Epistemic Containment Kaivon Fintel Sabineiatridou

Epistemic Containment Kaivon Fintel SabineIatridou

Thisarticle concerns a newconstraint on theinteraction of quantifier phrasesand epistemic modals. It is argued that QPs cannotbind their tracesacross an epistemic modal, though it is shown that scoping mechanismsof adifferentnature are permitted to cross epistemic modals.The nature and source of thisconstraint are investigated.

Keywords: quantification,modality, scope ambiguity, raising, clause structure,reconstruction

1Introduction Whena sentencecontains two quantificational items, it is quite often ambiguous between two logicalstructures that differ inthe relative scope of the two quantifiers. Until recently, it was widelyassumed that scope ambiguities are generally due to aninstanceof Move a calledQuantifier Raising(QR), whichwas assumedto operatequite freely (although it was knownto beconstrained byfinite clause boundaries). 1 Theassumption was, andis, that the investigation of QR will yieldan increased understanding of thelevel of syntacticrepresentation that interfaces with the interpretivecomponent. Alongthe way, consensus has been converging toward certain beliefs. Research has shown manyways in which scopal relations are much more rigid than previously suspected. 2 At the sametime, some scopal relations seem freer thanone might expect, especially if one looks at the propertiesof indefinite quantifiers (e.g., indefinites can take scope out of islands). In order to explainthe unexpectedly wide scope of indefinites, a currentlypopular approach is thatindefinite quantifiershave alternative ways of scopetaking, including choice functions, Skolem functions, andunselective binding by overt or covert operators. 3 Inother words, it is only an illusion that

Thisarticle is arevisedversion of a manuscriptwe circulatedunder the title ‘ ‘Onthe Interaction of Modals,Quantifiers, and If-Clauses.’’ One partof thatancestral paperhas nowbecome astand-alonework entitled ‘ ‘If andWhen If-Clauses CanRestrict Quantifiers’’ (vonFintel and Iatridou 2002). Material fromthis article has beenpresented at colloquiaat theUniversity of Massachusetts at Amherst, theUniversity of Arizona, and the University of Connecticut;at theGLOW conference inBraga, Portugal; and at Sinnund Bedeutung 7 inKonstanz, Germany. We thankthe audiences for their kindremarks andhelpful criticism. We thanktwo anonymous reviewers for LinguisticInquiry fortheir very useful comments. We alsothank Noam Chomsky,Bridget Copley, Danny Fox, Irene Heim, AngelikaKratzer, HowardLasnik, WinnieLechner, Lisa Matthewson,David Pesetsky, Norvin Richards, and Peter Svenoniusfor valuable input. Our names are listedin alphabetical order. 1 See amongothers Rodman 1976 and May 1977, 1985. 2 See amongothers Aoun and Li 1993,Szabolcsi 1997, Fox 2000, and Bruening 2001. Early work already contained muchevidence for strong constraints on quantifier scope (see, e.g.,Ioup 1975). 3 See Heim 1982for unselective binding. See Reinhart1997a,b, Winter 1997, and Kratzer 1998for choice-function indefinites.

Linguistic Inquiry, Volume 34, Number 2,Spring 2003 173–198 q 2003 bythe Massachusetts Institute ofTechnology 173 174 KAIVONFINTELANDSABINEIATRIDOU

QRasa movementis freer thanovert (A ¯ -)movement;when indefinites take higher scope than expected,it is not QR thatis responsible. On the other hand, the conclusion that QR ismore restrictedthan originally thought is takento bereal, as researchuncovers the existence of different sortsof interventioneffects and other constraints that block QR. Here, we addto this exploration a newobservation, a constrainton the scopal interaction ofquantifierswith epistemic modals 4 thatwe callthe EpistemicContainment Principle. 5 (1) TheEpistemic Containment Principle (ECP) Aquantifiercannot have scope over an epistemic modal. Thestructure of the article is as follows. In section 2, we presenttwo kinds of evidencefor theECP, one involving simple scope judgments and one involving binding (im)possibilities. In section3, we considerthe nature of the ECP. We concludethat it is avaliddescriptive generaliza- tion,but we donotdecide between two ways of integrating it into the grammar. In section 4, we discussreconstruction of A-moved quantifiers. In section 5, we brieflymention some ways of scopetaking that do not involve quantifier movement.

2Evidencefor the ECP 2.1Scope Judgments TheECP saysthat a quantifiercannot have scope over an epistemic modal —anunexpected claim, sincesentences containing quantifiers and modal or temporal elements routinely show clear scope ambiguities.Here isanexampleof a scopeambiguity between quantifiers and temporaloperators: (2) Mostof ourstudents will be professors in a few years. (3) Mostof ourstudents will be foreigners in a few years. Themost natural readings for thesetwo sentences involve different scopes: (2) saysthat most of our current studentswill in a few yearsbe professors ( mostof our students s will), while (3)

4 Very recently,we became aware ofa manuscriptby Drubig (2001), who independently claims thatquantifiers cannottake scopeover epistemic modals.As far as we can tell,Drubig gives no evidence to support his conclusion, however.On page 15, he cites Leech (1971:73)as havingobserved that quantifiers seem tobeunableto take scopeover epistemic modals.Drubig cites thisexample fromLeech: (i)Every clause mustcontain a finiteverb. He thencomments, ‘‘Thedeontic reading of [(i)] is compatiblewith wide and narrow quantifier scope, whereas the epistemic modalalways can onlyhave wide scope’ ’ (p.15). But of coursewith two expressions having universal force, every and must, scopeambiguities are exceedinglyhard to diagnose. That is why we use at least onenonuniversal expressionin the quantifier scope examples discussedbelow. From Leech’ s originalpassage, itis clear thathe didnot infact anticipateour ECP but correctly observed that for principled reasons, no scope ambiguity is detectable in(i). 5 Thename ofthisconstraint was carefullychosen by oneof theauthors. The other author warned that he vaguely remembered fromsyntax courses hetook as agraduatestudent that there was anotherprinciple that went by theabbreviation ECP.But he was assured thatthat had been a temporaryaberration in thescience andthat the competing principle had nowdeservedly sunk into near oblivion. EPISTEMICCONTAINMENT 175 saysthat in afew yearsmost of our students at that time will be foreigners ( will s most of our students).6 Quantifiersalso appear to show scope ambiguities with deontic modal operators. (4) Mostof ourstudents must get outside funding — a.for thedepartment budget to work out. b.the others have already been given university fellowships. Thesentence in (4) isambiguous,an ambiguity brought out by the different continuations. (4a) bringsout the reading where mostof our students takesscope below the deontic modal: for the budgetto work, it needs to bethecase that most of thestudents get outside funding. (4b) makes salienta widescope reading for mostof our students, sincethe obligation is imposed on those specificstudents who have not already been given fellowships. Whatthe ECP issaying is that parallel facts do not obtain with epistemic modals. So, it predictsthat there is no scope ambiguity in (5), with only the order in (5a) beingpossible, not theone in (5b). (5) Mostof ourstudents must be home by now. a. must s mostof ourstudents b.*most of ourstudents s must We predictthat (5) onlyhas a readingin which the quantifier mostof our students has narrow scopeunder the epistemic modal must. Thismeans that in order for (5) tosatisfy the ECP, the quantifierneeds to establish narrow scope by LF. Of course,it is very hard to figure out what the relative scope of most andepistemic must in(5) is.Consequently, our evidence for theECP willhave to involve different combinations of quantifiersand epistemic modals. We arestanding in front of anundergraduate residence at the Institute. Some lights are on andsome are off. We don’t knowwhere particular students live but we knowthat they are all conscientiousand turn their lights off whenthey leave. So, we clearlyknow that not all of the studentsare out (some lights are on andthey wouldn’ t beonif the students were away).It could infact be that all of themare home (the ones whose lights are off mayalready be asleep). But itis also possible that some of them are away. Since we don’t knowwhich student goes with whichlight, for everyparticular student it is compatible with our evidence that he orshehas left. Withthis background, consider the following sentence: (6) Everystudent may have left. a.every student x (mayx haveleft) true, *ECP b.may (every student have left) false, O K ECP Informantsreliably judge (6) tobe false in the scenario just sketched. This is predicted if the

6 Of course,many issues are glossedover here. See Musan1997 for in-depth exploration of thistopic. 176 KAIVONFINTELANDSABINEIATRIDOU

ECPisoperative. It would force (6) tobereadwith narrow scope for thequantifier everystudent, whichgives rise to a readingthat is false in our scenario. The ECP prohibitsthe reading where everystudent hasscope over the modal, a readingthat would be true in the given scenario. A raw truth-valuejudgment then supports our claim that there is anECP. Thejudgment is perhaps even sharper when we considera continuationof (6) thatwould beconsistentonly with the scope every s may, whichis ruled out by theECP. We expecta clear feelingof contradiction, and indeed, (7) isbeyond repair. 7 (7) *Everystudent may have left but not every one of themhas. Here isanother case showing that every cannothave scope over may: (8) #Everystudent is the tallest person in the department. (9) #Everystudent may be the tallest person in the department. a.every student x (mayx bethetallest) sensible,*ECP b.may (every student be the tallest) nonsense, O K ECP If therewere noECP, (9) wouldbe predicted to be sensible since the quantifier would be able tohavescope over may. Butbecause of theECP, this is not possible and the obligatory narrow scopeof thequantifier leads to a pragmaticallyodd reading on a parwith the oddness of (8). Imaginea groupof peoplewho were exposedto an infectious agent. From anonymoustest results,we haveconcluded that half of the people are healthy, but we don’t knowwhich ones theyare. For anygiven person, then, we don’t knowwhether he or she is infected. If theECP didn’t hold,we wouldbe able to report this situation to the people involved as follows: (10)Half ofyouare healthy. #But everyone may be infected. a.every person x (mayx beinfected) consistent,*ECP b.may (every person be infected) inconsistent, O K ECP However,this sequence seems utterly inconsistent. The reason is bynowclear: the second sentence canonly be readas asserting the possibility that everyone is infected, which contradicts the first sentence.8 TheECP effectalso arises with quantifiers other than universal every. Considerthe following example,which indicates that the scope order must s fewerthan half ispossible:

7 We havesometimes encounteredpeople who thought that with the right kind of intonation they might get readings thatthe ECP would rule out. We are unsureabout whether this is sometimes possible.But (7) at least is aclear case where noamountof prosodicacrobatics helps.What does help with (7) and (9), as well as manyof ourother examples, is using any instead of every. We are well aware ofthatand ask thereader towait untilsection 5.2 for an explanation. 8 Bythe way, the ECP seems toconcernnot only epistemic modalsin the form of auxiliaries.Example (10) remains unacceptablewhen we switchto an adverbial modal. (i)#Half of you are healthy.But everyone is possibly infected. An LI reviewer pointsout further that the (at least quasi-epistemic)modal be bound to alsogives rise toECP effects. However,at thispoint we are notin a positionto widen our inquiry beyond a small class ofmodalauxiliaries, so we willleave thematter withjust these observations. EPISTEMICCONTAINMENT 177

(11)Fewer thanhalf of thestudents must have passed the test. (Otherwise there wouldn’ t bethisuproar.) We claimthat (11) lacks the reading ‘ Fewer thanhalf of the students are such that it can be said withcertainty about them that they have passed the test’ . Thisis the reading ruled out by the ECP.We canshow that the reading is absentby combiningthe sentence with a continuationthat wouldforce the wide scope reading of thequantifier over the modal, producing a contradictory statement. (12)#Fewer thanhalf of the students must have passed the test, but perhaps all of them did. (12)is contradictory since the ECP forcesan interpretation like ‘ Itmust be the case that fewer thanhalf the students passed the test, but perhaps all of themdid’ . Onemight wonder whether the ECP concernsonly strong quantifiers. But we thinkthere is clearevidence that all quantifiers are subject to the constraint. (13)#Two friends of John’s from Texasmay have come to visithim this weekend, but they can’t bothhave come (because they hate each other). Ascenariothat would go withthe sentence in (13):regarding two friends of John’s from Texas, we knowthat either one may well have visited him, but we alsoknow that they did not both visit him.But the ECP forcesthe quantifier two friends tohavenarrow scope under the modal, which expressesthat it is possible that both of them came, which in turnis inconsistent with the follow- up.(13) is therefore unacceptable. Indefinites raise other issues as well, which we willlook at insection 5. We wouldlike to add one last instance of the ECP, noted by Heim (2001). Compare (14) and (15). (14)The paper is allowed to be less long than that. ReadingA: Itis allowed that the paper be shorterthan that. ReadingB: Itis not allowed that the paper be thatlong. (15)The paper might be less long than that. ReadingA: Itmight be thecase that the paper is shorterthan that. ReadingB: Itis not possible that the paper is that long. (unavailable!) Heimsuggests that the B readingof (14)is dueto LFmovementof thedegree-quantifier phrase acrossthe modal operator. She observes that (15) does not have a Breading.She conjectures thatthis effect is dueto the ECP.

2.2Binding (Im)possibilities

We saidearlier that the ECP cannotbe detectedin simple scope judgments involving every and must, becausethey are both universal quantifiers. But even with this combination, the ECP can 178 KAIVONFINTELANDSABINEIATRIDOU beshown to rule out a structurewhere the quantifier has scope over the epistemic modal. The trickis todetermine where the quantifier is located by looking at what it can bind. Youare going to the same dorm as before, planning to visit Chris but worrying whether youmight wake him up. Kim tellsyou to lookat hisdorm window because Chris must be awake ifhislight is on. So Kim says: (16)Chris must be awakeif hislight is on. Kim’s implicitsuggestion is to find the window of Chris’s roomand check whether the light is on.If itis on, you can conclude that Chris is awake. You ask Kim howshe can be sosure. She tellsyou that Chris is averyconscientious guy and would never leave the light on whenhe leaves hisroom or whenhe goes to sleep. In fact, she thinks that every student who lives in this dorm hasthis property. So, she might be temptedto assert that

(17)*Every student i mustbe awake if his j/* i light is on. Thepuzzling fact is that Kim can’t saywhat she is tempted to say. The sentence in (17) is unacceptableunder the intended meaning (where thequantifier everystudent bindsthe pronoun his).Armed withthe ECP, we correctlypredict the actual status of (17).Without the ECP, (17) ispredicted to have two possible LF configurations,depending on whether the quantifier has scopeover the modal (as in(18a)) or themodal has scope over the quantifier (as in(18b)). 9 (18) a. every s must ` if every [lx.student x]

[lx. must [lw. if hisx light isw on] [lw. x bew awake]] b. must ` if s every

must [lw. if hisx light isw on]

[lw. every [lx. studentx ] [lx. x bew awake]] Thestructure that comes closest to being available for (17)is the one in (18b). But note that in (18b),the quantifier cannot bind the pronoun, since the requisite c-command relation does not holdbetween the quantifier and the pronoun. 1 0

9 We are assuminghere withoutargument that the if-clause in(17) is restrictingthe epistemic modal must, in the sense oftheLewis-Kratzer viewof if-clauses (see Lewis 1975,Kratzer 1978,1986, Heim 1982).We assume furtherthat torestrict themodal, the if-clause needsto be local to themodal at LF.Weignorehere thepossibility that the if-clause mightbe restrictingthe quantifier every student, functioningin a way likea relative clause. Invon Fintel and Iatridou 2002,we showthat if-clauses cannotrestrict nominalquantifiers like every student. 10 We ofcourse assume thefollowing near triviality: (i) LFconditionon variablebinding Quantifiers(or the l-operatorsassociated withthem) can onlybind variables in their LF c-command domain. Inaddition, we adoptthe following common assumption: (ii) LFlocalityof restrictor At LF,restrictors are localto the elements theyrestrict. EPISTEMICCONTAINMENT 179

So,the pronoun in (18b) remains free. Indeed,a sentencepronounced the same as (17) has the(18b) reading, where the pronoun remains unbound by the quantifier. We canunderstand it asmeaning something like ‘ If his(e.g., the janitor’ s, or some other salient individual’ s) lightis on,it must be thecase that every student is awake’ . Butthe fact remains that there is no reading wherethe pronoun is bound,as indicatedin (17). The structure in (18a) is cruciallyruled out by theECP. The result is the judgment indicated in (17). Notethat the ECP onlyconcerns the interaction of quantifiers with epistemic modals. As soonas we usea nonepistemicmodal, then, we shouldfind acceptable structures— acorrect prediction.

(19)Every student i mustcontact the dean if he i istoosick to attend the exam.

Inthis sentence, the modal must hasa deonticreading stating an obligation following from some bodyof rulesand regulations.

3TheNature ofthe ECP

Our initialformulation of the ECP in(1) statedthat a quantifiercannot have scope over an epistemicmodal. In this section, we explorefurther the nature of thisprinciple. We willconsider twopossible ways of understanding the ECP: (a) asan intervention effect blocking quantifier- variablerelations, and (b) asareflectionof thestrict ordering of projectionsin the topology of clauses.

3.1Against a RadicalIntervention Account

Oneway to derivethe ECP ina naturalfashion would be to claimthat for deepreasons epistemic modalsdo not tolerate complements that contain variables bound from outside.The intuition mightbe that epistemic modals do not tolerate ‘ ‘openpropositions’ ’ astheir complement. This iscertainly a viewworth considering. In fact, there is oneplace in the literature where just such aphenomenonhas been described: Kratzer’ s workon the fact that the German discourse particle ja doesnot tolerate binding across it (Kratzer 1999).While this area of inquiry is exciting and we haveevery intention of pursuingit ourselves in future work, we donot think that epistemic modalsbehave in this manner. In fact, we haveconvincing counterexamples to the hypothesis thatepistemic modals do not allow binding of variablesacross them. First,a quantifiercan bind a pronounacross an epistemic modal.

(20)Every student i isconvincedthat I thinkthat the dean may well expel him i .

(21)Every man in the room i was askedto come here because he i maywell have seen the accident.

(22)a. Every student i thinksthat he i maywell be agenius.

b.Every student i thinksthat Mary may/ mustlike him i the most. 180 KAIVONFINTELANDSABINEIATRIDOU

Second, a wh-phrasecan bind its trace across an epistemic modal. 1 1

(23) Whoi mustshe have hired t i for thatjob?

(24) Whoi didBill say that Susan must have married t i ?

(25) Whoi mustBill have said that Susan married t i ?

(26) Wherei didBill say that Bill must have seen Susan t i ?

(27) Wherei mustBill have said that Bill saw Susant i ? (28)It didn’ t takeme long to findout which workers the boss must have reprimanded.

What about wh-in-situ?Consider a typicalmultiple question in English. (29)Who thinks that Susan talked with who?

Whathappens when an epistemic modal is inserted in the path between the two wh-phrases?1 2

(30)Who said that Susan must have talked with who?

Mostspeakers we haveconsulted find no contrast between (29) and (30). Depending on what analysisof structures like (29) one assumes, one can conclude either (a) thatepistemic modals donot block the relation between a wh-phrasemoved covertly at LF andits trace, or (b) that epistemicmodals do not block the dependency between a wh-phrasein situ and whatever it is relatedto inthematrix C position. 1 3 Inany case, the conclusion is that epistemic modals do not block wh-in-situdependencies. 1 4

11 We shouldnote that Drubig (2001) claims thatepistemic modalscannot occur in wh-questions.He cites McDowell 1987in supportof thisclaim andgives examples likethese: (i)Why must/ may Johnleave early tonight? (ii)Who must/ may leave early tonight? Notethat these involveeventive/ nonstativepredicates, which Drubig elsewhere inhispaper correctly discusses as resisting epistemic readingsanyway. (Similarly, Drubig uses thewrong predicate toillustrate that epistemic modalsare excluded fromthe protasis of a conditional. (iii)If Johnmust/ may leave early tonight,. ..) Thetest reallyrequires stative predicates suchas theperfect tense oneswe use inthe text. The following additional examples are fromBrennan 1993:24: (iv)What may hehavedone? (v)Where mustthe murder weapon be hidden,in your view? (vi)Who can haveleft thisbaby on mydoorstep? 12 AndyBarss (personalcommunication) prompted us toconsider such cases. 13 Foranalyses ofmultiplequestions that involve, not covert movement of the lower wh-phrase,but some otherkind ofpossiblylong-distance dependency, see Baker 1970,Tsai 1994,Reinhart 1997b, Pesetsky 2000. 14 TheECP might be reminiscent ofsome effects concerningcovert movements of wh-elements. Beck (1996a) discusses avarietyof wh-related movementconstructions where there isa choiceof leavingsome material insitu at S- Structure.One kindof example is this(Pesetsky 2000, cited from Beck 1996a): EPISTEMICCONTAINMENT 181

We concludethat under the intervention effect view, the ECP mustbe statedas a condition thatis specificallyabout the relationship between quantifiers and their traces at LF. As such,the ECPprohibitsLF raisingof a quantifieracross an epistemic modal and forces quantifiers that haveraised on the surface past an epistemic modal to take scope back under the modal. So, insteadof ‘ ‘Aquantifiercannot have scopeover an epistemic modal,’ ’ thisanalysis now says, ‘‘Aquantifiercannot take scopeover an epistemic modal.’ ’ Moreprecisely: (31) TheECP revisedas a conditionon QR At LF,aquantifiercannot bind its trace across an epistemic modal.

*Qi ...[EpistemicModal (. ..t i . . .)] We neededthe ECP toexplain the fact we observedabout (17). If oursuspicion about the ECPisright, we expectthat the following ways of expressingthe intended content of our puzzle sentence(17) should be acceptable,since they involve variable binding across the modal but not quantifiermovement.

(i) Who ` among DP a. Noseparation, no intervener [Wenvon den Musikern] hat Hans —— getroffen? whomof the musicians has Hans —— met ‘Whoamong the musicians has Hans met?’ b. Separation,no intervener Wenhat Hans [ —— vonden Musikern] getroffen? c. Noseparation, intervener [Wenvon den Musikern] hat keine Studentin getroffen? whomof the musicians has nostudent met ‘Whoamong the musicians has nostudentmet?’ d. Separation,intervener ??Wen hatkeine Studentin [ —— vonden Musikern] getroffen? As (ia) and(ib) show, the German wh-word wen can beseparated fromthe rest ofthe noun phrase it belongs to by movingjust it andnot the whole phrase. The choice between (ia) and(ib) is entirely optional. But when a quantifierlike keine Studentin intervenesbetween thesource position of the phrase and its movement target, the whole phrase has to move:(ic) is acceptable, while(id) is not. Beck suggeststhat at LFtheleftover material vonden Musikern has to move to join the wh-word.This LF movement,she argues,cannot cross anintervening quantifier. There isnointervener blocking theLF rendezvousbetween thetwo partners in (ib), but there is onein (id). Wedonot think that the ECP can bereducedto the effect thatBeck discusses.Epistemic modalsdo not appear to act as blockinginterveners in the structures Beck discusses. (ii) Wen mu¨sste Maria behauptethaben [[ —— vonden Musikern] getroffen zu haben]? whomust Maria claimed have ofthe musicians met tohave ‘Whomust Maria haveclaimed tohave met ofthe musicians?’ Here, theepistemic modal mu¨sste lies onthepath between thefronted wh-wordand the rest ofits phrase. Nevertheless, theseparation does not make thesentence unacceptable,in contrast to (id). Sigrid Beck (personalcommunication) has providedus with a furtherdemonstration that the ECP is notsubsumed by herintervention effects. InBeck 1996b,she showedthat partially reconstructed readings of how many–questionsare notavailable across thekind of barriers she had identified.But again, epistemic modalsdo notdisrupt such readings. (iii)Wieviele Bu ¨cher mu¨sste er derMaria versprochenhaben mitzubringen? how-manybooks must he the Maria promisedhave with-bring ‘Howmany books must he have promised Maria tobring with him?’ Possibleanswer: He musthave promised her to bring six books. 182 KAIVONFINTELANDSABINEIATRIDOU

(32)Every student (in this dorm) is suchthat he mustbe homeif his light is on. (33)We knowof everystudent (in this dorm) that he must be homeif his light is on. (34)?Every student (in this dorm) is so reliable that he must be homeif his light is on. Our predictionappears to be correct: these examples are somewhat stilted but are acceptable. 1 5

3.2The Clause Topology Account Thedata we havediscussed so far arealso compatible with the obvious competitor to the interventioneffect view— namely, a viewthat takes the ECP tobe theresult of thehierarchical structure ofthe clause. One could adopt from Cinque(1999) the idea of a veryrigid and articulated functionalhierarchy in thearchitecture of sentences, and in particularthe commonly held notion thatepistemic modality is located very high in the clause. One could then combine this with the ideaassociated with Beghelli and Stowell (1997) that quantifier movement targets particular positionsin the sentence (rather than being an adjunction process that can target many kinds of projection).If allthe quantifier targets are below the high position for epistemicmodals, we derivethe ECP for monoclausalcases. The fact that wh-dependenciescan cross epistemic modals evenin monoclausal structures would result from thefact that the checking position for wh is higherthan the position of epistemicmodals. What would remain to beexplained is why epistemic modalsare so high in the structure. 1 6

3.3Clause Topology or InterventionEffect? Inthe remainder of this section, we willconsider two facts that may inform our choice between theintervention effect account and the clause topology account. We willlook at (a) antecedent- containeddeletion cases that are in principleproblematic for eitheraccount but perhaps more so for theintervention effect account, and (b) factsinvolving the ECP withepistemic modals that arearguably quite low in the clausal topology, thus constituting cases that are quite problematic for theclause topology account. 3.3.1ACD andthe ECP Antecedent-containeddeletion (ACD) hasrepeatedly been seen as a goodargument for theexistence of LFmovementof quantifiers(Sag 1976, Williams 1977, May 1985,Fiengo and May 1994). It has more recently been used to argue against feature-checking, A-movementaccounts of quantifiers such as Beghelli and Stowell’ s (1997)analysis mentioned insection 3.2 (see Kennedy1997). We havefound that there are cases of ACD thatappear to violate the ECP. We say‘ ‘appear to’’ becausethe judgments are notoriously hard and we havefound it very difficult to makesure

15 An LI reviewer disagrees andfinds these examples justas unacceptableas (17).We don’t knowwhat to say about that. 16 Drubig(2001) offers some intriguingspeculations, but we willnot explore this question further here. EPISTEMICCONTAINMENT 183 thatthe epistemic modal is indeedincluded in the composition of theinterpretation. Nevertheless, thefollowing example seems to be quite good: (35)John thinks that Sarah must have played on everypiano that we hadpredicted he would. For ACD toberesolved, the QP [ everypiano that we had predicted he woulde ]mustLF-move overthe modal must intothe higher clause. Considernow what this means for theECP. Under the clause topology account, (35) should probablybe seen as an example of ACD makingavailable an additional avenue of quantifier interpretation.It violates the locality of feature checking; but as soonas it does this by moving thequantifier into a higherclause, we infact expect the ECP tobe circumvented (since it was nothingmore than the claim that there is no feature-checking position for quantifiersabove an epistemicmodal in theclause of thatmodal). Under the intervention effect view, we needto say thatthe quantifier chains created by ACD aredifferent from thosecreated by ‘ ‘normal’’ QR. Bothviews then need to appeal to theview that ACD isdifferent from normalQR inimportant ways. Thelatter might appear to bea desperatemove. But in fact, ACD-resolving QR (henceforth ACD-QR)iswell known to have different properties from non-ACD-resolvingQR (henceforth Scope-QR). · ACD-QR canmove a QPoutof atensedclause, while Scope-QR cannot. (36)John said that (they wrote that) Mary played on everypiano that we predictedhe would. (37)A different/ Somestudent said that Mary likes every boy. (* every s a different/some) · ACD-QR bleedsCondition C, whileScope-QR does not (Fiengo and May 1994, Fox 2000).

(38)a. ??I reportedhim i to[every cop that John i was afraidof].

b.I reportedhim i to[every cop that John i was afraidI would]. · ACD-QR cancross negation, while Scope-QR cannot. Imagine John, who for yearshas beenaccusing Mary of beinga racistand a sexist.Now thereis an exam coming up and we predictthat John will take this opportunity to makethe same point; that is, we predict thathe will say that Mary will fail women and minority students. In this scenario, the followingsentence seems fine: (39)John said that Mary will not pass every student that we predictedhe would. Inother words, the QP [ everystudent that we predicted he would e ]raisesout of the embeddedclause over sentential negation. On the other hand, Scope-QR cannot easily crosssentential negation. (40)John didn’ t touchevery dessert. (*? every s not) Thesame holds for movementover negative quantifiers. Imagine that John, who took part ina dessertcompetition, is alwaysnegative about certain rival cooks. We predictthat he 184 KAIVONFINTELANDSABINEIATRIDOU

willsay that people don’ t likethe desserts made by these cooks. In this scenario, the followingstatement seems fine: (41)John said that nobody will touch every dessert that we predictedhe would. Onthe other hand, Scope-QR cannot easily cross a negativequantifier. (42)Nobody touched every dessert. (*? every s nobody) Clearly,the differences between ACD-QR andScope-QR are not negligible. In this context, thefact that ACD-QR andScope-QR also differ in their sensitivity to the ECP doesnot appear allthat surprising. In fact, across all these facts we findScope-QR subject to constraints that ACD-QR appearsto be able to violate. We couldthen adopt the attitude that the fact that ACD behavesdifferently from normal scopetaking is aproblemfor allaccounts and hence not one that either of ourcandidate accounts oftheECP isdirectlyresponsible for. 1 7 3.3.2Low Epistemic Modals Inthe spirit of the clause topology account that we havebeen exploring,Drubig (2001) claims that negation cannot have scope over epistemic modals. Relevant examplesinclude these:

(43)a. John must not be athome. ( M ¥ )

b.John may not be at home. ( L ¥ )

cf.John may not leave. (deontic: L ¥ ) Theepistemic modals must and may cannotbe in the scope of negations,whereas deontic may can.We concedethat an attractive account of this proceeds along the lines we haveoutlined above.But we haveto add that it isinfactpossible to findcases where epistemic modal expressions areapparently low enough to be in the scope of negation.Relevant examples include the more ‘‘verbal’’ modal have to.

(44)John does not have to be athome. ( ¥ M )He mightbe atwork. Furthermore,there appear to be perfectlyepistemic modals that are almost specialized as narrow scopeepistemic modals.

(45) a. John need not be home. ( ¥ M )He mightbe at work.

b. John can’t be at home. ( L ¥ )He mustbe at work. If thereare indeed epistemic modals that are low enough to be inthescope of negation, and since quantifierscan have scope over negation (especially if they are in subject position), we can conductan interesting experiment. Consider (46) and (47).

17 Chomsky(2001:23) takes thissituation to indicate that ACD doesnot involve anything like QR butis tobe reducedto some kindof base-generated afterthoughtconstruction followed by VP-ellipsis. EPISTEMICCONTAINMENT 185

(46)Every student can’ t behome.(* ; ¥ L , only ¥ L ;)

(47)Every student need not be home. (* ; ¥ M , only ¥ M ;) As thesesentences show, even when a quantifieroccurs in a clausewith one of the epistemic modalsthat are arguably quite low in theclause topology, the quantifier has narrow scope. This isprima facie an impressive argument for theintervention effect account. Cinque (1999) claims thatthese modals are not really epistemic modals but ‘ ‘alethic’’ onesand that alethic ones are lowenough to be in the scope of negation. We findthis somewhat dubious, but in any case it doesnot help with the analysis of (46)and (47), unless one argues that alethic modals do allow negationon top but that there is neverthelessno positionabove the negation for thequantifier to land. We concludethat the ECP ismost likely an LFinterventioneffect. A quantifier-tracechain cannotcross an epistemic modal. Having established this, one might explore how thisintervention effectcould be derived. There are two possible perspectives to consider:(a) theepistemic modal isunhappy, because it does not tolerate the foot of a quantifier-tracechain below it or does not toleratesuch a chaincrossing it, and (b) thequantifier is unhappy, because the modal does somethingto the chain to disruptit so that the quantifier is unable to maintainits relationship to itstrace. We willnot pursue this interesting question here, wanting merely to lay out the logical space.

4TheECP, A-Movement, QR, andReconstruction We haveshown that when a quantifierand an epistemicmodal are in thesame minimal sentence, theonly possible scopal order permitted by theECP isthe one where the modal has scope over theQP. However,the quantifier of courseoften precedes the epistemic modal at S-Structure,as in (48). (48)Every student must be home. Thismeans that at LF we mustsomehow undo what has happened in (48).There are at least two imaginableways of doingthis. The first is tomove the epistemic modal to apositionabove the S-Structureposition (or LFposition)of thequantifier. The second would be todiminishthe scope ofthequantifier by a mechanismusually called reconstruction. 1 8 Thequestion then arises, Are sentenceslike (48) brought into compliance with the ECP byraising of the modal or by scope diminishmentof thequantifier?

4.1Raising the Modal We willstart with the possibility of raisingthe modal over the QP in(48). There are two ways tolook at this possibility. In the clause topology account introduced in section 3, we couldsay thatthe position where the epistemic modal checks certain relevant (e.g., ‘ ‘epistemic’’) features

18 We willopt for the more neutralterm scopediminishment tostay away fromthe exact nature(syntactic or semantic) anddetails (lowering, interpretation of alowercopy, etc.) ofthese mechanisms. 186 KAIVONFINTELANDSABINEIATRIDOU happensto be higher than the projection(s) where quantifiers end up. So, before everything is saidand done the modal has to raiseover the subject quantifier. In a morestandard QR account, onewould say that the movement of the modal to a higherposition is allowed and then made obligatoryin the relevant examples because without it the ECP wouldrule out the structure. We1 9 areskeptical about the possibility of allowingraising of themodal. One reason is that ifmodaloperators are allowed to move,we wouldexpect to findscope ambiguities between two modaloperators, since one of them might covertly move over the other. It seems that this never happens.Sentences with stacked modal verbs seem to be unambiguous and show only those readingswhere the scopes of theoperators reflect their surface hierarchy. (49)a. I haveto be allowed to graduate. b.#I amallowed to have to graduate. If in(49b) the lower modal were ableto take wide scope over the higher modal, the example wouldhave a sensiblereading equivalent to theone (49a) has. But (49b) only has the rather odd surfacescope reading. This is strong evidence that modals cannot take nonsurface scope. Of course,this might be explained by appropriateconstraints on themovement of modaloperators, andsuch constraints may even come for free inthe right syntactic theory. Also, we should investigatethe empirical facts much more comprehensively before we reachany verdict. If itis true,however, that modal operators engage in scope interaction only with DPs andnever with eachother, then a theorythat does not allow any movement of modals at all could claim the advantageof having a simpleand principled explanation for thisfact. Anotherway in whichraising the modal would be peculiaris thatthe movement would only havethe desired effect of giving a widescope interpretation of themodal if themovement left behindno semantically active trace. If themovement did leave behind a trace,that trace would haveto beof the same type as themodal; otherwise, type mismatches would result. But whenever anexpression leaves behind a traceof the same type as itself, the movement is semantically vacuous,which is notwhat we wanthere. Tosummarize, satisfying the ECP byraising the modal does not seem very promising to us.

4.2Scope Diminishment Scopediminishment is a phenomenondiscussed since May 1977 (see alsoLechner 1997, Lebeaux 1998,Fox 1999, 2000, Boeckx 2001, Elbourne and Sauerland 2002), motivated by the fact that insentences like (50) the indefinite can be interpreted above or below guaranteed.

(50)Someone from New York i isguaranteed t i towin the lottery. At thispoint, we couldhave just suggested that readers pick a favoriteamong the proposals about howto manage to interpret the QP essentiallyin its trace position in (51).

19 Thematerial inthis paragraph and the next is taken from class noteson intensionalsemantics byKai vonFintel andIrene Heim. EPISTEMICCONTAINMENT 187

(51)Every student i must have ti left. Allwe wouldhave needed to add was thatever since Perlmutter 1970 and 1971 and Jackendoff 1972,the idea has been on thetable that epistemic modals involve raising, as evidenced by classic testslike the appearance of expletivesand idioms in subject position. (52)There must have been many chairs in that room. (53)The shit must have hit the fan. However,recently there have been rumors to theeffect that no suchthing as scope diminishment inA-chains exists. We thereforefeel we haveto addresssome of thosepoints. We willtry to do sowhile still remaining neutral about the specifics of scopediminishment mechanisms. In particular,we willaddress the skepticism about scope diminishment in Chomsky 1995 and Lasnik 1999. 4.2.1Chomsky (1995) on the Absence of A-Reconstruction Chomsky(1995) argues that there isno such thing as reconstruction in A-movement —thatreconstruction occurs only in A ¯ -movement.His empiricalarguments are as follows. Consider (54) (Chomsky 1995:327).

(54) *Johni expected[him i toseem to me [t i tobe intelligent]]. Sentence(54) is unacceptablewhen John and him arecoindexed. This is presumablya Condition

Bviolation.If reconstructionto the position of tracet i were possible,Chomsky argues, (54) with coindexationwould not result in a ConditionB violationas the antecedent would be outsidethe bindingdomain of the pronoun, as in (55). (55)John expected [to seem to me [him to be intelligent]]. Chomsky’s secondempirical argument is based on the contrast between (56a –b)and (56c). In (56a–b),the QP cantake scope under negation; in (56c), it cannot. (56)a. (It seemsthat) everyone isn’ t hereyet. b.I expected[everyone not to be thereyet]. c.Everyone seems [t not to be there yet]. InChomsky’ s framework,all this means that there is noA-reconstruction. 2 0 Butthen what about theambiguity of caseslike (57)?

20 Chomsky’s thirdempirical argumentis this.There is reasonto believe that PRO has movedfrom the position marked t towhere itappears in(i). (i)[PRO to appear [tto beintelligent]]is harderthan one might think. Does PROreconstructto t? No,Chomsky says. Note that there isa feelingof ‘ ‘quasiagentivity’ ’ withderived subjects, as in (ii). (ii)John deliberately appears tobe stupid. We findthe same feelingof quasi agentivity in (i). In fact, thisfeeling is impossible to shakein (i) (and (ii)). This quasi agentivityappears tobeasideeffect ofderivedsubjecthood. So, being unable to shake off quasi agentivity in (i) means beingunable to remove (reconstruct)PRO from its derived subject position. Hence, noreconstruction has takenplace. We willnot address thisargument here. 188 KAIVONFINTELANDSABINEIATRIDOU

(57)Someone from New Yorkis guaranteedt towin the lottery. Chomskysays that no reconstructiontakes place here either. What corresponds to thereconstructed reading(i.e., the reading where the DP isinterpretedbelow the matrix predicate) Chomsky says isthe result of a separateprocess of Quantifier Lowering (QL). QL wouldbring the DP notto theposition of its trace but to an IP-adjoined position from whichit can still c-command its Spec,IPtrace. 2 1 We havea potentialparadox: there appear to be noreconstructioneffects for bindingpurposes butthere do appearto bereconstructioneffects for scopepurposes. Chomsky’ s solutionis tosay thatthere is no reconstruction at all and that the apparent scope inversion in (57) is due to the separatephenomenon of QL. Ina way,for ourpurposes all that we needis somethingthat will bringabout scope diminishment; Chomsky’ s rejectionof reconstruction in favor of QL causes ouraccount no harm, as a mechanismfor scopediminishment would still exist. However, for completeness’sake, we wouldlike to make some quick comments. Torebut Chomsky’ s firstargument, we refer thereader to Lebeaux1998 and cite the follow- ingpassage from Wurmbrandand Bobaljik 1999, as itcapturesin few wordsthe conclusion that argumentsfrom ConditionB thatprove or disprovereconstruction in A-chains are unlikely to be found:2 2

‘‘ConditionB effects’’ [(58)]a. *John expected [him to seemto me [t to beintelligent]] Chomsky 1995:327 N Reconstructionof him (tot) wouldavoid a conditionB violationat LF b.*He seems to John/him[t tobeexpected[t ¢ towin]] Lebeaux 1995 N Reconstructionof him (to t¢)wouldavoid a conditionB/ Cviolationat LF Yes, but . . . Lackof reconstruction (or lack of A-trace, as Lasnik suggests) would also yield problems for ConditionB: [(59)]a. *John/ Heseems to me[t to be expected to like him] (vs. himself) b.*He was expected [t to seem to him[t tolike[her]]] [example amended/ KvF,SI] · Forprinciple B, theentire A-chain (base, surface and intermediate positions) of the pronoun arerelevant.

21 Thisis infact themechanism arguedfor by May(1977) for cases like(57). 22 ConditionA is compatiblewith reconstruction.

(i)Pictures of himself i seem toJohn t i tobethebest possible present for Mary. However,as has beenpointed out many times, theproblem is the apparent discrepancy between ConditionsA andB/ C. ConditionB isdiscussed in the main text.With respect toCondition C, ifreconstruction were possible,we couldavoid aConditionC violationin (ii), as thenit would be like(iii).

(ii) *Hei seems toJohn i’s mother ti tobethebest candidate.

(iii)It seems toJohn i’smotherthat he i isthebest candidate. See Fox2000 and references thereinon the Condition A andCondition C arguments. EPISTEMICCONTAINMENT 189

[(58a)]is correctly excluded iff the surface position of thepronoun is relevant. [(59a)]is correctly excluded iff the base position of the pronoun is relevant. [(59b)]is correctly excluded iff the intermediate position of the pronoun is relevant. [(60)]The following assumptions cannot both be true: BindingTheory (Principle B, andmaybe C) holdsonly at LF Pronounsoccupy one and only one position at LF · Whateverthese facts challenge though, they do not pose a specificchallenge to atheorythat countenancesA-reconstruction. I.e., [(59)] is no more a challengeto A-reconstruction than is [(60)]an argumentfor it. · Theobservation is that Conditions B andC seemto require reference to the whole A-chain, i.e.,in terms of Lasnik and Saito 1992 they are ‘ ‘everywhereconditions’ ’ (asopposed to ConditionA whichneeds to be satisfied only once, and perhaps requires c-command at LF). Thisis the upshot of Lebeaux 1995. One may choose to state this as derivational conditions, withoutan intermediatelevel of S-structure with the same effect. Howeverthis is to be implemented,it isorthogonal to thequestion of A-reconstructionif scope istrulya questionof syntactic positions . ..(Wurmbrandand Bobaljik 1999:14 –15)

Withrespect to Chomsky’ s secondargument, based on the ambiguity in (56a –b)and its absencein (56c), one could observe the following. On the one hand, to obtainthe reading where theQP takesscope under negation in (56a– b), a mechanismfor scopediminishment would be needed,since moving negation over the quantifier would have the same problems as discussed insection 4.1 for raisingthe modal. So (56a– b) showthat intraclausal (henceforth, short) scope diminishment)exists. On the other hand, (56c) would show that strong quantifiers cannot undergo interclausal( long)scopediminishment. 4.2.2Lasnik (1999) on theNonexistence of Scope Diminishment Lasnik(1999) claims that there isno scope diminishment in A-chains whatsoever. He attributesthis to the position that A- movementdoes not leave traces, unlike A ¯ -movement,which does leave traces/ copiesand which showsreconstruction effects. We willnot go into the question of howto account for movement withouttraces here. Lasnikpresents arguments from (lackof) reconstructionfor bindingpurposes and (lack of) scopediminishment. We willnot address his examples supporting lack of A-reconstruction for bindingpurposes since we lookedat such cases above. What we willdo isto address (a) Lasnik’s skepticismregarding May’ s originalclaim about lower readings of quantifiers 2 3 and(b) Lasnik’s discoveryof constraintson scope diminishment. 4.2.2.1The Interpretive Evidence for ScopeDiminishment Lasnik writes that ‘ ‘itis not entirely clearprecisely what the phenomenon of QLis’’ (1999:205).He seemsdoubtful about the assump- tionthat paraphrases are a goodtool for bringingout the alleged ambiguity. In this, we haveto agree—atleast to a point.Sometimes one reads discussions according to which the lowered/ reconstructed/dedictoreading is supposed to be representedby aparaphrasethat turns out to be

23 See alsoBoeckx 2001 for a critiqueof Lasnik’ s positionon this point. 190 KAIVONFINTELANDSABINEIATRIDOU justas ambiguousas the target sentence itself. This was alreadynoted by Montague(1973), who said that (61)A unicornappears to be approaching. and (62)It appears that a unicornis approaching. aresynonymous, with two readings each. (Of course,the fact that (62) has a widescope/ dere reading of a unicorn isactually hard to explain (under the assumption that QR doesnot easily reachbeyond finite clause boundaries), making most analysts reach for specialmechanisms like worldindices and choice functions. But that is not the point of ourdebate here, which is instead whether(61) has a lowered/dedicto reading that necessitates some kind of reconstruction. And toestablish this, comparing (61) and (62) is indeed fruitless. As we seeit, the proper argument is this: (a) we needto show that (61) indeed has a de dictoreading (and to say what that means in precise terms), and(b) we needto show that this readingcannot be derived from anLF representationin which the subject has wide scope over themodal operator. The conclusion would be that(61) has a readingthat can only be accounted for usingsome kind of reconstruction. Theeasiest way of constructing the argument is this.We showthat (61) has a readingthat isnot the de re reading.We thenshow that when we forcethe subject to have wide scope, we getonly the de re reading.This means that we needto start off bycharacterizingwhat the de re readingof (61)would be. This can be done via a paraphrase: (63)There is a unicornwhich appears to be approaching. (63)is unambiguous and has the important property that it entails that there is a unicorn.Now, we observethat we arenot forced to read (61) as having that entailment. In other words, (61) butnot (63) could be uttered sincerely by someone who does not believe in the existence of unicorns. Fox’s (2000)example (64)Somebody from New Yorkis likelyto win the lottery. andits unambiguous de re paraphrase (65)There is somebody from New Yorkwho is likely to win the lottery. canalso be used to make the point. (65) can only be uttered about a bizarrelottery: either it is fixed/fraudulentso that there is somebodyfrom New Yorkwho has been predetermined to have agreatchance of winning or thereis somebodyfrom New Yorkwho has bought a largemajority ofthetickets. We observethat (64) can also be uttered in a moreunremarkable scenario where whathappened is simplythat most of the tickets were soldto New Yorkers. We thinkthese considerations establish the first part of the argument, that (61) and (64) havereadings that are not the de re readingsconveyed by (63)and (65). The second part of the EPISTEMICCONTAINMENT 191 argumentcomes from May’s observation.When the subject binds a pronounin an additional argumentof the modal operator, it cannot be seen as having narrow scope inside the clausal argumentof themodal operator. We thenobserve that the resulting sentences no longer have a readingdistinct from thede re reading.Relevant examples of courseare these: (66)A unicornappeared to its owner to be approaching. (67)Somebody from New Yorkseems to himself to be likely to win the lottery. Thesesentences have only de re readings.(66) entails the existence of aunicorn.(67) can only beutteredabout a bizarrekind of lottery(or adeludedNew Yorker). Theconclusion is that(61) and(64) have additional readings because their subjects are not forced to have wide scope. Therefore,there must be amechanismthat gives their subjects narrow scope. Lasnik(p. 206) has this to say about May’ s argument:

[(68)]some professor i isbelieved by his i studentsto be atyrant Itis certainly true that [(68)] cannot be paraphrased as [(69)].

[(69)]*it is believed by his i studentsthat some professor i is a tyrant Butit is not clear what we can conclude from the fact that a sentencecannot be paraphrased by an ungrammaticalsentence (in this case, one that violates the Weak Crossover Constraint). I willthus continueto tentativelyassume that it is nota syntacticoperation that is responsiblefor the ‘ ‘lowered’’ readingof raised indefinites.

Butas far aswe cansee, there is no appeal to (69) in the standard argument. So we don’t see howLasnik’ s pointundermines the argument. We havetried to present the standard view here withoutmaking precise what the de dicto and de re readingsare. This can be (andhas been) done ina preciseand explicit semantic system. Once this is done,the empirical argument we rehearsed hereis strengthened:de dictoreadings are the result of a DPtakingnarrow scope under a modal operator.Since (61) and (64) have de dictoreadings, there must be some kind of reconstruction. Allthis still leaves open the choice among many different ways of givingthe subject narrow scope.It also leaves open the issue of what restrictions there are on which subjects can be ‘‘lowered,’’ towhich we nowturn. 4.2.2.2Restrictions on Scope Diminishment The most important contribution of Lasnik’ s (1999)paper lies in hisdiscussion of theabsence of scope diminishment in certainenvironments. We believe,however, that he draws toostrong a conclusionfrom thedata he presents. We think thathe is absolutely right in claiming that there is no scope diminishment in most of the cases hepresents.However, we donotthink that he can therefore conclude that there is no suchthing asscope diminishment. One cannot point to islands for extractionand conclude that extraction doesnot exist.So we donotthink that his arguments counteract the considerableevidence amassed overthe years that scope diminishment in A-chainsexists. Let us takea closerlook at what we willconsider to be theconstraints on scope diminishment that Lasnik has found. Lasnikshows that negative DPs donot exhibit scope diminishment. The sentences in (70) (Lasnik’s (64)and (65)) lackthe readings in (71) (Lasnik’ s (66)and (67)). 192 KAIVONFINTELANDSABINEIATRIDOU

(70)a. No large Mersenne number was provento be prime. b.No one is certain to solve the problem. (71)a. It was proventhat no large Mersenne number is prime. b.It is certainthat no one will solve the problem. Lasnikalso points out that there is noscopediminishment in (72),as (72)lacks the interpretation that(73) has (his (68) and (69)). (72)Every coin is 3% likely to land heads. (73)It is 3% likely that every coin will land heads. Lasnikmakes the assumptions that (a) aQPheadedby every asin (72) would be expected to undergoscope diminishment, if there were sucha thingas reconstruction, and (b) because likely isaraisingpredicate, 3% likely isalso (and therefore, if there were sucha thingas scopediminishment,it would manifest itself with 3% likely). We agreewith Lasnik: it appears that negative quantifiers cannot undergo short or long scopediminishment. In combination with the ECP, this constraint on scopediminishment makes acertainprediction about sentences with epistemic modals and negative QP subjects:namely, thatthey should be ungrammatical. 2 4 (74)*Nobody may have pushed him. (Maybe he just fell.) (cf. Maybenobody pushed him. Maybe he just fell.) a. may s nobody O K ECP,*Lasnik b. nobody s may *ECP, O K Lasnik (75)*No student may have solved the biology problem. (cf. Itmay be the case that no student solved the biology problem.) (76)*Nobody may be home. (cf. Maybenobody is home.It may be the case that . ..) These,then, are the constraints identified so far: ConstraintA Strongquantifiers can undergo short scope diminishment. They cannot undergo long scope diminishment. ConstraintB Negativequantifiers cannot undergo short or long scope diminishment (modulo footnote 24). Let’s goon to (72) and (73) (Lasnik’ s (68)and (69)). Given constraint A, thesesentences donot yet show anything about the predicate 3% likely. However,even with QPs thatcan in principleundergo long scope diminishment, this predicate indeed blocks scope diminishment. In

24 Itseems thatthese examples improvesomewhat withthe modal must. Sincewe don’t know why negativequantifiers don’t reconstructso easily, we don’t knowwhy sometimes they can reconstruct,apparently back under must. EPISTEMICCONTAINMENT 193

(77),only surface scope is possible; in (78), the pronoun cannot be bound by the universal quantifier;and in (79), a Russian isinterpreted as having wide scope over the elided modal (see Fox2000 for thistest). (77)A soldieris 3% likelyto die in every battle. (78)Someone from hisclass seems 3% likely to every professor to be a genius. (79)An American runner is 3% likely to win a goldmedal and a Russiantoo. Inother words, even with a DPthatcan in principle undergo scope diminishment, 3% likely does notpermit scope diminishment across it. Why should this be? Either 3% likely blocksscope diminishment,or likely isa raisingpredicate but 3% likely isnot. The latter is indeeda possibility. For onething, likely cantake an expletive subject, but 3% likely cannot. (80)There are likely to be many students waiting for you. (81)*There are 3% likely to be manystudents waiting for you. Thesame holds with the other classical test for raising,namely, idioms. (82)Advantage is likelyto have been taken of John. (83)*Advantage is 3%likely to have been taken of John. So perhaps 3% likely isacontrolpredicate but likely isaraisingpredicate, or possibly,ambiguous betweena controland a raisingpredicate. With a controlpredicate, of course,we donot expect scopediminishment. This would explain why scope diminishment is notpossible across 3% likely. Infact, it was Lasnikand Saito (1992) who first proposed that likely isambiguous between controland raising and that how likely behaveslike a controlpredicate. Building on their work, Martin(1992) (as reportedin Boeckx 2001) observes that how likely bringsabout scopally un- ambiguoussentences. In (84), only the someone s likely readingis permitted;that is, no scope diminishmenttakes place. (84)How likelyto win the lottery is someone from New York? So if 3% likely isindeedonly a controlpredicate, then we expectthe absence of scopediminishment.Of course,the question remains why 3% likely couldnot be a raisingpredicate. This is indeeda hardquestion. 2 5 Thereis onething we wouldlike to setthe reader’ s mindat ease about, though.

25 Partlyfor this reason, it has beenproposed that 3% likely is indeeda raisingpredicate butthat scope diminishment is blockedin it. This is theposition taken by Boeckx (2001), who agrees withSauerland (1999) that there isonly one likely andit is always araisingpredicate. He claims that 3% isa quantificationalelement andthe reason that it does not displayscope diminishment is that it blocks the lowering of theQP (forBoeckx, scope diminishment in A-chains is due toactual lowering). Thestatus of (81) is attributedby Boeckx to a similar blockingeffect: thepresence ofa degree headblocks the Agree relationshipbetween theexpletive and its associate. Itis unclearwhat he wouldsay about(83), as hedoes not discusssuch examples. Isitpossible for a quantificationalhead to cause interventioneffects inan Agree relationship? Maybea quantificational/degree element can cause anintervention effect inQL, as thisis compatiblewith the spirit of RelativizedMinimality to which Boeckx attributes his reasoning. But that an intervening quantificational/ degree element 194 KAIVONFINTELANDSABINEIATRIDOU

Itmay not be anattractive solution to saythat 3% likely, unlike likely, isa controlpredicate. However,that may be just how things are. It has proven impossible to tellwhat sets the class of raisingpredicates apart. Of allpropositional predicates, which ones permit a raisingalternative? What,for example,is the difference between likely and possible suchthat the former isa raising predicateand the latter is not? It almost seems like a lexicallyidiosyncratic property. 2 6 And it couldbe, for instance,that the adjective likely, whenit isamount-modifiedlike 3% likely or how likely, is more like possible thanit is like unmodified likely. After all, 3% likely iscloserto the meaning of possible thanit is tothe meaning of likely, whichindicates high probability. Inshort, maybe 3% likely doesnot show scope diminishment because it is not a raising environmentto begin with. 2 7 Tosummarize: There are constraints on A-reconstruction.These constraints are very interesting,and future research will reveal much about the nature of A-chains. However, we haveno reasonto abandonthe position that scope diminishment in A-chains does exist. This means that itispossibleto appeal to scopediminishment to satisfy the ECP incases where the QP linearly precedesthe epistemic modal.

5ScopingMechanisms without Movement

We havesaid that the ECP constrainsthe relation between a quantifierand its trace. We therefore predictthat if there are scoping mechanisms that do notinvolve movement, they should be able toapply across epistemic domains. Recent work on the syntax-semantics interface has indeed exploredseveral such mechanisms. We willlook at twoprominent kinds of cases:(a) pseudoscope indefinitesand (b) genericindefinites.

5.1Pseudoscope Indefinites Here aretwo examples of indefinites that can be read with wide scope for theindefinite even thoughthe indefinite appears in whatis otherwiseassumed to bea domainout of which scope-

shouldcause problemsfor an Agree relationshipdoes not follow straightforwardly from the spirit of Relativized Minimality. Noris it clear thatit actuallydoes. (i)There isalways achair inthis room. Soitis not clear whythe absence ofscopediminishment with 3% likely shouldbe due to a blockingeffect. 26 We wouldlike to note that of themany adjectives thatQuirk et al.(1972) list as occurringin the frame it is A that . . ., onlythree are raisingpredicates. Anything can besaid about such a small class. 27 We wouldlike to add that, even though 3% likely, ifproven to be a controlpredicate, may notbelong to this class, itmay well bethatthere are raisingpredicates thatdo not permit scope diminishment of DPs thatdo otherwise (i.e.,with other predicates) undergoscope diminishment. For example, (i) is ambiguous,but with raising predicates of ‘‘universalforce,’ ’ modifiednumerals seem tous to have only wide scope. That is, scope diminishment is blockedin (ii). (i)At least/At most50 cheetahs are likelyto be born this year. (ii)At least/At most50 cheetahs are certain/guaranteedto be bornthis year. EPISTEMICCONTAINMENT 195 affectingquantifier movement is impossible: (85)Everyone reported that John had insulted Max and some lady. (86)Each teacher overheard the rumor that the dean had summoned a studentof hers. Amuch-exploredapproach to this phenomenon employs choice-functions in interpreting such indefinites. 2 8 Itassumes representations roughly like these: (87) ’f:everyone reported that John had insulted Max and f(lady)

(88)each teacher i ’f: xi overheardthe rumor that the dean had summoned f(student of hers i ) Now, ashintedabove, since this choice-function mechanism does not involve long-distance quantifier movementof theindefinite, we expectit to be able to apply across epistemic modals. (89)Everyone reported that John may have insulted Max and some lady. Thissentence is correctly predicted to havea readingequivalent to ‘ Thereis some lady such that everyonereported that it may be the case that John insulted Max and her’ . Butit achieves that readingwithout moving some lady (or Maxand some lady )acrossthe epistemic modal and out oftheembedded clause —andtherefore, it isnot a counterexampleto theECP. The same applies toa variantof (86). (90)Each teacher overheard the rumor that the dean may have summoned a studentof hers. Thepower of the choice-function mechanism should of course also apply in simpler examples. So,we expecta pseudo-wide-scopereading of the indefinite even in examples where the indefinite isnot buried in some lower clause. (91)John may have invited two friends of his from Texas. Thiswill be interpreted as follows: ‘ Thereis a wayf ofchoosing elements from setssuch that itmay be thecase that John invited the element that f choosesfrom theset of pluralitieswhich aretwo friends of John’s’ . Thisreduces to ‘ Thereis a particularplurality of twofriends of John’s suchthat it may be the case that John invited this plurality’ . Itis crucialthat the reading that the choice-function mechanism produces for (91)be a wide scopecollective reading of twofriends of John’ s. Themechanism cannot produce a widescope distributivereading. Such a readingcould only be producedby treating twofriends of John’s as adistributivequantifier and moving it across the modal, which is ruled out by the ECP. Hence theunacceptability of example(13), repeated here: (92)#Two friends of John’s from Texasmay have come to visithim this weekend, but they can’t bothhave come (because they hate each other).

28 See Reinhart1997a,b, Winter 1997, and Kratzer 1998,among others. 196 KAIVONFINTELANDSABINEIATRIDOU

5.2Generic Indefinites We wouldlike to mentionone other kind of scopingmechanism that does not involve quantifier movement.We assumethat generic readings of indefinitesare produced by havinga covertgeneric operatorbind the variable introduced by an indefinite. 2 9 Theindefinite itself is not a quantifier. So,even if the indefinite needs to moveto ensure that it ends up in therestrictor of the generic operator(which is for examplepart of the analysis of such sentences in Heim 1982), this will notresult in a configurationwhere a quantifierbinds its trace across an epistemic modal. We thereforeexpect generic indefinites to be able to have scope over epistemic modals. (93)a. A dogmay well be asleep when his eyes are open. Dogsmay well be asleep when their eyes are open.

b. Gen [lx. dog x & hisx eyesare open] [ lx.may [x be asleep]] Our expectationseems to be borne out. 3 0 Thispossibility offers anotherway of approximating ina grammaticalfashion the meaning that (17) tries to express but cannot. (94)a. A/ Anystudent whose light is on must be awake. b. Gen [lx.studentx &x’s lightis on] [ lx.must [x be awake]] Here aswell, the generic quantifier binds the variable introduced by the indefinite a student or the‘ ‘free-choice’’ indefinite anystudent. Noquantifier has moved across the epistemic modal. Thesentence is acceptableand expresses something very close to the intended meaning of (17). Whatis different is that this is a genericquantification and not the merely universal quantification that(17) wants to express. 3 1

References Aoun,Joseph, and Yen-hui Audrey Li. 1993. Thesyntax of scope. Cambridge,Mass.: MIT Press. Baker,C. L.1970.Notes on the description of Englishquestions: The role of anabstractquestion morpheme. Foundationsof Language 6:197–219. Beck,Sigrid. 1996a. Quantified structures as barriers for LF movement. NaturalLanguage Semantics 4: 1–56. Beck,Sigrid. 1996b. Wh-constructionsand transparent logical form. Doctoral dissertation, Universita ¨t Tu¨- bingen.

29 Thisanalysis is foundin Heim 1982and many subsequent works. It is ofcoursecontroversial, so we willneed toinvestigate whether our view of theECP is anargument for this approach against one thatbuilds the generic quantifica- tionalforce intothe meaning of the indefinite NP itself.For now, we willsimply assume thatthe account involving a covertgeneric operator is correct. 30 Notethat examples can alsobe constructedwhere genericclaims are inthescope of an epistemic modaloperator. (i)Q: Whyis Chaucer sogrumpy? A:Idon’t knowmuch about dogs. a. Maybedogs do notlike it when you blow in their faces. b.Dogs may notlike it when you blow in theirfaces. 31 Notethat we havenow provided the discussion promised in footnote7. Thisconcludes the article. EPISTEMICCONTAINMENT 197

Beghelli,Filippo, and Tim Stowell.1997. Distributivity and negation: The syntax of each and every. In Waysof scope taking, ed.by AnnaSzabolcsi, 71 –107.Dordrecht: Kluwer. Boeckx,Cedric. 2001. Scope reconstruction and A-movement. NaturalLanguage & LinguisticTheory 19: 503–548. Brennan,Virginia. 1993. Root and epistemic modal auxiliary verbs. Doctoral dissertation, University of Massachusetts,Amherst. Bruening,Benjamin. 2001. QR obeysSuperiority: Frozen scope and ACD. LinguisticInquiry 32:233–273. Chomsky,Noam. 1995. Categories and transformations. In TheMinimalist Program, 219–394.Cambridge, Mass.:MIT Press. Chomsky,Noam. 2001. Beyond explanatory adequacy. MIT OccasionalPapers in Linguistics20. Cambridge, Mass.:MIT, Department of Linguisticsand Philosophy, MITWPL. Cinque,Guglielmo. 1999. Adverbsand functional heads: A cross-linguisticperspective. Oxford:Oxford UniversityPress. Drubig,Hans Bernhard. 2001. On the syntactic form of epistemic modality. Available at http:/ / www.sfb441.uni-tuebingen.de/b2/papers/DrubigModality.pdf. Elbourne,Paul, and Uli Sauerland. 2002. Total reconstruction, PF movement,and derivational order. Linguis- tic Inquiry 33:283–319. Fiengo,Robert, and Robert May. 1994. Indicesand identity. Cambridge,Mass.: MIT Press. vonFintel, Kai, and Sabine Iatridou. 2002. If andwhen if-clausescan restrict quantifiers. Ms., MIT, Cam- bridge,Mass. Available at http:/ /web.mit.edu/fintel/lpw.mich.pdf. Fox,Danny. 1999. Reconstruction, binding theory, and the interpretation of chains. LinguisticInquiry 30: 157–196. Fox,Danny. 2000. Economyand semantic interpretation. Cambridge,Mass.: MIT Press. Heim,Irene. 1982. The semantics of definiteand indefinite noun phrases. Doctoral dissertation, University ofMassachusetts,Amherst. Heim,Irene. 2001. Degree operators and scope. In Audiaturvox sapientia: A festschriftfor Arnim von Stechow, ed.by CarolineFery and Wolfgang Sternefeld, 214 –239.Berlin: Akademie Verlag. Ioup,Georgette. 1975. The treatment of quantifier scope in a transformationalgrammar. Doctoral dissertation, CityUniversity of New York. Jackendoff,Ray. 1972. Semanticinterpretation in generativegrammar. Cambridge,Mass.: MIT Press. Kennedy,Christopher. 1997. Antecedent-contained deletion and the syntax of quantification. Linguistic Inquiry 28:662–688. Kratzer,Angelika. 1978. Semantikder Rede: Kontexttheorie - Modalwo¨rter- Konditionalsa¨tze. Ko¨nigstein/ Taunus:Scriptor. Kratzer,Angelika. 1986. Conditionals. In CLS 22, vol.2, ed.by AnneM. Farley,Peter T. Farley,and Karl- ErikMcCullough, 1 –15.Chicago: University of Chicago, Chicago Linguistic Society. Kratzer,Angelika. 1998. Scope or pseudoscope?Are there wide-scope indefinites? In Eventsand grammar, ed.by Susan Rothstein, 163– 196. Dordrecht: Kluwer. Kratzer,Angelika. 1999. Beyond ouch and oops: How descriptiveand expressive meaning interact. Available athttp:/ /semanticsarchive.net/Archive/WEwNGUyO/. Lasnik,Howard. 1999. Chains of arguments. In Workingminimalism, ed.by Samuel David Epstein and NorbertHornstein, 189– 215. Cambridge, Mass.: MIT Press. Lasnik,Howard, and Mamoru Saito. 1992. Move a. Cambridge,Mass.: MIT Press. Lebeaux,David. 1995. Where does the binding theory apply? In Papersin syntax, syntax-semantics interface, andphonology, ed.by RicardoEchepare and Viola Miglio, 63– 88. University of MarylandWorking Papersin Linguistics3. CollegePark: University of Maryland, Department of Linguistics. Lebeaux,David. 1998. Where does the binding theory apply? Ms. (second draft), NEC ResearchInstitute. Lechner,Winfried. 1997. Two kindsof reconstruction. StudiaLinguistica 52:276–310. 198 KAIVONFINTELANDSABINEIATRIDOU

Leech,Geoffrey. 1971. Meaningand the English verb. London:Longman. Lewis,David. 1975. Adverbs of quantification. In Formalsemantics of natural language, ed.by Edward Keenan,3 –15.Cambridge: Cambridge University Press. Martin,Roger. 1992. On the distribution and content of PRO. Ms.,University of Connecticut, Storrs. May,Robert. 1977. The grammar of quantification. Doctoral dissertation, MIT, Cambridge, Mass. May,Robert. 1985. LogicalForm: Its structure and derivation. Cambridge,Mass.: MIT Press. McDowell,Joyce. 1987. Assertion and modality. Doctoral dissertation, University of SouthernCalifornia, LosAngeles. Musan,Renate. 1997. Onthe temporal interpretation of nounphrases. New York:Garland. Montague,Richard. 1973. The proper treatment of quantification in ordinary English. In Approachesto naturallanguage, ed.by K.J.J.Hintikka,J. M.E. Moravcsik,and P. Suppes,221 –242.Dordrecht: Reidel. Perlmutter,David. 1970. The two verbs begin. In Readingsin English transformational grammar, ed. by RoderickJacobs and Peter S. Rosenbaum,107 –119.Waltham, Mass.: Ginn. Perlmutter,David. 1971. Deepand surface structure constraints in syntax. New York:Holt, Rinehart and Winston. Pesetsky,David. 2000. Phrasalmovement and its kin. Cambridge,Mass.: MIT Press. Quirk,Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1972. Agrammarof contemporary English. London:Longman. Reinhart,Tanya. 1997a. Quantifier scope: How laboris divided between QR andchoice functions. Linguistics andPhilosophy 20:335–397. Reinhart,Tanya. 1997b. Wh-in-situin theframework of the Minimalist Program. NaturalLanguage Seman- tics 6:29–56. Rodman,Robert. 1976. Scope phenomena, movement ‘ ‘transformations,’’ andrelative clauses. In Montague Grammar, ed.by Barbara Partee, 165 –176.New York:Academic Press. Sag,Ivan. 1976. Deletion and Logical Form. Doctoral dissertation, MIT, Cambridge, Mass. Sauerland,Uli. 1999. Scope reconstruction without reconstruction. In Proceedingsof the17th West Coast Conferenceon FormalLinguistics, ed.byKimaryShahin, Susan Blake, and Eun-Sook Kim, 582 –596. Stanford,Calif.: CSLI Publications. Szabolcsi,Anna, ed. 1997. Waysof scopetaking. Dordrecht:Kluwer. Tsai,Wei-Tien. 1994. On economizing the theory of A-bar dependencies. Doctoral dissertation, MIT, Cam- bridge,Mass. Williams,Edwin. 1977. Discourse and Logical Form. LinguisticInquiry 8:101–139. Winter,Yoad. 1997. Choice functions and the scopal semantics of indefinites. Linguisticsand Philosophy 20:399–467. Wurmbrand,Susi, and Jonathan David Bobaljik. 1999. Modals, raising, and A-reconstruction. Handout of atalkgiven at Leiden University (October 1999) and at the University of Salzburg (December 1999).Available at http:/ /www.arts.mcgill.ca/programs/linguistics/faculty/wurmbrand/research/files/ Leiden-1999.pdf.

Departmentof Linguistics and Philosophy MIT E39-245,77 MassachusettsAvenue Cambridge,Massachusetts 02139 [email protected] http://web.mit.edu /fintel [email protected]