NEG.head

CL.3PL.DAT-CL.3SG.ACC.F

citesc. read.3SG.PR

I dont read it to them anymore. b. Nemaicitindu-le-o,. NEG.affix.ADV.read.GER.-CL.3PL.DAT-CL.3SG.ACC.F Not reading it anymore to them,
Depending on the theoretical approach adopted, the negative affix ne- in (18b), can be taken to display a morphologically subcategorized position to which a given root element must raise (in this case, the V + adverb complex), in order to saturate the affix. This negated verb complex would then raise to the position hosting the gerundive affix -ind, itself unsaturated. Alternatively, in minimalist terms, the entire gerundive form, nemaicitind not reading anymore is analysed as inserted fully inflected from within the lexicon. Since at Spell-Out, it precedes pronominal clitics, the inflected verb will have raised to check its morpho-syntactic gerundive feature in C (or M). Though we adopt this latter perspective, the essence of our story is not theory dependent. As a sentence negator, we assume the free negative morpheme nu not to be a syntactic clitic whose domain of cliticization is IP. Insofar as subordinate particles are concerned, namely subjunctive sa and infinitive a, we assume they cannot be viewed as affixes since they appear to the left of clitic material. However, we have seen that pronominal and auxiliary clitics do not block verb raising to their left, while the subjunctive particle sa has been argued (Dobrovie-Sorin 1994a) to block head-movement in examples
such as (19) below (i.e., the verb cannot raise to Comp because of the intervening head, sa). (19) a. Sa ntmple ce s-o happen what SE-CL.3SG.ACC.F May whatever happen, happen! se ntmpla! happen ntmpla! happen ntmpla! happen ntmpla! happen

SUBJ SE

* S ntmpl-se ce s-o * SUBJ happen-SE what SE-CL.3SG.ACC.F May whatever happen, happen! *ntmpl-se s ce s-o happen-SE SUBJ what SE-CL.3SG.ACC.F May whatever happen, happen! ntmpl-se ce s-o happen-SE what SE-CL.3SG.ACC.F May whatever happen, happen!
(19a) is a grammatical 'surrogate' imperative sentence (i.e., an imperative realized with the subjunctive, rather than with imperative morphology), in which the pronominal clitic se is situated above the lexical verb. In (19b-d), the lexical verb has raised above this pronominal clitic. In this case, the only grammatical imperative sentence is (19d), in which there is no subjunctive s morpheme. Notice that examples such as (19) only show that s is in complementary distribution with lexical verb raising in imperative sentences. This need not imply that s blocks head movement, as suggested by Dobrovie-Sorin (1994a). Under whatever theory of cliticization we adopt, it would be undesirable to have certain clitics block head movement while others fail to do so. We suggest that s does not, in fact, block head movement. Rather, the presence or absence of s in sentences like (19) is directly dependent on their imperative status. We assume that in imperatives, a null imperative operator will have to check its feature against a head marked for [+ imperative]. We further assume that the inflectional head hosting the [+ imperative] feature is the M(ood) head, which also serves as a host for the subjunctive particle s. Since the operator is null, the [+ imperative] feature will require a lexical host for retrieval at PF (phonological interface). This lexical host is either s or the raised verb, as illustrated in (20). (20) a. MP ty OP

either by a subject in Spec,IP or by the presence of subject agreement on the verb in I (i.e., null-subject languages). We assume that uninterpretable formal features (FFs) are essentially of two kinds: (i) selectional (or strong) and (ii) non-selectional (or weak), an option parametrized across languages and FF type. Non-selectional features will be defined as features which check/erase in-situ, without dislocation, as a result of the operation Agree, which only requires feature matching (i.e., identity) and closest c-command. Selectional features will be defined as features which can only be checked in a strict locality relationship, which we assume to involve either a Spec-Head or a head-adjunction configuration, depending on whether the respective formal feature triggers movement of an X or an XP. Notice that we depart from Chomsky (1998) in that we assume both the Spec-Head configuration and the head-adjunction configuration to be indicative of a feature checking relationship. By definition, selectional features will require agreement (i.e., feature matching) and movement (i.e., second Merge). We propose that parametric variation across languages is dependent on the nature of uninterpretable features. These assumptions are consistent with economy conditions since they eliminate movement unless absolutely necessary: movement is not an intrinsic requirement of feature-checking, but a result of parametrized formal feature properties. Crucially, under these assumptions, formal feature-checking will always be overt. Insofar as the EPP feature is concerned, we follow Chomsky (1998) who claims it to be a selectional feature cross-linguistically. Specifically, we view the EPP feature as a non-thematic position licenser, which is universally present on I, being, in effect, the privileged feature of Massam and Smallwood (1996). It therefore requires obligatory checking in a manner that will ensure the realization or validation of positions not forced by the Projection Principle, but by dislocation/movement and second Merge. Under our proposed feature dichotomy, selectional features may be checked either as an instance of the Spec-Head configuration or as an instance of head-adjunction. Consequently, the EPP feature on I may be in principle checked by verb raising to I or by subject NP dislocation to Spec,IP, depending on the nature of this feature. Specifically, we suggest that the EPP feature is not universally synonymous to [D] (cf. Massam and Smallwood 1996, but contra Alexiadou and Anagnostopoulou 1999). In Romanian, for example, we assume the EPP feature to be equivalent to a strong [V] feature on I. This strong [V] feature attracts verb movement to I, thus activating the IP domain. To conclude, we assume a universal EPP feature whose realization is parametrized across languages. Languages seem to vary as to whether they require [D], [V], or [T] as their EPP (selectional/privileged) feature. Let us call

In addition to the affixal passive, labelled canonical in Spencer (1991), Romance languages have a passive construction realized with pronominal se. This type of passive, misleadingly labelled reflexive passive (Spencer 1991) is extremely common in Romance languages and has all the relevant properties of the canonical passive. We illustrate with French and Romanian examples, in (61) and (62), respectively. (61) a. On mange cette racine. one eats this root People / One eats this root. (French)
Cette racine se mange (par tout le monde). (French) this root.NOM SE eats (by all the world) This root is edible. / This root is being eaten by everybody. Toata lumea mannca all people-the eat.3SG.PR Everybody eats apples. Se

(62) a.

mere. apples
mannca mere (de toata lumea). eat.3SG.PR apples.NOM (by all people-the) Apples are being eaten by everybody. * Mere se mannca (de toata lumea). apples. NOM. SE eat.3SG.PR (by all people-the) Apples are being eaten by everybody.
In (61)-(62), the logical subject is again demoted, to the discourse-prominence
benefit of the direct object. In Romanian the affixal passive construction is in free variation with the se passive, both being in effect instances of canonical passive, as shown in (63). (63) a. affixal passive: Au

nchise

portile
(de catre soldati). (by the soldiers)
been locked gates-the.NOM The gates have been locked (by the soldiers).
se passive: S-au nchis portile SE AUX.3PL locked gates-the The gates have been locked (by the soldiers).
Examples such as (63), showing the Romanian affixal passive in free variation with the se passive suggest that, syntactically speaking, a unitary analysis should be available for both of these constructions. We assume that clitic se essentially plays the same role that passive morphology does in the canonical passive; namely, it absorbs the external theta-role and Accusative Case (see, for example, Baker, Johnson and Roberts 1989). Moreover, notice that the same interpretational requirements hold for both types of passive structures. In order to be able to raise to the preverbal (IP-related) position, the Nominative NPs have to be specific (see 60b and 62c). The default position of the logical object promoted to grammatical subject, is the postverbal one. We now need to address the nature of this postverbal position in the syntactic tree, alongside the Nominative Case-licensing mechanism.

Passives and Minimalism

We have seen that in passives, the noun phrase marked for Nominative Case can surface pre- or post-verbally. The preverbal position is semantically restricted in the same manner as was discussed to be relevant for all active voice predicates. In section 2.4.4., we concluded that there is no preverbal canonical subject position in Romanian. Theoretically, this follows from the specifics of the EPP feature in this language. Therefore, for the purposes of Nominative Case-licensing, we are only interested in the postverbal position. The logical direct object (turned grammatical subject) of passives is merged as a complement of the verb. The issue we are concerned with is whether this object noun phrase can check/erase Nominative in-situ, or whether it needs to

Vikner (1992) argues that the object NP alle die Bucher all the books undergoes A-movement (i.e. object shift) out of the VP delimited by the negative adverb nicht not. From this derived A-position, the object scrambles to an A-bar IP-related position, stranding its quantifier in its first landing-site. As a result of the specificity constraint associated with object raising to an argumental position in Germanic, object shift has often been analysed as an instance of semantically driven movement (e.g. Diesing 1992, de Hoop 1996, Runner 1994). These analyses view object shift as a result of interpretation conditions applying in the syntax-semantics mapping which induce movement of NPs with an intrinsic or acquired definite/specific/strong interpretation out of the nuclear scope (i.e., the v/VP). Diesing (1992) follows Heim (1982) and assumes that quantificational structures at LF are tripartite. She proposes that the interface between the syntactic representation and the semantic representation takes the form of a mapping procedure that splits the syntactic tree into two parts; the two parts of the sentence are then mapped into the two major parts of the logical representation: the restrictive clause and the nuclear scope, as in (35). (35) The Mapping Hypothesis (Diesing 1992) (i) Material from the VP is mapped into the nuclear scope. (ii) Material from the IP is mapped into the restrictive clause. Runner (1994) offers a proposal which is closely related to Diesings. The author argues that object Agreement phrases (AgrOPs, rephrased as vP in the Minimalist theory) correspond to presupposed or specific information, namely material that is linked to the discourse (r la Pesetsky 1987). Object NPs
raise to Agreement whenever they are discourse linked. De Hoop (1996:134) argues that NP interpretation is associated with Case type. According to her, structural Case is divided as in (36); (36) (i) WEAK Case = the default structural Case, assigned at D-structure in a specific syntactic configuration and dependent upon verb-adjacency; (ii) STRONG Case = the structural Case assigned at S-structure and acquired as a result of movement (i.e., DP raising). Under this analysis, NPs assigned weak Case will have a WEAK semantic interpretation and will reside within the VP throughout the derivation. NPs with a strong Case will raise (out of the VP to AGRO) and will bear a STRONG reading (i.e. referential, partitive, generic, and generic collective). In this system, Case is viewed as a type-shifter, since, by definition, NPs that raise out of their base-generated position will be interpreted as semantically strong. Mahajan (1991) also suggests there is a link between object specificity and structural Case. Due to the fact that AGR is pronominal (and thus specific), the author argues that only specific NPs can (and must) be structurally Case marked by AGR. Non-specific NPs must receive structural case in some other manner. (1991: 265). In section 3.3, we argued that object raising in Romanian involves A-movement, due to lack of weak crossover effects, reversal of binding interactions and the possibility of dislocating quantified object NPs. However, clause-medial object raising in Romanian is not restricted in the manner illustrated for Germanic, since there is no specificity requirement involved (see discussion in section 3.1). Since objects of all semantic types can yield the VOS word order in Romanian, we will refrain from labelling this type of object raising as object shift (the term currently used for Germanic clause-medial object A-movement). While it is true that object raising in Romanian VOS constructions entails de-focusing of the object (in the sense discussed in section 3.1), a strong, topical interpretation is neither required nor acquired by these objects. What is crucial is that the raised object is outside the rhematic domain of the Romanian sentence (i.e., out of its VP-internal position). By escaping the rhematic domain, the raised objects in VOS constructions will be understood as part of the presupposition/the theme together with the verb, and never as topics of the sentence. We suggest the following pragmatic domains, centred around the verbal complex in I, to be operative in the Romanian clause: (37) (topic XP*) IP (V-to-v-to-I) (?P*) vP (Merge domain) z--------------------m z-----m theme rheme In (XP)VSO, for example, the subject and object NP, being situated within the

The fact that wh-phrases cluster together, points to a unique host and to feature-checking against a single head. Since in Romanian wh-phrases are required to check their [+wh] feature in a strict locality relationship, all of the wh-phrases in a multiple question must move up to the closest interrogative host, whether they are base-generated in the matrix or in the embedded clauses. This can mean extracting more than one wh-phrase out of an embedded clause, as in (20a), or extracting wh-phrases from different clauses, as in (20b). (20) a. Cinei cej ziceai [ CP ca si nchipuie ti tj]? who what say.2SG.PAST [CP that REFL imagines.3SG.PR ti tj]? Who did you say imagines what?
b. Cinei cej ziceai [ CP ca si nchipuie who what say.2SG.PAST [CP that REFL imagines.3SG.PR ti [CP ca ai spus tj]]? ti [CP that AUX.2SG said t j]] Who did you say imagines that youve said what? Moreover, multiple fronted wh-phrases cannot raise randomly in Romanian. They must obey a rigid Subject Object word order. Consider the examples in (21); (21) a. Cinei who what cek a dat given ti lui Mihai ti to Mihai tk ? tk
what who AUX.3SG given Who has given what to Mihai?

ti to Mihai

The object wh-phrase in (21) cannot precede the subject wh-phrase. This word order constraint can be accounted via Superiority (see also Comorovski 1996, Motapanyane 1998a, in press). Superiority is a concept originally introduced to account for the sequencing of moved elements. Pesetsky (1987:104), following earlier work by Chomsky, defines the following Superiority Condition, In a multiple interrogative, where a wh-phrase is in Comp and another is in situ, the S-structure trace of the phrase in Comp must c-command the S-structure position of the wh in situ. Informally then, Superiority will be defined as a constraint that forbids movement of a phrase over another phrase that is superior to it (where X is superior to Y if every maximal projection dominating X dominates Y but not conversely). According to Superiority then, the subject wh-phrase in (21) should raise before the object wh-phrase. Under the assumption that the order at the landing-site reflects the order of movement, we would expect to see the object precede the subject. What we observe is a Superiority effect that apparently affects the landing site, since this is where the wh-subject must precede any other wh-constituent. This can be formalized as the (Anti)-Superiority effect in (22). (22) (Anti)-Superiority: Overt movement into multiple specifiers is well-formed only if the c-command sequence of the moved wh-operators parallels the c-command sequence of their traces. Nevertheless, we suggest Superiority is still observed in Romanian. Specifically, we do not take linear order to reflect order of movement. Movement of the wh-object before the wh-subject would also violate economy conditions, formalized as the Minimal Link Condition of Chomsky (1995) (see chapter 1, section 1.2). Given that the subject is the closest candidate (Goal) of the Probe (i.e., the functional head X endowed with the [+wh] feature which needs to delete), it should move first. Notice, however, that in ditransitive clauses, the (Anti)-Superiority effect is somewhat relaxed insofar as the ordering of objects with respect to each other is concerned. Compare for example, (23) and (24) with (25) and (26), respectively. (23) a. b. Pe cine ce PE who what a

Citing examples with D-linked wh-phrases borrowed from Comorovksi (1986), Rudin (1988) concludes that Romanian is a language free of wh-islands. Richards (1997) assumes Rudin (1988) to be correct and makes the same predictions for Romanian. Recall, however, that Comorovski (1986, 1996) refers to wh-island violations in Romanian only in relationship to D-linked wh-phrases. Non-D-linked wh-phrases, however, observe wh-islands, on a par with their counterparts in IP-absorption languages. Consider the illustrations in (70a-e), which show wh-island effects to be operative in Romanian, too.
?? Cinei m-ai ntrebat [cej poate face whoi CL.1SG.ACC-AUX.2SG asked [whatj can do ' * Who did you ask what can do?' * Cinei m-ai ntrebat [ cuij whoi CL.1SG.ACC-AUX.2SG asked [to whom i-a dat ti tj cartea ]? CL.3SG.DAT-AUX.3SG given t i tj book-the ] '* Who did you ask gave the book to whom?' * Cinei ncearca Ion s [sa afle who try.3SG.PR Ions [SUBJ find out [pe cinej a ucis ti tj]] ? [PE whom AUX.3SG killed ti tj]] * Who is Ion trying to find out killed whom ? * Pe cine j m-ai ntrebat PE whom CL.1SG.ACC-AUX.2SG asked [cinei iubeste t i tj]]? [who loves.3SG.PR ti tj]] * Whom did you ask me who loves ? * Cej m-ai ntrebat what CL.1SG.ACC-AUX.2SG asked [cinei sa mannce ti tj ]]? [who SUBJ eat ti tj ]] * What did you ask me who should eat? pros pros

ti tj ]? ti t j ]

The examples in (70a-e), in which wh-phrases move across an embedded interrogative are all ruled out in Romanian. In all of these examples, the wh-phrases can felicitously check their [+wh] formal features in the embedded interrogative, thereby becoming inactive for further attraction (cf. MP98). Consequently, the matrix clause interrogative feature cannot attract further movement of these wh-phrases and the derivation crashes, yielding ungrammatical results. Chomsky (1998) argues that wh-islands act as a defective intervention constraint, since the effects of matching a higher probe should be blocked. Specifically, feature-checking should proceed against the first Agreeing functional head (i.e., Probe), after which the wh-phrase should become inactive to further attraction for the purposes of checking a higher [+ wh] FF. This effect is illustrated in the Romanian examples in (70), which can

interpretation of sentences such as (69) repeated here as (75), which contrast in interpretation with (76), in which both wh-phrases reside within the wh-island. Consider (75) and (76) below. (75) a. Pe care copii i PE which children i ii-a nu stii [ cinej not know.2SG [ who j invitat tj ti]? CL.3PL.ACC.M-AUX.3SG invited tj t i] '* Which children don't you know who invited?' [- distributive]
b. Cu care candidat i nu stii [ cinej a votat tj ti]? with which candidate i not know.2SG [ who j AUX.3SG voted t j ti] '* For which candidate don't you know who voted?' [- distributive] (76) a. Nu not ii-a [pe care copii i [PE which children i tj ti ]? CL.3PL.ACC.M-AUX.3SG invited tj t i] 'Don't you know who invited which children?' [+ distributive] [cu care candidat i cinej a stii know.2SG invitat cinej whoj

b. Nu stii ti]?

votat tj
not know.2SG [with which candidatei whoj AUX.3SG voted t j ti] 'Don't you know who voted for which candidate?' [+ distributive] In (75), with raising of the D-linked wh-phrases out of the embedded interrogative, the only available reading is the one in which these wh-phrases are interprteted as the topics of the following discourse. In contrast, in (76), in which the wh-phrases reside within the embedded question, the only available reading is a distributive, 'pair-list' one (in the sense of Beghelli 1997). In view of the interpretational differences in (75) and (76), we claim that obviation of wh-island effects with D-linked wh-phrases in Romanian is independent of [+ wh] feature-checking and conclude that Romanian shows wh-island effects, at least with non-D-linked wh-phrases. Recall that this is a property associated with IP-absorption languages.
4.7.2. Local scrambling: A- or A-bar movement
Another characteristic of IP-absorption languages is that they exhibit local scrambling with properties of A-movement (Richards 1997). According to Richards, the Hungarian example in (77a) is ungrammatical due to a weak crossover violation. When object quantifier scrambling applies, as in (77b), the weak crossover violations in (77a) are fixed, indicating A-movement. (77) a. b. *Nem not szeret loves az the proi anyja mother-his mindenkiti. everybody.ACC
Nem szeret mindenkiti az proi anyja ti. Not loves everybody.ACC the mother-his His mother does not love everybody (Hungarian, Richards 1997:30; author does not provide traces)
In (77a), the quantifier object undergoes A-bar moment at LF leaving behind a variable, which is illicitly coindexed with a pronoun to its left; hence, the ungrammaticality. In (77b), the object quantifier has scrambled locally. In this case, there is no weak crossover violation, which means that the trace left behind this local move is not a variable (since it allows coindexation by a pronominal on its left). Richards further shows that Serbo-Croatian and Japanese (both IP-absorption languages) pattern identically. On the other hand, a CP-absorption language such as Bulgarian, lacks the above switch in grammaticality. Consider (78). (78) a. b. *Majka mu obica vseki covek. Mother his love every person *Vseki covek obica majka mu. Every person his mother love Hisi mother loves everyone i. (Bulgarian, Richards 1997:31)

structure-dependent functions assigned in some specific structural relation (i.e., an appropriate specifier-head relation). This analysis draws on Rizzis earlier assumptions (1991) regarding affective operators (i.e., [wh]- and negative operators). Consider the WH-Criterion (Rizzi 1991) introduced in chapter 4 and repeated below as (6). (6) WH]; B. An X [+ WH] must be in a Spec-Head configuration with a WH Operator. Rizzis Topic/Focus Criteria are a means of formalizing licensing conditions for Topic, Focus (as well as other affective operators). The entire format is similar to Chomskys feature checking mechanism. Rizzi also addresses some of the distinguishing properties between topic and focus and we offer a summary of the most salient differences. In Romance, the topic-comment articulation is typically expressed by the construction that Cinque (1990) has called Clitic Left Dislocation (CLLD), involving a resumptive clitic coreferential to the topic, as in (7). (7) Il tuo libro, lo ho letto. Your book, I have read it. (Italian, Rizzi 1995:5) WH-CRITERION (Rizzi 1991) A. A WH Operator must be in a Spec-Head configuration with X [+
The focus-presupposition articulation can be expressed in Italian by preposing the focal element and assigning it special focal stress, as in (8). Rizzi argues that in Italian this structural option is restricted to contrastive focus (i.e., (8) presupposes that you believe that I have read something different from your book, and corrects this belief). (8) IL TUO LIBRO ho letto (, non il suo) Your book I read (, not his) (Italian, Rizzi 1995:5)
Both topic and focus are argued to involve A- constructions, but whereas topics involve resumptive clitics, focalized constituents disallow them, as illustrated in (9). (9) * IL TUO LIBRO lo ho comprato (, non il suo) Your book I bought it (, not his) (Italian, Rizzi 1995:8)
This, coupled with the fact that a topic does not give rise to weak crossover, which is consistently detectable with focus, points to the major conclusion that focus is quantificational while topic is not. Rizzis (1995/97) analysis for contrastive focus then is very much in line with Kiss (1995b, 1998).
5.1.3. Zubizarreta (1998)
In a vein reminiscent of Kiss (1995b, 1998), Zubizarreta (1998) equates focus in a statement with that part of statement that substitutes for the wh-phrase in the context question (see also Kiss 1998). The author further distinguishes between new information (presentational) focus and contrastive focus. The conclusions with regards to the syntax and semantics of these two types of focus are strikingly similar to those presented in section 5.1.1. and will not be discussed in any detail here. Instead, we briefly outline Zubizarretas analysis for contrastive focus, since it bears interestingly on the Romanian data. In this theory, contrastive focus is argued to have two effects. It negates the value assigned to a variable in the context statement (as indicated by the implicit or explicit negative tag associated with contrastive focus), and at the same time, it introduces an alternative value for such a variable. Consider for illustration the contrastive utterance and its context statement (in square brackets) in (10): (10) John is wearing a RED shirt today (not a blue shirt). [John is wearing a blue shirt today.] (Zubizarreta 1998:7) Zubizarreta (1998) further discusses properties of the preverbal field in Spanish and Italian and concludes that the two languages have different structural realizations for focus. For Spanish, the author argues for a generalized TP analysis, proposing that within a view of syntactic structure in which heads consist of features that need to be checked against other heads, languages with a generalized TP may be said to allow a certain amount of feature syncretism. (Zubizarreta 1998:100). Consequently, Tense is viewed as a syncretic category (in the sense of Giorgi and Pianesi 1996), in which the feature T(ense) may combine with discourse-based functional features, such as topic, focus, or emphasis, yielding the syntactic categories T/topic, T/focus, T/emphasis. Such an analysis is argued to be desirable in view of Chomskys (1995) minimalist approach to syntactic structure, since feature syncretism will ensure a minimal structure in a given derivation. In Spanish, different types of

constituents may occupy Spec,TP: topics, emphatics, focused phrases (including wh-phrases) and subjects. T is thus seen to play a crucial role in checking nominative Case, as well as discourse-based features that belong to the outer layer of the clausal structure. However, a phrase may not check more than one type of feature in a given specifier-head configuration. In other words, a phrase may not simultaneously check an intrinsically grammatical feature such as Case and a discourse-based feature such as topic, emphasis, or focus. Moreover, while there can be several topics in Spanish, at most one functional focus feature is allowed per sentence for focus-checking purposes. For Italian, on the other hand, a different analysis is adopted. It is argued, following Belletti and Shlonsky (1995), that Spec,TP is occupied exclusively by the subject and that fronted focused phrases, emphatics and topics are left-dislocated (that is, they occupy a position above TP). Several differences between Spanish and Italian support such a dichotomy. For example, in contrast to Spanish, Italian disallows VSO word order and post-verbal subjects are right-dislocated, where right-dislocation is derived from left-dislocation via leftward adjunction (following Kayne 1994). These facts suggest that in Italian, nominative Case must always be checked overtly in Spec,TP. Moreover, in Italian, but not in Spanish, the preverbal focused or emphatic constituent need not be adjacent to the verb. Zubizarreta (1998) cites the examples in (11). (11) a. QUESTO Gianni ti dira (non quello che pensavi). this Gianni to-you will-say (not what (you) thought) (Italian, Rizzi 1995:48) Qualcosa, di sicuro, io farn. something surely I will do (Italian, Cinque 1990:15)
These facts are taken to suggest that Italian has a Focus or Emphasis projection located between CP and TP (cf. Rizzi 1995/97). In effect, the functional feature T in Italian cannot constitute a syncretic category with the functional feature topic, focus or emphasis (as is argued for Spanish). To sum up, Zubizarreta views focus as a syntactic feature incorporated onto T in generalized TP languages, such as Spanish, while allowing for the projection of a Focus Phrase in languages for which there is enough empirical evidence to support a distinct Focus head (i.e., Italian).
5.1.4. Erteschik-Shir (1997)
Erteschik-Shir (1997) uses the term focus structure (f-structure) to
characterize structural descriptions (SDs) annotated for topic and focus constituents. F-structure feeds both PF, since this level provides explicit phonetic intonation, as well as semantics (i.e., it is accessible and visible to both). F-structure theory is a pragmatic theory which is concerned with felicity conditions on the relation between sentences and context. Thus, the function topic can only be assigned to constituents which are already in the hearers attention. Focus is shown to be of two types. New information/presentational focus (plain focus in the authors terminology), which is defined as the (intension of a) constituent c of S which the speaker intends to direct the attention of his/her hearer(s) to, by uttering S. (Erteschik-Shir 1997:11). This type of focus is a discourse property which is assigned to a constituent in a context of conversation. Contrastive focus (operator focus) is argued to be contextually constrained to occur only if a contrast set is available. In (12), for example, if PETER is to be interpreted contrastively, {Susan, Peter} must be members of a contextually defined set. (12) Speaker A: You saw Susan at the party. Speaker B: No, I didnt see SUSAN, I saw PETER. (Erteschik-Shir 1997:121) Contrastive foci are by definition metalinguistic, since a previous utterance (possibly implied) is being objected to. Moreover, contrastive focus is assumed to be unique, since one cannot object to more than one implied utterance at a time. Erteschik-Shir (1997:121) further suggests the f-structure in (13b) for the sentence in (13a) with a contrastive interpretation. (13) a. b. A MAN is intelligent. [ [a man FOC] [a woman ] TOP

V-adjacenc wh-phrases Focus Non-D-linked quantifiers Topics D-linked quantifiers + + + Complementary distribution with other operators + + + Unordered co-occurrence + +
5.4. Evidence for A-bar movement
The verb-adjacency and interaction properties summed up in table (47) point toward a uniform treatment of sentence-initial operators in Romanian. Intuitively speaking then, these operators are expected to show parallel properties under a movement analysis. In this section, we discuss shared A-bar properties between focus- and wh-movement, as well as further parallels between contrastive focus and bare quantifiers.

5.4.1.

Contrastive wh-movement

focus-movement

It has been argued (cf. Cinque 1990, Rizzi 1990) that there are two ways in which a gap can be related to its antecedent. Non-NPs are identified through antecedent government, which is a local relation, while NPs are identified by binding, a non-local relation. Non-NPs (i.e., adjuncts) cannot be identified by binding, since binding requires that the antecedent and the bound constituent have the same index. Cinque (1990) has argued that since only NPs can have referential indices, only NPs can be identified via binding. The difference in manner of gap identification has obvious consequences on the length of movement. While long movements of NPs can in principle produce well-formed chains, only local movements are allowed for non-NPs. To theorize this, Cinque (1990) argues the two types of identification are subject to different types of barriers, which produce two types of islands (i.e., strong and weak islands). Strong islands affect both NPs and non-NPs, while weak islands exclusively affect non-NPs. There is evidence in Romanian that focus movement is subject to the same weak and strong island constraints as movement of wh-phrases. Let us first consider strong island constraints, which include extraction out of a clause dominated by a noun phrase (CNPC) and extraction out of an adjunct clause, a consequence of conditions on extraction domains (CED). In (48) through (51), NP and non-NP preverbal focused elements are shown to be subject to both CNPC and CED, in a manner parallel to moved NP and non-NP wh-phrases. Let us first consider extraction out of a clause dominated by a noun phrase (see 48-49). A. (i) STRONG ISLANDS: CNPC (extraction out of a clause dominated by a noun phrase): ntlnit un elev [care a scris met a student [which AUX.3SG written o scrisoare foarte ngrijit]. a letter very carefully] I met a student who worded a letter with great care. * Ce ai ntlnit un elev [care a what AUX.2SG met a student [which AUX.3SG scris t foarte ingrijit]? written t very carefully] * What did you meet a student who had worded very carefully?

PE nobody

iubeste mama luii. mother-the his * His i mother loves anyone i. (note that this sentence is ungrammatical in English)
PE anywho CL.3SG.ACC.M loves
To sum up, evidence from both parasitic gaps and weak cossover in Romanian point toward an analysis of definite focused elements as non-quantificational operators, on a par with definite wh-phrases. This is a desirable conclusion in view of the semantic and syntactic similarities between the two types of operators.

5.4.5. Summing up

In the sections throughout 5.3 - 5.4, we focused on the interaction among the elements present in the Romanian preverbal field. We saw that bare quantifiers, wh-phrases and focused phrases behave alike in terms of A-bar movement properties. We showed that co-occurrence among these operators is illicit in the preverbal field, a constraint directly following from the verb-adjacency requirement, which is a specific licensing condition on these operators. These requirements were seen to be distinct from those involved in topicalization or D-linked quantifier movement, which do not require verb-adjacency or special ordering. We therefore conclude that verb-adjacent constituents target Spec, IP, while topics (including D-linked quantifiers) scramble and adjoin to IP. We further discussed the types of chains involved in operator movement and concluded that a distinction needs to be made between quantificational chains, which prohibit clitic doubling, and anaphoric chains, which require clitic doubling. Specifically, A-bar movement into the left-periphery will involve quantificational chains when the moved element lacks a coindexed clitic (i.e., with non-D-linked or non-distributive constituents), and anaphoric chains when the moved element requires a coindexed clitic (ie., is D-linked or distributive). Under this analysis, topics (including D-linked quantifiers) form anaphoric chains in Romanian, given that they require clitic doubling. Our findings are summed up in table (71). (71)
Vadjacency omplement-ar distribution with other operators + + + + A-bar mvt. to Spec, IP + + + + A-bar mvt. as cramblin to IP resenc of clitic oublin + +
indefinite wh-phrase (e.g. cine who) D-linked wh-phrases (e.g. care which) indefinite Focus (e.g. DRAGOSTE love) definite Focus (e.g. DRAGOSTEA the love) non-D-linked (indefinite) non-distributive quantifier (e.g., BQ: nimeni nobody,

+ + + +

cineva someone) non-D-linked (indefinite) distributive quantifier (e.g., oricine anyone, fiece every) Topic (e.g. dragostea the love) D-linked distributive quantifier (e.g., fiecare each)
The properties summed up in table (71) point toward a uniform analysis of verb-adjacent operators in terms of licensing conditions. We suggest that the major distinction involved in preverbal operators in Romanian can be related to the presence versus absence of feature-driven movement. Topic movement is not feature-driven, while verb-adjacent operator movement is.

Romanian). This is consistent with Richards account, which predicts that selectional features on functional heads will instruct PF to pronounce the upper copy. When the [+ focus] FF is present in the derivation (incorporated on I, as with the [+ wh] feature), it too will need an LI with which to establish a checking relationship (otherwise the derivation will crash and the utterance will be ungrammatical). The respective LI must share the [+ focus] feature (i.e., must match) in order for checking to occur. However, we have argued that movement to Spec,IP is also involved, irrespective of whether the focused constituent is pronounced preverbally or in its base position. Consequently, we assume that a non-trivial chain obtains with [+ focus] feature-checking, on a par with [+wh]-checking. Nevertheless, in contrast to wh-movement, when the [+ focus] FF is present in the derivation, we have seen there is a choice in pronouncing the upper or the lower copy in Romanian. Recall that according to Richards (1999), whenever a functional feature is strong, pre-Spell-Out movement is involved, a non-trivial chain is formed, and PF is instructed to pronounce the upper copy (i.e., the head of the chain). Given that with wh-movement in Romanian it is always the upper copies that need to be pronounced, while with focus-movement there is a choice, we suggest that while the [+wh] feature on I is selectional, the [+ focus] feature on I is non-selectional. Since the [+ focus] feature on I is non-selectional, PF will not be instructed as to which of the two copies to pronounce. In contrast to wh-phrases, which are inserted with a selectional [+ wh] feature directly from the lexicon, focused phrases are not inserted marked [+ focus] from the lexicon. We assume the [+ focus] feature on lexical items is acquired after lexical insertion, via phonology (hence the prosodic stress requirement which identifies an LI as contrastively focused). In other words, we propose that, while the [+ focus] feature on I is a formal feature (FF), the [+ focus] feature on the lexical item is a phonological feature (P-feature). The account proposed here views contrastive focus in Romanian as a representational property of phonosyntax, that is, the intersection between syntax and phonology (see also Bring 1997, Dchaine 1998 and references therein). The [+ focus] FF on I is checked against a lexical item bearing a [+ focus] P-feature in phonosyntax, as in (82). (82) phonosyntax Spell-Out
Given that a non-trivial chain is always formed with contrastive focus, we assume the [+ focus] feature on the lexical item to be selectional in nature. As opposed to constructions which involve a non-selectional FF (weak in Richards terms), in which, even though PF does not receive any instructions, there is a single suitable candidate to be pronounced (since there are no copies), with the [+ focus] FF there are two copies available to PF, but no instruction as to which of the two copies to pronounce. Given that the [+ focus] FF on I is non-selectional, and does not itself trigger Attraction, the syntactic component will send no instructions to PF as to which of the two copies to be pronounced. Since economy considerations do not apply at PF, for the purposes of PF it will not matter which copy is uttered. Notice, however, that in derivations with both [+ wh] and [+ focus] features, it will always be the lower copy of the contrastive element that is pronounced. Consider (83). (83) Ce i (*COPILULUI) a spus el COPILULUI ti whati (*child-the. DAT) AUX.3SG said he child-the.DAT ti (, nu vecinei)? (not, friend. DAT) 'What is it that it is to the child that he said (, not to the neighbour)?' Despite the impossibility of simultaneous pronunciation in the preverbal field, there is evidence that even in the presence of wh-phrases, focus raising still applies. Consider the examples in (84). (84) a. Ce i a spus mama luij copilului j ti whati AUX.3SG said mother his j child-the. DATj ti (, nu vecinei)? (not, friend. DAT) 'What did hisi mother say to the child i (, not the neighbour)?' * Cei a spus mama luij COPILULUI j ti whati AUX.3SG said mother hisj child-the. DATj ti (, nu vecinei)? (not, friend. DAT) '*What is it that hisi mother said to the child i (, not the neighbour)?'

S. Vikner, Studies in Comparative Germanic Syntax, volume I, 241-269, Dordrecht: Kluwer Academic Publishers. Reinholtz, C. 1989. V-2 in Mainland Scandinavian: Finite Verb Movement to Agr, in Working Papers in Scandinavian Syntax 44, 101-117. Reinholtz, C. 1995. Quantificational Variability and Operator Licensing, in P. Koskinen (ed), Proceedings of the 1995 Annual Conference of the Canadian Linguistic Association, 519-530, Toronto: Toronto Working Papers in Linguistics. Reinholtz, C. 1999. Quantifier Bound Pronouns in Danish, ms Queens University. Richards, N. 1997. What Moves where in Which Language?, Doctoral dissertation, MIT. Richards, N. 1998a. Shortest Moves to (Anti)-Superiority, in The Proceedings of the Sixteenth West Coast Conference of Formal Linguistics, E. Curtis, J. Lyle, & G. Webster (eds), 335-351, Stanford: CSLI Publications. Richards, N. 1998b. The Principle of Minimal Compliance, in Linguistic Inquiry 29: 599-629. Richards, N. 1999. Subject Extraction Without Subjects, paper presented at the LSA Summer Institute, University of Illinois. van Riemsdijk, H. 1998. Head Movement and Adjacency, in Natural language and Linguistic Theory 16, 633-678. Rivero, M.-L. 1989. Barriers and Rumanian, in C. Kirschner and J. Decesaris, Studies in Romance Linguistics, 289-313, Amsterdam: John Benjamins. Rivero, M.-L. 1994. The structure of the Clause and V-movement in the Languages of the Balkans, in Natural Language and Linguistic Theory 12: 63-120. Rivero, M.-L. 1997. On two locations for complement clitic pronouns: Serbo-Croatian, Bulgarian and Old Spanish, in A. van Kemenade and N. Vincent, Parameters of Morphosyntactic Change, 170-207, Cambridge University Press. Rizzi, L. 1982. Issues in Italian Syntax, Dordrecht: Foris. Rizzi, L. 1986a. Null Objects in Italian and the Theory of pro, in Linguistic Inquiry 17, 501-557. Rizzi, L. 1986b. On the Syntax of Subject Clitics in Romance, in O. Jaeggli and C. Silva-Corvaln (eds), Studies in Romance Linguistics, 391-421, Dordrecht: Foris. Rizzi, L. 1990. Relativized Minimality, Cambridge: MIT Press. Rizzi, L. 1991. Residual verb second and the WH Criterion, Ms. Universit de Genve. Rizzi, L. 1994. Argument/Adjunct (A)symmetries, in G. Cinque, J. Koster, J-Y. Pollock, L. Rizzi and R. Zanuttini (eds), Paths Towards Universal Grammar, 361-377, Washington: Georgetown University Press. Rizzi, L. 1995. The Fine Structure of the Left Periphery, ms. Universit de Genve. Published in 1997, in L. Haegeman (ed), Elements of Grammar: Handbook of Generative Syntax, 281-339, Dordercht: Kluwer Academic Publishers. Roberts, C. 1998. Focus, Flow of Information, and Universal Grammar, in P. Culicover and L. McNally (eds), The Limits of Syntax, 109-161, San Diego: Academic Press. Roberts, I. 1992. Verbs and Diachronc Syntax, Dordrecht: Kluwer Academic Publishers. Rochemont, M. 1986. Focus in Generative Grammar. Amsterdam: John Benjamins. Rochemont, M. and P. Culicover. 1990. English Focus Constructions and the Theory of Grammar, Cambridge University Press. Rochemont, M. 1998. Phonological Focus and Structural Focus, in P. Culicover and L. McNally (eds), The Limits of Syntax, 337-365, San Diego: Academic Press. Rohrbacher, B. W. 1999. Morphology-Driven Syntax: A Theory of V to I Raising and pro-Drop, Amsterdam: John Benjamins. Rooth, M. 1996. Focus, in S. Lappin (ed), The Handbook of Contemporary Semantic Theory, 271-297, Oxford: Blackwell. Ross, J. R. 1967. Constraints on Variables in Syntax, Doctoral dissertation, MIT.