147 AGREEMENT AND MOVEMENT IN ICELANDIC RAISING CONSTRUCTIONS Anders Holmberg School of Linguistics and Language University of Durham Thorbjörg Hróarsdóttir Department of Linguistics University of Tromsö Abstract An intervening dative experiencer argument in an Icelandic raising verb construction blocks agreement between the matrix verb (the matrix T) and the embedded subject of the infinitival clause, as well as blocking raising of the embedded subject. If the experiencer is wh-moved (or relativized or topicalized), it still blocks agreement, but does not block raising. The facts show unequivocally that the whP moves directly from specVP to specCP. The facts are explained in terms of a theory of spell-out and the EPP: A whP is not spelled out before it enters an Agree relation with a C with a matching feature. Spelled-out or not, a whP in specVP blocks Agree between T and the embedded subject. A whP not spelled out does not block Stylistic Fronting, an EPP-driven movement affecting only spelled-out categories. Raising across a whP is shown to be Stylistic Fronting, not standard A-movement. 0. Introduction* The problem we address in the present paper is the following: Seem-type raising verbs take a usually optional experiencer argument, assigned dative in Icelandic. The dative NP is usually moved to subject position, as in (1) but can remain in specVP if it is indefinite, as in (2). In the former case the matrix verb can optionally agree with the nominative subject of the infinitival clause. In the latter case it can’t, but must have default 3SG form. The intervening NP blocks agreement. (1) (2) Mér vir›ast tNP [hestarnir vera seinir] meDAT seemPL the-horsesNOM be slow ‘It seems to me that the horses are slow.’ fia› vir›ist/*vir›ast einhverjum manni [hestarnir vera seinir] there seems/seem some manDAT the-horsesNOM be slow ‘It seems to some man that the horses are slow.’ If the dative NP is wh-moved, it still blocks agreement. (3) * Hva›a manni veist flú a› vir›ist/*vir›ast twh [hestarnir vera seinir] which manDAT know you that seems/seem the horses be slow ‘To which man do you know that the horses seem to be slow’ We wish to thank the following people for their assistance with data and/or their comments on an earlier version of the paper: A›alhei›ur fiorsteinsdóttir, Paolo Aquaviva, Halldór Á. Sigur›sson, fiórhallur Eyflórsson. 148 A way of formulating the generalization is that a wh-trace blocks agreement, but an NP-trace doesn’t. But this presupposes that the wh-phrase moves directly to specCP, without passing through specIP, or the relevant trace wouldn’t be a whtrace. The main challenge posed by these facts is how to ensure this. Furthermore, raising of the embedded subject across the dative NP is impossible in Icelandic. If the dative NP is wh-moved, raising is possible. (4) (5) *Hestarnir vir›ast mér [ tNP vera seinir] the-horsesNOM seem meDAT be slow Hverjum hafa hestarnir virst twh [tNP vera seinir]? whoDAT have the-horsesNOM seemed be slow So agreement with the embedded subject across a wh-trace is impossible, but movement of the embedded subject across the wh-trace is possible. We will show how these facts all fall into place, in a derivational framework, given a particular theory of agreement (mostly following Chomsky 2001), movement and spell-out. 1. Background: Object agreement in Icelandic It is a well know fact that Icelandic exhibits oblique subjects, also called quirky subjects, mainly in construction with experiencer or psychological predicates. This is shown in (6). (6) a. fieim var hjálpa› themDAT was helped b. Hennar var sakna› herGEN was missed Here the NP which precedes the verb in the stylistically unmarked word order has oblique case. Where an NP also follows the verb, it appears in nominative case, as shown in (7). (7) a. Henni líku›u hestarnir herDAT liked the horsesNOM ‘She liked the horses’ b. Henni flykir Ólafur lei›inlegur herDAT thinks OlafNOM boringNOM ‘She finds Olaf boring’ 149 That the preverbal NP in these examples is the subject, not for example a topicalized object, has been demonstrated beyond doubt in a large number of works; see Thráinsson (1979: 462-476), Zaenen & al. 1985, Sigur›sson (1989/1992: 198-209). In such constructions, Icelandic optionally allows agreement between the verb and the following nominative plural object or small clause/infinitival clause subject. This happens whether the matrix oblique subject has a singular or a plural form, as shown in (8) and (9) for singular subjects, and (10) for plural subjects. If object agreement is not chosen, the verb appears with a default 3SG form, as shown in the (a) examples. (8) a. Mér finnst tölvurnar ljótar meDAT find-SG the computersNOM uglyNOM b. Mér finnast tölvurnar ljótar meDAT find-PL the computersNOM uglyNOM ‘I find the computers ugly’ (9) a. Einhverjum stúdent finnst tölvurnar ljótar some studentDAT find-SG the computersNOM uglyNOM b. Einhverjum stúdent finnast tölvurnar ljótar some studentDAT find-PL the computersNOM uglyNOM ‘Some student finds the computers ugly’ (10) a. Einhverjum stúdentum finnst tölvurnar ljótar some studentsDAT find-SG the computersNOM uglyNOM b. Einhverjum stúdentum finnast tölvurnar ljótar some studentsDAT find-PL the computersNOM uglyNOM ‘Some students find the computers ugly’ In this paper we will discuss agreement in raising verb constructions between the matrix T and the subject of the small clause or infinitival clause, as in (8,9,10). We will continue to refer to this as (a special case of) object agreement. In particular, we will discuss a set of cases where such object agreement fails, due to an intervening dative experiencer NP. 2. Where object agreement fails In subject raising constructions, such as (8-10), agreement with the embedded subject of the small clause or raising infinitival is ungrammatical under certain conditions. These conditions include cases with (a) an overt, unmoved 150 experiencer, that is, in transitive expletive constructions, b) where the experiencer is a wh-question word, c) where the experiencer is introduced by a relative clause, and d) where it is topicalized.1 2.1 Transitive expletive constructions Transitive expletive constructions in Icelandic clearly block object agreement. There is a clear difference between a singular and a plural experiencer: In clauses with a plural subject, both object agreement and the default form of the verb are possible, as shown in (13), while only the default form is possible in clauses with a singular subject, as shown in (12) and (14). (11) a. Manninum vir›ist hestarnir vera seinir the manDAT seem-SG the horsesNOM be slowNOM b. Manninum vir›ast hestarnir vera seinir the manDAT seem-PL the horsesNOM be slowNOM ‘The man finds the horses slow’ (12) a. fia› vir›ist einhverjum manni hestarnir vera seinir it-EXPL seem-SG some manDAT the horsesNOM be slowNOM b. *fia› vir›ast einhverjum manni hestarnir vera seinir it-EXPL seem-PL some manDAT the horsesNOM be slowNOM ‘A man finds the horses slow’ (13) a. fia› finnst mörgum stúdentum tölvurnar ljótar it-EXPL find-SG many studentsDAT the computersNOM uglyNOM b. ? fia› finnast mörgum stúdentum tölvurnar ljótar it-EXPL find-PL many studentsDAT the computersNOM uglyNOM ‘Many students find the computers ugly’ (14) a. fia› finnst einhverjum stúdent tölvurnar ljótar it-EXPL find-SG some studentDAT the computersNOM uglyNOM b *fia› finnast einhverjum stúdent tölvurnar ljótar it-EXPL find-PL some studentDAT the computersNOM uglyNOM ‘Some student finds the computers ugly’ 1 Object agreement is restricted to 3rd person. The dichotomy between 1st/2nd person and 3rd person agreement discussed in Sigur›sson 1996 is not directly relevant to our concerns in this paper. 151 2.2 Wh-movement Object agreement also disappears in clauses where the experiencer is introduced by a wh-question, as shown in (15) through (18). As in the case of relative clauses, the lack of agreement becomes clearer in cases where the experiencer is singular, as shown in (16) and (18). It should be noted that for one of our three informants, there is a difference between the short wh-movement in (15) and (16), on the one hand, and the long wh-movement in (17) and (18) on the other hand, with regard to object agreement: While this informant allows both object agreement and the default form in conjunction with short wh-movement, object agreement is clearly out in clauses with long wh-movement and a singular subject. The two other informants dislike object agreement in clauses with whmovement and a singular subject, irrespective of whether the wh-movement is short or long (cf. 16b and 18b). (15) a. Hva›a stúdentum finnst tölvurnar ljótar? which studentsDAT find-SG the computersNOM uglyNOM b. ?Hva›a stúdentum finnast tölvurnar ljótar? which studentsDAT find-PL the computersNOM uglyNOM ‘Which students find the computers ugly’ (16) a. Hva›a stúdent finnst tölvurnar ljótar? which studentDAT find-SG the computersNOM uglyNOM b. ??Hva›a stúdent finnast tölvurnar ljótar? which studentDAT find-PL the computersNOM uglyNOM ‘Which student finds the computers ugly?’ (17) a. Hva›a stúdentum veist flú a› finnst tölvurnar ljótar? which studentsDAT know you that find-SG the computersNOM uglyNOM b. ??Hva›a stúdentum veist flú a› finnast tölvurnar ljótar? which studentsDAT know you that find-PL the computersNOM uglyNOM ‘Which students do you know consider the computers ugly?’ (18) a. Hva›a stúdent veist flú a› finnst tölvurnar ljótar? which studentDAT know you that find-SG the computersNOM uglyNOM b. *Hva›a stúdent veist flú a› finnast tölvurnar ljótar? which studentDAT know you that find-PL the computersNOM uglyNOM ‘Which student do you know considers the computers ugly?’ 152 2.3 Relative clauses As shown in (19) and (20), only the default form of the verb is possible in relative clauses where the experiencer is relativized, independent of whether the experiencer has the singular or the plural form, although the lack of agreement becomes even clearer in clauses where it is singular as in (20). (19) a. fietta eru stúdentarnir sem finnst tölvurnar ljótar these are the studentsNOM that find-SG the computersNOM uglyNOM b. ?? fietta eru stúdentarnir sem finnast tölvurnar ljótar these are the studentsNOM that find-PL the computersNOM uglyNOM ‘These are the students that find the computers ugly’ (20) a. fietta er stúdentinn sem finnst tölvurnar ljótar this is the studentNOM that find-SG the computersNOM uglyNOM b. *fietta er stúdentinn sem finnast tölvurnar ljótar this is the studentNOM that find-PL the computersNOM uglyNOM ‘This is the student that finds the computers ugly’ 2.4 Topicalization Finally, topicalization clearly blocks object agreement. There is a very clear difference here between a singular and a plural experiencer, as in most of the above cases: While both object agreement and the default form are possible in clauses with a plural experiencer, object agreement is ruled out in clauses with a singular experiencer. (21) a. fiessum stúdentum veit ég a› vir›ist tölvurnar ljótar these studentsDAT know I that seem-SG the computersNOM uglyNOM b. fiessum stúdentum veit ég a› vir›ast tölvurnar ljótar this studentsDAT know I that seem-PL the computersNOM uglyNOM ‘I know that these students find the computers ugly’ (22) a. fiessum stúdent veit ég a› vir›ist tölvurnar ljótar this studentDAT know I that seem-SG the computersNOM uglyNOM b. *fiessum stúdent veit ég a› vir›ast tölvurnar ljótar this studentDAT know I that seem-PL the computersNOM uglyNOM ‘I know that this student finds the computers ugly’ 153 3. Subject raising across a wh-trace Summarizing, agreement between the verb or auxiliary in T and the embedded subject of the small clause or infinitival clause in the raising construction is blocked (a) by an intervening dative experiencer NP, and (b) by the trace of such an NP, when it is wh-moved, relativized or topicalized. We repeat the relevant examples: (23) a. *fia› finnast einhverjum stúdent tölvurnar ljótar it-EXPL find-PL some studentDAT the computersNOM uglyNOM b. *Hva›a stúdent veist flú a› finnast t tölvurnar ljótar? which studentDAT know you that find-PL the computersNOM uglyNOM c. *fietta er stúdentinn sem finnast t tölvurnar ljótar this is the studentNOM that find-PL the computersNOM uglyNOM d. *fiessum stúdent veit ég a› finnast t tölvurnar ljótar this studentDAT know I that find-PL the computersNOM uglyNOM Now contrast this with the following observation: In experiencer-less raising constructions the embedded subject raises to matrix subject position.2 (24) a. Ólafur vir›ist [t vera gáfa›ur] Olaf-NOM seem-SG be intelligent ‘Olaf seems to be intelligent’ b. Ólafur hefur virst [t vera gáfa›ur] Olaf-NOM has seemed be intelligent ‘Olaf has seemed to be intelligent’ c. Stúlkan flykir [t vera falleg] the girl-NOM think-SG be beautiful ‘The girl is thought to be beautiful’ d. Stúlkan hefur flótt [t vera falleg] the girl-NOM has thought be beautiful ‘The girl has been thought to be beautiful’ The presence of an overt dative experiencer blocks raising. (25) a. *Ólafur vir›ist mér [t vera gáfa›ur] Olaf-NOM seem-SG me-DAT be intelligent ‘I find Olaf intelligent’ 2 We only use the verbs vir›ast and flykja in these examples, as the verb finnast, used in some of our earlier examples, has an obligatory experiencer. 154 b. *Ólafur hefur virst mér [t vera gáfa›ur] Olaf-NOM has seemed me-DAT be intelligent ‘I have found Olaf intelligent’ c. *Stúlkan flykir mér [t vera falleg] the girl-NOM think-SG me-DAT be beautiful ‘I find the girl beautiful’ d. *Stúlkan hefur flótt mér [t vera falleg] the girl-NOM has thought me-DAT be beautiful ‘I have found the girl beautiful’ This looks like a straightforward Minimal Link Condition effect (Chomsky 1995: ch. 4): The dative experiencer is closer to T than the embedded subject, hence the EPP feature attracts the former, not the latter; see Torrego 1996, Boeckx 2000. Somewhat surprisingly, if the dative experiencer is moved to specCP, as in wh-questions, relatives, or topicalization, raising of the embedded subject is possible (because of V2 we use a compound tense, in order to clearly show that raising has taken place).3 The structure of (26a) is roughly (b). (26) a. b. Hverjum hefur Ólafur virst vera gáfa›ur? who-DAT has Olaf-NOM seem-SG be intelligent ‘Who has found Olaf intelligent?’ hverjumi hefurj [TP Ólafurk tj [VP virst t i [IP tk vera gáfa›ur]]] That is to say, raising and Agree are both blocked by an intervening overt dative NP, but while Agree is blocked also by an intervening wh-trace, raising seems not to be. This is somewhat reminiscent of the situation in Italian, where a full lexical experiencer blocks raising, while a topicalized experiencer appears not to do so, as noted by Rizzi 1996, Torrego 1996, 1998, Boeckx 2000. We return in section 7 to a brief comparison of Icelandic with Italian and Spanish, as regards raising across the experiencer. 4. Agree and intervention We assume the following underlying structure for the raising verb construction with a dative experiencer. We ignore the precise structure or categorial make-up of the embedded small clause (SC) or infinitival clause (InfinP). 3 This observation is due to Halldór Á. Sigur›sson (p.c.). 155 (27) TP vP T v VP vir›ist v NPDAT V’ tV SC NP AP The experiencer NP is assigned dative by the verb, an instance of inherent (lexical) case-marking. The verb invariably moves to a position preceding the experiencer; we assume this is movement to v. We will return to the question whether the relevant verbs are properly characterized as transitive verbs. As a methodological rule of thumb we assume only as much functional structure as we need for the problem at hand. We assume the theory of Agree in Chomsky 2001. According to this theory, finite T comes with a set of unspecified -features and (subject to parametric variation) an EPP-feature. The function of the EPP-feature is to ensure that specTP gets filled in the course of the derivation. In order to be spelled out, and thereby be eliminated from narrow syntax but accessible to phonology, the unspecified -features need to be assigned a value. There are two ways that this can happen, in Icelandic: (a) T probes its c-command domain seeking an NP which is syntactically active, that is, hasn’t already been assigned structural case. If there is such an NP, it can assign values to T’s -features, and have its unspecified Case-feature valued nominative in return. The NP can also then move to specTP, to check, and thereby eliminate, T’s EPP-feature. (b) In the absence of a syntactically active and accessible NP, T’s -features can be assigned the default value 3SG. As noted, default feature assignment is always an option in modern Icelandic. The dative experiencer NP is syntactically active and visible to T, indicated by the fact that it can move to specTP (see 8-10), but it cannot value T’s -features; somehow this is prevented by the dative case. This is not a problem if the experiencer NP moves to specTP; in that case T can probe deeper into the VP, target the embedded subject NP, and have its features valued by this NP. But if the experiencer NP does not move, it will block access to the embedded subject for T, leaving default agreement as the only option. This is what Chomsky 2000 refers to as defective intervention: Even though the intervening NP cannot itself value T, it can prevent a lower NP from valueing T. The structure of, for 156 example, (12a), repeated here as (28a) at the relevant point in the derivation is (28a’): The expletive checks the EPP-feature on T; No Agree is possible between T and the experiencer NP, due to the dative Case, or between T and the embedded subject, due to defective intervention, leaving default agreement as the only option. The structure of (11b), on the other hand, repeated here as (28b) is (28b’): The dative NP is attracted to specTP, checking the EPP-feature. The NP-trace is not visible to T, which therefore can target the embedded subject, and have its -features valued by it. (28) a. a´. fia› vir›ist einhverjum manni hestarnir vera seinir [TP fia› T [vP NP V [InfinP NP VP]]] [,EPP] [DAT, SG] [PL] b. Manninum vir›ast hestarnir vera seinir b’. [TP NP T [vP tNP V [IP NP VP]]] [DAT SG] [,EPP] [PL] In (28b), the embedded subject NP conceivably gets its nominative case from T through Agree. How does it get its nominative case in (28a), though? It has recently been proposed (Jonas, to appear) that the raising verbs are transitive in Icelandic, in the sense that, like believe-type verbs, they assign structural case to the subject of a non-finite complement clause. However, unlike believe-verbs, the raising verbs assign nominative case (in Icelandic; as discussed by Jonas (to appear), in Faroese they assign accusative).4 We adopt this hypothesis here. To be more precise, we assume that the raising verbs in question come in two varieties: One is intransitive, meaning that it does not assign Case to the subject of its complement clause, which therefore relies on Agree with the matrix T for a Case feature value. Since the embedded subject will then value the -features of T, the result is agreement between the matrix verb and the embedded subject. The other variety is transitive, and assigns nominative to the subject of its complement clause, as proposed by Jonas, to appear. This NP is thereby inactivated, and matrix T has to be valued by default.5 4 See Sigur›sson 2001 for a different account of these facts. Halldór Á. Sigur›sson has informed us that he finds some lexical variation among the raising verbs regarding their inclination to agree, such that for example flykja agrees more readily than vir›ast or finnast. We (that is, the native Icelandic speaker among us) do not share his intuitions on this point. Nevertheless, in the present theory such lexical variation can be accounted for in terms of the propensity of a raising verb to assign nominative Case. If the verb prefers to assign nominative to the embedded subject, then agreement between T and the 5 157 Now consider (23), repeated here as (29). (29) a. *fia› finnast einhverjum stúdent tölvurnar ljótar it-EXPL find-PL some studentDAT the computersNOM uglyNOM *Hva›a stúdent veist flú a› finnast t tölvurnar ljótar? which studentDAT know you that find-PL the computersNOM uglyNOM *fietta er stúdentinn sem finnast t tölvurnar ljótar this is the studentNOM that find-PL the computersNOM uglyNOM *fiessum stúdent veit ég a› finnast t tölvurnar ljótar this studentDAT know I that find-PL the computersNOM uglyNOM The structures (29b,c,d) appear to exhibit defective intervention, even though the finite verb and the subject of the small clause are separated only by a trace, just as they are in (28b). However, the trace is an NP-trace in (28b), but a wh-trace in (29b,c,d). Let us first ignore relativization and topicalization, and focus on whmovement in questions, returning to the other cases later. How is the defective intervention accounted for? There are two options: (a) the dative whP hasn’t yet moved when Agree between T and NP applies, or (b) the trace of the moved dative whP blocks Agree just like a lexical NP does. We adopt hypothesis (a). The defective intervention effect in (29) then means that Agree must apply at a point after the dative NP has moved to embedded subject is correspondingly less preferred. An investigation of a wider range of Icelandic speakers’ judgments of raising verb constructions would clearly be in order. The judgments reported in this paper represent the shared judgments of three linguistically sophisticated native speakers. Certain facts noted above suggest an alternative to Jonas’s (to appear) account of the nominative Case of non-agreeing embedded subjects. Consider again the contrast between i. ?fia› finnast mörgum stúdentum tölvurnar ljótar there find-PL many students the-computers ugly ii *fia› finnast einhverjum stúdent tölvurnar ljótar there find-PL some students the computers ugly Number agreement between T and the embedded subject is marginally possible if the intervening dative NP and the embedded subject NP have the same number. Part of the problem in (ii), then, is the number conflict between the two NPs. This implies that T can ‘see’ the -features of the dative NP, although it cannot be valued by them (cf. Boeckx 1999, Sigur›sson 2001). Furthermore, T must see the -features of two NPs simultaneously, in order for Agree between T and the embedded subject to be excluded if the NPs have conflicting features. But if T can ‘see’ the embedded subject NP across the dative NP, it seems not implausible that it could assign nominative Case to the NP, although, as shown, T cannot have its -features valued by the NP, if there is an intervening dative NP with conflicting -feature values. 158 specTP, as in (28b), but while a dative whP is still in specVP. But this means that the whP must move directly from specVP to specCP, without passing through specTP. If it did pass through specTP, Agree would not be able to tell the difference between (28b), where the experiencer is a plain NP, and (29b), where it is a whP.6 This, in turn, means that the whP must have a mark of some kind, which prevents it from being attracted by the EPP the way the NP counterpart is. We propose the following: i. The EPP-feature of T only attracts phonologically visible, that is spelledout, categories (at least in Icelandic); ii. At the relevant point in the derivation the whP is not yet spelled out, while the NP counterpart is. Consequently the NP moves to specTP, attracted by the EPP-feature, but the whP does not. The whP is moved only when C, encoding the relevant attracting feature is merged. If at that point it is too late for Agree to apply between T and the embedded subject, we derive the result we want, namely that raising constructions where the experiencer is wh-moved all have default agreement. That a subject whP moves directly from specVP to specCP, and that it is phonologically empty before moving to specCP, was proposed, for independent reasons, by Holmberg 2000. The motivation was that a subject whP does not block Stylistic Fronting of a VP-internal category, while a subject NP does. Stylistic Fronting is an operation which moves a category to specTP to satisfy the EPP-feature of T in constructions where the subject is not in specTP, either because it is moved to specCP, or because the construction has no subject. In (30a,b) the category in boldface has undergone SF from the position indicated by the trace. (30) a. Hver heldur flú a› stoli› hafi t hjólinu? who think you that stolen has the bike ‘Who do you think has stolen the bike?’ b. fieir sem í Ósló hafa búi› t, segja a› fla› sé fínn bær they who in Oslo have lived, say that it is a nice town ‘Those who have lived in Oslo, say that it is a nice town’ 6 Note that this holds also if we adopt option (b) above; in that case the trace in specVP must be a wh-trace, distinct from an NP-trace, which presupposes that the whP moves directly to specCP. 159 The category moved may be an adverb, a nonfinite verb, a verb particle, a PP or an object NP; in fact almost any category. According to Holmberg 2000 the trigger of SF is the EPP-feature in T, which crucially attracts overt categories only, subject to locality: The overt category attracted is the closest overt category (see Holmberg 2000 for details). This means that SF of any VP-internal category is blocked by an overt VP-external category, such as a negation or adverb. In that case the negation or adverb is the only category that can undergo SF. (31) a. *Hver heldur flú a› stoli› hafi sennilega hjólinu? who think you that stolen has probably the bike b. Hver heldur flú a› sennilega hafi stoli› hjólinu? who think you that probably has stolen the bike c. *fieir sem í Ósló hafa ekki búi›, segja a›… they who in Oslo have not lived, say that … d. fieir sem ekki hafa búi› í Ósló, segja a› … they who not have lived in Oslo, say that … The subject whP in specvP does not, however, block SF, or there could be no SF in either (30a) or (b). The underlying structure of the embedded clause in (30a), for example, at the relevant point in the derivation is roughly (32): (32) hafi+T [vP HVER [v’ stoli› hjólinu]]] This follows if the subject whP in specvP is phonologically empty (indicated by the capitals in (32)). The structure of (30a) following SF, but before merge of C and subsequent wh-movement will then be (33). (33) [TP stoli›j [T’ hafi+T [vP HVER [v’ tj hjólinu]]] Now consider in this light the raising construction in (26a), repeated here as (34a), contrasted with (25b), repeated here as (34b). (34) a. b. Hverjum hefur Ólafur virst vera gáfa›ur? who-DAT has Olaf-NOM seem-SG be intelligent *Ólafur hefur virst mér vera gáfa›ur Olaf-NOM has seemed me-DAT be intelligent 160 Raising of the embedded subject is blocked by the NP but not by the whP. This follows if the raising of the embedded subject in (34a) is the result of SF. If so, the prediction is that it should be blocked by a negation or adverb situated between T and VP. The prediction is borne out: (35) a. *Hverjum hefur Ólafur alltaf virst vera gáfa›ur? who-DAT has Olaf-NOM always seem-SG be intelligent b. Hverjum hefur alltaf virst Ólafur vera gáfa›ur? who has always seem-SG Olaf-NOM be intelligent ‘Who has always found Olaf intelligent?’ (36) a. *Hverjum hefur María ekki flótt vera falleg? who-DAT has Mary-NOM not thought be beautiful b. Hverjum hefur ekki flótt María vera falleg? who-DAT has not thought Mary-NOM be beautiful ‘Who has not found Mary beautiful?’ (37) a. *Hann spur›i hverjum hafi fundist María ekki vera falleg he asked who-DAT has found Mary-NOM not be beautiful b. Hann spur›i hverjum hafi ekki fundist María vera falleg he asked who-DAT has not found Mary-NOM be beautiful ‘He asked who had not found Mary beautiful’ We conclude that there are two ways that the embedded subject can be raised to specTP in Icelandic : a. by ‘standard’ subject raising, a case of A-movement, as in (38a). This movement is not sensitive to the presence of intervening adverbs, but is blocked by an intervening argument, overt or covert, as shown in (38b). (38) a. b. Ólafur hefur (alltaf) virst vera gáfa›ur Olaf-NOM has (always) seemed be intelligent *Ólafur hefur (alltaf) virst mér vera gáfa›ur Olaf-NOM has (always) seemed me-DAT be intelligent b. by SF. This movement is sensitive to the presence of any overt categories, including adverbs. It is not sensitive to the presence of covert categories, though, including a whP in situ.7 7 Interestingly, SF of the embedded subject across the whP makes agreement between the fronted subject and T possible (thanks to Halldór Á. Sigur›sson for pointing this out). 161 5. Spell-out, Agree, and movement The crucial assumptions in the previous section were: a. the experiencer whP in specVP is phonologically empty, therefore invisible to the EPP-feature on T; b. Agree between T and the embedded subject applies before the whP has moved to specCP. If Agree is sensitive only to the -feature content of a category, not to whether it is overt or covert, as seems entirely plausible, an experiencer still in situ, whether overt or covert, blocks Agree between T and the embedded subject. The example of an overt blocking experiencer was (12b), repeated here as (39a). One of the examples of a covert blocking experiencer was (18b), repeated here as (39b): (39) a. b. *fia› vir›ast einhverjum manni hestarnir vera seinir it-EXPL seem-PL some manDAT the horsesNOM be slowNOM *Hva›a stúdent veist flú a› finnast tölvurnar ljótar? which studentDAT know you that find-PL the computersNOM uglyNOM How do we derive assumptions (a) and (b)? We will adopt a version of the theory of derivation proposed by Svenonius 2001. Essentially following Svenonius, we assume that a category is spelled out when it is ‘syntactically complete’, in the sense that any unspecified features are assigned a value, and all (other) uninterpretable features are eliminated as well; this is dubbed ‘Impatient Spell-Out’ by Svenonius. In our terms, this means that for example an NP (more correctly, DP) is spelled out at once when its Case-feature is valued. After that point, the contents of the NP are no longer accessible to syntax. There can, for example, be no movement out of the NP. The spelled-out NP itself can be moved, though, if attracted by an EPP-feature targeting spelled-out categories. 8,9 (i) Hverjum hafa strákarnir virst vera gáfa›ir? who-DAT have-PL the boys seemed be intelligent That is to say, SF feeds agreement: Matrix T cannot agree with the subject in the embedded infinitival clause (or its trace), as they are separated by the not-yet-moved whP, but can agree once the subject has been fronted by SF. 8 This should account for why argument NPs are, in general, islands. On the other hand, there are exceptions to this condition. In particular, indefinite (non-specific) NPs are not always islands (Who did you see pictures of?). This could be because such NPs permit movement of a whP to their edge, where they are accessible even after spell-out (see Chomsky 2001 on the notion edge in connection with phases). Alternatively these NPs are assigned Case in a 162 Elaborating on a suggestion in Chomsky 2000 we assume that the mechanism for feature checking and movement in wh-constructions is essentially Agree: The C of a question (for example) has an uninterpretable feature, let us call it [o] (suggesting ‘operator’) while a whP has an interpretable O-feature. So when C is merged, it probes the sentence for a whP able to check and eliminate C’s uninterpretable o-feature. If, in addition, C has an EPPfeature, the wh-phrase moves to specCP. The whP also has an uninterpretable feature, let us call it [c].10 Elaborating on Chomsky’s 2000 proposal, we assume that the uninterpretable c-feature of whP is assigned a value by an interpretable C-feature in C, where the value is [+/– question], plus for questions, minus for relatives, and perhaps values such as [focus] and [topic]. The point is, the whP cannot be spelled out before its c-feature is checked/assigned a value, that is before a C with the right feature content is merged. So, an experiencer NP is spelled out once it is assigned dative Case by V. Alternatively, assuming with Sigur›sson 2001 and Boeckx 1999 that there is an agreement relation between T and the dative experiencer (in addition to a possible agreement relation between T and the embedded subject) even though it is not reflected morphologically, the NP is spelled out when T is merged and Agree has applied. In either case, the dative NP will thereby be visible to the EPP feature of T. A dative whP is not, however, spelled out at this point, since it still has its uninterpretable c-feature, assigned a value only when C is merged. We still need to ensure that Agree between T and the embedded subject applies before whP has had a chance to get spelled out. Chomsky 2001 proposes that a phase P is spelled out when the highest head of the next phase, P+1, is merged. Given that the sentential phases are vP and CP, vP is spelled out when C is merged, and (embedded) CP is spelled out when the matrix v is merged. Adopting Chomsky’s (2001) spell-out mechanism in addition to our version of Svenonius’s Impatient Spell-Out would have the desired result, forcing Agree between T and the embedded subject to apply before C is merged. However, it would also introduce some redundancy. Note that spell-out of different manner than their specific counterparts (see Mahajan 1992), in such a way as to allow for extraction before the NP is spelled out. 9 Svenonius (2001) does not, in fact, propose that any category is spelled out as soon as it is complete, but instead that certain sentential projections, corresponding to Chomsky’s 2000, 2001 strong phases, are spelled out at once when they are complete in the sense that all their uninterpretable features are eliminated. 10 So this proposal differs crucially from the theory in Chomsky 1995: ch. 4, where only C but not the whP has an uninterpretable feature. In that theory C needs the whP but not vice versa. In the theory sketched here, following Chomsky 2000 (and in part Lasnik 1995), C and whP need each other. See Tanaka 2002 for arguments for a theory of wh-movement along these lines. 163 embedded CP when v is merged follows without special stipulation in the present framework if embedded CP is assigned Case by v.11 Instead we propose the following: When (finite) C is merged, it ‘assigns Case’ to T, the morphological reflex of which is that T’s -features, unless they are already assigned values by Agree, are assigned default values. After this, there can obviously be no Agree between T and the embedded subject. If the embedded subject is not assigned Case by other means, the derivation crashes on account of containing an unspecified (hence uninterpretable) Case-feature. That is to say, we can let Agree apply when it can. In raising constructions with an experiencer, it can apply once the experiencer has been spelled out and moved to specTP. It must apply before finite C is merged, since finite C assigns default values to T’s -features. 6. Remaining issues: Relativization and topicalization With regard to relativization and topicalizations there are two options to consider: A. They are derived by wh-movement, more or less as proposed in Chomsky 1977. In other words, in the case of relatives, the head NP is merged with the relative clause, in which there is a whP which undergoes movement to specCP. The result is then interpreted, perhaps by a rule of predication (Williams 1980), such that the relative head is effectively reconstructed into the position of the trace. The moved whP is frequently deleted (or not spelled out, or spelled out as null). In the case of topicalization, the topicalized NP is merged with CP, containing a whP undergoing movement to specCP. Again a rule of interpretation will effectively reconstruct the topic NP into the position of the wh-trace. In that case, no modification need be made to the theory sketched above, to accommodate relativization and topicalization. B. Alternatively, relativization and topicalization are derived by movement of the relativized or topicalized category itself. This movement would then be subject to the same conditions as wh-movement. In particular, relativization or topicalization of the experiencer in raising verb constructions must be movement directly from vP to specCP, and the moved category must not be spelled out before C is merged. That is to say, we have to assume that, just like the whP, the NP to be relativized or topicalized has an uninterpretable 11 Movement out of a complement CP is still possible given that the ‘edge’ of a spelled out category, in this case CP, can be accessible even after the category is spelled out, as proposed by Chomsky 2000, 2001. If the edge of CP is specCP, then specCP functions as an escape hatch for movement out of CP. 164 feature assigned a value only by C. This presupposes that every category that can be relativized or topicalized may come supplied with a c-feature and an o-feature. As the facts discussed in this paper do not provide any grounds for choosing between these two hypotheses, we remain agnostic. Raising across the experiencer in Spanish and Italian As discussed in Torrego 1996, 1998, Spanish does not tolerate raising across the experiencer at all (examples adapted from Torrego 1996). (40) a. b. c. d. e. Este taxista parece [t estar cansado]. this taxi driver seems be tired *Este taxista me parece [t estar cansado]. this taxi driver me seems be tired *Este taxista parece a Maria [t estar cansado]. this taxi driver seems to Maria be tired Les parecio (a mis amigos) [que Maria estaba cansada]. them seemed (to my friends) that Maria was tired ‘It seemed to them/my friends that Maria was tired’ *Les parecio (a mis amigos) [Maria estar cansada]. them seemed (to my friends) Maria be tired Whether the experiencer is a clitic pronoun or not, the embedded subject cannot be raised across it. In this respect Spanish is like Icelandic. As discussed by Torrego, here illustrated by (40d), the experiencer can itself satisfy the EPP, or more neutrally, whatever feature it is that triggers raising in (40a), but it does so not by movement but by clitic doubling: Without the clitic, (40d) is ungrammatical. It follows that raising across the experiencer will always be ruled out by the Minimal Link Condition (Chomsky 1995: ch. 4), as Torrego (ibid.) has argued: The experiencer can satisfy the EPP in T (by virtue of clitic doubling), and will always be closer to T than the embedded subject, so it will always block raising of the subject. Given that clitic doubling of the experiencer is obligatory (however that is formally construed), it does not help if the experiencer is wh-moved: The clitic-with-doubled-NP will always satisfy the EPP (or whatever feature triggers raising in Spanish). Note that, unlike Icelandic, Spanish seem-verbs with an experiencer can only be constructed with a finite embedded clause: compare (40d,e) and (1). We have a straightforward partial explanation of this: The experiencer always blocks Agree between T and the embedded subject in Spanish. Icelandic has the option 165 of moving the experiencer to specTP, thus making the embedded subject of the infinitival clause accessible to Agree by T. Spanish does not make use of this option. Instead of movement, the EPP can be satisfied by clitic doubling, but this leaves the doubled experiencer in situ, blocking Agree. We assume that the clitic is always doubled by an NP in situ, where the NP can be a visible [a NP] or pro. We have a complete explanation of the difference between Spanish and Icelandic as regards (40d), if it is the case that the subject of the infinitival in Spanish cannot get Case except by Agree with the matrix T. Recall that Icelandic has the means to assign nominative to the embedded subject, even without Agree. Above we adopted Jonas’s (to appear) proposal that the nominative case can be assigned by seem-type verbs, in Icelandic. Furthermore, once assigned nominative by T, the embedded subject is necessarily attracted to specTP, which accounts for why (41) is out as well. (41) *Parece [Maria estar cansada]. As mentioned earlier, the situation in Italian is, on the face of it, more similar to Icelandic. (42) a. ?*Gianni sembra a Piero [t non fare il suo devere]. Gianni seems to Piero not to do his duty b. Gianni non gli sembra fare il suo devere. Gianni not to-him seems to do his duty c. A Piero, Gianni non sembra fare il suo devere. (Rizzi 1986) A full lexical experiencer (or a stressed pronoun or anaphor) blocks raising, but a clitic experiencer does not.12 Furthermore, if the experiencer is clause-initial, as in (42c), raising is possible (and necessary). However, on closer inspection, Italian is not like Icelandic. Consider (43a,b,c): (43) a. b. 12 ?*l'uomo a cui Gianni sembra non fare il suo dovere the man to whom Gianni seems not to do his duty ?*L'unico osservatore a cui la situazione sembra (non) essere migliorata e' ... the only observer to whom the situation seems not to have improved is… This has been claimed to be the case in French, as well. However, Boeckx 2000 shows that once certain phonological factors are taken into account, French in fact allows raising across a lexical experiencer. 166 c. *Landolfi era l'unico scrittore a cui la vita sembrava t wh [ tNP averci tolto il diritto di vivere]. Landolfi is the only author to whom life appeared to have-us takenaway the will to live ‘…to have taken away from us the will to live’. We are grateful to Paolo Aquaviva for providing these sentences and discussing them with us. According to him they are no better than (42a), but have roughly the same very marginal status (possibly apart from (43c), which is worse). Recall that raising across the wh-trace (i.e. raising across a not-yet-moved whP) in Icelandic proved to be Stylistic Fronting. If Italian simply does not have Stylistic Fronting (presumably shown already by the fact that it is a null subject language), we do not expect raising across a wh-trace/whP to be possible in Italian, any more than raising across a lexical NP/PP experiencer. The problem remains how to account for the contrast between (42a,c). We suggest that the experiencer in (42c) is merged (‘first-merged’, i.e. not moved) directly with TP, receiving Case as well as theta-role from the preposition (alternatively, it is merged in a designated topic position). As for the contrast between Spanish and Italian, specifically (40d) and (42b), we concur with Torrego that it has to do with clitic-doubling. Italian does not have clitic-doubling. We take this to mean that the clitic in (42b) moves by A-movement from the experiencer base-position in specVP.13 In Spanish, on the other hand, the presence of a clitic indicates the presence of a doubling NP, either lexical or pro. This means that T can agree with the embedded subject in Italian. Like Spanish, Italian cannot license a subject in situ in a raising infinitival (with or without an experiencer, with or without agreement). (44) *(Gli) sembra [Gianni essere stanco] him seems Gianni be tired For some reason which remains obscure, Agree between T and the embedded subject forces movement of the latter to specTP.14 It follows that (42b) is well formed while (44) is ill formed. 13 More precisely, we assume with Cardinaletti and Starke 1999 that the clitic first moves by A-movement to a ‘weak pronoun position’, whence it undergoes head-movement, adjoining to T. 14 The triggering feature can’t be the Icelandic type of EPP, since (i) is well formed: (i) Sembra [che Gianni non faccia il suo dovere]. (it) seems that Gianni does not do his duty 167 References Boeckx, Cedric: 1999, ‘Conflicting C-command Requirements’, Studia Linguistica 53, 227-250. Boeckx, Cedric: 2000, ‘Raising and Experiencers, Cross-linguistically’, ms. University of Connecticut. Cardinaletti, Anna & Michal Starke: 1999, ‘The typology of structural deficiency: A case study of the three cl;asses of pronouns. In H. van Riemsdijk (ed.) Clitics in the languages of Europe, Mouton de Gruyter, Berlin & New York. Chomsky, Noam: 1977, ‘On Wh-movement’. In P. Culicover and T. Wasow (eds) Formal Syntax, MIT Press, Cambridge MA., pp. 71-132. Chomsky, Noam: 1995, The Minimalist Program, MIT Press, Cambridge MA. Chomsky, Noam: 2000, ‘Minimalist Inquiries: the Framework’. In R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik , MIT Press, Cambridge, Mass. Chomsky, Noam: 2001, ‘Derivation by Phase’. In M. Kenstowicz (ed.), Ken Hale: A Life in Language, MIT Press, Cambridge Mass. Holmberg, Anders: 2000, ‘Scandinavian Stylistic Fronting’: How Any Category Can Become an Expletive’, Linguistic Inquiry 31, 445-484. Jonas, Dianne. To appear. ‘Icelandic raising constructions.’ Ms. Department of Linguistics, Yale University. Lasnik, Howard: 1995, ‘Case and Expletives: On greed and other human failings’. Linguistic Inquiry 26, 615-634. Mahajan, Anoop. 1992. ‘The specificity condition and the CED’, Linguistic Inquiry 23, 510-516. Sigur›sson, Halldór Á: 1989,Verbal Syntax and Case in Icelandic. Ph.D. dissertation, published by Department of Scandinavian Languages, University of Lund. Reprinted by the University of Iceland, 1992. Sigur›sson, Halldór Á: 1996, ‘Icelandic Finite Verb Agreement’, Working Papers in Scandinavian Syntax 57, 1-46., Department of Scandinavian Languages, University of Lund, pp. 103-151. Sigur›sson, Halldór Á: 2001, ‘Case: Abstract vs. Morphological’. Working Papers in Scandinavian Syntax 67, Department of Scandinavian Languages, University of Lund, pp. 103-151. Svenonius, Peter: 2001, ‘Impersonal Passives: a Phase-based Analysis’. In A. Holmer, J-O. Svantesson and Å. Viberg (eds.) Proceedings of the 18th Italian is a null subject language, so the EPP-feature in T has a different status, if there is any at all. 168 Scandinavian Conference of Linguistics, Travaux de l’Institut de Linguistique de Lund. Tanaka, Hidekazu: 2002, ‘Large-scale pied-piping and feature-interpretability’. Ms. Institute of Linguistics, English as a Second Language, and Slavic Languages and Literatures, University of Minnesota. Torrego, Esther: 1996, ‘Experiencers and raising verbs’, in R. Freidin (ed.) Current Issues in Comparative Grammar. Dordrecht: Kluwer. Torrego, Esther: 1998, The dependencies of objects, MIT Press, Cambridge Massachusetts. Thráinsson, Höskuldur: 1979, On Complementation in Icelandic, Garland Publishing, New York. Williams, Edwin: 1980, ‘Predication’, Linguistic Inquiry 11, 203-238. Zaenen, Annie, Joan Maling & Höskuldur Thráinsson: 1985, ‘Case and Grammatical Functions: The Icelandic Passive’, Natural Language and Linguistic Theory 3, 441-483.