You are on page 1of 39

The Subject-in-Situ Generalization and the Role of Case in Driving Computations

Artemis Alexiadou Elena Anagnostopoulou


The article establishes a novel generalization concerning the placement of arguments by Spell-Out. It centers on the principles that force arguments to leave the VP across languages. The empirical domain consists of constructions where subject movement is not required for reasons that have to do with the Extended Projection Principle. In these environments and whenever a sentence contains both a subject and a direct object, one of the arguments must vacate the VP. We argue that argument externalization is related to Case. It is forced because movement of both arguments to a single head T0 that contains two active Case features in the covert component is banned. Keywords: Case, Extended Projection Principle, computational system, features, Spell-Out, subject inversion

In this article we investigate the syntactic conditions on the availability of VP-internal subjects and objects across languages in constructions in which the Extended Projection Principle (EPP) requirement of T is satisfied by a third element, for example, an expletive. We argue that the generalization in (1) holds. (1) By Spell-Out VP can contain no more than one argument with an unchecked Case feature. (1) expresses the fact that there is a ban against leaving, in the overt component, more than one DP argument in the position in which they are merged. If the sentence contains a subject and a

Preliminary versions of this article were presented at the Workshop on Minimalism at the University of Ottawa in March 1997, at the 7th Colloquium on Generative Grammar in Oviedo in April 1997, at the 3rd Languages and Grammar Colloquium in Paris in June 1997, at the 13th Comparative Germanic Syntax Workshop in Ithaca in July 1997, at the Ling-Lunch at MIT in October 1997, at the University of Tubingen in November 1997, at the 21st GLOW Colloquium in Tilburg in April 1998, and at the Workshop on Inversion in Romance in Amsterdam in May 1998. We wish to thank the audiences for their comments. We thank Jonathan Bobaljik, Marie Claude Boivin, Hans Broekhuis, Noam Chomsky, David Embick, Danny Fox, Sabine Iatridou, Giuseppe Longobardi, Paola Monachesi, David Pesetsky, Christer Platzack, Cecilia Poletto, Josep Quer, Norvin Richards, Ian Roberts, and Peter Svenonius for comments and discussions. Special thanks go to Martin Hackl, Eric Haeberli, Winfried Lechner, Gereon Muller, and in particular Alec Marantz for extensive discussions and comments on earlier versions of the article. We are indebted to two anonymous reviewers for their insightful suggestions and comments.
Linguistic Inquiry, Volume 32, Number 2, Spring 2001 193231 2001 by the Massachusetts Institute of Technology

193

194

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

direct object, one of them must vacate the VP. (1) does not discriminate between the subject and the object; it does not matter which argument moves, as long as one of them does. This is a surprising fact because usually argument externalization is related to the EPP and thus affects only subjects. Here we have a case not reducible to the EPP. The movement requirement in (1) applies equally to subjects and objects, when these have an active Case feature. Therefore, we argue that (1) expresses a requirement for argument externalization related to Case. To account for (1), we propose that there is a constraint against multiple covert Case checking, which is due to the fact that it is not possible for both a subject and an object to move to a single head (T0 ) containing two active Case features. The theoretical implication of this proposal is that there are two formal features manipulated by the computational system, one triggering EPP-related phenomena and one triggering externalization phenomena related to Case. The article is organized as follows. In section 1 we examine two A-movement constructions in which the subject and the object remain VP-internal in languages that generally show an intransitivity restriction on subject inversion. The cases in point are stylistic inversion in French and quotative inversion in English. These constructions lead us to formulate (1). In section 2 we investigate transitive constructions in languages without an intransitivity restriction on subject inversionspecifically, transitive expletive constructions in Icelandic, VSO constructions in Celtic and Arabic, and inverted (VOS) constructions in Romance (Italian/Catalan). At first sight these constructions seem to be counterexamples to (1). However, we argue that (1) holds for these languages as well, once adverb and participle placement facts are taken into account. In fact, there are certain restrictions on inversion in these languages that cannot be accounted for unless (1) is correct. In section 3 we discuss two more cases that arguably fall under (1): locative inversion in English and ga/no conversion in Japanese. In section 4 we discuss the criteria one can use to diagnose VP-internal argumenthood, focusing on the distribution of adverbs in quotative inversion. In section 5 we argue that (1) is never related to the EPP and that it is related to Case. We propose our analysis, which crucially relies on the assumption that whenever Case-assigning heads with unchecked Case features form a complex head (T0 ), the derivation crashes. This forces DP-externalization for reasons of Case checking before such a complex head is formed. In section 6 we illustrate how our proposal works by looking at the derivations of the constructions discussed in sections 13. In section 7 we look at a number of true counterexamples to (1), and we indicate how we would deal with them in this system. 1 Formulating (1): Stylistic Inversion in French and Quotative Inversion in English 1.1 Transitivity Restriction in French and English In French and English there is a transitivity restriction on subject inversion in constructions containing an expletive in subject position. While expletive constructions are well formed with intransitive verbs ((2a) and (3a)), transitive expletive constructions are ungrammatical ((2b) and (3b)).

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

195

(2) a. Il

est arrive un homme. is arrived a man There has arrived a man. ` b. *Il a lu un eleve le livre. EXPL has read a student-NOM the book-ACC There has read a student the book.
EXPL

Expl-VS

*Expl-VSO

(3) a. There arrived a man. b. *There finished somebody the assignment.

Expl-VS *Expl-VSO

It is generally agreed that in (2a) and (3a) the subject remains in a VP-internal position (see Bobaljik and Jonas 1996 and references therein for English, Deprez 1990 and references therein for French). However, these languages do have constructions where the subject can remain VP-internal with transitive predicates. These constructions involve A-movement of an element other than the subject. In French and English this is the case with stylistic inversion and quotative inversion, respectively. In these constructions the subject can remain VP-internal only when the object either has moved out of the VP or is a PP. In other words, the two DP arguments cannot both remain inside the VP. In the following sections we examine these two constructions in turn. 1.2 Stylistic Inversion in French Stylistic inversion (SI), discussed by Kayne and Pollock (1978), Deprez (1990), Collins and Branigan (1997), and Watanabe (1996), among many others, involves postposing of the subject in wh-questions (4a), relative clauses (4b), and subjunctive sentential complements (4c) (see Deprez 1990:4849). (4) a. Je me demande quand partira Marie. I wonder when will-leave Marie I wonder when Marie will leave. b. les resultats que nous donnent ces experiences the results that us give these experiments the results that these experiments give us c. Il ne faut pas que soit arrete ton ami. it must not be that be arrested your friend Your friend must not be arrested. Kayne and Pollock (1998) have argued that SI in subjunctive sentential complements is not the same phenomenon as the one found with wh-questions and relative clauses. In what follows we restrict our attention to the type of SI found in wh-constructions. SI in French is inapplicable when no wh-movement takes place, as shown in (5).

196

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(5) *Partira ton ami. will-leave your friend Your friend will leave. Deprez (1990) argues extensively that in SI the subject remains VP-internal. Her arguments are based on combien extraction, stage- versus individual-level subjects, and floating quantifiers. In the next paragraph we briefly summarize the floating quantifier argument. In French there are roughly two positions in which floating quantifiers occur, namely, following a tensed verb or following an auxiliary. Sportiche (1988) has argued that floating quantifiers are in fact stranded quantifiers that mark the original DP position of the subject. Now consider (6). (6) *Quont tous fait les enfants? what have all done the children What did the children all do? If postverbal subjects in SI have undergone no movement, then stranded quantifiers are expected to be excluded from SI constructions. The ungrammaticality of (6) is unexpected under an analysis of SI in terms of rightward movement. As noted by Kayne and Pollock (1978:604, 1998), Valois and Dupuis (1992), Collins and Branigan (1997), and Espanol-Echevarra, Pinto, and de Wind (1998), among others, SI is disal lowed when the VP contains a direct object. ` (7) *Je me demande quand acheteront les consommateurs les pommes. I wonder when will-buy the consumers-NOM the apples-ACC On the other hand, when the direct object itself is either wh-extracted or cliticized, SI becomes possible again. This is illustrated in (8). (8) a. Que crois-tu que manquent un grand nombre detudiants? what believe-you that be-absent-from a great number of students What do you think that a large number of students are missing? ` b. Tes cours, a quelle occasion les ont manques un grand nombre your courses at which occasion them-have been-absent-from a great number detudiants? of students On what occasion have a large number of students missed your courses? Another option is that the object be a PP (Collins and Branigan 1997, citing Kayne 1972). ` ` (9) ?Quand ecrira ton frere a sa petite amie? when will-write your brother to his little friend When will your brother write to his girlfriend?

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

197

The French facts clearly show that in order for a subject to remain VP-internal, either the DP object must be moved out of the VP or the object must be a PP. A similar state of affairs is found in English quotative inversion. 1.3 Quotative Inversion in English English sentences with direct speech complements (quotes) to verbs of saying, thinking, and writing allow a kind of inversion called quotative inversion (QI) by Collins and Branigan (1997) and Collins (1997). Example (10) is an instance of QI. (10) I am happy, said Mary. Collins and Branigan point out that QI is similar to SI in many respects. As with SI, the subject remains in VP-internal position, and the evidence again comes from floating quantifiers. Assuming Sportiches (1988) analysis of quantifier float, quantifiers that appear to the right of a DP occupy a position to the left of the DP-trace. Quantifiers in [Spec, VP] should therefore appear to the right of the subject DP only if the subject has raised to a higher position. The data in (11) show that quantifiers cannot appear to the right of the subject in QI (11c), although they can appear to the right of the subject when there is no inversion (11a) and they can appear to the immediate left of the subject in QI (11b). (11) a. We must do this again, the guests all declared to Tony. b. We must do this again, declared all the guests to Tony. c. *We must do this again, declared the guests all to Tony. The crucial example is thus (11c), which shows that the subject is VP-internal. Similarly to SI, when the quote triggers inversion in sentences with transitive verbs containing an indirect object, the result is ungrammatical (12a), unless the indirect object is expressed as a PP (12b). No such problem arises when no inversion takes place (12c). (12) a. *What is the exchange rate? asked Mary John. b. What is the exchange rate? asked Mary of John. c. What is the exchange rate? Mary asked John. As with SI, then, in QI the transitive subject can remain VP-internal when there is no VP-internal object. If there is a VP-internal object, it must be a PP. All the above factsthe transitivity restriction, stylistic inversion, and quotative inversioncan be correctly described in terms of (13). (13) When the subject is VP-internal and there is also a VP-internal object, subject inversion is not possible unless the object is a PP.

198

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Let us call (13) the Subject-in-Situ Generalization (SSG).1 We argue that the SSG is not a property of French and English only. Instead, it holds universally; in other words, there is no language in which both the subject and the object are VPinternal. In the next section we look at apparent counterexamples to the SSG and show that they actually support it, once adverb and participle placement facts are taken into account. 2 Apparent Counterexamples to the SSG and How They Support It 2.1 Transitive Expletive Constructions in Icelandic Transitive expletive constructions (TECs) in languages like Icelandic constitute the first apparent counterexample to the SSG. Standardly, TECs have been analyzed as involving VP-internal subjects and objects (see Vikner 1995 and references therein). If this analysis is correct, then the SSG holds only for languages like French and English that have an intransitivity constraint on inversion. However, it has been argued that this analysis is not correct, and it can be shown that the SSG holds for Icelandic as well. Unlike English, Mainland Scandinavian, and French, Icelandic has TECs (see, e.g., Vikner 1995, Hoekstra and Mulder 1990, Levin and Rappaport 1995, SigurLsson 1989, Bobaljik and Jonas 1996). This difference is illustrated in (14). (14) a. *There finished a student the assignment. b. aL klaruLu margar mys ostinn. there finished many mice the-cheese *Expl-VSO Expl-VSO

Bobaljik and Jonas (1996) argue that in TECs the subject is always VP-external. Evidence for this comes from the distribution of adverbs that are low relative to the subject.2 As shown in (15), the subject precedes the adverb completely, which is taken to mark the left edge of the VP. ostinn]. (15) aL klaruLu margar mys [ VP alveg there finished many mice completely the-cheese Bobaljik and Jonas specifically argue that in Icelandic TECs the subject is situated in [Spec, TP]. Hence, the structural representation for (15) is as in (16). It is well known that in Icelandic the direct object can be VP-external in certain contexts (Holmbergs Generalization contexts). (17) shows that the object itself precedes the adverb that marks the left edge of the VP.

1 Bures (1992) and Collins and Branigan (1997) were the first to point out that such a generalization concerning intransitivity exists. Here we substantiate their point by providing a systematic description of a variety of constructions attested crosslinguistically, and we aim to provide a unified account for all these cases. 2 As stressed by Bobaljik and Jonas, it is crucial that adverbs of the right kind be used. Sentential adverbs like probably cannot diagnose whether the subject is VP-internal or -external because these attach very high in the tree (see Alexiadou 1997, Cinque 1999).

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

199

(16)

AgrSP Spec a AgrS AgrS klaruui Spec TP

margar mysj

AgrOP AgrO Adv VP

VP

alveg
(17) aL klaruLu margar mys ostinn [ VP alveg]. there finished many mice the-cheese completely Many mice completely finished the cheese.

tj ti ostinn

In other words, in (17) both arguments are VP-external. Bobaljik and Jonas (1996:208, fn. 17) note that Icelandic also (marginally) allows sequences in which the object precedes and a quantificational subject follows the adverb, as in (18).3
3 Citing Thrainsson (1986), Jonas and Bobaljik (1993) note three examples in which a quantificational subject remains VP-internal in the presence of another complement. In two of them either the object has moved out of the VP, as in (18), or the object is a PP, as in (i) (from Jonas and Bobaljik 1993:30), conforming with the SSG. (i) aL stingur smjorinu [ VP einhver i vasann]. there put the-butter someone in the-pocket Somebody put the butter in the pocket. However, there is one example in which both the subject and the object remain VP-internal. ((ii) is from Collins and Thrainsson 1993.) (ii) aL sagLi Sveini [ VP einhver sogu]. there told Svein-DAT somebody-NOM a-story-ACC There told somebody Svein a story. Crucially, this is an example of a double object verb where the indirect object undergoes object shift to a position preceding the subject. For double object verbs it has been argued that there are two VP-internal Case-checking positions: the v introducing the subject, and the v introducing the indirect object (Collins 1997, McGinnis 1998). The v introducing the indirect object may check the Case feature of the theme, in which case the theme has undergone string-vacuous movement VP-internally. We assume that this is a possible derivation for the Icelandic example in (ii), which is then not a counterexample to the SSG. See the discussion in footnote 37.

200

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(18) aL maluLu sennilega husiL [ VP vandlega margir studentar]. there painted probably the-house carefully many students Many students probably painted the house carefully. Following a suggestion by Jonas and Bobaljik (1993), we assume that in these examples the subject is VP-internal, while the object has moved out of the VP. Crucially, what is not possible is for both the subject and a DP object to remain VP-internal. This is shown in (19). (19) *aL klaruLu [ VP alveg margar mys ostinn]. there finished completely many mice the-cheese Thus, Icelandic does not provide counterexamples to the SSG. In fact, it supports the SSG.4 The crucial example illustrating this is (19), which is the only case where both the subject and the object are VP-internal, and which is ungrammatical. 2.2 VSO Constructions in Celtic and Arabic VSO constructions in Celtic and Arabic constitute another apparent counterexample to the SSG. These were at one time also analyzed as involving two VP-internal arguments (see, e.g., the references in McCloskey 1991 and Fassi-Fehri 1993). However, later research has shown that in Celtic and Arabic VSO constructions the subject is always VP-external (and perhaps also the object; see, e.g., Demirdache 1988, Bobaljik and Carnie 1992, McCloskey 1996a,b, Benmamoun 1998).5 Consider (20a), an Irish example from McCloskey 1996b:269 (comparable data can be cited from Arabic). In (20a) the subject precedes the aspectual adverb, located low in the left periphery of the VP. This is evidence that the subject occupies a position outside the VP. When the adverb is placed before the subject, the sentence is ungrammatical. This is shown in (20b). (20) a. Chuala Roise go minic roimhe an t-amhran sin. heard Roise often before-it that-song Roise had often heard that song before. b. *Chuala minic Roise an t-amhran sin. heard often Roise that-song Hence, the correct analysis of VSO constructions in Irish is the one given in (21a) and not the one given in (21b).

4 Intransitive expletive constructions in Icelandic and English are similar in that in both languages the subject may remain VP-internal. Icelandic has the additional option that the subject may also surface VP-externally, an option not available in English (see, e.g., SigurLsson 1990, Jonas and Bobaljik 1993, Schutze 1997). In Icelandic transitive expletive constructions, on the other hand, the subject must be VP-external (except for the few Expl-VOS constructions found in the language). 5 It is worth pointing out that in Irish the subject must always be VP-external regardless of the transitivity of the predicate (cf. footnote 4 on Icelandic). Only when the subject of an unaccusative is a PP does it remain VP-internal (McCloskey 1996b).

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

201

(21) a. chuala Roisei minic [ VP ti an t-amhran sin] b. *chuala minic [ VP Roise an t-amhran sin] Thus, once adverbs are taken into account, it can be shown that VSO constructions in Celtic and Arabic are not counterexamples to the SSG either. 2.3 Subject Inversion in Romance A third case that could appear to challenge the SSG is that of subject inversion in Romance. As is well known, Italian (22) and Catalan (23) have VOS constructions. (22) Ha letto tutto bene Gianni. has read all well Gianni Gianni has read everything well. (23) Va llegir el llibre en Joan. read the book John John read the book. These have been analyzed as involving either a subject right-adjoined to the VP (Rizzi 1982) or a subject in a right specifier of VP (Bonet 1990). Under the right-specifier analysis, VOS constructions constitute a counterexample to the SSG. However, a number of researchers have argued that the right-specifier analysis is not correct. Recently two lines of argument have been proposed that both drop the right-specifier hypothesis. According to Cardinaletti (1999), who builds on work by Ordonez (1994, 1997), there is evidence that in VOS constructions the object undergoes leftward movement to a position higher than the subject. As (24) shows, in Italian a quantificational object can bind a pronoun contained within the subject in VOS constructions.6 (24) Ha visitato [ogni soldati]i suai madre. has visited every soldier his mother *His mother has visited every soldier. Josep Quer (personal communication) points out that similar binding facts hold in Catalan.7
6

In SVO orders a quantificational object cannot bind a pronoun contained within the subject (weak crossover). (i) ?*Suai madre ha visitato [ogni soldati]i . his mother has visited every soldier 7 There are some complications concerning the Italian and Catalan facts that have to do with whether or not the postverbal subject is focused and also with the type of quantificational elements involved in the binding relations (i.e., the binding facts do not hold with a negative quantifier), as pointed out to us by Cecilia Poletto, Giuseppe Longobardi, and Maria-Luisa Zubizarreta. Moreover, the transitivity restriction on inverted orders affects the position of PPs. In unaccusative constructions PPs may follow the subject, while in transitive constructions this is not possible unless the PP is right-dislocated (Bonet 1990, Ordonez 1997, Pinto 1997). This appears to correlate with the fact that in VOS constructions the subject necessarily carries contrastive focus, while in VS constructions this is not the case. Apparently, to describe these facts we need to take into account how the syntactic structure interacts with prosodic structure in these cases. Within Kaynes (1994) framework, Ordonez (1997) proposes that after the object has moved across the subject, additional movements take place that are responsible for the fact that PPs surface before the subject.

202

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Table 1 Position of subject and object in subject inversion constructions Icelandic a. b. c. d. [(Expl)-VS[VP O]] [(Expl)-VSO[VP ]] [(Expl)-VO[VP S]] [(Expl)-V[VP SO]] Celtic ? * * English/French * * * * Romance * * *

Alternatively, Ordonez (1997) proposes that VOS constructions in Romance are best analyzed as involving movement in three steps: (a) movement of the subject to a focus position; (b) movement of the object to a position over the subject, from which it can c-command the subject; and (c) movement of the remnant TP past this position. Leaving aside the question of which analysis is more successful, for us the main thing is that in both analyses both arguments cannot remain inside the VP; at least one of them has raised. Hence, in all apparent counterexamples to the SSG either the subject or the object or both are VP-external. Sequences in which both arguments can be shown to be VP-internal are ungrammatical in all languages. Table 1 summarizes the factual situation. Cases (ac) conform with the SSG, and they are, in principle, possible. For this reason, they are attested in one or the other language group depending on further properties of the language. What the SSG excludes is option (d), and indeed it is not attested in any of the languages we have considered. 3 More Cases That Fall under the SSG In this section we present two more cases that fall under the SSG: locative inversion in English, and ga/no conversion in Japanese. 3.1 Locative Inversion in English As is well known, locative inversion (LI) in English is licensed only when the verb is intransitive (Levin 1993). There is some controversy over which verb classes permit LI. Bresnan and Kanerva (1989) and Hoekstra and Mulder (1990) argue that it is limited to unaccusative and passive verbs, while other researchers, most notably Levin and Rappaport (1995:224227), point out that it is also found with representatives of several major subclasses of unergative verbs. (25) In the distance appeared the towers of a town. (unaccusative) (26) In the church sang a choir. (unergative)8 (27) In this cave was found an ancient treasure trove. (passive)
8 This example, as well as examples like In the woods lives a woman, cannot be claimed to have been shifted into an unaccusative frame by the PP, because it retains its atelic character while unergatives becoming unaccusatives via directional PPs shift from atelic to telic (see Levin and Rappaport 1995 for discussion).

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

203

Crucially, LI is not found with transitive verbs, while it is found with unergative verbs. This strongly suggests that the reason for the transitivity restriction on LI is not just locality. Specifically, if we were to propose that LI is ungrammatical in transitive constructions because the locative PP cannot skip across the external argument (Relativized Minimality, Attract Closest),9 then we would expect the same ungrammaticality to arise in the case of unergative verbs, which also have an external argument, a prediction contrary to the facts. On the other hand, the SSG correctly distinguishes between transitive and unergative verbs. In the former case both the subject and the object would remain inside the VP, violating the SSG, while in the latter case there is no overt object VP-internally. 3.2 Japanese ga/no Conversion Ga/No conversion in Japanese is a phenomenon in which the subject can appear either with a nominative case marker ga or with a genitive case marker no (see Watanabe 1996, Miyagawa 1997). It is possible in relative clauses and complex NPs but impossible in independent clauses. (28) John-ga/no nihon-e kaetta hi John-NOM/GEN Japan-to returned day the day on which John came back to Japan When an object is present, the presence of the genitive case marker is not acceptable. (29) John-ga/*no LGB-o kashita hito person John-NOM/GEN LGB-ACC lent the person to whom John lent LGB However, when the object itself is relativized, the presence of the genitive marker is again licit. (30) [John ga/no t katta] hon John-NOM/GEN bought book the book John bought Watanabe (1996) and Miyagawa (1997) argue that ga shows up whenever overt subject movement takes place, while no is a reflex of covert movement to AgrS (Watanabe 1996) or D (Miyagawa 1997).10 Thus, (29) and (30) are similar to the French SI examples (7) and (8), respectively. In both Japanese and French the subject cannot remain in its VP-internal base position in the presence of an object, unless the object moves out of the VP.11

9 Under the same reasoning, QI examples in which both the subject and the indirect object are VP-internal could be argued to be double violations of the SSG and Locality (see Collins 1997, where a locality approach is taken; for Collins it is crucial that LI takes place only in unaccusative constructions that do not include the external argument introduced by a light transitive v). 10 Ochi (1999) argues that no-marked NPs may also optionally raise overtly to check Case. This is incompatible with the present proposal. 11 VP-topicalization can be used to show that argument placement in German also falls under the same generalization,

204

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

4 A Digression on the Diagnostics: Adverbs in Quotative Inversion In the preceding sections we offered evidence from a wide variety of crosslinguistic constructions to support the SSG (13). It is important at this point to stress that the diagnostics by which each construction can be shown to have either a VP-internal subject or a VP-internal object are not identical from language to language, owing to language-specific and/or construction-specific properties. For instance, Costa (1998) has argued that in Portuguese only light manner adverbs (e.g., well/hard) can

as pointed out to us by Gereon Muller (see Haider 1993). (i) a. [Langusten gegessen] haben sie noch nie. lobsters-ACC eaten have they-NOM still never They have never eaten lobsters. b. (?)[Linguisten gegessen] haben das noch nie. have that-ACC still never linguists-NOM eaten Linguists have never eaten this. c. *[Linguisten Langusten gegessen] haben (aber nun wirklich) noch nie. linguists-NOM lobsters-ACC eaten have but now really still never While VP-topicalization of the object and the participle (ia) or of the subject and the participle (ib) is possible, topicalization of the participle and both arguments is ungrammatical (ic). Note that (ib) is grammatical only when the scrambled object is a pronoun, not when it is a DP (i.e., *Linguisten (NOM) gegessen haben diese Langusten (ACC) noch nie). Winfried Lechner (personal communication) points out that if scrambling of DPs is not taken to be Case-driven movement while pronominal movement is, then these facts provide evidence that the argument externalization requirement of the SSG reduces to Case, as will be argued in section 5. As David Pesetsky (personal communication) points out, the ban on multiple of-phrases in nominalizations also can be claimed to fall under the same pattern, if of is viewed as a structural Case marker (for discussion, see Alexiadou, to appear). (ii) a. the enemys destruction of the city b. the destruction of the city by the enemy c. *the destruction of the enemy of the city The effects of the SSG are also found in antipassive constructions. In focus antipassive constructions found in Mayan (see (iii), from Mam), a transitive subject exchanges its ergative morphology for absolutive agreement, a special oblique morphology must occur on the object, and the transitive subject undergoes focus movement (see Tada 1993, Watanabe 1996, Hale and Storto 1996, Richards 1997). (iii) a. Aa cheej ma -kub ky-tzyu-7n xiinaq. DEM horse ASP 3SG.A-DIR 3PL.E-grab-DS man The men grabbed the horse. b. Xiinaq x-chi kubtzyuu-n t-e cheej. man DEP.ASP-3PL.A DIR grab-APASS 3SG-RN horse The men grabbed the horse. Richards (1997) proposes that these facts can be regarded in terms of a constraint on argument overcrowding. In focus antipassives the subject must be extracted from a VP-internal position. In other words, absolutive agreement is a reflex of the fact that the subject has been extracted from a VP-internal position. In order for the subject to remain VP-internal (before movement), the object must be made into an oblique. An anonymous reviewer further points out that Mayanists sometimes distinguish between two types of antipassives, the agentive antipassive and the absolutive antipassive (SmithStark 1978, Nakamura 1996:106ff.). These two types are signaled by the same verbal morphology in some languages (e.g., Kekchi; see Berinstein 1985). The agentive antipassive allows the object to retain its argument status and requires extraction of the subject. On the other hand, the absolutive antipassive is of the familiar kind: the object is a PP and the subject need not be extracted. The reviewer suggests that if Richards (1997) is right that the antipassive morpheme signals that the subject remains VP-internal, then the two types of antipassives would represent the two ways in which a construction with both a VP-internal subject and a VP-internal object can be saved: either one argument (in the agentive antipassive, the subject) must be extracted, or one argument (in the absolutive antipassive, the object) must be made into a PP.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

205

be used to diagnose VP-internal material, unlike in English and French, where all types of manner adverbs can be used. Alexiadou (1997) and Cinque (1999) argue for Greek and Romance in particular that it is a combination of adverb and participle placement facts, not just adverbs, that diagnose whether arguments remain VP-internal or not. Kratzer (1995) and Diesing (1992) make use of the particles ja doch to diagnose VP-external scrambling in German. We have shown that aspectual and especially manner adverbs in combination with the relative position of participles can be used as diagnostics for whether arguments occur VP-externally or -internally, for principled reasons discussed in Alexiadou 1997 and Cinque 1999. Thus, in Icelandic and Irish (see section 2) the placement of arguments relative to adverbs is crucial. To a certain extent, the same holds for Catalan and Italian. For these languages, however, the mere existence of VOS constructions is sufficient to support the SSG once the right-specifier analysis for VOS has been argued to be inadequate. There is one case, however, where placement of arguments relative to adverbs can, in fact, be used to argue against the evidence we appealed to in order to establish the SSG for Icelandic and Irish, rather than simply being irrelevant as a diagnostic.12 This is the case of QI in English. As noted by Collins and Branigan (1997), adverbs cannot appear between the verb and the subject in this construction. (31) a. *Dont touch that dial! suggested abruptly the TV screen. b. Dont touch that dial! suggested the TV screen abruptly. To account for these facts, Collins (1997) proposes that adverbs may not be adjoined to vP (Collins and Branigan propose that adverbs cannot adjoin to VP), a proposal that cannot be adopted here given the discussion of Icelandic and Irish in section 2.13 But if we maintain that adverbs are adjoined to vP, as the facts in section 2 indicate, then the grammaticality of the subject-adverb order in (31b) poses a potential problem. This order could be viewed as an argument that in QI the subject is VP-external and, hence, that the intransitivity restriction this construction shows does not follow from the SSG.14 A similar problem is posed by the ungrammaticality of the verbadverb-subject order in (31a). The ungrammaticality of this example is difficult to explain if adverbs are indeed VP-external and subjects are indeed VP-internal in QI: if the verb raises, the subject stays in its base position as a specifier of vP, and the adverb is adjoined to vP, this would be exactly the order we would expect, contrary to fact. Nevertheless, we would like to maintain that (a) in QI the subject remains VP-internal and (b) in English (as in Icelandic, Irish, Romance, and Greek) manner and aspectual adverbs may generally adjoin to vP, marking its left edge. If the subject in QI moves out of the VP, then the floating quantifier facts discussed in (11) are left unexplained. In other words, the floating quanti-

We would like to thank an anonymous reviewer for pointing this out to us. Johnson (1991) proposes a similar explanation for English, in order to maintain that in English the verb raises to T, contra Emonds 1976, Pollock 1989. However, this leaves the classical crosslinguistic differences concerning V-raising in English and French unaccounted for. See the discussion in the main text below. 14 As we point out in section 5, the transitivity restriction facts found in QI could also be explained in terms of a locality constraint. In this case the transitivity restriction should follow from locality.
13

12

206

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

fier evidence directly contradicts the conclusion we could draw on the basis of the position of inverted subjects relative to adverbs in QI. If adverbs are not allowed to adjoin to vP, then it is hard to account for the systematic crosslinguistic patterns discussed in section 2. It would, perhaps, be conceivable to propose that Collins (1997) is right about English and we (following a large amount of literature) are right about Icelandic, Irish, and so on, but this would be the weakest possible position to take, for the following reasons. First, to suggest that the adjunction site of adverbs varies across languages is not a natural parameter. Second, if we assume that in English adverbs cannot be adjoined to vP, then we cannot account for the classical paradigm discussed in Emonds 1976 and Pollock 1989 concerning V-raising in French, as opposed to English (see footnote 13). A similar problem arises when we look at the position of manner adverbs relative to auxiliaries and main verbs in examples like (32), as acknowledged by Collins himself. (32) John will quickly solve the problem. For these reasons, we will maintain that adverbs in English may adjoin to vP. But then what accounts for the facts regarding adverbs in QI discussed above? We propose that adverbs are not allowed to adjoin to vP just in the case of QI because T and the subject must be adjacent in English for exactly the same reason v and the object must be adjacent. In the rest of this section, we outline our proposal in more detail. If it is correct, as the floating quantifier facts in (11) suggest, that in QI the subject remains vP-internal, then well-formed sequences like I am leaving! shouted John abruptly present evidence that manner adverbs can be adjoined lower than vP, presumably to VP (Chomsky 1995: 330ff.). On the other hand, the well-formedness of John often solved the problem shows that adverbs may also adjoin to vP, under the hypothesis that in English the verb remains VP-internal and undergoes short raising to v. Collins (1997) argues for this on the basis of the placement of seem relative to the experiencer in raising constructions like John seems to Mary to be intelligent (see also Chomsky 1995).15 Note, incidentally, that in QI manner adverbs may adjoin even higher than the verb, which is arguably in T (Collins 1997).16 For this reason, examples like ?I am leaving! abruptly shouted John are relatively acceptable. Indeed, manner adverbs are free to occupy various positions in the tree, as observed by Jackendoff (1972) and others. Suppose we are right that adverbs may adjoin to both VP and vP. Then two questions arise. First, what blocks adjunction to VP in examples like *John solved often the problem? This is ungrammatical in the vP-adjunction parse because in English the verb fails to raise beyond v; but it should be grammatical in the VP-adjunction parse, contrary to fact. A related question arises in QI, but this time it concerns adjunction to vP. The ungrammaticality of *I am leaving!

15 Johnson (1991) argues that well-formed coordination examples like Gary put the book on the table and the lamp on the sofa (Larson 1988) involve across-the-board V-raising and therefore provide evidence that the lexical verb in English raises to a higher position. Johnson proposes that lexical verbs move to , a position that can be identified with Kratzers (1994) Voice and Chomskys (1995) v. 16 Collins and Branigan (1997) propose that the verb raises only to AgrO . In a system that does away with AgrO , like the one assumed in Collins 1997 and adopted here, the next available landing site above v is T. Since the verb is to the left of the subject, it must be at least in T.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

207

shouted abruptly John shows that in this construction adverbs are not allowed to adjoin to vP, while the fact that they may follow the subject shows that they may adjoin to VP. We would like to treat the ungrammaticality of the two examples on a par, by proposing that adverbs cannot adjoin to VP in the first example and to vP in the second example for the same reason, namely, a Case adjacency condition. Since the seminal work of Stowell (1981), it has often been assumed that structural Case in English is assigned under adjacency. According to Stowell (1981) and Chomsky (1981), the assumption that Case is assigned under adjacency accounts for two sets of facts: first, that nominals precede all other verbal complements (see (33), from Johnson 1991:577); and second, that adverbs are not allowed to intervene between the verb and its DP complement, while they may intervene between the verb and its PP complement (see (34)(35), from Johnson 1991:580). (33) a. Gary introduced Mittie to Sam. b. *Gary introduced to Sam Mittie. (34) a. *Mickey visited quietly his parents. b. *Betsey sang loudly the anthem. (35) a. Chris walked quickly down the street. b. Mikey talked slowly to Gary. Abstracting away from the relative order of DPs and other complements that could perhaps be accounted for differently,17 we believe that the paradigm in (34) and (35) involving the relative position of adverbs with respect to DPs and PPs should be accounted for in terms of a Case adjacency condition, perhaps reducible to locality (Chomsky 1995). An alternative would be to suggest that this condition holds at PF and therefore is mainly found in languages with limited morphology (Stowell 1981, Chomsky 1981).18 Note that even if we were to assume, contrary to what we propose here, that adverbs like often and easily are necessarily vP-external, thus accounting for the ungrammaticality of *John solved quickly the problem by appealing exclusively to the failure of V-raising in English, we could not as easily do away with adjacency for examples like *John made last night a decision to leave town (as opposed to the well-formed John made a decision last night to leave town), which contain temporal adverbs that are never vP-external (see the discussion in Chomsky 1995: 330ff.). A Larsonian (1988) solution in terms of the assumption that temporal adverbs are necessar17 Examples containing a DP complement and a PP or clausal complement are typically cases falling under the dative alternation or the locative alternation, and there are many alternative explanations in the literature that account for the strict DP PP/clausal order without invoking adjacency. 18 Johnson (1991) accounts for the paradigm in (34) and (35) in terms of the proposal that (a) DP objects in English move to [Spec, VP], which is a Case position, and (b) adverbs may adjoin to V, thus preceding PPs and following DPs (see also Koopman and Sportiche 1998). We do not assume that adverbs adjoin to V (and intermediate positions in general); nor do we assume that Case-driven movement may target lexical projections. One could reformulate Johnsons proposal within the system assumed here and propose that English has a low Case position for objects below the position in which subjects are merged. This position could be identified with Traviss (1991) AspP. Such an analysis is incompatible with our account of the intransitivity restriction in English, which, we argue, follows from the SSG, which in turn ultimately reduces to Case.

208

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

ily generated lower than arguments because they are lower in the thematic hierarchy would not work either, in view of examples like ?John gave Bill yesterday a book (Pesetsky 1995:124), which are well formed with the temporal adverb to the left of the theme argument because there is no adjacency requirement between the verb and the theme in the double object construction (cf. the ill-formed *John gave yesterday Bill a book). So, let us conclude that in English there is an adjacency requirement between DP objects and the corresponding Case-checking head, which possibly holds at PF and which accounts for the fact that adverbs are not allowed to adjoin to VPs containing DP objects. If they did, they would prevent v (to which the verb moves overtly) from being adjacent to the DP object (which remains in VP). If there is an adjacency condition for accusative Case in English, then its effects should hold for nominative Case as well. In the general case these effects are difficult to detect because the subject raises to [Spec, TP].19 We propose, however, that QI presents such a case. In this construction adverbs are not allowed to adjoin to vP because if they did, they would interrupt the adjacency between T and the subject in [Spec, vP]. Because in QI the main verb raises to T (Collins 1997), it is always adjacent to the inverted subject. The effect of this is that in QI nothing may intervene between the verb and the subject, similarly to simple transitive sentences, where nothing may intervene between the main verb and the object. Thus, Case adjacency provides a parallel explanation for the relative position of adverbs with respect to the verb and the subject in QI and with respect to the verb and the object in transitive sentences without inversion. In the former case adverbs are not allowed to adjoin to vP; in the latter they are not allowed to adjoin to VP. One piece of evidence for this analysis is that particles may interrupt the adjacency between the verb and the subject in QI, exactly as in the case of objects. Compare (36) from Collins and Branigan 1997:45 with (37) from Johnson 1991. (36) Where do you want the concrete? called up Fanny to Max. (37) Mikey looked up the reference. Our proposal has one desirable consequence: namely, it can straightforwardly account for the fact that negation is not allowed to occur in QI (Collins 1997:3436). (38) a. *Lets eat, not said John just once. b. *Lets eat, said not John just once.

19 Johnson (1991:579580) argues that adjacency between T and the subject in English does not hold in view of grammatical examples like Gary probably has left or Gary in fact will leave, and he takes this as an argument against Case adjacency. We do not think, though, that these examples challenge Case adjacency. As pointed out by Pollock (1989) and discussed by Sportiche (1988), floating quantifiers in English may also occur in a pre-auxiliary position, unlike in French.

(i) The children all will leave. Sportiche suggests that in such examples the NP subject is topicalized, an option not possible in French, thus accounting for this difference between French and English. In Alexiadou and Anagnostopoulou 1998 we have argued that a topicalization analysis can account for the fact that adverbs may intervene between subjects and auxiliaries in English without resorting to adjunction to T. Under such an analysis, Johnsons examples are not counterarguments to adjacency.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

209

In Collinss (1997) analysis the ungrammaticality of (38a) is explained if negation is below TP and the verb moves to T. The ungrammaticality of (38b), however, is hard to explain because negation does not block overt V-raising to T, as auxiliary verbs show (see the discussion in Collins 1997). (39) John is not a teacher. In our account the ungrammaticality of (38b) is expected under the assumption that negation counts for adjacency. In addition, the account offers a uniform explanation for the adverb placement facts and the negation facts. 5 Case Externalization Let us now come back to the main topic of this article, namely, the SSG. How is the SSG to be understood? We argue that the SSG reflects a constraint on Case-checking configurations. Our argument crucially relies on the fact that the SSG is sensitive to the categorial status of arguments: namely, it discriminates between DPs and PPs.20 5.1 The SSG Reduces to Case Looked at more closely, the SSG breaks down into two parts: (a) when there are two DP arguments, at least one of them must externalize, and (b) if two arguments remain VP-internal, one of them must be a PP. For the moment let us concentrate on the first part. What it says is that there is a requirement for argument externalization quite independent of the EPP, which is standardly assumed to be the principle that forces arguments to move to a position other than their -position. Evidence for this way of looking at the SSG comes from two observations. First, it refers both to the subject and to the object. As long as one of them moves, the other one can remain in situ. Second, in the case of expletive constructions (discussed above) an expletive is checking the EPP-feature and yet some argument must also move. As a matter of fact, we propose that in all the cases discussed in the previous sections (except perhaps the ga/no case, which is a nominal construction), the EPP-feature is checked by an element other than the subject. Specifically, we distinguish four subcases of EPP checking: 1. Via an expletive in (T)ECs (Icelandic, English). 2. Via a locative PP in LI (Branigan 1993, Collins 1997). 3. Via V-raising in VSO languages (Celtic, Romance). Here we refer the reader to Alexiadou and Anagnostopoulou 1998 for detailed argumentation that in VSO/subject drop languages

20 Another possibility would be to suggest that the SSG is a filter banning multiple VP-internal arguments at PF. The problem with a filter approach, however, is that it is not the case that all XP arguments must vacate the VP. We have shown many examples where PPs can remain VP-internal. This is the case in SI, in QI, and in Icelandic, where quantificational subjects can be followed by PPs. If the SSG were correctly characterized as a PF filter, then it would not be clear why this filter distinguishes between PPs and DPs.

210

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

like Irish and Greek, V-raising to I0 checks and erases the EPP-feature. On this view, which we will be assuming throughout, the EPP is not a feature that can be checked only via Merge/Move XP. 4. Via a (wh)-phrase as in SI and QI. For SI we propose that the EPP-feature is checked in C and not in I and that I-to-C movement has the consequence that C and I match (see, e.g., Haider 1988, Muller and Sternefeld 1993, Watanabe 1996, Roberts and Roussou 1997; see also footnote 23). Evidence for this analysis comes from two facts: (a) SI is found in French, where subjects always have to move for EPP reasons, strongly suggesting that in this language the EPP is always satisfied by merging of an XP. This means that in SI some other element checks the EPP-feature so that the subject can remain in situ. (b) Deprez (1990) offers detailed argumentation that the verb is in C and not in I and hence that the EPP-feature of T can be checked by the phrase in [Spec, CP]. Essentially the same analysis holds for QI (Collins and Branigan 1997, Collins 1997). In QI as well, the EPP is checked by the quote or the null quotative operator postulated by Collins and Branigan (1997) and Collins (1997). Whether the quotative operator is in [Spec, CP] (Collins and Branigan 1997) or in [Spec, TP] (Collins 1997) is not crucial; what is crucial is that the EPP is checked not by the subject but by an operator, similarly to what happens in SI. We conclude that subject externalization in the cases discussed (TECs and VSO) is not triggered by the EPP, a conclusion that is reached for those languages by Bobaljik and Jonas (1996) and McCloskey (1996a,b) on quite independent grounds. We further propose that in these contexts subject externalization relates to Case. Externalization of the object also relates to Case, since it is sensitive to the category of the object: DP versus PP. It is usually assumed that DP objects require structural Case, while PPs do not (PPs are assumed either to have some sort of lexical Case or to assign Case to their DP complement inside their extended projection). On the basis of this reasoning, we propose to reformulate the SSG as in (40). (40) It is not possible for a subject and an object that need their Case to be checked both to be spelled out at the positions in which they are initially merged. Viewed this way, the SSG reduces to a constraint relating to multiple Case-driven movement of both the subject and the object without phrasal pied-piping.21 Why should there be such a constraint? Before we discuss this question, we must outline our theoretical assumptions.

21 An anonymous reviewer points out that this constraint is reminiscent of Bittner and Hales (1996) notion of Case competitor; in both theories a particular operation (Case marking, for Bittner and Hales account; movement out of the VP, for ours) must take place when two arguments occur within a certain domain. Note that our analysis supports views on ergativity according to which ergative languages are unaccusative systems (Nash 1995; for further discussion and references see Alexiadou, to appear).

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

211

5.2 Theoretical Assumptions Following Chomsky (1995, 2000), we assume that the computational system, using certain operations (Attract F and Move), manipulates a set of universal features to generate expressions. The operation Move involves matching of features between the target and the raised constituent and generalized pied-piping. Attract F followed by Move affects the phrase that is closest to the target and has the appropriate properties. Appropriateness depends on whether or not a feature F of the moved constituent may enter into a matching relation with a feature of the target. On the basis of the discussion in section 5.1, we further assume that there are two features associated with I (Chomsky 1995, Collins 1997): an EPP-feature and a Case feature.22 Both are formal features of the same type (i.e., uninterpretable nominal features on functional heads), and both are responsible for the movement operations performed in the computational system. Unlike Chomsky (1995, 2000), we assume that the EPP is not necessarily satisfied by Merge/Move XP. Merge/Move X0 can also check the EPP, in VSO pro-drop languages (see Alexiadou and Anagnostopoulou 1998 for detailed argumentation). In addition, we argue that the need to check Case features can also trigger the complex operation Move (which contains Merge XP as part of its definition; i.e., Case is not a reflex purely of agreement); otherwise, the facts discussed in this article, which crucially involve DPs with a structural Case feature, cannot be derived. We assume that the Case feature of v and the Case feature of T are formally identical in the strictest possible sense: that is, there is no nominative Case feature distinct from an accusative Case feature. There is just a Case feature on v and a Case feature on T. The two features are identical for the computational system (see Schutze 1997, Chomsky 2000). They come out differ ently in the morphological component, which spells out accusative on the DP relating to the lower feature in opposition to the DP relating to the higher feature by virtue of the fact that accusative is a dependent case downward in accordance with a case realization hierarchy suggested by Marantz (1992) (cf. Yip, Maling, and Jackendoff 1987). According to Marantz, dependent case can be assigned downward (accusative) or upward (ergative), when there is a chain distinct from the chain being assigned dependent case that does not have lexically governed case. Marantz argues that the EPP is the only principle that drives syntax and that abstract Case should be eliminated from the computational system entirely. We believe that abstract Case has a formal status inside the computational system, triggering argument externalization independently of the EPP; indeed, the main goal of the article is to argue for that view. However, we agree with Marantz that features such as nominative and accusative do not belong to the computational system, which only sees the formal feature [ Case] (see also Haeberli 1999 for a reanalysis of abstract Case in terms of categorial feature-matrices). Spelling out this Case feature as nominative or accusative is a matter of PF. In TECs the expletive and the subject relate to the two different features in I, the EPP-feature and the Case feature. There are a number of ways in which this relation can be structurally represented. In a layered-specifier approach the expletive and the raised subject are both specifiers
22

See also Boivin, to appear. What is crucial for us is to distinguish between the EPP and Case.

212

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

of TP (Chomsky 1995:sec. 4.10). In other approaches the expletive is in a higher projection (AgrSP, CP; see, e.g., Vikner 1995, Bobaljik and Jonas 1996). In what follows we will assume the layered-specifier approach. We further assume that there are cases where the EPP is checked in C by a nonsubject (the wh-phrase in SI, the quotative operator in QI), presupposing I-to-C movement (see the discussion in section 5.1), which permits the subject to remain in situ.23 Within the framework outlined above, there are two obvious ways of trying to account for (40). One could try to argue that if two DPs remain in the position in which they are merged, they pose a problem either for (a) locality or for (b) checking.24 In what follows we consider the two options in turn. We argue that under the most obvious assumptions concerning locality and checking (40) should not hold. 5.3 Not a Locality Constraint In principle, we could suggest that (40) reduces to a locality constraint, but this requires that locality be stricter in covert syntax than it is in overt syntax (see Collins and Branigan 1997).25

23 Note that if we assume a layered-specifier approach, in order to derive the Expl-VS surface order in TECs we must follow Chomsky (1995, to appear) in treating the verb-second phenomenon as a PF phenomenon. On the other hand, we treat residual verb-second in (e.g.) SI as crucially interacting with the computational system since we assume that EPP checking takes place in C following I-to-C raising (we thank an anonymous reviewer for raising this issue). In fact, the latter derivation is not strictly compatible with Chomskys (1995) system, where nothing can merge with a functional head containing a feature that requires phrasal pied-piping before this feature has been eliminated. Perhaps the right way to look at it is to suggest that T and C in these cases are merged together right from the start and that there is no I-to-C movement (Haider 1988). As for the Icelandic TECs, we could assume, following Nash and Rouveret (1997), that an Agr is inserted as a proxy, in order to create a landing site for the verb, without playing any other role in the checking relations. Adopting the layered-specifier model creates a potential problem for our treatment of VSO orders in Celtic and Semitic, where (at least in Celtic) the subject is in [Spec, TP]. In principle, one could argue that V-movement over the subject is PF-related or that the verb is situated in C. McCloskey (1996a,b) argues that V is not in C. Here again we could adopt the view that Agr is inserted as a proxy category in order to accommodate the facts. Notice that Benmamoun (1998) has argued that V-movement is PF-related for the Arabic VSO constructions. 24 There is another possibility, suggested by Chomsky (1995) and further developed by Espanol-Echevarra, Pinto, and de Wind (1998) (see also Nash and Rouveret 1998 for a related but different view): namely, that the problem is caused because the configuration resulting from covert movement of both the subject and the object is banned by bare output conditions at LF, because the object would end up c-commanding the subject. There are several reasons why we do not consider this option. One is that it is not obvious by which definition of c-command the object c-commands the subject, since the LF representation involves a head adjunction structure. Another is that we assume, following Richards (1997), that in principle it is possible to tuck in, leading to a configuration where the object is c-commanded by the subject. Finally, in a configurational -theory it is not clear why one has to look at the relative positions of the head and the tail of an A-chain to determine the hierarchical relations between the two arguments. We propose instead that the problem lies in the computational system. 25 Boskovic (1998) also proposes that locality is stricter for covert movement. An anonymous reviewer points out that Chomskys (2000) notion of phases would also limit covert movement provided that one continued to assume that covert phrasal movement exists, contra Chomsky (see Pesetsky 2000). Covert movement would only be able to attract phrases on the edge of a phase. If vP is a phase, then if the object does not move overtly, it should be unable to move covertly to T. Such an analysis could be implemented for objects if it were assumed that (a) objects check their Case in T (Pesetsky and Torrego, to appear) and (b) movement of the object to vP is triggered by an EPP-feature under a generalized EPP approach (Chomsky 2000). However, such an analysis would have nothing to say about the subject externalization cases discussed in section 2.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

213

If we stick to the assumption that locality works in the same way in both components, then locality principles cannot rule out a covert derivation in which both the subject and the object move to T. The specific construction we want to compare this derivation with is given in (41).26 (41) aL klaruLu margar mys ostinn [ VP alveg]. there finished many mice the-cheese completely There finished many mice completely the cheese. This is a well-formed Icelandic TEC with object shift. The expletive checks the EPP-feature and both arguments move overtly to their Case-checking positions. Any one of the ungrammatical constructions we have discussed so far would minimally differ from (41) in that the movements of the two arguments would take place covertly. Consider first how (41) is derived in the system developed in Chomsky 1995. First, v is merged with a Case feature, which requires phrasal piedpiping. The subject cannot check this Case feature, because it is in its -position and Case checking requires nontrivial chains. The object must therefore raise to the outer [Spec, vP], as shown in (42). It can raise across the subject because the subject and the target position are in the same minimal domain: both are specifiers of v.

(42)

vP

Objj Subj

VP

tj

Next, T is merged with a Case feature, which requires phrasal movement in order to be eliminated. The object cannot raise to T, because it no longer has a Case feature: its Case feature has been checked against v. The subject is the only XP that can be attracted by T, as shown in (42); it can raise to T across the object, since they are in the same minimal domain.

26 For QI it is possible that a locality approach also excludes the ungrammatical sentences because this construction involves a direct object skipping over both the subject and the indirect object (Collins 1997). However, this approach cannot be extended to the other cases we have discussed. See the discussion in footnotes 9 and 14.

214

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(43)

TP

Subji T0

vP

Vb

Objj ti

tv tV

VP tj

Let us now consider a covert derivation of this type. If in the covert component the derivation proceeds exactly as above, then it should converge for the same reason that it converges in (43). But the verb is already in T overtly; and in fact Chomsky (1995) proposes that after Spell-Out v raises to T, with the result that covert Case-driven movement of both the subject and the object always targets T to which v has adjoined. In this case as well, locality is not violated, even though the covert derivation in (44) is quite different from its overt counterpart in (43).

(44)

TP

Tmax

vP

Vb

Subj

tv V

VP

Obj

Tmax contains two attracting Case features, and there are two lower DPs with unchecked Case features that can move to T. By Attract Closest, the formal features of the subject move first to T because the subject is in a different minimal domain from both the object and T; then the formal

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

215

features of the object move to T. This movement is no longer blocked by the subject, since its intervening formal features have been removed. We conclude that under the most obvious derivations there is no way to derive (40) by locality. 5.4 No Problem with the Matching Relations Moreover, as we will argue, there is no obvious reason why the SSG should be operative from the point of view of the checking relations established. This will lead us to suggest that the SSG reflects the fact that one of the Case features on T or v must always be rendered inactive (i.e., eliminated by Merge/Move XP). We reinterpret this as an instruction to the computational system not to form a complex head that contains two Case features that have not been eliminated. This suggestion makes sense in a model of grammar that assumes there is a difference between the overt and covert components, but it cannot be phrased within a model that assumes the strict cycle and a single output (see Chomsky 2000, to appear). Consider first the checking relations resulting from the derivation (44), which we discussed in section 5.3 from the point of view of locality. In the ungrammatical (Expl)V[ VP SO] cases, V has raised overtly to T and S, and O will raise covertly to the complex Tmax, as in (45).

(45)

Tmax

FF(Obj)

FF(Subj)

Vb

If we assume that head movement results in adjunction structures and that FF(Subj) and FF(Obj) enter into checking relations with the most proximate sublabels of the target (see Chomsky 1995: 280) (i.e., they check the least deeply embedded feature of Tmax first), then checking should be no problem. By Attract Closest, FF(Subj) moves first to T, which contains two unchecked Case features, one on T and one on v. The least embedded feature is that of T. FF(Subj) enters into a checking relation with the feature of T, which is the closest feature to it. This checking relation is exactly the one we want. There is no feature mismatch between FF(Subj) and the Case features of T. Both features are erased. Next FF(Obj) raises to Tmax and checks its features against the only feature left, namely, the Case feature of v. Again the Case feature of the object and the Case feature of v match, and the derivation converges, incorrectly.

216

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

5.5 Locality and Checking in Chomskys (2000, to appear) Model Consider, finally, the state of affairs in the model discussed in Chomsky 2000, to appear, in which this derivation does not cause any problem from the point of view either of locality or of the checking relations established. There it is claimed that structural Case (nominative/accusative) is linked to -features (see George and Kornfilt 1981) in the sense that subject-verb and objectverb agreement are both reflexes of structural Case checking (Agree). The relation Agree can be satisfied by movement of the DP containing the matching feature, but it can also be established long-distance. In both cases the operation takes place in a strictly cyclic fashion; that is, the features of v are erased before T is merged. In the case of long-distance Agree, checking reduces to matching, which does not include covert movement of features. This being so, the derivation would proceed as follows: if v were merged without a Case feature whose elimination would require pied-piping of a phrase, an Agree relation would be established between the verb and the object. Similarly, an Agree relation would be established between T and the subject after T was merged. Thus, again there should be no reason why both the subject and the object could not remain inside the VP, contrary to fact. 5.6 A Proposal The facts in sections 2 and 3 show that at least one Case feature, either that of T or that of v, must be rendered inactive by overt movement of its respective DP. Crucially, in all the cases we have considered (except for the ones discussed in footnotes 34 and 36) the verb has raised overtly. We propose that there is a link between V-to-T raising and the SSG: the necessity for overt checking of at least one Case feature reflects a prohibition against the formation of a complex head that contains two active Case features. If the problem arises whenever a complex head is formed containing elements with active Case features, we can reformulate the SSG as follows: (46) a. By Spell-Out VP can contain no more than one argument with an unchecked Case feature. b. Interpreted as: v and T cannot both have active Case features when they form a complex head. We propose that such a head is an illicit item. Let us consider why. Formation of a complex head with two Case features would proceed as follows. First, v would have to move to T with its Case features still active. Under the assumption that head movement creates adjunction structures, the head created in this step would have the structure in (47).

(47)

Tmax

Vb

T Case

v Case

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

217

We assume a definition of c-command in terms of sisterhood and containment as in Chomsky 2000:116. (48) c-commands if is the sister of K that contains .

According to (48), heads cannot attract after adjunction because they fail to c-command outside the nonterminal node dominating them. At first sight, it appears that this would be sufficient to explain why (47) is problematic. In (47) the Case feature on T cannot attract the subject and the Case feature on v cannot attract the object, because of a failure of c-command. In fact, Chomsky (2000:117) points out that head adjunction is problematic if the head functions after adjunction in implementing agreement and movement, and we could suggest that the SSG is precisely an effect of this. On closer inspection, however, it turns out that such a proposal would be too strong because, crucially, if either the Case feature of T or the Case feature of v in (47) has been deleted prior to adjunction, the derivation converges. Let us suggest, then, following Chomsky (1995), that in head adjunction configurations the attractor is the node dominating the heads carrying Case and agreement features (in (47), Tmax ), because nonterminal nodes may c-command under sisterhood. We propose that in order for Tmax to attract, it must bear a Case feature that functions as the attractor. This means that in head adjunction structures, Case features must be passed up from terminal nodes to nonterminal nodes. There is an important constraint, however: only one active Case feature can be present on the nonterminal node. The presence of more than one Case feature on a single node is banned, for the same reason that verbs do not usually check two accusative Cases. As a result, in complex head formation inheritance of the ability to check Case is strictly limited by the condition in (49), which is inspired by Bakers (1988:122) Case Frame Preservation Principle (CFPP).27 (49) A complex X0 in a given language can have at most the maximal Case-checking features permitted to a simple X0 in that language.

27 Note that (49) could be a subcase of a more general principle regulating percolation of features. If this were so, one would expect to find effects of (49) in other areas of the grammar as well. An investigation of such areas lies outside the scope of this article. Bakers CFPP is formulated as follows: (i) A complex X0 of category A in a given language can have at most the maximal Case-assigning properties allowed to a morphologically simple item of category A in that language. There are some differences between our proposal and the approach underlying the CFPP. First, (49) involves head formation at LF, while Baker discusses morphological head formation. Second, Baker assumes that the CFPP holds at PF, while we assume that (49) holds at LF as well. Third, we assume that Tmax may inherit the Case-checking properties of v provided that T does not have active Case features, while the CFPP extended to configurations like (47) would allow Tmax to have only the same properties as T since the two are of the same category. Fourth, Baker assumes that the CFPP forces complex predicates to check the Case of one argument only, while we assume that the Case of more than one argument may be checked provided that each head checks Case prior to incorporation (see the discussion of double object constructions in footnotes 3 and 37). Note that (49) may also account for some of the constructions discussed in Baker 1988namely, the fact that either causatives of transitive verbs are completely ungrammatical (Berber) or the subject surfaces as an oblique (Chichewa, Malayalam, Italian, etc.). The restrictions on argument placement found in SI constructions and the intransitivity restrictions found in causative constructions can thus be given a uniform explanation. One could imagine that in certain languages a specific head is allowed to carry two Case features (see the discussion in Baker 1988). In this case one might expect that these languages would not show any transitivity restriction effects; that is, they would be counterexamples to the SSG.

218

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(49) is operative in configurations like (47), blocking both Case features from being present at the Tmax level. Tmax inherits the Case features of T, which is the least embedded head. Thus, T with an active Case feature is an intervener for the Case features of v.28 As a result, v fails to attract and the derivation crashes. On this conception it is necessary that at least one Case feature be checked before the complex head is formed, as stated in (50). (50) T Case or v Case must be eliminated before the complex head is formed. The intuition behind this is that there must be a very local relationship between an argument and its Case-checking head, which is destroyed as soon as a complex head is formed containing two unchecked Case features.29 (50) rules out multiple Case-driven movement operations targeting the same head. The difference between this derivation and other cases of multiple wh-movement that at first sight look similar is that the case we are discussing involves a many-to-many relationship, with many checkers and many checkees, while the cases of multiple wh-movement involve a one-to-many relationship, one wh-feature being checked against many wh-phrases (see Muller 1997, Richards 1997). Finally, the disjunction concerning PPs in the SSG can be understood if PPs are not attractable for Case. Collins (1997) has argued extensively that PPs are attractable for the EPP, but not for Case. Similar reasoning applies to DPs carrying inherent Case; they also do not count for Case attraction. Thus, we predict that the effects of the SSG will show up only with DPs that bear structural Case. Our analysis relies on the existence of two cycles, thus being incompatible with single-level models (ONeil and Groat 1995, Bobaljik 1995, Brody 1995, Chomsky 2000, to appear, Pesetsky 1998, Haeberli 1999). However, it limits the set of configurations that allow multiple XP-movement to a complex head, with the result that in many cases cyclic movement is enforced. One could try to reformulate our proposal in a single-level model by suggesting that either object checking is done within the vP phase, or it is postponed until v combines with T, in which case the problem arises. Since T does not constitute a phase, as claimed in Chomsky 2000, to appear,

This has been suggested to us by Alec Marantz (personal communication) and an anonymous LI reviewer. An alternative explanation for (50) can be stated in terms of an ambiguity for the purposes of Attract/Agree (Alexiadou and Anagnostopoulou, to appear). If the attractor is Tmax, the Case features of Tmax in (i) could be those of Vb or those of T.
29

28

(i)

T Case

Vb Case

T Case

v Case

Such an ambiguity is fatal; as a result, Tmax fails to attract. Since nothing can be attracted, the derivation is canceled.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

219

such a proposal can be implemented. We believe, however, that a reformulation along these lines would require a number of stipulations that are not necessary if we assume a distinction between an overt and a covert component. More specifically, in Chomskys (2000) system v would have to move to T before its features are checked, although computationally it would be simpler if the Case features of v were checked in the lower cycle. In a framework that assumes there is a difference between overt and covert operations, this problem does not arise, since in this case v and T always fall together, presumably because T, being an affix, cannot enter LF without being associated with lexical features for interpretational reasons at the output. Let us now look at how the ban on complex heads consisting of v and T with active Case features can account for the crosslinguistic patterns discussed in sections 2 and 3. In our discussion we do not include the Romance facts, because there are various ways in which the VOS order could be derived, as pointed out in section 2.3. 6 The Patterns 6.1 Icelandic In Icelandic the EPP-feature of T always requires phrasal pied-piping (see Bobaljik and Jonas 1996, Alexiadou and Anagnostopoulou 1998).30 For the cases we are discussing, the expletive is merged to check the EPP. The Case feature of T can be eliminated in two ways. One involves phrasal movement of the DP, a property related to the fact that Icelandic belongs to the group of languages that may allow [Spec, TP] for subjects (see Bobaljik and Jonas 1996). The other involves checking of the Case feature of T without phrasal movement, as shown by the fact that in intransitive constructions the associate of the expletive can be VP-external or VP-internal.31 This is shown in (51). (51) a. aL eru [ VP komni gestir hingaL]. there are arrived guests here There have arrived guests here.

30 In Chomsky 1995 this requirement was captured in the notion of strength, which was understood as the requirement that XP merge/move to eliminate a feature. Actually, in all the cases we are discussing the EPP must be satisfied in one of the ways suggested in section 5. However, the Case features of T and v can both require an XP to move, or not; that is, they can be both [ strong] and [ strong] (under the interpretation just given). In principle, the EPP-feature could be [ strong]; this is not excluded from our system. What is excluded is that the Case features of both T and v are [ strong]. Fanselow (1996, 1997) argues that this is the case in German. He also argues that in German the EPP-feature is [ strong], and he proposes that the Case features of the DPs are attracted by the corresponding verbal/Tense features in covert syntax. The Minimal Link Condition guarantees that the appropriate features arrive at the correct verbal/Tense Case feature complex no matter the order in which the DPs appear in VP before Spell-Out. Crucially, for Fanselow these types of feature chains are similar to clitic-doubling constructions. If this is the correct analysis for German, then this language does not pose a problem for the universality of the SSG (see the discussion on doubling in section 7.2). Thanks to Gereon Muller for pointing this out to us. 31 Here we differ from Bobaljik and Jonas, who assume that the formal feature checked in T is always strong (in the sense of footnote 30) and that in the cases where the subject remains VP-internal, T0 raises to AgrS0 and checks its feature against the expletive (see Alexiadou and Anagnostopoulou, to appear, for a critique of Bobaljik and Jonass analysis).

220

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

b. aL eru gestir [ VP komni hingaL]. there are guests arrived here There have arrived guests here. Moreover, the fact that object shift is optional leads us to propose that a similar situation holds for the Case feature of v. It can be checked by phrasal movement, but it need not be. Thus, in TECs the Merge XP requirement either on T or on v is forced by (46b)/(50).32 Let us examine the individual cases discussed in section 2 one by one. 6.1.1 [Expl-VO[ VP S]] The derivation resulting in an expletive-VOS order proceeds as follows. First, v specified as having a Case feature that requires phrasal pied-piping is merged. The object raises and eliminates the Case feature of v.33

(52)

vP

Objj Subj

VP

tj

Next T is merged: its EPP-feature always requires phrasal pied-piping, while its Case feature can be satisfied without phrasal movement. Since there is an expletive in the numeration, it is merged with T and eliminates the EPP-feature, because, by hypothesis, expletive Merge is less costly than XP-movement. Then v[ Case] raises to T overtly. After Spell-Out, FF(Subj) moves to the complex T, which contains only one active Case feature, that of T itself, as shown in (53).

32 Given that both features optionally require phrasal pied-piping, it is also possible to have a numeration where both features (the Case feature of T and the Case feature of v) require such movement. In this case (46b)/(50) does not require that both T and v be strong, only that one of them be. It is clear from the discussion of Icelandic in section 2.1 that Expl-VSO orders with VP-external subjects are much more frequent than Expl-VOS orders, which are judged quite marginal and are limited to quantificational subjects. As far as (46b)/(50) is concerned, both should be equally possible. At present we do not have an explanation for this. 33 We assume that it is unproblematic for the object to skip the subject since [Spec, vP] is not a potential checking position for the subject. This is a -position for the subject. v has a double function: it introduces the subject and checks Case. Of course, this does not account for Holmbergs Generalization effects. It is possible that V-movement plays a role for locality, in which case we would have to modify Chomskys suggestions. For our purposes what is crucial is that (Expl)-VOS orders do not violate locality, contrary to what Bobaljik and Jonas (1996) propose.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

221

(53) T0Case Vb V v

TP

vP

TCase Objj Subj

tv

VP

tV

tj

The movement is possible because the Case feature of v was checked overtly by the object before the complex head Tmax was formed. 6.1.2 [Expl-VSO[ VP ]] The second case to consider is an expletive-VSO order in which both arguments are VP-external. Here both v and T have Case features that require phrasal piedpiping. The derivation proceeds as above, the difference being that instead of covert movement of FF(Subj), overt XP-movement of the subject takes place, as shown in (54).

(54)

TP

Subji T0

vP

Vb

Objj ti

tv

VP

tV

tj

A similar derivation underlies Arabic and Celtic VSO constructions.

222

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

6.1.3 Expl-VS[ VP O] The third case is an expletive-VSO order in which the object remains VPinternal and the subject raises to T. Here, v does not have Case features that require XP-movement, and therefore the object does not move overtly. In this case the subject moves to T before the object ever reaches the complex Tmax. The subject moves overtly; the object moves covertly. It is never the case that the object checks its Case against T in the same cycle as the subject.34 6.1.4 Expl-V[ VP SO] Finally, consider the ungrammatical case in which both arguments remain VP-internal. Here the numeration contains a v and a T both bearing Case features that can be eliminated without phrasal pied-piping. The derivation proceeds as follows. First, v is merged, with the result that the object does not raise overtly. Then, T is merged. The expletive is merged next, eliminating the EPP-feature of T, and then v raises to T overtly, resulting in the formation of a complex head Tmax with two unchecked Case features. The derivation is canceled for the reasons discussed in section 5. We now turn to the languages that have an intransitivity restriction. 6.2 English and French English and French have neither TECs nor object shift.35 We take this to mean that the Case features of T and v can be satisfied without phrasal pied-piping. The EPP, on the other hand, always requires phrasal pied-piping. When the numeration contains an expletive checking the EPP-feature of T, the subject and the object must necessarily remain VP-internal. In these cases the derivation would proceed as in the ungrammatical case of Icelandic just noted in section 6.1.4.

34 Actually, this derivation is not so straightforward. As v does not have a Case feature that requires phrasal piedpiping, the derivation proceeds without movement of the object. Then T is merged, with an EPP-feature and a Case feature that both demand phrasal pied-piping. Before anything else happens, expletive Merge eliminates the EPP-feature of T. After expletive Merge takes place, there are two possible derivational steps to consider concerning the relative order of subject movement and v-movement to T: either (a) v raises to T first and then the subject raises, or (b) the subject moves to T first and then v raises. However, only option (b) conforms with (46b)/(50). The subject must be assumed to raise to T before v raises, checking and erasing the Case feature of T. Then v raises to T. 35 Johnson (1991) has argued, on the basis of particle constructions, that the object in English may undergo leftward movement. We believe that if there is movement in these constructions, then the target of movement is a position inside the VP, not a position where Case is checked (see the discussion in section 4). Evidence for this claim comes from the fact that comparable alternations exist in Icelandic but that crucially, such alternations do not depend on V-raising (Holmbergs Generalization), which has been used in the literature to diagnose movement of the object to a Case-checking position. To our knowledge, the most convincing cases for overt object movement have been made for exceptional-Casemarking (ECM) constructions (Lasnik 2000), where adverb scope relative to the ECM subject and binding facts suggest that the object moves overtly. If it is true that the object moves overtly in these constructions, then this analysis could be extended to other objects as well. Such an analysis would imply that the verb also raises in English and that the clausal architecture of English is quite different from what is usually assumed. Note, however, that there are obstacles to generalizing Lasniks arguments for object shift in English from ECM to other objects, as pointed out to us by an anonymous LI reviewer, because ECM objects obligatorily shift past particles, unlike other objects (compare (iab) with (ic)).

(i) a. John made Mary out to be a liar. b. *John made out Mary to be a liar. c. Mary looked (the information) up (the information). For French, as far as we know, it has not been argued that the object moves, unless it undergoes cliticization, where participle agreement has been taken to indicate that the object has passed through the specifier of the participle.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

223

(46b) cannot be avoided. As a result, there is an intransitivity constraint on subject inversion in these languages.36 We noted in section 1 that both French and English display constructions that void the transitivity restriction on inversion. We start with QI and then proceed to SI. In QI the quote or the quotative operator, which is the theme argument, checks the EPPfeature, permitting the subject to remain VP-internal. As noted earlier, the PP construction is permitted; the double object construction is not. Following standard analyses, we will assume that in the PP construction the theme has structural Case, while in the double object construction the goal has structural Case.37 In the case under discussion the theme argument is the quote, or the quotative operator, and the goal argument is the licit PP or the illicit DP. Starting from the grammatical example containing a PP, we will follow a suggestion in Richards 1997 and propose that an XP can check a Case feature on its way to check an EPP-feature by means of A-movement. In the unlikely case that the quote, which is a direct speech sentence, has structural Case (see Stowells (1981) Case Resistance Principle), it will check vs Case feature on its way to

36 In footnote 34 we crucially had to assume that the subject raises to T first, followed by movement of v, in order to account for the grammatical Expl-VS[O] orders in Icelandic. This raises a problem for English, where the verb does not raise overtly. If the overt derivation for Icelandic could be replicated for the covert syntax of English, then it would not be obvious why Expl-SVO orders are excluded. FF(Subj) would raise to T prior to v and FF(Obj), and the Case feature of T would be erased. Then, v would move to T; and, finally, FF(Obj) would check the Case feature of v. The situation in (46b) would not arise. For this reason, we must assume that in the LF cycle v-raising always happens first, followed by argument raising. But it is not clear why. Suppose that, unlike overt derivations, covert derivations proceed in a strictly cyclic fashion (Bures 1992). At LF, featural cyclicity (which derives from the notion of strength in Chomsky 1995) cannot guide derivations. Thus, the simplest assumption is that cyclicity guides the derivational path. In such an account Shortest Move must be respected at LF along with cyclicity. Given that we do not assume that equidistance is defined over nontrivial chains (see Alexiadou and Anagnostopoulou, to appear, for arguments against this view), we will adopt a proposal by Richards (1997), who argues that object raising across the subject can be made contingent upon V-raising on the basis of the Principle of Minimal Compliance in (i). (i) For any dependency D that obeys constraint C, any elements that are relevant for determining whether D obeys C can be ignored for the rest of the derivation for purposes of determining whether any other dependency D obeys C. (Richards 1997:232) Under the assumption that the verb and the subject are equally close to T, v-raising to T respects Shortest Move, and therefore, by (i), it can be followed by object raising to T across the subject. The result of the whole set of assumptions is that at LF v-raising will always happen first, followed by object raising. 37 It is uncontroversial that the theme is assigned structural Case when the goal is a PP. The Case of the theme and goal arguments in the double object construction is more controversial (see Anagnostopoulou, to appear, for discussion and references). If both of them check structural Case, then we must assume that at least one of them moves overtly out of the domain in which it is merged for the same reason that either the subject or the object must move overtlythat is, in order to conform with the SSG. An anonymous reviewer points out that the placement of the two objects in particle constructions can be used as an argument that in double object constructions where both arguments have structural Case at least one of them must move overtly. Specifically, it has been claimed (Emonds 1976, Koizumi 1993) that at least for some English speakers, in the double object construction the particle must occur between the two objects (ia) while in PP constructions it need not necessarily do so (ib). (i) a. The secretary sent (the stockholders) out (*the stockholders) the schedules. b. The secretary sent (the schedules) out (the schedules) to the stockholders. This pattern can be taken to show that it is impossible for both objects to remain below some point in the structure when both are DPs but not when one of them is a PP. If this is the way to interpret the facts in (i), however, the question that arises is what happens with optional placement of DPs to the left of the particle in constructions with a single object. See the discussion in footnotes 18 and 35.

224

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

[Spec, CP]. Otherwise, we must assume that v does not have a Case feature to be checked and the quote moves to C (if there is a quotative operator, then it presumably checks Case). The subject may remain VP-internal along with the PP, which does not have a Case feature, and it may move covertly to the T located in C to check Ts Case feature. When the goal argument is a DP, the situation is different. The VP-internal goal argument needs to check its Case feature against v, and the subject also needs to check its Case against T. As a result, this derivation is excluded. In the case of SI, again the object must raise overtly so that the problem of having two VPinternal arguments does not arise. Overt movement can be either wh-movement or clitic movement (see the examples in (8)).38 7 Some True Counterexamples to the SSG Are Apparent Counterexamples 7.1 Spanish, Romanian, and Greek To our knowledge, there are languages that do not observe the SSG. Here we will discuss Spanish, Romanian, and Greek. These languages have VSO constructions in which both arguments are VP-internal. ((55a) is Spanish; (55b) is Romanian; (55c) and following examples are Greek.) (55) a. Leyo ayer [ VP Juan el libro]. read yesterday Juan the book Juan read the book yesterday. b. L-a scris [ VP Ion o scrisoare Mariei]. CL-has written Ion a letter to-Mariei Ion has written a letter to Mariei. c. Elise sosta [ VP o Petros tin askisi]. solved correctly the-Petros-NOM the-problem-ACC Petros solved the problem correctly. These languages are V-raising languages (V-to-I0 ) (see, e.g., Rivero 1994, Ordonez 1994, Dobrovie-Sorin 1987) in which VSO constructions cannot be analyzed as involving I-to-C movement, since they also occur in embedded contexts (non-CP-recursion contexts; see, e.g., Iatridou and Kroch 1992).

38 Our analysis of the French patterns suggests that il does not eliminate the Case of T, contrary to what has been proposed for this type of expletive construction in Cardinaletti 1997. The problem with the view that il checks Case is that it is not clear what happens with the Case features of the single argument of an unaccusative predicate. One might suggest that the DP bears partitive Case (see Belletti 1988). However, Italian dialects that also have an il-type expletive show that this is wrong. Crucially, the presence of transitive sentences with il-type expletives, as in (i) from Tortora 1999, suggests that no partitive Case is assigned to the transitive subject.

(i) El a fat tut la serva. it has done everything the maids There have the maids done everything.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

225

(56) i idisi oti episkeftike o Janis tin Ilektra the news that visited the-Janis-NOM the-Ilektra-ACC the news that Janis visited Ilektra (complex NP) Evidence for the VP-internal position of the subject and the object comes from a combination of participle and adverb placement facts in periphrastic constructions (see Alexiadou 1997, Alexiadou and Anagnostopoulou 1998; and cf. (17) and (20)). (57) An ehi idi diavasi prosektika [o Janis to vivlio] . . . if has already read carefully the-Janis-NOM the-book-ACC If Janis has already read the book carefully . . . Now, what is the property that correlates with the availability of V[ VP SO]? We believe that this property is clitic doubling, and we propose that the availability of V[ VP SO] and the availability of clitic doubling should be linked. 7.2 VSO and Clitic Doubling Greek, Romanian, and Spanish have extensive clitic doubling of objects (see, e.g., Anagnostopoulou 1994 for Greek, Dobrovie-Sorin 1993 for Romanian, Suner 1988 for Spanish). (58) Tin ide i Maria ti gata. saw-3SG the-Maria-NOM the-cat-ACC Maria saw the cat.
CL-ACC

Anagnostopoulou (to appear) argues that in clitic-doubling constructions, clitics are formal features of DPs that move overtly without phrasal pied-piping. This proposal is based on the generalization in (59) concerning the distribution of DP dative arguments in Greek. (59) When a nominative argument undergoes NP-movement to T in the presence of a dative DP argument, the dative DP is not allowed unless it is a clitic or is doubled by a clitic. According to (59), clitic doubling is obligatory in contexts in which a lower argument undergoes NP-movement to T as in passives (60), raising constructions, and unaccusatives. (60) To grama ?*(tu) tahidromithike tu Petru htes. the-letter-NOM CL-GEN mailed-N-ACT.3SG the-Petros-GEN yesterday The letter was mailed to Petros yesterday. Anagnostopoulou argues that undoubled dative DPs are ruled out in NP-movement contexts because they block movement of lower nominatives to T as a result of Attract Closest (Chomsky 1995). Cliticization/clitic doubling systematically constitutes an escape hatch for Attract Closest violations because in clitic constructions the intervening features of the dative move out of the way of the lower nominative, and thus Attract Closest is respected. The proposed derivation is given in (61).

226

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

(61)

TP

T D, Case

VP1 V

DP1 goal/exp D 1 V1

VP2/IP V 2/I

DP2 nom D, Case 2 V2/I

In (61) the formal features of the higher DP move to T before the lower DP moves to T, and thus the higher DP no longer counts as an intervener for the movement of the lower argument. This approach leads to a view of cliticization as feature movement to T. On this view cliticdoubling languages are languages permitting feature movement without phrasal pied-piping. We propose to link this analysis of clitic doubling of objects to a proposal made in Alexiadou and Anagnostopoulou 1998 concerning subjects, in order to capture the correlation between clitic doubling and the apparent SSG violation showing up in VSO contexts. Building on Alexiadou and Anagnostopoulou 1998, where we follow Sportiche 1992, we analyze the relation between subject agreement on V and the subject in languages permitting VSO constructions as an instantiation of clitic doubling. Specifically, we propose that languages that permit clitic doubling of objects also permit feature movement without phrasal pied-piping in the case of subjects. Verbal agreement in such languages is a clitic, and Agr-NP pairs are similar to clitic-NP pairs. Under a generalized view of clitic doubling as feature movement without phrasal pied-piping, the subject in VSO constructions remains VP-internal, but its formal features have raised and its Case has been checked before Spell-Out. Since the Agr-NP pair in VSO constructions is a special type of feature chain in the overt syntax (see also Fanselow 1996, 1997), these constructions are only apparent counterexamples to the SSG. This accounts for the fact that VSO constructions in these languages are possible. 8 Conclusion The implications of our analysis are straightforward. The EPP and Case both force argument externalization. Argument externalization for Case reasons is forced by the fact that multiple Case checking without phrasal pied-piping is banned. The proposed system limits the set of

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

227

configurations that allow multiple XP-movement to a complex head, with the result that in many cases cyclic movement is enforced. References
Alexiadou, Artemis. 1997. Adverb placement: A case study in antisymmetric syntax. Amsterdam: John Benjamins. Alexiadou, Artemis. To appear. Functional structure in nominals: Nominalization and ergativity. Amsterdam: John Benjamins. Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, verb-movement and EPP checking. Natural Language & Linguistic Theory 16:491539. Alexiadou, Artemis, and Elena Anagnostopoulou. To appear. Covert feature movement and the placement of arguments. In The minimalist parameter. Amsterdam: John Benjamins. Anagnostopoulou, Elena. 1994. Clitic dependencies in Modern Greek. Doctoral dissertation, University of Salzburg. Anagnostopoulou, Elena. To appear. The syntax of ditransitives: Evidence from clitics. Berlin: Mouton de Gruyter. Baker, Mark. 1988. Incorporation. Chicago: University of Chicago Press. Belletti, Adriana. 1988. The Case of unaccusatives. Linguistic Inquiry 19:134. Benmamoun, Elabbas. 1998. Spec-head agreement and overt Case in Arabic. In Specifiers: Minimalist approaches, ed. David Adger, Susan Pintzuk, Bernadette Plunkett, and George Tsoulas, 110125. Oxford: Oxford University Press. Berinstein, Ava. 1985. Evidence from multiattachment in Kechi Mayan. New York: Garland. Bittner, Maria, and Ken Hale. 1996. Ergativity: Toward a theory of a heterogeneous class. Linguistic Inquiry 27:531604. Bobaljik, Jonathan. 1995. Morphosyntax: The syntax of verbal inflection. Doctoral dissertation, MIT, Cambridge, Mass. Bobaljik, Jonathan, and Andrew Carnie. 1992. A minimalist approach to some problems of Irish word order. Ms., MIT, Cambridge, Mass. Bobaljik, Jonathan, and Dianne Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27: 195236. Boivin, Marie Claude. To appear. En-cliticization, raising to subject, and Case. In Proceedings of CONSOLE 6. ` Bonet, Eulalia. 1990. Subjects in Catalan. In MIT working papers in linguistics 13: Papers on wh-movement, 126. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Boskovic, Zeljko. 1998. LF movement and the Minimalist Program. In NELS 28, 4357. GLSA, University of Massachusetts, Amherst. Branigan, Phil. 1993. Locative inversion and the Extended Projection Principle. Doctoral dissertation, MIT, Cambridge, Mass. Bresnan, Joan, and Jonni M. Kanerva. 1989. Locative inversion in Chichewa: A case study of factorization in grammar. Linguistic Inquiry 20:150. Brody, Michael. 1995. Lexico-Logical Form: A radically minimalist theory. Cambridge, Mass.: MIT Press. Bures, Anton. 1992. Recycling expletives and other sentences. Generals paper, MIT, Cambridge, Mass. Cardinaletti, Anna. 1997. Agreement and control in expletive constructions. Linguistic Inquiry 28:521533. Cardinaletti, Anna. 1999. On Italian post-verbal subjects. Ms., University of Venice. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1995. Categories and transformations. In The Minimalist Program, 219394. Cambridge, Mass.: MIT Press.

228

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step, ed. Roger Martin, David Michaels, and Juan Uriagereka, 89155. Cambridge, Mass.: MIT Press. [Also available as MIT Occasional Paper in Linguistics 15, MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass.] Chomsky, Noam. To appear. Derivation by phase. In Ken Hale: A life in language, ed. Michael Kenstowicz. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1999. Adverbs and the universal hierarchy of functional projections. Oxford: Oxford University Press. Collins, Chris. 1997. Local economy. Cambridge, Mass.: MIT Press. Collins, Chris, and Phil Branigan. 1997. Quotative inversion. Natural Language & Linguistic Theory 15: 141. Collins, Chris, and Hoskuldur Thrainsson. 1993. Object shift in double object constructions and the theory of Case. In MIT working papers in linguistics 19: Papers on Case and agreement II, 131174. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Costa, Joao. 1998. Word order variation: A constraint-based approach. Doctoral dissertation, HIL, Leiden. Demirdache, Hamida. 1988. Nominative NPs in Modern Standard Arabic. Ms., MIT, Cambridge, Mass. Deprez, Viviane. 1990. Two ways of moving the verb in French. In MIT working papers in linguistics 13: Papers on wh-movement, 4785. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. ` Dobrovie-Sorin, Carmen. 1987. Syntaxe du roumain. These de doctorat detat, Universite de Paris 7. Dobrovie-Sorin, Carmen. 1993. The syntax of Romanian: Comparative studies in Romance. Berlin: Mouton de Gruyter. Emonds, Joseph. 1976. A transformational approach to English syntax. New York: Academic Press. Espanol-Echevarra, Manuel, Manuela Pinto, and Maerten de Wind. 1998. A uniform analysis of Case adjacency in Romance. Paper presented at the Workshop on Inversion in Romance, Amsterdam. Fanselow, Gisbert. 1996. Minimal Link, phi-features, Case and theta-linking. GLOW Newsletter 36:3031. Fanselow, Gisbert. 1997. Minimal Link, phi-features, Case and theta-checking. Ms., University of Potsdam. Fassi-Fehri, Abdelkader. 1993. Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer. George, Leland, and Jaklin Kornfilt. 1981. Finiteness and boundedness in Turkish. In Binding and filtering, ed. Frank Heny, 105127. London: Croom Helm. Haeberli, Eric. 1999. Features, categories and the syntax of A-positions: Synchronic and diachronic variation in the Germanic languages. Doctoral dissertation, University of Geneva. Haider, Hubert. 1988. Matching projections. In Constituent structure, ed. Anna Cardinaletti, Guglielmo Cinque, and Giuliana Giusti, 101123. Dordrecht: Foris. Haider, Hubert. 1993. Deutsche Syntax-Generativ. Tubingen: Gunter Narr. Hale, Ken, and Luciana Storto. 1996. Agreement and spurious anti-passives. Ms., MIT, Cambridge, Mass. Hoekstra, Teun, and Rene Mulder. 1990. Unergatives as copular verbs: Locational and existential predication. The Linguistic Review 7:179. Iatridou, Sabine, and Anthony Kroch. 1992. On the licensing of CP-recursion and its relevance to the Germanic verb-second phenomenon. Working Papers in Scandinavian Syntax 50:124. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, Mass.: MIT Press. Johnson, Kyle. 1991. Object positions. Natural Language & Linguistic Theory 9:577636. Jonas, Dianne, and Jonathan Bobaljik. 1993. Specs for subjects. In MIT working papers in linguistics 18: Papers on Case and agreement I, 5998. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

229

Kayne, Richard. 1972. French relative que. In Current studies in Romance linguistics, ed. Frank Hensey and Martha Lujan, 255299. Washington, D.C.: Georgetown University Press. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, Mass.: MIT Press. Kayne, Richard, and Jean-Yves Pollock. 1978. Stylistic inversion, successive cyclicity, and Move NP in French. Linguistic Inquiry 9:595621. Kayne, Richard, and Jean-Yves Pollock. 1998. New thoughts on stylistic inversion. Paper presented at the Workshop on Inversion in Romance, Amsterdam. Koizumi, Masatoshi. 1993. Object agreement phrases and the split VP hypothesis. In MIT working papers in linguistics 18: Papers on Case and agreement I, 99148. MITWPL, Department of Linguistics and Philosophy, MIT, Cambridge, Mass. Koopman, Hilda, and Dominique Sportiche. 1998. The position of subjects. In Partitions and atoms of clause structure, ed. Dominique Sportiche, 4487. London: Routledge. Kratzer, Angelika. 1994. On external arguments. In University of Massachusetts occasional papers 17, 103130. GLSA, University of Massachusetts, Amherst. Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In The generic book, ed. Gregory N. Carlson and Francis Jeffry Pelletier, 125175. Chicago: University of Chicago Press. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335391. Lasnik, Howard. 2000. Syntactic structures revisited: Contemporary lectures on classic transformational theory. Cambridge, Mass.: MIT Press. Levin, Beth. 1993. English verb classes and alternations. Chicago: University of Chicago Press. Levin, Beth, and Malka Rappaport. 1995. Unaccusativity: At the syntaxlexical semantics interface. Cambridge, Mass.: MIT Press. Marantz, Alec. 1992. Case and licensing. In ESCOL 91, 234253. CLC Publications, Cornell University, Ithaca, N.Y. McCloskey, James. 1991. Clause structure, ellipsis, and proper government in Irish. Lingua 85:259302. McCloskey, James. 1996a. On the scope of verb-movement in Irish. Natural Language & Linguistic Theory 14:47104. McCloskey, James. 1996b. Subjects and subject positions in Irish. In The syntax of Celtic languages, ed. Robert D. Borsley and Ian Roberts, 241283. Cambridge: Cambridge University Press. McGinnis, Martha. 1998. Locality in A-movement. Doctoral dissertation, MIT, Cambridge, Mass. Miyagawa, Shigeru. 1997. Against optional scrambling. Linguistic Inquiry 28:125. Muller, Gereon. 1995. A-syntax. Berlin: Mouton de Gruyter. Muller, Gereon. 1997. Parallel movement. Ms., University of Stuttgart. Muller, Gereon, and Wolfgang Sternefeld. 1993. Improper movement and unambiguous binding. Linguistic Inquiry 24:461507. Nakamura, Masanori. 1996. Economy of chain formation. Doctoral dissertation, McGill University, Montreal, Quebec. Nash, Lea. 1995. Portee argumentale et marquage casuel dans les langues SOV et dans les langues ergatives: ` Lexemple du georgien. These de doctorat detat, Universite de Paris 7. Nash, Lea, and Alain Rouveret. 1997. Functional heads and proxies: An asymmetric set. Ms., Universite de Paris 8. Nash, Lea, and Alain Rouveret. 1998. Feature fission and the syntax of argument DPs and clitics. GLOW Newsletter 40:5051. Ochi, Masao. 1999. Attract F and optionally overt A-movement. Ms., University of Connecticut, Storrs. ONeil, John, and Erich Groat. 1995. Unifying Spell-Out and the LF interface in the minimalist framework. GLOW Newsletter 34:4445.

230

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

Ordonez, Francisco. 1994. Post-verbal asymmetries in Spanish. GLOW Newsletter 32:4041. Ordonez, Francisco. 1997. Word order and clause structure in Spanish and other Romance languages. Doc toral dissertation, City University of New York. Pesetsky, David. 1995. Zero syntax. Cambridge, Mass.: MIT Press. Pesetsky, David. 1998. Some optimality principles of sentence pronunciation. In Is the best good enough?, ed. Pilar Barbosa, Danny Fox, Martha McGinnis, Paul Hagstrom, and David Pesetsky, 337383. Cambridge, Mass.: MIT Press and MITWPL. Pesetsky, David. 2000. Phrasal movement and its kin. Cambridge, Mass.: MIT Press. Pesetsky, David, and Esther Torrego. To appear. T-to-C movement: Causes and consequences. In Ken Hale: A life in language, ed. Michael Kenstowicz. Cambridge, Mass.: MIT Press. Pinto, Manuela. 1997. Licensing and interpretation of inverted subjects in Italian. Doctoral dissertation, University of Utrecht. Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20:365424. Richards, Norvin. 1997. What moves where when in which language? Doctoral dissertation, MIT, Cambridge, Mass. Rivero, Mara-Luisa. 1994. Verb movement and the structure of IP in the languages of the Balkans. Natural Language & Linguistic Theory 12:63120. Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris. Roberts, Ian, and Anna Roussou. 1997. The EPP as a requirement on T. Paper presented at the Troms Workshop on Subjects, Expletives and the EPP. Schutze, Carson. 1997. INFL in child and adult language: Agreement, Case and licensing. Doctoral disserta tion, MIT, Cambridge, Mass. SigurLsson, Halldor. 1989. Verbal syntax and Case in Icelandic. Doctoral dissertation, University of Lund. SigurLsson, Halldor. 1990. V1 declaratives and verb raising in Icelandic. In Modern Icelandic syntax, ed. Joan Maling and Annie Zaenen, 4170. San Diego, Calif.: Academic Press. Smith-Stark, Thom. 1978. The Mayan antipassive: Some facts and fictions. In Papers in Mayan linguistics, ed. N. England, 169187. University of Missouri. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19:425449. Sportiche, Dominique. 1992. Clitic constructions. Ms., UCLA, Los Angeles, Calif. Stowell, Tim. 1981. Origins of phrase structure. Doctoral dissertation, MIT, Cambridge, Mass. Suner, Margarita. 1988. The role of agreement in clitic doubling constructions. Natural Language & Linguis tic Theory 6:391434. Tada, Hiroaki. 1993. A/A-partition in derivation. Doctoral dissertation, MIT, Cambridge, Mass. Thrainsson, Hoskuldur. 1986. V1, V2, V3 in Icelandic. In Verb second phenomena in the Germanic lan guages, ed. Hubert Haider and Martin Prinzhorn, 169194. Dordrecht: Foris. Tortora, Christina. 1999. Agreement, Case, and i-subjects. In NELS 29, 397408. GLSA, University of Massachusetts, Amherst. Travis, Lisa. 1991. Inner aspect and the structure of VP. Ms., McGill University, Montreal, Quebec. Valois, Daniel, and Fernande Dupuis. 1992. On the status of verbal traces in French: The case of stylistic inversion. In Romance languages and modern linguistic theory, ed. P. Hirschbuhler and K. Koerner, 325338. Amsterdam: John Benjamins. Vikner, Sten. 1995. Verb movement and expletive subjects in the Germanic languages. Oxford: Oxford University Press. Watanabe, Akira. 1996. Case absorption and wh-agreement. Dordrecht: Kluwer. Yip, Moira, Joan Maling, and Ray Jackendoff. 1987. Case in tiers. Language 63:217250.

SUBJECT-IN-SITU GENERALIZATION AND ROLE OF CASE

231

(Alexiadou) Department of Linguistics University of Potsdam Postfach 601553 14415 Potsdam Germany artemis@ling.uni-potsdam.de (Anagnostopoulou) University of Crete Department of Philology Section of Linguistics 74100 Rethymno Greece elena@phl.uoc.gr

You might also like