Control in Icelandic and Theories of Control Author(s): Cedric Boeckx and Norbert Hornstein Reviewed work(s): Source: Linguistic

Inquiry, Vol. 37, No. 4 (Autumn, 2006), pp. 591-606 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4179385 . Accessed: 30/07/2012 14:13
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

Remarks and Replies

Control in Icelandic and Theories of Control Cedric Boeckx NorbertHornstein
This article examines a pervasive argumentagainst a movement approachto controlbased on Icelandicconcord facts. We show that the argumentdoes not underminethe movementapproachwhen the facts areconsideredin theirentirety.The facts divide into two basic groups: instances of quirkyCase assignmentand instances of structural Case sharing. The former require some theoretical adjustmentsregarding NPs in orderto be incorporated multiplyCase-marked into a movement We show thatthe adjustments neededmay be independently approach. required,and may be even more problematicfor alternativeviews on control. Keywords:Case, concord, control, Icelandic, movement, 0-role

1 Introduction Recently the natureof controlin Icelandichas animateddebatesin syntactictheory.In particular, Landau(2003:491-493) argues1that the control data discussed by Sigur6sson (1991) provide a strong argumentagainst a movement theory of control of the type advocatedin Hornstein 1999, 2001, 2003, and in Boeckx and Hornstein2003, 2004.2 We thinkthatcontrolin Icelandicis worth discussing at some length, because, as we show here, Sigur6sson'sdataappearto raise interesting issues and difficulties for all currentapproachesto control. Icelandicexhibits agreementpatterns in controlstructures which, understandard, PRO-based analyses of control,suggest thatthe antecedentand the controlledPRO carrydistinctCase values

This article grew out of materialdiscussed in a seminar on Icelandic syntax that we had the pleasure to co-teach with KjartanOttosson at the University of Marylandin the spring of 2003. We would like to thank Kjartanfor his invaluableinputand for checking the statusof countless Icelandicsentences for us. We also would like to thankthe other in the seminar,the audience at the LSA annualmeeting (January2005), Halld6rSigur6sson, and three very participants scrupulousLI reviewers. ' Although to our knowledge Landau2003 is the first article to make the argumentin print, the control facts in Icelandic were repeatedlybroughtup as problematicfor a movement analysis of control when the latter was presented at conferences. 2 The movement theory of control we have in mind denies the existence of PRO and claims that "control" arises from movement from one thematicposition to the other. We have nothing to say about approacheslike those advocated by Kayne (2002) or Martin(1996) that assume the existence of PRO while adopting a movement relation between the controllerand the controllee.
Linguistic Inquiry,Volume 37, Number4, Fall 2006 591-606 (? 2006 by the MassachusettsInstituteof Technology

591

592

REMARKS

AND

REPLIES

(see section 2 for data and discussion). The data all have the same form. In configurationslike (1) the controller,NP1, has a Case distinctfrom thatrealizedon a floating quantifieror secondary predicatein the embedded clause. The overt Case on the latter elements is taken to reveal the covert Case on the embeddedPRO. (We remainuncommittedat this point aboutwhat the internal structureof the complementclause is. For discussion, see section 3.)
(1)
...

NPaCase

...

...] [ ... PRO'pCasefloating Q/secondarypredicatepcase

It is temptingto conclude, on the basis of (1), that controlcannot be movement, since underthat view, PRO reduces to an unpronouncedcopy ("trace") of its antecedent,whose Case value is expected to match that of the "antecedent" copy.3 Put more positively, configurationslike (1) as supportingthe classical theory of control, which relies on the existence have been interpreted (A-)traces,to mediatethe control of PRO, an element differentin kind from movement-generated relation manifestedin obligatorycontrol constructions. The aim of this article is to argue that it pays to look before one leaps, even if it is merely to theoreticalconclusions. More specifically, we argue that no currentlyentertainedtheory of control can accommodatethe reportedIcelandic data without alteration.Thus, these data cannot by themselves argue eitherfor a PRO-basedaccount of control or against a movement-based account, as is tacitly assumed and occasionally asserted. Moreover,we believe thatthe Icelandicdata,when properlyconsidered,arguefor a stronger require conclusion. In section 2, we describe the Icelandic data and show that control structures Case matching in configurationslike (1) when structuralCase is at issue. In other words, the Case clash schematizedin (1) is only really possible when at least one of the two relevantCases is oblique. This suggests that whateverstory we tell for Icelandicmust be able to accountfor the lack of Case clash in some instances of control, and we argue that this is very easy to do if one assumes a movement theory and less easy if one does not. (For a related argument,based on Case concord in Latin and Italian, see Cecchetto and Oniga 2004.) In section 3, we focus on the theoreticalimplicationsof Icelandic control for a movement of the movementaccountappliedto Icelandic theory, and we provide a technicalimplementation in the form of multiple Case assignment to chains in well-defined contexts. In section 4, we discuss some of the consequences of our analysis for nonmovementtheories of control. Section 5 is the conclusion. 2 Control in Icelandic Icelandic has a morphologicallyrich Case/agreementsystem, and some of that richness shows Considersome basic facts. (For a comprehensivereview, see Sigur6sson up in controlstructures. 2003.) Overt morphologicalagreementon finite verbs (person,number)and passive past partici3This is not how Landauformulateshis argument.Landautakes the existence of Case-markedPRO to be lethal to Homstein's view of control. As we show below, it is importantto focus the discussion on Case values in order to understand the way control works in Icelandic. Landaualso observes that in contrastto control, raising disallows situationswhere a single NP appearsto receive two Cases. We returnto this difference between raising and control in section 3.

REMARKS

AND

REPLIES

593

Case (2). Quirky ples (Case, number,gender)can only take place with elementsbearingstructural subjects cannot participatein (overt) agreementin those contexts (3). (2) Straikarnir voru a6sto6a6ir/*a6sto6a6.
the.boys.NOM were.3PL aided.NoM.PL.MASC/NoM.SG.NEUT(default)

'The boys were helped.'

hjalpa6. (3) Strakunum var the.boys.DAT was.3sG helped.NoM.SG.NEUT(default)

'The boys were helped.' By contrast,all elements-both structurallyCase-markedand quirkily Case-marked-are able to agree (in Case, number,gender) with secondarypredicatesand floating quantifiers(4)-(5).4 i skola. komust allir (4) a. Strakamir in school all.NoM.PL.MAsc got the.boys.NoM.PL 'The boys all got to school.' drukknir. b. Strakamir hittu kennarann met the.teacher.Acc.sG.MAsc drunk.NoM.PL the.boys.NoM.PL 'The boys met the teacherdrunk(i.e., the boys were drunk).' i skola. (5) a. Strakunum leiddist ollum/*allir in school bored.3sGall.DAT.PL.MAsc/NoM.PL.MAsc the.boys.DAT 'The boys were all bored in school.' ovir6ingu drukknum. b. Strakarnir syndu kennaranum disrespect drunk.DAT.SG.MASC showed the.teacher.DAT.sG.MAsc the.boys.NoM 'The boys showed the teacherdisrespect (when he was) drunk.' Because secondary predicatesand floating quantifiersovertly display the Case of the NPs they relateto, Sigur6sson( 1991) uses themto addressthe questionof which Case, if any, the controlled element bears. (Sigur6ssonassumes thatthe controlledelement is PRO, and phrasesthe question in terms of PRO, but we want to formulateit in terms as neutralas possible.) of the facts is thatthe case morphologyon the floatingquantifier/ Sigur6sson's interpretation secondary predicatesuggests that the controlled element in Icelandic can bear either structural or quirkyCase, as sentences (6) and (7) show.5

4 This agreementis often called concord. We would like to avoid using this term, which is less theory-neutral than Case matching. At this point, we are agnostic about how Case matchingis achieved (e.g., by direct agreementbetween and FQ/SP; etc.). In controller(or PRO undersome theories) and FQ/SP; by simultaneousassignmentto controller/PRO of Case matching.For now, we ask the readerto understand section 4, we commit ourselves to a specific implementation featureson NPs and FQs/SPs. phenomenaas simply denoting occurrencesof identical/matching concord/agreement 5 An importantcaveat is in order here. Thrainsson(1979:chap. 5) notes that "Equi [/control] constructionsare one." Indeed, frequentlyratherbad when the deleted subject[i.e., PRO] should have been an oblique (or non-nominative) KjartanOttosson (pers. comm.) observes that there is a lot of uncertaintyconcerning the Case markingof FQs/SPs in control complements. This is confirmed by Sigurosson (1991:332), who reports that "many speakers are reluctantto embed floated quantifiersinto control infinitives, presumablyfor scope reasons." Sigur6sson's suggestion is puzzling since the correspondingEnglish sentences with floating quantifiersare fine. This strongly suggests that something less

594

REMARKS

AND

REPLIES

ba6 Bjaama a6 koma einan. (6) Jon Jon.NoM asked Bjarni.Accto come alone.Acc 'Jon asked Bjami to come alone.' (7) Jon ba6 Bjaama a6 lei6ast ekki einum. asked Bjami.Acc to be.borednot alone.DAT Jon.NoM 'Jon asked Bjarni not to be bored alone.' Closer examinationreveals that Icelandicoffers the whole range of logical possibilities when we look at the Case of the FQ/SP (floating quantifier/secondary predicate)in the embeddedclause in conjunctionwith the Case of the controller.(In what follows, we restrictattentionto the case morphologyof elements that we can see on the surface,leaving the questionof the existence and Case markingof PRO for subsequentsections.) Apartfrom "structuralCase on the antecedent and structural Case on the FQ/SP" (6) and "structuralCase on the antecedentand quirkyCase on the FQ/SP" (7), Icelandic allows "quirkyCase on the antecedentand structural Case on the FQ/SP" (8) and "quirky Case on the antecedentand quirkyCase on the FQ/SP" (9).6 (8) Bjaama langa6i ekki til a6 hlaupaeinn. Bjarni.Accwanted not to to run alone.NoM 'Bjami wanted not to run alone.' (9) Bjaama langabi ekki til a6 lei6ast einum. Bjami.Acc wanted not to to be.bored alone.DAT 'Bjami wanted not to be bored alone.' Summing up so far, we have observed the following situations: NP ... [structural (10) structural FQ/SP ...] (11) structural NP ... [quirkyFQ/SP ... ]

universal than scope is at issue. In fact, Sigur6sson's claim is only valid for instances of quirky-Case-marked floating quantifiers,as structurally Case-marked floating quantifiersinside controlcomplementclauses do not give rise to marginality. (i) Strakarnir vonast til [a6 PRO koma allir]. the.boys.NoNhope to to come all.NoM 'The boys were hoping to all come.' Despite these ratherimportantreservationsconcerning the acceptabilityof the relevant Icelandic data, we will restrict attentionto those speakersof Icelandic who have internalizedthe empiricalpicture describedin Sigur6sson 1991. For other speakers,such as Thrainsson,Icelandic behaves exactly like English. 6 Instancesof object controlconstructions where the object controllerbearsquirkyCase are very limited (we suspect thatthis is due to the very narrowsemanticrangeof verbs takingquirkyobjects (mostly experiencerpredicates)).Kjartan Ottosson (pers. comm.) provides the following example: (i) Maria skipa6i J6ni a6 vera twgur. Maria orderedJon.DAT to be well.behaved.NoM 'MariaorderedJon to be well behaved.' Examples like (i) can be readily accommodatedunderour analysis. They would be derived along the lines of (32)-(33), except that matrixTc'would not be involved; matrix V? would be the final landing site of the controller.

REMARKS AND REPLIES

595

(12) quirkyNP ... [structural FQ/SP ...] (13) quirkyNP ... [quirkyFQ/SP ... ] quirky}Case quirky}Case on the antecedentand {structural, Note that when we say "{structural, on the FQ/SP," we do not necessarilymean thatthe Case values must be identical.For example, they differ in (9): quirkyaccusative NP controllerand quirkydative FQ/SP. However, when the full rangeof Case values is takeninto account,the picturebecomes morecomplex and interesting. In particular,(10) splits into two possibilities, as illustratedin (14) and (15). a6 koma einan/??einn. (14) Jon bad Bjarna Jon.NoM asked Bjarni.Accto come alone.Acc/alone.NoM 'Jon asked Bjarni to come alone.' vonast til [a6 koma einn/*einan]. (15) Jon Jon.NoM hopes to to come alone.NoM/alone.Acc 'Jon hopes to come alone.' As (14) indicates,if the antecedentis accusative,the floating quantifiermay be either accusative or (quite marginally,for many speakers7)nominative.But, as can be gatheredfrom (15), if the antecedentis nominative,then the floatingquantifiercan only be nominative.This, to us, suggests Case of the antecedentand the structural Case of between the structural some interdependence the FQ/SP. This interdependence is supportedby (14): when the structural Case value of the antecedent is accusative, though nominative is marginallypossible on the floating quantifier,accusative is stronglypreferred.We take this to indicatethat in such situations,nominativeis really a marked default Case realization.8 The appearanceof default nominativeCase in nonfinite contexts is actually quite common in Icelandic.For example,it is attestedin othernonfinitecontexts such as (16), where the nominanominative.(See tive element fails to triggeragreementon the finite verb, unlike true structural Boeckx 2000, 2003b, Sigur6sson 1996, 2000, 2003 for extensive discussion of such examples. and Gutmann2006. On defaultCase, On defaultnominativeCase in Icelandic,see also Frampton see Schuitze2001b.) hefur alltaf fundist [vi6 gafud]. (16) a. Peim them.DAT have.3sGalways found we.NoMboring 'They have always found us boring.' b. Mer fannst/*fundust[henni lei6ast beir]. her.DAT be.boredthey.NoM me.DAT seemed.3sG/3PL 'I thought she was bored with them.'
7 Halld6rSigur6sson (pers. comm.) indicates that for him, the nominativeform in (14), (17), (18), and (20) is fully acceptable.Our interpretation of his judgments is that he has generalizedthe use of default nominativeCase in nonfinite contexts. 8 See Andrews 1982 and Hornstein 1990 for claims along those lines.

596

REMARKS

AND

REPLIES

What is importantfor presentpurposesis that once the rathermarkedavailabilityof default nominativeCase on floatingquantifiers is factoredout, (10) is just like English:the same structural on the Case must appear controllerand an embeddedFQ/SP in a control structure. (11) behaves as expected: structural Case on the antecedentmay be either nominative(17) or accusative (18), and Case on the FQ/SP must be quirky (dative in the example at hand).9 (17) Jon ba6 Bjaama a6 lei6ast ekki einum/*einan/*einn. asked Bjarmi.Acc to be.bored not alone.DAT/Acc/NoM Jon.NoM 'Jon asked Bjarninot to be bored alone.' (18) Jon vonast til [a6 lei6ast ekki einumn*einan/*einn].
Jon.NoM hopes to to be.bored not alone.DAT/Acc/NoM

'Jon hopes not to be bored alone.' For the situationin (12), the range of quirkyelements is not surprising.What may be unexCase is limited to nominative.Witness (19). pected is that structural (19) a. Bjarma langa6i ekki til a6 hlaupaeinn/*?einan. Bjarni.Accwanted not to to run alone.NoM/ACC 'Bjarniwanted not to run alone.' b. Bjaama leiddist a6 hlaupaeinn/*?einum. Bjarmi.DAT was.boredto run alone.NoM/DAT 'Bjarniwas bored to be runningalone.' Again, we take this nominative on the secondarypredicateto be a default Case, as there is no source for structural nominativein the embeddedclause. Finally, (13), illustratedin (20), shows the full range of quirky Case values for both the antecedentand the FQ/SP, and no interdependence between the two elements. Nominative Case (or accusative Case) on the floating quantifieris clearly ruled out. Quirky Case matching the Case of the antecedenton a floating quantifieris ruled out as well. (20) a. Bjarna langa6i ekki til ad lei6ast einum/*einan/*einn. Bjarni.Accwanted not to to be.bored alone.DAT/Acc/NoM 'Bjami wanted not to be bored alone.' b. Bjarna leiddist a6 vanta einan/*einum/*einni veisluna. was.boredto be.missing alone.Acc/DAT/NoM Bjarni.DAT from the.party 'Bjarniwas bored not to be alone at the party.' Because structural Case values are those thatreveal somethingaboutstructure, and aboutderivations, let us rewrite (10)-(13) as (2l)-(24) (glossing over the idiosyncrasies of quirky Case values).

9 Nominative Case on floating quantifiersin (18) is sometimes reportedto be ?* as opposed to *. Accusative Case in the same context is clearly *.

REMARKS

AND

REPLIES

597

(21) a. nominativeNP ... [nominativeFQ/SP ... ] default nominativeFQ/SP] b. accusative NP ... [accusativeFQ/SP/(marginally) (22) a. nominativeNP ... [quirkyFQ/SP ...] b. accusative NP ... [quirkyFQ/SP ...] (23) quirkyNP ... [defaultnominativeFQ/SP ...] (24) quirkyNP ... [quirkyFQ/SP ... ] There are some noteworthypatternsin the structuressummarizedin (21)-(24). What comes out of this is the following: (a) Icelandic speakersstronglypreferto match the Case is involved and (b) they require Cases on the antecedentand on the FQ/SP when structural the local quirkyCase on FQs/SPswhen availableand stronglyrejectremotequirkyCase matching. Case configurations Thus, speakersclearly keep instancesof quirkyCase separatefrom structural (see footnote 5). As already noted, when it comes to structuralCase values, Icelandic control is just like English control. The only genuine instances of multiple Case assignment arise where multiple quirkyCase values areassigned.However,if quirkyCase is a kindof inherentCase, as conventionally assumed(i.e., a Case that is tightly connectedto 0-role assignmentas opposed to agreement, as Chomsky (1986) argued),10 the Icelandic facts are once again not at all that different from view) bears a 0facts of English. The controlledelement in English ("PRO" underthe standard role distinct from that of its antecedent,just as the controlledelement in Icelandic (as reflected on the FQ/SP) bears a quirkyCase value distinctfrom that of its antecedent.The only difference between English and Icelandicis that,unlike English, Icelandicmorphologicallyreflects 0-marking in its Case system. 3 Control as Movement Armed with the facts schematizedin (21)-(24), we can now returnto the main theoreticalissue of this article:how does the movement theory of control fare in the face of the Icelandic facts? Although Landau (2003) claims that the Icelandic control facts are incompatible with a movementtheory,we would like to argueotherwise.The movementanalysisassumesthata chain bears one structural giving rise to a control structure Case (just like in raisingconstructions)and in two 0-roles (unlike in raising). For configurations which only structural Case is at issue, this implies thatthe Case on the FQ/SP shouldmatchthaton the controller.Modulo the rathermarked since option of default nominativeCase on FQs/SPs, this is exactly what we find. Furthermore, the essence of the movement theory of control is that it allows-in fact, requires-multiple 0role assignmentto a single chain, examplesof multipleinherent/quirky Cases in controlstructures in Icelandic simply follow from the connection between 0-role and inherentCase. This point bears emphasis. The argumentagainst a movement theory of control based on Icelandicrestson the premisethatthe movementtheorycannotbe rightbecauseonly one Case may
10

See, for example, Freidin and Sprouse 1991.

598

REMARKS

AND

REPLIES

be assignedto a chain (Landau'sargument). Call this the chain uniformity argument.Specifically, Icelandic shows that two Cases are assigned. Since two Cases are assigned, and since two copies of one and the same element cannotbeardistinctCases, two chains must be involved: one headed by PRO, and the other headed by PRO's antecedent.Grantedthe premise, this is a good prima facie argumentfor an independentPRO. However, the premisedoesn't hold. The argument raised against the movement theory does not make allowance for the central theoreticalclaim of the movement theory of control, namely, that movement into 0-positions is licit in principle. Once that assumptionis granted,it is not clear why we should keep to the assumptionthat only one inherentCase may be assigned per chain, given the thematicnatureof inherentCase. And, since we have shown that there is no evidence that multiple structural Cases are assigned to a chain, the argumentagainst a movement theory of control dissolves. Two questions remainto be addressedby the movementanalysis of control:(Ql) why only one Case value is reflected on the NP (the otherCase value, if any, being realizedonly indirectly on FQs/SPs), and (Q2) why that Case has the value it does. Before getting into technicaldetails, we would like to observe that we take it as a theoretical strengthof the movementtheory that the kind of answerwe must provide to questionslike those just raised is very clear. For instance,for Q2, it seems sufficient to assume that whateverensures that the highest copy of a moved element is pronouncedwill ensure that the Case assigned by the highest Case assigner will surface on the moved NP. The Case on an FQ/SP would be akin to pronouncingan element strandedundermovement. Its morphologywill be determinedby its immediateenvironment(e.g., by the adjacentverb)." Let us now be a bit more specific about the technical implementationswe would like to pursue. RegardingQl, let us assume, along with much recent work, that Case assignmenttakes place strictly cyclically, as soon as the properenvironmentis formed-that is, as soon as the probe is introducedinto the derivation.'2For Icelandic, this means specifically that quirkyCase is assigned under0-role assignment(first Merge), whereasstructural Case is assigned upon introduction of the relevantfunctionalhead (v for accusative, (finite) T for nominative). Regardingthe "Case-matching"facts discussedin section2, we would like to takeagreement on FQs/SPs to be similar in natureto past participleagreementas analyzed in Chomsky 2004, Hiraiwa2001, 2004, and Boeckx 2003a,c, 2004. That is, we would like to analyze these facts as an instance of multiple Agree-not as a direct agreementrelation between an NP and the FQ/ SP, but as an indirect relation between the two, mediated by a Case-assigning head. In other words, Case matching arises as a result of simultaneousvaluationof the NP and the FQ/SP by a functionalhead F, as schematizedin (25). '3 (For concreteness,we assume that the relevantNP
" The situation is reminiscent of theories of resumptionthat treat resumptive pronouns as overt copies of their antecedents.A wh-antecedentdoes not force wh-morphologyon the resumptivepronoun. 12 This is sometimes referredto as the "virus" theory of the cycle, following Uriagereka(1998). 13 We do not claim that all instances of the agreement configurationsconsideredhere are cases of multiple Agree. See Boeckx 2004:33n6 for empirical argumentsthat some such cases can be achieved directly (agreementbetween NP and predicate)or indirectly(multiple Agree). It is in fact not implausible to think that different features may participatedifferently in such agreements:say, interpretable/valued featureslike + on NP may agree directly with secondarypredicates,while uninterpretable/unvalued

REMARKS

AND

REPLIES

599

and the FQ/SP enterthe derivationas a small clause, in a mutualc-commandrelation.An adjoined structure,where the FQ/SP is an XP adjoinedto the NP it modifies, would also be compatible with our analysis.) (25) F
...

[[NP] [FQ/SP]]

simultaneousCase assignment Once this assumptionconcerning the natureof Case matchingis made, it follows that FQs/SPs will bear quirkyCase if the relevantNP receives quirkyCase upon first Merge; otherwise, they Case is introduced.'4The two possibilities will have to wait until the probe assigning structural schemas; are schematizedin (26)-(3 1). (Relevantexamples arerepeatedwith theircorresponding original example numbersare in brackets.(27) is the derivationwe propose for (26), (29) is the derivationwe propose for (28), and (31) is the derivationwe propose for (30).)15 (26) a. nominativeNP ... [quirkyFQ/SP ... ] (= (22a)) b. Jon vonast til [a6 lei6ast ekki einum]. (= (18)) Jon.NoM hopes to to be.borednot alone.DAT 'Jon hopes not to be bored alone.' (27) NPi To ... t'i V0 ... [Tinf. .. V0 [ti FQ]] Step 1: embeddedV? assigns a 0-role/quirkyCase to NP and quirkyCase to FQ Step 2: matrixV0 attractsNP and assigns a 0-role to it Case to NP, which moves to check EPP Step 3: matrixTo assigns structural (28) a. quirkyNP ... [quirkyFQ/SP ... ] (= (24)) b. Bjarna langa6i ekki til a6 lei6ast einum. (= (20a)) Bjarni.Accwanted not to to be.bored alone.DAT 'Bjarniwanted not to be bored alone.' (29) NP, To . .. t'i V0 ... [Tinf... V0 [ti FQ]] Step 1: embeddedV0 assigns a 0-role/quirkyCase to NP and quirky Case to FQ Step 2: matrixV0 attractsNP and assigns a 0-role/quirkyCase to it Step 3: NP raises to matrixTo to check EPP

features like Case-on which we focus here-may come to be shared via multiple Agree. We hope to returnto these issues in future work. 4 It is plausible to assume that this delay makes room for a marked,default Case assignmentoption. See below. 5In all the derivationsthat follow, we remain agnostic about whetherthe controllermoves throughthe embedded Spec,TP (the position standardlyassumed to be PRO's final landing site). The issue of movement throughthe specifier position of a nonfinite clause touches on the natureof the Extended ProjectionPrinciple (EPP), which is tangentialto the questionsraised in this article.As far as we can tell, our resultsare unaffectedby movementof the controllerthrough embedded Spec,TP or lack thereof.

600

REMARKS AND REPLIES

(30) a. nominativeNP ... [nominativeFQ/SP ... ] (= (21a)) b. Jon vonast til [a6 koma einn]. (= (15)) Jon.NoM hopes to to come alone.NoM 'Jon hopes to come alone.' (31) NPi To ... t'i V0 ... [Tinf ... V0 [ti FQ]] Step 1: embeddedV? assigns a 0-role to NP Step 2: matrix V0 attractsNP and assigns a 0-role to it nominativeCase to NP and FQ by multipleAgree Step 3: matrixTo assigns structural Now because quirkyCase assignmentcannottake place long-distance(a fact aboutinherentCase in general), an embeddedFQ/SP will have to bear default (nominative)Case (a morphological aboutIcelandic)if no quirkyCase is assignedin the embeddedclause andno structural requirement Case is assigned in the matrixclause (32)-(33). (32) a. quirkyNP ... [nominativeFQ/SP...] (= (23)) b. Bjarna leiddist a6 hlaupa einn. (= (19b)) was.boredto run alone.NoM Bjarni.DAT 'Bjarniwas bored to be runningalone.' (33) NP, To ... t'i V0 ... [Ti,f ... V? [ti FQ]] Step 1: embeddedV? assigns a 0-role to NP Step 2: matrixV0 attractsNP and assigns a 0-role/quirkyCase to it Step 3: NP raises to matrixTo to check EPP Step 4: FQ receives default Case (26)-(33) captureall the facts in (21)-(24). As for the Case value that surfaces on the moving element (Q2 above), it is always the highest Case value. We claim that this follows from the same mechanismthat ensures that the highest copy in a chain is pronounced(see especially Nunes 1999, 2004 for extensive discussion). A more subtle question (Qi) is this: why don't the two Case values surface in situations like (21)-(24)? We could stipulate this as a morphological fact.16 Alternatively, we could follow Marantz(1991) and Harley (1995) and arguethat Case values are assigned in the morphological component.One way of formalizingthis intuitionis to say that Case is checked in narrowsyntax (i.e., Case gets valued as [ + Case-marked] as opposed to [- Case-marked], but it is morphologically fixed (as nominative,dative, etc.) upon Spell-Out(or Transfer,in Chomsky's (2004) terminology)). This would fit nicely with a DistributedMorphology view of things. Say that this is so. It is then plausible to assume that Case is morphologicallyrealized only once (just as only

16 This has been done for exceptionalCase marking(ECM) in Korean-stylelanguagesby those who offer a raising/ Case-stackinganalysis (see Yoon 1991).

REMARKS

AND

REPLIES

601

one memberof a chain is pronounced),17 accordingto the context in which the NP is pronounced (highest copy).18 Let us now turn to a question we have not yet addressed:why can nominative Case be assigned indirectlyin controlas a rathermarkedoption (accusativeCase on the moving element and nominative on secondarypredicatesas in (21b)), whereas raising cannot avail itself of this default nominativeoption (see (34))?19 (34) Jon taldi Bjarnaj hafa hlaupi6 einanj/*einnj.
alone.Acc/NoM Jon.NOMconsideredBjarni.Acchave run

'Jon consideredBjarnito have run alone.' One way to make sense of the contrastis by looking at how the derivationsproceed in (21b) and (34). For (34), we propose the following derivation: (35) NP, vo ... V0 ... [Ti,f ... V0 [ti FQ]] Step 1: embeddedV? assigns a 0-role to NP accusative Case to NP and FQ by multiple Agree Step 2: matrixvo assigns structural Step 3: NP raises to matrix Spec,vP (to check EPP) For (21b), repeatedwith the appropriate example, we propose the derivationin (37) (cf. (29)). (36) a. accusativeNP ... [accusativeFQ/SP/defaultnominativeFQ/SP...] a6 koma einan/??einn. (= (14)) b. Jon bad Bjarna asked Jon.NoM Bjarni.Accto come alone.Acc/NoM 'Jon asked Bjarnito come alone.' (= (21b))

17Let us draw an analogy. In the SPE system (Chomsky and Halle 1968), a syllable may be markedfor [ + stress] multiply (on more than one cycle), but this does not mean that that syllable is more stressedthan a syllable thatreceives stress only once (as in monosyllabic words, where there is only one cycle). 18The presentremarksmay help to explain why morphologicalCase stacking is possible (as in Korean(i)). (i) Na-eykey-ka paym-i mwusepta. I-DAT-NOM snake-NoM fearful '(Only) I am afraidof snakes.' As discussed by Yoon (1996, 2004), Schutze (2001a), Jo (2002), and Hong (2002), Case stackingin Koreaninduces very special interpretiveeffects such as topic/focus reading. It is not implausibleto think that multiple Case values may be overtly realized if each value is associated with a special outputeffect. 19Such a disparitybetween control and raising is sometimes taken to be lethal to a movement theory of control. Consider for instance Culicover and Jackendoff's (2001) argumentthat the movement theory of control is inadequate because control in nominals is licit (i), but raising is not (ii). (i) John's attemptto leave

(ii) *John's appearanceto leave As we point out in Boeckx and Hornstein2003:sec. 4, any contrastbetween raising and control would be lethal if a raising theoryof controlwere entertained. But no one (rightly)entertainssuch a theory.Whatis entertained is a movement theory of control. The fact that raising is the prototypicalcase of A-movement is a useful thing, as it can serve as a theoreticalreference at various points, but because the frameworkwe assume (the principles-and-parameters approach) dispenses with construction-specificrules, the parallelismbetween control and raising should be abandonedwhen the subtle differences between the two become crucial.

602

REMARKS

AND

REPLIES

(37) NP, vo ... t'i V0 ... [Ti,f ... V? [t, FQ]]

Step 1: embeddedV? assigns a 0-role to NP Step 2: matrixV0 attractsNP and assigns a 0-role to it accusative Case to NP and FQ by multiple Agree Step 3: matrixv? assigns structural Note thatupon structural Case assignment,the two goals are in differentclauses in (37) (because of remergerof NP into the matrix VP for thematic reasons). This is not the case in (35). We speculate that the markeddefault nominative Case on the floating quantifierin (14)/(36b) is a distance effect. That is, in contrast to what we find in the configurationsof multiple Agree discussed above, where both the NP and the FQ/SP are in a small clause, mutual c-command relation,in (37) the NP and the floating quantifierare separated by an infinitivalclause boundary. Although, as far as we can see, nothing in the existing formulationsof multipleAgree inherently blocks multiple Agree across clauses, speakerswho marginallyaccept nominative in (36) may relax the obligatorinessof multiple Agree in such cases. This is just a speculation,and clearly not a deep explanation.But it is importantto bear in mind that nominativeassignmentin (36) is marginal.So we may not want to let this markedoption emerge from a deep mechanismof the grammar. Treatingnominativeassignmentin (36) as a relaxationof the obligatorinessof multiple Agree resulting from the distance between the two targets of Agree is one way of capturing marginalfacts without deriving them in a deep way. Let us take stock of what we have done in this section. We have proposeda possible implementationof Case stacking in control. We have addressedtechnical issues that arise once Case stacking is involved: how many Case values are allowed to surface, and, if not all of them can surface,which one does? In the course of answeringthose questions,we have made the following two assumptions: (a) Case is valuedas soon as possible, and (b) Case values are fixed morphologically in the PF component. In addition, we have assumed, as is standard,that inherent(in our case, quirky)Case cannot be assigned long-distance,and that secondarypredicatesrequiresome Case/+-featuremarking(sometimes these are default markings)in Icelandic. It is importantto note that such assumptionsare not specific to control. In fact, all of them have been proposed independently. These assumptionsturnedout to be sufficient once the core factual descriptionwas made in section 2: thatmultipleCase assignmentis availableonly when quirkyCase is.20In the absence of quirky Case, structural Case is sharedby the controllerand the FQs/SPs. 4 Nonmovement Alternatives In the previous section, we developed an analysis of the controlfacts in Icelandic undera movement theory of control. In this section, we will briefly discuss some implications of the same
20 The same holds true in Korean Case-stackingconfigurations.The only acceptableexamples involve stacking a structuralCase on top of a quirky one. As mentionedin footnote 18, KoreanCase stacking has discourse effects such as introducinga focus or topic reading. However, which structural Case one finds on top of the quirky Case has been shown to fall underCase theory. See Hong 2002, Jo 2002, and Yoon 2004 for discussion. This overt ordermakes sense if quirky Case is lexical while structural Case is determinedby an element outside the domain of a lexical head.

REMARKS

AND

REPLIES

603

control facts for nonmovementalternatives.The alternativeswe have in mind are the null Case approach (ChomskyandLasnik 1993) and the classic accountbased on the PRO Theorem(Chomsky 1981). (We assume basic familiaritywith these approachesand will not review them here.) At first sight, the null Case approachto control appearsadequate,because the essence of the theory is that it allows-in fact, requires-PRO to be Case-markedto be licensed. This is exactly what the Icelandicfacts point to. However, upon closer scrutiny,the Icelandicfacts under discussion pose a nontrivial problem for the null Case theory. This is because the null Case approachrequiresPRO to bear a specific Case value (dubbed "null"), and not just any Case. Otherwise,proponentsof the null Case theory have no explanationfor why PRO is null, a fact thatany theoryof controlmust addressand answer.Of course, proponentsof the null Case theory could claim, as we did above, thatcontrolin IcelandicrequiresmultipleCase assignmentto PRO: null Case (ensuringthe phoneticallynull characterof PRO), and some other Case (structural or But this supports our claim that no extant theory of control can capture the inherent/quirky). Icelandic facts without modifications. The null Case approachfaces one problem that is worth mentioning. In contrast to our proposalabove, the proposalthat null Case can stack leads to the following problemwhen structuralCase occurs on FQs/SPs withincontrolclauses. It is unclearhow the structural Case assigner in the matrix clause would be allowed to see past a null-Case-bearing PRO to assign (nonnull) structural Case to the secondarypredicatewithout violating Minimality.The problemis schematized in (38).
(38) antecedent NP[structural Case]... [PRO[null Case]... FQ[structural Case]]
I X I

In short, the fact that structuralCase sharing occurs in control clauses follows neatly from a movement account but has no ready explanationin a PRO-basedaccount. Whatof a PRO Theorem-based approach? Sigur6sson(1991) arguedthatthe Icelandicdata that he reviewed requiredabandoningthe PRO Theorem approachto control. This seems right if it is assumedthatCase is assignedundergovernmentand thatPRO appearsonly in ungoverned positions-for then the fact that PRO can bear structural Case requiresthat it be both governed and ungoverned,not a happypairof requirements. However, given the discussion above concerning Case on a movementapproach, another possibilityarises.2'Supposewe couple a PROTheorem approachwith the assumptionthat Case assignmentmay take place not under government,but in some other way-say, under0-role assignmentfor inherentCase, or long-distanceagreement for structural Case as we proposedfor the movementaccount.This would allow PRO to be Casemarked(for quirky Case), while still remainingungoverned. This suggestion faces a serious problem,however. It loses the explanationfor the facts that PRO is null and that full lexical NPs cannot occupy positions occupied by PRO (unless we stipulatethatPRO remainsCaseless underlong-distancestructural Case assignment,which would

21 This

suggestion was made by an anonymousreviewer.

604

REMARKS

AND

REPLIES

then targetthe FQ/SP only). PROTheoremaccountsexplainedthe null statusof PROby assuming thatCase was a necessaryconditionfor phoneticvisibility. If Case is assignedundergovernment, as was standardlyassumed at the time, then PRO, being in an ungovernedposition, could not get Case and so would be expected to be phoneticallynull. Similarly,phoneticallyfull NPs need Case to be licensed but cannot get it in an ungovernedposition and so could not appearwhere PRO did. However, if Case is divorcedfrom government,as proposedhere, then this tie between Case and phonetic visibility is broken and we can no longer explain either why PRO is null or why full lexical NPs cannot occupy these positions. That is, while Case-markedPRO is not inherentlyincompatiblewith the PRO Theoremapproach,it is the phoneticallynull characterof PRO and the fact that positions occupied by PRO cannot be occupied by full NPs that fit poorly with a PRO Theorem-based account that dispenses with the assumptionthat Case is assigned/ checked undergovernment. In short, none of the currentapproachesto control account for the Icelandic Case facts as they stand. Thus, the control facts in Icelandic are, at the very least, no less problematicfor nonmovementtheories of control than they are for movement-basedapproaches. 5 Conclusion argumentagainsta moveIn this article, we have reviewed a pervasive, influential,underground ment approachto control based on Icelandic concord facts. We have shown that this inchoate argumentdoes not underminethe movement approachwhen the facts are considered in their entirety.The facts divide into two basic groups:instancesof quirkyCase assignmentandinstances of structural Case sharing.The formerrequiresome theoreticaladjustments with regardto multiply Case-markedNPs in order to be incorporated into a movement approach.We have shown that the needed adjustmentsmay be independentlyrequired.Furthermore, we have argued that no extant theory of control can accommodatethe Icelandic facts without (similar) adjustments. References
Andrews, Avery. 1982. The representation of Case in Modem Icelandic. In The mental representationof grammaticalrelations, ed. by Joan Bresnan,427-503. Cambridge,Mass.: MIT Press. Boeckx, Cedric. 2000. Quirkyagreement.Studia Linguistica53:451-480. Boeckx, Cedric. 2003a. Case mattersand minimalistconcerns. In Harvardworkingpapers in linguistics 8, ed. by Cedric Boeckx, Claire Bowem, and Jay H. Jasanoff, 159-197. Cambridge,Mass.: Harvard University, Departmentof Linguistics. Boeckx, Cedric. 2003b. Intricaciesof Icelandic agreement.Ms., Universityof Maryland,College Park,and HarvardUniversity, Cambridge,Mass. Boeckx, Cedric.2003c. Symmetriesand asymmetriesin multiplefeaturechecking. In Multiplewh-fronting, ed. by Cedric Boeckx and KleanthesK. Grohmann,17-26. Amsterdam:John Benjamins. Boeckx, Cedric.2004. Long-distanceagreementin Hindi: Some theoreticalimplications.StudiaLinguistica 58:23-36. Boeckx, Cedric, and NorbertHomstein. 2003. Reply to "Controlis not movement." LinguisticInquiry34: 269-280. Boeckx, Cedric, and NorbertHomstein. 2004. Movement undercontrol. LinguisticInquiry35:431-452.

REMARKS

AND

REPLIES

605

Cecchetto,Carlo,and RenatoOniga. 2004. A challenge to null Case theory.LinguisticInquiry35:141-149. Chomsky, Noam. 1981. Lectures on governmentand binding. Dordrecht:Foris. Chomsky, Noam. 1986. Knowledgeof language. New York: Praeger. Chomsky, Noam. 2004. Beyond explanatoryadequacy.In The cartographyof syntactic structures.Vol. 3, Structuresand beyond, ed. by AdrianaBelletti, 104-131. Oxford:Oxford University Press. Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper& Row. In Syntax:An internaChomsky,Noam, and HowardLasnik. 1993. The theoryof principlesandparameters. tional handbookof contemporaryresearch, ed. by JoachimJacobs, Arnim von Stechow, Wolfgang Sternefeld,and Theo Vennemann,506-569. Berlin: Mouton de Gruyter. Culicover, Peter, and Ray Jackendoff.2001. Controlis not movement. LinguisticInquiry32:493-512. Frampton,John, and Sam Gutmann.2006. How sentences grow in the mind. In Agreementsystems, ed. by Cedric Boeckx, 121-157. Amsterdam:John Benjamins. Freidin,Robert,andRex Sprouse. 1991. Lexical Case phenomena.In Principlesandparametersin comparative syntax, ed. by RobertFreidin, 392-416. Cambridge,Mass.: MIT Press. Harley, Heidi. 1995. Subjects,events, and licensing. Doctoral dissertation,MIT, Cambridge,Mass. Hiraiwa,Ken. 2001. Multiple Agree. Ms., MIT, Cambridge,Mass. Hiraiwa, Ken. 2004. Dimensions of agreement.Ms., MIT, Cambridge,Mass. Hong, Soo-min. 2002. Case and focus in Korean. General examinationpaper, University of Maryland, College Park. Homstein, Norbert.1990. Verbraisingin Icelandicinfinitives.In Proceedingsof NELS20, ed. by Juli Carter, Rose-MarieDechaine, Bill Philip, and Tim Sherer,215-229. Amherst:Universityof Massachusetts, GLSA. Homstein, Norbert. 1999. Movement and control. LinguisticInquiry30:69-96. Hornstein,Norbert.2001. Move! A minimalisttheory of construal. Oxford:Blackwell. Hornstein,Norbert.2003. On control.In Minimalistsyntax, ed. by RandallHendrick,6-81. Oxford:Blackwell. Jo, Jung-Min.2002. Chain reactionin the computation.Ms., University of Illinois, Urbana-Champaign. Kayne, RichardS. 2002. Pronounsand their antecedents.In Derivation and explanationin the Minimalist Program, ed. by Samuel David Epstein and T. Daniel Seely, 133-166. Oxford:Blackwell. Landau,Idan. 2003. Movement out of control.LinguisticInquiry34:471-498. Alec. 1991. Case andlicensing.In ESCOL'91: Proceedingsof theEighthEasternStates Conference Marantz, on Linguistics,ed. by GermanF. Westphal,BenjaminAo, and Hee-RahkChae, 234-253. Columbus: Ohio State University, Departmentof Linguistics. Martin,Roger. 1996. A minimalisttheoryof PRO and control.Doctoraldissertation, Universityof Connecticut, Storrs. Nunes, Jairo. 1999. Linearizationof chains and phoneticrealizationof chain links. In Working minimalism, ed. by Samuel David Epstein and NorbertHomstein, 217-249. Cambridge,Mass.: MIT Press. Nunes, Jairo. 2004. Linearizationof chains and sideward movement.Cambridge,Mass.: MIT Press. Schtitze,Carson.2001a. On Korean "Case stacking": The varied functions of the particleska and lul. The LinguisticReview 18:193-232. Schuitze,Carson. 2001b. On the natureof default Case. Syntax4:205-238. Sigur6sson, Halldor Armann. 1991. Icelandic Case-markedPRO and the licensing of lexical arguments. Natural Language & Linguistic Theory9:327-363. Sigur6sson,Halld6rArmann.1996. Icelandicfinite verbagreement.Working Papers in Scandinavian Syntax 57:1-46. Sigur6sson, Halld6r Armann.2000. The locus of Case and agreement. WorkingPapers in Scandinavian Syntax 65:65-108. Sigur6sson,Halld6rArmann.2003. Agree and agreement:Evidence from Germanic.Ms., LundUniversity.

606

REMARKS

AND

REPLIES

in Icelandic. Doctoral dissertation,HarvardUniversity, Thrainsson,Hbskuldur.1979. On complementation Cambridge,Mass. Uriagereka,Juan. 1998. Rhymeand reason. Cambridge,Mass.: MIT Press. Yoon, H.-S. James. 1996. Ambiguityof governmentandthe ChainCondition.NaturalLanguage& Linguistic Theory 14:105-162. Yoon, H.-S. James. 2004. Non-nominative(major)subjectsand Case-stackingin Korean.In Non-nominative and KarumuriVenkata Subbarao,265-314. Amsterdam: subjects, vol. 2, ed. by Peri Bhaskararao John Benjamins. Yoon, Jeong-Me. 1991. The syntax of A-chains: A typological study of ECM and scrambling.Doctoral dissertation,Cornell University, Ithaca,N.Y. (Boeckx) Departmentof Linguistics Harvard University Cambridge,Massachusetts02138 cboeckx@fas.harvard. edu (Hornstein) Departmentof Linguistics Universityof Maryland College Park, Maryland20742 @ umd. nhornste edu