TheCloudofUnknowing* LaurenceR.Horn,YaleUniversity 

Abstract: NegͲraising (NR) predicates satisfy the consistency and completeness properties of Zwarts 1986, ¬f(X)œf¬(X) (cf. Klooster 2003), strengthening formal contradictory negation to contrariety.Butwhencanthisstrengtheningoccur?Collins&Postal(inpress)argueforderiving “NEG(aVNRthatp)”grammaticallyfrom“aVNRthatNEGͲp”onthebasisofthepossibilityofsubjectͲ auxinversionunderhighernegationaswellasthelicensingofstrictNPIs,propertiestheyclaimto be restricted to classical NR governors. Closer inspection shows that no such restriction can be maintained,inlightoftheunderͲappreciatedpropertiesofnonͲfactive“know”. 

1.Introduction Asrelatedbytheusualinfallibleauthority,Wikipedia,TheCloudofUnknowing(orCloude ofUnknowyng)is“ananonymousmysticalworkwritteninMiddleEnglishinthelatterhalf of the 14th century” that “draws on the mystical tradition of PseudoͲDionysius the Areopagite and Christian Neoplatonism, which focuses on the via negativa road to discovering God as a pure entity” (http://tinyurl.com/6c295h). As a longtime traveler myselfonthevianegativa,Iampleasedtofollowthatroadintothatcloudwiththeaimof determining how exploring unknowing—or more exactly, the behavior of know in the scope of negation—might help elucidate (or at least fruitfully complicate) our understandingofnegͲraising,negativepolaritylicensing,inversion,andintervention.  2.Varietiesofcontra(dicto)rynegation A relation between negͲraising (NR), the lowerͲclause understanding of higherͲclause negation,andthelogicalpropertyofconsistency,¬f(X)f¬(X),isfirstsuggestedinZwarts 1986; cf. Zwarts 1991, Kas 1993, Tovena 2001.  NR predicates seem to allow this conditional to be perfected to a biconditional satisfying both Zwarts’s consistency and completeness properties, ¬f(X)œf¬(X). In such cases, not only does contrariety entail contradictory sentence negation (reading the biconditional right to left) but formal  * Someofthismaterialwaspresentedatothervenues,inparticularatNYUinJune2013.Iam grateful to commenters at those occasions and to Barbara Abbott, Betty Birner, Chris Collins, PierreLarrivée,PaulPostal,andGregoryWardforhelpfulcommentsanddiscussion.

179

contradictorynegationisstrengthenedtocontrariety(readinglefttoright).Butjustwhen andhowisthispossible? “The essence of negation”, according to the 19th century philosopher Bernard Bosanquet,“istoinvestthecontrarywiththecharacterofthecontradictory”(1888:306). Oneprimeinstanceofthisinvestmentpracticeis“...thehabitualuseofphrasessuchasI donotbelieveit,whichrefergrammaticallytoafactofmyintellectualstatebutactually serveasnegationsofsomethingascribedtoreality...Compareourcommonphrase‘Idon’t think that’—which is really equivalent to ‘I think that ___ not’” (Bosanquet 1888: 337). This, of course, is a protoͲdescription of the practical equivalence at the heart of negͲ raising,towhichwereturnshortly. Butfirstitisworthnotingsomeotherexamplesofthisstrengtheningtendencyatwork. InmanyinstancestotheLawofExcludedMiddle,inthatwhenadisjunctionbetweenlive contrariespvqcanbeinferred,theformalnegationofpresultsnotmerelyinasimple contradictory ¬p but in the affirmation of q.  Janet Fodor (1970: 158Ͳ168) describes the “ALLͲORͲNONE” effectresultinginthetendencyforanapparentsententialnegationwitha definitepluralorgenericbarepluraltoscopeunderthe(explicitorimplicit)quantification withinsubjectandobjectterms.Thuswhileseeingtheboysgenerallyamountstoseeing themall,thepossibilityfornegationtooutscopeauniversalasin(1a)isunavailablewith thepluraldefinitein(1b),while(2b),unlike(2a),“leavesnoroomfordisagreementsabout differentwomen.” 

(1) a.Ididn’tseealltheboys,butIdidseesomeofthem.  b.#Ididn’tseetheboys,butIdidseesomeofthem.  (2)a.Allwomenenjoy/donotenjoywashingdishes;Doallwomenenjoywashingdishes?   b.Womenenjoy/donotenjoywashingdishes;Dowomenenjoywashingdishes? 

In each case, the set in question behaves uniformly as a monolith, rendering any literal wideͲscopecontradictoryreadingofthenegationdifficultorimpossibletoget. Fodor’sallͲorͲnonewithimplicitquantificationinbareanddefinitepluralsresurfaces astheGENERICEXCLUDEDMIDDLEofvonFintel(1997:31)— When a kind is denied to have a generic property Pk, then any of its individuals cannothavethecorrespondingindividualͲlevelpropertyPi. —andthe HOMOGENEITY or UNIFORMITYPRESUPPOSITION ofLöbner(1985&seq.),basedonthe interpretation of negative responses to questions like Do mammals lay eggs or Are the childrenasleep?: IfthepredicatePisfalsefortheNP,itsnegationnotͲPistruefortheNP.

180

—laterreformulatedasthePRESUPPOSITIONOFINDIVISIBILITY: Wheneverapredicateisappliedtooneofitsarguments,itistrueorfalseofthe argumentasawhole. (Löbner2000:239) Ineachcase,themembersofasetAeitherhomogeneouslyexhibitaproperty(e.g.eggͲ laying) or homogeneously exhibit the opposite property (e.g. nonͲeggͲlaying); the possibility that there might be an aA in one camp and a bA in the opposite camp is ruled out. We have, then, an instance of the Zwarts biconditional in which both consistencyandcompletenessaresatisfied,sothat¬f(X)œf¬(X).Inthesecases,theLawof ExcludedMiddleistriggered,strengtheningapparentwideͲscopesententialnegation(‘No, it’s notthecasethatmammals [ingeneral]lay eggs’] intocontrarynegation (‘Mammals aresuchthey[don’tlayeggs]’),byvirtueofthenatureoftheimplicitlyquantifiedterms with which negation interacts, or rather fails to interact. In the case of negͲraising, to whichwenowturn,theeffectderivesfromthenatureofthepredicate. 

3.NegͲraisingandtheexcludedmiddle TheGoldenAnniversaryofthefirstanalysisofnegͲraisingwithinafragmentofgenerative grammar(Fillmore1963)hasnowbeenmarkedbyamajormonograph,Collins&Postal 2014, that marshals old and new evidence in support of a syntactic approach to negͲ raising. In motivating his original rule for the “Transposition of NOT (EVER)”—“Under certainconditions(e.g.afterverbslikeWANTorTHINKwhicharethemselvesnotnegated),a NOT in the embedded sentence may be moved in front of the main verb…”—Charles Fillmore cites no explicit evidence for a grammatically coͲderiving (3a,b), relying instead on the (purported) paraphrase relation between (3b) and one reading of the putatively ambiguous(3a)(Fillmore1963:220,fn.12).  (3) a.Idon’tbelievethathewantsmetothinkthathedidit.   b.Ibelievethathewantsmetothinkthathedidn’tdoit.  Infact,itis unclearwhether(3a,b)do constitutegoodparaphrases;as recognizedahalfͲ centuryearlierbythegrammarianPoutsma(1928:105),“theshiftingofnotoftenhasthe effectoftoningdownthenegativingofasentence.”  After some years of neglect, Fillmore’s syntactic approach to what had come to be callednegͲraising(NR),followingtheintroductionofthe NEGelementinKlima1964,began to be supported with direct evidence. The most influential argument was due by Robin Lakoff.CitinganearlierpersonalcommunicationfromMasaruKajita,Lakoff(1969)points

181

outthatstrictorstrongnegativepolarityitems(NPIs),e.g.untilmidnight,inweeks,which normally require a tautoclausal negative licenser, are nevertheless wellͲformed in the scopeofnegatedNRpredicateslikethinkorsuppose: (4)a.Ididn’tthinkJohnwouldleaveuntiltomorrow.  b.*Ididn’tsayJohnwouldleaveuntiltomorrow. The functional transparency of NR predicates that overrides the locality restriction on strict NPIs turns out to be a more complex matter than first appeared; see Horn 1978, Gajewski2007,andCollins&Postal2014fordetails,someofwhichweconsiderbelow. Over the years, empirical and theoretical considerations have gradually led linguists awayfromFillmoreandLakoffanddownthetrailblazedbyJackendoff(1971:291):“The synonymy between John thinks that Bill didn't go and onereading of John doesn’t think that Bill went is inferential in character and has nothing to do with the syntactic component—it may even have nothing to do with the semantic component.” (This analytictrendisreviewedinHorn1978andHorn1989:Chapter5.)Theresulthasbeena de facto rejection of the syntactic approach to NR and a return in essence to the traditionalviewpurveyedingrammarsofEnglishthattendtosubsumeNR(ifit’snoticed atall)underamoregeneraltendencyforformalcontradictorynegation(paraphrasableas “Itisnotthatcasethatp”)tobestrengthenedtoacontrary,aswhenShe’sunhappyor even She’s not happy are understood as making a stronger negative claim than a mere denialthatsheishappy.Thesecanbeviewedasexpressinglexicalizedandvirtual(oronͲ line) contrariety respectively. And similarly, Jespersen (1917: 53) presents the use of I don’tthinkto“reallymean”Ithinkhehasnotcomeasbothaninstanceofspecialization of negative meaning and an illustration of “the strong tendency in many languages to attract to the main verb a negative which should logically belong to the dependent nexus,”i.e.asacontrary,f¬(X),incontradictory’sclothing,¬f(X). Butjusthowandwhyisthehearerledtostrengthentheweakercontradictory,Idon’t believethatpunderstoodas¬(Ibelievethatp),totheforceofthecontrary,Ibelievethat ¬p,?TheclassicrecipeisgivenbyBartsch(1973)inapaperthatwearsitsconclusionon its sleeve: “‘Negative transportation’ gibt es nicht.” Like Jackendoff, Bartsch (1973: 3) rejectsanyambiguityfor(5),butarguesthatmirroringthesemanticentailmentfrom(5a) to(5b),therearisesincertaincontextsapragmaticimplicationfrom(5b)to(5a). (5) Peterdoesn’tbelievethatHansiscoming.    a.PeterbelievesthatHansisnotcoming.   b.ItisnotthecasethatPeterbelievesthatHansiscoming.

182

This implication derives from the assumption that the subject a can be assumed to havegivensomethoughttothetruthofthecomplementpandcometosomeconclusion about it, rather than that a hasn’t thought about p or is neutral as whether p or ¬p. Propositional attitudes (‘think’, ‘believe’, ‘want’) express the subject’s cognitive or psychological stance toward the complement, inducing a disjunctive pragmatic presuppositionoftheform“[abelievesthatp]or[abelievesthat¬p].”ThussoͲcallednegͲ raising is not a rule of grammar or semantic interpretation but a (mere) pragmatische Implikation;(5a,b)aresemanticallydistinctbutcanexpressthesameinformationrelative toagivenSprechsituation.Bartsch’sinferenceschemacanbegivenasfollows: 

(6)(i)F(a,p)vF(a,¬p)  (ii)“¬F(a,p)” (iii)F(a,¬p)

(thepragmaticallypresupposeddisjunction) (thepropositionactuallyasserted) (thepropositionconveyed)



Thekeystepistheassumeddisjunctionin(i):ifyoucannaturallyassumeIeitherwant togoorwanttostay(=notͲgo)andIsay(outofdiffidence,politeness,cowardice,etc.) thatIdon’twanttogo,youcaninferthatIwanttostay.Fromthedisjunctionporqand the expression of notͲp, infer q—this is a basic inferential schema variously known as DISJUNCTIVESYLLOGISM,MODUSTOLLENDOPONENS,REASONINGBYEXCLUSION,or(ifyou’reaStoic)THE FIFTHINDEMONSTRABLESYLLOGISM. Unfortunately,thisproposedsolutiontoNRcannothandlevariationwithinandacross languagesastojustwhichNRcandidatescansubstituteforFin(6).WhenisthemiddleͲ excludingdisjunctionin(6(i))actuallyassumed?Notjustanypredicateoftheknowledge andbeliefclasswilldo.Infact,membershipintheclassofpropositionalattitudesisnot necessary;negͲraisingwasfirstidentifiedbySaintAnselm(1033Ͳ1109),whodescribesthe strongtendencyfordebere‘ought’innondebetpeccare(i.e.‘[aman]notͲshouldsin’)to be interpreted as debere non peccare (i.e. ‘[a man] should notͲsin’] and similar readings areavailableforarangeofotherdeonticandepistemicpredicates,e.g.besupposedto, falloir‘must’,beadvisable,belikely(seeHorn1989:308ff.andreferencestherein).Noris propositionalͲattitudeͲhood a sufficient condition, given the fact that factives like know, realize, and regret and related strong epistemics like be certain or be sure fail to allow lowerͲnegorcontraryreadings. And then there is the problem of variation, both within and across languages. While German hoffen and Dutch hopen negͲraise, their English sister hope (usually) doesn’t; Latinsperare‘hope’negͲraisedbutitsFrenchderivativeespérerdoesn’t(whilesouhaiter ‘wish,hope’does).ParentheticalguessisanegͲraisingpropositionalattitudeinSouthern U.S.EnglishbutnotinotherU.S.orU.K.varieties.Andsoon:seeHorn&Bayer1984for

183

furtherdiscussionandaproposedfix.Morerecently,Gajewski(2007)takestheexistence of lexical exceptions (nonͲNR attitude predicates) as indicating that excluded middle for NR must be treated as a “soft” presupposition, while Collins & Postal (2014) see this muddle as further evidence for a grammatical rather than pragmatic account of NR readings. Klooster (2003) undertakes his own quasiͲBartschean analysis of NR by invoking the notion of dichotomy, or what he calls a ‘BLACK AND WHITE’ EFFECT, for verbs like think and want, given that “in a discourse where judgements and intentions are relevant, but reservingordeferringthemarenot,verbsoftheconsideredtypeareeasilyinterpretedas dichotomous” (Klooster 2003: 3Ͳ4). This suggests an analysis along the lines of the Zwartsiancompleteness+consistencybiconditional— Inasentencecontaininga matrixverbofthe[dichotomous]type in question,its contrarycanthus(indirectly)beexpressedsimplybyintroducingnegation.Thatis, wherePisanNRverb,xthesubject,andpthecomplementclause,thefollowing seemstohold: [7]¬P(x,p)iffP(x,¬p)





—but there’s many a slip between what seems to hold and what does hold. Klooster ultimatelyrejectshistentativelyadvancedbiconditionalfortheverbsatissue,giventhe nonͲsynonymyofthehigherͲclauseandlowerͲclauseversionsofpairslikethatin(3a/3b) and (5/5a) and other considerations. Indeed, the one value of P that most truly fits the biconditional frame of (7) is not even a negͲraiser, given its nonͲtransparency for the licensingofstrictNPIs(cf.Horn1978:207,Klooster2003:4): 

(8)a.ItistruethatJohnwillnot{leaveuntiltomorrow/doafuckingthing}.  b.*ItisnottruethatJohnwill{leaveuntiltomorrow/doafuckingthing}. 

ThefailureofnottruethattolicenseNPIs(asopposedto,say,notlikelythat)isofcourse gristforthesyntacticNRmillofCollinsandPostal,astheyrecognize(2014:§12.4),butI wanttoturntheinvestigationinanotherdirection.   4.Waysofnotknowing Characterizing the class of potential negͲraising predicates has been an avocation of linguists for well over a century, from Tobler (1882) and Kalepky (1891) to G. Lakoff (1970),Horn(1975),andCollins&Postal(2014).Oneconstanthasbeentheostracismof

184

factive predicates, whether epistemic (know, realize, point out) or emotive (regret, conceal, be surprising). The difference between contradictory (Lee didn’t realize it was snowing)andcontrary(Leerealizeditwasn’tsnowing)insuchcasesisjusttoobroadfor these predicates to be viable candidates for “dichotomous” readings. Further, since negation outside the scope of such predicates can’t be interpreted as derived from an embedded clause, these predicates should be uniformly opaque with respect to the diagnosticsforwhichnegͲraisingpredicatesaretransparent.Butasweareabouttosee, whatwefindwhenwelookcloselyenoughissometimesnotopacitybuttranslucence.In particular,considerthecaseofknow. FromKiparsky&Kiparsky(1970)on,knowhasstoodastheepitomeoffactivity.But howfactiveisknow?AgoodstartingpointisHintikka(1962)andhislogicofknowledge andbelief.ForHintikka,negatingaknowclausebringsoutanambiguity,asseeninthis summaryofhistranslations:  (9)a.aknowsthatp   a’.Kap  b.aknowswhetherp   b’.KapvKa̱p c”.p&̱Kap  c.adoesnotknowthatp   c’.̱Kap   d.adoesnotknowwhetherp  d’.̱(KapvKa̱p)ḻKap&̱Ka̱p  Theprincipalcomplication,Hintikkaobserves,residesintheapparentambiguityof(9c),as seen in thedual translation possibilities.  The conjoined (9c”) reading is taken to be the appropriate one for “the most typical cases”, especially third person examples1; (9c’), however, is needed for e.g. an imagined reaction of Russell’s in (10a) to rumors of Wittgenstein’s death in or a challenge in (10b) to an earlier claim that someone is a perjurer(Hintikka1962:13Ͳ14).Asimplerexampleisthatin(10c).  (10) a.Idon’tknowthathe’sdead;therumoursmaynotbetrue,forallthatIreallyknow.   b.Youdon’tknowthatheis;thewitnessesyou’rerelyingonareuntrustworthy.   c.Idon’tknowthatIcanmakeitthereontime.    1 NotethatHintikkaassumesasimpleconjunctionhereratherthanthepresenceorabsenceofa presupposition,althoughhecharacterizessuchascasesasthoseinwhich“thespeakerassumes thatpisinfacttrue”(1962:13).Inthesame“prolegomena”,Hintikkanoticesanddismissesthe “peculiarity” of a does not believe that p in that “it is often used as if were equivalent to “a believes that ̱p”, i.e. as Ba̱p  rather than as ̱Bap, thus dismissing negͲraising as an unfortunatequirkofEnglish.SeeHorn1978:129forsimilarremarksbyQuineandothers.

185

IshallassumethatHintikka’s (9c’)isthebasic compositionalmeaningof(onereading of) not know sentences, remaining neutral as to whether factivity (the presuppositional counterpartof(9c”))representsaforkofatrueambiguityorarisesinsomeothermanner. Butwhatisittonotknow(or,ifyouinsist,nottoknow),andwhatisn’tit?Tobeagnostic istonotknowwhetherornotp;tobeignorantistonotknowpinacontextinwhichpis presupposed.Tonotknowthatpisneithertobeagnosticaboutp(sinceit’scompatible withknowingthat¬p2)nortobeignorantaboutp(sinceit’scompatiblewithcontextsin which ¬p is a live option).  If knowledge is justified true belief as in Plato’s Theaetetus (which it isn’t; see Ichikawa & Steup for a helpful review), to say nonͲfactively that a doesn’tknowthatptendstodenythepresenceofjustificationorevidence. We can begin by leaving aside cases in which a knows that p can be used in positive utterances of (9a) without an entailment that p is the case.  Such cases exist, but are clearly marginal,oftenutilizingscarequotesortheirintonationalcounterpart;one can’t reallyknowwhatisn’ttrue,butatbest“know”whatisn’ttrue. 

(11)a.Thatpeacocksfantheirtailsinthesightofpoisonwasagoodthingtoknow,even ifitwasn'ttrue.”(—DavidR.Slavitt(1971),Anagrams)   b.Everybodyknewthatdoinganythingtoreduceinequalitywouldhaveatleast somenegativeimpactonG.D.P.Butitappearsthatwhateveryoneknewisn’ttrue.    (—PaulKrugmancolumn,NYT10March2014,http://tinyurl.com/nytt4fn) 

Crucially,thereisasharpcontrastbetween(12a,b), 

(12)a.JohnknowsthatMaryisinParis(#butsheisn’t).   b.Johndoesn’tknowthatMaryisinParis,becausesheisn’t. 

and while (12b) may be seen as an instance of presupposition cancellation (as it is for Chierchia2013:405;cf.alsoBeaver2010),thisdoesn’tgeneralizetothecasesin(10).Soif anentailmentispresentinthepositive(9a)butis(sometimes)absentinthenegative(9b), knowcannot(always)beafactiveinthelinguisticsense(Kiparsky&Kiparsky1970).We mightcallit VERIDICALinthesenseofZwarts19953,wereitnotforthegeneraluseofthis term to encompass verbs like believe where no entailment is present (cf. Giannakidou 1998andsubsequentworks).IntheterminologyofKarttunen1971a,(nonͲfactive)knowis aoneͲwayimplicative,alongwithprove,imply,andmake.  2  KnowingthatnotͲp unilaterally entails (nonͲfactive) not knowing than p, assuming a scale of theform. 3

AoneͲplacepropositionaloperatorOpisveridicaliffOp(p)Æp.

186

FinessingtheambiguityquestionfirstraisedbyHintikka,letusrefertotheknowof(9c’) and (10) as knowͲNF, remaining agnostic (notͲknowingͲwhether) as to the status of this categoryasauseordistinctsenseofknow.Oneargumentforambiguityisthatapparent synonyms of know lack nonͲfactive occurrences in simple negative contexts. Thus, by uttering(13a)Imaydisappointyou,butbyuttering(13b)insteadIwilljustconfuseyou. 

(13)a.Idon’tknowthatIcantrustyou.   b.#Idon’trealizethatIcantrustyou. 

ItmaybethoughtthatknowͲNFisrestrictedtothepresenceofnegation,butitshows upmoregenerallyinquestions,asHintikkapointsout(“Doyouknowthathe’sreliable?”). SecondͲperson questions are in fact as natural a context for knowͲNF as first person declaratives,althoughinbothcasestenseiscrucial: (14)a.Howdoyouknowhe’shere?(knowͲNFpreferred;у‘Whatmakesyouthink…’) b.Howdidyouknowhewashere?(knowͲNFdispreferred;у‘Howdidyoufindout…’) 

Other nonͲveridical contexts allow or favor nonͲfactive readings of know, yielding a meaningdistinctfromaknowͲwhetherparaphrase.Thusinthefirstcoupletofthisverse fromCaroleKingandGerryGoffin’s1961popclassicWillYouStillLoveMeTomorrow, 

(15)

I'dliketoknowthatyourlove, IsloveIcanbesureof, SotellmenowandIwon'taskagain, Willyoustilllovemetomorrow?

thesinger’sexpresseddesireisnotequivalentincontent(ormeter)withI’dliketoknow whetheryourlove/IsloveIcanbesureof.Similarly,ifItellmydoctorthatI’dliketoknow thatI’llsurvivetheoperation,orthatI’dagreetotheoperationifIknewI’dsurviveit,this isnotthesameaswishingIknewwhetherornotI’dsurviveit;knowingwhethermaywell benecessarybutnotsufficient. AnothercorrelateofknowͲNFisthecomplementizer.Formanyspeakers(asGoogling confirms),notͲknowͲNFinsistsonthepresenceofthat,whilefactiveknowisindifferent. Thus,thesequence“Idon’tknowthatI’mallthat…”has133Googlehits,comparedtojust 3for “I don’tknow I’mall that…”(Thefirst person present tenseandthe strongNPI all that both rule out factive readings here.) Similarly, compare the nearͲobligatory complementizer in (nonͲfactive) I don’t know that I can and its optionality in (usually factive)Ididn’tknow(that)Icould. While, as we’ve seen, knowͲNF is most comfortable in first person declaratives (and second person interrogatives), the real key is whether the author of the (original)

187

utterance is the possessor of the epistemic state.  Thus, knowͲNF seems to require coreference of John and he in (16b) (cf. Chierchia’s (12b) above, where the nonͲfactive readingonlyemergeswithdirectcancellation);ifJohn=he,(16b)isessentiallyaninstance oftheimpeccablynonͲfactive(16a).  (16) a.Idon’tknowthatMaryisinParis. b.Johnsayshedoesn’tknowthatMaryisinParis.  Compare(17)(followingmyestablishedpractice,IuseJtomarkGoogledexamples): 

(17) Jhesaysthathedoesn'tknowthatI'mthewomanhewantstospendtherestof hislifewitheventhoughatthemomenthelovesme.(http://tinyurl.com/l5w8e75) 

AsDearCupid’scorrespondent,nineweekspregnant,knowsalltoowell,herboyfriend’s notͲknowingisnonͲfactive. Inhisrecenttreatiseonthelogicofpolarityandimplicature,Chierchia(2013:Ch.7) investigates the role of factive presuppositions as intervener that block the licensing of even weak NPIs by negation outside their scope. He points out that this effect is parameterized across languages, so that John doesn’t know that Mary stole anything, while grammatical in English, has no grammatical counterpart in Italian. On Chierchia’s account,thepresuppositionoffactivesispredictedtointerferewithpolaritylicensing,but the facts as reported appear to be admittedly unsystematic (2013: 401), yielding arrays likethatin(18)(=Chierchia’s(60)). 

(18)a.LeodoesnotbelievethatJohnstoleanyvase   b.?EventhoughJohndidstealavase,Leodoesn’tknowthatJohnstoleanyvase.   c.Itisn’tatallstrangethatMarywroteanybooks.   d.Pauldidn’tdiscoverthatMaryeverwenttoNewYork. Actually,oncloserexamination,thefactsaren’tasmurkyastheyappear.In(18b),know isclearlyfactiveandtheresultisindeedprettybad.Complicatingmattersistheunlikelya vase…any vase sequence; (18b) remains an odd sentence even if doesn’t think replaces doesn’tknow.Butcomparethefollowing,wheretheknow/thinkdistinctionisclearer,and thefactsgoaspredicted:  (18’) Leodoesn’t{think/#know}thatyouhaveanyfriends,eventhoughyoudo.  

188

In (18c), strange is indeed factive and any is indeed fine. However, it’s not the upstairs negation that licenses the NPI any (although that negation does license at all in the matrix),butrathertheemotivefactivestrangeitself:  (18)c’.It’sstrangethatMarywroteanybooks.  Thedistinctionbetweencognitive/epistemicfactivesandemotivefactivesisanimportant one (as the Kiparskys first recognized) and its ramifications extend well beyond polarity licensing. In (18d), discover is at most a semiͲfactive, lacking any presupposition in a negative statement(especiallyinthepresenceofanNPI);cf.Karttunen1971b,Beaver2010. Chierchia does not distinguish factive from nonͲfactive occurrences of know (and nor doesSchueler2005inhisstudyofinterventioneffects),butthisisdirectlyrelevanttoits interventional potential. When know behaves as a factive, its interventional potential is revived.  While the naturally nonfactive first person cases—I don't know that I did anythingwrong—areimpeccable,eventhethirdpersoncasesaremorenaturaliftheycan be read as knowͲNF: compare (18’) with Leo says he doesn’t know that you have any friends.Ifwereplaceknowwithpointout(anepistemicpredicatethatdoesn'thaveanonͲ factiveuse),theresult,Ididn'tpointoutthatIhaddoneanythingwrong,seemshopeless. AnotherapproachtothefactivityfactoristoGooglesequencesinwhichknowͲNFiseither favored(inthefirstcase)ordisfavored(intheothertwo): (19)“don’tknowthatanythingever”:   “doesn’tknowthatanythingever”:         “didn’tknowthatanythingever”: 

91hits 3hits,ofwhichtwoareclearlynonͲfactive ("Hesayshedoesn'tknow…") 5hits

Why these particular strings? The “doesn't know” eliminates first and second person cases, where the knowͲNF is easiest to get, and the “anything ever” is easily searchable whileknockingoutfreeͲchoiceany,whichinfectsexampleswithanythingtoutcourt,and past tense favors factive readings. So this tends to support the hypothesis that when a negationoutsideknowlicensesNPIs,it’sgenerallyaninstanceofknowͲNF,whencealso the intervention properties of persistent factives like point out. Of course the equally persistentemotivefactiveslicenseNPIsontheirown,butthat'sadifferentproblem(see Linebarger1987andeveryonesince). The NPIs licensed in the cases under consideration here are the weak ones, in particular, ever and (polarityͲsensitive) any and its derivatives. Predictably, factives

189

intervenetoblocklicensingofstrongNPIs—afterall,iffactivescan’tbenegͲraisersandif onlynegͲraiserscanbetransparenttothelicensingofstrongNPIsinembeddedclauses, howcouldnegationoverknow,evenknowͲNF,licenseastrongNPI?Theproblemisthat strongNPIsdooccurundernegatedknow—ormorepreciselyundernegatedknowͲNF,as weshallseebelow.Idon’tknowthatChierchia2013orCollins&Postal2014,interalia, wouldbeallthatpleasedwiththisfact,butafactitremains. 5.Whentonotknowistoimply:PattyHearst’srevenge  Following Horn 1975: 283 and Horn 1978, Collins & Postal (2014: Chapter 13; cf. also Collins2013)constructanargumentforthesyntacticnatureofNRbasedonthepossibility ofembeddedsubjectͲauxinversion.TheycitethefollowingpassagefromHorn1978:168Ͳ 69;IfollowCollins&Postalinpreservingtheoriginalexamplenumbering.  “ConsiderthefollowingobservationmadebyatelevisionnewsmancoveringthePatty Hearstabduction: (96)Idon’tthinkthateverbeforehavethemediaplayedsuchamajorroleina kidnapping. In this sentence, SubjectͲAuxiliary Inversion has not only applied in a subordinate clause,ithasapparentlybeentriggeredbyanegativewhich appearsinthesurface structureintheclauseabovethatwheretheinversion hastakenplace. Inatheory withasyntacticruleofnegͲraising,wemightpositasan intermediatestringinthe derivationof(96)somethingoftheform: (97)IbelievethatNEG[themediahave(at)sometimebeforeplayedsuchamajor roleinakidnapping] [...]IfthehigherpredicateisnotanNRͲgovernor,only…NegativeIncorporation[but notSubjͲAuxInversion]isavailable: (99)a.Iclaimthatneverbeforehavethemediaplayed...   b.*Idon’tclaimthateverbeforehavethemediaplayed...” 

 McCawley (1998: 598), endorsing this argument for NR, adding the (constructed) examplesin(20),towhichCollinsandPostaladdtheattestedandconstructedexamples in(21)illustratingwhattheytermHORNCLAUSES(HCS):  (20) a.Idon’tsupposethatunderanycircumstanceswouldhehelpme.   b.Wedidn’tanticipatethatatanytimewouldourworkcreatedifficulties. 

190

(21)      

a.JIdon’tbelievethatatanytimedidtrafficcometostandstill. b.JIdon’tbelievethatatANYtimedidRockstarconsiderclosingshop. c.Ididn’texpectthatforanyreasonwouldsheagreetothat. d.Ididn’timaginethateitherofthemwouldshebeanxioustomarry.



The problem with this otherwise elegant argument for a syntactic rule of NR is that “dichotomous” status is not a necessary property for triggering embedded subjectͲaux inversionfollowingafrontedNPI,asrevealedbyouroldfriendknowͲNF,astaunchnonͲ negͲraiser(Idon’tknowthatpтIknowthat¬p): 

(22) a.JIthasbeenmyfortunetohavemanysuchcases,butIdon’tknowthatever beforehaveIhadthreesuchcasesinsuccession.(TheClinique,Vol.4,1883)  b.JIdon’tknowthateverbeforeorsinceinmylifehaveIfeltsuchexhaustion. (HughWalpole,TheDarkForest,1916)  c.Idon’tknowthatEVERbeforehadallthreeboysnappedsimultaneously. (http://www.vrbo.com/194521/reviews)  d.Idon’tknowthateverbeforehadtheArmycommanderbeeninchargeofthe Navalforces.(http://www.americancivilwarforum.com)  e.“Idon’tknowthateverbeforehastheworldbeensosmallinthesenseofmedia gettingouttothepeople.”(http://tinyurl.com/kvdjn3f)  f.“Idon’tknowthateveragainwillwehavefour6Ͳ10guysontheteamatthesame time.Andtheyallcanplay.”(http://tinyurl.com/lbtc7mm) HCsarelikewisepossibleundercan’tsayandnotawareͲNF,whichareequallynonͲraisers (I can’t say that p т I can say that ¬p); as above, the googled examples are all fully acceptableinmyowndialect.  (23)a. Fortherecord,Iview“hopingthatthereisnoChristianGod”tobeutterlysillyand useless.Ican’tsaythatatanytimehaveIeverhopedthatthereisnoChristian God(hah!Asifthere’sjustone.(http://tinyurl.com/mdralo7)   b.Ican’tsaythatatanytimedidIhaveaproblemwithanyofthecustomerservice team.(http://www.reviewcentre.com/reviews182288.html)   c.Ican’tsaythatatanyofthosesituationswithanyofthosewomenhasit [=belowͲaveragepenissize]everpresenteditselfasarealissue.   (http://www.documentaryͲlog.com/penisͲsizeͲinsecurityͲbyͲmen)

191

(24)

“I’mnotawarethateverinthehistoryofNewYorkStatehassomethinglikethis happened.It’sanunfortunatefirst.NormallywhenStateagencieslaypeopleoff, thereisanannouncementthatlayoffswillhavetooccur.” (http://tinyurl.com/m8qpygq)  Note that although no lowerͲneg reading is available in such cases, there is a negative proposition that is at issue in such cases; in (24), for example, the speaker’s point is to indicatethathesuspectsthatsuchabruptlayoffshaveneverhappenedbefore.4 Crucially, this same class of nonͲnegͲraisers tends to allow strict NPIs in embedded clauses,againcontrathereceivedwisdom,andagainwiththeimplicitsuggestionthatthe negationoftheembeddedpropositionholds,sothatIsabel,theauthorof(25c),istaken assuggestingthatshehasn’tcookedmyselfafullmealinweeks.  (25) a.JIdon'tknowthatSantacomesaroundthesepartsuntilChristmasEve. (http://www.sowonderfulsomarvelous.com/2013/12/santaͲvisit.html)  b.J[W]ithmy1stpregnancy…Idon'tthinkIgotbreastmilk.Idon'tknowthatIhave itthistimeeither.(http://tinyurl.com/n35lw62)  c.JIcan'tsayI'vecookedmyselfafullmealinweeks,ifnotmonths.   (http://starnickel.com/undercoverkinkster/blog1.php/longͲtimeͲnoͲblog)  d.JI’mnotawarethatherbidentificationwalkswereacommonpracticeat*any* veterinarymeetinguntilVBMAbeganofferingoneatitsyearlyconference.   (http://www.vbma.org/Susan%20Wynn.html)  

This is not a new observation. As remarked in Horn 1978: 148, following Lindholm (1969:153Ͳ4),nonclausematenegationcanlicenseembeddeduntilphrases“despitethe unavailability of either a paraphrase with the negation in the lower clause or, concomitantly, any possible NR derivation.” Indeed, this observationgoes back toBaker (1970),whopointsoutthat(26a)isacceptablebyvirtueofitsimplicationthatwhatever yousay,Iwillgoonbelievingthathe DIDN’Tliftafinger.Wherethisimplicationisabsent, asitisinthesyntacticallyparallel(26b),thepolarityitemishardertoget. 

(26)a.Youcan’tmakemebelievethatheliftedafingertohelp.   b.#Youshouldn’tmake{me/her}believethatheliftedafingertohelp. 

 4  I argue in Horn (to appear) that the distribution of NPIs—both weak and strong—is robustly linkedtothepresenceofatͲissuenegativepropositions;cf.Linebarger(1987)forarelatedview.

192

Cornulier (1974) makes a similar point about French based on the data of (27) (glossesaremine)involvingahighregistermainͲclausesubjunctiveexpressingsavoirͲNF; cf. http://tinyurl.com/kh5okw6 for another take on this somewhat restricted doubleͲ subjunctiveconstruction.  

(27)a.Jenesache[SUBJ]pasqu’ilait[SUBJ]jamaisditcela.    ‘Idon’tknowthathe’seversaidthat’   b.Jesaisqu’iln’a[INDIC]jamaisditcela.    ‘Iknowthathe(has)neversaidthat’   c.Iln’ajamaisditcela{quejesache/àmaconnaissance}.    ‘He(has)neversaidthat,{thatIknowof/to(thebestof)myknowledge}’  

(27a)cannotbeparaphrasedby(27b)—asitwouldhavetobeifsavoir‘know’wereinfact an NRͲverb of the croire (‘believe’) stripe—but rather by (27c), an admission of lack of positive knowledge rather than an assertion of knowledge to the contrary. (Notice that thisistrue oftheEnglish glossesaswell.)Yetthespeaker’simplicationthatthat sucha negativepropositionholdsissufficienttolicenseanembeddedjamaisintheabsenceofa tautoclausalnegationoraNRpredicate. Letustouchononemoreinstanceofa(putatively)strictNPIlicensedunderknowͲ NF.Collins&PostalrankallthatintherosterofstrictNPIs.Theirjudgmentsincludethe paradigmin(28)(=their(21)inChapter9).  (28)a.Arnoldis*(not)allthatintelligent.   b.Lucindadoesn’tbelieve/thinkthatArnoldisallthatintelligent.   c.*Lucindadoesn’tknow/realizethatArnoldisallthatintelligent. However,wellͲformedanaloguesof(28c)arenotallthathardtofind,asintheexchange in(29),fromClintEastwood’s1992westernUnforgiven.  (29)NedLogan:Howlonghasitbeensinceyoufiredagunataman,Will?Nine,tenyears?   WillMunny:Eleven.   NedLogan:Easy,huh?Hell,Idon'tknowthatitwasallthateasyevenbackthen.And wewasyoungandfullofbeans. 

Relevantinstancesof“Idon’tknowthat*isallthat…”areepidemicontheweb,butwe alsofindanoccasionalthirdͲpersonexampleintheframeof(28c),e.g. 

193

(30)JButshedoesn'tknowthatit'sallthatbad.Ahousethat'sdustyandfullofstuff.   (http://crookedkat.tumblr.com/) 

Crucially, however, these examples all feature know, i.e. knowͲNF, rather than realize, despite the equivalence Collins & Postal seem to be suggesting in (28). There are, for example,dozensofGooglehitsfor“Idon’tknowthat___is/’sallthatcommon”—about thesamenumberasforbelieveorthink—asagainstnoneatallforrealize,which(while permitting presupposition cancellation in some environments, as shown by Karttunen 1971b)doesnotoccurasacoherentnonͲfactiveintheframeIdon’t___thatp. We have seen that nonfactive occurrences (especially but not only in first person cases)ofadoesn’tknowthatpoftentendsuggestthatthespeakerthinksthat¬pislikely tobethecase,butwedohaveexampleslikethatin(31),wheretheradiocommentator was struggling with certain counterfactual truth conditions (this was uttered after the 49ersbenchedSmithinfavorofColinKaepernickandwentontogettothe2013Super Bowl).  (31)Idon’tknowthatthe49erswouldbeintheSuperBowlifAlexSmithwerestillthe  quarterback.Idon’tknowthattheywouldn’tbe.But…  Butevenhere,despitethedisclaimerinthesecondsentence,thepointoftheutteranceis todiscountthepossibleworldsinwhichSmithremainedtheQBandthe49erssucceeded. ThecanonicalcaseinwhichHintikka’s(9c),adoesn’tknowthatp,isreadas(9c’),knowͲ NF, rather than as (9c”), is one in which the speaker (or the subject, when these are distinct)undertakestohighlightherbeliefinthepossibilityorlikelihoodof¬p,NegͲraising isasufficientconditionfortriggeringthiseffect,tobesure,butnotanecessaryone.And itisthishighlightednegativepropositionthatresultsinthecorrelateswehaveobserved, fromtheconditionsoninterventiontothedistributionof“Hornclauses”andstrictNPIs. Thisbringsustothefollowinghypothesis,withwhichweconclude:  ThecrucialfactorinlicensingembeddedsubjectͲauxinversion(“Hornclauses”)and strictNPIswhentheseoccurunderhighernegationinthesequence[aNEGͲFsthat p] is not the requirement that F be a NR predicate per se but the existence of a robust association between a being in a NEGͲF relation to p and a being in anF´ relationto¬p,whereF´=ForF´