<<

Clitics, Doubling and Resumption: An account of obviated WCO effects ∗∗∗

Giorgos Spathas

Abstract. This thesis examines the absence of WCO effects in clitic constructions in Greek and Romanian. It focuses on wh-interrogatives and QPs and argues that, although the mechanism that causes standard WCO effects is uniform, the obviation of the WCO effect is not. Doubled wh- operators in Romanian and doubled QPs in Greek, on the one hand, and doubled wh-operators in Greek, on the other, show systematic differences; the former, but not the latter, show island sensitivity, obviate WCO effects only in a very local domain and exhibit a specificity requirement. These differences are best captured if we see the obviation of WCO effects in the former case as a side effect of clitic doubling and the absence of WCO effects in the latter case as an instance of resumption. The proposal has consequences for an analysis on both clitic doubling and resumption and provides additional evidence in favour of a theory of in terms of logical representations.

1 Introduction

1.1 The Weak Crossover effect

Consider the in (1a-b).

(1) a. Who i t i loves his i mother ?

b. ?? Who i does his i mother love t i ?

In (1a) the anaphoric can be interpreted as a variable bound by who . The same is not possible in (1b), where who undergoes A’-movement to Spec, CP from the object position. The same contrast obtains with QPs like every boy in (2).

(2) a. Every boy i loves his i mother

b. ?? His i mother loves every boy i

∗ This MA thesis was completed in August 2005 at Utrecht Institute of Linguistics-OTS. It was written under the supervision of Eric Reuland. I would like to thank him for his support and encouragement. I would also like to thank Eddy Ruys and Elena Anagnostopoulou for helpful discussions, as well as all the friends and colleagues who helped me with their judgements. Note that after applying Raising to every boy as in (3) the positioning of traces and in (1) and (3) ends up identical.

(3) a. Every boy i t loves i his mother

b. ?? every boy i his i mother loves t i

The effect in (1b) and (3b) has been named the ‘Weak Crossover’ (WCO) effect; a quantified expression cannot bind a pronoun over which it crosses. The effect is ‘weak’ when compared with similar cases in which the intervening pronoun is in subject position and the ungrammaticality of the sentences is more severe. Those are cases of ‘Strong Crossover’ (SCO).

(4) a. * Who i does he i like t i ?

b. * He i likes every boy i

c. * every boy i he i likes t i

At a first glance, the examples can be described with regard to the position of the traces. In the grammatical examples ((1a) and (2a)) the trace occupies a position on the left side of the pronoun; in all ungrammatical examples the trace occupies a position on the right side of the pronoun. This generalization has been formulated as the Leftness Principle by Chomsky (1976).

(5) Leftness Principle A variable cannot be the antecedent of a pronoun to its left

The descriptive generalization in (5) fails, though, when we consider cases like the one in (6), which shows the same effect; the pronoun cannot be interpreted as a bound variable.

(6) a. * A party without every actress i annoys her i

b. every actress i a party without t annoys her i

Thus, the facts should be explained by the workings of a deeper syntactic generalization. Two main approaches have dominated the relevant literature; one that prohibits quantified expressions to locally A’-bind pronouns and one that requires pronouns coindexed with quantificational expressions to be locally A-bound. In fact, these two approaches correspond to two different approaches to binding. In the next section I will present their basics in turn.

2 1.2 The theories

1.2.1 Canonical binding theory

The canonical binding theory first formulated in Chomsky (1981) and adopted in the Government and Binding framework (and, subsequently in the Minimalist Program) seeks to derive the linguistic rules that regulate the possibilities of linguistic elements like R-Expressions, pronoun and reflexive anaphors. Thus, the central problem is to regulate the possible relations between those linguistic elements or between a linguistic element and a referent as they are (dis)allowed by the computational system of human language. These relations are relations of encoded in the computational system by the use of indices and are captured by the three Binding

Conditions.

(7) Conditions A: An anaphor is bound in its governing category B: A pronoun is free in its governing category C: An R-expression is free

Condition C determines that coreference between John and he will not be possible in (8a) but will be in (8b). Condition B determines that coreference will not be possible in (9a) but will be in (9b-c).

(8) a. * He i said that Mary likes John i

b. Most of her i friends adore Lucie i

(9) a. * John i likes him i

b. John i likes his i picture

c. John i said that he i likes Mary

The pronouns in all the sentences above can have a second option; instead of referring to the intra- sentential antecedent they can refer to an antecedent in the discourse (and then examples (8a) and (9a) are grammatical). But the reflexive anaphor in (10) does not have this option.

(10) John i likes himself i

The anaphor himself refers necessarily to John , its antecedent within the sentence; it cannot pick up its referent from the discourse. Thus, the relation between an anaphor and its antecedent is not a simple relation of coreference. It is a relation of binding. Condition A regulates then that (10) is grammatical but (11) is not.

(11) John i said that Mary likes himself i

3 Since the of himself is dependant on the reference of its binder it is interpreted as a variable. Pronouns can also enter binding relations and be interpreted as variables. When the antecedent of a pronoun is a quantified expression, the interpretation of the pronoun depends on its binder; the pronoun cannot simply pick up a referent from the discourse.

(12) Every woman i likes her i picture

But it still has to obey Condition B.

(13) * Every woman i likes her i

Binding relations are defined as follows.

(14) Definition of binding An element α is bound by an element β iff α and β are coindexed and β c-commands α

Empty categories are also interpreted as variables entering binding relations. That is the standard interpretation of traces of A’-movement, like wh-movement and Quantifier Raising.

(15) a. Who i does Mary love t i ? b. Mary likes every boy

c. Every boy i Mary likes t i

Empty categories that are interpreted as variables fall under the definition in (16).

(16) Definition of variable α is a variable iff α is an empty category and occupies an A-position and is case marked

Note now that the WCO cases cannot be handled by the Conditions of the binding theory.

(17) a. ?? Who i does his i mother love t i ?

b. ?? Every boy i his i mother loves t i

The traces satisfy the definition in (16) so they can be variables. The pronouns satisfy the definition of binding so they should also be able to be bound by the quantified expressions. Yet they cannot be bound. So, examples of WCO are considered to manifest the workings of an additional generalization that regulates the binding possibilities of pronouns with quantificational antecedents. The most standard generalization assumed in the canonical binding theory is the Bijection Principle (BP) (Koopman and Sportiche 1982).

4 (18) Bijection Principle There is a bijective correspondence between variables and A’ positions

The Bijection Principle prohibits an operator to bind more than one variable at the same time. In effect it poses a negative requirement on the binding possibilities of pronouns: pronouns are not allowed to be A’-bound (since an A’-moved operator will always have to bind its trace as well). Following Ruys (2004), I will call conditions on WCO that are formulated within the canonical binding theory and pose a negative requirement on WCO ‘configurational approaches to WCO’. 1 The SCO cases, on the other hand, do not need an additional generalization to be ruled out.

(19) a. * Who i does he i like t i ?

b. * He i likes every boy i

c. * every boy i he i likes t i

In the canonical binding theory traces left after A’-movement are, by definition, equivalent to R- Expression. Thus, they are subject to Condition C. Since the pronoun in (19a) and (19c) is in a position to A-bind the trace, a Condition C violation arises and the SCO cases can be ruled out without the need for an additional generalization.

1.2.2. Reinhart’s Conception of Binding

Reinhart’s Conception of Binding (RCB) proposes an alternative to the canonical binding theory. RCB was first formulated in Reinhart (1983). I present it here as stated in Grodzinsky and Reinhart (1993) but I will still refer to it as RCB. In the canonical binding theory, binding conditions are stated to regulate ‘coreference’, i.e. a relation between an anaphor and some referential antecedent. The binding conditions only fail to give the right result in the standard WCO cases, which are then considered to ‘reflect a peculiar property of quantification in natural language’ (Grodzinsky and Reinhart 1993:p.72) formulated as, e.g., the Bijection Principle. In RCB the opposite view is taken; binding rules determine the conditions allowing bound variable anaphora and it is coreference that is governed by a different module, namely by a pragmatic rule (Rule I). The departure from the GB framework implies that all anaphoric elements (bound variable pronouns, reflexive anaphors, wh-traces, NP-traces, PRO) fall under the same syntactic and semantic generalizations: in order to be interpretable they must be syntactically bound (i.e. c-commanded by a coindexed antecedent) and they are all interpreted as bound variables. Pronouns, being able to choose their antecedent freely from the discourse, are the exceptional case.

1 There is another influential configurational approach to WCO that I don’t discuss here, Safir’s (1984) Parallelism Constraint on Operator Binding (PCOB). PCOB allows operators to bind more than one variable only when they are of the same kind (both pronouns or both wh-traces). I will discuss a reformulation of this approach in Safir (1996) in section 4.4.4.

5 Their bound variable interpretation is still regulated by the binding rules, but their referential interpretation is regulated by a pragmatic rule, Rule I. 2 So, the relation between a pronoun and an antecedent can be either one of binding or of coreference, as indicated by the of (20).

(20) Only Alfred thinks he is a great cook a. only Alfred (λx (x thinks x is a great cook)) b. only Alfred (λx (x thinks he is a great cook)) and he=Alfred

(20) can either mean that Alfred is the only person that thinks that Alfred is a great cook (20b) or that Alfred is the only person that thinks of himself as a great cook (all others think of themselves as great linguists) (20a). (20) is important for another difference between RCB and the canonical binding theory. Whereas in the canonical binding theory pronouns were only allowed to have bound readings with quantified antecedents, in RCB their bound variable interpretation does not depend on the status of the antecedent; in (20) the pronoun is bound by a referential DP. Thus, the possibility of a pronoun to be a bound variable is independent of the of its antecedent. If this is true for bound pronouns and if all anaphoric elements obey the same conditions on their interpretation, as noted above, then, within RCB, it is not possible to pose any extra restriction on the semantic type of the antecedent of the empty categories we have discussed so far. Coreference, on the other hand, is sensitive to the type of the antecedent; the antecedent has to be referential. Quantifiers, not being able to form a discourse entity, are not available antecedents for referential pronouns. Thus, in the examples in (21), where the syntactic conditions on binding are not met and only coreference is possible, (21a) is grammatical, while (21b) is not.

(21) a. A party without Lucie i annoys her i

b. * A party without every actress i annoys her i

In short, in RCB the standard definition of binding is assumed, as in (22a). Binding conditions, on the other hand, since they only regulate variable binding and not coreference, do not include Condition C of the binding theory, as seen in (22b). The translation definition in (22c) is responsible for the interpretation of indices at LF.

(22) a. Definition An element α is bound by an element β iff α and β are coindexed and β c-commands α b. Conditions A: An anaphor is bound in its governing category B: A pronoun is free in its governing category

2 For more details concerning Rule I see section 4.1.

6 c. Translation definition An NP is a variable iff either i. it is empty and A’-bound, or ii. it is A-bound and lacks lexical content Other cases of NP coindexation are uninterpretable

The standard WCO and SCO examples, repeated here in (23) and (24), respectively, are excluded by the translation definition.

(23) a. * Who i does his i mother like t i ?

b. * His i mother likes every boy i

c. * every boy i his i mother likes t i

(24) a. * Who i does he i like t i ?

b. * He i likes every boy i

c. * every boy i he i likes t i

In (223a), ‘his’ (an NP that lacks lexical content) cannot be interpreted as a variable because it is not A-bound. The same holds for (23b), even after QR applies and derives the LF representation in (23c); the pronoun is only A’-bound but not A-bound by the operator and cannot be interpreted as a bound variable. SCO effects are excluded in the exact same way; the pronoun in subject position is not c- commanded by an antecedent in an A-position and cannot be translated into a variable. Note that the possibility of the translation definition to apply either before or after QR is a departure from Reinhart’s earlier formulations of the RCB, where it was stated explicitly that the translation definition had to apply at S-Structure.

1.3 The basic facts

In this section I present the basic facts I seek to explain. The basic WCO violation is present in Greek, as shown in (25a-b). 3

(25) a. * Pjon i ajapai i mitera tu i t i ? who loves the mother his ‘Who does his mother love ?’

3 The fact that WCO effects arise in Greek has been challenged in the literature. Indeed, speakers seem to form two groups, one that detects WCO effects in (25a) and (26a) (and includes myself) and one that it doesn’t (and whose judgement has initiated a long discussion about the configurational nature of Greek). Both groups are consistent with their judgements. I only deal with the first group in this thesis. Facts similar to the ones I present here can be found in Anagnostopoulou (2003) and Alexiadou and Anagnostopoulou (1997) for cases of QPs and in Hornstein (1995) for wh-questions.

7 b. Pjos i t i ajapai ti mitera tu i ? who loves the mother his ‘Who loves his mother ?’

The same is true for cases of WCO induced by QR.

(26) a. * Ajapai i mitera tu i to kathe pedhi i loves the mother his the every child ‘His mother loves every child’

b. Ajapai to kathe pedhi i ti mitera tu i loves the every child the mother his ‘Every child loves his mother’

In both cases the WCO effect is obviated when a clitic ‘doubles’ the wh-element (in (27a)) and the QP (in (27b)).

(27) a. Pjon i ton ajapai i mitera tu i ? who him loves the mother his ‘Who does his mother love ?’

b. To ajapai i mitera tu i to kathe pedhi i it loves the mother his the every child ‘His mother loves every child’

However, the distribution of the clitic is different in the two cases. Clitics are generally disallowed in Greek wh-questions.

(28) a. * Pjon ton ajapai i Maria ? who him loves the Mary ‘Who does Mary love ?’ b. * Pjon mathiti ton idhes ? which student him saw-you ‘Which student did you see ?’

Clitics can only be licensed when there is additional information (as in (29a)) or when the wh-element has an (overt) partitive reading (as in (30b)). Even then it is only optional.

(29) a. Pjon (ton) idhes ehthes na filai ti Maria ? who him saw-you yesterday subj kiss-he the Mary ‘Who did you see yesterday kissing Mary?’

8 b. Pjon apo tus fitites (ton) filise i Eleni ? who from the students him kissed the Helen ‘Which of the students did Helen kiss ?’

The optionality of the clitic in (29a-b) contrasts with its obligatory presence in (27a), where it is licensed by the presence of an intervening coindexed pronoun. In the case of QPs, on the other hand, it seems that the clitic is not licensed by the coindexed pronoun; QPs in Greek can be optionally ‘doubled’ by clitics, with the proviso that the QP is accompanied by the definite determiner to . QPs without to cannot be clitic doubled (30b) and show the standard WCO violation (30c).

(30) a. (To) ajapai i Maria to kathe pedhi (it) loves the Marythe every child ‘Mary loves every child’ b. (*To) ajapai i Maria kathe pedhi (it) loves the Mary every child ‘Mary loves every child’

c. * Ajapai i mitera tu i kathe pedhi i loves the mother his every child ‘His mother loves every child’

Similar facts have been observed in other languages. Romanian (Dobrovie-Sorin 1990) and Bulgarian (Alexandrova 1999) show similar effects with wh-questions, as in (31) and (32) respectively; only the variants with the clitic in (31b) and (32b) show no WCO effects.

(31) a. * Pe cine i a certat mama lui i t t ? Pe who has scolded mother his ‘Who did his mother scold ?’

b. Pe care i l-a certat mama lui i t i ? Pe which him-has scolded mother his ‘Which one did his mother scold ?’

(32) a. * Koe dete i obi a majka mu i t i ? which child loves the mother his ‘Which child does his mother love ?’

b. Koe dete i go obi a majka mu i t i ? which child him loves the mother his ‘Which child does his mother love ?’

The same holds with QPs in River Plate Spanish; the variant with the clitic in (33b) shows no WCO effects (Suner 1988).

9 (33) a. * Su i madre quiere a todos i their mother likes a everyone ‘Everyone is loved by his own mother’

b. Su i madre los quiere a todos i their mother them likes a everyone ‘Everyone is loved by his own mother’

This thesis will seek to explain the data above while evaluating the competing theories of binding I presented in the previous section. In the next chapter I discuss and dismiss theories that attribute the obviation of WCO effects to the non-quantificational nature of certain configurations. Chapter three presents an alternative that argues that the absence of WCO effects can be explained by assuming a specific analysis of clitic doubling. This option works for some of cases (Romanian care -questions and Greek QPs) but not for others (Greek wh-questions). The fourth chapter proposes that although the former can be seen as a side-effect of clitic doubling, the latter are better analyzed as cases of resumption. Both options are dealt within a logical syntax conception of binding. In chapter five the findings of the previous chapters are used to propose an answer to a long-standing question wrt Clitic Left Dislocation (CLLD) structures in Greek and Italian. Conclusions follow.

10 2 NonNon----quantificationalquantificational WCO

In this chapter I will present and evaluate three analyses that deal with obviated WCO effects and make the same basic assumptions; namely, they adopt (some version of) the canonical binding theory and they attribute the amnesty of specific configurations to the WCO effect to the non-quantificational nature of the antecedent. I start with Dobrovie-Sorin’s (1990) analysis of Romanian interrogatives, which show very similar properties with the Greek facts I presented in the previous section.

2.1 DobrovieDobrovie----SorinSorin 1990

Romanian exhibits two kinds of wh-elements: cine that corresponds to who , and care that corresponds to which person . Wh-questions formed with cine do not tolerate a clitic (34a-b) while wh-questions formed with care require it (35a-b).

(34) a. Pe cine ai vazut ? pe who have-you seen b. * Pe cine l-ai vazut ? pe who him-have-you seen ‘Who have you seen ?’

(35) a. * Pe care (baiat) ai vazut ? pe which (boy) have-you seen b. Pe care (baiat) l-ai vazut ? pe which (boy) him-have-you seen ‘Which (boy) have you seen ?’

Both constructions being sensitive to islands, Dobrovie-Sorin assumes a movement analysis of (34a) and (35b), namely movement of the wh-element out of a clitic doubling structure to the sentence initial A’-position, as shown in (36).

(36) wh i cl e i

In order to account for the different pattern of cine - and care -questions, she relies on the following generalization: ‘wh-variables cannot be doubled by accusative clitics’. This should be so following Chomsky’s (1981) definition of a variable (‘a is variable iff a is an empty category that (a) occupies an A-position, (b) is bound by a quantifier and (c) is case marked’) and the, at the time, widely held view that the clitic absorbs case. If the clitic absorbs case then the empty category in (34b), though it sits in an A-position and is bound by a quantifier, cannot count as a variable, since it bears no case. No problem arises in (34a), where no case-absorbing clitic is involved.

11 It follows that for care -questions to require the presence of a clitic (2b) care N should not be a true quantifier and that no quantification is involved in the formation of care -questions. In other words, the empty category in (35b) is not a variable. Dobrovie-Sorin provides the following LF representation for care -questions.

(37) [NPi for which x, x is a boy] you saw him i

In (37) the domain of quantification of care is limited to the NP it belongs and not to the whole clause. The empty category in argument position is interpreted as an empty category of a pronominal kind that is coindexed (coreferential) with the NP. So the NP that contains care does not function as a quantifier. That depends on the nature (meaning the semantics) of the wh-element itself and not to a syntactic property of the NP (e.g. the syntactic realization of a restriction). The obligatory presence of the clitic indicates that care N does not function as a quantifier. Dobrovie-Sorin argues that the D- linking requirement on care -questions, namely the fact that the set denoted by care N constitutes shared pragmatic knowledge (see Pesetsky 1987), follows from the LF representation in (37). In effect, it is (the semantics of) care that licenses the clitic and not the clitic that licenses a certain interpretation of the wh-element. According to Dobrovie-Sorin there is also independent evidence that care -questions do not involve quantification. Unlike the cine -question in (38a), the care -question in (38b) does not show any WCO effect.

(38) a. * Pe cine i a certat mama lui i e i ? pe who has scolded mother his ‘Who did his mother scold ?’

b. Pe care i l-a certat mama lui i e i ? pe which him-has scolded mother his ‘Which one did his mother scold ?’

The Bijection Principle prohibits an operator from binding two variables simultaneously. Since, there is no WCO effects in (38b) then it follows that the empty category is not a variable. But traces left after A’-movement of quantificational NPs are always variables. It follows, then, that for an empty category not to be a variable, no quantification should be involved. Under this reasoning, an empty category that does not show WCO effects is not a variable but an empty category of a different type. Dobrovie- Sorin argues that it is pronominal category. So in (38b) movement of the non-quantificational element care does not leave behind a variable but an empty category of a pronominal. Since care N behaves in all respects like a usual NP then the coindexing between care N , lui and the empty category is a simple case of coreference. The example in (38a), on the other hand, shows the standard WCO effect. The Bijection Principle easily rules out (38a); cine , a true operator, binds two variables simultaneously, lui and its trace.

12 It is true that under the assumptions made about the structure of doubled wh-questions the BJ cannot account for (38b), without additional assumptions about the nature of quantification in these structures. If care N was a true quantifier, the BJ would rule (38b) out; even under the assumption that care only binds the accusative clitic, then it would be the clitic that simultaneously binds the empty category and lui . Note, that RCB wouldn’t do better; no matter whether care N is base-generated or moved to the sentence initial A’-position, there is still no A-binder to license lui as a bound variable. But if a configurational approach to WCO is correct and if there is no quantification involved in care - questions, the facts follow. However, the analysis faces empirical problems. Consider the contrast in (39a-b).

(39) a. Pe care i spune Mary ca il iubeste mama lui i ? pe which one said Mary that him loves mother his ‘Who did Mary say that his mother loves him ? ’

b. * Pe care i spune mama lui i ca il iubeste Mary ? pe which one said mother his that him loves Mary ‘Who did his mother say that Mary loves him ?’

Although the WCO effect is obviated in (39a), where lui appears in the embedded clause, it is not in (39b), where lui appears in the matrix clause. If the obviation of the WCO effect is simply a matter of coreference between lui and care N , then the position of lui should be irrelevant. The contrast in (39a- b) clearly indicates that binding is involved. A second potential empirical problem comes from the investigation of cine -questions. Dobrovie-Sorin mentions that, according to an LI reviewer, the WCO violation in (38a) improves when an accusative clitic is added, as in (40).

(40) Pe cine i l-a certat mama lui i ? pe who him-has scolded mother his ‘Who did his mother scold ?’

Dobrovie-Sorin dismisses the example in (40) noting that the interpretation of cine changes when an accusative clitic is added; it shifts from cine ('who') to care (‘which one’). She argues that this does not invalidate the hypothesis according to which clitics are excluded from quantifier-variable configurations, because cine is reanalyzed as care in the presence of a clitic, and the absence of a WCO effect follows. But note that this is a reversal of her previous argument. Recall that it was the semantics of the wh-element that licensed the clitic in care -questions, not the clitic that altered the interpretation of the wh-element. If the example in (40) shows indeed an obviation of the WCO effect, then Dobrovie-Sorin’s analysis fails to explain the fact that it is the intervening pronoun that licenses the clitic. More problems appear when trying to implement the Greek data in Dobrovie-Sorin’s analysis. Greek has only one wh-element ( pjos ) that can appear with or without an overt restriction. There is no

13 specificity or D-linking requirement and neither can license a doubling clitic, as already seen in (28a- b). So there is already a first problem when trying to apply Dobrovie-Sorin's analysis in the Greek data, since no care-cine distinction is visible. Clitics are optional in Greek wh-questions only when there is either additional information or when the wh-element has a partitive reading, as we have seen in (29a) and (29b) respectively, repeated here as (41).

(41) a. Pjon (ton) idhes ehthes na filai ti Maria ? who him saw-you yesterday subj kiss the Mary ‘Who did you see yesterday kissing Mary?’ b. Pjon apo tus fitites (ton) filise I Eleni ? who from the students him kissed the Helen ‘Which of the students did Helen kiss ?’

So, there is a correlation between the semantics of the wh-element and the appearance of a doubling clitic; specificity in Greek can license a clitic. The question is whether this effects the quantificational force of the dependency between the wh-element and the empty category. I will consider this question in the last part of this section. Note, however, that the answer to the question is independent of the issue of the obviation of the WCO effect (at least in the case of Greek wh-questions). As shown in (26a-b), the clitic is obligatory in order to obtain a bound variable reading with an intervening pronoun, without this having any effect on the semantics of pjon , as is evident by the fact that there in no need for additional information or a partitive reading to license the clitic. In effect, Dobrovie-Sorin’s generalization that clitics cannot double wh-traces does not hold. The contrast found with Romanian care -questions in (39) is also found in Greek. In cases of WCO induced by QR, the bound variable reading is saved by the appearance of a clitic but only when the pronoun appears in the same clause. The WCO effect is present when the pronoun is embedded in the subject of the matrix clause. 4

(42) a. I Maria ipe oti to ajapai i mitera tu i to kathe pedhi i the Mary said that it loves the mother his the every child 'Mary said that his mother loves every child'

b. * I mitera tu i ipe oti to ajapai i Maria to kathe pedhi i the mother his said that it loves the Mary the every child 'His mother said that Mary loves every child'

What the data above indicate is that, at least in Greek, it is not possible to attribute the presence of the clitic to the semantics of the wh-element; pjos does not come from the lexicon with a special semantics

4 WCO effects obviated in Greek wh-questions by the presence of a clitic do not show this sensitivity. More on this point in section 3.3 and section 4.4 for an explanation.

14 that can license the clitic. The clitic is licensed only in case additional material has been added. This material seems indeed to correlate with D-linking or specificity. Both (41a) and (41b) require that the answer to the question belongs to a set of elements that has already been introduced in the discourse. The additional information in (41a) helps identifying this set; partitives on the other hand are necessarily specific (Enç 1991). But even then the clitic is only optional. The appearance of a coindexed intervening pronoun, though, forces the presence of the clitic. Dobrovie-Sorin fails to explain this correlation.

2.2 An AA’’’’----boundbound pro

A proposal very similar to Dobrovie-Sorin’s can be found in Cingue (1990). Cinque, who also assumes the basics of the canonical binding theory. He discusses cases of long wh-movement and parasitic gaps and argues that they actually don't involve movement but base-generated empty categories. He also observes that they circumvent the WCO effect, as shown in (43) for the parasitic gap case.

(43) Who i did you gossip about t i NO despite his i teacher having vouched for e i ?

In (43), the standard WCO configuration is visible and the Bijection Principle would rule it out; the wh- operator simultaneously binds his and the empty category. Cinque concludes that the empty category involved cannot be a variable, but, like Dobrovie-Sorin, an empty category of pronominal character, not subject to the WCO generalization (because it is bound by a Null Operator (NO) that is not a true quantifier). In fact, the empty category is a base-generated pro ('an empty resumptive pronominal') . Unlike Dobrovie-Sorin, though, Cinque argues that the relation between the empty category and its antecedent is not one of coreference, but one of binding. Extending Rizzi's (1986) analysis for the null subject in pro-drop languages, like Italian, he proposes that an object pro needs to be identified. Subject pro s get identified via rich morphological agreement. Cinque proposes that object pro s get identified via A'-binding. In this sense, object pro s constitute a new type of empty category; though they are of pronominal character, they allow A'-binding, like variables. In fact, Cinque argues, they cannot even tolerate A-binding. The argument goes as follows. Cinque's pro , being pronominal, is expected to fall under Principle B but not under Principle C. In effect, it is expected to circumvent SCO effects the same way it circumvents WCO effects. Yet, the SCO configuration in (44) renders the sentence ungrammatical.

(44) * Who i did they find t i hostile before he i realized they wanted to help e i ?

15 The difference between (44) and (43), the WCO case, is that in the former an A-binder intervenes between pro and its A'-binder. Cinque argues that the intervening A-binder prevents pro from being identified by its A'-binder. Thus, cases of SCO are reduced to a failure of pro to be identified. 5 Implementing Cinque's analysis to the Greek and Romanian data would require the acceptance of a base-generation analysis for the doubled wh-element. The clitic could then be considered to be an overt resumptive pronoun bound by care , in, e.g., the Romanian case. That could be problematic for Romanian where care -questions are sensitive to islands, but not for Greek. 'Doubled' wh-elements in Greek do not show island sensitivity, as can be seen for an adjunct island and a complex NP in (45) and (46), respectively.

(45) Pja efimeridha apokimithikes dhiavazontas tin ? which newspaper fell-asleep-you reading it 'Which newspaper did you fall asleep while reading ?' (46) Pjon akuses tin idhisi oti ton sinelavan ? who heard-you the news that him arrested-they 'Who did you hear the news that they arrested him ?'

Still, the approach inherits all of Dobrovie-Sorin’s problems. The clause-boundedness of the obviation of the WCO effect in (39) and (42) is still mysterious; The antecedent is high enough to bind the intervening pronoun, regardless of whether it is embedded in the subject of the embedded (39a/42a) or the matrix clause (39b/42b). The problems with the distribution of clitics remain as well; Cinque would have to say that specificity licenses the resumptive clitic. That is not true for both Greek wh- elements where it is the coindexed intervening pronoun that licenses the clitics (with no specificity requirement imposed on it) and Greek QPs like to kathe pedhi which can only optionally be doubled by a clitic.

2.3 A null epitepithethet

There is yet another analysis that accepts that the immunity of WCO effects that certain constructions exhibit should be attributed to the non-quantificational nature of the antecedent and the special status of the empty category it leaves after movement. Lasnik and Stowell (1991) discuss cases similar to the ones that are discussed in Cinque (1990) and observe that they are licit in WCO configurations. They group those cases as cases of ‘weakest crossover’. Tough -movement in (47a), too -movement in (47b), the parasitic gap construction in (47c) and topicalization in (47d) are all instances of weakest crossover.

5 Dobrovie-Sorin notes that Romanian care -questions show SCO effects. She derives SCO effects by assuming that care N undergoes reconstruction in the argument position where it is c-commanded by the pronoun in subject position. Since care N in her analysis behaves like a referential NP in all respects, it is also subject to Principle C. Hence, the presence of SCO effects.

16 (47) a. Who i t i will be easy for us NO i to get his i mother to talk to e i ?

b. This book i was too obscene to have its i author publicize e i

c. Who i did you stay with t i NO i before his i wife had spoken to e i ?

d. This book i I expect its i author to buy e i

Lasnik and Stowell assume the presence of a Null Operator (NO) that binds the intervening pronoun and the empty category in object position. This is the standard WCO configuration that should lead to a WCO violation. Yet, the sentences in (47) are grammatical. In order to explain (47a-d) Lasnik and Stowell draw heavily on the existence of the NO (or a referential DP in the case of topicalization). They assume that NOs and referential DPs differ from ‘true QPs’ (QPs and wh-elements) in that they involve quantification over a singleton set) Thus, leave behind a different kind of trace, a ‘null epithet’, that is the covert equivalent of overt epithets like the idiot , the bastard etc. Thus, [-a, -p] categories are distinguished in variables and null epithets. The former are subject to the WCO generalization, while the latter are not, thus the grammaticality of (47a-d). It is obvious that this approach is very similar to the one proposed in Cinque (1990); essentially they only differ in that the empty category in Lasnik and Stowell’s approach is the result of movement (so, it cannot be pro ) and on their account of SCO. In fact, Lasnik and Stowell argue against Cinque’s account on both conceptual and empirical basis. They argue that if the identification of Cinque’s pro is simply a case of A’-binding, then there is no reason why an intervening A-binder in the SCO cases should block A’-binding; the presence of an A-binder should be irrelevant. The second problem they note is that for Cinque cases of ‘long wh-movement’ should pattern exactly like parasitic gaps. But as can be seen below, the ‘long distance wh-movement’ case in (48) does show WCO effects.

(48) ?* Who does his mother wonder whether you like t ?

Lasnik and Stowell’s approach overcomes the problem with the SCO cases by assuming that the empty category is a null epithet. Overt epithets are subject to Principle C of the Binding theory, as can be seen in the contrast in (49). The empty category in (50) is a null epithet that is also subject to Principle C. The intervening pronoun in (50) c-commands the null epithet resulting to a Principle C violation.

(49) a. * Every criminal i praises the bastard i ’s lawyer

b. Every criminal i ’s wife praises the bastard i ’s lawyer

(50) * Who i did they find t i hostile before he i realized they wanted to help e i ?

A possible way to implement Lasnik and Stowell’s analysis in the Greek and Romanian data can be found in Cecchetto (2000). Cecchetto discusses cases of Clitic Left Dislocation in Italian as in

17 (51). CLLD is also discussed in Cinque (1990) who argues that no A'-movement is involved in CLLD, but that the left-dislocated NP in (51) is base-generated in the specifier of a TopicPhrase. Italian CLLD also shows no WCO effects, as can be seen in (52).

(51) Gianni, lo ama Maria John, him loves Mary ‘As for John, Mary loves him’

(52) Gianni i, lo ama sua i madre John, him loves his mother ‘As for John, his mother loves him’

Cecchetto (2000), based on reconstruction effects found in Italian CLLD structures, challenges Cinque's conclusion that the left-dislocated DP is base-generated. He proposes that CLLD structures are derivationally linked to clitic doubling structures. He assumes a variant of Torrego's (1992) analysis of clitic doubling according to which the clitic and the doubled DP initially form a 'big DP' in object position; the clitic is the head of the 'big DP' and the doubled DP sits in the specifier. The doubled DP undergoes A’-movement to the sentence initial position and the ‘big DP’ A-moves to the IP domain carrying along the trace of the doubled DP. In order to account for the absence of WCO effects Cecchetto further argues that the empty category left at the specifier position after movement of the doubled DP is not subject to the WCO generalization. He excludes all possibilities one by one (NP-trace, PRO, pro, and, most crucially, wh-trace) to conclude that the empty category is equivalent to an A'-bound pro . As we saw in Cinque's account, the empty category being a pro it needs to be identified. Cecchetto follows Rizzi's (1986) analysis that a pro can only be licensed in the presence of overt morphology and argues that the relevant morphology is the accusative clitic, that is, the trace of the left dislocated DP is only legitimated if the clitic head of the 'big DP' is overt. Since Italian CLLD is derived from a clitic doubling configuration, Cecchetto argues that the same analysis can be extended to account for the absence of WCO effects in clitic doubling structures in the Romanian and Rio Plate Spanish examples, we saw in section 1.3. Cecchetto provides no explicit implementation of how this should be done. Cecchetto's analysis relies crucially on the fact that it is the trace of the doubled DP that qualifies as an A'-bound pro and, as such, is not subject to WCO. Following Lasnik and Stowell he assumes that A’-bound pro s belong to the type of traces that cannot be interpreted as variables. Cecchetto does not state explicitly why traces of DPs that participate in a clitic doubling construction should be of this kind. But following a widely held assumption that clitic doubling structures encode specificity, one could say that it is the specific nature of the doubled DP that is crucial. The analysis then is straightforward for the WCO cases. Consider for example Romanian care -questions. care N sits in the specifier of the ‘big DP’ in argument position. It undergoes A’- movement to Spec, CP leaving behind a trace of the A’-bound pro / null epithet kind. The ‘big DP’ A- moves to the IP domain carrying along this trace. Since the trace is immune to the WCO effect no

18 ungrammaticality arises. The same analysis can be used for Greek doubled QPs, with the sole difference that movement of to kathe N is covert. 6 The truth is, however, that we don’t gain much by adopting this analysis. An analysis that relies on clitic doubling does provide a more meaningful answer to the issue of the distribution of clitics (see the next chapter for details), but it still inherits the main empirical problem of the clause- boundedness of the obviation of the WCO effect in (39) and (42); the antecedent in Spec, CP sits in a position from which it c-commands the intervening pronoun, regardless of whether this is embedded in the subject of the embedded or the matrix clause. Consider now the following conceptual problem that Lasnik and Stowell’s and Cecchetto’s analyses face. Although they both assume a configurational approach to WCO they do not adopt any specific theory of WCO. Lasnik and Stowell only provide the descriptive generalization in (53).

(53) In a configuration where a category C A’-binds a pronoun P and a trace T, P may not be contained in an argument phrase XP that c-commands T.

Ruys (2004) argues extensively that the exception of epithets from the WCO configuration cannot be incorporated to any configuration account of WCO. Let’s follow Ruys’ argument for the Bijection Principle, which we have discussed above. The Bijection Principle works because in a WCO configuration it does not distinguish between the trace of the operator and the A’-bound pronoun (recall that for the Bijection Principle A’-bound pronouns count for variables). We have already seen in (49) that overt epithets do count as variables. For Lasnik and Stowell’s account of null epithets to work, then, only null epithets should be exempted; overt epithets do cause WCO violations when it is not the trace but the anaphor that is an epithet, as can be seen in (54).

(54) a. * Who i does the bastard i ’s mother love t i ?

b. * The idiot i ’s father hates every boy i

So, the exemption should only regard the trace and not the anaphor. Taking ‘variable’ in the Bijection Principle to exclude epithets will allow the weakest crossover cases in (47), but it will also allow (54). The only alternative for Lasnik and Stowell, Ruys concludes, is to incorporate the exemption in their descriptive generalization, by adding, e.g., ‘unless T is an epithet’. But this is merely a stipulation that gives us no insight into the WCO phenomenon.

2.4 Quantification and specificity

Apart from the specifics of each analysis presented in this chapter they all share one main characteristic; since they all assume a configurational analysis to WCO that prohibits local A’-binding

6 Since this analysis involves movement it can fit better care -questions that show island sensitivity but not doubled wh-operators in Greek which are not sensitive to islands.

19 of variables under certain circumstances, they seek to explain the absence of WCO effects in certain configurations by denying the existence of variables. Since variables are elements of structures that involve quantification, then structures that obviate WCO effects should have no quantificational character. All analyses in this chapter ground this claim by appealing to the special semantic nature of the antecedent involved: non-quantificational wh-elements in Dobrovie-Sorin 1990, operators that quantify over singleton sets in Lasnik and Stowell 1991, specificity/ D-linking imposed by clitic doubling in Cecchetto 2000. We have already seen that the most meaningful way to implement an analysis like that in the cases where clitics seem to obviate WCO effects is to assume that it is the specificity/ D- linking requirement of doubled elements that is responsible for the alleged non-quantificational nature of the structures. In this section I will argue (in a rather informal way) against this view. There have been various attempts in the literature to define a notion of specificity or D-linking (Partee 1972, Fodor and Sag 1982, Pesetsky 1987, Enç 1991). Pesetsky’s (1987) definition of D- linking seems to be particularly relevant for the discussion here. According to Pestetsky, wh-phrases of the type which N are appropriate only when the individuals named in the answer come from a select set in the domain of discourse; this phenomenon is called D-linking and is exactly what Dobrovie-Sorin has argued to be the interpretation of care -questions in Romanian. The same is true for Greek wh-questions that can license clitics, repeated here in (55).

(55) a. Pjon (ton) idhes ehthes na filai ti Maria ? who him saw-you yesterday subj kiss-he the Mary ‘Who did you see yesterday kissing Mary?’ b. Pjon apo tus fitites (ton) filise i Eleni ? who from the students him kissed the Helen ‘Which of the students did Helen kiss ?’

The questions in (55) exhibit the exact same requirement, the answer to the question must come from a set that has already been established in the discourse. For example, the question in (55a) is felicitous (with a clitic) when the interlocutors are gossiping about what happened in the party they both attended the previous night; all guys that were present in the party are known to the interlocutors and form a set in the discourse. A felicitous answer can be ‘John’ or even a specific indefinite (‘enan pu foruse ena kokino pukamiso/ one that was wearing a red shirt’), but not a non-specific indefinite (#‘kathe tipo pu tis kernuse ena poto/ every guy that gave her a drink’). I have already mentioned a couple of times that no specificity/ D-linking requirement is imposed on wh-elements that obviate a WCO effect, as in (56).

(56) Pjon i *(ton) ajapane prajmatika ta pedhia tu i ? who him love really the children his ‘Who do his children really love ?’

20 ‘John’ or the guy with the red shirt are, again, good answers to these questions; but so is ‘kathe patera pu tus dhini pola lefta/ every father that gives them a lot of money’. There is no D-linking requirement in these cases. But even if it was one, like in the case of care -questions, it would be highly imlpausible to attribute the absence of WCO effects to specificity alone (under any of the analyses presented in this chapter). Specificity/ D-linking does not suffice to remove a WCO violation. Specific wh-phrases in English, for example, do show WCO effects (the example from Ruys 2000).

(57) * Which boy i does his i mother like t i ?

In any way, the specificity requirement posed on some wh-questions does not, in any sense that I can understand, affect the quantificational status of the dependency between the wh-element and its trace. A question like ‘Who did Mary see?’ clearly involves quantification and its semantics are usually translated into something like ‘who is the person x, such that Mary love x’. How does this translation change if the set over which the wh-operator ranges is known to the speaker’s? It doesn’t; the specificity requirement works on top of this translation (possibly on a pragmatic level) but it doesn’t affect the semantics in any relevant way. The same argumentation holds for Greek QPs like to kathe pedhi (‘the every child’). One might be tempted by the obligatory presence of the definite determiner to to assume that the quantifier is, in some way or another, ‘referential’. In fact, the definite determiner in this case does not behave like the definite determiner in cases of simple referential NPs. For example, definite NPs in Greek show a phenomenon called Determiner Spreading (see, e.g., Alexiadou and Wilder 1998) when modified by an adjective (58a); it is possible that the adjective and the noun both appear with their own definite determiner (58b). Determiner Spreading is restricted to definite determiners; adjectives modifying indefinite NPs (59a) cannot appear with their own indefinite determiner (59b).

(58) a. to kokino pukamiso the red shirt b. to kokino to pukamiso the red the shirt (59) a. ena kokino pukamiso a red shirt b. * ena kokino ena pukamiso a red a shirt

Phrases like to kathe pedhi are not subject to Determiner Spreading even though they come with the definite determiner (60a-b); they pattern with indefinite NPs.

(60) a. to kathe eksipno pedhi the every smart child

21 b. * to (kathe) eksipno to (kathe) pedhi the every smart the every child

It seems that the definite determiner in phrases like to kathe pedhi enforces a distributive reading on the quantifier. It resembles more the quantifier each than the quantifier every . For example, it takes wide easier than the corresponding kathe pedhi (‘every child’), that lacks the definite determiner.

(61) a. Kapjos kathijitis ekstase kathe mathiti kapjos > kathe , * kathe > kapjos some professor examined every student ‘Some professor examined every student’ b. Kapjos kathijitis ekstase ton kathe mathiti kapjos > kathe , kathe > kapjos some professor examined the every student ‘Some professor examined each student’

If to kathe pedhi is more like ‘each’ than like ‘every’ then it is already specific even without the clitic. Indeed, a sentence like (62) is strange when uttered out of the blue. But if we know that Mary that Mary is a teacher and we are talking about her behavior in class, then it is perfectly fine.

(62) I Maria ajapai to kathe pedhi the Mary loves the every child ‘Mary loves every/ each child’

It is obvious then that specificity cannot, by itself, obviate WCO effects, since both to kathe pedhi (63a) and each (63b) do show the WCO effect.

(63) a. * Ajapai i mitera tu i to kathe pedhi i loves the mother his the every child ‘His mother loves every/ each child’

b. * His i mother loves each child i

Only when a clitic is present does the WCO effect disappear.

(64) To ajapai i mitera tu i to kathe pedhi i it loves the mother his the every child ‘His mother loves every/ each child’

I conclude, then, that if we want account for the immunity to WCO effects in clitic constructions, we don’t need to go further than the clitic itself.

22 3 Clitic doubling

The basic problem we have seen so far is that under the analyses of doubled wh-elements we have seen so far there is no A-binder to license the bound variable reading of the intervening pronoun. An A-binder would immediately solve the problem posed by the obviation of the WCO effect in doubling structures (both interrogatives and cases of QR) without any need for additional stipulations. In fact, there are accounts of clitic doubling that do provide a potential A-binder and the idea that the answer to the obviation of the WCO effect may lie on the account of clitic doubling one assumes has been proposed in the literature. In this section I will demonstrate this idea using Sportiche's (1992) analysis of clitic doubling. Sportiche's analysis not only provides the missing A-binder, but has also something to say about the interpretational effects of clitic doubling.

3.1 Clitics as heads

Sportiche (1992) proposes an analysis of cliticization according to which clitics are functional heads licensing specificity on an argument with which they agree in φ-features. The clitic projects its own maximal projection located above VP and below TP/ IP. Licensing between the clitic and the argument is established in a spec-head configuration after (overt or covert) movement of the argument to the specifier of the CliticPhrase. The relevant configuration is given in (66).

(66) [ClP XP i [ Cl’ cl [ VP V t i ]]]

In clitic doubling structures the object undergoes covert movement to Spec, ClP. When no double is present then it is a pro that is generated in object position and undergoes movement to Spec, ClP, deriving the fact that in cases of simple cliticization the clitic behaves like a pronominal element. Sportiche further argues that the configuration in (66) allows a unification of the syntax of clitic doubling with that of scrambling in languages like German and Dutch. The only difference is that the movement of the argument is overt and that the clitic is not phonetically realized. So, both the XP and the clitic head may be phonetically realized or not. The Doubly Filled Voice Filter, that states that in a spec-head configuration where a head licenses some property P with an XP, it is not allowed to overtly encode both the XP and the head, regulates the PF representation of clitic doubling and scrambling structures. The crucial question for our discussion is whether the movement of XP to Spec, ClP is an instance of A- or A’-movement. Sportiche actually favors the latter. He presents some binding facts (failure of licensing anaphors and Principle C effects) concerning en -cliticization in French that point towards Spec, ClP being an A’-position. 7 He then proposes to extend this to all clitics ‘’because of uniformity’’ (Sportiche 1992: p. 46).

7 Sportiche presents more arguments for his claim, like agreement phenomena and the licensing of parasitic gaps. I do not discuss parasitic gaps in detail in this thesis, so I will only consider binding effects.

23 Binding effects in Greek, however, point towards the opposite direction. Consider the contrast between the SVO order in (67a) and the VSO order in (67b).

(67) a. I mitera tis Sofias i tin i ajapai poli the mother the Sophia_gen her loves a lot ‘Sophia’s mother lover her a lot’

b. * Tin i ajapai i mitera tis Sofias i poli ehr loves the mother the Sophia_gen a lot ‘Sophia’s mother loves her a lot’

The contrast in (67) indicates an asymmetry between the SVO and VSO order in Greek; coreference is possible in the former but not in the latter. Assuming the independently motivated claim that subjects in VSO orders in pro-drop languages stay in their VP internal position (see, e.g., Alexiadou and Anagnostopoulou 1998), the examples in (67a-b) correspond to the LF representations in (68a-b).

(68) a. [ TP i mitera tis Sofias i [ T’ ajapai j [ ClP pro i [ Cl’ tin [ VP t j t i ]]]]

b. [ TP ajapai j [ ClP pro i [ Cl’ tin [ VP i mitera tis Sofias i [ V’ t j t i ]]]]

In (68a) there is no c-command between Sofias and pro and coreference is possible regardless of where pro is interpreted. (68b) is more revealing. If Spec, ClP were an A’-position coreference should be possible. But if it is an A-position a Principle C violation arises and the ungrammaticality of (67b) follows. The same contrast arises in cases of variable binding. Consider the examples in (69) and their corresponding LF representations in (70).

(69) a. Kathe jineka i ton ajapai ton antra tis i every woman him loves the husband hers ‘Every woman loves her husband’

b. * Ton ajapai kathe jineka i ton antra tis i him loves every woman the husband hers ‘Every woman loves her husband’

(70) a. [ TP kathe jineka [ T’ ajapai j [ ClP ton antra tis [ Cl’ ton [ VP t j t i ]]]]

b. [ TP ajapai j [ ClP ton antra tis [ Cl’ ton [ VP kathe jineka i [ V’ t j t i ]]]]

The option of A’-movement would fail again. If movement to Spec, ClP were A’-movement, then it should be able to undergo reconstruction, a well documented property of A’-chains (see, e.g. Sportiche 2003). Reconstruction to the object position would put the pronoun in the scope of the operator and binding should be possible. A-movement, on the other hand, shows no reconstruction effects; the moved element necessarily gets interpreted in the target position. Thus, A-movement of ton antra tis to Spec, ClP explains the ungrammaticality of (b) since the NP is interpreted in a position

24 where it is not c-commanded by the operator and binding of the pronoun is impossible. I conclude that Spec, ClP is an A-position. 8

3.2 Advantages

3.2.1 WCO and Specificity

If clitic doubling licenses specificity it follows that only specific NPs can participate in clitic doubling constructions. This explains straightforwardly the correlation of clitics and specificity. Recall the Greek facts regarding optional clitics with specific wh-elements. When additional information is added or when the wh-element is necessarily specific because it has a partitive reading, the wh-element is able to check the specificity feature of the clitic head and, thus, can be clitic doubled. The same holds for to kathe N ('the every N'); it is necessarily specific, so it can participate in a clitic doubling structure. In order to explain the obligatory presence of the clitic in Romanian care -questions we will have to assume (with Dobrovie-Sorin) that the of care force it to be necessarily specific and that this specificity necessarily needs to be checked by a clitic head, but (unlike Dobrovie-Sorin) without this having any effect on care 's quantificational force. 9 Recall that Dobrovie-Sorin's main argument for this latter claim was the absence of WCO effects. The absence of WCO effects can now be considered to be a side-effect of clitic doubling; the doubled wh-element undergoes a first step of A-movement to Spec, ClP before moving further to Spec,CP. Under the standard analysis that the subject in inverted orders in pro-drop languages is located within the VP, the wh-element (or its trace) at Spec, ClP c-commands and binds the pronoun. The relevant structure for (38b), repeated here as (71a) would be as (71b).

(71) a. Pe care i l-a certat mama lui i ? pe who him-has scolded mother his ‘Which one did his mother scold ?’

b. [ CP pe care i [ TP certat j [ ClP t i [ Cl’ cl [ VP mama lui [ V' t j t i ]]]]]]

In a similar way the LF for (72a) after QR has applied, would be as in (72b).

8 The ‘uniformity’ of clitics Sportiche appeals to has been challenged in the literature. The issue is to complicated to even start presenting it here. I refer to Anagnostopoulou (to appear) for an overview. 9 Sportiche, not accepting that Spec, ClP is an A-position, explains the obviation of the WCO effect in Romanian (and similar Spanish data) appealing to the specificity requirement induced by clitic doubling. Sportiche and Dobrovie-Sorin claim that specificity has two effects: it licenses the appearance of a clitic and shows no WCO. The two effects are not connected to each other. My claim is that they are; specificity can license a clitic (clitic doubling) and, when it does, it is the specific configuration of clitic doubling that obviates WCO effects.

25 (72) a. To ajapai i mitera tu to kathe pedhi it loves the mother his the every child ‘His mother loves every boy’

b. [ CP to kathe pedhi i [ TP ajapai j [ ClP t i [ Cl’ cl [ VP i mitera tu [ V' t j t i ]]]]]]

In both cases the trace of the operator lies in an A-position where it c-commands and binds the pronoun in the VP internal subject position. Note that the grammaticality of the LFs in (71b) and (72b) is predicted by both configurational and licensing approaches to WCO. Their grammaticality is explained by whatever explains the grammaticality of (73).

(73) Who i t i loves his i mother ?

In a configurational approach, the Bijection Principle is not violated because the wh-operator only binds its trace and it is the trace that binds the pronoun. Similarly, in e.g. (72b), the operator only binds its trace in Spec, ClP and it is the trace that binds the pronoun. Under a licensing approach, the trace of the operator at Spec, ClP counts as an A-binder for the pronoun and, following the translation procedure, the pronoun can be interpreted as a bound variable. 10

3.2.2 Scrambling

Sportiche's analysis, unifying the syntax of clitic doubling and scrambling, makes a prediction; scrambled operators should also show no WCO effects. Facts confirming this prediction have been presented in a number of studies. I will present here some German data from Hiterhoelzl andPili (2003). 11 The basic contrast is the one between (74) and (75) where the object has undergone scrambling. The bound variable reading is unacceptable in (74), but acceptable in (75).

10 Following RCB's earlier proposal that the interpretational procedure should apply at S-Structure, Sportiche's analysis wouldn't work. Anagnostopoulou (2003) proposes an analysis that could work in this case. Anagnostopoulou discusses the obviation of WCO in cases of clitic doubled quantifiers. She proposes that the clitic in clitic doubling structures is the spell-out of the quantifiers D-feature. Under the assumption that it is the D- feature of a quantifier that is responsible for binding, the clitic c-commands and binds the pronoun in the VP internal position. 11 Alexiadou and Anagnostopoulou (1997) argue, as well, for a unification of the syntax of scrambling and clitic doubling. With regard to the absence of WCO they give similar facts from double object constructions in Greek and German. Note that Alexiadou and Anagnostopoulou do not adopt Sportiche's analysis, but argue that both scrambling and clitic doubling involve movement to an ObjectAgreement Projection. What is crucial is that this movement counts as A-movement. Hiterhoelzl and Pili assume an analysis that also makes use of CliticPhrases, without any explicit reference to Sportiche (1992).

26 (74) *...weil seine i Mutter jeden i liebt ...because his mother everyone loves

(75) ...weil jeden i seine i Mutter liebt ...because everyone his mother loves

Again, the facts follow if jeden moves to an A-position, the specifier of a phrase projected by a null clitic. 12 Hinterhoelzl and Pili argue that scrambling, if seen as A-movement, can also explain the well known fact that wh-movement in German does not show the WCO effect, as can be seen in (76).

(76) Wen i liebt seine i Mutter nicht ? who loves his mother not

(76) can be explained if the wh-element undergoes a short step of A-movement before raising to Spec, CP, just like in the care -question in (71b). Consider the following contrast.

(77) Wen i glaubt Peter dass seine i Mutter nicht liebt? who think you that his mother love

(78) * Wen i glaubt seine i Mutter dass Peter liebt ? who think his mother that Mary loves

The contrast shows that it is not a peculiar property of wh-movement in German that obviates the WCO effect. There are cases, like (78), where the effect is visible. What makes the difference is the position of the DP containing the intervening pronoun. In (77) it lies in a position where the pronoun can be licensed as a bound variable if wen undergoes scrambling before A'-movement. In (78) it isn't; after scrambling the trace of the wh-element is in the embedded clause where it cannot license the bound pronoun in the matrix clause. The contrast follows simply from the fact that scrambling (like clitic doubling) is clause-bound.

3.3 Disadvantages

So far, applying an analysis of clitic doubling a la Sportiche to the Greek and Romanian data has provided us with a clearer picture on the distribution of clitics in wh-questions and strong quantifiers and a straightforward explanation of the obviation of the WCO effect in clitic doubling constructions without any additional stipulations about the nature of quantifiers and the binding possibilities of empty categories. Moreover, it makes a correlation with similar facts in German and other languages exhibiting scrambling. Yet, there are failings.

12 I refer to Hinterhoelzl and Pili for more arguments in favor of an analysis of German scrambling in terms of A- movement.

27 In the analysis presented here the obviation of the WCO effect is just a by-product of clitic doubling. That might fare well with the obviation of WCO induced by QR in Greek and with Romanian care -questions, but not with Greek wh-questions. We have seen there that it is the other way around; it is the presence of an intervening pronoun that triggers the appearance of a clitic in order for the bound variable reading to be obtained. Simply accepting a Sportiche like analysis of clitic doubling leaves this issue unexplained. Moreover, treating doubled wh-elements in Greek on a par with Romanian care -questions would leave unexplained the fact that the latter, but not the former, are sensitive to islands. Recall now that care -questions, doubled QPs in Greek and scrambled QPs in German all obviated the WCO effects only when the pronoun occupied a position within the domain of the clitic. In the case of Greek doubled wh-elements, on the other hand, WCO effects are obviated regardless of the position of the intervening pronoun. In (79b) the pronoun is embedded within the subject of the main clause and in (79c) it is in the position of the indirect object of the verb of saying. 13

(79) a. Pjon i ipe i Maria oti ton i ajapai i mitera tu i ? who said the Mary that him loves the mother his 'Who did Mary say that his mother loves him ?'

b. Pjon i ipe i mitera tu i oti prepi na ton i kalesume sto parti ? who said the mother his that have-to him we-invite to-the party 'Who did his mother say that we should invite him to the party ?'

c. Pjon i tu i ipes oti tha ton i kalesume sto parti ? who him_gen told-you that will him we-invite to-the party 'Who did you tell him that we will invite him to the party ?'

If clitic doubling is responsible for the obviation of the WCO effect in Greek wh-questions, then the grammaticality of (79b-c) is unexpected. The contrast between these examples and the ones above is

13 The equivalents with no clitic are, as expected, ungrammatical with the intended reading, as can be seen in (i- iii). i. * Pjon ipe i Maria oti ajapai i mitera tu ? who_acc said the Mary that loves the mother his 'Who did Mary say that his mother loves ?' ii. * Pjon ipe i mitera tu oti prepi na kalesume sto parti ? who_acc said the mother his that have-to we-invite to-the party 'Who did his mother say that we should invite to the party ?' iii. * Pjon tu ipes oti tha kalesume sto parti ? who_acc him_gen that will we-invite to-the party 'Who did you tell him that we will invite to the party ?'

28 clear and significant. I will discuss it in more detail in the next chapter. For now it suffices to show that the clitic doubling analysis cannot be the whole story for Greek wh-questions. Another empirical problem arises wrt SCO effects; the analysis, as proposed here, predicts an obviation of the SCO effect, contrary to facts. That is the same for all three cases we have been examining (Romanian care-questions, doubled quantifiers and wh-elements in Greek) and regardless of the binding theory one accepts. The facts are repeated in (80) and (81).

(80) * Pjon i ton kalese pro i sto parti ? who him invited-he to-the party 'Who did he invite to the party ?'

(81) * To ajapai pro i to kathe pedhi i it loves-he the every child ‘He loves every child’

In the canonical binding theory SCO is excluded because of the Condition C restriction on wh-traces. If we accept that subjects (including pros) are located in the VP internal position and that clitic doubling involves a first step of A-movement of the doubled phrase in a position that c-commands the subject, then the wh-trace left after movement to the sentence initial position (to Spec, CP in wh- questions or after QR in the case of the strong quantifier) is not c-commanded by pro , as can be seen in the LF representations in (82). Note that it is the trace at Spec, ClP that counts since this is the one interpreted as a variable bound by the operator.

(82) a. [ CP pjon i [ TP ajapai j [ ClP t i [ Cl’ cl [ VP pro [ V' t j t i ]]]]]]

b. [ CP to kathe pedhi i [ TP ajapai j [ ClP t [cl [ VP pro [ V' t j t i ]]]]]]

In RCB, SCO is handled like WCO; both manifest the failure of a pronoun to be licensed as a bound variable from a c-commanding antecedent in an A-position Recall that in RCB, Principle C, being a condition on coreference, is not part of the binding theory. Thus, SCO violations cannot be ruled out in the same way as in the canonical binding theory, where wh-traces where subject to Principle C. Note that even if Principle C was assumed in this framework it would be a theoretically undesirable stipulation to attribute SCO effects to a difference between wh-traces and bound variable pronouns; in RCB they are both translated into variables under similar syntactic and semantic conditions and no difference between them is to be expected. In effect, SCO effects in RCB are ruled out in the same way WCO effects are; the intervening pronoun (in a subject position in the case of SCO) is not A- bound at any level of representation and fails to be translated into a variable. Therefore, we expect that every time WCO is not visible, no SCO effect should arise as well. But if the trace of the quantifier

29 at Spec, ClP can license the bound variable reading in the WCO case, it should also be able to license a bound variable reading for pro in (82). Yet, it doesn't. 14

3.4 The theories

3.4.1 SimilaritiSimilaritieses

It is obvious then that the analysis presented in this section cannot distinguish between configurational and licensing approaches to clitic doubling (and, in effect, between the canonical binding theory and RCB). If one assumes that the obviation of the WCO effect in the cases we have seen so far is a consequence of the clitic doubling configuration, then both approaches predict the grammaticality of the examples. Similarly, they both face the same empirical faults, since both wrongly predict an obviation of the SCO effect that never happens. RCB and the canonical binding theory also share a conceptual problem; they all make explicit use of indices. That is so because both assume a syntactic notion of binding formulated as identity of variables (indices). Binding, in this respect, manifests a relation between arguments, where one occurrence of a variable binds another, under certain syntactic conditions. Defining these syntactic conditions has been a problem for such approaches, the SCO and WCO cases, I am discussing here, being the most persistent. We have already seen that accounting for both SCO and WCO, either needs additional stipulations (in configurational approaches) or does not distinguish between the two (in RCB). Following Reinhart (2000), I will demonstrate this problem again for the SCO case. Consider the contrast in (83).

(83) a. Who t said we should invite him ? b. * Who did he say we should invite t ?

If binding is the relation between indices, or variables, then it is hard to distinguish between (83a) and (83b), where the relation between the trace and the pronoun (both translated into variables) is identical. In order to do so, a syntactic restriction was assumed, a syntactic restriction posed on wh- traces; wh-traces, unlike pronouns, are subject to Principle C. But we have already seen when presenting RCB that a distinction between the two is not feasible; pronouns and wh-traces pattern alike in all syntactic and semantic respects. But RCB didn’t provide a solution either treating SCO on a par with WCO.

14 Of course, there is no independent reason for pro not to be able to be interpreted as a bound variable. It clearly does so in, e.g., (i).

i. Kathe mathitis i pistevi oti ine pro i o kaliteros stin taksi every student believes that is-he the best in-the class 'Every student believes that he is the best in class'

30 Moreover, within the Minimalist Program (Chomsky 1995) indices cannot even be stipulated to be part of the computational system. In MP, the syntactic component operates exclusively on morpho- lexical elements selected from the lexicon. This requirement has been formulated as the Inclusiveness Condition. The Inclusiveness Condition dictates that the input vocabulary to the computational system of the human language is identical to the output vocabulary. So, no additional devices, like indices, can be added during the derivation. That implies that the interface levels, being part of the computational system, contain nothing more than lexical features. In effect, a syntactic definition of binding formulated as regulating the relations between variables (indices) is not conceivable within the Minimalist Program and binding theory should apply at the interface level with the interpretive component. But even at this level, indices should be expelled from binding theory.

3.4.1 And differences

It is not true, however, that licensing and configurational approaches behave similarly in all the cases I have presented so far. In what follows I will present Ruys’ (2004) argument that the weakest crossover cases of Lasnik and Stowell (1991) can be handled by a licensing approach to WCO without any need for additional stipulations. Ruys observes that configurational and licensing approaches end up with identical empirical results, since at the majority of cases, a ban on A’-binding equals a failure of A-binding. He then argues that the weakest crossover cases in Lasnik and Stowell provide the relevant construction where the two approaches actually make different prediction. Consider again example (84).

(84) Who did you stay with t NO before his wife had spoken to e ?

In (84) there are two available antecedents for the pronoun, the NO, assumed by Lasnik and Stowell, in an A’-position and the trace of who in an A-position. A configurational approach, e.g. the Bijection Principle, predicts the sentence to be ungrammatical (since the NO binds the pronoun and the empty category at the same time); a licensing approach, on the other hand, requires a pronoun to be bound by some operator in an A-position. Such an operator is available in (84) ( who and its trace) and predicts the sentence to be grammatical. 15 In RCB’s terms, both the empty category (which can be thought to be a pronominal category that lacks lexical content) and the pronoun are translated into variables, bound by the trace of who . Thus, in cases where an ‘inappropriate’ and an ‘appropriate’ antecedent compete, the latter wins. Ruys concludes that the additional assumptions made by Lasnik

15 The fact that the trace of who can license the trace as a variable is supported by identical examples with pronouns in the place of the empty category. These examples are also grammatical. Consider, e.g., the equivalent of (84) in (i).

i. Who did you stay with before his i wife had spoken to him i ? If the trace of who can license a bound variable reading of him in (i), there is no reason why it shouldn’t be able to do the same for the empty category in (84).

31 and Stowell about the nature of NOs and referential antecedents and the empty categories they leave behind after movement, are linked to their particular choice of WCO theory, the configurational approach. Licensing approaches explain the weakest crossover cases with no need for additional stipulations and are, thus, to be preferred. In what follows I will adopt this conclusion, but I will not adopt RCB as it has been presented here. In the next section I will present Reinhart’s latest modification of RCB (Reinhart 2000), an approach that solves the conceptual problems of the use of indices and that, I will argue, allows a more promising interpretation of the facts I have been discussing and of the empirical problems that are still unsolved, if some modifications are to accepted.

32 4 Binding in logicallogical----syntaxsyntax

In this section I will present Reinhart's latest formulation of RCB as presented in Reinhart (2000). As it stands, it takes care of the conceptual problems previous theories of binding and, accepting Reuland's (1998) modification, helps address the empirical problems I have been discussing so far. I will argue that although WCO effects are induced by the same mechanisms in cases of wh-movement and QPs, the obviation of the WCO effect will be attributed to two different mechanisms: clitic-doubling in Romanian care -questions and Greek QPs, and resumption in Greek wh-questions.

4.1 Reinhart 2000

Reinhart (2000) maintains RCB’s basic assumptions; principles of the binding theory should capture the mechanisms of variable binding and not of coreference; bound pronouns and all empty categories are translated into variables and interpreted as such, licensed in similar syntactic and semantic environments; coreference is not governed by the computational system, but by a pragmatic rule, Rule I. The basic innovation that Reinhart proposes (basically the return to an idea first proposed in Reinhart (1983)) is that binding should not be understood as a relation between indices (variables), but should be captured in terms of the traditional logical syntax conception of binding. In that respect, variables are only bound by operators; the relation between two variables, say α and β, is now mediated via operator binding. The definition of binding is thus reformulated as in (85).

(85) A-Binding (logical syntax definition): α binds β iff α is the sister of a λ- whose operator binds β

The definition in (85) immediately locates binding outside syntax proper and into a level where syntactic derivations are translated into logical-syntax forms. For concreteness, I will follow Reuland (1998) in assuming that this level is the Conceptional-Intensional interface. In that level, indices being additional non morpho-syntactic objects are not only disallowed but are also superfluous. After the formation of the λ-predicate empty categories are translated into variables; pronouns have an option, they can either be bound by the operator or stay open and wait for discourse to provide a relevant antecedent. In effect, being in the c-command domain of a l-operator is enough to license an empty category or a pronoun as a bound variable. The binding relation between the variable and the argument of the l-predicate follows directly with no need for indices. Let me demonstrate how binding in logical syntax works using the standard ambiguous example in (86a), where both binding and coreference are possible. After the λ-predicate is formed (following the standard assumptions that subjects are the arguments of λ-predicates), as in (86b), there are two possibilities for the pronoun; it can either be bound by Lucie , as in (86c), or be covalued (coreferent) with Lucie , as in (86d).

33 (86) a. Only Lucie respects her husband b. only Lucie (λx (x respects y's husband)) c. only Lucie (λx (x respects x's husband)) d. only Lucie (λx (x respects y's husband) & y=Lucie)

(86c) and (86d) correspond to the two meanings; (86c) implies that all other women in the world do not respect their husbands and (86d) that no other woman, but Lucie, respects Lucie’s husband. The question now is what regulates binding and coreference. Reinhart proposes that coreference is regulated by Rule I. 16

(87) Rule I : and β cannot be covalued (coreferent) in a derivation D, if a. α is in a configuration to A-bind β, and b. α cannot A-bind β in D, and c. the covaluation interpretation is indistinguishable from what would be obtained if α A-binds β

If we apply Rule I to (86) we see that the first clause is met; Lucie is in a configuration to bind her . So the second clause needs to be checked; Lucie can bind her (there is no independent reason to exclude binding, like e.g. a Principle B violation). So the third clause needs to be checked; the interpretations obtained by binding and coreference are, indeed, distinguishable, so coreference is possible in (86). What the formulation of Rule I implies is that whenever binding is possible, coreference will also be possible (if it leads to a different interpretation) and that, every time binding is not possible, coreference is not possible as well, when it competes with binding. In what follows I will not consider the details of Rule I. I will assume that whenever binding is not possible, coreference is not possible as well, in all the cases where the two compete. 17 Consider now how Reinhart’s system excludes SCO configurations. Consider again the example in (88a) and its logical syntax representation in (88b).

(88) a. * Who i did he i say we should invite t i ? b. who (λx (he said we should invite x))

16 Reinhart views Rule I as an interface strategy that regulates the workings of different modules. 17 Reinhart argues that coreference/ covaluation is not restricted to referential antecedents but extends to quantified antecedents, as well. There still is a difference, though. Pronouns cannot be directly covalued with a quantified antecedent, because, unlike a referential antecedent, it does not form a discourse entity. Covaluation is only indirect, between a pronoun and a variable bound by the quantified antecedent. Note that this innovation in Reinhart’s conception of binding does not disturb my criticism in section 2.1 against Dobrovie-Sorin’s analysis that required coreference between care N and the pronoun. If I am correct that D-linked QPs do form a set, then coreference between the pronoun and the operator should be direct and direct coreference is still not allowed.

34

The representation in (88b) indicates that movement of who creates a λ-predicate under which who binds x. But the subject pronoun he is also the argument of a λ-predicate with x in its c-command domain, as in (89).

(89) who (λx (he (λy said we should invite x)))

The question now is what prevents binding of x by he , as in (90).

(90) who (λx (he (λy said we should invite y)))

The answer is that, given the definition of binding in (85), binding is sensitive to standard laws of the relation between operators and variables in logical syntax; in this case, that only free variables can be bound by a given operator. Since x in (89) is already bound by who , it cannot be bound again by he (and since binding is not possible, covaluation is not possible as well). Note, however, that the system, as it stands, does not exclude WCO configurations. Consider (91a).

(91) a. * Who i does his i mother love t i ? b. who (λx (y’s mother loves x))) c. who (λx (x’s mother loves x)))

After wh-movement, the λ-predicate in (91b) is formed. There is no independent reason to exclude binding of y by who , as in (91c); the pronoun is free and there is no Principle B violation. The same applies to cases of WCO induced by QR. After QRing every boy , nothing prevents it from binding y.

(92) a. * His i mother loves every boy i b. every boy (λx (y’s mother loves x))) c. every boy (λx (x’s mother loves x)))

Reinhart assumes that WCO is to be handled by some additional generalization. Reuland (1998), on the other hand, proposes that a more principled explanation might be available.

4.2 Reuland 1998

Reuland follows Reinhart’s (2000) basic assumption that binding is to be captured in terms of logical syntax. He further attempts to derive the notion of c-command by independent procedures of interpretation. The specifics of Reuland’s analysis will not concern me here. What is of particular

35 interest for this discussion is his observation that the elimination of indices from syntax results in a reinterpretation of chains. Since chains are no longer defined in terms of indices, they are interpreted in terms of the operation Move. The copy theory of movement assumes that chain members are copies of each other, i.e. they constitute one object distributed over a number of syntactic positions. If this is true, then we can no longer define binding relations in terms of the syntactic positions that the relevant elements occupy in X’-structure at the relevant level of representation. Either of the positions linked be a chain may be the relevant one. The issue now is to define what part of the element is interpreted in what position. The question has initiated a long discussion that is still a hot issue in current generative linguistics. Chomsky (1995) and Fox (1998) propose that in cases of wh-movement and QR it is only the quantificational part of the expression that is interpreted in the moved position, the rest (crucially including φ-features) is interpreted in the argument position. Reuland makes a very similar suggestion. He proposes that the structure of (93a) is (93b).

(93) a. Who does his mother love t ? b. Q [ y’s mother loves (who (λx (x))]

That is, who and the variable it binds form one single object interpreted in the argument position. If who moves, it does so to satisfy the checking requirement of some feature in Q. The interpretive effects associated with this movement are due to the features of Q, not to the features of who . 18 If this is true for wh-movement it should be true for any kind of A’-movement, like the QR case in (94).

(94) a. His mother loves every boy b. [ y’s mother loves (everyone (λx (x))]

A’-movement satisfies the requirements of the target position and it carries along only those features that are needed in that position. Hence the possibility of reconstruction in cases of A’-movement. So, who in (93) and every in (94) are interpreted in the argument position for all relevant purposes; in that position they bind x, but they cannot bind y. In other words, the copy of, e.g., who in Q, does not bear the relevant features (basically φ-features) to qualify as the argument of a λ- predicate. A-movement on the other hand, will always qualify as the producer of a λ-predicate, since A-movement carries all the features of the element along. Hence the impossibility of reconstruction in cases of A-movement.

18 I believe that Reuland’s analysis of wh-movement is compatible with Reinhart’s (1997) proposal that wh-in-situ is interpreted through choice-functions and not through covert wh-movement, if we extend her analysis to apply to all wh-questions. This is the line taken in Ruys (2000) who extends the choice-function analysis to apply to cases of QR as well.

36 Let’s assume that something along these lines is correct. Though Reuland does not discuss SCO, it is clear that Reinhart’s account of it should be reformulated. It is not difficult to do so, maintaining the gist of her analysis. Under current assumptions, the logical-syntax representation of (95a) would be (95b).

(95) a. * Who i did he i say we should invite t i ? b. Q [he (λy (y said we should invite (who (λx (x)))))] c. Q [he (λy (y said we should invite (who (λx (y)))))]

As in Reinhart (2000), x is bound by who . Being bound it cannot be bound by he , as in (95c). So, not only is the pronoun not in the c-command domain of who , as in the WCO cases, but, in addition, it attempts to bind a variable that is already bound; hence the more severe ungrammaticality of SCO configurations when compared with WCO ones.

4.3 Clitic doubling in logical syntax

These are the basics of the theory I will be assuming. I will now proceed to implement it in the Greek and Romania data. Remember that in the previous chapters I draw a distinction between cases for which the obviation of the WCO is more plausible to be seen as a side-effect of clitic doubling (Romanian care -questions and Greek QPs) and cases for which this solution does not seem to take us too far (Greek wh-questions). In this section I will discuss the latter. There are two issues to be considered; first, to provide a translation of clitic doubling structures into logical syntax representations; second, to see if this solves the empirical problem with SCO. The first issue is pretty straightforward. In a Sportiche-like analysis of clitic doubling clitics are inflections. In the logical syntax representations I have given so far I have followed Reinhart in that movement of the subject to TP forms a λ-predicate. This is independent of the actual property that T licenses (tense). Extending this analysis for clitic doubling requires the assumption that movement of the object to Spec, ClP forms a λ-predicate whose argument is the raised object. The clitic is, thus, translated into a λ-predicate, regardless of the property it licenses (specificity). I don't see any reason for this not to be so; if a predicate can be formed via A-movement of the subject from its VP internal position to the IP domain, the same should be possible for objects. The logical syntax representation for Romanian care -questions would then be as in (96b) and (96c) after binding.

(96) a. Pe care i l-a certat mama lui i e i ? pe which him-has scolded mother his ‘which one did his mother scold e ?’ b. Q [ pe care (λx ( mama y certat x)] c. Q [ pe care (λx ( mama x certat x)]

37

Note that it is the A-movement to Spec, ClP that counts for the formation of the λ-predicate. Further A'- movement to Spec, CP does not count for the creation of a λ-predicate. So, care at Spec, ClP is the argument of a λ-predicate that binds all free variables within it, namely lui (again under the assumption that the postverbal subject is located within the VP). The same holds for the Greek case in (97a) (regardless of whether to kathe pedhi undergoes QR). The relevant logical syntax representation is (97b) and (97c) after binding.

(97) a. To ajapai i mitera tu i to kathe pedhi i it loves the mother his the every child 'His mother loves every child' b. to kathe pedhi (λx (i mitera y ajapai x)) c. to kathe pedhi (λx (i mitera x ajapai x))

Recall that the clitic doubling analysis I have proposed in the previous chapter faced an empirical problem; it predicted an obviation of the SCO effect, contrary to fact. That was so for both traditional GB approaches to binding and RCB. The logical syntax definition of binding may be able to solve this problem. I can see two possible ways to do that. Neither is straightforward, but they might give us some insight. So far, we have seen one restriction on binding, the logical restriction that operators can only bind free variables. We have mentioned another one, Principle B. Reinhart argues that Principle B is not a restriction imposed by logic but a condition specific to the computational system. Logic does not exclude the binding construal (98b) for (98a) but the CS Principle B does.

(98) a. Max touched him b. Max (λx (x touched x))

Consider now the SCO violation in (99a) and its logical-syntax representation after binding in (99b).

(99) a. * To ajapai pro i to kathe pedhi i it loves-he the every child 'He loves every child' b. to kathe pedhi (λx (x ajapai x))

The logical-syntax representations in (99b) and (99b) are identical. If Principle B is to be understood as a restriction on binding that operates on logical-syntax representations excluding the construal in (98b), then the ungrammaticality of (99b) could be understood as a Principle B violation. The alternative is to appeal to the first restriction on binding, the logical restriction that operators can only bind free variables. This would be possible under the assumptions that the pro subject in (100) forms a λ-predicate (100a) that tries to bind the variable in object position, a variable already bound by to kathe pedhi , as in (100b).

38

(100) a. to kathe pedhi (λx (pro (λy (y ajapai x)))) b. to kathe pedhi (λx (pro (λy (y ajapai y))))

However, that contradicts with the conditions on λ-predicate formation I have assumed so far, namely that subjects become arguments of λ-predicates after A-movement from the VP internal position to Spec, TP. If the postverbal argument remains in its VP internal position, it is not clear how the λ- predicate is formed.

4.4 Resumption

In this section I will consider the obviation of the WCO effect in Greek doubled wh-questions. Recall that those cases exhibited three main properties that distinguished them with the cases above: they show no island sensitivity; the obviation of the WCO effect is not restricted to the domain of the clitic; the presence of the clitic is not triggered by specificity/ D-linking but by the presence of a coindexed pronoun. In what follows I will argue that the two first properties (and possibly also the third) can be accounted for if one assumes that we are actually dealing with a case of resumption. In order to make the link with resumption, though, it is required that we go one step back and examine whether resumption obviates crossover effects.

4.4.1 Resumption and Strong Crossover

One long-standing issue in the literature of resumption has been to determine whether a resumptive pronoun exhibits properties of pronouns or of wh-traces. 19 Since SCO effects are considered to be characteristic of wh-traces, then the presence of SCO effects in cases of resumption would help solving the issue. There is a consensus in the literature that resumption does not obviate SCO effects (see, e.g., McCloskey to appear). The way to demonstrate that requires a rather complicated argument that first appeared in Dermidache (1991). According to this argument, it is not enough to replace the trace in a SCO configuration with a resumptive pronoun and see the effects. Even in languages where resumption in wh-questions is possible, the argument claims, a potentially grammatical resumptive equivalent of (101a) with a resumptive pronoun, as in (101b), will not prove that resumption obviates SCO effects.

(101) a. * Who i did he i say we should invite t i ? b. Who did he say we should invite him ?

19 Note that this discussion is only meaningful under a GB approach to binding.

39 The reason is that in these cases it can be that the higher of the two pronouns is the resumptive one. If this is so, then the (potential) grammaticality of (101b) is explained by whatever explains the grammaticality of (102).

(102) Who i t i said that we should invite him i ?

In order to distinguish which is the resumptive pronoun we need to use an element that gives rise to SCO effects but cannot function as a resumptive, the argument proceeds. Epithets are assumed to be such elements. Replacing the first pronoun with an epithet, as in the Irish example in (103) (McCloskey to appear) the SCO is present. Thus, resumption does not obviate SCO effects.

(103) * Sin an fear i ar dhúirt an bastard i co maró sé i muid that the man C said the bastard C kill he us ‘That’s the man that the bastard said that he would kill us’

There are two arguments against this reasoning. The first has to do with the alleged ambiguity wrt which of the two pronouns is the resumptive one. The argument seems to stand in languages like English, but n languages that exhibit overt case marking of the wh-element, the wh-element and the pronoun necessarily have to agree in case. Greek is a language that belongs to this group. Consider the contrasts in (104a-b) and (105a-b). 20

(104) a. * Pjos ipe i Maria oti tha ton kalesume sto party ? who_nom said the Mary that will him_acc invite-we to-the party ‘Who did Mary say that we will invite him to the party ?’ b. Pjon ipe i Maria oti that ton kalesume sto party ? who_acc said the Mary that will him_acc invite-we to-the party ‘Who did Mary say that we will invite him to the party ?’ (105) a. * Pjon ipe i Maria oti tha mas kalesi pro sto party ? who_acc said the Mary that will us invite-he to-the party ‘Who did Mary say that he will invite us to the party ?’ b. Pjos ipe i Maria oti that tha mas kalesi sto party ?

20 One might argue that case agreement between the antecedent and the resumptive pronoun in, e.g. (105b) is a result of clitic doubling. We have seen however that only additional material or partitive readings allow the option of clitic doubling in Greek. No such addition is required in this case for the clitic to be licensed. The data in (105b) and (116b) pose another problem, though. Recall that the resumptive element and the antecedent cannot appear in the same clause, as in (i). i. * Pjon ton ajapai i Maria ? who him loves the Mary ‘Who does Mary love ?’ The ungrammaticality of (i) contrasts with the grammaticality of (105b). A resumptive element can be licensed when it appears in the embedded clause. I don’t know how to explain this contrast (but see footnote 23).

40 who_nom said the Mary that will us invite-he to-the party ‘Who did Mary say that he will invite us to the party ?’

The contrasts above show that in Greek there is an independent way to see which pronoun is the resumptive one; it is the one that agrees in case with the antecedent. So, in the Greek equivalent of (101b) in (106b) it is the second pronoun that serves the resumptive function (106a shows the standard SCO violation in Greek).

(106) a. * Pjon i ipe pro i oti tha kalesume t i ? who_acc said-he that will invite-we ‘Who did he say that we will invite ?’

b. * Pjon i ipe pro i oti tha ton i kalesume ? who_acc said-he that will him_acc invite-we ‘Who did he say that we will invite him ?’

The ungrammaticality of (106b) seems to confirm that resumption does not obviate SCO effects. Consider, however, the contrast between (106b) and (107).

(107) Pjon i nomizi i Maria oti ipe pro i oti tha ton i kalesume ? who_acc thinks the Mary that said-he that will him_acc invite-we ‘Who does Mary think that he say that we will invite him ?’

The SCO effect is, indeed, obviated in (107). The only difference between (106b) and (107) is that in the former the intervening pronoun appears in the embedded clause. Since there is no other difference wrt to resumtion between the two cases, it is reasonable to attribute the ungrammaticality of (106b) to some independent property and not to resumption. I will discuss this issue latter in this section. For now, I conclude that the example in (107) is a clear case of a SCO effect obviated by the use of a resumptive pronoun. So, the only case in which the ambiguity wrt which one of the two pronouns serves the resumptive function can arise in Greek is when both pronouns bear the same case as the antecedent. (108) provides an example.

(108) Pjos i nomizi i Maria oti ipe pro i oti tha mas kalesi pro i sto parti ? Who_nom thinks the Mary that said-he that will us invite-he to-the party ‘Who does Mary think that he said that he will invite us to the party ?’

One could use an epithet to solve this out, like in the Irish example in (114).

(109) Pjos i nomizi i Maria oti ipe o malakas i oti tha mas kalesi pro i sto parti ?

41 Who_nom thinks the Mary that said the jerk that will us invite-he to-the party ‘Who does Mary think that the jerk said that he will invite us to the party ?’

The example in (109) is grammatical, indicating again that resumption does obviate SCO effects. However, in the Irish example the epithet is in the matrix clause, not in the embedded clause like in (109). Regulating this difference results in ungrammaticality.

(110) * Pjos i ipe o malakas i oti tha mas kalesi pro i sto party ? who_nom said the jerk that will us invite-he to-the party ‘Who did the jerk say that he will invite us to the party ?’

The data above actually show us two thinks; examples like (107) and (109) show that resumption does obviate SCO effects; the similar behavior of pronouns and epithets shows that epithets actually behave like pronouns; 21 the ungrammaticality of (106b) and (110) does not depend on whether the intervening element is a pronoun or an epithet, but should be attributed to some independent reason. This second observation puts in doubt one of the basic assumptions of Dermidache's argument we started with in this section, namely that epithets cannot function as resumptives. In fact, Safir (1996) gives two English examples where epithets can be resumptives.

(111) a. The judge will release anyone i who i it is unclear what the jerk i did

b. There was one prisoner i who i we couldn’t even figure out why the guy i was in jail

Other languages as well have been reported to allow epithets as resumptive elements. Aoun, Choueiri and Hornstein (2001) make this claim for Lebanese Arabic. Also Greek allows epithets to be resumptives, as in (112).

(112) Pjos i nomizi i Maria oti ipe o malakas i oti o Janis tin filise ? Who_nom thinks the Mary that said the jerk that John her kissed ‘Who does Mary think that the jerk said that John kissed her ?’

Again, this is not possible when the epithet appears in the subject position of the matrix clause. 22

21 We have already seen in section 2.3 that epithets can be bound variables, when the c-command restriction is regulated. Since there is no A-binding involved in the cases of resumption I am considering here, it would be highly unexpected for epithets not behave as pronouns. See Aoun, Choueiri and Hornstein (2001) for a similar argument. 22 The equivalent of (113) with a pro in the place of the epithet, as in (i), is not revealing. Since Greek is a null subject language (i) seems indistinguishable with the alternative in (ii) that is created by movement.

i. Pjos i nomizi pro i oti o Janis filise ti Maria ?

ii. Pjos i nomizi t i oti o Janis filise ti Maria ?

42

(113) * Pjos i nomizi o malakas i oti o Janis filise ti Maria ? Who_nom thinks the jerk that John kissed the Mary ‘Who does the jerk think that John kissed Mary ?’

What the contrast between (112) and (113) reveals is that the ungrammaticality of cases like (103) and (110) cannot be attributed to the alleged non-compatibility of epithets with resumptive elements. Instead, it is to be attributed to the position of the pronoun/ epithet: epithets/ pronouns cannot be bound by antecedents in an A’-position when they occupy a subjacent subject position. There have been various attempts in the literature to formulate a restriction like that, which is usually referred to as the A’-Disjointness Requirement (see, e.g. McCloskey 1990, Aoun and Li 1989, Aoun 2000). I will follow McCloskey (1990) in calling it the Highest Subject Restriction and will formulate it in what appears to me to be its simplest possible form. 23

(114) The Highest Subject Restriction: A pronoun in subject position cannot be bound by an operator in a subjacent A’-position

The restriction in (114) seems to be (and probably is) highly stipulative and the rough version of it I adopt here does not give any real insight on why it should hold. However, since it is the only way I can see to derive the facts above I will adopt it. On the basis of facts where independent reasons can distinguish between the resumptive pronoun and the intervening pronoun I have argued in this section that resumption does obviate SCO effects. The ungrammaticality of similar cases with epithets intervening between the antecedent and the resumptive element is to be explained under the assumption that epithets behave like pronouns in all relevant respects and are, in effects, also subject to the Highest Subject Restriction.

4.4.2 Resumption and Weak Crossover

who thinks e that the John kissed the Mary ‘Who thinnks that John kissed Mary ?’ 23 My formulation differs from previous formulations that either narrow the restriction to apply only to resumptive pronouns (e.g. McCloskey 1990) or extend it to cover pronouns in other positions within the same clause (e.g. Aoun 2000). The former cannot be adopted here, as the contrast between (106b) and (107) shows; the pronoun in (107) is not the resumptive element, but still the example is ungrammatical. The latter would wrongly rule out cases like (i).

i. Pjanu i ipes sti mitera tu i oti filise o Janis ti Maria ? whose_gen told-you to-the mother his_gen that kissed the John the Mary ‘Who did you tell his mother that John kissed Mary ?’ It would, however, explain why resumptives are possible when they appear in the embedded clause and not when they appear in the matrix clause (see footnote 20), as suggested by Tsimpli (1999), but the restriction would be violated when an intervening coindexed pronoun appears that triggers the appearance of a clitic.

43

Although the literature fails to see that resumption obviates SCO effects, the same does not hold for WCO effects. The consensus in the literature is that resumption does obviate WCO effects (see McCloskey to appear). This is somehow surprising. One would think that Dermidache’s argument should hold in the case of WCO effects as well and the same discussion should start all over again (using epithets in the place of pronouns etc. etc.). However, there is a clear difference between SCO and WCO effects in cases of resumption; the Highest Subject restriction does not obscure the picture in the latter though it does in the former. Replacing an intervening pronoun with an epithet does not yield a violation of the Highest Subject Restriction in WCO configurations (the intervening pronoun is not in subject position) but it does so in SCO configurations (when the subject is subjacent to the antecedent). I will not go into the same discussion again. I will simply provide two cases of resumption that obviate the WCO effect and mention some Greek facts that confirm the prediction. Safir (1996) mentions one such Hebrew example from Sells (1984). The structure with no resumptive is ungrammatical, whereas the structure with a resumptive is grammatical.

(115) a. * Ha-ʔiš še ʔim-o ʔohevet the-man that mother-his loves t

‘The man i that his i mother loves’ b. Ha-ʔiš še ʔim-o ʔohevet oto the-man that mother-his loves him

‘The man i that his i mother loves him i’

McCloskey (to appear) gives an Irish example.

(116) An fear ar fhág a bhean é The man C left his wife him

‘The man i that his i wife left him i’

In Greek, case agreement can distinguish between the resumptive and the intervening pronoun as shown by the contrasts in (117a-b) and (118a-b).

(117) a. * Pjon ipe i Maria oti i mitera tu mas kalese sto party ? who_acc said the Mary that the mother his_gen us invited to-the party ‘Who did Mary say that his mother invited us to the party ?’ b. ? Pjanu ipe i Maria oti i mitera tu mas kalese sto party ? who_gen said the Mary that the mother his_gen us invited to-the party ‘Who did Mary say that his mother invited us to the party ?’

(118) a. * Pjanu ipes sti Maria oti tha ton kalesume sto party ?

44 who_gen told-you to-the Mary that will him_acc invite-we to-the party ‘Who did you tell Mary that we will invite him to the party ?’

b. Pjon ipes sti Maria oti that ton kalesume sto party ? who_acc told-you to-the Mary that will him_acc invite-we to-the party ‘Who did you tell Mary that we will invite him to the party ?’

Again, only pronouns that agree with the antecedent in case can function as resumptive elements. In effect, in the WCO configurations in (119) it is the accusative clitic that is the resumptive pronoun and the WCO effect is obviated.

(119) a. Pjon i ton ajapai i mitera tu i ? who_acc him_acc loves the mother his_gen ‘Who does his mother love him ?’

b. Pjon i ipe i mitera tu i oti ton i ajapai i Eleni ? who_acc said the mother his_gen that him_acc loves the Helen ‘Who did his mother say that Helen loves him ?’

4.4.3 Resumption in logical syntax

What we saw above is that resumption can, indeed, obviate all crossover effects despite the arguments against this view that have been proposed in the literature. Arguing against these dominant assumptions in the literature I have just shown that there is no a priori reason not to consider the possibility that the obviation of crossover effects can be attributed to resumption. I have said nothing about why this should be so. In what follows I give a short presentation on why the obviation of crossover effects in cases of resumption poses a problem for most theories of binding and that, under the current theory, this obviation cannot only be accounted for, but is, in fact, predicted. Note, first, that although the two previous sections were as descriptive as possible, some minimal theoretical assumptions were implicit. One was that elements in A’-positions linked with a resumptive pronoun actually bind the resumptive element and the bound element is translated into a variable. This assumption poses a problem for the GB approach to binding and its definition of a variable. If a resumptive pronoun (overt or covert) is translated into a variable after A’-binding, then Chomsky’s derivational definition of a variable should be reformulated. According to Chomsky’s definition pronouns are translated into variables under A-binding and empty categories are translated into variables when they are traces left after A’-movement. A resumptive pronoun translated into a variable via A’-binding does not fit in this picture. That is the reason why there have been various attempts to reformulate the definition of variables within the GB approach to binding. Recall, e.g.,

45 Cinque’s (1990) reformulation that considered object pro s that were identified via A'-binding; Cingue’s definition is, in effect, a configurational definition: a variable is an element that is (locally) A’-bound. 24 When it comes to crossover effects a simple reformulation of the definition of a variable does not solve the issue. Consider the Hebrew example in (120) (from Grodzinsky and Reinhart 1993). (120) Kol saxen se+ha+raas mealav hifria lo kara lamistara Every neighbor that the noise above him bothered him called (to) the police

The Bijection Principle, that assumes that A’-bound pronouns are variables, would rule (120) out; ‘every neighbor’ simultaneously binds two variables. If one wants to deal with (120) and the similar cases we have seen in the previous sections within a GB approach to binding, then something more needs to be said. Safir (1996) proposes an analysis that deals with these cases. I will delay a discussion of it until the next section. The same problem arises with RCB when it comes to cases of resumption like the one in (120). At no point in the derivation are the pronouns bound by an antecedent in an A-position. The translation procedures fail to recognize the pronouns as bound variables; they are both A’-bound but not empty, they have no lexical content but they are not A-bound. The logical syntax definition of binding does not face the same problems. If the antecedent in the resumptive cases I have been considering is the argument of a λ-predicate then it can automatically bind all pronouns (and empty categories) within that predicate. So far, we have seen only one configuration in which NPs become arguments of λ-predicates; after A-movement to the IP domain. Since the antecedents in all cases of resumption we have seen occupy A’-positions this option is excluded. A’-movement, on the other hand, did not qualify because the ϕ-features of the moved element stayed in the base-position. If this is so, it is expected in the logical syntax framework proposed by Reuland that if the antecedents in cases of resumption are not moved but base- generated in their position, their feature specification is intact and can, thus, be arguments of λ- predicates. 25 A base-generation analysis is, indeed, the standard analysis for cases of resumption, usually grounded on the fact that they amnesty island violations. The same has been shown to hold for Greek doubled wh-elements in (45) and (46). 26 Before concluding this section it is important to make clear what are the data that actually support the analysis. In order to support the claim that base-generation allows an element to be the argument of a λ-predicate, Reuland gives the following two English examples in (121a-b) for WCO and SCO, respectively.

24 See Safir (1996) for an overview of the issues discussed in this paragraph. 25 In this sense the A/A’ distinction collapses in its straightforward formulation that has been assumed in the literature. When the antecedent occupies an A’-position there is a second distinction to be considered; if it has been moved to the A'-position, binding will not be possible; if, on the other hand, the element is base-generated, binding will be possible. 26 In fact, the binding data presented here could be seen as an independent argument against movement analyses of resumptive configurations, as in, e.g., Grohmann (2000).

46

(121) a. That guy i Mary thinks his i friend said Jane loved him i

b. That guy i Mary thinks he i said Jane loved him i

However, the examples above do not necessarily prove that binding is involved. They could simply be cases of coreference; the NP occupies an A’-position, so no Principle C issue arises and nothing excludes the possibility that that guy and the pronouns are simply coreferent. The same could be argued for the obviated crossover effects in the Irish and Hebrew data with referential antecedents in the previous sections. Nevertheless, this complication does not arise with Greek wh-questions and the Hebrew example in (120); these examples do not involve referential antecedent but quantified antecedents. The coreference option is thus excluded and binding should be involved. I repeat here the relevant Greek data and their corresponding binding construals.

(122) a. Pjon i ipe i mitera tu i oti tha ton i kalesume sto parti ? who said the mother his that will him we-invite to-the party 'Who did his mother say that we will invite him to the party ?' b. pjon (λx ( ipe i mitera x oti tha kalesume x sto parti))

(123) a. Pjon i tu i ipes oti tha ton i kalesume sto parti ? who him told-you that will him we-invite to-the party 'Who did you tell him that we will invite him to the party ?' b. pjon (λx ( ipes x oti tha kalesume x sto parti))

(124) a. Pjon i nomizi i Maria oti ipe pro i oti tha ton i kalesume sto parti ? Who thinks the Mary that said-he that will him invite-we to-the party ‘Who does Mary think that he said that we will invite him to the party ?’ b. pjon (λx (nomizi i Maria oti x (λy (y ipe oti tha kalesume y sto parti))))

The SCO example in (125a) and its binding construal in (125b) would then be ruled out by the Highest Subject Restriction, which, I propose here, should be seen as a third restriction on binding.

(125) a. * Pjon i ipe pro i oti tha ton i kalesume ? who_acc said-he that will him_acc invite-we ‘Who did he say that we will invite him ?’ b. pjon (λx (x (λy (y ipe oti tha kalesume y))))

4.4.4 Safir 1996

It is clear that in the framework assumed here the debate on the exact formulation of the definition of a variable is redundant. There is, however, an account of crossover within the GB approach to binding that reaches very similar conclusions with the ones I have presented in the previous section. Safir (1996) argues that some instances of A’-binding are derivational whereas others are representational.

47 Derivational A’-chains are the product of an operator moving and leaving behind a trace (e.g. all instances of wh-movement) and exhibit d(erivational)A’-binding. Representational A’-chains, on the other hand, are not the results of any syntactic operation; they are formed due to identity of indices. Safir argues that cases of resumption are cases of r(epresentational)A’-binding. Moreover, Safir proposes the A’-Consistency condition in (126).

(126) A’-Consistency An A’-chain is either consistently dA’-binding or consistently rA’-binding

Consider now the standard WCO violation in (127a) and the grammatical (127b).

(127) a. * Who i does his i mother love t i ?

b. Who i t i loves his i mother ?

In (127a) both his and the trace are locally A’-bound by who . Thus, the chain headed by who has two tails. But the chain is not consistent, who dA’-binds the trace but it rA’-binds his . In (127b), on the other hand, there is only one A’-chain, namely (who, t) and the trace is dA’-bound. The pronoun is not locally A’-bound (it is A-bound by the trace) so it does not participate in the A’-chain and (127b) conforms with the A’-Consistency condition. Consider now a case of resumption, e.g. the obviated WCO effect in Greek wh-questions.

(128) a. Pjon i ton i ajapai i mitera tu i ? who him loves the mother his ‘Who does his mother love him ?’

b. [ CP pjon i [ IP ajapai j [ ClP pro i [ Cl’ ton [ VP i mitera tu [ VP’ t j t i ]]]]]]

As in (127b) there is one A’-chain headed by pjon , that has two tails, tu and pro . But in this case the A’-chain is consistent; the relation between pjon and tu and pjon and pro is in both cases rA’-binding. So, the A’-chain is consistent. It is easy to see how this reasoning can be applied to all obviated crossover effects in cases of resumption I have presented so far (with the proviso that Safir adopts the Highest Subject Restriction, as I have done). In fact, Reuland sees his proposal as an explanation of why the A’-Consistency Condition should hold. The A’-Consistency condition can be seen as a descriptive generalization for all cases where A’-movement prevents the formation of a binding configuration for the reasons I have presented in section 4.2. Although the empirical coverage of the two proposals seems to be identical in all clear cases of resumption, I still think that we can distinguish between the two. The weakest crossover cases come in handy once more. Consider again the weakest crossover configuration in (129).

(129) John i will be easy NO i to get his i wife to vouch for e i

48 Safir argues that weakest crossover examples like (129) should be treated like cases of resumption. He argues that the empty category in (129) is not a null epithet but an A’-bound resumptive pro , in the spirit of Cinque (1990). The obvious disadvantage of the resumptive analysis, namely that weakest crossover configurations obey islands, is circumvented, Safir argues, if one assumes that the Null Operator is not base-generated in it’s A’-position, but has landed there after movement. The trace it leaves behind is turned into a resumptive only at LF. If the empty category in (129) is a resumptive pronoun then it obviates the WCO effects like all cases of resumption; turning the trace to a resumptive at LF results in turning an dA’-chain into a rA’-chain. If so, the A’-chain headed by the NO and tailed by his and the (now resumptive) empty category is consistent. Safir proposes that conversion of a trace to a resumptive is regulated by the resumption conversion rule in (130).

(130) A derivational variable v is converted to a (resumptive) pronoun iff v is not bound by a true quantifier.

In effect, Safir maintains Lasnik and Stowell’s basic distinction between ‘true-’ and ‘non-true-QPs’, but focuses the discussion on the empty category itself rather on the antecedent. This being said, it is obvious that Safir’s analysis faces the same basic problems as Lasnik and Stowell’s. I have already mentioned some in section 2.3. Here I will again follow Ruys (2004) and mention two more problems as they manifest themself in Safir's analysis. 27 One important point to be mentioned is that whereas in Lasnik and Stowell's analysis the obviation of the WCO effect is solely dependant on the existence of a moved NO or referential NP that leaves behind a null epithet trace Safir's resumption conversion rule is stated as to apply to the actual antecedent. In Safir's (1996: p.329) words: 'The conclusion that empty category variables are pronominal if they are locally A'-bound by antecedents that are not true quantifiers has been supported by the parallel with overt pronominal variables'. In other words, whether an empty category can be converted to a resumptive depends on the actual antecedent that binds the empty category (left after movement of the NO). Thus, if the ultimate binder is a true QP WCO effects are expected to arise. As

27 Ruys gives more arguments against Safir's analysis. Since he favors a licensing approach to the weakest crossover cases, that is an approach that interprets pronouns as bound variables iff they are bound by an antecedent in A-position, he argues that resumptive pronouns do not actually obviate WCO effects. He supports this claim with two arguments; the first one is that most reported cases involve referential antecedents and can, thus, be analyzed as cases of coreference. His second argument is that Dermidache's argument should also hold for WCO cases and, in effect, a grammatical counterpart of the standard WCO violation that includes a resumptive pronoun does not prove that resumption does not cause WCO violations, because the first pronoun could be the resumptive one. As for the first argument, I have given examples with quantified expresions that obviate the WCO effect in section 4.4.3 (and see also section 5.1.1). The same Greek examples can be used against Ruys' second argument, since in Greek case distinguishes between the resumptive and the intervening pronoun. Ruys gives one real counterexample from Vata, where resumptive pronouns can be distinguished from normal pronouns, and a WCO effect ariese. If this is so there is some evidence that resumption does not always save WCO violations.

49 Safir notes himself there are counterexamples to this claim. Look at the parasitic gap and tough- movement examples in (131a) and (131b), respectively; both contain true QPs but no WCO effect arises (the examples are from Ruys 2004).

(131) a. Who i did you stay with t i NO i before his i wife had spoken to e i ? b. No child is old enough for us NO PRO to ask his mother to give up e

The second argument concerns topicalization structures that Lasnik and Stowell also include in the weakest crossover configurations (once more the argument is taken from Ruys 2004 who presents it to contradict Lasnik and Stowell's analysis). Consider the contrast in (132a-b).

(132) a. This book i, (NO i) I expect its i author to buy e i

b. * Everybody else i, (NO i) I told his i wife that I had called e i

At a first glance, the contrast seems to confirm the resumption conversion rule's distinction between true- and non-QPs. Under the assumption that topicalization involves movement of a NO, Safir can argue that the trace is converted to a resumptive pronoun when bound by a referential antecedent (132a) (and, in effect, does not show the WCO effect) but not when bound by a true QP, as in (132b). Note, however, that this analysis predicts that the referential topic actually binds the pronoun and that the grammaticality (132a) is not a simple case of coreference. Ruys provides a range of arguments that (132a) actually involves coreference. I will only present one here. Using one of the usual disambiguators ( only and even ), it turns out that (133) only allows the coreference reading. It can only mean 'even Winnie belongs to the set of people that Winnie's mother likes' (the coreference reading) and not 'even Winnie belongs to the set of people that have mothers that love them'.

(133) Even Winnie his mother likes

So, for Ruys, (132a) is a case of coreference. (132b), on the other hand, is excluded because topics cannot variable-bind pronouns; they occupy an A'-position and, under the licensing approach Ruys adopts, they don't qualify as proper antecedents for a pronoun to be interpreted as a bound variable. The version of the logical syntax approach to binding I have adopted overcomes both problems. Consider first the topicalization structures. If no NO operator is involved, it is the topicalized element itself that undergoes A'-movement to the sentence initial position. So, for interpretational purposes both topicalized elements are interpreted in their source positions, as in (134a-b).

(134) a. I expect its author to buy this book b. I told his wife that I had called (everybody else (λx (x)))

The representation in (134a) clearly allows coreference (and only coreference); its does not c- command the book at the relevant level of representation. Since everybody else in (134b) is a QP the

50 coreference option is not available; but neither is binding, since its is not c-commanded by the operator. The fact that there are weakest crossover cases that allow true QPs as antecedents poses no problem as well. In the logical syntax framework the licensing analysis of Ruys presented in section 3.4.2 can be adopted with almost no modifications. What is crucial for the logical syntax analysis is that a λ-predicate is formed and that the antecedent (quantified or referential, in A- or A’-position) is the argument of this predicate. It can then bind all free pronouns and empty categories in that λ- predicate, regardless of how they were formed (e.g. by operator movement of a NO). For concreteness, consider again the tough -movement example in (135). Following Ruys (2004) we can assume that the NO functions as the λ-operator that changes the AP into a predicate that applies to the matrix subject. The logical syntax representation of (135a) is then (135b). The same holds when the antecedent is quantificational as in (136a-b), with no need for additional stipulations.

(135) a. John i will be easy NO i to get his i wife to vouch for e i b. John (λx (it will be easy to get x’s wife to vouch for x))

(136) a. Who i t i will be easy to get his i wife to vouch for e i ? b. who (λx (it will be easy to get x’s wife to vouch for x))

4.4.5 Resumption and economy

In the analysis I have proposed in the previous sections three out of four of the major differences between Romanian care -questions and doubled Greek QPs, on the one hand, and Greek 'doubled' wh-questions, on the other, can be accounted for straightforwardly. The former are specific/ D-linked, show island sensitivity and obviate WCO effects only in the local domain of the clitic. All three properties are the result of clitic doubling. The latter are not specific/ D-linked, do not show island sensitivity and obviate WCO effects across the board. All three properties are the result of resumption. However, the fourth (and probably the most intriguing) difference has not yet been addressed. Although the fact that the former cases show no WCO effects can be seen as a side effect of clitic doubling, the obviation of WCO effects in the case of Greek 'doubled' wh-elements cannot be seen as a side effect of resumption. Recall that wh-questions in Greek do not generally license clitics; (137) is ungrammatical.

(137) a. * Pjon ton ajapai i Maria ? who him loves the Mary ‘Who does Mary love ?’

But the clitic is obligatory when an intervening coindexed pronoun appears, with no consequenses for the specificity of the wh-element (there is no specificity requirement on pjon in (149b)).

(138) a. * Pjon i ajapai i mitera tu i ?

51 who him loves the mother his ‘Who does his mother love ?’

b. Pjon i ton ajapai i mitera tu i ? who him loves the mother his ‘Who does his mother love ?’

It seems, thus, that the presence of the clitic licenses a certain interpretation, the bound variable reading of tu . I do not have a definite explanation for this observation. In what follows I will make a suggestion, namely that the pattern found in (137) and (138) is the result of economy considerations. ‘Economy considerations are believed to choose the most optimal derivation from a set of competitors (a 'reference set')’ (Fox 1995, p.283). Two questions immediately arise; how is optimality achieved and what determines whether a given derivation should be part of the reference set or not. One possible answer to the first question (e.g. the one promoted in Chomsky (1992) and Fox (1995)) is that optimality is achieved by reducing instances of moving or by minimizing their length. Another possible answer is that optimality is achieved by reducing the number of steps needed in a derivation (Aoun, Choueiri and Hornstein 2001). Let me assume the latter option, although both answers are equivalent for what I am proposing. The second question is more controversial. It is generally assumed that the reference set is composed by derivations that start with the same Numeration and end up in identical LFs. Reinhart (1997) proposes to expand the reference set and include derivations that end up with interpretively equivalent LFs (what is usually referred to as ‘global economy’). Let me assume something along these lines. So, according to the conception of economy I adopt here, one convergent derivation blocks another if it has fewer steps (Merge/ Move) and results in an interpretively equivalent logical syntax representation. Derivations resulting in non-equivalent logical syntax representations are not compared, hence cannot block each other. An idea along these lines has been influential in order to regulate the application of QR (Reinhart 1997, Fox 1995); QR will only apply if it results in an interpretation that would otherwise not be available (inverse scope). My suggestion is essentially identical: a clitic (resumption) is licensed in Greek wh-questions when it results in an interpretation (the bound variable reading of the intervening pronoun) that would otherwise not be available. Consider first the example in (139). (139a) can have two possible interpretations, as indicated by the indices in (139b) and (139c). Let’s call the first interpretation DiscourseReference (DR) (139b) and the second Binding (B) (139c).

(139) a. Pjon ajapai i mitera tu ? ‘Who does his mother love ?’

b. Pjon i ajapai i mitera tu j ?

c. Pjon i ajapai i mitera tu i ?

Let’s assumed first that DR is the designated interpretation. The optimal logical syntax representation would then be the one in (140) where tu is not bound by pjon .

52

(140) ajapai i mitera tu (pjon (λx (x))) If, on the other hand, the designated interpretation is B, then the optimal logical syntax representation is one in which tu is bound by pjon , as in (141).

(141) (pjon (λx (i mitera x ajapai x))

The intervening pronoun can only be bound by the wh-operator if it is in the c-command domain of it. Under current assumptions this can only be done if pjon is base generated at Spec, CP, in ither words, if a strategy of resumption is employed. Note, that the logical syntax representation in (141) does not exclude the DR interpretation, as in (142).

(142) (pjon (λx (i mitera tu ajapai x))

Why, then, is it not chosen for (139b)? Because (142) is derived via resumption and resumption is more costly than simple wh-movement to Spec, CP. Consider the competing derivations.

(143) a. [ CP pjon i [ IP ajapai j [ VP i mitera tu [ VP’ t j t i ]]]]]]

b. [ CP pjon i [ IP ajapai j [ ClP pro i [ Cl’ ton [ VP i mitera tu [ VP’ t j t i ]]]]]]

(143a) is more economical than (143b), since the derivation in (143b) requires more steps (merging of the clitic, movement of pro to Spec, ClP etc.). Thus, the derivation in (143a) will be chosen, a derivation that will end up with the logical syntax representation in (140). But in order to derive interpretation B, (143b) that results in the logical syntax representation (141), is the only option. Note that even if optimality is achieved by comparing logical syntax representations and not syntactic derivations, then (142) is still more costly than (140). (140) can only have a binding construal, whereas (142) can have: a construal in which tu picks up a discourse referent (142), a binding construal (141) or even a coreference/ covaluation construal, as in (144), which requires the strenuous computation of Rule I.

(144) (pjon (λx (i mitera tu ajapai x)) and tu=x

All these construals must be computed, before choosing the intended interpretation derived by (142). Economy considerations will, thus, choose (143a) and its logical syntax representation (140) for the designated interpretation DR. For the designated interpretation B, on the other hand, derivation (143b) will be chosen since it is the only derivation that can end up in the logical syntax representation (142). Since the logical syntax representations correspond to different interpretations, they are never compared with each other, so both (145a) and (145b) are allowed.

(145) a. Pjon i ajapai i mitera tu j ?

53 b. Pjon i ton ajapai i mitera tu i ?

Consider now the sentence in (146).

(146) Pjon ajapai i Maria ? who loves the Mary ‘Who does Mary love ?’

(146) will have the same interpretation under both logical syntax representations in (147a-b).

(147) a. ajapai i Maria (pjon (λx (x))) b. (pjon (λx (i Maria ajapai x))

The two representations are interpretively equivalent, but the first is derived via wh-movement and the second via resumption. The competing derivations are given in (148a) and (148b), respectively.

(148) a. [ CP pjon i [ IP ajapai j [ VP i Maria [ VP’ t j t i ]]]]]]

b. [ CP pjon i [ IP ajapai j [ ClP pro i [ Cl’ ton [ VP i Maria [ VP’ t j t i ]]]]]]

As before, (148b) is more costly than (148a), so it will never be chosen. Thus, (149a) is grammatical, but (149b) is ungrammatical.

(149) a. Pjon ajapai i Maria ? b. * Pjon ton ajapai i Maria ?

54

5 Clitic Left Dislocation

In this chapter I will present another test case for the conclusions I have drawn in the previous chapter, namely Clitic Left Dislocation (CLLD). CLLD is a topicalization structure in which a topicalized element in a left peripheral position is linked with an accusative clitic within the clause with which it agrees in case (accusative). As I have already mentioned in section 2.3 for Italian, the literature on CLLD is divided between a base-generation and a movement analysis of the phenomenon. I have argued that the patterns of crossover effects can distinguish between the two. In what follows I will apply the same reasoning in CLLD cases in Greek and make a suggestion for Italian CLLD structures.

5.1 CLLD in Greek

The movement vs. base-generation debate manifests itself in the literature on CLLD in Greek by two competing analyses; one that argues for the former and treats CLLD as a case of resumption in which the clitic is the resumptive element (Anagnostopoulou 1994, 1997, Iatridou 1991) and one that argues for the latter and treats CLLD as involving movement out of a clitic doubling configuration (Agouraki 1993). Under current assumptions the two analyses should be reformulated as follows: in the base- generation analysis it is not the clitic itself that functions as a resumptive but an object pro , identified in a clitic doubling configuration. In the movement analysis the doubled NP undergoes a first step of A- movement to Spec, ClP before landing in the peripheral A’-position. 28 The two competing derivations for the example in (150a) are given in (150b-c) for the base-generation and the movement analyses, respectively.

(150) a. To Jani ton ajapai i Maria the John_acc him_acc loves the Mary ‘As for John, Mary loves him’

b. [ TopP to jani [ IP ajapai j [ ClP pro i [ Cl’ ton [ VP i Maria [ V’ t j t i ]]]]]]

c. [ TopP to jani i [ IP ajapai j [ ClP t i [ Cl’ ton [ VP i Maria [ V’ t j t i ]]]]]]

5.1.1 Crossover effects

Note that using crossover effects to distinguish between the two analyses requires some caution. One cannot simply create a WCO configuration by embedding a pronoun within the subject, as in (151a), and draw conclusions on its (un)grammaticality. Both analyses predict that (151a) will be grammatical

28 I won't take stand on the question what exactly the left peripheral position is. For concreteness I will assume, like Cecchetto 2000, that the NP occupies the specifier of a TopicPhrase. See Anagnostopoulou (1997) for arguments that the dislocated element is adjoined to IP.

55 (as it is), since the pronoun occupies a position within a λ-predicate whose sister is to Jani , in the logical syntax representation of both a base-generation (151b) and a movement (151c) analysis. The logical syntax representations end up identical since further A’-movement of to Jani from Spec, ClP to Spec, TopP in (151c) does not create a λ-predicate.

(151) a. To Jani i ton ajapai i mitera tu i the John him loves the mother his ‘As for John, his mother loves him’ b. to Jani (λx (i mitera x ajapai x)) c. to Jani (λx (i mitera x ajapai x))

In fact, both analyses predict that to Jani actually binds tu and that the grammaticality of (151a) cannot be attributed to coreference. 29 This is confirmed when using mono (‘only’) in order to distinguish between a coreference and a binding reading, as in (152).

(152) Mono to Jani i ton ajapai i mitera tu i only the John him loves his mother ‘Only John, his mother loves him’

The example in (152) is ambiguous; it can either mean ‘only John belongs to the set of people that are loved by their mothers’ (the binding reading) or ‘only John belongs to the set of people that John’s mother loves’ (the coreference reading). In fact, there is a strong preference for the binding reading. 30 The fact that (151a) is not the right example to look at is a result of the clause-boundedness of clitic doubling. Since the intervening pronoun is within the domain of the clitic it is expected to be bound. So, let’s try two examples in which the intervening pronoun occupies a position in a matrix clause. The examples with their corresponding logical syntax representations for the competing analyses are given in (153) and (154).

(153) a. To Jani i ipe i mitera tu i ipe oti i Maria ton i ajapai

29 Coreference could be possible under a movement analysis in which the dislocated NP moves out of a clitic doubling configuration directly to Spec, TopP without passing through an intermediate A-position, in the spirit of Dobrovie-Sorin’s (1990) analysis of Romanian care -questions (see section 2.1). 30 Note that the same example can be used to distinguish between a base-generation and a movement analysis. The clitic doubling equivalent of (152) is not grammatical to begin with, as shown in (i). i. * I Maria ton ajapai mono to Jani the Mary him loves only the John ‘Mary loves only John’ Mono can only modify the NP when it appears in a dislocated position. Thus, the ungrammaticality of (i) speaks for a base-generation analysis. See Anagnostopoulou (1997) on further differences wrt to the types of elements that can appear in dislocated and clitic doubling structures.

56 the John said the mother his that the Mary him loves ‘As for John, his mother said that Mary loves him’ b. to Jani (λx (i mitera x ipe oti i Maria ajapai x)) c. i mitera y ipe oti (to Jani (λx (i Maria ajapai x)))

(154) a. To Jani i i Eleni tu i ipe oti i Maria ton i ajapai the John the Helen told him that the Mary loves him ‘As for John, Helen told him that Mary loves him’ b. to Jani (λx (i Eleni ipe x oti i Maria ajapai x)) c. i Eleni ipe y oti (to Jani (λx (i Maria ajapai x)))

The examples in (153) and (154) are no more revealing than the example in (151). That is so because nothing excludes the option of coreference between y and to Jani in (153c) and (154c). 31 However, the option of coreference can be excluded when we use a quantified rather that a referential NP. (155a) is the equivalent of the unrevealing example in (151). The examples in (156) and (157) with their corresponding logical syntax representations for the two analyses are more helpful.

(155) a. Kathe andra i ton ajapai i mitera tu i every man him loves the mother his ‘Every man, his mother loves him’ b. kathe andra (λx (i mitera x ajapai x)) c. kathe andra (λx (i mitera x ajapai x))

(156) a. Kathe andra i ipe i mitera tu i oti i Maria ton i ajapai every man said the mother his that the Mary him loves ‘Every man, his mother said that Mary loves him’ b. kathe andra (λx (i mitera x ipe oti i Maria ajapai x)) c. * i mitera y ipe oti (kathe andra (λx (i Maria ajapai x))) and y=kathe andra

(157) a. Kathe andra i i Eleni tu i ipe oti i Maria ton i ajapai every man the Helen told him that the Mary loves him ‘Every man, Helen told him that Mary loves him’ b. kathe andra (λx (i Eleni ipe x oti i Maria ajapai x)) c. * i Eleni ipe y oti (kathe andra (λx (i Maria ajapai x))) and y=kathe andra

All the examples above are grammatical. (156c) and (156c) show that the movement analysis cannot predict this grammaticality; y is not c-commanded by the antecedent and direct coreference with it is not possible since it is a quantified expression. The base-generation analysis faces no problem. Since

31 My intuitions collapse when I try to use only in order to disambiguate the examples in (154) and (155).

57 the dislocated NP is a base-generated NP in the left peripheral position qualifies as the argument of a λ-predicate and binds all variables within that predicate. I conclude on the basis of these facts that CLLD in Greek does not involve movement of the dislocated NP; the NP is merged directly in the left peripheral position. SCO effects confirm this conclusion. As predicted, they show the same pattern as in the cases of doubled interrogatives, I discussed in the previous section. That means that for both referential and quantified NPs CLLD shows no SCO effects, with the exception of the examples in (158a) and (160a) that are ruled out by the Highest Subject Restriction.

(158) a. * To Jani i ipe pro i oti tha ton i kalesume the John said-he that will him invite-we ‘As for John, he said that we will invite him’ b. to Jani (λx (x (λy (y ipe oti tha kalesume y))))

(159) a. To Jani i nomizi i Maria oti ipe pro i oti tha ton i kalesume the John thinks the Mary that said-he that will him invite-we ‘As for John, Mary thinks that he said that we will invite him’ b. to Jani (λx (nomizi i Maria oti x (λy (y ipe oti tha kalesume y))))

(160) a. * Kathe andra i ipe pro i oti tha ton i kalesume every man said-he that will him invite-we ‘Every man, he said that we will invite him’ b. kathe andra (λx (x (λy (y ipe oti tha kalesume y))))

(161) a. Kathe andra i nomizi i Maria oti ipe pro i oti tha ton i kalesume ? kathe andra thinks the Mary that said-he that will him invite-we ‘Every man, Mary thinks that he said that we will invite him’ b. kathe andra (λx (nomizi i Maria oti x (λy (y ipe oti tha kalesume y))))

5.1.2 Obligatory clitics

There are two more issues that should be addressed wrt CLLD in Greek; the island sensitivity of CLLD and the obligatory presence of the clitic. Altghough the issues seem unrelated I will follow Anagnostopoulou (1997) in arguing that they are connected. Anagnostopoulou shows that CLLD in Greek does not amnesty island violations. She presents the following examples.32

32 Although CLLD is sensitive to strong islands like the ones presented in the main text, it is not sensitive to weak islands, like the wh-island in (i). i. To forema dhen ksero pu na to valo the dress not know_i where to it put ‘The dress, I don’t know where to put it’

58

(162) CNPC (relative) * Ton Jani sinantisa tin kopela pu ton idhe the John met-I the girl that him saw ‘As for John, I met the girl that show him’ (163) CNPC (noun complement) * Ton Jani dhiavasa tin idhisi oti ton sinelavan the John read-I the news that him arrested-they ‘As for John, I read the news that they arrested him’ (164) Adjunct island * Tin efimeridha apokimithike dhiavazontas tin the newspaper fell-asleep-he reading it ‘The newspaper, he fell asleep while reading it’

The island sensitivity of CLLD seems to contradict our prior conclusion that it is a case of resumption. Anagnostopoulou, who also argues in favor of a base-generation approach, proposes (following Iatridou 1991) that the clitic is actually an operator that mediates the relation between the dislocated element and the empty category. The empty category is left after movement of the clitic from the argument position in a position where it heads a ‘predicate variable chain’ (Anagnostopoulou 1997, p. 158). In other words, the clitic in this analysis is the exact counterpart in Greek of the NO in English weakest crossover cases. Movement of the clitic, Anagnostopoulou argues, is covert and this explains the selective island sensitivity of CLLD. So, the derivation in (165a) is translated into (165b).

(165) a. [ TopP NP [ CP cl…t…]]

b. [ TopP NP [ CP Op…t…]]

We have already seen that Cecchetto (2000) proposes a meaningful alternative. Cecchetto explains the obligatory presence of clitics in Italian CLLD structures arguing that the empty category is a pro . Being a pro it needs to be identified. Subject pro s are identified through rich morphological agreement on the verb. Objects pro s, on the other hand, get identified (check φ-features) in a clitic doubling configuration. Under an analysis of clitic doubling, like the one I have adopted here, where the clitic is a head translated into a λ-operator, nothing prevents us from adopting Cecchetto’s analysis for CLLD. The clitic has then a dual function; it licenses the empty category and then moves to form the λ-predicate, whose argument is the dislocated NP. Note that in both proposals the clitic is obligatory; without it no λ-predicate can be formed and the structure is uninterpretable. If this is true and if it is also true that the clitic in a clitic doubling configuration is the Greek equivalent of the NO in English we expect that a clitic will be present in all

59 Greek equivalents of English constructions that involve NO operators. Consider, e.g., the equivalent of the tough -movement construction in (166) (the example from Tsimpli 1999). 33,34

(166) I Maria ine omorfi na *(tin) kitas the Mary is pretty to her look-you ‘Mary is pretty to look at’

5.2 CLLD in Italian

This section will try to answer the same question for Italian CLLD, where the base-generation vs. movement debate has been posed in the literature in exactly the same way as it has been for Greek; Cinque (1990) proposes a base-generation analysis, Cecchetto (2000) argues for a movement out of clitic doubling configuration analysis. 35 I will only present a limited set of facts of crossover effects, so any conclusion should be handled with caution. Both Cinque and Cecchetto give the example in (167).

(167) Gianni i, sua i madre lo ama John his mother him loves ‘John, his mother loves him’

As I argued before this example is not revealing. Consider the example in (168), from Cinque (1990).

33 Equivalents of the other weakest crossover cases are hard to construct in Greek. I leave here aside the case of relative clauses where the distribution of clitics is far more complex than all the cases I have presented here. If topicalization is considered to be a weakest crossover configuration then Greek CLLD confirms the prediction as well. 34 Tsimpli (1999) argues in the spirit of Lasnik and Stowell (1991) that clitics in Greek can only be licensed when bound by a non-quantificational antecedent (if I understand correctly they are the equivalents of null epithets). The proposal collapses though when Tsimpli argues that cases like (i) actually instantiate coreference and not binding.

i. Kathe pedhi i ajapai ti mitera tu i every boy loves the mother his ‘Every boy loves his mother’ However, this cannot be so. Even in the logical syntax conception of binding where coreference is possible between a pronoun and a variable bound by kathe pedhi , as in (ii), ii. kathe pedhi (λx (x ajapai ti mitera tu)) and tu=x Rule I (its third clause) will always exclude coreference in this case since the interpretations produced from binding and coreference will never be distinguishable. 35 Cecchetto uses a variant of Torrego's (1992) analysis of clitic doubling. I will keep using Sportiche's analysis. For the failings of Cecchetto’s analysis wrt to crossover effects see section 2.3.

60 (168) a. * Ogni bambino i, sua i madre lo ama every boy his mother him loves ‘Every boy, his mother loves him’ b. * sua madre (λy (ogni bambino (λx ( y ama x)) and sua=ogni bambino

Both examples above are different from the corresponding Greek examples in (151) and (155) in that the subjects occupy preverbal positions in the former but postverbal positions in the latter. Thus, the pronoun in the Italian examples lies outside of the domain of the clitic. If the dislocated NP was base- generated in the TopP, then it would be able to bind sua in (168). Since it doesn't the λ-predicate must be formed lower, presumably in ClP. If a movement analysis is on the right track and a λ-predicate is formed after A-movement to Spec, ClP it is expected that the dislocated NP can bind a preverbal subject. The prediction is borne out.

(169) a. Ogni bambino i, lo ama sua i madre every boy him loves his mother 'Every boy, his mother loves him' b. ogni bambino (λx (x madre ama x))

That the distinction between (168) and (169) is real and that the position of the pronoun is relevant, is confirmed by the ungrammaticality of (170) where the pronoun is embedded in the subject of the matrix clause.

(170) a. * Ogni bambino i, sua i madre pensa che lo i ama Helena every boy his mother thinks that him loves Helen ‘Every boy, his mother thinks that Helen loves him’ b. * sua madre pensa che ogni bambino (λx (Helena ama x)) and sua=ogni bambimo

The equivalent with a referential NP in (171) is, as expected, grammatical, since coreference is possible.

(171) a. Gianni i, sua i madre pensa che lo i ama Helena John his mother thinks that him loves Helen ‘John, his mother thinks that Helen loves him’ b. sua madre pensa che Gianni (λx (Helena ama x)) and sua=Gianni

The grammaticality of (170) is also an independent argument against Ceccheto's analysis of clitic doubling. Recall that in his analysis the dislocated NP A'-moved out of a 'big DP' in argument position directly to the left-peripheral position (the big DP would then perform A-movement to the IP domain carrying along the trace). The obviation of the WCO effect was attributed to the special kind of trace left after movement out of a 'big DP' (a trace of the null epithet kind). But, if, as in Cecchetto's

61 analysis, the dislocated element binds pronouns within the clause from its left-peripheral position, then the contrast between the preverbal and the postverbal subject is unexpected; they both occupy positions within the c-command domain of ogni bambino , so they both should be equally able to be bound by it (again without any WCO violation because the trace of the dislocated NP is a null epithet). If, on the other hand, clitic doubling involves A-movement to a position above the VP but below the IP, as I have argued here, the contrast between preverbal and postverbal subjects follows. Note that no contrast between postverbal and preverbal subject appears in Greek, as can be seen in (172a-b). This exactly what we expect, if, as I have argued in the previous section, the dislocated NP in Greek is merged directly to the left peripheral position. 36

(172) a. Kathe andra i ton ajapai i mitera tu i every man him loves the mother his ‘Every man. his mother loves him’

b. Kathe andra i i mitera tu i ton ajapai every man the mother hishim loves ‘Every man, his mother loves him’

36 If it is true that Italian CLLD involves movement out of a clitic doubling configuration and Greek CLLD involves base-generation of the dislocated element, then CLLD should be dissociated from the availability of clitic doubling. As it is well known Italian does not allow clitic doubling configurations. Obviously, further research on what determines the movement/ base-generation parameter of dislocated structures needs to be done.

62 6 Conclusions

This thesis began with a very basic set of data that showed an unexpected absence of WCO effects in clitic constructions. Its aim was to provide an explanation for the obviation of the WCO effects by evaluating existing theories of binding and WCO. The exploration of the theories led to a simple view of WCO; WCO effects in wh-questions and cases of QR are to be treated similarly. All instances of A’-movement of an element with quantificational force will fail to license a pronoun as a bound variable; A’-movement doesn’t relocate all the relevant features for the formation of λ-predicate, which is the building stone for any binding relation, a property that can be reduced to more basic elements of the theory. In effect, WCO is induced because there is no ‘crossing over’ of the relevant features that would establish a binding relation. Although the mechanism that generates the WCO effect is uniform, the mechanisms that obviate it can vary, even if the relevant cases seem, at a first glance, to be very similar, as happens with the cases discussed in this thesis. WCO is absent when the relevant features that allow the creation of a λ-predicate do cross over the ‘intervening’ pronoun (when, e.g., the object A-moves in the IP domain, like in the case of clitic doubling) or when the relevant features are already in a position to c-command the pronoun (when, e.g., the antecedent is base generated in an A’-position, like in cases of resumption). It so happens that both options can be used in the Greek and Romanian data I have presented here and that both doubling and resumption require the presence of a clitic. But they do so for different reasons; clitic doubling requires a clitic head that licenses a certain property, specificity; a resumptive pro , on the other hand, needs to be identified and can only be so in a clitic doubling configuration. There are even cases where these two properties of clitics collapse, as in the case of CLLD in Greek, but it turns out that this is irrelevant from the obviation of the WCO effect. WCO effects do not disappear in clitic constructions BECAUSE clitics license specificity or BECAUSE resumptive pro s are identified in a clitic doubling configuration. WCO effects are obviated because clitic doubling forces the doubled element to undergo a step of A-movement and because resumption forces the antecedent to be merged in a position above the clitic. The answer to the question that naturally follows is pretty easy for the former case but not for the latter. What licenses clitic doubling? It is specificity, it is generally assumed. What licenses resumption? The question is well debated but it remains answered. In this thesis I do nothing more than follow the basic assumption that resumption is some kind of last resort strategy, used to escape island violations is a repeated theme in the literatue, used to create a bound variable reading, I argue here. Implementing this idea of resumption being a last resort strategy in the theory is far from being straightforward. I used one possible way (the global economy analysis) that seems fruitful. Of course more empirical research should be done on this matter.

63 ReferenceReferencessss

Agouraki G. 1993. Spec-Head Licensing: the Case of Foci, Clitic-Constructions and Polarity Items: a Case Study of Modern Greek. PhD Dissertation. University College London Alexandrova G. 1999. Nonquantificational WCO and the role of sublabel AGRO. Talk presented at the Third Conference on Formal Approaches to South Slavic and Balkan Languages . Plovdiv, University of Plovdiv. Alexiadou, A. and E.Anagnastopoulou 1997. Toward a uniform account of scrambling and clitic doubling. In A.Werner and E.van Gelderen. German: Syntactic Problems Problematic Syntax. Tubingen: Niemeyer. Alexiadou A. and E. Anagnostopoulou 1998. Parametrizing Agr: word order, verb- movement and EPP-checking. Natural Language and Linguistic Theory 16.3: 491-539. Alexiadou A. and C.Wilder 1998. Adjectival modification and multiple determiners. In A. Alexiadou and C. Wilder (eds). Possessors, Predicates and Movement in the DP . Amsterdam: Benjamins. Anagnostopoulou, Elena. 1994. Clitic Dependencies in Modern Greek. PhD Dissertation, Universität Salzburg. Anagnostopoulou, Elena. 1997. Clitic Left Dislocation and Contrastive Left Dislocation. In E. Anagnostopoulou, H. van Riemsdijk and F.Zwarts (eds). Materials in Left Dislocation . Amsterdam, Philadelphia: John Benjamins Publishing Company. Anagnostopoulou, Elena. 2003. The syntax of ditransitives: Evidence from clitics . Berlin/New York: Mouton de Gruyter. Anagnostopoulou E. Clitic doubling. To appear in the Syncom Project . Aoun, J. 2000. Resumption and Last Resort. D.E.L.T.A. 16: 13-43. Aoun J., L. Choueiri and N.Hornstein. 2001. Resumption, Movement, and Derivational Economy. Linguistic Inquiry 32: 371-403(33) Aoun J. and Y.-H.A. Li. 1989. Scope and constituency. Linguistic Inquiry 20:219251. Cecchetto C. 2000. Doubling Structures and Reconstruction. Probus 12, 93-126. Chomsky N. 1981. Lectures on Government and Binding . Dordrecht: Foris. Chomsky N. 1995. The Minimalist Program . Cambridge, Mass.: MIT Cinque G. 1990. Types of A-bar Dependencies . Cambridge, MA: MIT Press. Demirdache H. 1991. Resumptive Chains in Restrictive Relatives, Appositives and Dislocation Structures . Doctoral dissertation. MIT, Cambridge. Dobrovie-Sorin C. 1990. Clitic Doubling, Wh-Movement, and Quantification in Rumanian. Linguistic Inquiry 21, 351-397. Enç M. 1991. The semantics of specificity. Linguistic Inquiry 22: 1-25. Fodor J.D. and I.Sag 1982. Referential and quantificational indefinites. Linguistics and Philosophy 5:355-398. Fox D. 1995. Economy and scope. Natural Language Semantics 3: 283-341. Fox D. 1998. Economy and Semantic Interpretation . Cambridge,Mass.: MIT Press

64 Grodzinsky Y. and T. Reinhart 1993. The innateness of binding and coreference. Linguistic Inquiry 24(1), 69-101. Grohmann K. 2000. Copy Left Dislocation. In R.Billerey and B. Lillehaugen (eds). Proceedings of the Nineteenth West Coast Conference on Formal Linguistics . Somerville, MA: Cascadilla Press, 139-152. Hinterhoelzl R. and D.Pili. 2003. Argument shift phenomena across language types. ms. Humboldt Universitaet zu Berlin and Università di Siena. Hornstein N. 1995. Logical Form: From GB to Minimalism . Oxford: Blackwell. Iatridou S. 1991. Clitics and Island Effects. ms. MIT Koopman H. and D.Sportiche 1982. Variables and the Bijection Principle. The Linguistic Review 2:139-160. Lasnik H. and T.Stowell. 1991. Weakest crossover. Linguistic Inquiry 22:687-720. McCloskey J. Resumption. To appear in the Syncom Project . Partee B. 1972. Opacity, coreference and pronouns. In D.Davidson and G.Harman (eds). Semantics of Natural Language . Reidel, Dordrecht. Pesetsky D. 1987. Wh-in situ: Movement and Unselective Binding. In A.G.B. ter Meulen and E. Reuland (eds). The Representation of (In) . Cambridge, MA: MIT Press, 98-129. Reinhart T. 1983. Anaphora and semantic interpretation . London: Croom Helm. Reinhart T. 1997. Quantifier-Scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335-397 Reinhart, T. 2000. Strategies of anaphora resolution. In H.Bennis, M.Everaert and E.Reuland (eds). Interface strategies . Amsterdam: Royal Academy of Arts and Sciences. Reuland E. 1998. Structural conditions on chains and binding. In NELS 28, 341- 356. GLSA, University of Massachusetts, Amherst. Rizzi L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501-557. Ruys E.G. 2000. Weak crossover as a scope phenomenon, Linguistic Inquiry 31: 513-539 Ruys E.G. 2004. A Note on weakest crossover. Linguistic Inquiry 35: 124-140 Safir K. 1984. Multiple variable binding. Linguistic Inguiry 15: 603-638 Safir K. 1996. Derivation, representation and resumption: the domain of weak crossover. Linguistic Inquiry 27: 313-339. Sportiche D. 1992. Clitic constructions. ms. UCLA Sportiche D. 2003. Reconstruction, Binding and Scope. ms. UCLA Suner M. 1988. The role of Agreement in clitic-doubled constructions. Natural Language and Linguistic Inquiry 6: 391-434. Tsimpli M.-I. 1999. Null operators, clitics and identification. In A.Alexiadou, G.Horrocks and M.Stavrou (eds). Studies in Greek Syntax . Dordrecht: Kluwer Academic Publishers. Torrego E. 1992. Case and argument structure. ms. U/Mass, Boston.

65