thatthesemanticsoffutureinvolvesamodalcomponentegseeBertinetto1979En19 - PDF document

1 thatthesemanticsoffutureinvolvesamodalco
thatthesemanticsoffutureinvolvesamodalcomponent(e.g.seeBertinetto,1979;Enç1996,Copley,2002;Squartini,2004;Kaufmann2005;Mari,2009,2010,toappear,Klecha2014,Gi-annakidou2012,GiannakidouandMari2012a,b,BroekhuisandVerkuyl2014).EvenapurelytemporalanalysissuchasKissine2008positsepistemicmodalitywiththefuture.Futurewords,oftenandinmanylanguages,exhibitpurelymodalreadings.ConsiderasaninitialexampletheEnglishmodalverbwill.(1)a.TheFrenchwillbeonholidaythisweek.b.AsfarasIknow,oilwilloatonthewater.(Haegeman1993)Thesentencesheremakenoreferencetothefuture,butseemtoconveyepistemicmodality,e.g.,givenwhatIknow,theFrenchareonholidaythisweek.Modalusesforwillarecommon(seePalmer1987,Tsangalidis1980),andlikewiseforfutureexpressionsinotherEuropeanlanguage—e.g.Dutch(BroekhuisandVerkuyl2014),andinourownworkonGreekandItalianthatwerelyonhere.Weusethesecrosslinguisticobservationsasthestartingpointforourpresentpaper.BroekhuisandVerkuylclaimthattheDutchfutureverbzullenisanepistemicmodalopera-torexpressingthattheprejacentpropositionistheresultofreasoningbasedoninformationthatjudgedas`reliableandwell-founded'.Theygoontosaythatwhenusingzullen`thespeakersfeelsufcientlycondenttosaypistrueatn[now]oristobemadetrueati[laterthannow].Thiscondencereliesoninformationjudgedasreliableandwell-founded.Itmaytakeall

2 sortsofformdependentonthesituation:asahy
sortsofformdependentonthesituation:asahypothesis,acondentexpectation,areassurance,etc.Whatthesecircumscriptionshaveincommonisthatthespeakerhasentrancetosufcientlymanyworldstobeabletopickouttheonesthatseemconvincing.'(BroekhuisandVerkuylibid.:Conclusions).ThispassagerendersDutchzullenakintoapurelyepistemicmodallikemust,asnotedinGiannakidou2014andillustratedbelow:(2)Context:HeinIzalcan't(wel)seeinHein.de/opseezijn.`Hemustbeatsea(swimming/onaboat).'(3)Context:Iknowforsure:#Heinzalinde/opseezijn.#Heinmustbeatsea.#HeinistwohlaufZee.Weseeherethatzullenisbeingusedepistemically,asanequivalenttomust.AndthoughtheinferencethatHeinisatseaisdrawnbasedon`reliableandwell-founded'informationasBroekhuisandVerkuylputit,ifthespeakeractuallyknowsthatHeinisatthesea,shecannotusezullenormust.Wewillcallthisuse`epistemicfuture',andtheseareitspuzzles:likemust,epistemicfutureseemstoconveyastrongstatement,acertaintythatthepropositionitappliestoholds;butatthesametime,andagainlikemust,itisincompatiblewithfullknowledgeoftheproposition,thusmakingaweakerstatementthanthesentencewithoutit.Noticethatzullencanbeaccompaniedbythemodalparticlewel,acognateofGermanwohl,alsoknowntohaveepistemicuseslikethis,i.e.combiningcertaintyandweakness(seeZimmermann2011,andGiannakidou2014whoisthersttoobservethesimilarityoffutureandm

3 odalparticles).Thesentenceswithwiththepa
odalparticles).Thesentenceswithwiththeparticlesareepistemicallyweakerthanthesentenceswithoutthem.Zullenreceivespurelyepistemicreadingsalsowithpasttense:(4)A:Heissogrumpy.Hijzalwelslechtgeslapenhebben!`Hemusthavesleptreallybad!'2 future.Thereadingwegetis,toallintendsandpurposes,equivalenttoepistemicmust,anditisfullyproductive.ThepredictivefutureofFUTisillustratedbelow:(10)a.GianniJohnarriveràarrive-FUT.3sgdomani.tomcorrow.`Johnwill/#mustarrivetomorrow.'b.OTheJanisJohnthaFUTftasiarrive.nonpast.perf.3sgavrio.tomorrow.`Johnwill/#arriveat5pm/tomorrow.'Inthisreadingwehavetheexpectedreadingofprediction,whichwewillnotdiscussasithasbeendiscussedindetailelsewhere(GiannakidouandMari,2013,2014).Regardingtheepistemicreading,aswesaidearlier,thespeakerhasastrongattitudetowardstheproposition,i.e.shejudgesptobelikelytobetrue,butatthesametime,shedoesnotknowthatpisactuallytrue.ThisiswhyshechoosestoaddtheFUT.ThisbecomesparticularlyobviousifwecomparetheFUTsentencewithanunmodalizedone,inbothGreek(asentences)andItalian(bsentences):(11)a.IAriadneine/itanarosti,#aladhenimekeendelossigouri.b.Giacomoèmalato,#manonsonosicura.`Ariadne/Giacomois/wassick,#butIamnotentirelysure.'(12)a.IAriadnethainearosti,aladhenimekeendelossigouri.b.Giacomosaràmalato,manonsonosicura.`Ariadnemustbesick,butIamnotentirelysure.'Intheunmo

4 dalizedpresentandpastsentences,thespeake
dalizedpresentandpastsentences,thespeakerisfullycommittedtothetruthofthesentencesbyassertingthem.ByassertingthatAriadneandGiacomoweresick,thespeakerknowsthatAriadneandGiacomoweresick,henceacontinuationthatquestionsthisknowledgeisimpossiblebecauseitwouldattributetothespeakerinconsistentknowledge.Theepistemicfuture,ontheotherhand,isnewithIamnotentirelysure.Thesentencesindicateindeedanepistemicstatewherethespeakerconsidersapropositionveryplausiblytobetruewhilenotbe-ingfullycertainaboutit.WedonotgetaMoore-paradox,unlikewiththepositiveunmodalizedassertionwherethespeakerhasnodoubtthatpistrueintheactualworld.Wecanimpressionisticallydescribethisdifferencebysayingthatthefuture/mustsentencesare`weaker'thantheunmodalizedassertion(asiscommonsinceKartunnen1971).Below,wegiveexamplesinGreekandItalianwiththeequivalentsofmustillustratingthesamething(inGreekprepitakesasubjunctivena-complement,likeallmodalverbs;Giannakidou2009):(13)a.ItheAriadneAriadneprepimustnasubjtroieat.non-past3sgtora,now.alladenimekeendelossigouri.b.GiacomoGiacomodevemuststarbemangiando.eat-gerund.'Giacomo/Ariadnemust/willbeeatingnow,butIamnotentirelysure.'(14)ItheAriadneAriadneprepimustnasubjmilisetalk.past.3sgxthes,yesterdayalladenimekeendelossigouri.`Ariadnemusthavespokenyesterday,butIamnotentirelysure.'4 2.1VeridicalityandNonve

5 ridicalityMontagueintroducesthetermverid
ridicalityMontagueintroducesthetermveridicalitytocharacterizesentenceswithdirectperceptionverbssuchassee.Zwarts1995,andGiannakidou(1997,1998,1999)dene(non)veridicalityasrelatingtotruth(seeGiannakidou2013aforarelationbetweenthetwodenitions).Veridicalityisunderstoodobjectivelyastruthintheactualworld:asentenceisveridicalifthepropositionitdenotesistrueintheactualworld(Zwarts1995,Giannakidou1997,Egré2008),andnon-veridicalotherwise.Authorshavealsousedotherlabels,e.g.factivity,factualitytorefertoveridicality(Kartunnen1971,KartunnenandZaenen2005,KiparksyandKiparsky1970),aswellasactuality(Bhatt2006,Hacquard2010).FollowingZwarts(1995)andGiannakidou(1997,1998),wedeneveridicalityobjectivelyasapropertyofpropositionalfunctionssuchthatifanexpressionentailsthetruthofitspropo-sitionitisveridical:(18)Def1.Objectiveveridicality.AfunctionFisobjectivelyveridicaliffFpentailsp;otherwiseFisobjectivelynon-veridical.Functionsthathaveveridicalityandnonveridicality,ascanbeseen,arepropositionalfunctions(butseeBernardi2001fortype-exibledenitions).AveridicalornonveridicalfunctionFtakesthedenotationofasentence,i.e.apropositionp,asanargumentandcreatesaveridicalornonveridicalproposition.Sentencesthatdenoteveridicalandnonveridicalpropositionscanthenalsobecalledveridicalandnonveridical:(19)Def.2.Objectiv

6 e(non)veridicalityofsentences.a.Asentenc
e(non)veridicalityofsentences.a.AsentencethatdenotesapropositionoftheformFp,whereFisobjectivelyveridical,isanobjectivelyveridicalsentence.b.AsentencethatdenotesapropositionoftheformFp,whereFisobjectivelynonveridical,isanobjectivelynonveridicalsentence.Veridicalityobjectivelyisequivalenttothetraditionalrealis:averidicalsentenceistrueintheactualworld,i.e.itreferstoafact.Anysentencethatdoesnotrefertoafactisnonveridi-cal.Unmodalized,non-negated,sentencesinthesimplepastorpresent(whichisthepresentprogressiveinEnglish)areobjectivelyveridical:(20)Nicholasbroughtdessert.(21)Nicholasiswashingthedishes.Thesentenceshererefertoeventsthathappenedinthepastorarehappeningrightnow;inthepresent,thespeakermayevenbewitnessingtheeventunfolding.PASTisanobjectivelyveridi-calfunction.PASTpentailsp.ThesameforPRESp.Veridicalsentencesarethereforeactuallytrue—andtheactualityentailmentsdiscussedintheliterature(Bhatt2006,Hacquard2010)areveridicalityentailments.ThefuturesentenceNicholaswillwashthedishes,ontheotherhand,isobjectivelynonveridical,i.e.itdoesnotentailactualtruth.Allprospectivedomains(future,subjunctive,optative,bouleticanddeonticdomains),andtraditionalirrealisdomains,lackveridicality(seee.g.Condoravdi,2002;Copley,2002;Kaufmann,2005;Giannakidou1998,1999,2014).Modalstatementsasaclassarenonveridical(Giannak

7 idou1997,1998,2013),andlikewiseBeaverand
idou1997,1998,2013),andlikewiseBeaverandFrazee2011,creditingGiannakidou,presentnonveridicalityasadeningpropertyofthecategorymodality.Consider:6 PASTorPRESimposehomogenousepistemicstateswhichareincludedinp.(26)a.Johnwontherace.b.[[Johnwontherace]]M(speaker)=1iff8w[w2M(speaker)!w2fw0jJohnwontheraceinw0g]IfthespeakerassertsJohnwontherace,shemustbelieveorknowthatJohnwontherace,henceallworldsinM(speaker)areJohn-won-theraceworlds:M(speaker)p.Thepast(andpresent)sentenceisthereforeequivalenttothespeakerknowsthatp.Thisisalsousefulwhenwethinkofevidentialcontrastsine.g.languagesthathaveso-called"indirect"evidentials,andwhichformminimalpairswithsimplepastorpresent.Thesimplepastorpresentissaidtodrawson"direct"evidenceinthesensethatitconveysthemorereliable,undisputedknowledge(seeGiannakidouandMari2014formorediscussion).Subjectivenonveridicality,ontheotherhand,comeswithepistemicstatesthatonlyintersectwithp,andthereforecontain:pworlds:(27)Def.5.SubjectivenonveridicalityAfunctionFthattakesapropositionpasitsargumentissubjectivelynonveridicalwithrespecttoanepistemicstateM(i)iffM(i)�p6=;.From(27),itfollowsthat9w02M(i)::p(w0).Hence,asubjectivelynonveridicalfunctionimposesnon-homogeneityontheepistemicstate,sincethereisatleastonenonpworld.Again,asubjectively(non)veridicalfunctionFcreatesasubjectively

8 (non)veridicalproposition,whichcharacter
(non)veridicalproposition,whichcharacterizesa(non)veridicalsentence:(28)Def.6.Subjective(non)veridicalityofsentencesa.AsentencethatdenotesapropositionoftheformFp,whereFissubjectivelyveridical,isasubjectivelyveridicalsentence.b.AsentencethatdenotesapropositionoftheformFp,whereFissubjectivelynonveridical,isasubjectivelynonveridicalsentence.ModalsandtheFUTareobjectivelynonveridical,asmentionedearlier,butalsosubjectively:themodalbases(whicharesubsetsofM(speaker)interactwithpbutarenotincludedinit,andM(speaker)isalsonotincludedinp.Fromtheaboveitbecomesclearthatsubjectiveveridicalitycanbeextendedtocharacterizetheepistemicstatesthemselves.Averidicalepistemicstateisanon-partitioned,homogenousepistemicstate.Anonveridicalepistemicstate,ontheotherhand,isaspacepartitionedintopand:pworlds.(29)Def.7Veridical,nonveridicalepistemicstatesandcommitmenta.Anepistemicstate(asetofworlds)M(i)relativetoanindividualanchoriisveridicalwithrespecttoapropositionpiffallworldsinM(i)arep-worlds.(fullcommitment).b.IfthereisatleastoneworldinM(i)thatisa:pworld,thenM(i)isnonveridical(weakenedcommitment,uncertainty).c.IfallworldsinM(i)are:pworlds,thenM(i)isantiveridical(counter-commitment).Averidicalepistemicstateisanon-partitioned,homogenousepistemicstate,astateoffullcom-mitment.Aknowledgestateisveridical;aswesaid,umodalizeds

9 entencesinthepastrevealveridicalstates.A
entencesinthepastrevealveridicalstates.AnonveridicalstateM(i),ontheotherhand,isdenedasonethatcontainsat8 (35)Def.13.Projectedtruth.pisprojectivelytrueiffpistrueinalltheworldsofthesupportsetW:Wp.Inotherwords,anonveridicalmodalspaceMsupportsapropositionifthethereisasupportsetWforthepropositioninM.SincethesupportsetisthesetofBestworlds,thisstructurerevealsabiastowardsBestworlds.Modalsthatcomewithsupportsets,suchasMUSTandFUTarenonveridical,likeallmodals,butarebiased:(36)Def.14.Biasedmodals.AmodaloperatorFisbiasedifitsmodalbaseMcontainsasupportsetWM:Wp.ThesupportsetistheinnerdomainofthemodalF,andthemodalbaseisitsouterdomain.Inthisstructure,itbecomesclearthatstrengthdoesnotmeanthatthemodalentailsactualtruth(veridicality).Rather,thebiasedmodalentailspintheinnerdomain,i.e.inthesupportset,whilepremainsunsettledinthemodalbasewhichisnonveridicalandallowsnon�pworlds.Bothbiasedandunbiasedmodalsareweakerthanunmodalizedassertionsbecausetheyarenonveridical;buttherearetwokindsofnonveridicalepistemicmodalspaces:thosethatcontainasupportsetforaproposition(thebiasedmodals),andthepossibilitymodalsthatconveynonveridicalequilibriumbetweenpand:p(Giannakidou2013):(37)Def15.Nonveridicalequilibrium(Giannakidou2013b).AnepistemicstateMisinnonveridicalequilibriumiffMispartitionedintopand:p,andthere

10 arenoBestworlds.Anonveridicalstatewitheq
arenoBestworlds.Anonveridicalstatewithequilibriumrevealsnopreferencebecausethereisnoordering.TakeforexampleItmightraintomorrow.Thisisamerepossibilitystatement,andthereisnoor-deringthatcouldcreateasupportsetfortheproposition`itrainstomorrow'.Orderingsourcesaddinformationrestrictingsetsofpossibilitiesandcreatingsupportsets,thusprivilegingonesubsetofthemodalbaseoveritscomplement(:p).Intermsofcommitment,weproposethefollowingscaleofcommitmentstrength,fromstrongesttoweakest:(38)CommitmentstrengthMorecommittedunmodalizedp,MUSTp,POSSIBLYp&#x-250;lesscommittedThemodalspaceMofanunmodalizedsentencecontainsonlypworlds.WhenallworldsinMarepworlds,wehaveveridicality,andthisconveysthestrongestcommitment.WithabiasedmodallikeMUST,wehaveanonveridicalspacewithasetofbestworldswherepistrue(Giannakidou2013b,GiannakidouandMari2013).Inthiscase,thepworldsarethesupportsetofp,butthemodalbaseandM(speaker)stillallownon�pworlds.Thepossibilitysentence,ontheotherhand,conveysequilibriumbetweenpandnon-p(Giannakidou2013,GiannakidouandMari2014,thisvolume),i.e.thereisnopreferencetowardsthepornon-pworlds,nobestolds,nosupportofp.Thisissobecausethereisnoorderingwiththepossibilitymodal.Wheneverthereisorderingtherearebestworlds,theuniversalmodalwillthereforegiverisetostrongercommitmentthanthepossibilitymodal.Inotherworlds,theuni

11 versalmodalisstronginthesenseofpartially
versalmodalisstronginthesenseofpartiallysupportingpinthebestworlds,butitisstillweakwithrespecttothenonmodalveridicalsentencewhichconveysfullcommitment.(39)Commitmentweakeningofepistemicmodalsi.Allepistemicmodalsarenonveridical:theyconveyweakerspeakercommitment.ii.Biasedepistemicmodalsconveystrongercommitmentthanpossibilitymodalsbe-causetheycontainbestworldsthatsupportp.10 wesee,andwetakethistoshowthattheyconveyun-partitioned,veridical,epistemicstatesthatareincludedinp.Wecomebacktothesesentencesinsection4.Intheliterature,mustisknowntobe`weaker'thantheunmodalizedassertion,andtheideathatMUSTisweakgoesbacktoKartunnen1971(seealsodiscussioninvonFintelandGillies2010andreferencestherein,aswellasGiannakidou1997,GiannakidouandMari,2014).Be-low,wegiveexamples,inGreekandItalianwiththeverbequivalentsofMUST,andnotethattheypatternwithFUT,andcontrastwithunmodalizedassertions:(45)a.ItheAriadneAriadneprepimustnasubjtroieat.non-past3sgtoranowalabutdhennotimebe.1sgkeandendelosabsolutelysigouri.sureb.GiacomoGiacomodevemuststarbemangiando,eat-gerund,mabutnonnotsonoamtotalmentetotallysicura.certain.`Giacomo/Ariadnemust/willbeeatingnow,butIamnotentirelysure.'(46)a.ItheAriadneAriadneprepimustnasubjmilisetalk.past.3sgxthes,yesterday,alabutdhennotimebe.1sgkeandendelosabsolutelysigouri.sure`Ariadnemusthavespokenyeste

12 rday,butIamnotentirelysure.'b.GianniGian
rday,butIamnotentirelysure.'b.GianniGiannidevemustaverhaveparlatospokenieri.yesterday,manonsonototalmentesicura.`Giannimusthavespokenyesterday,butIamnotentirelysure.'ThesentenceswithMUSTandthesentenceswithepistemicFUTareequivalentinthespeaker'sintuitions.FUTandMUSTcanactuallycombine—aninstanceofmodalconcord.Thereadingremainsthesame:(47)ItheAriadneAriadnethafutprepimustnasubjmilisetalk.past.3sgxthes,yesterday,alabutdhennotimebe.1sgkeandendelosabsolutelysigouri.sure.`AriadnemusthavespokenyesterdaybutIamnotentirelysure.'(48)GiacomoGiacomodovràmust-FUT.3sgaverhaveparlatospokenieri.yesterday.`Giacomomusthavespokenyesterday,butIamnotentirelysure'.Giventheepistemicnon-predictiveusageofFUTandtheparallelwithMUST,itbecomesveryappealingtoarguethatwiththeepistemicfuturethereadingcanbederivedparalleltomust.InGiannakidouandMari(2014b)weshowthattheanalysisoftheepistemicreadingofMUSTrunsparalleltotheanalysisofFUTinthepredictivereading,differingonlyinthemodalbase.Inthepresentpaper,wefocusstrictlyontheepistemicreading.Intheepistemicuse,FUTassociateswithanepistemic,notmetaphysical,modalbase.Specically,themodalbaseisthesetofpropositionsknownbythespeaker(w0istheac-tualworld):\fepistemic(w0)=w0:w0iscompatiblewithwhatisknownbyi(thespeaker)inw0.Notethat\fepistemic(w0)M(speaker).(Ourepistemicmo

13 dalityisthussubjective,seetheobjectivevs
dalityisthussubjective,seetheobjectivevs.subjectivedistinctionofPapafragou2006.Infact,giventhatwerelativizewithrespecttoindividualanchors,therecanbenoobjectivemodality,strictlyspeaking,inoursystem).Givenwhatthespeakerknows,themodalbasecontainspworlds,butalso:pworlds;12 Howabouttheactualworld?Bestworldsarethoseinwhichstrangethingsdonothappen.Typicallytheactualworldtendstobenon-extraordinary(Portner,2009),butwealsoknowthatstrangethingshappen.Asaconsequence,wedonotclaimthattheepistemicagentactuallyknowsthattheactualworldbelongstothesetofbestworlds.Giventhattheaccessibilityrelationisepistemicandthereforereexive,itisensuredthattheactualworldisinthemodalbase(seeMatthewsonetal.2007;Portner,2009),butitisnotguaranteedthattheactualworldbelongstothepworlds.WithuniversalquanticationoverthesetofBestworlds,however,truthisprojectedwithinthesupportset,andthereforebiasisrevealedtowardsthepworlds,asisthecasealsothepredictivereading.BiasisresponsibleforthesenseofstrengththatcomeswithFUTandMUST,butitmustnotbeconfusedwithveridicality,whichexpressesfullcommitmenttopandthereforedoesnotallowforthepossibilityofnotpwithinM.53Theevidentialcomponent:perception,missingpremises,andpartialknowledgeInhisseminalwork,Kartunnen(1972)heldthattheweaknessofMUSTisintimatelyrelatedtotheweaknessofthesourceofinformation.Th

14 eviewthatweholdhereisthatepistemicweaken
eviewthatweholdhereisthatepistemicweakeningmakesthebiasedmodalstatementcompatibleonlywithapartitioned,non-veridicalepistemicstateconsistingofasubspaceofbestworldsthatsupportsp(theinnerdomain),andasubspacethatdoesn't(theouterdomain,themodalbase).AnimportantdifferencebetweenourviewandKartunnen'sisthattheepistemicweakeningisnotduetothefactthatknowledgeisindirect,buttothefactthatknowledgeispartial.Kartunnenties`weakness'toindirectevidence:whenthespeakerhasindirectevidencethattheprejacentistrue,sheusesthemodaltosignalthatsheisuncertainaboutthetruthoftheprejacent.VonFintelandGillies(2010:361)challengethisposition:`Ourpointissimple:weaknessandindirectnessarenottwosidesofasinglecoinatall.Theyarejustdifferent'.Theirclaimisthattheepistemicmodalmustpresupposesindirectevidence,butitis`strong'.Inourviewofepistemicweakening,theindirectnessisreducedtoamereside-effect,notarealphenomenon.Thekeyispartialknowledge:epistemicweaknessarisesbecausethespeakerisreasoningwithpartialknowledge,andsheknowsthatshedoesnothaveallthefacts(seealsoMari,2010).Recallthepartitionthatliesattheheartofthetruthconditionofuniversal,biasedmodals:thepartitionbetweenbestandnon-bestworlds.Whenthespeakerreasonswithauniversalmodal,sheisawarethatshedoesnothavealltheknowledgesheneedstodrawavalidcon-clusioninallworldsinthemodalbase.Whensheh

15 ascompleteknowledge,shecannotuseuniversa
ascompleteknowledge,shecannotuseuniversalmodal,asevidencedindirectvisualperceptioncontexts:(55)Context:Directvisualperceptionofraina.#Itmustberaining.b.#ThaFUTvrexi.rain. 5Inarecentwork,Mari(2014)arguesthatsomeothermodalsareveridicalwithrespectarestricteddomainthatcontainstheactualworld,andnonveridicalwithrespecttoanouterdomain.This,sheargues,iswhathappenswithpastmodalswithpresentorientation,triggeringtheveridicality(a.k.a.actualityentailments).WhatiscommontoMUST/FUTandthesemodals,isthatthemodaldomainisarticulatedintotwosubdomains,withtheinneronebeingveridicalandtheouterone,beingnonveridical.14 premise:ifsoundthereforerain).Onlyinthebestworldsisthesoundofrainingduetorain.Auditoryperceptionthereforeprovidesonlyincompleteknowledge,andthemodalisallowed.Inotherwords,theapparentevidentialeffectofuniversalepistemicmodalsisineffectduetothefactthattheyareindicatorsofreasoningwithnonveridicalityandincompleteknowledge.Wesummarizethisinthefollowing:(58)EvidentialcomponentofUniversalEpistemicModals(UEMs):partialknowledgea.UEMscanonlyeffectivelyweakenapropositionp,ifthespeaker'sknowledgethatsupportspisnotcomplete.b.Completeknowledgeisknowledgeofalltherelevantfactsforp.Moretechnically,itisasetofpropositionsthatentailsp.c.Allotherknowledgeispartial.ThegeneralizationsthatweestablishhereforUEMsareveryrelev

16 antforthediscussionofevidentiality,espec
antforthediscussionofevidentiality,especiallyinlanguagesthathaveindirectevidentialmorphemesbutdonotmarkdirectperception(e.g.NativeAmericanlanguagessuchasCheyenne,Murraytoappear,andTurkish,Bulgarian,Smirnova2013a).The`direct'evidentialistypicallyanunmarkedpastorpresent,andthemarkedformistheso-calledindirectevidential,whichindicatesthatthesourceofinformationisnotrsthandknowledgeofthespeaker.Theindirectevidential—liketheUEMinEnglish,GreekandItalian—isanonveridicalmarkerthatmarksthereducedspeakercommitmenttop.Directperception,ontheotherhand,aswediscussedinthecaseofrain,entailsfullknowledge.IfIseetherain,Iknowitisraining.Thisisthecaseofveridicalcommitment,offullcompleteknowledge.Directperception,naturallygivesthatkindofknowledge6Givenpartialknowledge,pisentailedonlyinthesupportset(thebestworlds,orvonFintelandGillies'skernel).Otherwise,inthemodalbasedpremainsunsettled.Currentclassicationofsourcesofevidencefocusonthedistinctionbetweendirectandindirectknowledge,withvisualevidencecountingasdirectevidenceentailingfullknowledge,andreprobativeandotherinternalevidencecountingasindirect,thusimplyingincompleteknowledge.Ourclaimisthateverytypeofsourceofinformationcomesaseithercompleteorpartial.Completeknowledgeisasetofpropositionsthatentailpandpartialknowledgeisasetofpropositionthatisonl

17 ycompatiblewithp(seealsoMari,2010).Greek
ycompatiblewithp(seealsoMari,2010).GreekandItaliandohaveareportativeevidentialform(lei,ipan,sidice,dicono,pare):FUTcanco-occurwithit:(59)OtheJanisJohnthaFUTginibecome.3sgkalawell,eipan.say.3pl.(60)GianniGianniguarirà,recover-FUT.3sg,pare.itseems.`Johnmustrecover,itseems.However,thereportativecontextbyitselfisnotsufcienttotriggerUEM.WereproducehereanexamplefromSmirnova2013b,toshowthecontrastbetweentheGreek/ItalianFUT 6Note,atthesametime,thatdirectaccessinthesenseofWillett(1988)doesnotcountperseasacaseofcompleteknowledge.Thereisadifferencebetween`seeingsomefactsthatarecompatiblewithpbeingtrue'and`knowingthatpistrueinvirtueofvisualevidence'.AsLee2012explains,witnessingpisnotequivalenttohavevisualevidenceforp.Likewise,visualevidencecanbeincompleteevidenceforassessingthetruthofp(in(56),Iseetheumbrella,butIdonotseetherain).Similarly,inferentialknowledgecanbecompleteorincomplete:itiscompleteifallthepremisesallowingtoconcludethatpistruearegivenandincompleteiftherearesomemissingpremisesforconcludingthatpistrue.16 pleteknowledge.Themirativeuseistreatedasanadditionalinferencethattheevidentialtrig-gersinviewofthefactthatitcannothaveitsregularcontribution.TheUEMs,clearly,lackthisinference.Havingillustratedtherelevanceofpartialknowledge,wenowturntoanalchallenge.4Moore-likeeffectswithFUTandinfor

18 mationalconictInthislastsection,wewa
mationalconictInthislastsection,wewanttodiscussapotentialchallengeforouranalysis:FUTcangiverisetoeffectsthatappeartobeMoore-paradoxical.TheliteratureontheMooreparadoxisvast,andwewillnotattemptageneralanalysisofithere,sinceourtopicisnottheparadoxitself.OurnewobservationsarethatwendMoore-effectwithFUT,butdifferentvariantsofMoore'sparadoxaffectfuturesentencesindifferentways.Toexplainthevariation,weproposethatMoore-effectsdonotnecessarilyrevealanepistemic(veridicality)conict,butmanifestalsosensitivitytoinformationalowthatpreviouslyhasescapedattention.TheclassicalMooreparadoxitselfariseswithsentenceslikebelow:(65)#ItisrainingandIdon'tknowthatitisraining.(66)#ItisrainingandIdon'tbelieveit.Theusualreactionisthatthesentencesaboveareodd,contradictory-sounding,andunassertable.Intheliterature,thesentencesaretreatedasdefectiveinthattheyinvolvethespeakerinsomekindofepistemicconict.Inourterms:aswesaidinsection2.1,apositiveunmodalizedasser-tionissubjectivelyveridical,i.e.thespeakeristypicallyunderstoodasknowingthatpistrue.Ifthisisso,theninthesentencesabovethespeaker'sepistemicstateM(speaker)ispresentedasbothbeingincludedinpandallowing:pworlds.Thisisacontradictoryepistemicstate,andthesentencesaredefectivebecauseofthisveridicalityconictimposedbythetwoconjuncts.Yalcininamorerecentdis

19 cussion(Yalcin2007)coinsthetermepistemic
cussion(Yalcin2007)coinsthetermepistemiccontradictionsforMoorevariantswithlogicalformssuchasanditisnotpossiblethat,e.g:(67)#Itisraininganditmightnotberaining.(68)#Itisraininganditispossiblethatitisnotraining.Yalcincallstheseepistemiccontradictions.Again,theconictappearstobebetweenaveridi-calepistemicstateestablishedbytheunmodalizedrstconjunct(whereallworldsarerainingworlds),andanon-veridicalstate,allowingrainingandnon-rainingworldsinthesecondcon-junct.So,bothYalcin'sexamplesandtheclassicMooreparadoxexamplesinvolveanepistemicconictwhichrevealsasubjectiveveridicalityconict.Interestinglyfromourperspective,futureandMUSTsentencesgiverisetowhatappearstobeaMooreparadoxicaleffect.Weobserveitbelow.(Wereplaceandwithbuttomakethesentencesmorenaturalsounding,butascanbeseen,theeffectisobserved):(69)#Itmustberaining,butitmightnotberaining.(70)#ThaFUTvrehi,rain.imperf.non-past.3sgalaalaineispithanopossiblenasubjminnotvrehi.rain.imperf.non-past.3sg'#Itmustberaining,butitispossiblethatititmightnotberaining.(epistemic)'(71)GianniJohnsaràbe.3sg.futarrivato,arrived,#mabutèispossibilepossiblechethatnonnotsiabe.subj.3sgarrivato.arrived.`Johnmusthavearrived,butitispossiblethathehasnotarrived'(epistemic).18 opposedtoIamnotsurewhichisaneutralsentencewithnobiastowardsp7).Hence,thetwoconju

20 nctsmakereferencetononveridicalmodalspac
nctsmakereferencetononveridicalmodalspaceswhichareinagreementandnotinconict,sincetheyarebothnonveridicalandbiased.Wewillcallthissituationinformationalharmony.WedonotgetaMoore-paradoxsituationwithinformationalharmony,unlikewiththepositiveunmodalizedassertionwhereIamnotentirelysurethatpcreatesexactlythekindofcontradictiontheclassicalMoorecontinuationgives(acontradictoryepistemicstate).Hence,IamnotentirelysurefullysupportsournonveridicaltreatmentofepistemicFUTandMUST.Whatwejustsaidreliedonthenotionofinformationalstrength.Thetwosentenceswereofequalinformationalweight,wesaid.WhatgoeswrongintheclassicalMoorecasesandinourFUT/MUSTvariantsofthemisthatthesentencesdonothavethesameinformationalweight.Thiscreatesinformationalconictthatmanifestsitselfintwoways:(a)asbreakdownofinformationow,whichnormallyproceedsfromweakertostronger(aswedeneitbelow),and(b)asan"informationalcontradiction".TheMooreeffectswithFUTandMUSTareduetothese.Considerrsthowtheclassicalcaseillustratesbreakdownofinformationow:(76)#Itisrainingand/butitmightnotberaining.TherstconjunctItisrainingpresentstherainasanactualfact(objectiveveridicality).Thestatementisalsosubjectivelyveridical,i.e.inallworldscompatiblewiththespeaker'sknowl-edgeitisraining.Thesecondconjunctconveysaweakerinformation,i.e.thatthespeakercon

21 sidersitpossiblethatitisnotraining.Thisd
sidersitpossiblethatitisnotraining.Thisdiscourseisodd,andthehearermustconcludethatthespeakerisnotbeingco-operative.Shesaidsomethingfalseeitherintherstorinthesecondconjunct,inbothcasesviolatingquality,thusbeingmisleading.Weendupwithaconict,aswellasabreakdownofwhatcanbethoughtofasnormalinformationow.Proceedingfromweakertostrongeristhenormalcourseofinformationow,expectedbyGriceanpragmatics:(77)NormalcyconditionsoninformationowInformationowisconsiderednormaliff:(i)InformationgoesfromweakerinformationAtostrongerinformationB.Or,(ii)AandBdonotinformationallycontradicteachother.Theseconditionsarenothingextraordinary,butmeresummaryofrun-of-the-millversionsofGriceanviewsofhowinformationnormallyproceeds.Weakerandstrongeraretheinforma-tionalalternativescompared,i.e.thepropositionsdenotedbythesentences.Theproblem,crucially,inthesecondconjunct,comesfromthefactthatastrongerinformationwasestab-lishedrst:averidicalsentenceisinformationallystrongerthananon-veridicalsentence.Andwithinnon-veridicalsentences,SwithbiasisstrongerthatSwithequilibrium.Recallthattheorderingsourcerevealsbias.Withequilibrium,thereisnoorderingsource.Withorderingsourcesthedomainofquanticationismorerestrictedandisthusinformationallyricher(àlaStalnaker).Belowwegivetherelevantscale:(78)Informationalstrengt

22 horderingrelevantforMoore'scontrastsweak
horderingrelevantforMoore'scontrastsweakerhmight:p,MUSTp,pistrongerLetusrepresentMoore'ssentencesSasapairofalternativeshS1;S2i:(79)Moore'svariant:pandmight:pAlternatives:hS1:p;S2:might:pi 7Notethatthe`Iamnotsure'continuationisodd.20 Accordingtothenormalcycondition,informationowrequiresthestrongerstatementtobesecond.TheinformationowintheMooresentenceisthusnotnormal,andthesentenceisdefectiveforthisreason.Notice,crucially,thatifwereversetheorderaswedobelow,thesequenceisimproved,andshowsnoMooreeffect:(85)a.Ineispithanopossiblenasubjminnotvreksi,rain.perf.non-past.3sg,allabutthaFUTvreksi,rain.perf.non-past.3sg,thaFUTdhis.see.2sgb.ÈIspossibilepossiblechethatnonnotpiova,rain.3sg.subj,mabutpioverà,rain.3sg.fut,vedrai.see.2sg.fut.`Itispossiblethatitwillnotrain,butitwillrain,you'llsee.'(86)a.Ineispithanopossiblenasubjminnotvrexi,rain.imperf.non-past.3sg,allabutmalonprobablythaFUTvrexi.rain.imperf.non-past.3sg.`Itispossiblethatisnotraining,butmostlikelyitmustberaining.'b.ÈIspossibilepossiblechethatnonstay.3sg.subjstianotpiovendo,rain.gerund,mabutstaràstay.3sg.futpiovendo,rain.gerund,vedrai.see.2sg.`Itispossiblethatitisnotraining,butIamprettycertainitisraining,you'llsee.'Thesediscoursesarenormalbecausethestrongersentencefollowstheweakerone.ThisfactillustratesthattheproblemwiththeMooresente

23 nceswithstrongmodalsintherstconjunct
nceswithstrongmodalsintherstconjunctfollowedbynegationsofweakermodals,atleastwhenwethinkoftheirGreekandItaliancounterparts,isnotaveridicalityconict,butbreakdownofnormalinformationow.Thetwoalternativesarenotepistemicallyinconsistentbutinformationallynot-normal.Consider,nally,thecontinuationswithIdon'tbelieveit:(87)GianniJohnsaràbe.3sg.futarrivato,arrived,#mabutnonnotlothatcredo.believe.1sg.`#Johnmusthavearrived,butIdonotbelieveit.'Here,thealternativesare:(88)Moore'svariant:MUSTpandIdonotbelievethatpAlternatives:hS1:MUSTp,S2:IdonotbelievethatpiThespeakerestablishesbiastowardspwiththeuseofauniversalmodalintherstconjunct.ShecontinuesthenbysayingthatshedoesnotbelievethatJohnarrived.Importantly,believeisaneg-raisingverb,sonotbelievethatptypicallystrengthenstobelievethatnotp;(seeHorn1979foraclassicalpieceonneg-raisingwithbeliefverbs).Inthestrengthenedreading,bothconjunctsappeartobeinformationallyequallystrong,but,crucially,conictingwitheachother:therstconjunctconveysbiastowardsthepset(Johnarrived)andthesecondalternativestrengthenstothebeliefthatJohndidn'tarrive(therearenopworldsinthespeakersepistemicstate,countercommitmenttop).Thiscreatesaconictinthejoinutterancebecausethespeakerisrequiredtobothhavebiastowardspandcountercommitmenttoit.Thisisaninformational"contrad

24 iction",aconictthatcannotberepaired.
iction",aconictthatcannotberepaired.Asweseebelow,changeoforderhasnoeffect:22 Biasedmodalsthusexpressweakerepistemiccommitmentofthespeakertowardstheproposi-tionalcontent–buttheyarestillstrongerthanunbiasedpossibilitymodalsthatsimplyraisethepossibilitythatpanddonotcontainasupportsetforp.ThethirdcontributionconcernstheevidentialcomponentofMUSTandepistemicfuture.Weshowedthatcontrarytowhathasbeenclaimedintheliteratureaboutindirectness,thekeytounderstandingtheevidentialityofMUSTandepistemicfutureispartialknowledge.Partialknowledgerepresentsapartitioned,non-homogenous,thereforenon-veridicalepistemicstate,andexplainsnicelywhytheuseofuniversalepistemicmodalsisconstrainedinsuchstates.Wealsoshowedthatvisualperceptionofaneventisprivilegedbecauseitleadstocompleteknowledge,andthereforeFUTandMUST,beingnon-veridical,cannotbeusedwithdirectvisualperception.AckowledgementsWepresentedthismaterialattheworkshoponCategoriesinWroclaw,attheInternationalCongressofLinguistsinGeneva,andincolloquiaattheUniversityofBrusselsandGroningen.Wethanktheaudiencesfortheirinsightfulfeedback.WearealsoverygratefulforcommentsandhelpfuldiscussiontoKaivonFintel,JackHoeksema,MikhailKissine,JasonMerchant,andMalteWiller–aswellastheanonymousreviewersofthisvolumefortheirhelpandsuggestions.ReferencesBertinetto,P.M.1979.Alcune

25 ipotesisulnostrofuturo(conalcuneosservaz
ipotesisulnostrofuturo(conalcuneosservazionisupotereedovere),Rivistadigrammaticagenerativa4:77-138.Broekhuis,H.andVerkuyl,H.(toappear).BinaryTenseandModality.NaturalLanguageandLinguisticTheory.Ciardelli,I,Groenendijk,JandRoelofsen,F.2013.Inquistitivesemantics:anewnotionofmeaning.LanguageandLinguisticCompass,7:459-476.Chiou,Michael.2012.TheinterpretationofGreekfuture:apragmaticapproach.Ms.Condoravdi,C.2002.Temporalinterpretationofmodals:modalsforthepresentandforthepast.InTheConstructionofMeaning,edsD.Beaver,LuisD.CassillasMaritinez,BradyZ.Clark,S.Kaufmann.Condoravdi,C.2003.MoodsandModalitiesforWillandWould,InvitedcommunicationatAmsterdamColloquium.Copley,B.2002.Thesemanticsofthefuture,PhDMIT.vonFintel,K.andGillies,A.2010.Must...stay...strong!Nat.LanguageSemantics18:351-383.vonFintel,K.andHeim,I.2007.Intensionality.Ms.MIT.vonFintel,K.andIatridou,S.2006.HowtoSayOughtinForeign:TheCompositionofWeakNecessityModals.Ms.MIT.Giannakidou,A.1997.TheLandscapeofPolarityItems.PhDthesis,UniversityofGronin-gen.Amsterdam.Giannakidou,A.1998.PolaritySensitivityas(Non)veridicalDependency.JohnBenjamins.Amsterdam.Giannakidou,A.1999.Affectivedependencies.LinguisticsandPhilosophy22:367-421.Giannakidou,A.2009.Thedependencyofthesubjunctiverevisited:temporalsemanticsandpolarity.Lingua120:1883-1908.24 Kratzer,A.1991.M

26 odality.InSemantics:AnInternationalHandb
odality.InSemantics:AnInternationalHandbookofContemporaryResearch,ed.ByA.vonStechow&D.Wunderlich,639-650.Berlin:deGruyter.Lassiter,D.2011.MeasurementandModality:theScalarBasisforModalSemantics.PhDthesis,NYU.Lassiter,D.2013.Theweaknessofmust:Indefenseofamantra.CSLIWorkshopPerspec-tivesonModality.Montague,R.1969.Onthenatureofcertainphilosophicalentities.TheMonist53,159-194.deMarneffe,M.,C.Manning,andC.Potts.2012.Didithappen?Thepragmaticcomplexityoftheveridicalityjudgement.ComputationalLinguistics38:300-333.Matthweson,L.2010.Crosslinguisticvariationinmodalitysystems:theroleofmood.SemanticsandPragmatics3:1-74.Matthewson,L.Rullmann,H.,andDavis,H.2007.EvidentialsasEpistemicModals:Evi-dencefromSt'at'imcets.ThelinguisticVariationYearbook7.Mari,A.2009.DisambiguatingtheItalianFuture.InProceedingsofGenerativeLexicon,209-216.Mari,A.2010.OntheevidentialnatureoftheItalianfuture.Ms.IJNhttp://jeannicod.ccsd.cnrs.fr/docs/00/67/85/49/PDF/Evidential_Future_Italian.pdfMari,A.2013.Eachother,asymmetryandreasonablefutures.JournalofSemantics.doi:10.1093/jos/fft003.Mari,A.2014.ActualityEntailments:broadeningthespaceofthepossibilities.Ms.InstitutJeanNicodandUniversityofChicago.Mari,A.toappear.Modalitésettemps.PeterLangAG.McCready,E.andN.Ogata.2007.Evidentiality,modality,andprobability.LinguisticsandPhilosophy30:14

27 7-206.Murray,S.Toappear.EvidentialsandIl
7-206.Murray,S.Toappear.EvidentialsandIllocutionaryMoodinCheyenne.ToappearinInter-nationalJournalofAmericanLinguistics.Palmer,F.R.1987.MoodandModality.CambridgeUP.Pietrandrea,P.(2005).Epistemicmodality:functionalpropertiesandtheItaliansystem.Amsterdam:JohnBenjamin.Portner,Paul1998.TheProgressiveinModalSemantics.Language74(4):760-87Portner,P.2009.Modality.OxfordUniversityPress.Prior,A.1957.TimeandModality.Oxford:OxfordUniversityPress.Smirnova,A.2013a.ThemeaningoftheBulgarianandTurkishevidentials.ContrastiveLinguistics2013,2/3:205-223.Smirnova,A.2013b.EvidentialityinBulgarian:Temporality,epistemicmodality,andin-formationsource.JournalofSemantics2013,30(4):479-532.Squartini,M.2004.DisentanglingevidentialityandepistemicmodalityinRomance.Lin-gua,114:873-895.Squartini,M.2012.Evidentialityininteraction:TheconcessiveuseoftheItalianFuturebe-tweengrammaranddiscourse.JournalofPragamtics.doi.org/10.1016/j.pragma.2012.09.008Thomason,R.1984.CombinationofTenseandModality,inD.Gabbay;F.Guenthner(eds.),HandbookofPhilosophicalLogic:ExtensionsofClassicalLogic,Dordrecht:Reidel,135-165.Trnavac,R.andTaboada,M.2012.Thecontributionofnonveridicalrhetoricalrelationstoevaluationindiscourse.InLanguageSciences34:301-318.Tsangalidis,A.1998.Willandtha:aComparativeStudyoftheCategoryFuture.Thessa-loniki:UniversityStudioPre