Weak Crossover As a Scope Phenomenon

Weak Crossover as aScope Phenomenon E. G. Ruys

Thisarticle investigates the proper characterization of the condition thatis responsible for weak crossover effects. It arguesthat the relevant conditionbelongs to scope theory and that weak crossover arises from theway in which scope is determined in syntax. This implies that weakcrossover can occur whenever an operator must take scope over apronoun,even when the pronoun and the operator are not coindexed andthe intended interpretation of the pronoun is not as a variable boundby the operator. It also implies that, when an operator is for somereason assigned scope in an exceptional manner and escapes theusual syntactic restrictions on scope assignment, bound variable licensingwill be exceptionally allowed as well. Keywords: weakcrossover, scope, variables, binding

1Introduction Considerthe examples of strong crossover (SCO) andweak crossover (WCO) in(1).

(1) a. Whoi ti likes hisi mother?

b. *Whoi does hei like ti ?

c. ??Whoi does hisi motherlike t i ? In(1a) itis possible for thepronoun to be interpreted as a variablebound by the wh-operator, yieldingthe interpretation ‘ for whichperson x, x likes x’smother’. In(1b) and (1c) asimilar readingis unavailable. At firstglance the last two examples resemble (2) (see Chomsky1977), whichalso excludes a boundvariable reading for thepronoun.

(2) ??Everysoldier i hasa gun.But will he i shoot? However,(2) isusuallygiven a differenttreatment than the examples in (1).The operator in (2) doesnot take scope over the pronoun, so the unavailability of thebound reading in this case may simplybe attributed to the fact that, as a matterof logic, operators can only bind variables in theirscope. In (1), on theother hand, the operators c-command the pronouns. By currenttheories ofscopeassignment, this means that the operators do take scope over the pronouns. The absence

Iwouldlike to thank Jan Don,Johan Kerstens, Iris Mulders,Ad Neeleman, BillPhilip, Tanya Reinhart, Maaike Schoorlemmer,Henk Verkuyl, Fred Weerman, YoadWinter, audiences at Utrecht University,Tilburg University, Potsdam University,and CUNY, andanonymous LI referees forhelpful comments andsuggestions. All remaining errors are my own.

Linguistic Inquiry, Volume 31, Number 3,Summer 2000 513–539 q 2000 bythe Massachusetts Institute ofTechnology 513 514 E . G . R U Y S ofvariablebinding in (1b–c) thusappears to revealthat a boundvariable pronoun not only must bein the scope of the operator, but also is subject to some additional, presumably syntactic constraint. Thesyntactic generalization governing the facts in (1) hasoften been characterized as follows: pronounsthat are coindexed with operators must be locally A-bound, or alternatively,operators mustnot locally A ¯ -bindpronouns. Many principles of grammar havebeen proposed that express thisbasic generalization. It is well known, however, that this generalization does not cover the fullrange of relevant examples; various refinements have been introduced to dealwith different typesof counterexamples. Below, I willdiscuss several classes of such counterexamples. My investigationof these nonstandard cases will lead to a generalconclusion regarding the proper characterizationof crossover,which I willoutline here. Mostanalyses of crossover assume that an operator and a pronounare subject to WCO only if theyare coindexed. For instance,(1b) and (1c) areill formed only on the intended reading indicatedby the indexing subscripts; otherwise, the operator is free tocross over the pronoun. Hence,crossover conditions are usually formulated as applying to coindexed operator-pronoun pairs.Such conditions typically make use of notions borrowed from someexisting module of grammar thatemploys the coindexing relation: binding theory or, sometimes, movement theory. 1 Insome cases the crossover condition is actually incorporated into binding theory. 2 For ease of reference,I willrefer toall such conditions as binding-theoretic conditions. Below,I willpresent several classes of exampleswhere an operator and a pronounare not coindexed,but we stillfind a ‘‘WCO configuration’’— inthe sense that a particularreading of thesentence is available only if the operator c-commands the pronoun from anA-position, and isunavailable if theoperator merely c-commands the pronoun from anA ¯ -position.I willargue thatthis is thecasewhenever that reading depends on the operator’s takingscope over the pronoun. Naturally,one might claim that in allsuch examples the operator and the pronoun really are coindexed.But the question then arises how we canprovide independent evidence that a pronoun andan operator bear the same index in a givenstructure. In the relevant examples the pronoun isnotinterpreted as a variablebound by the operator: they are not coreferential or ‘‘covalued.’’ Nonetheless,one might still claim that some coindexing relation obtains, but that the relevant coindexingrelation does not denote coreference or covaluation. Of course,this is a gratuitous move,unless one can demonstrate the presence of these indices by some other means— for instance,by showing that they are visible to Condition A, B,orCofbinding theory. I willargue

1 Someanalyses thatreturn to a movementtheory perspective of crossover(see Postal1971) are May’s (1985)path- theoreticaccount, Ruys’ s (1994b)analysis that derives WCO fromChomsky’ s (1993)Shortest Link, and the analysis proposedby Georgopoulos(1991). See Ruys1992 for a discussionof May1985 and other references mentionedin the text. 2 Forexample, Reinhart 1983, 1987, Koopman and Sportiche 1982, Ha ¨õ k1984,Safir 1984, 1996, Stowell, to appear. Someauthors, notably Higginbotham (1980, 1983), have described the phenomenon in terms ofconditionsand relations thatare onlyindirectly related tobinding theory (e.g., Higginbotham’ s Dependenceand the associated Accessibility Condition).The problem with such an approach is that it begs the question of whythe proposed condition has theproperties itis defined to have, and why it applies to the class ofrelationsit issaidto apply to. I willreturn to such binding theory extensionsbelow. WEAKCROSSOVERASASCOPEPHENOMENON 515 thata WCOconfigurationsometimes obtains between a pronounand an operator that are not ‘‘related’’ inany way (coreferentiality, binding theory) except that the operator, in the intended reading,takes scope over the pronoun. Iwillthus argue that the binding-theoretic approach faces severe empirical inadequacies. I proposeinstead that the condition responsible for WCObelongsto scope theory: both (1b –c) and(2) reflecta failureon the part of the operator to take scope over the pronoun. I willargue thatthis view holds the promise of aprincipleddefinition of the class of pronoun-operator pairs thatare subject to WCO. 3 Somany different ‘ ‘weakcrossover conditions’ ’ havebeen proposed in the literature that spacedoes not permit me to discuss each of themseparately. Indeed, this is not strictly necessary, asI aimto argue against whole classes of weak crossover conditions. The exact way in which myarguments apply to each particular implementation known from theliterature must be leftfor thereader to verify (see, however, footnote 21, and Ruys 1992:chap. 4). Inorder to facilitate the discussion, I willformulate two ‘ ‘straw man’’ theoriesof thekind Iwillargue against. Two such straw menare needed, as the class of analyses I argueagainst dividesinto two separate subclasses that are affected by my arguments in different ways. These straw mantheories are stated in (3). (3) a. Binding-theoreticlicensing principle PronounB maybe interpreted as a variablebound by A onlyif A A-bindsB. b. Binding-theoreticconfigurational principle PronounB maynot be locallyA ¯ -bound. Twotypes of crossover principles are distinguished here, which I willrefer toas licensing principles and configurational principles.Licensing principles, in this sense, state that for apronoun tobe interpretable as a boundvariable, it mustenter into a licensingrelation with some (operator) expressionat some stage in the derivation. In this manner, the bound variable interpretation is ruled in undercertain conditions. This category includes the proposals in Higginbotham 1980, 1983and Reinhart 1976, 1983. The licensing principle (3a) isbasically Reinhart’ s condition. Configurationalprinciples, on the other hand, state that a pronounmay not enter into a given relationshipwith any (operator)expression at a givenstage in the derivation. In this manner, structurescontaining a boundpronoun are ruled out undercertain conditions. This category in- cludesthe proposals in Koopman and Sportiche 1982 and Safir 1984.The configurational principle (3b)is basicallyKoopman and Sportiche’ s BijectionPrinciple. Whatthe two conditions in (3) havein commonis thata pronounand an operator must obey them,on pain of producing a WCO violation,just in case the pronoun and the operator are

3 As forthe WCO/ SCOdistinction, there is disagreement intheliterature about whether these phenomenashould begivena unifiedaccount. Reinhart (1983), for example, excludes (1b) and (1c) with the same condition.The Bijection Principle(Koopman and Sportiche 1982), on the other hand, does not exclude (1b); this example isinsteadconsidered tobe aConditionC violation.The crossover account to be developed here rulesout WCO andSCO inthe same way. ButSCO configurationspresumably display a ConditionC effect inaddition to the crossover effect. Althoughnothing hingeson this,I willrestrict attentionto WCO configurationsso as topreempt any interfering Condition C effects. 516 E . G . R U Y S coindexed.I argueinstead that WCO asitobtainsin (1),where the pronoun and the operator are coindexed,is just a specialcase of amoregeneral phenomenon. WCO arisesin all cases where anoperator needs to take scope over a pronoun,but fails to c-command it from anA-position. Again,for easeof exposition, I willformulate a conditionthat implements this, although in a verysimple way (see section3 for somediscussion of detailsof implementation). 4 (4) ScopeLicensing Aissyntacticallylicensed to take scope over B iff a.A c-commandsB, Banoperator; or b.A c-commandsB from anA-position. Iwillbriefly illustrate how condition (4) appliesto the matters at hand.Clause (4a) recapitulates May’s (1977)Scope Principle. It accounts for theambiguity of (5a) (after QuantifierRaising (QR) everyone c-commands someone in(5b), or viceversa in (5c)) andfor thenonambiguity of (6a) (for every man x totake scope over awomanthat x loves, itmust c-command that phrase; hence,QR mustviolate the Complex NP Constraintin (6b)). (5) a.Someone loves everyone.

b. [IP everyonej [IP someonei [IP ti loves tj ]]]

c. [IP someonei [IP everyonej [IP ti loves tj ]]] (6) a.I knowa womanthat every man loves.

b. *[IP every mani [IP I know [NP a woman [CP that ti loves]]]] Clause(4b) is theinteresting part of thecondition: it is responsiblefor WCO. Letus seehow it accountsfor regularexamples of WCO.

(7) a.Everyone i loves hisi mother.

b. ??Hisi motherloves everyone i .

(8) a. Whoi ti loves hisi mother?

b. ??Whoi does hisi motherlove t i ? Assumethat for apronounB tobe interpreted as a variablebound by an operator A, AandB mustbe coindexed, and A musthave scope over B. Thenin (7a) his maybe interpreted as a variablebound by everyone:everyone c-commands his from anA-position in accordance with (4b),hence may take scope over it. In (7b) everyone willc-command his from an A¯ -positionat best(after QR), henceis not licensed to takescope over it by(4b).This prevents a boundvariable readingfor thepronoun. 5 Similarly,in the well-formed case (8a) theoperator who maytake scope

4 Notethat (4) is a licensingprinciple in the sense explicatedabove. Elsewhere (Ruys,in preparation)I arguethat ‘‘weakest crossover’’ phenomena(Lasnik and Stowell 1991) indicate that this will ultimately turn out to be thecorrect approach(see alsoHornstein 1995). 5 Already,issues ofimplementation present themselves. We mustnot wind up preventing everyone in (7a) from takingscope over the pronoun when it movesto an A ¯ -positionat LF.Thisquestion touches on an unresolved issue in thetheory of scope. Various facts indicatethat scope is determinednot only by operator positions at LF,butalso by theirpositions at previousstages inthe derivation. Aoun and Li (1989), for example, attempted to solve this problem by allowingtheir scope principle to refer totrace positions.Elsewhere (Ruys1992) I haverejected thisview (partly for WEAKCROSSOVERASASCOPEPHENOMENON 517 overthe pronoun by clause (4b) since it (or itstrace, see footnote 5) c-commands his from an A-position,whereas in (8b) the operator is not licensed to take scope over the pronoun, so that thebound variable reading cannot obtain. How exactlydoes this account differ from binding-theoreticaccounts such as (3)? Thepresent accountassumes that for apronounB tobe interpreted as a variablebound by an operator A, twothings are required: A andB mustbe coindexed, and A musthave scope over B. Binding- theoreticaccounts assume that A andB mustbe coindexed, A musthave scope over B, 6 and someadditional constraint such as (3) mustbe met. Instead of postulating such an additional constraintthat refers tothe coindexing relation, the present account describes WCO througha complicationin the definition of scope,namely, clause (4b). Binding-theoretic approaches typi- callyemploy a muchsimpler definition of scope taking (e.g., c-command at LF). Inthe following section I willreview a rangeof observationsthat can be arguedto support ascope-theoreticapproach to WCO. Inmost cases (sections 2.1 –2.4)I willargue that WCO obtainseven though there is nobinding-theoretic(coindexing) relation between the expressions involved,only a scoperelation. In one other case (section 2.5) I willshow that where relative scopeis exceptionallydetermined by other factors than the usual c-command condition laid down in(4), so isvariablebinding; exceptional scope and exceptional variable binding go handin hand.

2FiveArguments fora ScopeTheory of WCO 2.1Transitivity Thefollowing well-known paradigm (see Higginbotham1980) exemplifies what I callin Ruys 1992 the transitivity propertyof bound variable licensing: 7

(9) a. [a Every boy’si mother]loves him i .

b.[[Every boy’ s i mother’s] husband]loves him i .

reasons alsonoted in Aounand Li 1993), proposing instead that licensing in the sense of(4) may take place at anypoint inthe derivation. For the present I willstick to this view. In any case thedistinction between (7a)and (7b) remains clear: in(7b) no element ofthechain of everyone c-commands his froman A-position at anystage inthe derivation. See section 3forfurther discussion, in particular footnote 23. 6 As pointedout above, it is naturalto assume, beitas apropertyof syntax or semantics, thata pronounmay only beinterpretedas avariablebound by a quantifiedexpression if itis inthe scope of that expression. It is importantto notethat this assumption is not unique to my approach, but is oneit has incommon with virtually all theoriesof bound anaphora,including those that attribute WCO toa binding-theoreticprinciple such as (3).Accounts of boundanaphora generallyrely on this(usually tacit) assumption;it is sometimes knownas theScope Condition (Koopman and Sportiche 1982).For instance, (2) is not excluded by any configurational WCO conditionlike (3b) and is generally attributed to a scopefailure. And although the licensing condition (3a) happens to exclude the bound reading in (2), it must (when taken toapply at S-Structure)resort to the Scope Condition to explain the nonambiguity of (i).

(i)Every man i lovessome paintingin his i house. See Ruys1992 for an overview of variousWCO hypothesesand their reliance onthisScope Condition. As for(i), condition (4) allows every man totake scopeover the pronoun; but if some paintingin his house then takes widescope, scope relations will be contradictory.The reverse holdsin Somepiece youwant him to will be played byevery musician, where every musician may take scopeover the subject (by (4a)) but not over the pronoun (by (4b)); hence,the bound reading is excluded. 7 Alsosee Ruys1992 for a review ofpreviousanalyses ofthetransitivity effect. 518 E . G . R U Y S

(10) a. [a Which boy’si mother]j tj loves himi ?

b. [a Whosei mother]j tj loves himi ?

(11) [a Someonein every city i] hates iti .

(12) [a Whichpicture of whichman i ]j tj pleases himi ? Thesedata indicate that an operator-type expression need not A-bind a pronounin order to variable-bindit. If theoperator is embeddedin a largerconstituent a ,eitheras a complementor asaspecifier,the pronoun may still be interpreted as a boundvariable. Thereare just two provisos. First, of course, the operator must take wide scope with respect to a (presumablyby wayof inverse linking, as indicated in (9a ¢ )); whenthe wide scope reading isblocked,as in (13), the bound reading is alsoblocked for obviousreasons.

(9) a¢ .Everyboy’ s i [ti mother]loves him i .

(13)??Every boy’ s i bestfriend, Baden-Powell, loves him i . Second,the transitivity effect occurs only in case a isitself in an A-position and c-commands thepronoun (or, as in (9b),is itselfcontained in yeta largerconstituent that obeys this condition). If thecontaining constituent a isnot in a c-commandingA-position, WCO obtainsbetween it andthe pronoun, and variable binding fails.

(14)a. ??His i fatherloves [ a every boy’si mother].

b. ??[a Whosei father]j does hisi motherhate t j ?

(15)a. ??Its i mayorloves [ a someonein every city i ].

b. ??[a Whichpicture of which man i ]j did hisi agent sell tj? How canthis regularity be explained? Most authors agree that a in(9)– (12) is in some sense actingas a licenserfor thebound pronoun by proxy. Althoughthe bound pronoun must be licensed bysome c-commanding element in an A-position, this need not be the binding operator itself; someother constituent may substitute for theoperator. 8 Butin none of the examples above is therea bindingor coindexingrelation between the proxy constituent a andthe pronoun. Hence, the pair k a , pronounl cannotviolate or satisfy a WCOconditionin terms of coindexing. As a result,the binding-theoretic conditions (3) failto distinguishthe well-formed examples in (9)–(12) from theill-formed ones in (14)– (15). Binding-theoretictreatments of WCOneedto invokesome extension of thenotionof coindex- ingto capture the transitivity effect. But such attempts to incorporate the effect into binding theoryare self-defeating, as they must be accompanied by a correspondingrevision to exempt theusual binding conditions from transitivity.If bindingtheory allowed a constituentto act as

8 Reinhart(1983, 1987) takes adifferentapproach, denying the relevance of a andclaiming that the antecedent itself licenses thepronoun. Specifically, she proposesthat B is boundby A onlyif A c-commands B,orA is thespecifier ofC,andC meets thiscondition. This accounts (only) for (9) and (10), but (16) remains aproblem. WEAKCROSSOVERASASCOPEPHENOMENON 519 proxyfor adistinct,noncoindexed constituent, we wouldtrivially expect (16a) to be wellformed and(16b –c)tobe ill formed. 9

(16) a. *[a Every boy’si mother]loves himself i. b. [a Every boy’si mother]loves him i . (4 (9a)) c. [a Every boy’si mother]loves the little rascal i . Safir (1984),for instance,proposes that in transitivity cases the index of theoperator is attached tothe index of a , allowing a (or itstrace) to A-bind the pronoun in accordancewith (3) (theQ- ChainConvention). But these added indices must somehow be rendered invisible for allother purposes:they play no rolein the binding conditions, witness (16), nor do theyreceive the same semanticinterpretation as regular indices. The same is true of Higginbotham’s (1980,1983) more straightforwardanalysis. There, the licensing relation between the antecedent, a ,andthe pronoun isnot one of coindexing (or linking),but one of accessibilitythrough a V-chain,a notionspecifi- callydesigned to describe such facts. This avoids any unwelcome predictions about (16), but raisesthe same question as Safir’ s solution.Both solutions postulate a relationbetween the antecedentoperator and a ,andbetween a andthe pronoun, which appears to be both syntactically andsemantically vacuous: it merely serves to describe the transitivity effect. If thepostulated relation,be it accessibility through a V-chainor index sharing through a Q-chain,is not the familiarcoindexing/ coreferencerelation, then what, if anything, does the relation mean, and how canit be independentlyestablished? Inother words, if the transitivity effect observed above is a distinctiveproperty of bound variablelicensing, then we areunlikely to be successfulin attempting to reduce bound variable licensingto bindingtheory, which predicts that this effect will not obtain. 10 Instead,it wouldbe preferableto derive the transitivity effect from somecomponent of the grammar thatnaturally relates a tothe antecedent operator and the pronoun. Scope theory is an obvious candidate.

9 Note,incidentally, that the considerations presented here detract fromKayne’ s (1994)claim that(9) and (10) evidencean adjoined position for specifiers; thisclaim leaves (16)unexplained, and it makes anunwarranted distinction between (9)–(10) and (11)– (12), which my analysis does not. 10 Obviously,we cannotclaim thatit is strictlyimpossible to formulate an extension of bindingtheory that allows forA-binding in the transitivity cases whilestill deriving the correct judgmentsin (16). Reinhart (1987), for example, proposesthat Condition A states thatan anaphor and its antecedent must be containedin the same minimalgoverning category(MGC). This correctly disallows (16a); a similar reformulationof ConditionB willcorrectly allow (16b). This particularreformulation ran into trouble vis-a `-visexample (i),

(i) [S1 I believe [S2 Johni tolike himself i ]]. where MGC(John) 4 S1 ? S2 4 MGC(himself),butthe structure is well formednevertheless (similarly for Condition B).Also, finding a reformulationof ConditionC thatdistinguishes (iia) from(iib) while maintaining that the direct object is A-boundin both cases willpresumably prove more difficult.

(ii)a. *Everyboy i likesthe little rascal i . b.Every boy’ s i motherlikes the little rascal i. Similarproblems arise withSafir’ s (1984)reformulation of bindingtheory (see Ruys1992 for discussion). Nonethe- less, theargument cannot be thata descriptivelyadequate incorporation of conditionson variable binding into binding theoryis impossible in principle. But the complex immunizations this would involve are evidencethat the conditions on variablebinding cannot be shown to be related to,let alonederivable from, the conditions on obligatory coreference and disjointreference. 520 E . G . R U Y S

Inthe well-formed examples (9)– (12), clause (4b) allows a totake scope over the pronoun —a predictionthat is known to be correct. The antecedent operator in turn may take scope over a byclause(4a) afterQR —alsoa well-knownfact. 11 Finally,relations of relativescope are transitive ingeneral,so we expectthat the antecedentoperator may take scope over the pronoun and nothing barsa boundvariable interpretation. In the ill-formed examples (14) and (15), on theother hand, neitherthe antecedent nor its container a islicensed to take scope over the pronoun. Allowing thesejudgments to follow from scopetheory should therefore be entirelystraightforward. Thereare various ways of implementing this analysis. One option is to add to (4) athird clausestating that scope is transitive.However, it may be morenatural to assume that transitivity ofscopefollows directly from propertiesof thesemantic component. We needto assume, then, thatsuch inherent properties of the semantic component may license a relativescope ordering whenit isnot syntactically licensed; we willsee another possible example of this below. In any casereducing bound anaphora to scope theory will allow us to explain the observations in this section.

2.2Donkey Anaphora Donkeyanaphora presents a secondexample of a syntacticrelation that is subject to WCO but thatcannot be described in terms of bindingor coindexing. Consider the contrast in (17).

(17) a. [a Everyfarmer whoowns a donkey i ] beats iti .

b. *Itsi former ownerenvies [ a everyfarmer whoowns a donkey i ]. Thecoindexed pair it – a donkey cannotsatisfy the WCO conditionin (17a) and violate it in (17b),given that this condition involves some form ofc-command.Instead, as in the case of the transitivityexamples discussed above, the contrast appears to be due to some relation between thepronoun and a ,thequantified NP containing a donkey. Inview of the prevalent semantic analysesof donkeyanaphora, this is predicted by the scope-theoretic approach to WCO. A donkey in(17)does not take scope over the pronoun in eitherexample. This must be the caseon syntacticgrounds, since a donkey cannotbe extractedfrom thecomplex NP thatcontains it.Furthermore, assigning a donkey widescope so as toallow it to bind the pronoun would not resultin the correct semantic representation; a donkey isinterpreted in the scope of the universal quantifier every, insideits restrictive clause. Hence, the scope-theoretic approach predicts, correctly,that the relation between the pronoun and a donkey isirrelevant to WCO. Oneprevalent analysis of donkey anaphora says that it in(17a) is interpretedas a variable thatis unselectively bound by the universal quantifier, resulting in auniversalquantification over farmer/donkeypairs (see Lewis1975, Heim 1982, Kamp 1981, Ruys 1992, Kamp and Reyle 1993,and references cited there). Another influential view is that it isan ‘ ‘E-type’’ pronoun, whichis interpretedroughly as a definiteNP equivalentto thedonkey that x owns, where x is a variablebound by everyfarmer who owns a donkey (see Evans1977, Cooper 1979a, Heim 1990,

11 Assumingthat a countsas anoperator in terms of(4a); see May1985 for arguments that QR is possiblefor such expressions. WEAKCROSSOVERASASCOPEPHENOMENON 521 andreferences cited there). There is no need to discuss the relative merits of these views here; itis sufficientto notethat both types of analysisrequire the larger constituent a thatcontains the antecedentto take scope over (the interpretation of) theanaphor, either to bind it as a variable orto bind a variablecontained in it. Hence, the scope-theoretic approach to WCO predicts, correctly,that the pair k a , pronounl issubject to the WCO condition. Therequirement that the containing constituent must have scope over the pronoun is confirmedby the absence of donkeyanaphora in (18a) (May 1985:chap. 3, (40)),(18b) (Ha ¨õ k 1984: (56)),and (18c), and by theabsence of scopeambiguity in (18d).

(18)a. *Your shouting at every owner of a donkey i frightenedit i.

b.*Everyone who owns a donkey i came,and Mary bought it i .

c.*Every farmer whoowns a donkey i feeds iti , but will iti grow?

d.Every farmer whoowns a donkey i hatessome of its i habits. Inspectionof thesyntactic contexts that allow donkey anaphora shows that the anaphor and thecontainer of itsantecedent must meet the WCO conditionexactly. (19) (Reinhart 1987:(37b)) and(17b) violate WCO.

(19) *Heri mothervisited every knight who courted a lady i . Thesame is truein case the container is a wh-element,as the contrast in (20) shows.

(20)a. Which man who owns a donkey i hates itsi former owner?

b.*Which man who owns a donkey i does itsi former ownerhate? Andagain, we findthe transitivity effect, in (21a) (May 1985:chap. 3, (41)) and(21b).

(21)a. A friendof every owner of a donkey i beats iti .

b.Someone in every city with a cathedral i hates iti . Thereis of course no syntacticbinding relation between the donkeyanaphor and the container a ofitsantecedent. Consequently, binding-based licensing theories of boundanaphora (3a) would indiscriminatelyrule out all examples cited above, and binding-based configurational conditions (3b)would rule them all in (except (18), given the Scope Condition; see footnote 6). Again, as inthe case of the transitivity effect, binding-based theories of bound anaphora have been aug- mentedwith index-copying mechanisms that induce some type of binding relation between the pronounand the antecedent-containing constituent (see, e.g., Ha ¨õ k1984,Reinhart 1987). But this cannotlead to a realunification of binding and bound variable licensing, as these additional mechanismsmust be voided of anypossible effects other than those needed for boundvariable licensing,or we wouldexpect (17a) and (22a) to be illformed, and (22b) to be well formed. 12

(22)a. Every farmer whoowns a donkey i beatsthe stupid animal i .

b.*Every farmer whoowns a donkey i beatsitself i .

12 See Ruys1992 for a review ofpreviousanalyses ofWCO withdonkey anaphora. 522 E . G . R U Y S

Althoughthere is no syntactic binding between the donkey anaphor and its licenser, there canbe littledoubt in viewof the semantics of theseconstructions that the container of the donkey antecedentmust be able to take scope over the anaphor for donkeyanaphora to be possible (as confirmedby (18)). Hence, my reduction of bound variable licensing to scope theory predicts thewell-formedness of (17a),(20a), and (21), and the WCO effectin (17b), (19), and (20b) on aparwith normal cases of boundanaphora. The implementation of this reduction in (4) applies straightforwardly:in thewell-formed examples the licenser a c-commandsthe pronoun from an A-positionin conformity with (4b); in the ill-formed examples it fails to do so.

2.3Pronouns of Laziness We havetwo more classes of examplesto discusswhere WCO obtainsin astructurethat involves scopedependency, but not coindexing. This section deals with pronouns of laziness,exemplified in (23). (23)The man who gave his paycheck to his wife was wiser thanthe man who gave it to hismistress. (23)is apaychecksentence (see Karttunen1969), with a readingfor it asa pronounof laziness; onthis reading it isinterpretedjust as a copyof hispaycheck. Toset the stage, consider first some examples of sloppyidentity in VP-deletion contexts. It iswellknown that sloppy identity requires that the pronounin theantecedent clause be interpreted asaboundvariable.

(24)a. John loves his mother, and Peter does too [ VP e]. b.The woman who emulated Harry believeshe is intelligent and the woman who

emulatedBill does too [ VP e]. (24a)allows a sloppyreading (Peter loveshis own mother), but (24b) does not, because (the l - operatorassociated with) Harry cannotvariable-bind he (Lasnik1976). Variable binding in these constructionsobeys the usual WCO conditions(Reinhart 1983). Thesame conditions hold for theantecedent of a pronounof laziness;a pronouncan be a pronounof lazinessonly if thepronoun contained in its antecedent can in turn be interpretedas aboundvariable. (25)a. Every boy under ten loves his cousin, and every boy over ten wants to marry her. b.His cousinloves every boy under ten, and every boy over ten wants to marry her. c.When every boy was atschool I talkedabout his cousin, and every boy bought her flowers. Thisexplains why (25a) is a paychecksentence, but (25b –c) arenot; that would require variable bindinginto the antecedent his cousin, violatingWCO in(25b) and the Scope Condition (see footnote6) in (25c). Sofar thereis littleto indicate whether the structural relation needed for sloppinessis more likebinding or scope.Things change when we observethat the conditions on variable licensing WEAKCROSSOVERASASCOPEPHENOMENON 523 mustbe metnot only in the antecedent for apronounof laziness,but also in thecopy. Consider (26)and (27). (26)Every boy under ten loves his cousin, but because every boy over ten is so mature, sheis unpopular. (27)a. Every boy under ten loves his cousin, but she hates every boy over ten. b.Every boy under ten loves his cousin, but her husband hates every boy over ten. In(25b –c) theantecedent fails to meet the conditions on bound variable licensing. In (26) and (27),however, only the intended pronoun of lazinessfails to meet these conditions, and yet these examplesdo not allow a lazinessreading ( she, her canonly be deictic). In (26) everyboy over ten cannottake scope out of theadjunct clause; (27a –b)reveala crossovereffect. Thefollowing examples show that the transitivity effect occurs with pronouns of laziness aswell, both in the antecedent clause and in the clause containing the pronoun of laziness: (28)a. Every boy’ s motherloves his girlfriends, and every boy’ s fatherhates them. b.Someone in everyItalian city hates its cathedrals, and someone in every American cityloves them. Itappears, then, that the configuration required for boundvariable licensing must hold be- tweenthe pronoun of laziness itself and the expression that is interpreted as binding a variable inthe interpretation of thepronoun —callthat expression the licenser (i.e.,between she/her and everyboy over ten in(25),(26), and (27)). Obviously, this requirement cannot be statedin terms ofbinding theory, as none of the expressions involved are ever coindexed. At best,there is a bindingrelation at a semanticlevel of representation between the interpretation of the licenser anda variableproperly contained in the interpretation of the licensee. Assumefor concreteness,following Cooper (1979a), that a pronounof laziness is interpreted as (29),

(29) l K ’ z (; y [S(y)(u) z 4 y] ` K(z)) $ whichreads, roughly, ‘ (thegeneralized quantifier corresponding to) the unique object that u has therelation S with’(where, in (25a), S refers tothe relation beingcousin of, andthe variable u isbound by the quantifier that is the interpretation of the licenser, i.e., everyboy over ten ). It is obviousthat if the correct interpretation of pronouns of laziness is even remotely like this, the licensermust take scope over the pronoun of laziness (as confirmedby (26)), namely, in order tobind the variable u. Hence,(4) explainsthe observations discussed above. But the variable u isnot present in thesyntactic representation, so that there can be noconfigurationalor licensing conditionin terms of syntactic binding or coindexing that makes any distinction between the well-formedand ill-formed examples cited above, other than through some extension of binding theoryfor whichthere is noindependent evidence outside the domain of crossovereffects.

2.4Non-NP Dependencies Afinal,striking case of a WCO-sensitivedependency not reducible to binding is found with determinerssuch as another. 524 E . G . R U Y S

(30)Every student i kissedanother (i) student. In (30) another isinterpretivelydependent upon everystudent, inthatfor eachchoice of akissing student x, thestudent being kissed must not equal x (Ihaveindicated the dependency with a parenthesizedindex). The dependency involved can be clearly perceived by notingthe ambiguity of Everyfemale student saw every male student kiss another student, wherethe student being kissedsystematically differs eitherfrom themale student or from thefemale student. As shownin (31),the dependent element must be inthescopeof thequantifier it isdependent upon.Also, as observedin Barwise1987, the dependency is sensitive to crossover.This is shown in (32).

(31)a. ??John kissed every student i ,whichannoyed another (i) student.

b.Every student i kissedsome other (i) student.

(32)??Another (i) studentkissed every student i . Sentence(32) only has a readingwhere the kissing student differs from somecontextually deter- minedstudent. The same holds for (31a);(31b) has the dependent reading, but only with narrow scopefor theobject. Thetransitivity effect applies as expected.

(33)Each girl’ s i boyfriendtook another (i) girl home. Noneof these observations can be statedin termsof binding theory. First of all,the dependent elementis a determiner,possibly an adjective,but notanNPsubjectto binding in termsof binding theory.And even if the NP modifiedwith another were consideredthe dependent element, we wouldneed a counterindexingmechanism in thiscase, not the coindexing mechanism needed in previousexamples. Similarremarks hold for othertypes of dependencydiscussed in Barwise 1987. (34)a. Every linguist is inawe ofa tallerlinguist. b.??A tallerlinguist contradicted every linguist. c.??Because every linguist presented a paper,a tallerlinguist became angry. Again,the interpretation of taller candepend on the interpretation of the element with wider scope.The availability of thedependent interpretation is subjectto theconditions that determine relativescope: WCO in(34b), the Adjunct Condition that constrains QR in(34c). Deriving these datafrom someversion of binding theory would be less than straightforward. The relevance of scopetheory is unmistakable; we merelyneed to finda suitablecharacterization of thesyntactic conditionson relative scope in order to derive the observations in this section. Assuming that another and taller arenot operators in the sense of (4a) maybe sufficient. 13 Inthis section and in sections 2.1 –2.3I havepresented examples where a WCOeffectis

13 Similareffects obtainwith expressions such as thenearest, the local, similar, opposite, different (Partee 1989), somethingelse (Culicoverand Jackendoff 1995); see alsoCarlson 1987, Keenan 1988,Chierchia 1995 for relevant examples. See especially Partee 1989for arguments that no hiddenpronoun is presentin these expressions. WEAKCROSSOVERASASCOPEPHENOMENON 525 found,or foundto be absent, although binding-based theories of WCOwouldpositively predict theopposite. In each case we saw thatthe structural requirements relevant to WCO were being met,or failedto be met, by pairsof constituentsthat were notrelated in binding-theoretic terms. Ratherthan complicating binding theory in orderto circumventits predictions, I haveproposed thatWCO berecognized as a failedscope relation, as scope theory is obviously relevant in all thesecases.

2.5Exceptional Scope Inthis section I willconsider arguments supporting my thesis that have a ratherdifferent logical structurethan used so far. Briefly,consider an operator whose scope is notsubject to condition (4):its scope is not determined in syntax in the usual way, but follows from someexceptional syntacticor semantic rule. Suppose further that, binding-theoretically ,thereis no indication that thisoperator behaves exceptionally —itsanaphor-binding capacity, for instance,is unexceptional. Our expectationis thatthis operator will also show exceptional variable-binding properties, since theability to variable-binddepends solely on the ability to take scope. A binding-theoreticlicen- singcondition such as (3a),on theother hand, predicts that, because scope taking is notasufficient conditionfor variablebinding, this operator will not exceptionally license bound variables. Inthis section I willdiscuss two potential arguments of this type very briefly, and a third insome detail. Consider first the examples in (35). (35)a. As eachsoldier appeared on theplatform, a flagwas hoistedand a gunwent off. b.As everysoldier appeared on theplatform, a flagwas hoistedand a gunwent off.

c.As eachsoldier i cameinto the room, he i was givena gun.

d.??As everysoldier i cameinto the room, he i was givena gun. In(35a) and (35c) eachsoldier maytake scope over the entire sentence, whereas the scope of everysoldier in(35b)and (35d) is restrictedto theadjunct clause in accordance with conditions onmovement. (35a) thus illustrates the well-known fact that NPs specifiedby the QP each can takeexceptionally wide scope, in apparent violation of locality constraints. For someunknown reason,the scope of such NPs isnot subject to condition (4). Example (35c) then shows that the exceptionalityof each interms of scopeassignment carries over to thevariable-binding properties ofNPs specifiedwith each. Thecontrast between (35c) and (35d) is asexpected, provided that variablebinding depends only upon the operator’ s beingable to take scope over the pronoun. Butif one supposes that variable binding requires a binding-theoreticlicensing relation, such as A-binding,in addition to the scope requirement, then the contrast is unexpected, since there is noevidencethat each isexceptional in termsof binding theory. The conditions under which each N canbind an anaphor, for instance,do not differ from theconditions that allow every N to do so.14

14 Notethat this line of reasoningargues only against a licensingcondition (3a), not against a configurationalcondition (3b): eachsoldier in(35c) does not A ¯ -bind he at anylevel, so the condition is not violated; the same holdsfor the other structuresdiscussed in this section. Nonetheless, I believethat this detracts onlyslightly from the importance of the presentargument type, as licensingconditions are themore promisingWCO conditions(cf. footnote4). 526 E . G . R U Y S

Asecondexample with the required properties might be theVP-conjunction donkey anaphora structure(36).

(36)Every farmer [ VP[VP ownssome donkeys i ] and [VP feeds themi at night]]. Theindefinite somedonkeys in(36) does not A-bind the pronoun them; indeed,it does not c-command them atany syntactic level of representation. By condition (3a), we wouldexpect them notto be interpretable as a boundvariable, as it is not bound, let alone A-bound, by any syntacticexpression at any level. Presumably, it isbound by adefaultexistential closure operator associatedwith somedonkeys. As suchan operator is not a syntacticobject, its scope is not determinedby syntactic constraints such as (4), so that it is free totake scope over, hence, by myreasoning, to bindthe pronoun (see Heim1982, Kamp and Reyle 1993, Ruys 1992, to appear, for furtherdiscussion). Iwillconsider a finalexample of this type in some detail. Examples like (37) have been discussedby, for example,Geach (1962), Cooper (1979b), and Hornstein (1984).

(37) a. [NP Thewoman that every Englishman i lovesmost] is his i mother.

b. [NP Theanimal that every Englishman i cherishesmost] is his i own dog. Theseexamples are remarkable in several respects. First, the quantified NP everyEnglishman seemsto takescope over the complex NP thatcontains it, although the Complex NP Constraint andother constraints should prevent it from movingout of this NP. Second, everyEnglishman in(37a) can bind the pronoun his asa variable,although it does not c-command the pronoun from anA-position. But not only do exceptionalscope and exceptional variable binding cooccur inthisconstruction, it isalso exceptional from abinding-theoreticpoint of view.In (37b) his own, presumablyan anaphor (see Higginbotham1985:fn. 28), is coindexedonly with everyEnglishman, whichdoes not A-bind it, but the structure is wellformed nonetheless. Hornstein(1984) proposes that the well-formedness of (37a) be viewed as the result of a reconstructionprocess presumed operative in copula constructions. A similaranalysis, suggested earlierby Cooper(1979b), proposes that the reconstruction effect in (37)could be reducedto the connectednesseffect found in pseudocleft constructions. These authors claimed that, when the copula be isreplaced with a lexicalverb, binding is no longer possible. (38a) is from Hornstein 1984,(38b) from Cooper1979b.

(38) a. *[NP Thewoman that every Englishman i likes]kissed his i mother.

b. *[NP Theanimal every Englishman i cherishes]hates his i own dog. If exceptionalvariable binding correlates both with exceptional scope assignment and with excep- tionalA-binding in (37), then this construction does not offer anyprima facie evidence about whethervariable binding is dependent upon scope or A-binding.And the reconstructionhypothesis doesnot decidethis issue, since both A-binding and scope assignment can be shownto beaffected byreconstruction processes. Nevertheless,I believethat whatever the value of Hornstein’ s andCooper’ s proposals,and whateverthe ultimate explanation of (37b),examples of this type can be used to justify reducing WEAKCROSSOVERASASCOPEPHENOMENON 527 boundvariable licensing to scope theory. In noncopula constructions the observed three-way exceptionalitybreaks down. In such constructions exceptional A-binding is no longer possible. Butin many cases (unlike (38a)) scope assignment remains exceptional, and bound variable licensingthen sides with scope assignment, not with A-binding. This can be concludedfrom the examplesin (39).

(39) a. [NP Theman who builds each television set i ]alsorepairs it i/*itselfi .

b. [NP Thevery woman that every boy i lovedmost] came out to save him i ; Won- derwomancame out to save Bill, Spiderwoman came to save Peter, . ..

c. [NP Thewoman that every Englishman i lovesmost of all]loves him i/*himselfi least of all. Ineach of these cases the universally quantified NP takesscope over the entire sentence in spiteof the Complex NP Constraint,and it variable-binds the pronoun although A-binding is impossible. 15 Althoughthe wide scope readings may be somewhat marginal in these examples, thepoint is that,when a widescope reading is obtained,variable binding is possibleas well. This confirmsmy central hypothesis. Thisconclusion would be strengthened if a plausibleexplanation were providedfor the exceptionalityof thisconstruction, from whichits interpretive properties could be made to follow. Suchan explanation might run roughly as follows. Observethat in (37) and (39) the values assigned to the complex NP area functionof the valuesassigned to the universal quantifier. (37a), for instance,is interpretedsuch that for every Englishman x, thereis exactly one woman that x lovesmost. Now supposethat in semantics, a quantifiercontained in anNP maybe given scope over that NP (evenif thisis not syntactically licensed)in casethe NP istheninterpreted as just such a function.Other factors in (37) and (39) allconspire to bringa functionalinterpretation about; this would then be enoughto triggera rule ofexceptionalscope inversion in semantics. The factors contributing to the functional interpreta- tionin (37) and (39) include the use of most and,in particular, the use of thedefinite article. 16 If thisassumption is correct, the possibility of variable binding in these examples follows: theuniversal quantifier is assigned scope over the complex NP, whichin turn is syntactically licensedby (4b) to take scope over the pronoun; the universal quantifier then takes scope over thepronoun by transitivity(see section2.1), which is sufficientto allowa boundvariable reading. 17

15 (39a),taken from Cooper 1979b, is slightlybetter than the others, presumably because itcombinesa structureof thetype discussed here withthe use oftheQP each, whichin itself has exceptionalscope properties (see thebeginning ofthissection). For more sophisticatedanalysis of these examples see Sharvit1999, which unfortunately was notavailable at thetime thisarticle was written. 16 See Lo¨bner1985 for a plausibleargument that nouns specified with the definite article are always interpretedas functions. 17 There is theexpected contrast between (37)and (i).

(i) ??Hisi motherhates [ a thewoman that every boy i lovesmost of all]. Theill-formedness of (i) can beexplained as follows.Even if every boy may take scopeover the NP a thatcontains it, itmay notthereby take scopeover his since a isnot licensed to take scopeover his, sothat transitivity does not provide thenecessary scoperelations. Myanalysisin terms ofexceptional scope assignment does not preclude, of course, that connectedness does play a 528 E . G . R U Y S

Thishypothesis correctly predicts that the wide scope reading is blocked in casewe remove thesuperlative and the definite determiner (and other relevant factors). The complex NPs in(40) couldnot be interpreted as functions even if theembedded universal quantifiers were givenwide scope;hence, the wide scope reading is not available.

(40) a. ?[NP Someman who admires each television set i ]describesit i to his wife.

b. ??[NP Somewoman that every boy i loved]came out to save him i .

c. ??[NP Awomanthat every Englishman i loves]loves him i as well. Mynonsyntactic approach to the wide scope readings in (39) is moreplausible than a syntactic approachwould be for variousreasons. The likelihood of afunctionalinterpretation partly depends onthe interaction of idiosyncratic properties of lexical items; it is highly unlikely that their combinedinfluence upon scope assignment (if thatis indeed the correct observation) could be explainedin syntactic terms. A syntacticexplanation for thecontrast between (39) and (40) is allthe more unlikely because replacing the definite article with the indefinite article should, if anything,make extraction of the quantifier out of the complex NP easier. 18 If thewide scope readingin (39)were dueto QR, we shouldexpect this reading to become more easily available in(40), instead of beingruled out. Thefunctional scope assignment hypothesis proposed here has the added advantage of possi- blyproviding some insight into hitherto ill-understood restrictions on inverse linking. Consider (41).19

rolewhen the copula is present.But Cooper’ s andHornstein’ s connectednessanalysis is less thanplausible because it doesnot explain the well-formedness of(ii),where every Englishman doesnot c-command thegap indirectly associated with his mother, butvariable binding is possiblenonetheless. Similarly in (iii).

(ii)The woman that t lovesevery Englishman i most is hisi mother.

(iii)The man thatt interviewedevery president i insultedhim i. Afurtherargument against the connectedness analysis might be basedon (iv).

(iv) Hisi motheris thewoman that every Englishman i likest best. (iv)is well formed,but only with contrastive stress on his mother. Supposingthat this indicates movement of his mother fromthe predicate positionof the copula (of the type discussed in, e.g., Emonds 1976:34– 40, Hoekstra and Mulder 1990), we can explainthis as follows.When his mother isthesubject, the pronoun cannot be marked inthe scope of the complex NP thewoman that every Englishmanlikes best; hence, every Englishman cannottake scopeover the pronoun by transitivity.If his mother originatesfrom a positionc-commanded bythecomplex NP insubject position, the latter can take scopeover the pronoun and variable binding is fineby transitivity.A binding-theoreticcondition on variablebinding thatinvokes connectedness to explain the data inthis section cannot make thisdistinction since his mother isconnected tothe trace positionin the complex NP undereither analysis. Finally,note that, even if the bound variables in (37) are licensedthrough reconstruction, this alone does not explain thebound variable reading. It must still be assumed thatthese examples showexceptional scope assignment (of the embeddedquantifier over the complex NP) as well,since thisscope relation is at least anecessary conditionfor variable bindingunder any analysis. 18 Because oftheSpecificity Constraint on movement.See Chomsky1973, Fiengo and Higginbotham 1981. 19 These data andtheir relevance were pointedout to me byTanyaReinhart. WEAKCROSSOVERASASCOPEPHENOMENON 529

(41)a. Someone in every city i hates iti .

b.??An owner of [everycar in the street] i shouldtow it i inthe morning.

c.The owner of [every car in the street] i shouldtow it i inthe morning. Althoughinverse linking and variable binding are possible in the well-known case (41a), they arenot in the superficially similar case (41b). Although I haveno suggestionsto offer aboutwhy thisnoun-complement case should be worse thanthe noun-adjunct case, 20 we cannow understand theotherwise puzzling fact that (41b) can be improved by replacing the indefinite article with thedefinite article, as in (41c). In (41c) the subject NP willbe interpreted as a functionof its internalargument; hence the option of exceptional‘ ‘inverselinking’ ’ insemantics, and concomi- tantvariable binding. Note again that we canhardly suppose the distinction between (41b) and (41c)to be due to restrictionson syntacticmovement. We wouldthen expect the definite article in(41c) to make extraction of thequantifier less acceptable, not more acceptable, owing to the SpecificityConstraint. Inconclusion, the data in (39)indicate first of all,from arelativelypretheoretical viewpoint, thatvariable binding derives from scopetheory, not from bindingtheory. In addition, if my suggestionsregarding the proper theoretical characterization of theexceptional scope assignment phenomenondiscussed in this section are at all plausible, the bound variable readings observed canbe derivedfrom mytheoretical assumptions. Then, not only do (37)and (39) show that scope maybe assigned in other ways than on the basis of (4), thereby evading (4b) and permitting exceptionalvariable binding, but we mayalso begin to understand why (4) canbe circumvented inthese examples.

2.6Conclusions Inthis section I haveargued that the single requirement that must be met for apronounto be interpretableas a variablebound by a coindexedoperator is that the operator must be interpreted withscope over the pronoun. I havealso argued that for aquantifierto takescope over a pronoun, itmust meet stricter conditions than for itto takescope over, say, another quantifier. Specifically, Ihaveassumed that the quantifier must c-command the pronoun from anA-position (at some stagein the derivation). This restriction explains the WCO effect. Variouspredictions follow. First, pronouns that are not interpreted as variablesbound by a quantifierQ maynevertheless show a WCOeffectrelative to Qiftheyare interpreted as scopally dependentupon Q insome other way (the pronouns of laziness of section 2.3 and, under some analyses,the donkey anaphors of section2.2). Also, extending the A-c-command requirement to otherlexical categories explains the WCO effectwith nonpronouns (section 2.4). Second,although c-command from anA-position is anecessarycondition for syntacticscope licensing,a quantifiermay in additionbe licensedto take scope over a pronounthrough properties ofsemantics.The transitivity effect in section 2.1 may be oneexample: if someother quantifier Rislicensed to take scope over a pronounP, andQ maytake scope over R, thenQ maytake

20 See Hornstein1995 for an interesting suggestion in this regard. 530 E . G . R U Y S scopeover P, sothatvariable binding and other scope-dependent relations between Q andP are allowed.In some cases (section 2.5) the effect is compounded: if R takesscope over P, andQ isexceptionallylicensed (through properties of semantics)to takescope over R, thenQ maytake scopeover P, hencevariable-bind it. Ihavealso argued that these predictions cannot be madeto followfrom variousalternative analyses.In section 1 Imadea distinctionbetween two types of conditions on bound variables: configurationalprinciples and licensing principles. The examples discussed in thepresent section showthat the relevant condition (of whatevertype) must be basednot on bindingtheory, but on scopetheory. The transitivity examples show the effects of the A-c-command condition on the scoperelation between pronoun P andthe ‘ ‘mediating’’ quantifierR, althoughthere is nobinding- theoreticor coindexing relation between P andR. Similarly,donkey anaphora shows effects of theA-c-command condition between a quantifierand a pronounit must take scope over (under anyanalysis, as an E-type pronoun or as a boundvariable), although again there is no binding- theoreticrelation between the quantifier and the pronoun. Similarly with non-NP scope dependen- ciessubject to A-c-command, and with pronouns of laziness.These classes of examplesall argue againstboth types of binding-based bound variable conditions: licensing principles based on coindexingcannot rule in the well-formed cases, and configurational principles cannot rule out theill-formed ones. Furthermore, licensing principles based on binding cannot allow for the possibilityof bound variable interpretations in the exceptional scope assignment examples of section2.5. 21 Of course,a relationof syntactic coindexing might simply be postulated for allpairs of expressionssubject to WCO examinedhere. But there would be no syntactic evidence for this coindexingrelation (from bindingconditions or other principles) except for theWCO effectit wouldbe used to describe. Nor wouldthere be a semanticinterpretation for thishypothetical coindexingrelation (such as coreference or variable binding); it would thus remain completely stipulative. 22

21 Space limitationsprevent me fromseparately discussingall, or even most, representatives ofthe binding-theoretic orcoindexingapproach to WCO.To mention just a few, though:the strictly binding-theoretic view of Reinhart(1983) (basicallycondition (3a)) predicts none of the examples discussedin this section; its ‘ ‘configurational’’ counterpart,the BijectionPrinciple (Koopman and Sportiche 1982) fails onall butthe ‘ ‘exceptionallicensing’ ’ examples insection 2.5. Byextending the notion of coindexingor replacingit witha derivednotion, various authors have succeeded indescribing some oftheother facts discussedhere. Thus, Ha ¨õ k’s(1984)indirect binding correctly predicts the donkey anaphora facts (exceptthose involving transitivity) in section 2.2; Safir’ s (1984)Parallelism Constrainton Operator Binding, coupled withan index-copying mechanism, describes thetransitivity effect (section2.1); a similar index-copyingmechanism accountsfor the donkey anaphora facts ofsection 2.2 in Reinhart 1987. Higginbotham’ s (1980,1983) Accessibility Condition(see alsoHuang 1995, Hornstein 1995), which employs mechanisms furtherremoved from strict coindexing thanthe other proposals, captures boththe transitivity effect andthe donkey anaphora facts insection 2.2. It is important tonote that attempted explanations of WCO interms ofmovementtheory (such as thosementioned in footnote1) also cannotcope with any WCO examples where thepronoun and the relevant quantifier are notcoindexed. 22 Tobe more precise (ifnot pedantic), I see onepossible semantic interpretationfor this hypothetical coindexing relation:it might be taken to encode the relation of relative scope(or ‘ ‘interpretivedependence,’ ’ orsome suchnotion), thusimplementing the scope-theoretic approach defended here (see alsofootnote 23). WEAKCROSSOVERASASCOPEPHENOMENON 531

Myconclusions here are empirical in nature, but they are not restricted to proposing a particulargrammatical mechanism. I amnot primarily interested here in defending the scope- theoretichypothesis as it is expressedby condition (4). Rather, I haveattempted to establishthe moregeneral point that, whatever the ultimate condition responsible for WCO turnsout to be, it mustbelong to the theory of scope. WCO canbe explained with a properformulation of the conditionsthat allow a quantifierto take scope over a pronoun.

3WCOinthe MinimalistFramework Sofar boththeoretical and empirical progress have been made. I haveargued first ofall that, evenfrom arelativelypretheoretical stance, considerable insight can be gainedsimply by viewing theconditions on bound variable licensing as belonging to scope theory. But I havealso shown thata relativelyminor adjustment of theScope Principle (clause (4b)) actuallyderives many of theexamples under discussion. In short, a reinterpretationof WCOasascopephenomenon holds thepromise of not just describing but explaining various exceptional cases of crossover. Butin order to be empirically adequate, I haveresorted to a concomitantcomplication in scopetheory, to witclause (4b). The scope principle (4) raisesseveral major questions, such as: canwe givea principledaccount of theclass of syntacticcategories subject to clause (4b); and whatother indications are there that the semantic interface level LF isinterpreted by means of ascopeprinciple of thistype? In addition, various technical issues of implementation arise. These questionsare addressed at length, and a moreprecise implementation is provided, in Ruys 1992. 23 Here, Iwantto explore a differentroute. Amajorpotential advantage of thescope-theoretic perspective on WCO(althoughone that was nottaken advantage of in Ruys 1992; see footnote 23) is that it does not rely on asyntactic

23 InRuys 1992 I proposean implementation that uses adiacriticscope-marking mechanism, whichdeals withscope reconstructionand specificity as well.These are themain rules: (i) ScopeMarking a. Assign to a theindex of any c-commanding category b as asuperscript,at anylevel of representation (obligatoryif a the trace of b ); except b.Pronouns may bescope-markedonly by XP in A-position. Afurtherconstraint, which we neednot go into,implies that weak NPs,unlike strong QNPs, may remain unmarkedby c-commandingoperators. LF representationsenriched with scope-marking superscripts are interpretedby thefollowing rulesof interpretation: (ii) ScopePrinciple Aisinthe scope of Biff a. Ais superscriptedby B;or b.A is superscriptedby C,andC is inthe scope of B(transitivity). (iii) ScopeCondition proi is interpretedas xn iff NPi is associated with xn andthe operator that binds xn may take scopeover the interpretationof pro i. TheScope Principle (ii) is superficiallysimilar tothat proposed in Ha ¨õ k1984;see Ruys1992 for a discussionof crucial differences between thetwo proposals. To mention just one, it is essential toHa ¨õ k’saccountthat her scope-marking slash indicescount as referential indices,as she intendsto unify scope theory and binding theory. It should be clear fromthe foregoingdiscussion that nothing could be more distinctfrom my views. See alsofootnote 21. 532 E . G . R U Y S relationof coindexing. I haveshown above that this is empirically well motivated. Within the MinimalistProgram (Chomsky 1995), it is also conceptually preferred. None of theother WCO hypothesesdiscussed above will fit into the minimalist framework, as theyrefer toasyntactically annotatedrelation between subexpressions that is neither maximally local nor conceptually un- avoidable.This holds good to an even greater extent with regard to those extensions of the ‘‘coindexing’’ relation(slash indexing, etc.) that do not denote coreference or covaluation and aremere technicaldevices intended to capture exceptional cases of WCO. Withinthe Minimalist Programone wants to diagnose WCO asreflectingthe mode of operation of theC-I (conceptual- intentional)interface component, in terms of a minimalset of notions that must be assumed to beoperative at that level: say, c-command, covaluation/ coreference,and relative scope. There aregood indications that scope theory as it is developing in the Minimalist Program will yield thescope relations between operators and pronouns that I haveclaimed are revealed by theWCO phenomenon. Inthis section I willsketch the outlines of aminimalistview of WCOthat,although making useof thefindings of previoussections, further reduces the stipulations required in orderto derive theconditions on variablebinding, specifically (4b). Although at presentI cannotaccount for all theobservations discussed in previous sections, I hopeto be able to show that further research alongthese lines may prove fruitful. Consideragain a simpleexample of WCOin(42).

(42)??Which boy i does hisi motherlike t i ? In (42) the wh-operatorhas undeniably moved to a positionwhere it c-commandsthe pronoun. Hence,one should expect there to benodifficultyin assigning it scopeover the pronoun. Further- more,the question operator in (42) is certainly interpreted with scope over the whole clause. Why,then, should the intended bound variable reading be unavailable if, as I claim,all that is requiredfor sucha readingis that the binding operator have scope over the pronoun? The answer comesfrom distinguishingthe scope of the question operator from thescope of the wh-NP. As iswell known, the observable position of an operator expression does not determine its scope. Andthe fact that the scope of thequestion operator in (42) is thewhole clause does not necessarily implythat the scope of the wh-word isthe same. 24 Recentdevelopments in the interpretation of wh-elementsmay allow us tosee why the scope of a wh-NP isnot identical to the scope of thequestion operator. Proceeding on the minimalist assumptionthat wh-in-situdo notmove at LFbutremain in situ (Chomsky 1993), Reinhart (1992, 1998)considers how wh-in-situare interpreted. She observes that in exampleslike (43a) existing analyseslead to incorrect predictions. The assumption that philosopher remainsin situ and is

24 In fact, Ha¨õ k(1984)cites evidencethat the scope of a wh-NP cannot beequated with the scope of thequestion operator. (i)Which men didsomeone say thatMary likes t? Accordingto Ha ¨õ k, someone in(i) is not interpreted in the scope of plural which men, althoughit isin the scope of the questionoperator. WEAKCROSSOVERASASCOPEPHENOMENON 533 predicatedover an individual variable bound by an operator in the position of who (through unselectivebinding, absorption (Higginbotham and May 1981),or movementof just which (Chom- sky1993)) implies that (43a) is interpreted as ‘ for which x and y isit the case that x will be offended if y isa philosopher,and we invite y’. (43)a. Who will be offended if we invitewhich philosopher? b. $ P | ’ k x,yl [P 4 V((y isa philosopher ` we invite y) (x willbe offended)) ` ! vP]} (43b)is theinterpretation that results if we adoptKarttunen’ s (1977)view that a questiondenotes theset of its true answers. The set (43b) will for examplecontain the proposition that Lucy will beoffended if we inviteDonald Duck (as DonaldDuck is not a philosopher,the antecedent of theimplication will be false; hence, P willbe true). The incorrect prediction then follows that Lucywill be offended if we inviteDonald Duck isatrueanswer to (43a). Inorder to remedythe situation, Reinhart proposes that wh-in-situare interpreted as follows. Whatis unselectively bound by thequestionoperator is a variablethat ranges over choice functions (whichassign to eachset in their domain a memberof theset); this function is applied to the in- siturestriction on the wh-in-situ.This gives for (43a)the interpretation (44), which is correct.

(44) $ P | ’ k x,fCF l [P 4 V(we invite f (philosopher) x willbe offended) ` vP]} ! (43a)now roughly paraphrases as ‘ for which x andwhich choice function f isit the case that x willbe offended if we invitethe individual that f selectsfrom theset of philosophers’ . The propositionthat Lucy will be offended if we inviteDonald Duck is nota memberof the set of trueanswers (44), since there can be no f thatselects Donald Duck from theset of philosophers. 25 Theresult of Reinhart’ s proposalis that wh-operatorsmay be interpreted in two distinct manners.In case they happen to move(overtly), they are interpreted through existential quantificationover individual variables; in case they do not move, they are interpreted through existential quantificationover choice functions. But in fact we onlyhave semantic evidence that the latter optionis required.Let us thereforesuppose, in order to arrive at a uniforminterpretation of wh- operators,that all wh-elements,including those that are preposed in syntax, must be interpreted throughquantification over choice functions. This implies that wh-NPs mustalways undergo reconstructionat somelevel, since the result of applyinga choicefunction to asetis an individual, notan operator, hence cannot be interpretedin operator position. 26 Itfollows that the LF representationfor (42)is in relevantrespects as givenin (45a), which inturn is interpreted as in(45b).

25 Forelaboration and refinement ofthe choice function approach, see Reinhart1997, Winter 1997, 1998. 26 Thisaccount suggests of course that the preference principleproposed by Chomsky(1993) that says thatreconstruc- tionof preposed wh-elements mustbe maximal is areflectionof the need to massively reconstruct wh-operatorsfor interpretivepurposes. The fact thatanaphors contained in wh-phrases may escape thereconstruction process may bedue toan operation that extracts them fromthe wh-phraseprior to reconstruction. However, this leaves theeffects ofidiomatic readingson the wh-phraseunaccounted for. 534 E . G . R U Y S

(45)a. ??which j does hisi motherlike [t j boy]i b. $ P | ’ fC F [P 4 V(hisi motherlikes f(boy)i ) ` vP]} Deriving(45a) is enough in itself to explain the WCO effectif we assumethat a pronounmay notbe coindexed with a non-DP: which may not bind his, andthe intended reading is ruledout. We caneven do without this condition on coindexing if we acceptthe choice function analysis (45b).In (45b) the existential quantifier takes scope over the complete sentence, including the pronoun.However, it cannotbind the pronoun as avariable,if we assumethat pronouns can only beinterpreted as variablesranging over individuals, not as choice function variables. Eitherway, we stillcorrectly predict that variable binding is possible in (46).

(46) a. [CP Which boyi [IP ti likes hisi mother]]?

b. [CP whichj [IP [tj boy]i likes hisi mother]] c. $ P | ’ fCF [P 4 V(f(boy) l x(x likes x’s mother)) ` vP]} Inthe LF representation(46b) the pronoun is bound by the in-situ remnant of the wh-DP. The interpretiveprocess is illustrated in (46c). According to theusual analysis (Lasnik 1976, Reinhart 1983),the VP isinterpreted as a l -expressionoperating on the subject: the l -operatorhas scope over,hence can bind, the pronoun. Canwe extendthis reasoning to apply to quantificational NPs? Consider(47).

(47) ??Hisi motherlikes every boy i . Insome ways, this is aweakertype of example.A puzzlearises only if we assumethat the object every boy iscapable of takingscope over the subject by QR: thequestion then is whyit cannot bindthe pronoun, and a WCOexplanationis required. That the object can take scope over the subjectis assumed on the basis of the ambiguity of such examples as (48). (48)Someone loves everyone. Butit is well known that intuitions regarding the wide scope reading in (48)are far from secure. Nativespeakers unfamiliar with this field of linguistic research are notoriously reluctant to accept thewide scope reading for everyone.27 Amonglinguists working in thisfield, intuitions have also beensubject to debate and shifts in opinion. Even those who firmly believe that a widescope readingis available will usually be more than ready to admit that the reading that reflects the surfaceorder of thequantifiers is much more easily available than the inverted order. 28 In view

27 Thisis confirmedin experimental work; see forexample Kurtzman andMacDonald 1993 and references cited there.Kurtzman andMacDonald report a strongpreference forthe observable order in [ a N] V [every N]structures. Also, in [NP ...NP]cases theyfound the inversely linked reading to be preferred,confirming the ‘ ‘transitivity’’ observation. 28 Theexplanation for the difference inavailability in performance between thetwo readings might run roughly as follows(as suggestedin Ruys 1994a). The reason that QR is possibleat all,given considerations of economy (as in Chomsky1993), is thatquantified expressions are notinterpretable at LFunlessoccurring as thehead of anoperator- variablechain (see Epstein1992). This overcomes Last Resort,but only allows a QNP tomove to thenearest possible A¯ -position,in viewof ShortestLink. Therefore, the object in (48)should not be able toraise byQRtoa positionwhere itc-commands thesubject at all.However, relativizing Shortest Link to interpretation(as inGolan 1993, Reinhart 1998, Fox1995, Ruys 1994a,b), so that only derivations that result in equivalent interpretations are comparedwith respect to WEAKCROSSOVERASASCOPEPHENOMENON 535 ofthis, it is not clear that the difficulty in obtaining the bound variable reading in (47) differs significantlyfrom thedifficulty in obtaining the wide scope reading for theobject in (48). In otherwords, the WCO effectdetected in (47)may well reflect the general difficulty of assigning widescope to the QNP inobject position. If thisis indeed the correct way of looking at these examples,the WCO phenomenonin (47) simply disappears. 29 Nonetheless,a morepositive approach is also available in theMinimalist Program. Chomsky (1993)argues that only the quantificational specifier every in every boy undergoesQR (Move Spec);a laterversion (Chomsky 1995:chap. 4) hasit thatonly the formal features of theDP move (MoveF). Thisleads to a reasoningsimilar to the one employed above with regard to wh- expressions.(47) is assignedthe LF representation(49a). 30

(49)a. every j [hisi motherlikes [t j boy]i ]

b. every fCF [hisi motherlikes f(boy)i] The DP every boy nevertakes scope over the pronoun; only the specifier every does.Again, we mayrule out a boundreading for (49a)by assumingthat a pronounmay only be bound by a DP. Or we canmove one step further and derive this assumption from theway in which every is itself interpreted.In Ruys 1997 I suggestthat the problematic semantics of quantificationalspecifiers thatare separated through Move Spec from theirrestricting N ¢ maybe resolvedby assuming that, like wh-specifiers,they range over choice functions. If so,(49a) is interpreted as (49b), and we againderive the absence of a boundreading by assuming that pronouns never have a choice function–typereading. And again, a boundreading can only be obtainedin (50a), as illustrated here.

(50)a. Every boy i likes hisi mother.

b. everyj [IP[DP tj boy]i likes hisi mother]

c. every fCF [ f(boy) l x(x likes x’s mother)] TheMove Spec/ choicefunction hypothesis exactly predicts the A/ A ¯ bifurcationstipulated bymany accounts of the crossover effect. It predicts that QR and wh-movementdo not affect thevariable-binding options of quantifiedand wh-expressions.Rather, these expressions behave asif theyremain in anA-position,where they have the same binding options as any other term

thelength of their links, may overcomethis in principle.Suppose now that, in performance, theeffect ofShortest Link is thata QNP is assumed tomove to the nearest possibleA ¯ -positionunless longer movement has beendetermined to resultin a nonequivalentrepresentation. This implies that the reading for (48) that results from the shortest possible movementwill be easily detected,whereas obtainingthe reading that requires a longermovement requires calculating interpretiveequivalences, which entails a heavyprocessing burden. 29 Intuitionswith WCO are at least as volatileas intuitionson scopeambiguity. For example, His mothergave every soldiera goodluck charmto wear intobattle ismuch better than (47). WCO intuitionsare muchstricter with wh- antecedents,suggesting they may requirea differentexplanation, as above.See Wasow 1979for many examples of variationin acceptability with WCO. 30 Theanalysis presented here is basedon Ruys1997, where Iarguethat Move Spec and Move F make thesame predictions.Here Iwillconcentrate on the Move Spec option: every moves, [tevery boy]remains insitu. For a more detaileddiscussion, see Ruys1997. 536 E . G . R U Y S inanA-position. This in effect derives the conditions (3) and(4) for apromisingrange of cases. Furthermore,the choice function account attributes the WCOphenomenonto a semanticmismatch (pronounsfunction as individual variables at best, which cannot be bound by choice function quantifiers),so that WCO isseento reflect a propertyof theC-I interface. Theconception described in this section requires significant further research. It has ramifica- tionsin such areas as wh-movement,reconstruction, and relativeclause construction, which remain tobe explored. Nevertheless, it should be appreciated that this project of further reducing the WCOeffectto theconditions on scope assignment among operator-type expressions can only be undertakennow that it has been established, in the earlier sections of thisarticle, that the conditions onvariable binding are indeed to be explainedas belonging to scope theory proper.

4Conclusions Inthis article I haveattempted to establish the following point. The conditions under which an anaphoricexpression may be interpreted as a boundvariable must be explained by a principle belongingto scope theory, so that WCO mustbe considereda scopephenomenon. This conclusion was basedon arangeof facts that indicate, first, that the common denominator that unites various typesof WCO violationscannot be expressed in termsof binding(or movement),but presumably isamenable to analysis in terms of relativescope (pronouns of laziness,non-NP dependencies, donkeyanaphors); and second, that various semantic factors involved in the determination of scoperelations may play a rolein licensing bound anaphora (transitivity, exceptional scope). This leavesus with this desideratum for theoptimal theory of bound variable anaphora: it must reduce conditionson bound anaphora to the conditions that license scope relations among linguistic expressions. Besidesformulating and defending this desideratum for atheoryof boundanaphora, I have presented,partly for purposesof expositoryconcreteness, a scope-licensingcondition that allows themajority of datadiscussed here to be described, and partly explained. Theabundance of WCOanalysesthat have been proposed in the literature over the past 30 yearssuggests that these analyses may all be seriously underdetermined by the empirical data (whichare often confusing and seldom allow definitive generalizations). In view of thisrecord, itmay well be the case that further progress in this area will not come from thestudy of these particularphenomena themselves, but must await developments in related areas to provide a theoreticalframework for amoreprincipled description of conditionson bound anaphora. Rather thanwanting to addyet another, equally underdetermined WCO analysisto thegrowing catalogue, Ihavepreferred therefore to examinesome of thepreconditions that must be metfor atheoryof boundanaphora to be successful, and to determine the area of linguistic research from which progressis mostlikely to be forthcoming. Perhaps, the program outlined in section 3 willprove fruitfulin this sense.

References Aoun,Joseph, and Yen-hui Audrey Li. 1989. Scope and constituency. LinguisticInquiry 20:141–172. WEAKCROSSOVERASASCOPEPHENOMENON 537

Aoun,Joseph, and Yen-hui Audrey Li. 1993. Thesyntax of scope. Cambridge,Mass.: MIT Press. Barwise,Jon. 1987. Noun phrases, generalized quantifiers and anaphora. In Generalizedquantifiers: Linguis- ticand logical approaches, ed. Peter Ga¨rdenfors,1– 29. Dordrecht: Reidel. Carlson,Greg. 1987. Same and different: Some consequences for syntax and semantics. Linguisticsand Philosophy 10:531–565. Chierchia,Gennaro. 1995. Dynamicsof meaning: Anaphora, presupposition and the theory of grammar. Chicago:University of ChicagoPress. Chomsky,Noam. 1973. Conditions on transformations. In Afestschriftfor Morris Halle, ed.Stephen R. Andersonand Paul Kiparsky, 232– 286. New York:Holt, Rinehart and Winston. Chomsky,Noam. 1977. On wh-movement.In Formalsyntax, ed.Peter W. Culicover,Thomas Wasow, and AdrianAkmajian, 71– 132. New York:Academic Press. Chomsky,Noam. 1993. A minimalistprogram for linguistic theory. In Theview from Building 20: Essays inlinguistics in honor of Sylvain Bromberger, ed.Kenneth Hale and Samuel Jay Keyser, 1 –52. Cambridge,Mass.: MIT Press. Chomsky,Noam. 1995. TheMinimalist Program. Cambridge,Mass.: MIT Press. Cooper,Robin. 1979a. The interpretation of pronouns. In Syntaxand semantics 10: Selections from the ThirdGroningen Round Table, ed.Frank Heny and Helmut S. Schnelle,61 –92.New York:Academic Press. Cooper,Robin. 1979b. Variable binding and relative clauses. In Formalsemantics and pragmatics for naturallanguages, ed.F. Guenthnerand S. J.Schmidt,131 –169.Dordrecht: Reidel. Culicover,Peter, and Ray Jackendoff. 1995. Somethingelse forthe binding theory. LinguisticInquiry 26: 249–275. Emonds,Joseph E. 1976. Atransformationalapproach to English syntax. New York:Academic Press. Epstein,Samuel David. 1992. Derivational constraints on A ¯ -chainformation. LinguisticInquiry 23:235–259. Evans,Gareth. 1977. Pronouns, quantifiers, and relative clauses (I). TheCanadian Journal of Philosophy 7:467–536.[Reprinted in Gareth Evans, Collectedpapers, 76–152.Oxford: Clarendon Press (1985).] Fiengo,Robert, and James Higginbotham. 1981. Opacity in NP. LinguisticAnalysis 7:395–421. Fox,Danny. 1995. Economy and scope. NaturalLanguage Semantics 3:283–341. Geach,Peter. 1962. Referenceand generality. Ithaca,N.Y.: Cornell University Press. Georgopoulos,Carol. 1991. Canonical government and the specifier parameter: An ECP accountof weak crossover. NaturalLanguage & LinguisticTheory 9:1–46. Golan,Yael. 1993. Node crossing economy, superiority and D-linking. Ms., Tel Aviv University. Ha¨õ k,Isabelle. 1984. Indirect binding. LinguisticInquiry 15:185–223. Heim,Irene. 1982. The semantics of definiteand indefinite noun phrases. Doctoral dissertation, University ofMassachusetts,Amherst. Heim,Irene. 1990. E-type pronouns and donkey anaphora. Linguisticsand Philosophy 13:137–177. Higginbotham,James. 1980. Anaphora and GB: Somepreliminary remarks. In Proceedingsof the Tenth AnnualMeeting of NELS, 223–236. (Cahierslinguistiques d’ Ottawa 9.4.)Department of Linguistics, Universityof Ottawa, Ontario. Higginbotham,James. 1983. Logical Form, binding, and nominals. LinguisticInquiry 14:395–420. Higginbotham,James. 1985. On semantics. LinguisticInquiry 16:547–594. Higginbotham,James, and Robert May. 1981. Questions, quantifiers and crossing. TheLinguistic Review 1:41–80. Hoekstra,Teun, and Rene´ Mulder.1990. Unergatives as copular verbs: Locational and existential predication. TheLinguistic Review 7:1–79. Hornstein,Norbert. 1984. Logicas grammar: An approach to meaning in natural language. Cambridge, Mass.:MIT Press. Hornstein,Norbert. 1995. LogicalForm: From GB tominimalism. Oxford:Blackwell. 538 E . G . R U Y S

Huang,C.-T. James. 1995. Logical Form. In Governmentand Binding Theory and the Minimalist Program, ed.Gert Webelhuth, 125 –175.Oxford: Blackwell. Kamp,Hans. 1981. A theoryof truth and semantic representation. In Formalmethods in the study of language, ed.J. A.G. Groenendijk,T. M.V.Janssen,and M. B. J.Stokhof,277 –322.Mathematical Centre,Amsterdam. [Reprinted in Truth,interpretation and information: Selected papers from the ThirdAmsterdam Colloquium (GRASS 2),ed. Jeroen Groenendijk, Theo M. V.Janssen,and Martin Stokhof,1– 41. Dordrecht: Foris.] Kamp,Hans, and Uwe Reyle.1993. Fromdiscourse to logic:Introduction to modeltheoretic semantics of naturallanguage, formal logic and Discourse Representation Theory. Dordrecht:Kluwer. Karttunen,Lauri. 1969. Pronouns and variables. In Papersfrom the Fifth Regional Meeting of theChicago LinguisticSociety, 108–116.Chicago Linguistic Society, University of Chicago,Chicago, Ill. Karttunen,Lauri. 1977. Syntax and semantics of questions. Linguisticsand Philosophy 1:3–44. Kayne,Richard. 1994. Theantisymmetry of syntax. Cambridge,Mass.: MIT Press. Keenan,Edward L. 1988.Complex anaphors and Bind Alpha. In CLS 24, 216–232.Chicago Linguistic Society,University of Chicago,Chicago, Ill. Koopman,Hilda, and Dominique Sportiche. 1982. Variables and the Bijection Principle. TheLinguistic Review 2:139–160. Kurtzman,Howard S., and Maryellen C. MacDonald.1993. Resolution of quantifier scope ambiguities. Cognition 48:243–279. Lasnik,Howard. 1976. Remarks on coreference. LinguisticAnalysis 2:1–22. Lasnik,Howard, and Tim Stowell.1991. Weakest crossover. LinguisticInquiry 22:687–720. Lewis,David. 1975. Adverbs of quantification. In Formalsemantics of natural language, ed.Edward L. Keenan,3 –15.Cambridge: Cambridge University Press. Lo¨bner,Sebastian. 1985. Definites. Journalof Semantics 4:279–326. May,Robert. 1977. The grammar of quantification. Doctoral dissertation, MIT, Cambridge, Mass. May,Robert. 1985. LogicalForm: Its structure and derivation. Cambridge,Mass.: MIT Press. Partee,Barbara. 1989. Binding implicit variables in quantified contexts. In CLS 25, 342–365.Chicago LinguisticSociety, University of Chicago,Chicago, Ill. Postal,Paul. 1971. Cross-overphenomena. New York:Holt, Rinehart and Winston. Reinhart,Tanya. 1976. The syntactic domain of anaphora.Doctoral dissertation, MIT, Cambridge, Mass. Reinhart,Tanya. 1983. Anaphoraand semantic interpretation. London:Croom Helm. Reinhart,Tanya. 1987. Specifier and operator binding. In Therepresentation of (in)definiteness, ed. Eric J. Reulandand Alice G. B.terMeulen, 130– 167. Cambridge, Mass.: MIT Press. Reinhart,Tanya. 1992. Wh-in-situ:An apparent paradox. In Proceedingsof the Eighth Amsterdam Collo- quium, 483–492.ITLI, University of Amsterdam. Reinhart,Tanya. 1997. Quantifier scope: How laboris divided between QR andchoice functions. Linguistics andPhilosophy 20:335–397. Reinhart,Tanya. 1998. Wh-in-situin the framework of theMinimalist Program. NaturalLanguage Semantics 6:29–56. Ruys,E. G.1992.The scope of indefinites. Doctoral dissertation, Utrecht University. Ruys,E. G.1994a.Economy conditions and the syntax-semantics interface. Ms., Utrecht University. Ruys,E. G.1994b.A globaleconomy analysis of weak crossover. In Linguisticsin theNetherlands 1994, ed.Reineke Bok-Bennema and Crit Cremers, 223– 234. Amsterdam: John Benjamins. Ruys,E. G.1997.On quantifier raising and ‘ ‘Move-F.’’ In UtrechtInstitute of Linguistics OTS yearbook 1997, 83–104. Utrecht Institute of Linguistics OTS, UtrechtUniversity. Ruys,E. G.Toappear. Unexpected wide scope phenomena. In Thesyntax companion, ed.Henk van Riems- dijkand Martin Everaert. Ruys,E. G.Inpreparation. On ‘ ‘Weakestcrossover.’ ’ Ms.,Utrecht University. WEAKCROSSOVERASASCOPEPHENOMENON 539

Safir,Ken. 1984. Multiple variable binding. LinguisticInquiry 15:603–638. Safir,Ken. 1996. Derivation, representation, and resumption: The domain of weak crossover. Linguistic Inquiry 27:313–339. Sharvit,Yael. 1999. Functional relative clauses. Linguisticsand Philosophy 22:447–478. Stowell,Tim. Toappear. Adjuncts, arguments, and crossover. NaturalLanguage & LinguisticTheory. Wasow,Thomas. 1979. Anaphorain generative grammar. Ghent:E. Story-Scientia. Winter,Yoad. 1997. Choice functions and the scopal semantics of indefinites. Linguisticsand Philosophy 20:399–466. Winter,Yoad. 1998. Flexible Boolean semantics: Coordination, plurality and scope in natural language. Doctoraldissertation, Utrecht University.

UtrechtInstitute of LinguisticsOTS UtrechtUniversity Trans 10 NL-3512JK Utrecht TheNetherlands [email protected]