RAISING IN ROMANIAN: MOVE AND AGREE. MARIA LUISA RIVERO University of Ottawa. DANA GEBER University of Ottawa

Transcription

1 RAISING IN ROMANIAN: MOVE AND AGREE MARIA LUISA RIVERO University of Ottawa DANA GEBER University of Ottawa 0. Introduction 1 Romanian raising constructions have been the source of debates in the past because they seem to challenge basic tenets of the Government and Binding framework (Grosu and Horvath 1984, Rivero 1989, Dobrovie-Sorin 2001, Roussou 2001, among others). As the minimalist program has developed, such constructions have not attracted as much attention. In this paper, we argue that in Romanian, raising is interesting for recent innovations such as (a) the operation Agree (Chomsky 2001a), (b) similarities and differences between Move and Agree, and (c) locality effects of experiencers on raising debated under the label Intervention. In this section, we introduce our main proposals in informal terms. First, consider the constructions with a raising verb in (1a-b), which differ in the position of nominative noi we. We claim that such sentences prove of interest for the recent operation called Agree, and its comparison with Move. (1) a. Noi părem [ să începem [să lucrăm bine]]. we.nom Seem.1Pl Subj begin.1pl Subj work.1pl well b. Părem [ să începem [să lucrăm bine noi]]. Seem.1Pl Subj begin.1pl Subj work.1pl well we.nom We seem to begin to work well. In (1a-b), all verbs agree in person and number with the nominative, the matrix verb is in the indicative mood, and the two embedded verbs are in the subjunctive mood. Examples of type (1a) have attracted attention under the assumption that nominative noi moves step by step and licenses some form of agreement on each verb. Examples of type (1b) have received less attention. Partially inspired by Chomsky (2001a-b), in this paper we argue that such patterns involve a type of multiple Agree that is established between each of the verbs and the nominative in the embedded clause. We view indicative verbs as complete because they contain both phi-features and tense, and subjunctives as defective because they contain just phi-features. In (1b), noi in the embedded clause has a nominative feature that is uninterpretable, and is valued by the complete verb părem, and each verb has person and number uninterpretable features valued by noi. On this view, (1b) illustrates multiple long-distance Agree. T (Tense) in the matrix is one Probe, Person 1 Research for this paper was partially subsidized by the Social Sciences and Humanities Research Council of Canada through Research Grant to María Luisa Rivero. Grammaticality judgments come from several native speakers of Romanian. We give special thanks to Rodica Diaconescu for help throughout the development of the ideas in this paper.

2 MARIA LUISA RIVERO & DANA GEBER 2 and Number on matrix and embedded Ts are other Probes, and noi is the Goal for all such Probes. Now consider (2a-b). We argue that such constructions are of interest for intervention effects of matrix experiencers on both Move and Agree. (2) a. Copiii îi par Mariei să lucreze bine. Children.Nom.the Cl.Dat.3 seem.3pl Mary.Dat Subj work.3pl well b. Mariei îi par să lucreze bine copiii. Mary.Dat Cl.Dat.3 seem.3pl work.3pl well children.nom.the The children seem to Mary to work well. All the verbs in (2a-b) agree in person and number with nominative copiii, and this nominative is in the matrix clause in (2a), and in the embedded clause in (2b), which provides a second illustration of multiple Agree. Romanian is not mentioned in recent discussions concerned with crosslinguistic variation of matrix experiencer effects that we briefly recall in 1, so (2a-b) are interesting to show that in Romanian, experiencers are not interveners for nominatives from a double perspective. On the one hand, (2a) shows that a dative argument of the matrix does not block Move of the nominative. On the other hand, (2b) shows that a dative argument of the matrix does not block Agree between matrix T and nominative in situ. Thus, we can conclude that in Romanian, matrix experiencers are transparent for the two operations that can affect nominatives. The raising constructions in (3a-b) have escaped notice in the literature, and also prove of interest for experiencer intervention effects on both Move and Agree. (3) a. Noi părem [să îi plăcem lui Ion]. We.Nom seem.1pl Subj Cl.Dat.3 like.1pl John.Dat b. Lui Ion părem [să îi plăcem noi]. John.Dat seem.1pl Subj Cl.Dat.3 like.1pl we.nom John seems to like us. The complement of the raising verb in (3a-b) is a quirky clause with a psych verb equipped with its own dative experiencer argument, and a nominative theme that agrees with the verb. In such constructions, the nominative can raise, (3a). Alternatively, the embedded dative can also raise, (3b). Under either option, both matrix and embedded verbs must agree in person and number with the nominative, so (3b) is another instance of long distance Agree. In our view, such sentences show that in Romanian a dative argument of the embedded verb is not an intervener for Move by the nominative, (3a), or for long distance Agree between matrix T and nominative, (3b). So far then, Romanian raising constructions seem free of intervention effects. However, sentences of type (3) are ungrammatical if they contain a matrix experiencer in addition to the embedded experiencer, as illustrated in (4). (4) *Lui Ion îţi părem [să îi plăcem noi]. John.Dat Cl.Dat.2 seem.1pl Subj Cl.Dat.3 like.1pl we.nom *John seems to you to like us.

3 3 RAISING IN ROMANIAN: MOVE AND AGREE In this paper, we argue that (4) represents a Minimal Link Condition violation caused by the presence of the two dative experiencers. In constructions of type (1b), structural case must be licensed by a complete T, which is the matrix T, and not by a defective T such as a subjunctive. We consider Romanian dative experiencers quirky subjects with inherent and structural case that must be licensed by a complete T and not a defective T. When two dative quirky subjects cooccur as in (4), the matrix experiencer prevents the embedded experiencer from licensing structural case at the matrix clause level. Our view then is that Romanian dative subjects interfere with one another, but dative subjects are transparent for nominatives. The paper is organized as follows. Section 2 first recalls crosslinguistic variation on matrix experiencer effects on raising by nominatives, goes on to introduce general characteristics of raising in Romanian relevant for Move and Agree and their comparison, and concludes by establishing (the absence of) matrix experiencers effects on Move in this language, which makes it similar to Greek and distinguishes it from Spanish. Section 3 develops an analysis of long distance and multiple Agree for constructions of the type in (1b). Section 4 looks at intervention effects on both Move and Agree, and examines the effects not only of matrix experiencers as in (2a-b), but also of embedded experiencers as in (3a-b), and of the two types combined, as in (4), which to our knowledge have not been discussed in the previous literature. We show that Romanian experiencers are transparent for the two operations on nominatives, that is Move and Agree, and that datives experiencers interfere with one another in a Minimal Link Condition fashion, leading to the impression that they block both Move and Agree for nominatives. Section 6 concludes the paper. 1. Raising in Romanian Before we turn to raising constructions in Romanian, let us first recall crosslinguistic variation on matrix experiencers effects on raising nominatives, which is considerable. In English raising constructions, a subject is able to move across an experiencer, as in (5): (5) John seems to Mary [t to be the best candidate]. Holmberg and Hróarsdóttir (2003) point to a different situation in Icelandic. A dative experiencer blocks both agreement between matrix T and the embedded subject of the infinitival clause, and raising of the embedded subject, as in (6a). If the experiencer is wh-moved, agreement is still blocked since the raising verb is in default form, but raising is possible, as in (6b). (6) a. *Herstarnir virdast mer [t vera seinir]. the horses.nom seem.3pl I.Dat be slow The horses seem to me to be slow. b. Hverjum hafa hestarnir virst [t vera seinir]? Who.Dat have the horses.nom seemed be slow? To whom have the horses seemed to be slow?

4 MARIA LUISA RIVERO & DANA GEBER 4 The situation of Italian and French has also attracted attention (see Cuervo (to appear) for an overview of different proposals and references). In these two languages, matrix experiencers allow raising when they are clitics, (7a) and (8a), but disallow raising when they are NPs, (7b) and (8b). (7) a. Gianni non gli sembra [t fare il suo dovere]. John.Nom Neg Cl.Dat.3 seems do.inf the his duty John does not seem to him to do his duty. b. *Gianni sembra a Piero [t non fare il suo dovere] John.Nom seems to Peter Neg do.inf the his duty *John seems to Peter not to do his duty. (8) a. Paul lui semble [ t être malade]. Paul Cl.Dat.3 seems be.inf sick Paul seems to him to be sick. b. *Paul semble à Valerie [t être malade]. Paul seems to Valerie be.inf sick *Paul seems to Valerie to be sick. Languages with clitic doubling also show variation. In Spanish and Greek, dative or genitive experiencers are necessarily encoded in a clitic, which can be doubled by a NP or appear without NP. In Spanish, the clitic blocks raising, with or without the doubling lexical NP (Torrego 1996, 1998), as in (9): (9) *Juan le parece (a María) [t ser el mejor candidato]. John.Nom Cl.Dat.3 seems (Mary.Dat) be.inf the best candidate *John seems to Mary to be the best candidate. A recent idea to account for the Spanish situation is that parecer seem is a raising verb without an experiencer, and a control verb with an experiencer (Ausín and Depiante 2000, Cuervo (to appear)). In Greek, the embedded subject can move across the clitic with or without the doubling lexical NP, (10). Anagnastopoulou (2003) suggests that clitics facilitate extraction, so tis makes it possible for ta pedhia to raise. (10) Ta pedhia dhen tis fenonte (tis Maria) [t na meletun]. The children.nom Neg Cl.Gen.3 seem.3pl the Mary.Gen Subj study.3pl The children do not seem to Mary to study. Romanian raising constructions offer a unique combination of characteristics reminiscent of Greek, or of Spanish, or of both. In Romanian, raising is possible out of subjunctive or să-clauses, (11a), and out of infinitival or a-clauses, (11b). In raising from subjunctives, all verbs agree in person and number with the nominative, as in (12). (11) a. Copiii par [să fie inteligenţi]. Children.Nom.the seem.3pl [Subj be.3pl intelligent] The children seem to be intelligent.

5 5 RAISING IN ROMANIAN: MOVE AND AGREE b. Copiii par [ a fi inteligenţi]. Children.Nom.the seem.3pl Inf be intelligent The children seem to be intelligent. (12) Voi păreţi [să lucraţi bine]. You.Nom seem.2pl [Subj work.2pl well] You seem to work well. The option to raise out of subjunctives makes Romanian similar to Greek, as the comparison of (13a) and (13b) shows. Greek extraction is out of na-clauses, raising verbs and embedded verbs also agree in person and number with the nominative, and there are no infinitive complements. (13) a. Copiii par [să lucreze bine]. Children.Nom.the seem.3pl [Subj work.3pl well] The children seem to work well. b. Ta pedhia dhen fenonte [ na dulevun]. The children.nom Neg seem.3pl [ Subj work.3pl] The children do not seem to work. The option to raise out of infinitives makes Romanian resemble the major Romance languages including Spanish: (14a-b) vs. (15a-b). (14) a. Copiii par [ a lucra bine]. Children.Nom.the seem.3pl [Inf work well] The children seem to work well. b. Voi păreţi [ a lucra bine]. You.Nom.Pl seem.2pl [Inf work well] You seem to work well. (15) a. Los niños parecen [trabajar bien]. The children.nom seem.3pl work.inf well The children seem to work well. b. Vosotros pareceis [trabajar bien]. You.Nom.Pl seem.2pl [work.inf well] You seem to work well. Romanian is a clitic doubling language where (dative) experiencers must also be obligatorily encoded in a clitic, which can either be doubled by an NP or stand alone. Romanian allows raising of nominatives across a clitic with or without the double NP, both out of subjunctive and out of infinitive embedded clauses. The sentences in (16a-b) illustrate extraction out of subjunctives across a clitic, (16a), or across clitic and doubling NP, (16b). Raising out of infinitives is shown in (17a-b). (16) a. Copiii îmi par să fie inteligenţi. Children.Nom.the Cl.Dat.1 seem.3pl Subj be.3pl intelligent The children seem to me to be intelligent.

6 MARIA LUISA RIVERO & DANA GEBER 6 b. Copiii îi par Mariei să lucreze bine. Children.Nom.the Cl.Dat.3 seem.3pl Mary.Dat Subj work.3pl well The children seem to Mary to work well. (17) a. Copiii îmi par a lucra bine. Children.Nom.the Cl.Dat.1 seem.3pl Inf work well The children seem to me to work well. b. Copiii îi par Mariei a lucra bine. Children.Nom.the Cl.Dat.3 seem.3pl Mary.Dat Inf work well The children seem to Mary to work well. Recall that Spanish and Greek differ as to matrix experiencer effects. The comparison of Romanian (16a-b) with Greek (10) and with Spanish (9) shows that Romanian resembles Greek and differs from Spanish. Romanian also differs from Icelandic where a dative NP in situ blocks raising by the nominative, as the comparison of (16b) and (17b) with (6a) shows. In sum, Rumanian matrix experiencers are transparent for Move by nominatives. 2. Long Distance Agree and Romanian raising constructions In this section, we examine in situ nominatives, which have attracted by far less attention in generative grammar, and have escaped notice in Romanian. In Romanian, a nominative NP may remain in situ in raising constructions, but must agree in person and number with the raising verb, as illustrated in (18a-b). (18) a. Par [ să lucreze bine Ion şi Maria]. Seem.3Pl Subj work.3pl well John and Mary.Nom John and Mary seem to work well. b. Par [ să înceapă [să lucreze bine Ion şi Maria]]. Seem.3Pl Subj begin.3pl Subj work.3pl well John and Mary.Nom John and Mary seem to begin to work well. Worth noting again is the fact that in raising with subjunctive complements of type (18), the nominative agrees in person and number with all verbs. In (18b), Ion şi Maria agrees with the raising verbs par and înceapă, and the lexical verb lucreze. A partially similar situation is found in Spanish, (19a-b), but agreement obtains just between matrix verb and nominative, since embedded verbs are infinitives. (19) a. Hoy parecemos [trabajar bien nosotros]. Today seem.1pl work.inf well we.nom Today we seem to work well. b. Hoy parecemos [ empezar [ a trabajar bien nosotros]]. Today seem.1pl begin.inf work.inf well we.nom Today we seem to begin to work well. We propose that Romanian (18a-b) involve multiple and long distance Agree as in (Chomsky 2001a-b).

7 7 RAISING IN ROMANIAN: MOVE AND AGREE We assume that uninterpretable features for the Case/Agree system in (18a-b) are (a) the phi-features of the different Probes, which are the matrix indicative T and each embedded subjunctive T, and (b) the structural case of the nominative Goal Ion şi Maria John and Mary in the embedded clause. Let us first look at how the superordinate indicative matrix T is licensed. This T seeks an element able to satisfy its phi-features, and the first item it finds that can fulfill such a requirement is the nominative NP in the embedded clause. Therefore, long distance Agree between matrix par and embedded nominative Ion şi Maria takes place. The nominative agrees in person and number with all the verbs, as (20) clearly illustrates. Such example shows that person can be first and number can be plural, so uninterpretable phi-features on each T are not defective, and need to be satisfied. (20) Părem [ să începem [să lucrăm bine noi]]. Seem.1Pl Subj begin.1pl Subj work.1pl well we.nom We seem to begin to work well. Our general idea is that agreement patterns such as those in (18a-b) and (20) indicate multiple partial Agree, and are possible because subjunctive să -clauses constitute weak phases with a defective T, not strong phases with a complete T. Subjunctive Ts are defective because they contain Person and Number, but no Tense feature. Thus, they can determine phi-features, but are unable to determine or license Case. Case is licensed by the complete T in the matrix, which contains not only Person and Number, but also Tense. That is, we adopt for Romanian raising constructions the idea for Greek control in (Iatridou 1993), which is that only the matrix verb with a complete T determines nominative. In more recent terms, we adopt the proposal for Agree in Carstens (2001), which is that this operation deletes the Case goal only if the Probe has an intrinsic case feature, contra Chomsky (2001b). Translating these proposals to the system of uninterpretable features in Chomsky (2001b) adopted here, the general idea is that matrix T in (20) is the only complete category in the structure, and as such counts as the only candidate that can value the uninterpretable Case feature on the Goal. This proposal receives motivation in the last section of this paper, when we propose that embedded quirky subjects must also license their structural case through the matrix T of raising constructions. As stated, each defective T contains Person and Number as uninterpretable phifeatures, so partial Agree also operates in each clause in the following way. Agree is established phase by phase, starting with the most embedded subjunctive clause, which is să lucrăm bine noi. This is a weak phase, not a strong phase, where Person and Number are satisfied by Match of the verb with the nominative NP. We follow Chomsky (2001a), and assume that defective elements are unable to delete matched features, which remain visible for further computation into the next phase, which in the case under discussion is the intermediate subjunctive clause să începem. On this view, the features of the nominative NP will still be available when exiting this intermediate weak phase. In the last cycle, which is the strong phase with a complete T, the matrix verb will Match with the nominative via Agree, the nominative will value the phi-features on T, and T will value Case on the Goal at the same time.

8 MARIA LUISA RIVERO & DANA GEBER 8 To summarize, embedded clauses are subjunctives, so they contain a defective T and count as weak phases. In cyclic fashion, each verb enters phase by phase into partial Agree with the Nominative NP, whose phi-features remain visible until the strong phase, which is the matrix clause with a complete (tensed) T. In the matrix cycle, matrix T has its phi-features valued by the nominative, and values in turn the Nominative feature on the Goal, which cannot be valued by the defective Ts in each of the embedded clauses. In the next section we will further examine Agree from the perspective of experiencer effects. 3. Experiencer Effects in Raising Constructions In this section, we examine intervention effects of matrix experiencers, of embedded experiencers, and of matrix and embedded experiencers combined first on Move and subsequently on Agree. Our view is that in Romanian such experiencers are transparent for all operations involving nominatives. 3.1.Matrix Experiencers and Move We saw in 1 that Romanian resembles Spanish and Greek, with dative experiencers signaled by an obligatory clitic, which can be doubled by a NP, or may appear on its own. The parallelism between Romanian and Spanish is illustrated with raising verbs in (21-22). Romanian experiencers with psychological verbs are in (23). All the c. sentences are ungrammatical because they contain a dative NP without a corresponding clitic. (21) a. Lui Ion îi pare că sîntem obosiţi. John.Dat Dat.Cl.3 seem.3sg IND be.1pl tired It seems to John that we are tired. b. Îi pare că sîntem obosiţi. Dat.Cl.3 seem.3sg IND be.1pl tired It seems to him that we are tired. b. * Lui Ion pare că sîntem obosiţi. *It seems to John that we are tired. (22) a. A Juan le parece que estamos cansados. John.Dat Dat.Cl.3 seem.3sg that be.1pl tired It seems to John that we are tired. b. Le parece que estamos cansados. Dat.Cl.3 seem.3sg that be.1pl tired It seems to him that we are tired. c. *A Juan parece que estamos cansados. *It seems to John that we are tired. (23) a. Lui Ion îi plăcem noi. John.Dat Dat.Cl.3 like.1pl we.nom John likes us. b. Îi plăcem noi. Cl.Dat.3 like.1pl we.nom He likes us.

9 9 RAISING IN ROMANIAN: MOVE AND AGREE c. *Lui Ion plăcem noi. John.Dat like.1pl we.nom *John likes us. Recall that Rumanian allows raising out of subjunctives and infinitives across a matrix experiencer, as illustrated in (16-17) now partially repeated in (24). (24) a. Copiii îi par Mariei [ să lucreze bine]. The children seem to Mary to work well. b. Copiii îi par Mariei [a lucra din greu]. The children seem to Mary to work hard. Thus, a matrix experiencer does not block Move by a nominative in Romanian. In this way, Romanian is parallel to Greek, as (24a-b) compared with (10) partially repeated in (25a) shows, and differs from Spanish, as (24a-b) compared with (9) partially repeated in (25b) shows. (25) a. Ta pedhia dhen tis fenonte tis Maria na meletun. The children do not seem to Mary to study. b. *Juan le parece a María ser el mejor candidato. *John seems to Mary to be the best candidate. From the above, it can be concluded that Romanian a părea behaves like a raising verb when combined with an experiencer, unlike what is proposed for Spanish parecer. In the next section we see further motivation for this conclusion when we discuss Agree. Romanian is similar to Greek, so it seems that clitic experiencers in this language facilitate extraction. We do not develop this second idea formally due to space limitations, but it has been implemented in the recent literature in at least two ways. On the one hand, the general proposal of Anagnostopoulou (2003:5) is that in clitic constructions, the intervening features of the dative clitic move out of the way, and the dative argument ceases to count as an intervener for the nominative. On the other hand, MacGinnis (2004 and her earlier work) proposes that in some Romance languages, clitics head the type of Applicative Phrase that provides an escape hatch for NP movement. These two options concern Move by nominatives. However, in the next section we show as well that in Romanian, dative experiencers do not block long distance Agree between matrix T and embedded nominatives. Thus, another option that should be explored in future work could be some version of Relativized Minimality where the feature content of dative experiencers is transparent for nominatives for all types of relations, namely those based on Move, and also those based on Agree. 3.2.Matrix Experiencers and Agree. The effect of matrix experiencers on Agree has attracted less attention in the literature, and has not been discussed in the case of Romanian. To this effect, consider the raising constructions in (26-27), with verbs that agree with the nominative in the embedded clause, and an experiencer clitic that, doubled or not, is part of the argument structure of the matrix.

10 MARIA LUISA RIVERO & DANA GEBER 10 (26) a. Îmi par [să fie inteligenţi copiii]. Cl.Dat seem.3pl Subj be.3pl intelligent children.nom The children seem to me to be intelligent. b. Lui Ion îi par [ să înceapă John.Dat Cl.Dat.3 seem.3pl Subj begin.3pl [să înveţe bine copiii]]. Subj study.3pl well children.nom The children seem to John to begin to study well. (27) a. Îmi par [a fi inteligenţi copiii]. Cl.Dat.1 seem.3pl Inf be intelligent children.nom The children seem to me to be intelligent. b. Mariei îi par [a fi inteligenţi copiii]. Mary.Dat Cl.Dat.3 seem.3pl Inf be intelligent children.nom The children seem to Mary to be intelligent. Such examples show that a matrix clitic experiencer with or without NP (a) allows long distance Agree between matrix T and nominative in situ with subjunctives, (26a-b), or infinitives, (27), and (b) that partial Agree obtains between the various subjunctive Ts and the nominative. By contrast, a Spanish matrix experiencer blocks long distance Agree with a nominative, (28), which is compatible with the idea that parecer with an experiencer is a control verb. (28) *A María le parecen ser inteligentes los niños. Mary.Dat Cl.Dat.3 seem.3pl be.inf intelligent the children *The children seem to Mary to be intelligent. Thus, in contrast with the situation found in Spanish, Romanian matrix experiencers do not block Move by nominatives as we saw in the previous section, and they do not block Agree. This situation suggests that Romanian dative experiencers have a feature composition that makes them generally transparent to nominatives in raising constructions. This idea receives further support from the behavior in raising constructions of experiencers that belong to the embedded clause, not the matrix clause. 3.3.Embedded Experiencers, Move, and Agree. We noted in (23) a construction that has (a) a dative experiencer clitic, (b) a dative NP doubling such a clitic, (c) and a nominative that agrees with the verb: Lui Ion îi plăcem noi. John likes us. Such psych patterns are interesting to study the effect of embedded experiencers on raising, but they seem to have escaped notice from such a perspective in the relevant literature. Consider (29), where the raising verb embeds a psych verb with the characteristics just described.

11 11 RAISING IN ROMANIAN: MOVE AND AGREE (29) a. Noi părem [să îi plăcem lui Ion]. We.Nom seem.1pl Subj Cl.Dat.3 like.1pl John.Dat John seems to like us. b. Noi părem [să îi plăcem] We.Nom seem.1pl Subj Cl.Dat.3 like.1pl {He/she} seems to like us. c. Noi părem [ să începem [ să We.Nom seem.1pl Subj begin.1pl Subj îi plăcem (lui Ion)]]. Cl.Dat.3 like.1pl (John.Dat) John/ he seems to begin to like us. The sentences in (29) show that a doubled or nondoubled clitic argument of the embedded clause allows fronting of the nominative. Thus, embedded experiencers do not block Move of nominatives, and the same is true in Spanish (30). (30) Nosotros parecemos [empezar [a gustarle a Juan]]. We.Nom seem.1pl begin.inf like.inf Cl.Dat.3 John.Dat John seems to begin to like us. Turning to Agree, (31) shows that an experiencer NP argument of the embedded clause may raise to the matrix, and allows long distance Agree with the nominative in situ. Partial Agree between subjunctive Ts and nominative in situ applies as well. (31) a. Lui Ion părem să îi plăcem (noi). John.Dat seem.1pl Subj Cl.Dat.3 like.1pl (we.nom). John seems to like us. b. Lui Ion părem [ să începem [să îi plăcem (noi)]]. John.Dat seem.1pl Subj begin.1pl Subj Cl.Dat.3 like.1pl we.nom John seems to begin to like us. Therefore, Romanian embedded experiencers do not block long distance and multiple Agree for nominatives. Spanish bears a partial resemblance to Romanian. Agree between a complete matrix T and a nominative in situ is possible when an embedded experiencer NP raises, as in (32). (32) A Juan parecemos [empezar [a gustarle nosotros]]. John.Dat seem.1pl begin.inf like.inf Cl.Dat.3 we.nom John seems to begin to like us. 3.4.Matrix and embedded experiencers combined. The combination of matrix and embedded experiencers in raising constructions seems to block both Move and Agree involving nominatives. In this section, we propose that such effects are due to the interaction of the two experiencers. That is, the matrix experiencer is a quirky subject that blocks a necessary relation between the embedded

12 MARIA LUISA RIVERO & DANA GEBER 12 experiencer as quirky subject and the matrix T, which results in a Minimal Link Condition. Let us motivate this idea. First compare (33a) and (33b). violation of the (33) a. Părem [să îi plăcem noi]. Seem.1Pl Subj Cl.Dat.3 like.1pl we.nom {He/she} seems to like us. b. *Îţi părem [să îi plăcem noi]. Cl.Dat.2 seem.1pl Subj Cl.Dat like.1pl we.nom *{He/she} seems to you to like us. Without matrix experiencer in (33a), long distance Agree applies between the raising verb and the nominative in the embedded clause with another dative experiencer. Partial Agree also obtains. Adding a matrix experiencer as in (33b), however, makes the construction ungrammatical. At this point, it looks as if the combination of a matrix experiencer and an embedded experiencer blocks long distance Agree for the nominative. However, matrix experiencers and embedded experiencers on their own do not have a blocking effect on Agree, as illustrated in (26a-b) and (31a-b). Thus, our idea is that in (33b) the matrix dative interferes with the embedded dative, not with the nominative. Now compare (34a) and (34b). (34) a. Noi părem [să îi plăcem]. We.Nom seem.1pl Subj Cl.Dat.3 like.1pl {He/she} seems to like us. b. *Noi îti părem [să îi plăcem]. We.Nom Dat.Cl.2 seem.1pl Subj Cl.Dat.3 like.1pl *{He/she} seems to you to like us. Without matrix experiencer in (34a), the nominative raises from a clause with another experiencer. However, when matrix experiencer and embedded experiencer cooccur as in (34b), raising the nominative gives ungrammatical results, making Romanian (34b) resemble Spanish (35). (35) * Nosotros te parecemos gustarle. We.Nom Dat.Cl.2 seem.1pl like.inf. Cl.Dat.3 *{He/she} seems to you to like us. It thus seems that Move by the nominative is blocked by the combination of two experiencers in Romanian (34a-b). However, we saw above that if either matrix or embedded experiencer occurs in isolation, Move by an embedded nominative is unproblematic, as illustrated in (24b) and (28a-b). If it is now assumed that in (34b) the matrix dative interferes with the embedded one, this proposal can accommodate the lack of interference on the nominative when only one experiencer is present. Another advantage of the idea that Romanian datives interact with each other and not with nominatives is that it can capture why Romanian and Spanish differ in relation to

13 13 RAISING IN ROMANIAN: MOVE AND AGREE constructions with only one experiencer, and they are parallel when speaking of constructions that combine matrix and embedded experiencers. Finally, compare (36a) and (36b). (36) a. Lui Ion părem [să îi plăcem (noi)]. John.Dat seem.1pl Subj Cl.Dat.3 like.1pl (we.nom). John seems to like us. b. *Lui Ion îţi părem [să îi plăcem noi]. John.Dat Cl.Dat.2 seem.1pl Subj Cl.Dat.3 like.1pl we.nom. *John seems to you to like us. Sentence (36a) shows an embedded experiencer NP raising to the matrix, and Agree between matrix T and embedded nominative. However, with two experiencers as in (36b), the sentence is ungrammatical. As illustrated in (37), the Spanish equivalent for (36b) is also deviant. (37) * A Juan te parecemos [gustarle nosotros]. John.Dat Cl.Dat.2 seem.1pl like.inf. Cl.Dat.3 we.nom. *John seems to you to like us. In this case, the Romanian matrix experiencer seems to block raising of the embedded experiencer, and in addition the two experiencers would seem to block long distance Agree between matrix T and nominative. Thus, it seems that Romanian matrix experiencers have a general blocking effect when the raising verb takes a quirky complement, which makes this language resemble Spanish. However, when a raising verb with its own experiencer takes a complement that is not quirky, there are no blocking effects in Romanian, which thus contrasts with Spanish. Our proposal is that, contrary to appearances, Romanian dative experiencers do not interfere with nominatives, while matrix datives act as interveners for embedded datives. On this view (33b), (34b) and (36b) represent Minimal Link Condition violations under the assumptions listed in (38). (38) a. Romanian dative Experiencers are quirky subjects (pace Dumitrescu and Masullo 1996, contra Rivero 1989). b. Romanian dative Experiencers as quirky subjects combine inherent Case and the equivalent of structural Case. c. A matrix verb with a complete T determines all structural case relations in Romanian raising constructions, including those for quirky subjects. d. Defective Ts (i.e. Subjunctive Ts) do not determine structural case relations for Romanian nominatives, as we argued above. We

14 MARIA LUISA RIVERO & DANA GEBER 14 propose that defectives Ts do not determine structural case for quirky subjects either. e. On the above views, a matrix Experiencer is an intervener that prevents an embedded quirky Experiencer from licensing its structural Case with a complete T, leading to a Minimal Link Condition violation in (33b), (34b), and (36b). 6. Conclusions In Romanian raising constructions, both matrix and embedded dative experiencers allow Move and multiple long-distance Agree involving embedded nominatives. From this perspective, Romanian can be called a transparent language. However, when two dative experiencers cooccur in a raising-type construction, it seems as if both Move and Agree for nominatives are blocked. We have proposed that this is because dative experiencers have no blocking effect on nominatives, but interfere with each other. Dative experiencers are quirky subjects with structural case. In Romanian raising constructions, all structural case relations must be licensed by a complete T, that is, a matrix T with tense, and not by a defective T, a subjunctive T with phi-features and no tense. On such a view, an embedded quirky subject that appears in the complement of a raising verb must license its structural case via the matrix T, not an embedded T. When in a raising construction an experiencer belonging to the raising verb combines with an experiencer belonging to the complement verb, the matrix experiencer counts as an intervener that prevents the embedded experiencer from licensing case with the matrix T. Such a situation represents a violation of the Minimal Link Condition. Our proposal accounts for a complex web of Romanian raising constructions that had escaped notice in the literature. It can also capture in a principled way some complex differences and similarities between Romanian raising constructions and Spanish raising constructions that had also gone unnoticed in the past. REFERENCES Anagnostopoulou, Elena The syntax of ditransitives. Berlin: Mouton De Gruyter. Ausín, Adolfo & Marcela Depiante On the Syntax of parecer with and without Experiencer. In H. Campos (ed.), Hispanic Linguistics at the Turn of the Millennium, Somerville, Mass.: Cascadilla. Carstens, Vicki Multiple Agreement and Case deletion: against phi (in)completeness. Syntax Chomsky, Noam. 2001a. Derivation by Phase. In M. Kenstowicz (ed.), Ken Hale: A Life in Language. Cambridge: MIT Press Chomsky, Noam. 2001b. Beyond Explanatory Adequacy. MIT Occasional Papers in Linguistics Cuervo, María Cristina. To appear. A Control-Vs.-Raising Theory of Dative Experiencers. In A. T. Perez-Leroux and Y. Roberge (eds.), Romance Linguistics, Theory and Acquisition. Amsterdam: Benjamins.

15 15 RAISING IN ROMANIAN: MOVE AND AGREE Dobrovie-Sorin, Carmen Head-to-Head Merge in Balkan Subjunctives and Locality. In M. L. Rivero and A. Ralli (eds.), Comparative Syntax of Balkan Languages. N. Y.: Oxford University Press Dumitrescu, Domnita and Pascual Masullo Romanian and the Non-Nominative Subject Parameter. In C. Parodi, et al. (eds.), Aspects of Romance Linguistics. Washington : Georgetown University Press Grosu, Alex and Julia Horvath. The GB Theory and Raising in Rumanian. Linguistic Inquiry Holmberg A. & T. Hróarsdóttir Agreement and Movement in Icelandic Raising Constructions. Lingua Iatridou, Sabine On Nominative Case Assignment and a Few Related Things. MIT Working Papers in Linguistics MacGinnis, Martha Lethal Ambiguity. Linguistic Inquiry 35, Rivero, María Luisa Barriers and Rumanian. In C. Kirschner and J. DeCesaris (eds.), Studies in Romance Linguistics. Amsterdam: Benjamins Roussou, Anna Control and Raising in and out of Subjunctive Complements. In M. L. Rivero and A. Ralli (eds.), Comparative Syntax of Balkan Languages. N. Y. Oxford University Press Torrego, Esther Experiencers and Raising Verbs. In R. Freidin (ed.), Current Issues in Comparative Grammar. Dordrecht: Kluwer Torrego, Esther The dependencies of objects. Cambridge, MA: MIT Press. María Luisa Rivero Department of Linguistics University of Ottawa Ottawa ON, K1N 6N5 Canada. mrivero@aix1.uottawa.ca Dana Geber Department of Linguistics University of Ottawa Ottawa ON, K1N 6N5 Canada. dana.geber@sympatico.ca