1<br />

A<br />

of Bisvalid.<br />

D1<br />

A<br />

D<br />

B<br />

of A,theclosedproof<br />

•Aclosedproofof Anotinanintroductionformisvalidifitreduces,<br />

bymeansofthegivenjustifications,toavalidclosedproofof Ainan<br />

introductionform.<br />

Ifweareonlyinterestedinclosedproofs,thisdefinitionissufficient. In<br />

viewofthelastclause,itisageneralizedinductivedefinitionproceedingon<br />

thecomplexityofendformulasandthereductionsequencesgeneratedby<br />

justifications. Ifwealsowanttoconsideropenproofs,wewouldhaveto<br />

define:<br />

•Anopenproof<br />

D1<br />

A1<br />

,..., Dn<br />

,theproof<br />

An<br />

A1,...,An<br />

D<br />

B<br />

D1<br />

isvalidifforallclosedvalidproofs<br />

Dn<br />

A1,... ,An<br />

D<br />

B<br />

isavalidclosedproof.<br />

Giventhisclauseforopenproofs,thedefiningclauseforthevalidityofa<br />

closedproofof A →Binintroductionformmightbereplacedwith


250 PeterSchroeder-Heister<br />

•Aclosedproofof A →Bintheintroductionform<br />

(n)<br />

A<br />

D<br />

B (n)<br />

A → B<br />

isvalidifitsimmediateopensubproof<br />

isvalid,<br />

yieldingauniformclauseforallclosedproofsinintroductionform.However,<br />

thiswayofproceedingmakesthedefinitionsofvalidityofopenandclosed<br />

proofsintertwined,whichobscuresthefactthatthereisanindependent<br />

definitionofvalidityforclosedproofs.<br />

Accordingtothisdefinition,closedproofsareconceptuallypriortoopen<br />

proofs.Furthermore,assumptionsinopenproofsareconsideredtobeplaceholdersforclosedproofs,asthevalidityofopenproofsisdefinedbythevalidityoftheirclosedinstancesobtainedbysubstitutingafreeassumptionwithaclosedproofofit.Sowehaveidentifiedtwocentralfeaturesof<br />

standardproof-theoreticsemantics:<br />

A<br />

D<br />

B<br />

Theprimacyofclosedoveropenproofs (α)<br />

Theplaceholderviewofassumptions (β)<br />

Thedefinitionofvalidityshowsafurtherfeaturewhichisconnectedto(α)<br />

A<br />

and(β). Thefactthatinanopenproof Dtheopenassumption<br />

Aisa<br />

B<br />

placeholderforclosedproofs D1<br />

of A,yieldingaclosedproof<br />

A<br />

D1<br />

A<br />

D<br />

B<br />

A<br />

meansthatthevalidityof Disexpressedasthetransmissionofvalidity<br />

B<br />

from[theclosedproof] D1to[theclosedproof]<br />

D1<br />

A<br />

D<br />

B


BidirectionalNaturalDeduction 251<br />

A<br />

Ifoneconsidersanopenproof Dtobeaproofoftheconsequencestatement<br />

B<br />

that Bholdsunderthehypothesis A,thisexpresses<br />

Thetransmissionviewofconsequence (γ)<br />

i.e.,theideathatthevalidityofaconsequencestatementisbasedonthe<br />

transmissionofthevalidityofclosedproofsfromthepremissestotheconclusion.<br />

Thisideaiscloselyrelatedtotheclassicalapproachaccordingto<br />

whichhypotheticalconsequenceisdefinedasthetransmissionofcategorical<br />

truth(inamodel)fromthepremissestotheconclusion. Inthatrespect,<br />

Dummett–Prawitz-styleproof-theoreticsemanticsdoesnotdepartfromthe<br />

classicalviewpresentintruth-conditionsemantics(see(Schroeder-Heister,<br />

2008b)).Ofcourse,therearefundamentaldifferencesbetweentheclassical<br />

andconstructiveapproaches,whichmustnotbeblurredbythissimilarity,<br />

inparticularwithrespecttoepistemologicalissues(see(Prawitz,2009)). 7<br />

Afurtherpointshowingupinthedefinitionofvalidityistheassumption<br />

ofglobalreductionproceduresforproofs(called“justifications”). Thisis<br />

whatmakesthe(generalized)inductiononthereductionsequenceforproofs<br />

possible.Itisassumedthatitisnotindividualvalidproofstepsthatgenerateavalidproof,buttheoverallproofwhichmayreducetoaproofofa<br />

particularform(viz.,aproofinintroductionform).Wecallthis<br />

Theglobalviewofproofs (δ)<br />

Thesefourfeaturesareintimatelyconnectedtothemodelofnaturaldeductionasitsformalbackground.<br />

Thisholdsespeciallyfor(β)and(δ),<br />

whichspecify(α)and(γ),respectively.Naturaldeductionpermitstoplace<br />

aderivationontopofanotherone,anditisnaturaldeductionwherewe<br />

havethenotionofproofreduction. Inthesequentcalculus,thissortof<br />

connectionisnotpresent.<br />

Inthesequentcalculus,logicalinferencesnotonlyconcerntherightside<br />

ofasequent(correspondingtotheendformulainnaturaldeduction)butthe<br />

rightandleftsidesofsequentslikewise.Inthissensethesequentcalculusis<br />

7<br />

Itmightbementionedthatthedefinitionofvalidityforaclosedproofof A → Bisclosely<br />

relatedtoLorenzen’sadmissibilityinterpretationofimplication.Accordingto(Lorenzen,<br />

1955), A →Bexpressestheadmissibilityoftherule A<br />

. Theclaimthateveryclosed<br />

B<br />

proofof Acanbetransformedintoaclosedproofof Bcanbevieweduponasexpressing<br />

admissibility.Atfirstglance,thiscontradictsthefactthatinnaturaldeductionanopen<br />

A<br />

...<br />

proof isaproofof Bfrom Aandshouldassuchbedistinguishedfromanadmissibility<br />

B<br />

statement. However,evenif,intheformalsystem,wearedealingwithproofsfrom<br />

assumptionsratherthanadmissibilitystatements,thesemanticinterpretationintermsof<br />

validitycomesveryclosetotheadmissibilityview.See(Schroeder-Heister,2008a).


252 PeterSchroeder-Heister<br />

inherentlybidirectionalcomparedtotheunidirectionalformalismofnatural<br />

deductionthatunderliesDummett–Prawitz-styleproof-theoreticsemantics.<br />

Inthefollowingwewillmakeacaseforthebidirectionalframework.<br />

3 Thesequentcalculusandbidirectionalnaturaldeduction<br />

Accordingtothetraditional,i.e.pre-natural-deductionreasoningmodel,we<br />

startwithtruesentencesandproceedbyinferenceswhichleadfromtrue<br />

sentencestotruesentences. Thisguaranteesthatwealwaysstayinthe<br />

realmoftruth. 8 Alternatively,wecouldstartwithassumptionsandassert<br />

sentencesunderhypotheses. Thisisthebackgroundofnaturaldeduction.<br />

Naturaldeductionaddsthefeatureofdischargingassumptions,i.e.,the<br />

dependencyonassumptionsmaydisappearinthecourseofanargument.<br />

Inthiswaythedynamicsofreasoningnotonlyaffectsassertionsbutat<br />

thesametimethehypothesesassumed. However,thisdynamicsisvery<br />

limitedastheonlyoptionsareintroducinganddischarging,sothereisno<br />

morethanayes/noattributiontohypotheses. Wecannotintroduceand<br />

eliminateassumptionsaccordingtotheirspecificmeaning,whichwouldbe<br />

amoresophisticateddynamics.Inthissensereasoninginstandardnatural<br />

deductionisassertioncentredandunidirectional.Thisisevenmoreso,as<br />

thehypothesesassumedareplaceholdersforclosedproofs. 9<br />

Agenuinelydifferentmodelisgivenbythesequentcalculus. Theparticularfeatureofthissystem,i.e.introductionrulesontheleftsideofthesequentsign,canbephilosophicallyunderstoodasthemeaning-specificintroductionofassumptions.Considerconjunctionwithleftsequentrules<br />

Γ,A⊢C<br />

Γ,A∧B ⊢C<br />

Γ,B ⊢ C<br />

Γ,A∧B ⊢ C<br />

Theserulescanbeinterpretedasfollows:Supposewehaveasserted Cunder<br />

thehypotheses Γand A. Thenwemayclaim Cbyassuming A∧Basan<br />

assumptionanddischarging Aasanassumption,andsimilarlyfor B.Writteninnatural-deductionstylethiscorrespondstothegeneralelimination<br />

rulesforconjunction<br />

A∧B<br />

C<br />

(n)<br />

A.<br />

C (n)<br />

A∧B<br />

C<br />

(n)<br />

B.<br />

C (n)<br />

8 Thiswas,forexample,thepicturedrawnbyBolzanoandFrege.<br />

9 Thisisnotessentiallychangedifwereplaceassertionwithdenialandinthissense<br />

dualizenaturaldeduction.Unidirectionalitywouldjustpointintotheoppositedirection.<br />

See(Tranchini,2009).


BidirectionalNaturalDeduction 253<br />

butwiththecrucialmodificationthatthemajorpremissmustnowbean<br />

assumption,i.e.,mustoccurintopposition 10 (thisishereindicatedbythe<br />

lineoverthemajorpremiss).Similarly,theleftimplicationrule<br />

Γ ⊢ A Γ,B ⊢ C<br />

Γ,A→B ⊢ C<br />

isinterpretedasfollows: Supposewehaveassertedboth Aunderthehypotheses<br />

Γ,and Cunderthehypotheses Γand B. Thenwemayclaim<br />

Cundertheassumption A → Binsteadof B,i.e.,discharge Bandassert<br />

A →Binstead. Writteninnatural-deductionstyle,thisyieldsthegeneral<br />

→-eliminationrule<br />

(n)<br />

B.<br />

A → B A C<br />

(n)<br />

C<br />

againwiththecrucialdifferencetothestandardgeneraleliminationrule<br />

thatthemajorpremissoccursintopposition. 11<br />

Bypresentingthesequent-calculusrulesinanatural-deductionframeworkwearenolongerworkingin“standard”or“genuine”naturaldeductionbutinthereasoningmodelsuggestedbythesequentcalculus,asthe<br />

restrictiononmajorpremissesofeliminationrulesrunscountertotheway<br />

premissesaretreatedinstandardnaturaldeduction.Wecallthismodified<br />

systembidirectionalnaturaldeductionasitactsonboththeassertionand<br />

theassumptionside,withrulesthatdependontheformsoftheformulasassumedorasserted.Sothepossibleoperationsonassumptionsarenolonger<br />

merelystructural. 12<br />

Inproposingbidirectionalnaturaldeduction,asanatural-deduction-style<br />

variantofthesequentcalculus,asourmodelofreasoning,weestablisha<br />

symmetrybetweenassertionsandassumptions. Likeassertions,assumptionscanbeintroducedaccordingtotheirmeaning,namelyasmajorpremissesofeliminationinferences.<br />

Byimposingtherestrictionthatmajor<br />

premissesmustalwaysbeassumptions,eliminationinferencesreceivean<br />

10 InTennant’s(Tennant,1992)terminology,themajorpremiss“standsproud”.<br />

11 Atranslationbetweensequentcalculusandnaturaldeductionwithgeneralelimination<br />

rulesiscarriedoutinfulldetailin(vonPlato,2001). Notethatforimplication,weare<br />

hereconsideringthegeneraleliminationruleusedbyvonPlato,astheycorrespondto<br />

theleftsequentcalculusrule,ratherthanthemorepowerfuloneproposedin(Schroeder-<br />

Heister,1984),whichextendsthestandardframeworkofnaturaldeductionwithrulesas<br />

assumptions.<br />

12 Wealsocallit“natural-deduction-stylesequentcalculus”,asitisconceptuallyasequent<br />

calculuswhichispresentedintheformofanaturaldeductionsystem(Schroeder-Heister,<br />

2004). In(Negri&vonPlato,2001),thistermisusedinadifferentsense,meaninga<br />

specificformofthesequentcalculus.


254 PeterSchroeder-Heister<br />

entirelydifferentreading. Theyarenolongerjustifiedbyreferencetothe<br />

waythemajorpremisscanbe(canonically)derived.Theyareratherviewed<br />

aswaysofintroducingcomplexassumptions,giventhederivationsofthe<br />

minorpremisses.Eliminationinferencesinbidirectionalnaturaldeduction<br />

combinetheintroductionofanassumptionwithaneliminationstepand<br />

canthusbeviewedasaspecialformofassumptionintroduction. Thereforewealsocallthem“upwardintroductions”,asopposedto“downward<br />

introductions”whicharethecommonintroductionrules.<br />

Assumptionswhicharemajorpremissesofeliminationinferencesareno<br />

longerplaceholdersforclosedproofsastheycannotbeinferredbymeans<br />

ofaninference. Theyarealwaysstartingpointsofeliminationinferences.<br />

Ofcourse,itmightbepossibletoshowthatgivenaproof D1<br />

of Aand<br />

A<br />

A<br />

aproof D2of<br />

Bfrom A,wecanobtainaproof<br />

B<br />

D<br />

of B. However,this<br />

B<br />

wouldhavetobeestablishedasatheoremcorrespondingtocutelimination<br />

forthesequentcalculus. Itisnolongeratrivialmatterasinstandard<br />

(unidirectional)naturaldeduction,since<br />

D1<br />

A<br />

D2<br />

B<br />

isnolongerawell-formedproofif Aisamajorpremissofanelimination<br />

inference.Thereforebidirectionalityovercomestheplaceholderviewofassumptions(β).Withthisitalsoovercomestheprimacyofclosedoveropen<br />

proofs(α)asclosedproofsarenolongerusedtointerpretassumptions.<br />

Onlyapremissofanintroductionrulecanbeviewedasaplaceholderfora<br />

closedproof,whichmeansthattheuniforminterpretationofassumptions<br />

byreferencetoclosedproofsisgivenup.<br />

Thetransmissionviewofconsequence(γ)disappearsaswell.Asassumptionscanbeintroducedinthecourseofaproof(inthesequentcalculusby<br />

leftintroduction,inbidirectionalnaturaldeductionasthemajorpremiss<br />

ofaneliminationinference),itisnolongeradefiningfeatureofthemthat<br />

theytransformclosedproofsintoclosedproofs. Ifthishappenstobethe<br />

case,thenitis“accidental”andhastobeproved. Theintroductionof<br />

anassumptionisjustasprimitiveastheintroductionofanassertion. In<br />

theterminologyofDummett–Prawitz-styleproof-theoreticsemantics,both<br />

theintroductionofanassertionandtheintroductionofanassumptionis<br />

acanonical,i.e.definitionalstep. Moreprecisely,thedistinctionbetween<br />

canonicalandnon-canonicalstepsdisappears.Inthissensetheconceptof<br />

validityismuchmorerule-orientedthanproof-oriented:Wenowconsidera<br />

prooftobevalidifitconsistsofproperapplicationsofrightandleftrules


BidirectionalNaturalDeduction 255<br />

(inthesequentcalculus)ordownwardandupwardintroductionrules(in<br />

bidirectionalnaturaldeduction)ratherthanifitreducestoaproofinintroductionformforallitsclosedinstances.Inthisway,theglobalviewof<br />

proofs(δ)alsodisappears,asitisbasedonthefundamentalassumption<br />

thatproofsareprimarytorulesandthatthevalidityofrulesisbasedon<br />

proofsandproofreduction.Theideaofbidirectionalreasoningisverymuch<br />

localratherthanglobal. 13<br />

Thisdoesnotmeanthatrightandleft(sequentcalculus),ordownward<br />

andupward(bidirectionalnaturaldeduction)introductionsareunrelatedto<br />

eachother.Wewillstillrequiresomenotionofharmonybetweenthetwo<br />

sortsofinferencesasanadequacycondition.However,thisharmonywillbe<br />

localratherthanglobal,andnotbasedonproofreduction. Onecriterion<br />

wouldbeuniquenessinthesenseof(Belnap,1961/62),whichmeansthat<br />

ifweduplicaterulesforaconstant ∗,yieldingaconstant ∗ ′ withthesame<br />

right(ordownward)andleft(orupward)rules,wecanprove A[∗] ⊣⊢ A[∗ ′ ]<br />

inthecombinedsystem. There A[∗]isanyexpressioncontaining ∗,and<br />

A[∗ ′ ]isobtainedfrom A[∗]byreplacing ∗with ∗ ′ .However,unlikeBelnap,<br />

wewouldnotrelyonconservativeness,asthisisaglobalconcept,but<br />

ratheronlocalinversioninthesensethatthedefiningconditionsfora<br />

constant ∗canbeobtainedbackfromthisconstant. Ourmaincriticism<br />

ofBelnap’sproposalofconservativenessanduniquenessinhisdiscussionof<br />

theconnective“tonk”isthathemixesalocalcondition(uniqueness)with<br />

aglobalone(conservativeness). 14<br />

4 Whygoinglocal?<br />

Whyshouldweswitchtoaconceptofhypotheticalreasoningwhichisdifferentfromthestandardonecharacterizedby(α)–(δ),andwhichisprevailing<br />

bothinclassicalandconstructivesemantics? Thelackofanintuitivejustificationoftheprinciples(α)–(δ)isnoreasonforabandoningthem,ifwe<br />

cannotalsotellwhythebidirectionalalternativehasgreaterexplanatory<br />

power.Infact,wegainaccesstoamuchwiderrangeofphenomena,ifwe<br />

sticktothebidirectionalparadigm.Wejustmentiontwopoints.<br />

Atomicreasoningandinductivedefinitions<br />

Thediscussioninproof-theoreticsemanticshastraditionallyfocusedonlogicalconstants.Logicalconstantsareaparticularlywell-behavedcasewherewecanapplytheglobalconsiderationscharacteristicofthestandardapproach.Natural-deduction-basedproof-theoreticsemanticshasbeendevel-<br />

13 Thelocalapproachtohypotheticalreasoningputforwardherewasoriginallyproposed<br />

byHallnäs(Hallnäs,1991,2006).<br />

14 Thispointwillbeworkedoutelsewhere.


256 PeterSchroeder-Heister<br />

opedasasemanticsoflogicalconstants. However,thisfocusismuchtoo<br />

narrow.Proof-theoreticdefinitionsoflogicalconstantsjustfeatureasparticularcasesofinductivedefinitions.<br />

Lookingatinductivedefinitionsas<br />

basicstructuralentitiesthatconfermeaningtoobjects,thedistinctionbetweenatomicandnon-atomic(i.e.logicallycompound)objectsdisappears.<br />

Mostgenerally,wewoulddealwithdefinitionalclausesoftheform<br />

a ⇐ C<br />

where aisanobjecttobedefinedand Cisadefiningcondition.Starting<br />

withadefinitionofthiskind,right(downward)andleft(upward)introductionrulescanbegeneratedfromthisinductivedefinitioninacanonicalway,representingawayofputtinginductivedefinitionsintoaction,andresultinginpowerfulclosureandreflectionprinciples.<br />

Theformofdefinitional<br />

clauseslooklikeclausesinlogicprogramming,andlogicprogramscanbe<br />

viewedasparticularcasesofinductivedefinitions.Wewouldevengeneralize<br />

theframeworksetupbylogicprogrammingbyconsideringclauseswhere<br />

thebody Cofaclausemaycontainhypotheticalstatementsandtherefore<br />

negativeoccurrencesofdefinedobjects. ThisgoesbeyondstandarddefiniteHornclauseprogrammingandeventranscendstheclassicalfieldof<br />

logicprogrammingwithnegation(Hallnäs&Schroeder-Heister,1990/91).<br />

Itdiffersfromsystemsinvestigatedin(Martin-Löf,1971)inthatitisnot<br />

mainlydirectedatinductionprinciplesbutratherthelocalinversionof<br />

rules. SystemsofthiskindhaverecentlybeenconsideredbyBrotherston<br />

andSimpson(Brotherston&Simpson,2007),wherealsotherelationship<br />

betweeninversion-basedreasoningandinductionprinciplesforiteratedinductivedefinitionsisdiscussed.Consideringinductivedefinitionsingeneral<br />

opensupawiderperspectiveathypotheticalreasoningwhichisnolonger<br />

basedonlogicalconstants. Itcanalsointegratesubatomicreasoningin<br />

thesenseof(Więckowski,2008),wherethevalidityofatomicsentencesis<br />

reducedtocertainassumptionsconcerningpredicatesandterms.<br />

Non-wellfoundedphenomena<br />

Theglobalreductionistperspectiveunderlyingunidirectionalnaturaldeductionexcludesnon-wellfoundedcasessuchastheparadoxes.<br />

Theinductive<br />

definitionofvalidityexpectsthatthereisnolooporinfinitedescentinthe<br />

reasoningchain.However,inthecaseoftheparadoxes,wehaveexactlythis<br />

situation. Ourlocalframeworkcaneasilyaccommodatesuchphenomena.<br />

Forexample,ifwedefine pby ¬qand qinturnby p,thenboth pand q<br />

arelocallydefined.Thegloballoopisirrelevantforthelocaldefinition.In<br />

suchasituationwecannolongerproveglobalpropertiesofproofssuchas<br />

cutelimination,butthiswedonotrequire.


BidirectionalNaturalDeduction 257<br />

Asitisnowamatterof(mathematical)factratherthanadefinitional<br />

requirementwhethercertainglobalpropertieshold,wedonotruleoutnonwellfoundedphenomenabydefinition.Thisisagreatadvantage,asitgives<br />

usabetterchancetounderstandthem.Following(Hallnäs,1991),wemight<br />

callthisapproachapartialapproachtomeaning.AccordingtoHallnäs,this<br />

wouldbeincloseanalogytorecursivefunctiontheory,whereitisapotential<br />

mathematicalresultthatagivenpartialrecursivefunctionistotal,rather<br />

thansomethingthathastobeestablishedforthefunctiondefinitionto<br />

makesense.<br />

Thereareotherapplicationsofthelocalapproachthatwecannotmentionheresuchasthepropertreatmentofsubstructuralissues,generalized<br />

inversionprinciples,evaluationstrategiesinextendedlogicprograms,etc.<br />

5 FinalDigression:Dialogues<br />

Wehavepleadedforabidirectionalviewofreasoningasitisincorporated<br />

inGentzen’ssequentcalculusandcanbegiventheformofbidirectional<br />

naturaldeduction.Astherearecertainadequacyconditionsgoverningsuch<br />

asystemthatrelateright/downwardsandleft/upwardsruleswithoneanother,sothattheyarelinkedtogetherinacertainway,wemightaskof<br />

whetheritwouldbepossibletoobtainthemfromasingleprinciple. One<br />

possibleanswermightbethedialogicalapproachproposedbyLorenzen<br />

(Lorenzen,1960)andhisfollowers.Ifonecarriesitsideasovertothecase<br />

ofinductiveclauses<br />

a ⇐ C1<br />

.<br />

a ⇐ Cn<br />

onewouldbeleadtoanapproachwhereanattackonthedefinedobject<br />

awouldhavetobedefendedbyachoiceamongthedefiningconditions<br />

Ci,whicharethemselvesattackedbychoosingoneofitscomponents.The<br />

distinctionbetweenrightandleftruleswouldthenbeobtainedbystrategy<br />

considerationsforandagainstcertainatoms. Inthiswayamoreunified<br />

approachcouldbeachieved. Thedialogicalmotivation,asbasedonlocal<br />

attackanddefencerules,wouldnotinvolveglobalreductivefeaturescomparedtovaliditynotionsinstandardproof-theoreticsemantics.Therefore,itappearstobemorefaithfultoourlocalapproach,astheglobalperspectiveisonlyintroducedatalaterstageintermsofstrategiesandtheir<br />

transformations.Inthiswaythedialogicalresearchprogrammepromisesa<br />

novelperspectiveatthelocal/globaldistinction.


258 PeterSchroeder-Heister<br />

PeterSchroeder-Heister<br />

Wilhelm–Schickard–InstitutfürInformatik,UniversitätTübingen<br />

Sand13,72076Tübingen,Germany<br />

psh@informatik.uni-tuebingen.de<br />

http://www-ls.informatik.uni-tuebingen.de/psh<br />

References<br />

Belnap,N.D.(1961/62).Tonk,plonkandplink.Analysis,22,130–134.<br />

Brotherston,J.,&Simpson,A. (2007). Completesequentcalculiforinduction<br />

andinfinitedescent.InProceedingsofthe22ndannualIEEEsymposiumonLogic<br />

inComputerScience(LICS)(pp.51–62).LosAlamitos:IEEEPress.<br />

Dummett,M.(1991).Thelogicalbasisofmetaphysics.London:Duckworth.<br />

Dyckhoff,R. (1988). Implementingasimpleproofassistant. InWorkshopon<br />

programmingforlogicteaching(Leeds,6–8July1987):Proceedings23/1988(pp.<br />

49–59).UniversityofLeeds:CentreforTheoreticalComputerScience.<br />

Fitch,F.B.(1952).Symboliclogic:Anintroduction.NewYork:RonaldPress.<br />

Gentzen,G.(1934/35).UntersuchungenüberdaslogischeSchließen.MathematischeZeitschrift,39,176–210,405–431.(Englishtranslationin:TheCollected<br />

PapersofGerhardGentzen(ed.M.E.Szabo),Amsterdam:NorthHolland(1969),<br />

pp.68–131.)<br />

Gentzen,G.(1935).DieWiderspruchsfreiheitderreinenZahlentheorie.MathematischeAnnalen,112,493–565.(Englishtranslationin:TheCollectedPapers<br />

ofGerhardGentzen(ed.M.E.Szabo),Amsterdam: NorthHolland(1969),pp.<br />

132–201.)<br />

Hallnäs,L.(1991).Partialinductivedefinitions.TheoreticalComputerScience,<br />

87,115–142.<br />

Hallnäs,L.(2006).Ontheproof-theoreticfoundationofgeneraldefinitiontheory.<br />

Synthese,148,589–602.<br />

Hallnäs,L.,&Schroeder-Heister,P. (1990/91). Aproof-theoreticapproachto<br />

logicprogramming: I.Clausesasrules.II.Programsasdefinitions. Journalof<br />

LogicandComputation,1,261–283,635–660.<br />

Jaśkowski,S.(1934).Ontherulesofsuppositionsinformallogic.StudiaLogica,<br />

1,5–32(reprintedin: S.McCall(ed.),PolishLogic1920–1939,Oxford1967,<br />

232–258).<br />

Lorenzen,P.(1955).EinführungindieoperativeLogikundMathematik.Berlin:<br />

Springer(2ndedition1969).<br />

Lorenzen,P.(1960).LogikundAgon.InAttidelXIICongressoInternazionale<br />

diFilosofia(Venezia,1958)(pp.187–194).Firenze:Sansoni.<br />

Martin-Löf,P.(1971).Hauptsatzfortheintuitionistictheoryofiteratedinductive<br />

definitions.InJ.E.Fenstad(Ed.),ProceedingsoftheSecondScandinavianLogic<br />

Symposium(pp.179–216).Amsterdam:North-Holland.


BidirectionalNaturalDeduction 259<br />

Negri,S.,&vonPlato,J. (2001). Sequentcalculusinnaturaldeductionstyle.<br />

JournalofSymbolicLogic,66,1803–1816.<br />

Prawitz,D. (1965). Naturaldeduction: Aproof-theoreticalstudy. Stockholm:<br />

Almqvist&Wiksell(ReprintedMineolaNY:DoverPubl.,2006).<br />

Prawitz,D. (2009). Inferenceandknowledge. InM.Peliˇs(Ed.),TheLogica<br />

Yearbook2008.London:CollegePublications[thisvolume].<br />

Schroeder-Heister,P.(1984).Anaturalextensionofnaturaldeduction.Journal<br />

ofSymbolicLogic,49,1284–1300.<br />

Schroeder-Heister,P. (1999). Gentzen-stylefeaturesinFrege. InAbstractsof<br />

the11thInternationalCongressofLogic,MethodologyandPhilosophyofScience,<br />

Cracow,Poland(August1999)(p.449).<br />

Schroeder-Heister,P. (2004). Onthenotionofassumptioninlogicalsystems.<br />

InR.Bluhm&C.Nimtz(Eds.),Selectedpaperscontributedtothesectionsof<br />

GAP5, FifthInternationalCongressoftheSocietyforAnalyticalPhilosophy,<br />

Bielefeld,22–26September2003(pp.27–48).Paderborn:mentis(Onlinepublication:http://www.gap5.de/proceedings).<br />

Schroeder-Heister,P.(2006).Validityconceptsinproof-theoreticsemantics.Synthese,148,525–571.<br />

Schroeder-Heister,P.(2008a).Lorenzen’soperativejustificationofintuitionistic<br />

logic.InM.vanAtten,P.Boldini,M.Bourdeau,&G.Heinzmann(Eds.),One<br />

hundredyearsofintuitionism(1907–2007):TheCerisyconference(pp.214–240<br />

[Referencesforwholevolume:391–416]).Basel:Birkhäuser.<br />

Schroeder-Heister,P. (2008b). Proof-theoreticversusmodel-theoreticconsequence.<br />

InM.Peliˇs(Ed.),TheLogicaYearbook2007 (pp.187–200). Prague:<br />

Filosofia.<br />

Schütte, K. (1960). Beweistheorie (revised: Proof Theory, 1977). Berlin:<br />

Springer.<br />

Tennant,N.(1992).Autologic.Edinburgh:EdinburghUniversityPress.<br />

Tennant,N.(2002).Ultimatenormalformsforparallelizednaturaldeductions.<br />

LogicJournaloftheIGPL,10,299–337.<br />

Tranchini,L. (2009).Theroleofnegationinproof-theoreticsemantics:Aproposal.InM.Peliˇs(Ed.),TheLogicaYearbook2008.London:CollegePublications<br />

[thisvolume].<br />

vonPlato,J.(2001).Naturaldeductionwithgeneraleliminationrules.Archive<br />

forMathematicalLogic,40,541–567.<br />

Więckowski,B. (2008). Predicationinfiction. InM.Peliˇs(Ed.),TheLogica<br />

Yearbook2007(pp.267–285).Prague:Filosofia.


Relatives of Robinson Arithmetic<br />

Vítězslav ˇ Svejdar ∗<br />

1 Introduction:numbers,orstrings?<br />

Robinsonarithmetic Qwasintroducedin(Tarski,Mostowski,&Robinson,<br />

1953)asabaseaxiomatictheoryforinvestigatingincompletenessandundecidability.Itisveryweak,butallitsrecursivelyaxiomatizableconsistent<br />

extensionsarebothincompleteandundecidable.Inlogictextbooks,itoften<br />

playstheroleoftheweakestreasonabletheorywiththisproperty.<br />

A.Grzegorczykrecentlyproposedtostudythetheoryofconcatenation<br />

asapossiblealternativetheoryforstudyingincompletenessandundecidability.<br />

UnlikeRobinson(orPeano)arithmetic,wheretheindividualsare<br />

numbersthatcanbeaddedormultiplied,inthetheoryofconcatenation<br />

onehasstrings(ortexts)thatcanbeconcatenated.Sointhelanguageof<br />

thetheoryofconcatenationthereisabinaryfunctionsymbol ⌢ forlaying<br />

twostringsendtoendtoformanewstring.Axiomsofthetheoryofconcatenationpostulate,e.g.,associativityoftheoperation<br />

⌢ ,ortheexistence<br />

ofirreducible,i.e.single-letter,strings.Particularvariantsofthetheoryof<br />

concatenationmaydifferinthenumberofirreduciblestrings(withtwoas<br />

themostobviouschoice),orintheexistenceoftheemptystring.<br />

BeforeGrzegorczyk,someaspectsofconcatenationwereconsideredand<br />

someaxiomswereformulatedbyQuine(1946)andTarski. Onevariant<br />

ofthetheory,calledtheory F,appearsalreadyinthebook(Tarskietal.,<br />

1953),whereitisclaimedbutnotprovedthat Fisessentiallyundecidable.<br />

Grzegorczyk’smotivationtostudythetheoryofconcatenationisphilosophical.Whenreasoningorwhenperformingacomputation,wedealwith<br />

texts.Ourhumancapacitytoperformtheseintellectualtasksdependson<br />

ourabilitytodiscerntexts.Thenitisnaturaltodefinenotionslikeundecidabilitydirectlyintermsoftexts,withoutreferencetonaturalnumbers.<br />

∗ ThisworkisapartoftheresearchplanMSM0021620839thatisfinancedbytheMinistry<br />

ofEducationoftheCzechRepublic.


262 Vítězslav ˇ Svejdar<br />

WhenprovingGödel1 st incompletenesstheorem,choosingthetheoryof<br />

concatenationasthebasetheorycouldbepreferabletochoosingPeano<br />

orRobinsonarithmetic,becausethenoneoftheessentialstepsintheincompletenessproof,formalizationoflogicalsyntax,wouldbepractically<br />

effortless.<br />

Wewilldiscusspropertiesoftwotheoriesofconcatenation,theory FdefinedinTarskietal.(1953)andtheory<br />

TCproposedbyGrzegorczyk. It<br />

appearsthatanappropriatemethodofshowingundecidabilityofallconsistentextensionsisprovingmutualinterpretabilityofthesetheorieswith<br />

Robinsonarithmetic Q.Wewillconsidermethodsofconstructinginterpretations,oneofthesebeingthewellknownSolovaymethodofshorteningofcuts.WewillalsodiscusstheGrzegorczyk’sprojectofreplacingRobinson’s<br />

Qbysomeversionoftheoryofconcatenationinmoredetails. The<br />

prosoftheprojectareobvious,buttherearealsosomecons.<br />

2 Somepreliminaries<br />

Foranaxiomatictheory T,let Thm(T)bethesetofallsentencesprovable<br />

in T,insymbols Thm(T) = {ϕ;T ⊢ ϕ},andlet Ref(T)bethesetofall<br />

sentencesrefutablein T,insymbols Ref(T) = {ϕ;T ⊢ ¬ϕ}.Atheory Tis<br />

consistentif Thm(T)∩Ref(T) = ∅,i.e.,ifnosentenceof Tissimultaneously<br />

provableandrefutablein T.Atheory Tiscompleteifitisconsistentand<br />

eachsentenceof Tiseitherprovableorrefutablein T.Atheory Tisrecursivelyaxiomatizableifitisequivalenttoatheory<br />

T ′ withanalgorithmically<br />

decidablesetofaxioms(i.e.with T ′ algorithmicallydecidable).Atheoryis<br />

decidableifthereexistsanalgorithmthatdecidesaboutitsprovability,i.e.,<br />

iftheset Thm(T)isalgorithmicallydecidable.<br />

Atheory Sisanextensionofatheory Tifthelanguageof T(i.e.the<br />

setofallnon-logicalsymbolsof T)isasubsetofthelanguageof S,and<br />

eachsentenceof Tprovablein Tisprovablealsoin S.Atheory Tisessentiallyincompleteifnorecursivelyaxiomatizableextensionof<br />

Tiscomplete;<br />

Tisessentiallyundecidableifnoconsistentextensionof Tisdecidable.It<br />

isknownthatatheoryisessentiallyincompleteifandonlyifitisessentiallyundecidable.Thusweusethesenotionsinterchangeablyor,following<br />

Grzegorczyk,wepreferablyspeakaboutessentialundecidability.<br />

Aninterpretationofatheory Tinatheory Sisamappingfromformulas<br />

of Ttoformulasof Sthatwell-behavesw.r.t.logicalsymbolsandmapsall<br />

axiomsof Ttosentencesprovablein S. Atheory Tisinterpretablein S<br />

ifthereexistsaninterpretationof Tin S.Thenotionofinterpretation,as<br />

wellasthenotionofessentialundecidability,firstappearedin(Tarskietal.,<br />

1953). Importantfactsaboutinterpretabilityarethefollowing:(i)if Tis<br />

interpretablein Sand Sisconsistentthen Tisconsistent,too;(ii)if Tis


RelativesofRobinsonArithmetic 263<br />

interpretablein Sand Tisessentiallyundecidablethenthen Sisessentially<br />

undecidable,too.Thenotionofinterpretabilitycanbeusedasameansto<br />

measurestrengthofaxiomatictheories:if Tisinterpretablein Sandvice<br />

versa,i.e.,if Tand Saremutuallyinterpretable,thenwecanthinkthat<br />

Tand Srepresentthesameexpressiveanddeductivestrength.<br />

3 TheimportanceofRobinsonarithmetic<br />

Robinsonarithmetic Qisanaxiomatictheoryhavingsevensimpleaxioms<br />

formulatedinthelanguage {+, ·,0,S}withsymbolsforadditionandmultiplication(ofnaturalnumbers),aconstantforthenumberzero,andaunary<br />

functionsymbol Sforthesuccessorfunction x ↦→ x+1.Peanoarithmetic PA<br />

isobtainedfrom Qbyaddingtheinductionschema.Thetheory I∆0islike<br />

Peanoarithmetic,butwiththeinductionschemarestrictedto ∆0-formulas<br />

(boundedformulas)only. Thetheory I∆0+Ω1is I∆0enhancedbytheaxiomassertingthetotalityofthefunction<br />

x ↦→ x log x .Foranon-expert,the<br />

propertiesofnaturalnumbersexpressibleby ∆0-formulasconstituteaclass<br />

thatisasubclassofallalgorithmicallydecidableproperties. Anexample<br />

ofa∆0-formulaistheformula ∃v(v · x = y),i.e.theformulathenumber x<br />

isadivisorofthenumber y.Twootherexamplesarethenumber xisprime<br />

andthenumber xisdivisiblebysomeprime.Anexampleofaformulathat<br />

isnot ∆0isthereexistsay> xsuchthat y �= 0and yisdivisiblebyall v<br />

suchthat v �= 0and v ≤ x;thisformulaspeaksaboutathingsimilarto<br />

thefactorialof x.Anotherexampleofanon-∆0formulaisthereexistsay<br />

suchthat y > xand yisprime. Inthetheory I∆0,onecannotprovethat<br />

afactorialof xexistsforeachnumber x,whileprovabilityofthesentencea<br />

prime y > xexistsforeach xisadifficultopenproblem.Bothsentencesare<br />

easilyprovedbyunrestrictedinduction,i.e.inPeanoarithmetic.<br />

Basicpropertiesofnaturalnumbers,likeassociativityandcommutativityofadditionandmultiplication,areprovablein<br />

I∆0,butunprovable<br />

in Q. Generally,universalsentencesareseldomprovablein Q. However,<br />

I∆0+Ω1isinterpretablein Q.Gödel1 st incompletenesstheorem,orbetter,<br />

itsRossergeneralization,saysthatanyrecursivelyaxiomatizableextension<br />

of Qisincomplete.So Qisessentiallyincomplete(essentiallyundecidable).<br />

ThemeaningofGödel2 nd incompletenesstheoremissomewhatquestionablefor<br />

Q.However,itsusualproofgoesthroughin I∆0+Ω1withoutany<br />

changes.<br />

ThusRobinsonarithmetic Qisaveryweakbutstillessentiallyundecidabletheory.<br />

Itrepresentsarich“degreeofinterpretability”becausealot<br />

ofstrongertheories,like I∆0+Ω1,areinterpretableinit.Sinceitisfinitely<br />

axiomatizable,itcanbeusedinastraightforwardproofofundecidabilityof<br />

classicalpredicatelogic.


264 Vítězslav ˇ Svejdar<br />

4 ThetheoryTC<br />

�<br />

x<br />

�� � �<br />

y<br />

�� �<br />

w<br />

w<br />

� �� �<br />

u<br />

� �� �<br />

v<br />

Figure1.Theeditorsaxiom<br />

Thetheoryofconcatenation TChasthelanguage { ⌢ ,ε,a,b}withabinaryfunctionsymbol<br />

⌢ ,aconstant εfortheemptystring,andtwoother<br />

constants aand b.Weusuallyomitthesymbol ⌢ ,i.e.,write xyfortheconcatenation<br />

x ⌢ yofthestrings xand y.Theaxiomsof TCarethefollowing:<br />

TC1: ∀x(xε = εx = x),<br />

TC2: ∀x∀y∀z(x(yz) = (xy)z),<br />

TC3: ∀x∀y∀u∀v(xy = uv →<br />

→ ∃w((xw = u & wv = y) ∨ (uw = x & wy = v))),<br />

TC4: a �= ε & ∀x∀y(xy = a → x = ε ∨ y = ε),<br />

TC5: b �= ε & ∀x∀y(xy = b → x = ε ∨ y = ε),<br />

TC6: a �= b.<br />

TheaxiomsTC1andTC2canbedescribedasaxiomsofsemigroups;byTC2<br />

wecanomitparenthesesinexpressionswheneverconvenient. Theaxioms<br />

TC4–TC6postulatethatthestrings aand baredifferent,andeachofthem<br />

isnon-emptyandirreducible(cannotbenon-triviallydecomposedintotwo<br />

strings).TheaxiomTC3iscallededitorsaxiomin(Grzegorczyk,2005).It<br />

describeswhathappensiftwoeditorsofalargeworkindependentlysuggest<br />

splittingthetextintotwovolumes. Iftheirsuggestionsare x,yand u,v<br />

respectively,asshownintheFigure1,thenthefirstvolumeofoneofthe<br />

editorsconsistsoftwoparts:theothereditor’sfirstvolume,andatext w<br />

(possiblyempty)thatsimultaneouslyoccursasastartingpartoftheother<br />

editor’ssecondvolume.In(Ganea,2007)thistext wiscalledaninterpolant<br />

(oftheequation xy = uv).<br />

Thetheoryofconcatenation TCwasdefinedin(Grzegorczyk,2005).<br />

However,theeditorsaxiomisattributedtoTarski,andtheideaaboutthe<br />

importanceofconcatenationinincompletenessproofscanbetracedback<br />

toQuine,whoin(Quine,1946)citesTarskiandHermesandsays:Gödel’s<br />

proof[...]dependedonconstructingamodelofconcatenationtheorywithin<br />

arithmetic. NotethatQuinedoesnotlistanyaxioms,andthuswhenhe


RelativesofRobinsonArithmetic 265<br />

says“concatenationtheory”,heinfactmeansitsstandardmodel(defined<br />

below). Grzegorczyk(2005)proved(mere)undecidabilityof TC. Later<br />

GrzegorczykandZdanowski(2008)showedessentialundecidabilityof TC<br />

andleftopenthequestionwhetherRobinsonarithmeticisinterpretable<br />

in TC.A.VisserandR.Sterken,see(Visser,2009),M.Ganeain(Ganea,<br />

2007),andthepresentauthorin( ˇ Svejdar,2009)independentlygaveapositiveanswertothisquestion.Moreaboutinterpretabilityin(andof)<br />

TCis<br />

inSection5below.<br />

Thepapers(Grzegorczyk,2005)and(Grzegorczyk&Zdanowski,2008)<br />

workwithavariantof TChavingnoemptystring. Then,forexample,<br />

theaxiomTC4hastheform ∀x∀y(xy �= a). Thepaper( ˇ Svejdar,2009)<br />

workswithavariantof TChavingthreeinsteadoftwoirreduciblestrings.<br />

Theexactchoiceofvariantofthetheoryisamatteroftastebecause,as<br />

shownin(Grzegorczyk&Zdanowski,2008),allvariantsofthetheoryof<br />

concatenationaremutuallyinterpretable,providedtheirreduciblestrings<br />

areatleasttwoinnumber.<br />

Let Abetheset {a,b} ∗ ofallstringsinthetwo-letteralphabet {a,b},<br />

andlet Abethestructurewith Aasauniverse,withconcatenationdefined<br />

“normally”andwithconstants aand brealizedby aand b,respectively.<br />

Then Aisthestandardmodelof TC. Thestructure Bhavingtheset<br />

B = {a,b,e} ∗ asitsuniverseandwithallsymbolsalsodefinednormallyis<br />

anotherexampleofamodelof TC. Let x ⊑ ymean ∃u∃v(uxv = y),and<br />

let x ymean ∃u(ux = y). Theformulas x ⊑ yand x ycanberead<br />

thestring xisasubstringof yandthestring yendsby xrespectively. The<br />

model Baboveshowsthatthesentence ∀x(x �= ε → a ⊑ x ∨b ⊑ x)isnot<br />

provablein TC.<br />

Thefollowingtheoremgivessomemoreexamplesofprovableandunprovablesentences.<br />

Itspurposeistogivethereadersomefeelingabout<br />

provabilityin TC.<br />

Theorem9.Thefollowingsentences(a)–(d)areprovablein TC,<br />

(a) ∀x(xa �= ε),<br />

(b) ∀x∀y(xy = ε → x = ε & y = ε),<br />

(c) ∀x∀y(xa = ya → x = y),<br />

(d) ∀x∀y(a xy → y = ε ∨a y),<br />

whilethefollowingsentence(e)isnotprovablein TC:<br />

(e) ∀x∀y∀z(xz = yz → x = y).<br />

Proof.(a)Assume xa = ε. Then,byTC1andTC2,wehave (bx)a = b.<br />

Irreducibilityof b,i.e.TC5,yields bx = εor a = ε.Thelatterisexcluded


266 Vítězslav ˇ Svejdar<br />

byTC4. Thenfrom bx = ε, (bx)a = b,andTC1wehave a = b,a<br />

contradictionwithTC6.<br />

(b)If xy = εthen x(ya) = ausingTC1andTC2.So x = εor ya = ε<br />

byTC4.From(a)wehave x = ε.Then xy = εyields y = ε.<br />

(c)Let xa = ya.BytheeditorsaxiomTC3,thereexistsawsuchthat<br />

xw = y & wa = aor yw = x & wa = a.Considerthefirstcase,thesecond<br />

oneissymmetric.From wa = aandirreducibilityof awehave w = ε.From<br />

thatand xw = yweindeedhave x = y.<br />

(d)Let a xy,andlet ubesuchthat ua = xy.TheaxiomTC3yields<br />

a wsatisfying uw = x & wy = a,or xw = u & wa = y.Inthesecondcase<br />

weobviouslyhave a y.Soconsiderthefirstcase.From wy = awehave<br />

w = εor y = ε.If y = εthenwearedone.If w = εthen y = a,andthus<br />

a y.<br />

(e)Let Dbethesetofallstringsin {a,b,e} ∗ thathavenooccurrences<br />

of ae.Realize aand bby aand brespectively,anddefine x+yaccordingly:<br />

x + yresultsfrom xybyrepeatingthesubstitution ae→ewhilepossible.<br />

Forexample, bab + eb = babeb,but baa + eb = beb.Onecancheck,incase<br />

ofTC3withalittleeffort,incaseoftheremainingaxiomsrathereasily,<br />

thatthestructure D = 〈D,+,ε,a,b〉isamodelofthetheory TC. In D<br />

wehave a + e = ε + e. Sotheformula x ⌢ z = y ⌢ zisnottruein Dif x,<br />

y, zareevaluatedby a,theemptystring,and erespectively,andthusthe<br />

sentence(e)isnotvalidin D.<br />

Anotherusefulsentenceis ∀x∀y(a ⊑ xy → a ⊑ x ∨ a ⊑ y). Weleave<br />

itsproofin TCasanexercise. Moreaboutthetheory TCandaboutits<br />

modelsisin(Visser,2009).<br />

5 ThetheoryF,interpretability<br />

Theorem10.Robinsonarithmetic Qisinterpretablein TC.<br />

Proof.Weonlygivethebasicideaoftheproofgivenin( ˇ Svejdar,2009).<br />

Thefullproofisrathertechnical.<br />

Whenconstructinganinterpretation,onefirsthastospecifyitsdomain,<br />

whichinourcasemeanstoworkin TCandselectstringsthatwillplaythe<br />

roleofnaturalnumbers.Itappearsthatthefollowingdefinitionworks:<br />

Num(x) ≡ ∀u(u ⊑ x & u �= ε → a u),<br />

astring xisanumberifeachnon-emptysubstringof xendsby a. Note<br />

that,inthemodel DintheproofofTheorem9(e),thestring estartsby a<br />

(since e = a + e). However, eisnotanumberbecauseitisanon-empty<br />

substringofitselfandcannotbewrittenas e = z +a,i.e.doesnotendby a.


RelativesofRobinsonArithmetic 267<br />

Havingnumbers,additionisinterpretedasconcatenation,zeroisinterpretedastheemptystring<br />

ε,andthesuccessorfunction Sisinterpretedas<br />

thefunction x ↦→ xa.Thesedefinitionsworkbecausein TConecanprove<br />

that εand aarenumbersandthatnumbersareclosedunderconcatenation.<br />

Allaxiomsof Qabout 0, S,and + translatetosentencesprovablein TC<br />

underthisinterpretation.<br />

Tointerpretmultiplication,astraightforwardideaistofirstdefinethe<br />

notionofawitnessingsequence. Asequenceofpairs [u0,v0],.. ,[uq,vq]is<br />

awitnessingsequencefor x · yif: u0 = v0 = ε,foreach i < qthepair<br />

[ui+1,vi+1]equals [uia,viy],and uq = x.Thenonecandefinethat x · y = z<br />

ifthereexistsawitnessingsequencefor x · ywith [x,z]asthelastmember.<br />

Theproblemhereisthatin TCitisnotpossibletoprovethatawitnessing<br />

sequenceexistsforeachchoiceof x,y,anditisalsonotpossibletoprove<br />

thatifitexists,itisuniquelydetermined. Awayhowtoovercomethis<br />

problemisinterpretingnotthefullRobinsonarithmetic Q,butratherits<br />

variant Q − inwhichadditionandmultiplicationarenon-totalfunctions.<br />

Thentheresultisobtainedbycombiningtheconstructedinterpretation<br />

of Q − in TCwithafactknownfrom( ˇ Svejdar,2007)that Qisinterpretable<br />

in Q − .<br />

Thetheory Q − usedintheproofofTheorem10wasalsointroduced<br />

byGrzegorczyk. Theinterpretationof Qin Q − in( ˇ Svejdar,2007)isconstructedusingtheSolovaymethodofshorteningofcuts.<br />

Thismethodis<br />

nowwidelyknown,butwasneverpublished:itisonlyexplainedinaletter<br />

toPetrHájek(Solovay,1976).M.Ganeain(Ganea,2007)givesadifferent<br />

proofofinterpretabilityof Qin TC,buthealsousesthedetourvia Q − .<br />

SterkenandVissergiveaproofnotusing Q − ,see(Visser,2009).<br />

Aconsequenceofthefactthat Qisinterpretablein TCisessentialundecidabilityof<br />

TC.Allproofsofinterpretabilityof Qin TCaresomewhat<br />

involved,butstillsimplerthanthedirectproofofessentialundecidability<br />

of TCgivenin(Grzegorczyk&Zdanowski,2008). Theseinterpretability<br />

proofsmightusesomeideasdevelopedbyGrzegorczykandZdanowski:that<br />

iscertainlytrueabouttheauthor’sproofin( ˇ Svejdar,2009).<br />

Since TCisinterpretablein I∆0,thetheories TCand Qaremutually<br />

interpretable;thustheyrepresentthesameexpressiveanddeductivepower.<br />

Thisisapieceofinformationmissingin(Grzegorczyk&Zdanowski,2008).<br />

Aninterestingalternativetheoryofconcatenationisthetheory F.Ithas<br />

thesamelanguageas TC,anditsaxiomsare:<br />

F1: ∀x(xε = εx = x),<br />

F2: ∀x∀y∀z(x(yz) = (xy)z),<br />

F3: ∀x∀y∀z(yx = zx ∨ xy = xz → y = z),


268 Vítězslav ˇ Svejdar<br />

F4: ∀x∀y(xa �= yb),<br />

F5: ∀x(x �= ε → ∃u(x = ua ∨ x = ub)).<br />

AxiomsF1andF2arethesameasaxiomsTC1andTC2of TC.Itiseasy<br />

toverifythataxiomF4isprovablein TC;axiomsF3andF5,asisevident<br />

frommodels Dand Bintheprevioussection,arenotprovablein TC.From<br />

theoppositepointofview,axiomsTC4–TC6andsentences(a)and(b)in<br />

Theorem9areexamplesofsentencesprovablein F;weleavetheirproofs<br />

tothereaderasaninterestingexercise. AlbertVisser,see(Visser,2009),<br />

hasconstructedamodel Mof Fsuchthat M |= / ∀x∀y(a ⊑ xy → a ⊑<br />

x ∨a ⊑ y).Thusin F,onecanhavestrings w1and w2suchthat a ⊑ w1w2,<br />

a �⊑ w1, a �⊑ w2;AlbertVisserdescribesthissituationascreatingaletter<br />

exnihilo. Aconsequenceoftheseremarksisthat Thm(TC)and Thm(F)<br />

areincomparablesetsofsentences.<br />

Itisclaimedin(Tarskietal.,1953)thatW.SmielewandA.Tarski<br />

provedessentialundecidabilityof Fbyinterpreting Qin F;however,no<br />

proofisgiven.Ganea(2007)constructedaninterpretationof TCin F.In<br />

conjunctionwithTheorem10,thisgivesaproofofthetheoremofSmielew<br />

andTarski. Wegive(aslightsimplificationof)Ganea’sproofbelowin<br />

Theorem11.Notehowever,thatitisstillaninterestinghistoricalproblem<br />

whatproofcouldSmielewandTarskihavehadinmind. Ours(Ganea’s)<br />

proofimplicitlyusestheSolovay’sshorteningtechnique,formulatedlong<br />

afterthebook(Tarskietal.,1953)waspublished. A.Visserhassome<br />

possibleexplanationofthishistoricalproblem.<br />

Theorem11(Ganea). TCisinterpretablein F.<br />

Proof.Workin Fanddefinetamestringsasfollows:<br />

Tame(x) ≡ ∀v∀z(z vx → z x ∨ x z),<br />

where hasthesamemeaningasin TC.<br />

(i)Wefirstshow(provewithin F)thattamestringsareclosedunder<br />

concatenation.Soassumethat xand yaretame,andlet vand zbesuch<br />

that z vxy. Weneedtoshowthat z xyor xy z. Since yistame,<br />

wehave z yor y z.If z ythen z xyandwearedone.Soassume<br />

that y zandtake tsuchthat ty = z. From z vxywehaveausuch<br />

that uz = vxy;thus uty = vxy.FromaxiomF3wehave ut = vx.Since x<br />

istame,wehave t xor x t.Then ty xyor xy ty.Since ty = z,<br />

weindeedhave z xyor xy z.<br />

(ii)Nextweshowthatif wyistame,thenalso wistame.Solet vand z<br />

besuchthat z vw.Wewanttoshowthat z wor w z.From z vw<br />

wehave zy vwy.Since ywistame,wehave zy wyor wy zy.Then<br />

astraightforwarduseofaxiomF3yields z wor w z.


RelativesofRobinsonArithmetic 269<br />

Nowwearereadytoverifythatthedomainoftamestrings,togetherwith<br />

theidenticalmappingofsymbols(a, b,and εto a, b,and εrespectively,<br />

concatenationtoconcatenation),definesaninterpretationof TCin F. It<br />

isnotdifficulttoverifythat a, b,and εaretame;thistogetherwith(i)<br />

meansthatthedomainoftamestringsisclosedunderalloperations.The<br />

axiomTC1translatestothesentence ∀x(Tame(x) → xε = εx = x). This<br />

sentenceisevidentlyprovablein F.Asimilarargumentshowsthataxioms<br />

TC2andTC4–TC6translatetosentencesprovablein Faswell.Thisisso<br />

easybecauseTC2andTC4–TC6areuniversalsentences.<br />

ThusitremainstoprovethethetranslationoftheeditorsaxiomTC3is<br />

provablein F.NotethatTC3istheonlyaxiomof TCthatisnotauniversal<br />

sentence;itcontainsanexistentialquantifier.Let x, y, u, v,betamestrings<br />

suchthat xy = uv.Wehavetoshowthatthereexistsatame wsatisfying<br />

xw = u & wv = yor uw = x & wy = v.Since yistame,from uv = xy<br />

wehave v yor y v.Itissufficienttoconsiderthelatter,theformeris<br />

symmetric.Wehaveawsuchthat wy = v.Then uwy = uvand uwy = xy.<br />

FromaxiomF3wehave uw = x.So wisaninterpolant.Since vistame,<br />

from wy = vand(ii)aboveweknowthat wistame.<br />

Since Fiseasilyinterpretablein I∆0,fromtheotherresultsmentionedin<br />

thispaperweknowthat Fand TCaredeductivelyincomparable,butfrom<br />

interpretabilitypointofviewtheyrepresentthesamedegreeofdeductive<br />

strength.Itmaybeofsomeinteresttodirectlyinterpret Fin TC.<br />

Theorem12. Fisinterpretablein TC.<br />

Proof.Nowin TC,workwithradicalstrings,where<br />

Rad(x) ≡ ∀y∀z(yx = zx → y = z).<br />

Itisnotdifficulttoshowthatradicalstringsinclude ε, a,and b,andthat<br />

thedomainofallradicalstringsthatareemptyorendineither aor bis<br />

closedunderconcatenationanddefinesaninterpretationof F.<br />

6 OntheGrzegorczyk’sproject<br />

LetusrepeatfromtheIntroductionthatGrzegorczyk’ssuggestionisto<br />

considerstringsandconcatenationonbothformalandmetamathematical<br />

level.Onformallevel,thetheoryofconcatenationcanserveasanalternativetoRobinsonarithmetic;onmetamathematicallevel,dealingwithtextsisphilosophicallybetterjustifiedbecauseintellectualactivitieslikereasoningandcomputinginvolveworkingwithtexts.<br />

Briefly,themotivationsof<br />

thatprojectcanbesummarizedasfollows:


270 Vítězslav ˇ Svejdar<br />

•inGödel’sargument,theonlyuseofnumbersiscodingofsyntactical<br />

objects,<br />

•thenGödeltheoremsarepresentedasapartofmathematics,buttheir<br />

significanceisbroader,<br />

•whenreasoning,communicating,orevencomputing,wedealwith<br />

texts,notwithnumbers,<br />

•onmetamathematicallevel,thenotionofcomputabilitycanbedefined<br />

withoutreferencetonumbers.<br />

Onecouldremarkthatmathematicsinnotnecessarilyidentifiedbyworking<br />

withnumbers;Gödeltheoremscouldbepresentedaspartofmathematics<br />

evenifreformulatedwithoutnumbers,andtheytranscedemathematicsregardlesswhethertheirformulationinvolvesstringsornumbers.Withthis<br />

littleremarkinmind,onecansaythattheargumentsproGrzegorczyk’s<br />

projectareclearandeasilyacceptable.Thedefinitionofrecursivenesswithoutusingnumbers,asdonein(Grzegorczyk,2005),isveryinteresting.<br />

However,itisalsopossibletofindsomeargumentsthatspeakcontrathat<br />

project,oratleastformodifyingorextendingit. First,whenreasoning<br />

orcomputing,wenotonlyconcatenate: wealsosubstitute. Creatinga<br />

grammaticallycorrectsentenceinanaturallanguagecanbedescribedas<br />

substitutingintopatterns.Inlogic,wehavesubstitutioninformulationof<br />

predicateaxioms. Soonecanthinkthatthetheoryofconcatenation,if<br />

enhancedbysomenotionofoccurrenceorsubstitution,couldbetterserve<br />

itspurpose.Second,whenprovingessentialundecidability,onealsoneeds<br />

anorder. Knownproofsusually(isitamistaketosayalways?) contain<br />

somesortofRossertrick,i.e.,speakaboutaneventthatoccursbeforesome<br />

otherevent. Onecanthinkthatconsideringorderismorenaturalinthe<br />

environmentofnumbersthanintheenvironmentofstrings.Infact,defining<br />

anorderofstringsisoneofcrucialandratherdifficultstepsintheessential<br />

undecidabilityproofof TCcontainedin(Grzegorczyk&Zdanowski,2008).<br />

Vítězslav ˇ Svejdar<br />

DepartmentofLogics,FacultyPhilosophy&Arts,CharlesUniversity<br />

nám.JanaPalacha2,11638Praha1,CzechRepublic<br />

vitezslav.svejdar@cuni.cz,http://www.cuni.cz/∼svejdar/<br />

References<br />

Ganea,M.(2007).ArithmeticonSemigroups.J.SymbolicLogic,74(1),265–278.<br />

Grzegorczyk,A.(2005).UndecidabilitywithoutArithmetization.StudiaLogica,<br />

79(2),163–230.


RelativesofRobinsonArithmetic 271<br />

Grzegorczyk,A.,&Zdanowski,K. (2008). Undecidabilityandconcatenation.<br />

InA.Ehrenfeucht,V.W.Marek,&M.Srebrny(Eds.),AndrzejMostowskiand<br />

foundationalstudies(pp.72–91).Amsterdam:IOSPress.<br />

Quine,W.V.O. (1946). Concatenationasabasisforarithmetic. J.Symbolic<br />

Logic,11(4),105–114.<br />

Solovay,R.M.(1976).Interpretabilityinsettheories.(UnpublishedlettertoP.<br />

Hájek,Aug.17,1976,http://www.cs.cas.cz/hajek/RSolovayZFGB.pdf.)<br />

ˇSvejdar,V.(2007).AnInterpretationofRobinsonArithmeticinitsGrzegorczyk’s<br />

WeakerVariant.FundamentaInformaticae,81(1–3),347–354.<br />

ˇSvejdar,V. (2009). Oninterpretabilityinthetheoryofconcatenation. Notre<br />

DameJ.ofFormalLogic,50(1),87–95.<br />

Tarski,A.,Mostowski,A.,&Robinson,R.M. (1953). Undecidabletheories.<br />

Amsterdam:North-Holland.<br />

Visser,A.(2009).GrowingCommas:AStudyofSequentialityandConcatenation.<br />

NotreDameJ.ofFormalLogic,50(1),61–85.


The Role of Negation in Proof-theoretic Semantics:<br />

a Proposal<br />

Luca Tranchini ∗<br />

Proof-theoreticsemantics,asdevelopedbyauthorssuchasDummettand<br />

Prawitz,triestoaccountforthemeaningoflogicalconstantsthroughthe<br />

usemadeoftheminpractice. Thetypicalcontextinwhichtheyfigure<br />

isdeduction,sotheprogrambecomestheoneofshowinghowtherules<br />

governingdeductivepracticesfixthemeaningoflogicalconstants. The<br />

theoreticalrequirementruleshavetosatisfyisharmony,whichisendorsed<br />

inGentzen’sinversionprinciple.<br />

Oneofthedistinctivefeaturesofhumanlanguageiscompositionality,<br />

thatisthepossibilityofproducingsentencesofarbitrarycomplexityby<br />

meansoflogicaloperators. Henceproof-theoreticsemanticscanaimat<br />

beingthecoreofafully-fledgedtheoryofmeaning,thatisofanexplication<br />

ofspeakerslanguagecompetence.<br />

1 Verificationism:ProofandAssertion<br />

Theverificationisttheoryofmeaning,isgroundedonthechoiceofassertion<br />

asthebasiclinguisticact.Assertionistakentobegovernedbythefollowing<br />

principle<br />

Theassertionofasentenceiswarrantedonlyifitstruthisrecognized<br />

wheretheintuitivenotionoftruthrecognitionistobeexplainedbymeans<br />

ofthenotionofproof.<br />

Clearly,theproofsthatcountasevidenceforthetruthofsentencesare<br />

onlyclosedproofs,i.e.proofsinwhichtheconclusiondoesnotdependon<br />

anyassumption.<br />

Openproofsareonlymediatelyconnectedwithlinguisticacts.Takean<br />

openproofof Bfrom A<br />

∗ ThisworkhasbeensupportedbytheESFEUROCORESprogramme“LogiCCC—<br />

ModellingIntelligentInteraction”(DFGgrantSchr275/15–1).


274 LucaTranchini<br />

A.<br />

B<br />

thesentence Bcanbeassertedonlyifevidencefor Aisavailable. If A<br />

isatomic,evidencewillconsistintheopportunecomputation,if Aisa<br />

mathematicalsentence.Whenever Aisanempiricalsentencewecanthink<br />

ofevidenceforitasanopportuneempiricalobservation.<br />

Technically,toaccountforthese“atomicproofs”inastandardnatural<br />

deductionsystemweextendthevocabularywithaset Pofpropositional<br />

constants,standingfortheatomicsentencesoflanguage.Asubsetofsuch<br />

sentences, Twillconsistsofthesentencesforwhichextra-deductiveevidence<br />

isavailable.Thesetofopenassumptionsinadeductionwillberestricted<br />

tothosenotbelongingto T.<br />

If Aisacomplexsentence,onecantrytoobtainaproofofitfromatomic<br />

assumptionsin T.Still,itisnotalwayspossibletodoso.Nonetheless,even<br />

incasessuchasthesewecanobtainaclosedprooffromtheopenone,even<br />

thoughtheconclusionoftheclosedproofisnottheconclusionoftheopen<br />

one,butamorecomplexsentence. Typically,implicationisthedeviceby<br />

meansofwhichanopenproofistakenintoaclosedonehavingasconclusion<br />

theimplicationoftheassumptionandtheconclusionoftheopenproof:<br />

[A]<br />

.<br />

B<br />

A → B<br />

Ingeneral,wecansaythatverificationismfocusesontheroleofsentencesasconclusionsofdeductiveprocesses.<br />

Asaconsequence,therules<br />

thataretakentofixthemeaningoflogicalconstantsareintroductionrules.<br />

For,theyspecifytheconditionsunderwhichasentencehavingtherelevant<br />

constantasprincipaloperatorcanbeintroducedasconclusionofaderivation.Accordingly,thenotionofcanonicalproof(thatisofaproofinwhich<br />

introductionrulesplayaprominentrole)hasbeentakenastheexplicans<br />

ofthenotionofmeaning.Thatis,toknowthemeaningofasentenceisto<br />

knowwhatcountsasacanonicalproofofit.<br />

2 Negation<br />

Toaccountfornegationinverificationism,thesymbol ⊥,standingforabsurdity,isintroduced.Thenegationofasentence<br />

A, ¬A,isdefined,infull<br />

analogywiththeBHKclause,as A → ⊥.Sowehavetworulesfornegation<br />

(anintroductionandanelimination),whicharenothingbutspecialcases<br />

oftheimplicationrules.


NegationinProof-theoreticSemantics 275<br />

[A]<br />

.<br />

⊥<br />

¬A I¬<br />

.<br />

A<br />

.<br />

¬A<br />

⊥<br />

Obviously,theinversionprincipleholdsfortheserulesaswell(asaconsequenceofitsvalidityforimplication).<br />

Theproblem<br />

Suchacharacterizationgraspsallpropertiesofintuitionisticnegationexcept<br />

thefactthatnoconstructionoffersevidencefortheabsurdity.So,further<br />

ruleshavetobeadded,tofixtheintendedmeaningof ⊥.<br />

Itisnosimpletasktoexplicitlyexpresswitharulethefactthatno<br />

constructionsatisfiestheabsurdity:rulesspecifywaysofobtainingproofs,<br />

whileouraimistospecifytheabsenceofproofs.<br />

IntuitionisticNaturalDeductionNJisobtainedbyextendingminimal<br />

logicwiththefollowingrule,socalledexfalsoquodlibet:<br />

.<br />

⊥<br />

A ef<br />

Oncetheabsurdityhasbeenderived,itispossibletoderiveeverything.<br />

Theruleisintendedtoholdforanysentence A. Still,withoutlossof<br />

generality,itisusefultorestrictittoatomicsentences.Asaconsequence,<br />

⊥canbetakenasanabbreviationofaself-contradictoryatomicsentence,<br />

forinstance‘0 = 1’,deductivelycharacterizedbythefactthatitentailsall<br />

otheratomicsentences.<br />

Aswiththeexfalsowecanformallyseizeintuitionisticlogic,itisnatural<br />

tothinkofitasgraspingtheintendedmeaningoftheabsurdity. Yet,as<br />

someauthorshavenoticed,itisdubiousthattheexfalsoconveysto ⊥the<br />

desiredmeaning.Thefactthatitistobereadasabsurdityseemstodepend<br />

onwhichatomicsentencesareprovable.Indeedifallatomicsentenceswere<br />

provable,therewouldbenothingwronginasserting ⊥asonlytruesentences<br />

couldbeinferredfromit.Butif ⊥hastobeconsideredasstandingforthe<br />

absurdity,thenitshouldnotbeassertibleinanysituation.<br />

Apossiblewayout(Dummett,1991)istobanthepossibilitythatall<br />

atomicsentencescanbesimultaneouslyasserted,thatistoassumethatat<br />

leasttwoatomicsentencesaremutuallyincompatible. Nonetheless,this<br />

restrictionsoundsdefinitelyadhoc:wehavenoreasontoproposeit,apart<br />

fromtheneedofwarrantingthattheexfalsoconveystheexpectedmeaning<br />

to ⊥and,consequently,tonegation.<br />


276 LucaTranchini<br />

Brouweronabsurdityandnegation<br />

InBrouwer’searlywritingswefindtheideathatthenegationofasentence<br />

iswarrantedwhen<br />

wearrivebyaconstructionatthearrestmentoftheprocesswhich<br />

wouldleadto[aconstructionforthesentence]. 1<br />

Toclarifythispoint,onecanimagineamathematician(or,rather,the<br />

idealizedmathematician)attemptingtoproduceaconstructionforasentence<br />

A. Unfortunately,itisimpossibletoobtainaconstructionfor A.<br />

Hence,eachattemptreachesacertainpointafterwhichitisnotpossibleto<br />

carryouttheconstruction,apointbeyondwhich,inBrouwer’swords,“the<br />

constructionnolongergoes.”<br />

AccordingtoBrouwer,whenthemathematicianfindsherself(i.e.,she<br />

producesaconstructionshowingthatsheis)insuchasituationshecan<br />

declarethesentencefalseor,equivalently,acceptitsnegationaswarranted.<br />

Itisonlylaterthattheideaofarrestintheprocessofconstructionis<br />

substitutedbytheoneofcontradiction,glorifiedintheBHKspecification<br />

ofthesemanticsfortheintuitionisticlogicalconstants. Thetreatmentof<br />

contradictionasasentence,implicitintheintuitionisticinformalsemantics<br />

hasbeenfullyfledged,aswebrieflyshowed,withthemarriagebetween<br />

intuitionismandnaturaldeduction,throughtheopportunereadingofthe<br />

exfalsorule.<br />

Tennanton ⊥<br />

Recently,Tennant(1999)triedtochallengetheverificationistaccountof<br />

whatisaproofforthenegationofasentence. Evenifhedoesn’tmake<br />

explicitreferencetoBrouwer,itisquitenaturaltoputtheconceptionof ⊥<br />

heproposessidebysidewiththeideathatacontradictionisnothingbuta<br />

deadendintheprocessofconstruction.<br />

Tennantstartsfromtherefusalofconsideringproofsofthenegationof<br />

asentenceasmethodstoobtainaproofofafalsesentence, ⊥.Rather,he<br />

proposesconsidering ⊥asamarkerofadeadendintheprocessofconstruction.<br />

Clearly, ⊥isdevoidofsententialcontent,i.e.,itisnomorean<br />

abbreviationfor‘0 = 1’. Hence,weareforcedtowithdrawtheinterpretationof<br />

¬Aas A → ⊥: as ⊥hasnosententialcontentwecan’tapply<br />

toitsententialoperators,inparticularimplication.Wecanconcludewith<br />

Tennant’sthat,<br />

accordingly,anoccurrenceof ⊥isappropriateonlywithinaproof,as<br />

akindofknot—theknotofpatentabsurdity,orofself-contradiction. 2<br />

1 (Brouwer,1908,p.109).<br />

2 (Tennant,1999,pp.203–204).


NegationinProof-theoreticSemantics 277<br />

However,TennantdoesnotcompletelyembodyBrouwer’ssolution.For,<br />

accordingtoBrouwer,provingthenegationof Ameansfindingadeadend<br />

intheroutetowardtheproofof A.Ontheotherhand,Tennantthinksthat<br />

theroleof Aisnotthatofanunreachablegoal,butratherastartingpoint.<br />

Weprovethenegationof Awhenwereachadeadendstartingfrom A.It<br />

isimportanttoobservethatthisisnottosaythatwestartfromahypotheticalconstructionfor<br />

A,asinHeytinginterpretationofthemeaningof<br />

negation.Ratherweareperforminganactivitywhichisdifferentfromthe<br />

productionofproof.Suchanactivityisnotorientedbytheconclusionthat<br />

wewanttoreach,butratherbythepointfromwhichwestart. Tennant<br />

introducesanewprimitivenotiontorefertothisalternativeactivity:disproof.Theactivityofconstructionisthensplitintwodifferentsubspecies:<br />

theproductionofproofsandtheproductionofdisproof.Whenweattempt<br />

todisproofasentencewedonotstartfromaproofofitthatthenturnsout<br />

tobeimpossible.Wesimplylookforadisproofofit.<br />

Asaconsequence,theBHKnegationclauseisreformulatedas:<br />

•Aproofof ¬Aisadisproofof A<br />

droppingthe ⊥clause.So,insteadofanalyzingnegationintermsofimplicationandabsurdity,wetrytodothisintermsofdisproofs.<br />

Tennantpresentshisnotionofdisproofwithoutanyreferencetorelated<br />

work. Thoughitseemsquitenaturaltocomparetheseideaswithwhat<br />

DummettandPrawitzsaidaboutthepossibilityofdevelopingatheoryof<br />

meaningcenteredaroundthenotionofrefutation. 3<br />

3 Falsificationism:RefutationandDenial<br />

AccordingtobothDummettandPrawitz, 4 itispossibletothinkoftheories<br />

ofmeaningalternativetoverificationism,inparticulartooneinwhichthe<br />

meaningofsentencesisspecifiedbywhatcountsastheirrefutation. The<br />

primitivecharacterofrefutationscanbeendorsedbyconsideringthelinguisticactparalleltotheoneofassertion,thelinguisticmanifestationofthe<br />

possessionofarefutationofasentence:denial.Therelationshipofdenial<br />

torefutationisgovernedbytheprinciple:<br />

Thedenialofasentenceiswarrantedonlyifitsfalsityisrecognized<br />

wheretherecognitionofthefalsityofasentenceamountstothepossession<br />

ofarefutationofit.<br />

3 WhileTennantspeaksofdisproofs,wepreferrefutation. Aswillbecleared,theintuitionsbehindthetwonotionarecommon,eventhoughthedetailedtreatmentissensibly<br />

different.<br />

4 Theideaspresentedinthissectioncomefrom(Dummett,1991,Ch.13)and(Prawitz,<br />

1987, §6).


278 LucaTranchini<br />

ThistheoreticalperspectiveofPopperianflavorisgroundedontheintuitionaccordingtowhich,inacceptingasentence,aspeakermustalso<br />

bereadytoacceptallitsconsequences.Wheneveroneofitsconsequences<br />

turnsouttobeunacceptable,sotoomustthesentenceuponwhichitdependsberejected.Hencethecentralnotionofatheoryofmeaninginwhich<br />

denialplaysabasicrolewillbetheoneofconsequenceofasentence.Thus,<br />

themeaningofthelogicalconstantsisfixedbyeliminationrules,asthey<br />

typicallyspecifyhowasentencecanbeusedasassumptioninaderivation.<br />

AccordingtoDummettandPrawitz,oneneedsnottointroducenew<br />

technicaltoolstoaccountforrefutations. Ratherthenotionofrefutation<br />

canbedefinedinastandardnaturaldeductionframeworkprovidingan<br />

alternativeinterpretationofthedeductivesystem.<br />

Accordingtotheverificationistreading,onecaneasilyconstructproofsof<br />

morecomplexsentencesstartingfromproofsofsimpleroneswithintroductionrules.So,accordingtofalsificationism,onecanconstructrefutationsof<br />

morecomplexsentencesfromrefutationsofsimpleroneswithelimination<br />

rules.Forexampletakenarefutationof A:<br />

A.<br />

onecanobtainarefutationof A ∧Bwiththehelpofthe ∧eliminationrule:<br />

A ∧ B<br />

A E∧<br />

.<br />

Thecoreofthedeductiveprocesseswillthenbetheassumption,thatacts<br />

likeastartingpointofthederivationandthatonetriestorefute. Asa<br />

consequence,therulesthataretakentofixthemeaningoflogicalconstantsareeliminationrules.For,theyspecifytheconditionsunderwhicha<br />

sentencehavingtherelevantconstantasprincipaloperatorcanbeusedas<br />

assumptioninaderivation.Accordingly,thenotionofcanonicalrefutation<br />

(thatisofadeductioninwhicheliminationrulesplayaprominentrole)has<br />

beentakenastheexplicansofthenotionofmeaning.Thatis,toknowthe<br />

meaningofasentenceistoknowwhatcountsasacanonicalrefutationof<br />

it.<br />

Asintheverificationistframework,soherenotallderivationsaredirectly<br />

linkedtothebasiclinguisticact.Again,anopenderivationof Bfrom A<br />

A.<br />

B<br />

receivesahypotheticalreading:ifonecomesintopossessionofarefutation<br />

of B(i.e.,ifsheisinthepositionofdenying B),thenshewillalsobeinthe<br />

positionofdenying A.


NegationinProof-theoreticSemantics 279<br />

Justasintheverificationistcase,weintroduceanotionofevidencefor<br />

atomicsentencestowhichwereferasextra-deductiverefutingevidence.<br />

Suppose Bisthesentence‘Thecuponthetableisblue’,theempirical<br />

observationthatthecuponthetableisredcanbetakenasrefutingevidence<br />

for B.<br />

Technically,wedefineasubsetofthepropositionalconstants, F,containingtheatomsforwhichanextra-deductiverefutationisavailable.RefutationsendingwithatomsbelongingtoF,havingasopenassumptionsinstancesofonlyonesentence,willallowthedenialofthatsentence.<br />

Thedisanalogybetweenthetwoperspectivesconsistsinthelackofa<br />

connectiveactinginfalsificationismasimplicationdoesinverificationism.<br />

Suchaconnectiveshouldallowthedenialofasentencealsoinsituations<br />

inwhichonlyanopendeductionisathand. Asimplicationissaidto<br />

dischargetheopenassumption,sotheconnectiveinquestioncouldbesaid<br />

to“dischargetheconclusion”ofthededuction.<br />

Butdoesthestandardlanguagepossessatoolwhichcanbetakenin<br />

somesensetodischargeconclusions? Negationcanbe(partially)thought<br />

ofintheseterms.Consideraderivationof A<br />

.<br />

A.<br />

Thenegationeliminationrulecanbeseenasawayofclosingtheconclusion<br />

ofthederivationbyintroducingamorecomplexassumption:<br />

.<br />

A ¬A<br />

⊥ .<br />

Thisisactuallyinfullanalogywiththewayinwhichimplicationworks:it<br />

closesanassumptionandintroducesamorecomplexconclusion.<br />

4 Towardaunifiedframework<br />

Thetwotheoriesofmeaning,verificationistandfalsificationism,havebeen<br />

treatedbybothDummettandPrawitzastwodifferent(concurrent)theoreticalenterprises.Thatis,asemanticsforagivenlanguagecanbedeveloped<br />

eitheraccordingtotheverificationistorthefalsificationiststandpoint. 5<br />

Onthecontrary,Tennant’ssuggestionsontheroleof ⊥inadeduction<br />

areveryneartothefalsificationistperspective.Bylookingathisproposal<br />

inmoredetail,onerealizesthatitisnothingbutamixtureofthetwoviews<br />

onmeaning.<br />

5 Dummettactuallygivesreasonsfordevelopingsimultaneouslybothperspectives. But<br />

eveninsuchacasethetwotheoriesaredistinct.


280 LucaTranchini<br />

ThelimitsofTennant’sapproach<br />

Aswesaw,Tennant’ssuggeststoread ⊥asamarkerofdeadendsindeductions.<br />

Inotherwords,adeductionendingwith ⊥istakenasadeduction<br />

withnoconclusion.Hence,heproposestointerpretnaturaldeductionsystemsasprovidingthemeansforproducingopenproofs,closedproofsand<br />

disproofs.<br />

Supposeonehasalogicalsystemforwhichtheexistenceof<br />

proofsisindicatedbytheusualturnstile ⊢,arelationofexactdeducibilityholdingbetweenpremisesontheleftandaconclusion<br />

ontheright.Theintuitivemeaningof‘X ⊢ A’isthatthereisa<br />

proofwhoseconclusionis Aandwhosepremises(undischarged<br />

assumptions)formtheset X.[...]<br />

Therearetwoextremecases.<br />

1. Xisempty. Then‘⊢ A’means Aisatheorem. Thatis,<br />

thereisaproofof A‘fromnoassumptions’.[...]<br />

2. Ais‘empty’.Then‘X ⊢’meansthatthereisadisproofof<br />

X,thatis,adeductionshowingthat Xisinconsistent.<br />

Accordingly,insteadoftheusualinductivedefinitionofproof,Tennant’s<br />

givesasimultaneousdefinitionofthenotionsofproofanddisproof.<br />

Still,inthelightoftheconsiderationsonfalsificationism,Tennant’sapproachcanbecriticizedfortheasymmetryinthetreatmentofthetwonotions.Inparticular,totreatopendeductionsasopenproofsmeanstotreat<br />

themas“incompleteproofs”:theyaremeansofobtainingclosedproofsof<br />

theconclusions,providedclosedproofsoftheassumptions.Butwhyshould<br />

theynotbeconsideredas“incompleterefutations”,thatisasmeansof<br />

rejectingthepremisesoncerefutationsoftheconclusionsareprovided?<br />

ThisasymmetrycanalsobeseeninTennant’swayofdealingwithrules,<br />

ingivingthedefinitionofproofanddisproof.Introductionrulescanbeused<br />

onlytoproduceproofs. Eliminationrules,ontheotherhandcanbeused<br />

toproduceeitherproofsordisproofs,dependingonwhetherthedeductions<br />

oftheminorpremisesareproofsordisproofs. Herearethetwocasesfor<br />

disjunction:<br />

.<br />

A ∨ B<br />

[A]<br />

.<br />

C<br />

C<br />

[B]<br />

.<br />

C<br />

.<br />

A ∨ B<br />

[A]<br />

.<br />

⊥<br />

⊥<br />

Thepointisthatalsointroductionrulescanbeusedinproducingrefutations.Justlikeintheverificationistperspectiveoneproducesnon-canonical<br />

[B]<br />

.<br />


NegationinProof-theoreticSemantics 281<br />

proofswitheliminationrules,soinfalsificationismoneproducesnon-canonicalrefutationswithintroductionrules.<br />

Finally,itisimplicitinTennant’slineofargumentthattheroleof ⊥in<br />

disproofsisanalogoustotheroleofdischargedassumptionsinproofs.Asa<br />

closedproofisadeductionwithnoopenassumptions,soarefutationisadeductionwithnoconclusion.Furthermore,evenifTennantdoesnotconsiderit,wesawthatinordertouseanaturaldeductionsystemformeaningtheoreticalpurposesonealsohastoaccountforextra-deductiveevidence<br />

foratomicsentences.Thereisadeepanalogyoftheroleofextra-deductive<br />

probativeandrefutingevidenceforatomicsentencesand(respectively)the<br />

roleofdischargingtheassumptionsandreachingadeadendinadeduction.<br />

Forallthesearethemeansthroughwhichanopendeductionistakeninto<br />

aclosedone(eitheraprooforarefutation).<br />

Alltheseconsiderationssuggestthepossibilityofre-framingthenatural<br />

deductionsysteminordertoexplicitlyshowthesesymmetries.<br />

Top-closedandbottom-closedderivations<br />

Bothperspectivesonmeaningdistinguishbetweenderivationsthatimmediatelyallowalinguisticperformanceandthosethatdonot.Inverificationismwehaveadistinctionbetweenclosedandopenproofs.Itseemsnatural<br />

toadaptthisterminologytorefutations,sothatwehaveopenandclosed<br />

refutations.<br />

Aswesaw,Tennantproposestotreat(whatinthestandardframework<br />

areconsidered)derivationsofconclusion ⊥asdisproofs.For, ⊥hasnosententialcontentandhencecan’tbetakentobetheconclusionofadeductive<br />

process.Rather,itregistersthefactthatthedeductivepathleadingtothe<br />

conclusionofthederivationisadeadend,orinotherwords,itisclosed.<br />

Canthesetwonotionof“closure”,theoneregisteredby ⊥andtheoneof<br />

deductiveprocesseslinkedtolinguisticacts,betakenintoone?<br />

Toexplicitlystatetheanalogyweintroducethesign ⊤tomarkassumptionclosure.Sowheneveranassumptionisclosed,wewillmarkit<br />

⊤.Inthe<br />

caseofassumptionsdischargethroughimplicationthissimplyamountstoa<br />

notationalchange.Insteadofputtingthesentenceinbrackets(oroverlining<br />

it),weputthesign ⊤overit.So,theintroductionruleforimplicationwill<br />

appearas:<br />

⊤ A.<br />

B<br />

A → B<br />

Asweobservedtherearetwodifferentwaysinwhichanopendeduction<br />

canbetakenintoaclosedone. Onecancloseoneoftheedges(assump-


282 LucaTranchini<br />

tionsorconclusion)bylogicalmeans,inverificationismwithimplication,in<br />

falsificationismwithnegation;alternativelyonecantrytoreachtheatomic<br />

componentsofthesentencetobeprovedorrefuted,toseeifthereisextradeductiveevidencepurportingorrefutingsuchcomponents.<br />

Ifweconsiderverificationism,thenotionofclosure(bymeansofwhich<br />

weusuallyrefertodischargedassumptions)appliesquitewellalsotoatomic<br />

sentencesforwhichwehaveextra-deductiveprobativeevidence:anassumptionisclosedwhentheconclusionofthedeductiveprocessdoesnotdepend<br />

onit.Andclearly,notonlytheassumptionsdischargedthroughimplication<br />

areclosed,butalsotheatomiconesforwhichextra-deductiveevidenceis<br />

available.Thissuggeststheideaofextendingtheuseof ⊤tomarktheclosureoftheatomicassumptionsaswell.Accordingtothewayinwhichwe<br />

introducedatomicsentencesinnaturaldeductioninsection1,wecanuse<br />

⊤toexplicitlymarktheatomicsentencesbelongingtotheset Tofverified<br />

atoms.Todothis,weaddanewruletothenaturaldeductionsystem:<br />

if A ∈ Tthen<br />

isaderivationofconclusion Afromnoassumptions.<br />

⊤ A<br />

Forexample,supposetheweatheriswindy:insuchacase,theconclusion<br />

ofthederivation<br />

⊤ ⊤<br />

Itrains Itiswindy<br />

Itrainsanditiswindy I∧<br />

Ifitrainsthenitrainsanditiswindy<br />

I →<br />

canbeasserted,becauseitdoesnotdependonanyassumption,evenifthe<br />

twoassumptionsareclosedindifferentways:thefirstoneisdischargedby<br />

theapplicationofthe I →rule;thesecondoneisclosedbytheavailability<br />

oftheempiricalevidenceforit.<br />

Inanalogywiththis,infalsificationismwehavetwowaysoftakingan<br />

opendeductionintoarefutation:eitherrefutingevidenceisprovidedforthe<br />

conclusion;oralternatively,wecanusesomelanguagedevicesto“discharge”<br />

theconclusioninthecourseofthederivation.<br />

Ifwetakeacloselookatthefirstpossibility,Tennant’sideaof ⊥as<br />

registeringaknotofinconsistencyfitsthissituationquitewell.For, ⊥can<br />

betakentoregisteranincompatibilitybetweentheoutputofthedeductive<br />

processandtheavailableevidence.Thissuggeststhepossibilityofextendingtheuseof<br />

⊥,bymarkingwithittheatomicconclusionsofderivations,<br />

forwhichweareinpossessionofextra-deductiverefutingevidence.Wecan<br />

formallyachievethiswitharuleanalogoustotheoneforatomicassumptions:


NegationinProof-theoreticSemantics 283<br />

if A ∈ Fand<br />

isaderivationofconclusion Afromassumptions Γthen<br />

Γ.<br />

A<br />

Γ.<br />

A<br />

⊥<br />

isaderivationofnoconclusionfromassumptions Γ.<br />

Forexample,supposethecuponthetableisred:insuchacasewemark<br />

theconclusionofthefollowingderivationwith ⊥:<br />

Thecuponthetableisblueanditisfulloftea<br />

Thecuponthetableisblue<br />

⊥<br />

Thisextensionoftheuseof ⊥makesitpossibletoschematicallyrepresent<br />

thecoreprocessesoffalsificationismas<br />

A.<br />

⊥<br />

Thispatternstandsforarefutation(andhenceallowsthedenial)ofagiven<br />

sentence A.Wewillalsorefertosuchdeductivepatternsasbottom-closed<br />

derivations.Onceintroducedthesign ⊤inordertomarkclosedassumptions<br />

indeductions,itispossibletorepresentthecoreprocessesofverificationism<br />

withthescheme:<br />

⊤. A<br />

standingforaproof(andhenceallowingtheassertion)ofthesentence A.<br />

Wewillrefertosuchdeductivepatternsastop-closedderivations.<br />

Newhorizonsforproof-theoreticsemantics<br />

Aswepreviouslyunderscoredwhatweareproposingisaunifiedframework<br />

inwhichbothproofsandrefutationscanbeaccountedfor.Todothiswe<br />

havetoaddasetofpropositionalconstants Ptoastandardnaturaldeductionsystemandbothasubset<br />

Tofverifiedatomsandasubset Fofrefuted<br />

atomshavetobespecified.Atthispointwehavethattop-closedderivation<br />

andbottom-closedderivationscountasclosedproofandclosedrefutations


284 LucaTranchini<br />

forsentences,i.e.,theyallowtheassertionanddenialofsentences. Itremainsonlyadisanalogybetweenthetwokindsofderivations,namelythat<br />

whiletoatop-closedderivationalwayscorrespondstheassertionofthe<br />

conclusion,toabottom-closedderivationcorrespondsadenialonlyifthe<br />

assumptionsofthedeductionareoccurrencesofthesamesentence.Thisis<br />

duetothefactthatthenaturaldeductionframeworkallowsatmostone<br />

conclusionbutthereisnolimitonthenumberofpossibleassumptions.<br />

Besidethis,whatlooksreallyproblematicforthefulldevelopmentofthis<br />

perspectiveisthedefinitionofvalidity.For,inverificationismanopendeductionisvalidif,providedclosedderivationsoftheassumptions,itreduces<br />

toaclosedproof;infalsificationismanopendeductionisvalidif,provided<br />

aclosedderivationoftheconclusion,itreducestoaclosedrefutationof<br />

theassumption(s).Inotherwords,inbothperspectivesthecategoricalnotionofclosedderivationhasprimacyoverthehypotheticalnotionofopen<br />

derivation.Thepointisthatitisnotclearhowthenotionofvalidityisto<br />

beshapedinasysteminwhichwehavetwodistinctcategoricalnotions.<br />

Thedirectioninwhichthesolutioncanbefoundistherejectionofthe<br />

proof-theoreticdogmaaccordingtowhichthecategoricalnotionhasprimacyoverthehypotheticalone.<br />

Bydoingthiswecouldreallyembody<br />

Tennant’sintuitionaccordingtowhichclosedproofsarejustlimitcasesof<br />

openones.Intuitively,thismeansthatthegroundconceptofproof-theoretic<br />

semanticsistherecognitionofdeductivelinksamongsentences,thatonly<br />

inveryspecialoccasionscanbetakentobeorientedbytheconclusionor<br />

bytheassumption. Thisideacanbeseenatworkinreadingtherulefor<br />

makingassumptionsinnaturaldeduction:<br />

forany A<br />

isadeductionhaving Aasconclusionand Aasassumption<br />

A<br />

Bothverificationismandfalsificationismareforcedtoreadtheruleasproducing“incomplete”derivations,inthesenseofeitheranincompleteproof<br />

of Aoranincompleterefutationof A.Fromtheunifiedperspective,therule<br />

forassumptionisinterpretedsimplyas‘Consider A’:inconsidering Awe<br />

areneithercommittedtotheexpectationofaproofnortotheexpectation<br />

ofarefutationofit,weareopentoseewhatwillhappenatlaterstagesof<br />

thedevelopmentofthedeductiveprocess.<br />

Formalmodelswhichexplicitlyendorsethisintuitionaresequentcalculi.<br />

Insuchsystemsthefullsymmetrybetweenassumptionsandconclusion,<br />

i.e.assertionanddenial,isembodiedinthesymmetrybetweenleftand<br />

rightsideoftheturnstile.Comingbacktovalidity,itisinterestingtonote<br />

thatnoquestionofvalidityhaseverbeenaddressedforsequentcalculiand<br />

itisnotcompletelyclearhowtoformulateit. Itwouldnotbesurprising


NegationinProof-theoreticSemantics 285<br />

thatratherthanaglobaldefinitionofvaliditywhatisneededaresimply<br />

localcriteriatobeimposedonrules. 6 Butwedonotpushtheissuefurther.<br />

AbsurdityandConsistency<br />

Aswesaw,inordertofixthemeaningof ⊥viadeductiverules,theverificationisthastorequirethatallatomicsentencesoflanguagecan’tbe<br />

simultaneouslyasserted. Otherwisenothingbansthepossibilityofasserting<br />

⊥,violatingtheBHKclausethatstatesthat ⊥can’tbeassertedin<br />

anysituation.<br />

Theinterpretationweareproposingclearlymakestheproblemofthe<br />

assertionof ⊥disappear,as ⊥isnolongertobeconsideredasentential<br />

contentcapableofbeingasserted(ordenied). Nonethelesstheintuition<br />

that ⊥can’tbeassertedcanbereformulatedasfollows. Wenotedthata<br />

sentencecanbeassertedwhenweareinpossessionofaderivation,having<br />

thesentenceasconclusion,withnoopenassumptions. So,theexpression<br />

“⊥canbeasserted”appearsasaroughwaytorefertoasituationthatwe<br />

canschematicallyrepresentinthisway:<br />

⊤.<br />

⊥<br />

Howisthispatterntoberead? Itlookslikeaderivationinwhichboth<br />

assumptionsandconclusionhavebeenclosed. Tobetterunderstandit,<br />

considerasentence Afiguringinthederivation:<br />

⊤. A.<br />

⊥<br />

Ifwesplitupthisdeductivepatternwefindourselveswiththefollowing:<br />

⊤. A<br />

Accordingtothereadingof ⊥and ⊤,thesederivationsamounttoaproof<br />

of Aandtoarefutationof A.Thatis,thepossessionofboththetop-and<br />

bottom-closedderivationsallowsboththeassertionandthedenialof A.<br />

Ifwetakethesentence Atobeanatomicsentence,e.g.,‘Thecupon<br />

thetableisred’,thetop-andbottom-closeddeductivepatternisavailable<br />

6 ThisdirectionisstronglycalledforbySchroeder-Heister(seeforinstance(Schröder-<br />

Heister,2009)).<br />

A.<br />


286 LucaTranchini<br />

onlyifweareinpossessionbothofsupportingandrefutingevidenceforit.<br />

Obviously,thefactthatthesentence‘Thecuponthetableisred’canbe<br />

neitherprovednorrefuteddoesnotdependondeduction,butratheronthe<br />

factthatitisnotpossibletoseearedcupandagreencuponthetableat<br />

thesametime.<br />

Usuallyconsistencyistakentobetheimpossibilityofassertingtheabsurdity.Intheframeworkwearedevelopinganalternativenotionofconsistencycanbeputforward:namely,theimpossibilityofbeinginthepositionofassertinganddenyingasentenceatthesametime.Thisnotionofconsistencyamountstotheimpossibilityofobtainingdeductivepatternshaving<br />

bothassumptionsandconclusionsthatareclosed.<br />

JustlikeDummett,topreserveconsistencywehavetoimposearestrictiononatomicsentences:<br />

namely,wehavetorequirethateveryatomic<br />

sentencecan’tbebothassertedanddenied(wewillrefertothisasatomic<br />

consistency). Ourrestrictioncan’tbejustifiedonlogicalbasis,justas<br />

Dummett’sone.Nonetheless,itismuchmoreplausibletorequirethateach<br />

atomicsentencecan’tbeassertedanddeniedatthesametimeratherthan<br />

torequirethattheremustbemutuallyincompatibleatomicsentences. In<br />

particular,sucharestrictioncouldbefullyarguedfor,onthebackground<br />

ofconsiderationsonhumancognition.<br />

5 Conclusions<br />

Traditionally,thepossibilitiesofdevelopinganaccountofassertionandan<br />

accountofdenialhavebeenconsideredtwodifferententerprises.Togivean<br />

accountofthemeaningofnegationwesuggestedtodevelopauniqueframeworkinwhichthecentralroleisplayedbythenotionofopendeduction.<br />

Bymeansofthe ⊤and ⊥signswecangiveanaccountinwhichdeductive<br />

patternscountasproofsandrefutationsofsentences,i.e.,allowtheirassertionanddenial.Aswesaw,introductionrulesareatthecoreoftheprocess<br />

ofproof,whileeliminationsareatthecoreoftheprocessofrefutation.<br />

AtthispointwecanreconsiderthealternativetotheBHKclausefor<br />

negationTennantproposed:<br />

Aprooffor ¬Aisarefutationof A<br />

Inthelightoftheconventionsintroducedwecanschematizetheequivalence<br />

as:<br />

A.<br />

⊤.<br />

⊥ = ¬A<br />

Itseemsthatwearefacedwithasortofgeometricaloperationonderivation:byrotating180<br />

◦ arefutationof Acanbeturnedintoaproofof ¬A.


NegationinProof-theoreticSemantics 287<br />

Hence,negationappearsasalinguisticdevicethatstatesinanexplicitway<br />

theimplicitharmony,embodiedintheinversionprinciple,holdingbetween<br />

groundsforasentenceandconsequencesofasentence. Indeed,inversion<br />

governstherelationshipbetweenintroductionandeliminationrulesand<br />

negationtheonebetweenproofsandrefutationswhicharedirectlyconnectedtothetwosetsofrules.<br />

Afurtherquestionnaturallyarises,namelywhetherthestandardrules<br />

fornegationdoproperlyseizethiscrucialfeatureoftheconnective. If<br />

theanswertobegivenwerenegative,thanmovingapartfromstandard<br />

intuitionisticlogicwouldbenecessary.<br />

Inconclusion,webelievetohaveisolatedtheroleofnegationinthe<br />

architectureofdeductiveactivityasbeingradicallydifferentfromthatof<br />

otherconnectives.Eventhoughthedevelopmentofaunifiedframework,in<br />

whichtoaccountforboththeactivitiesofproofandrefutation,seemsto<br />

requirefurtherinvestigation,webelieveittobeanimportantsteptofully<br />

developaproof-theoreticaccountofthemeaningoflogicalconstants,asthe<br />

analysisofnegationemergingfromitshows.<br />

LucaTranchini<br />

Wilhelm-SchickardInstitutfürInformatik,TübingenUniversity<br />

Sand13,72076Tübingen,Germany<br />

luca.tranchini@gmail.com<br />

References<br />

Brouwer,L.E.J.(1908).Theunreliabilityofthelogicalprinciples.InA.Heyting<br />

(Ed.),Collectedworks(Vol.I,pp.443–446).<br />

Dummett,M.(1991).Thelogicalbasisofmetaphysics.London:Duckworth.<br />

Prawitz,D.(1987).Dummettonatheoryofmeaninganditsimpactonlogic.In<br />

B.M.Taylor(Ed.),MichealDummett.<br />

Schröder-Heister,P. (2009). Hypotheticalreasoning: AcritiqueofDummett-<br />

Prawitz-styleproof-theoreticsemantics.InTheLogicaYearbook2008.(Sequent<br />

CalculiandBidirectionalNaturalDeduction: OntheProperBasisofProof-<br />

TheoreticSemantics.)<br />

Tennant,N. (1999). Negation,absurdityandcontrariety. InD.M.Gabbay&<br />

H.Wansing(Eds.),Whatisnegation? KluwerAcademicPublishers.


Oiva Ketonen’s Logical Discovery<br />

Michael von Boguslawski<br />

1 ShortbiographyofOivaKetonen<br />

OivaToivoKetonenwasbornJanuary21,1913,inthemunicipalityofTeuva<br />

intheSouthernOstrobothniaregionofFinland. 1 Hewaschildnumbereight<br />

inafamilythatraisedaltogether13children.Alreadyatayoungagethe<br />

law-governednessofnaturemadeadeepimpressiononhimandapparently<br />

plantedtheseedforaninterestinthenaturalsciences. Ketonenwasthe<br />

onlyoneofthefamily’schildrentogetanyformofhighereducation.<br />

KetonengraduatedfromKristiinakaupunginLukio(roughlyequivalent<br />

tohighschool)in1932,andenrolledintothedepartmentofhistoryand<br />

linguistics(wherephilosophyinHelsinkiwastaughtatthattime)atthe<br />

universityofHelsinki. TheprofessorofphilosophyatthattimewasEino<br />

Kaila,whohadcloseconnectionswiththeWienerKreisanditwasdue<br />

tohispersonaleffortsthatlogicarrivedinFinland. Ketonenswitchedto<br />

thedepartmentofmathematicsayearlater,despitehavingdoubtsthat<br />

mathematicsalonewouldsatisfyhisacademicinterests.Ketonen’steacher<br />

inmathematicsbecameRolfNevanlinna,thefamouscomplexfunctiontheoretician,andwecantellfrompreservedcorrespondencethatNevanlinna<br />

wasextremelyimpressedbyKetonen’smathematicalabilities.<br />

Therewasonlyonetext-bookonlogicavailableinFinnishatthattime<br />

—ThiodolfRein’sMuodollinenlogiikka—Formallogic(freetranslation<br />

fromFinnish)whichtreatedonlyAristotelianlogic.Therewasachangein<br />

thecurriculum,however,andBertrandRussell’sTheproblemsofphilosophyandKaila’sNykyinenmaailmankäsitys—Thepresentworld-view(free<br />

translationfromFinnish),amongothers,wereintroduced.Theteachingof<br />

logicwas,accordingtoKetonen,confinedtothebasicsandcouldnotas<br />

1 AnextendedversionofthisarticlewillappearintheYearbookoftheViennaCircle<br />

institute.IwouldalsoliketothankOiva’ssonTimoforgenerouslyprovidingmewitha<br />

copyofOiva’sunfinishedautobiography.


290 MichaelvonBoguslawski<br />

such,Ketonenspeculates,causeanyinterest.WereadinKetonen’sstudy<br />

bookthathedidnottakeasinglecourseinlogic.<br />

AccordingtoTimoKetonen,Oiva’searlyinterestswerealgebraandnumbertheory,whichpavedthewayforthehugeinterestinGödel’sfirstincompletenesstheorem,ofwhichhewasmadeawarebyhisfellowstudent,Max<br />

Söderman.Nevanlinnaalsolatermentionedthetheorem. 2 Gödel’sfantastic<br />

resultwasprobablywhatignitedKetonen’sinterestinformallogic.Ketonenwritesintheautobiographythathefrequentlywenttoeveningmeetings<br />

ofwhathecalled“Thephilosophicalclub.” Thesemeetingsseemtohave<br />

beenquiteunofficial,usuallythegroupgatheredatthehomeofoneofthe<br />

professors,e.g.,KailaorYrjöReenpääandlogicwasamongthetopicsdiscussed.TheyalsogatheredatleastonceatSöderman’shome.Inthestudy<br />

diarywecanreadthathelateralsospentsomeeveningsattendingwhat<br />

hecalls“mathematical-logicalconferences”. Itisunclearatthismoment<br />

whethertheseconferencesandthemeetingsofthe“philosophicalclub”were<br />

thesame.<br />

NevanlinnatriedtoconvinceKetonentotakeupfunctiontheory—<br />

anotherwitnessofNevanlinna’sfaithinKetonen’sabilities—butKetonen,<br />

aftersomecontemplation,decidedtoworkonlogic.Hewrotehismaster’s<br />

thesisonaxiomaticlogic,arithmetic,andGödel’stheorem. Thefirstpart<br />

waspublished(Ketonen,1938)andusedbyKailaasatextbookforlogic<br />

courses. KetonenhadreceivedtheimpressionfromNevanlinnathatsome<br />

mathematicianssuspectedthattherewassomefaultinGödel’sproof,and<br />

thatthisfaultmightbeworthuncovering. Ketonenbelievedthatasa<br />

resultofhisworkwiththethesis,hesucceededinstreamliningGödel’s<br />

proofsomewhat. 3<br />

KetonenkeptworkingonGödel’sresultsandmadeasmallimprovementtoGödel’scompletenesstheoremforthepredicatecalculusin1941<br />

(Ketonen,1941). Gödelshowedthatthateitheraproposition Aisprovable,oritisimpossiblethattheredoesnotexistacounterexample.Ketonenimprovedthisresultsothatthiscounterexamplecanbefounddirectly.<br />

Söderman,whoresidedinViennaatthetime,reportedKetonen’sresultto<br />

Gödel,whoadmittedthatitwasindeedanimprovement(vonPlato,2004).<br />

2 TheDissertation—UntersuchungenzumPrädikatenkalkül<br />

Accordingtohisautobiography,Ketonenhaddecidedalreadyinthespring<br />

of1938togoforadissertationimmediately. Hewenttotheuniversityin<br />

Göttingen,mostprobablywiththeaidofNevanlinna’scontacts,whohad<br />

2<br />

HowwellNevanlinnawasacquaintedwithlogic,andwhathethoughtoftheatthetime<br />

completelynewdiscipline,remainsdebated.<br />

3<br />

Wehopetoinvestigatethisstreamlininginalaterwork.


Ketonen’sDiscovery 291<br />

workedattheuniversityasavisitingprofessorin1936–1937.Kailahadmet<br />

GentzeninMünsterin1936aswell.KetonenalsowenttoMünsterwhere<br />

hemet—amongothers—HeinrichScholzwithwhomtherewassome<br />

correspondence. Shockingly,theverysamenightthatKetonenarrivedin<br />

Göttingen,9–10November1938,laterbecameinfamousasthe“Kristallnacht”—“crystalnight”.<br />

InGöttingen,intheautumnof1938,Ketonen<br />

becameGerhardGentzen’spresumablyfirst—andalsolast—student,<br />

althoughKetonenhadtowaituntilChristmastoreceiveaproblemfrom<br />

Gentzentoworkon.HerecallsGentzenasasympatheticyoungmanwho<br />

“didnottalkmuch”butmentionedthathischiefassignmentasHilbert’s<br />

assistantwasthereading(apparentlyaloud)of“popular”scientificpublicationstohisprofessor.Thedissertation(Ketonen,1944),UntersuchungenzumPrädikatenkalkül,isdividedintothreeparts.ThefirstpartpresentsandimprovesGerhard<br />

Gentzenssequentcalculusbyintroducinginvertiblerulesforthecalculus’<br />

propositionalparts, 4 parttwodiscussesacertainSkolemnormalizationof<br />

derivations,andthethirdpartappliestheresultsfrompartsoneandtwoto<br />

produceaproofoftheunderivabilityofEuclid’sparallellpostulatefromthe<br />

restoftheSkolem-axiomsforEuclideangeometry.Ketonenwasthefirstto<br />

continueSkolem’sworkongeometry(vonPlato,2007b). Theinvertibility<br />

resultwillnowbepresentedindetail.<br />

InvertibilityofRulesinGentzen’sLK<br />

Asequentisoftheform A1,A2,... ,Am → B1,B2,...,Bn. Capitallatin<br />

letters A,B,C,...willbeusedtodenoteformulas,capitalgreekletters<br />

Γ,∆,Θ,...willbeusedtodenotethe(possible)contextofaderivation.<br />

Contextsaretreatedaslistsofformulas. Theformulastotheleftofthe<br />

sequentarrow →makeuptheantecedent,theformulastotherightthe<br />

succedent. Thesequentarrowcanconvenientlybereadas“gives”. Thus<br />

thesequent A&B → Cmeansthatfromtheassumptions Aand Btogether,<br />

theconclusion Cfollows. Thesequentisreadas“Aand Bgives C”. A<br />

sequentshouldbeviewedasageneralizationoftheconceptofderivability,<br />

withoneormoreassumptionsintheantecedentgivingoneormorepossible<br />

casesinthesuccedent. Weusetheparenthesesintheusualway,andall<br />

theconnectives ¬, ∨,&,and ⊃. Forthefalsesentence(andtodenotea<br />

contradiction),Gentzenusesaspecialsymbolbutwewillnotneedithere.<br />

Theonlyaxiomistheinitialsequent A → A. Tobeabletocarryout<br />

derivationsandproofswithinthesystem,weneedlogicalandstructural<br />

rules.Thelogicalrulesmanipulateconnectiveswhereasthestructuralrules<br />

manipulateformulas.Derivationsarein“tree-form”andbeginfrominitial<br />

4 Obviously,invertibilityinKetonen’ssensecannotholdforthepredicatepart.


292 MichaelvonBoguslawski<br />

sequents(andpossiblycontexts)attheendofbranches,andendwiththe<br />

provensequentatthebottomofthetree,the“root”. 5 Below 6 aregiventhe<br />

structuralandlogicalrulesofGentzen’sfirstsystemofsequentcalculus,<br />

whichwetodaycallGentzenLK:<br />

StructuralrulesforGentzenLK<br />

Γ → Θ<br />

LW<br />

A,Γ → Θ<br />

Γ → Θ<br />

RW<br />

Γ → Θ,A<br />

Leftweakening Rightweakening<br />

A,A,Γ → Θ<br />

LC<br />

A,Γ → Θ<br />

Γ → Θ,A,A<br />

RC<br />

Γ → Θ,A<br />

Leftcontraction Rightcontraction<br />

∆,B,A,Γ → Θ<br />

LE<br />

∆,A,B,Γ → Θ<br />

Γ → Θ,B,A,Λ<br />

RE<br />

Γ → Θ,A,B,Λ<br />

Leftexchange Rightexchange<br />

Γ → Θ,B B,∆ → Λ Cut<br />

Γ,∆ → Θ,Λ<br />

Cut<br />

LogicalrulesforGentzenLK<br />

Γ → Θ,A Γ → Θ,B<br />

R&<br />

Γ → Θ,A&B<br />

A,Γ → Θ B,Γ → Θ<br />

L∨<br />

A ∨ B,Γ → Θ<br />

Rightconjunction Leftdisjunction<br />

A,Γ → Θ<br />

L&1<br />

A&B,Γ → Θ<br />

B,Γ → Θ<br />

L&2<br />

A&B,Γ → Θ<br />

Leftconjunction1 Leftconjunction2<br />

Γ → Θ,A<br />

R∨1<br />

Γ → Θ,A ∨ B<br />

Γ → Θ,B<br />

R∨2<br />

Γ → Θ,A ∨ B<br />

Rightdisjunction1 Rightdisjunction2<br />

A,Γ → Θ<br />

R¬<br />

Γ → Θ, ¬A<br />

Γ → Θ,A<br />

L¬<br />

¬A,Γ → Θ<br />

Rightnegation Leftnegation<br />

A,Γ → Θ,B R ⊃<br />

Γ → Θ,A B,∆ → Λ L ⊃<br />

A ⊃ B,Γ,∆ → Θ,Λ<br />

Γ → Θ,A ⊃ B<br />

Rightimplication Leftimplication<br />

5 Aderivationmayofcoursehaveonlyonebranch,i.e.,haveonlyoneinitialsequentfrom<br />

whichsomeothersequentisproven.<br />

6 See(Gentzen,n.d.)fordetails.


Ketonen’sDiscovery 293<br />

Withinvertibilityismeantthatifasequentmatchestheconclusionofa<br />

rule,andifitisderivable,thenthecorrespondingpremissesarederivable.<br />

Gentzen’sLKisnotinvertible.ConsiderruleR∨2,forexample.Ifitwere<br />

invertible,thenthesequent A → Bwouldbederivablebecause A → A ∨ B<br />

isderivablefromtheinitialsequent A → A. A → Bisnotatallatautology<br />

soitclearlyshouldnotbederivablewithoutassumptionsinacompleteand<br />

consistentsystem. Thus,thelogicalrulesforleftconjunctionandright<br />

disjunctionneedtobereplacedwithinvertibleones,andKetonennotes<br />

thattheruleforleftimplicationwillhavetobereplacedwitharulewhich<br />

hasthesamecontextsinitstwopremisses:<br />

A,B,Γ → ∆ L&<br />

A&B,Γ → ∆<br />

Ketonen’sinvertiblerulesforGentzenLK<br />

Γ → ∆,A,B R∨<br />

Γ → ∆,A ∨ B<br />

Γ → ∆,A B,Γ → ∆ L ⊃<br />

A ⊃ B,Γ → ∆<br />

Theproofsoftheinvertibilityoftherulesareeasyandshort.Theones<br />

givenherediffersomewhatfromthosegivenbyKetonen,specificallysothat<br />

whenKetonenintroducestheconclusionofaruletheinvertibilityofwhich<br />

istobeprovedthroughaninstanceofitsnon-invertiblecounterpart,we<br />

simplyintroducetheconclusionaftertheverticaldotsthatindicatesome<br />

possiblederivation.<br />

A → A LW<br />

B,A → A LE<br />

ProofofinvertibilityofruleL&<br />

B → B LW<br />

A,B → B R&<br />

A,B → A .<br />

A,B → A&B A&B,Γ → Ω<br />

Cut<br />

A,B,Γ → Ω<br />

TheproofoftheinvertibilityofruleR∨issimplyahorizontal“mirror<br />

image”oftheproofabove. InordertoprovetheinvertibilityofruleL⊃,<br />

weshowthatbothpremissesarederivablefromtheconclusionbycut:<br />

A → A RW<br />

A → A,B R ⊃<br />

.<br />

→ A,A ⊃ B A ⊃ B,Γ → Ω<br />

Cut,RE<br />

Γ → Ω,A<br />

B → B LW<br />

A → A,B R ⊃<br />

.<br />

B → A ⊃ B A ⊃ B,Γ → Ω<br />

Cut<br />

B,Γ → Ω<br />

Withtheinvertiblerules,wecancarryouta“root-first”proofsearch<br />

inanalgorithmicfashion,beginningwiththesequentwewanttoprove,


294 MichaelvonBoguslawski<br />

andthenapplyingtherulesinreverseuntilwereachasituationwithonly<br />

initialsequents(andpossiblycontexts).Thisproofsearchwillterminate,so<br />

itcanintheorybedonebyacomputer.IndeeditispossiblethatKetonen’s<br />

sequentcalculusisthefirstsystemthatwouldpermitacomputertoproduce<br />

proofs.Itdoesnotmatterinwhichordertherulesareappliedinreverse,<br />

asthetwoproofsof → (A ⊃ B) ⊃ (¬B ⊃ ¬A)belowillustrate:<br />

A → A LW<br />

¬B,A → A R¬<br />

¬B → ¬A,A R ⊃<br />

B → B RW<br />

B → B, ¬A L¬<br />

B, ¬B → ¬A R ⊃<br />

B → ¬B ⊃ ¬A L ⊃<br />

→ ¬B ⊃ ¬A,A<br />

A ⊃ B → ¬B ⊃ ¬A<br />

→ (A ⊃ B) ⊃ (¬B ⊃ ¬A)<br />

A → A LW<br />

¬B,A → A R¬<br />

¬B → ¬A,A<br />

R ⊃<br />

B → B RW<br />

B → B, ¬A L¬<br />

B, ¬B → ¬A L ⊃<br />

A ⊃ B, ¬B ⊃ → A R ⊃<br />

A ⊃ B → ¬B ⊃ ¬A<br />

→ (A ⊃ B) ⊃ (¬B ⊃ ¬A)<br />

Themodificationoftherulesdoesnothamperthepropertiesofthe<br />

system,theHauptsatz,forexample,stillholds. KurtSchütteandHaskell<br />

Currygavecut-freeproofsofinvertibilityin1950and1963respectively,<br />

Currywiththeaddedresultthatinversionsareheightpreserving. 7<br />

Reactionstothethesisandfollow-up<br />

PaulBernays(Bernays,1945)wroteafavorablereviewofKetonen’sthesisinTheJournalofSymbolicLogicin1945,andKleenenotes(Kleene,<br />

1952)thatheknowsofKetonen’scalculusonlythroughthisreview. We<br />

knowthroughseveralsources,forexample(vonWright,1951),thatseveralresearchersincludingRichardFeys,andthealreadymentionedCurry,<br />

Kleene,andBernaysheldKetonen’sworkinhighregard.Curryreportedly<br />

(vonPlato,2004)heldKetonen’sworktobethebestthinginprooftheory<br />

sinceGentzen,andthepresentwriterhasseenaletterfromCurrytoKetonenwheretheformerasksforeverythingKetonenhaswrittenonlogic,<br />

eveninFinnish. ArendHeytingwroteareviewofthethesisin1947,but<br />

apparentlyfailedtoseeitsmainpointandappearsinsteadtoviewitas<br />

aworkongeometryratherthanonprooftheory. Thefirstinternational<br />

referencetoKetonen’sworkseemstobebyKarlPopperin1947(Popper,<br />

1947),andBethusespartsofKetonen’scalculusinhistableaumethod 8<br />

butcitesKleeneandGentzen,butnotKetonen.<br />

7 See(vonPlato,2007a).<br />

8 Seeforexample(Beth,1962).<br />

R ⊃


Ketonen’sDiscovery 295<br />

NomoreoriginalworkonlogicbyKetonenappearedafterthethesis,<br />

andexactlywhythisissoisnotcompletelyclear. Heisknowntohave<br />

beenworkingonforcinginsettheoryandevenrelativitytheory(inspired<br />

bypreviousworkonthesubjectbyKaila)butdidnotpublishanyown<br />

resultsevenifsurvivedcorrespondencesuggeststhatheindeedhadworked<br />

outsomeresultsofhisown. Hehasalsoworkedontheinterpretationof<br />

consistencyproofs,many-valuedlogics,andtheapplicationofsomeofthe<br />

resultsofthethesisonepistemology(vonWright,1951).Apossiblereason<br />

astowhyhedidnotcontinuewithlogiccouldbetheseveredisappointment<br />

heexperiencedwithphilosophyofscienceingeneralduringhisvisittothe<br />

UnitedStatesinthe1950’s(Ketonen&vonWright,1950)and,asishinted<br />

atintheautobiography,theeffectsthatthesecondWorldWarbrought<br />

withitwhichpossiblysteeredalsohisphilosophicalinterestsawayfrom<br />

theworldofmathematicstowardsbroaderphilosophicalenquiries. Only<br />

oneworkonlogicafterthedissertationhasbeenfoundasaveryrough<br />

manuscriptofabouttenpages,writtenonatypewriterbutwithseveral<br />

hand-writtencorrections,andcontainingsomenotesonepistemologyand<br />

geometry,butisnothinglikesuchapolishedversionmentionedbyvon<br />

Wright(Ketonen,1944–1950).Weknowforcertainhowever,fromsurvived<br />

correspondence,thatatleaststillinthelate1960’sKetonentriedtostay<br />

up-to-datewithrecentlogicalresearch.Healsogavelecturesinbasiclogic<br />

forstudentsattheuniversity. Whenhewasaskedinhislateryearswhy<br />

hehadabandonedlogic,Ketonenalwaysremarkedabruptly“logicgivesme<br />

suchheadache”.Onecouldperhapsspeculatethatlogicbecamesomething<br />

ofaspare-timeactivity,whilethemainattentionwasonuniversitypolitics,<br />

hisprofessorshipthatheheldforover25years,between1951–1977,andon<br />

aphilosophyincorporatingelementswhichfalloutsidethoseofthenatural<br />

sciences.<br />

MichaelvonBoguslawski<br />

Departmentofphilosophy,UniversityofHelsinki<br />

Siltavuorenpenger20A,P.O.Box9,00014Helsinki,Finland<br />

michael.vonboguslawski@helsinki.fi<br />

References<br />

Bernays, P. (1945). Review: Oiva Ketonen, Untersuchungen zum<br />

Prädikatenkalkül.TheJournalofSymbolicLogic,10(4),127–130.<br />

Beth,E.(1962).Formalmethods.Dordrecht:D.Reidel.<br />

Gentzen,G.(n.d.).UntersuchungenüberdaslogischeSchliessen.InM.E.Szabo<br />

(Ed.),ThecollectedpapersofGerhardGentzen.<br />

Ketonen,O. (1935–1936). Lahjomatontilintekijä(Theunbribableaccountant).<br />

(StudyDiary.)


296 MichaelvonBoguslawski<br />

Ketonen,O.(1937).Tutkimuksiaformaalisentodistamisenristiriidattomuudesta<br />

(Investigationsintotheconsistencyofformalproving). (ManuscriptofM.A.<br />

thesis.)<br />

Ketonen,O.(1938).Todistusteorianperusaatteet.Ajatus,IX,28–108.<br />

Ketonen,O.(1941).Predikaattilogiikantäydellisyydestä.Ajatus,X,77–92.<br />

Ketonen,O.(1944).UntersuchungenzumPrädikatenkalkül.AnnalesAcad.Sci.<br />

Fenn.,23.<br />

Ketonen,O. (1944–1950). Tietommeapriorisistaaineksista. (Manuscriptin<br />

NationalArchiveofFinland.)<br />

Ketonen,O.(2000).Unfinishedautobiography.<br />

Ketonen,O.,&vonWright,G. (1950). CorrespondencebetweenOivaKetonen<br />

andGeorgHenrikvonWright. (StoredbothinthenationallibraryinHelsinki,<br />

andinthenationalarchiveofFinland.)<br />

Kleene,S. (1952). PermutabilityofinferencesinGentzen’scalculiLKandLJ.<br />

MemoiresoftheAmericanMathematicalSociety(10),1–26.<br />

Menzler-Trott, E. (2007). Logic’slostgenius: thelifeofGerhardGentzen<br />

(Vol.33).Providence,RI:AmericanMathematicalSociety.<br />

Nevanlinna,R.(1938).LetterstoOivaKetonen.(KeptbyTimoKetonen.)<br />

Popper,K.(1947).Newfoundationsforlogic.Mind,56,193–235.<br />

vonPlato,J.(2004).Einleben,einWerk.Gedankenüberdaswissenschaftliche<br />

SchaffendesfinnischenLogikersOivaKetonen.InForm,Zahl,Ordnung—StudienzuWissenschafts-undTechnikgeschichte(pp.427–435).<br />

Stuttgart: Franz<br />

SteinerVerlag.<br />

vonPlato,J.(2007a).Gentzen’slogic.In(Vol.33).Providence,RI:American<br />

MathematicalSociety.<br />

vonPlato,J.(2007b).IntheshadowsoftheLöwenheim—Skolemtheorem:Early<br />

combinatorialanalysesofmathematicalproofs. TheBulletinofSymbolicLogic,<br />

13(2),189–225.<br />

vonWright,G.(1951).ExpertopiniononOivaKetonen’sapplicationforprofessorshipintheoreticalphilosophy.<br />

(Keptinthecentralarchivesoftheuniversity<br />

ofHelsinki.)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!