Abstract

In this article, we discuss two types of cooccurrence restrictions involving reflexive clitics in European Portuguese and examine their implications for obligatory control. We argue that these restrictions may shed some light on where the controller is generated, thus making it possible to empirically test three Minimalist approaches to control: the predicate attraction approach (see Manzini and Roussou 2000), the PRO-based approach (e.g., Chomsky and Lasnik 1993, Landau 2000, 2004, Martin 2001), and the movement approach (e.g., Hornstein 1999, 2001, Boeckx, Hornstein, and Nunes 2010). We show that none of the approaches is able to capture all the relevant data if pursued under a strong lexicalist perspective such as Chomsky’s (1993, 2000) and that only the movement approach can account for all the data in a uniform way under Chomsky’s (2001) weak lexicalist perspective.

1 Introduction

In the last two decades, the syntax of control has been the object of a rich and at times heated discussion within Minimalism, as figuring out what might be the best analysis of control has important theoretical consequences. This article aims at contributing to this debate by examining new empirical phenomena that may help us choose among three of the most prominent approaches to control within Minimalism, namely, Manzini and Roussou’s (2000) predicate attraction account, PRO-based accounts (e.g., Chomsky and Lasnik 1993, Landau 2000, 2004, Martin 2001), and movement accounts (e.g., Hornstein 1999, 2001, Boeckx, Hornstein, and Nunes 2010), as respectively sketched in (1a–c).

(1)

graphic

Leaving aside a detailed discussion of the approaches’ technical implementations, we will focus on two of their major architectural differences. The first one concerns the number of DPs required to encode a control relation. In Manzini and Roussou’s (2000) predicate attraction approach, a single DP (which corresponds to the controller in the other approaches) is involved: it is merged where it surfaces and attracts the relevant θ-features of both the matrix and the embedded predicate at LF (see (1a)). By contrast, the other approaches resort to two DPs, each of which occupies a θ-position at some point in the derivation. The second major difference is related to the nature of the controllee and distinguishes between these two last approaches: it is a lexical formative (PRO) under a PRO-based approach (see (1b)) and a trace/copy in a movement approach (see (1c)).

The data to be examined below involve two types of cooccurrence restrictions affecting reflexive clitics in European Portuguese, which we will refer to as identity avoidance.1 The first case, which to our knowledge has not been noticed in the literature, involves deletion of a reflexive clitic within the complement of perception and causative verbs when the embedded subject is an identical clitic, as illustrated in (2).2

(2)

  • a.

    A Maria viu-me sentar-(*me) naquele banco.

    the Maria saw-me sit-REFL1SG on.that bench

    ‘Maria saw me sit on that bench.’

  • b.

    A Maria ouviu-te lamentar-(*te).

    the Maria heard-you.SG lament-REFL2SG

    ‘Maria heard you lamenting.’

  • c.

    O barco balançou e o João sentiu-se desequilibrar-(*se).

    the boat lurched and the João felt-REFL3SG lose.balance-REFL3SG

    ‘The boat lurched and João felt himself lose his balance.’

  • d.

    O João fez-nos encontrar-(*nos) com o Pedro.

    the João made-us meet-REFL1PL with the Pedro

    ‘João made us meet with Pedro.’

The intrinsic interest of data such as (2a–d) to the current debate on control is that the identity avoidance effects they display are computed in a local domain. Sentences such as (3), for example, where the identical clitics are separated by a CP boundary, do not give rise to an identity avoidance effect, for the lower instance of me cannot be deleted.

(3)

  • Eu pergunto-me se vou arrepender-*(me) depois.

  • I ask-REFL1SG if go repent-REFL1SG after

  • ‘I wonder if I’m going to regret it later.’

As object control constructions may also give rise to identity avoidance effects between the controller and a reflexive in the embedded clause, as illustrated in (4), one is led to expect that the controller and the reflexive in (4) should also be in a local configuration, comparable to that of (2). Identity avoidance effects such as the ones in (4) thus set up a scenario for us to examine the empirical coverage of the approaches to control entertained here. As each of them generates the controller in a different position (see (1)), they may make different predictions about whether or not an identity avoidance effect should obtain.

(4)

  • a.

    Foi a mãe que te convenceu a pôr-(*te) em pé e tentar andar?

    was the mother that you convinced to put-REFL2SG on foot and try walk

    ‘Was it Mom that convinced you to get on your feet and try to walk?’

  • b.

    Os professores autorizaram-nos a sentar-(*nos) naquele banco.

    the teachers authorized-us to sit-REFL1PL on.that bench

    ‘The teachers allowed us to sit on that bench.’

  • c.

    O médico teve de obrigar-te a deitar-(*te) na maca.

    the doctor had of force-you to lie-REFL2SG in.the stretcher

    ‘The doctor had to force you to lie down on the stretcher.’

The second case of identity avoidance we will discuss involves the cooccurrence restriction between the indefinite clitic se and the reflexive clitic se, as illustrated in (5).3

(5)

  • *Levanta-se-se cedo neste país.

  • raises-SEREFL-SEIND early in.this country

  • ‘One gets up early in this country.’

(5) differs from (2) (and (4)) in that deletion of the lower instance of the clitic is not licensed as a repair strategy to allow the structure to comply with identity avoidance. Thus, deletion of the lower instance of se leads to a grammatical sentence in (2c), for instance, but not in (5). The sentence in (6), for example, is fully grammatical, but can only be interpreted as having a referential null subject and a reflexive object; an indefinite interpretation for the subject, which should obtain if deletion of the reflexive clitic in (5) were allowed, is totally excluded.

(6)

  • Levanta-se cedo neste país.

  • raises-SEREFL early in.this country

  • ‘He/She gets up early in this country.’/*‘One gets up early in this country.’

The type of identity avoidance illustrated in (5) also brings interesting locality issues to light. On the one hand, it appears to behave like the first type (see (3)) in that a CP boundary between the identical clitics prevents an identity avoidance effect from arising. (7), for example, is grammatical despite the cooccurrence of indefinite se with reflexive se in the whole sentence.

(7)

  • Soube-se ter-se ele suicidado.

  • knew-SEIND have-SEREFL he committed.suicide

  • ‘It was heard that he committed suicide.’

On the other hand, when control is at stake, the two types of identity avoidance effects seen in (2) and (5) do not always go hand in hand. In (8a), for instance, indefinite se is the controller of the subject control verb and triggers an identity avoidance effect with respect to reflexive se in the embedded clause. By contrast, the upper instance of me in (8b) is also the controller of a subject control verb, but does not induce an identity avoidance effect with respect to the embedded reflexive clitic, as deletion of the lower instance of me is blocked. Thus, this contrast raises the question of why the same control configuration gives rise to different results depending on the type of identity avoidance considered.

(8)

  • a.

    *Quer-se sentar-se (e não se pode).

    wants-SEIND sit-3SGREFL and not SEIND can

    ‘One wants to sit down but can’t.’

  • b.

    Ele fez-me tentar sentar-*(me) de outra maneira.

    he made-me try sit-REFL1SG of other manner

    ‘He made me try to sit down in another way.’

Before we start the discussion proper, we would like to make it clear that our goal is not to provide a specific analysis of identity avoidance itself 4 or to discuss the specific technical details of each approach to control entertained here. What we will do is to use identity avoidance effects in European Portuguese as diagnostics of the relevant configurations that an empirically adequate theory of control should yield. For this purpose, it suffices to rely on the major architectural differences underlying each approach sketched in (1), such as the number and the nature of DPs involved in a control relation.

The article is organized as follows. In section 2, we show how (2) and (5) can be accounted for in a phase-based system (see Chomsky 2000, 2001). In section 3, we show that a phase-based analysis of control constructions involving potential configurations of identity avoidance is unable to capture all the data if pursued under a strong lexicalist approach such as Chomsky’s (1993, 2000). In particular, the movement approach undergenerates in some cases and the predicate attraction and PRO-based approaches overgenerate in others. When the three approaches are implemented under Chomsky’s (2001) weak lexicalist approach instead, the movement approach is able to overcome its undergeneration problem, but the competing analyses still overgenerate. Thus, the overall conclusion, presented in section 4, is that the movement theory of control under a weak lexicalist approach is in better shape to handle identity avoidance effects in European Portuguese.

2 Identity Avoidance in European Portuguese and Phase-Based Computations

The ungrammaticality of (9a) (= (5)) along with (9b) shows that the cooccurrence restriction in (9a) is not simply a matter of adjacency,5 for (9b) displays an identity avoidance effect even though the identical clitics are not contiguous.6 Similar considerations apply to the type of identity avoidance effect illustrated in (10): the reflexive is deleted even though it is not adjacent to the upper instance of me.

(9)

  • a.

    *Levanta-se-se cedo neste país.

    raises-SEREFL-SEIND early in.this country

    ‘One gets up early in this country.’

  • b.

    *Vai-se levantar-se cedo amanhã.

    goes-SEIND raise-SEREFL early tomorrow

    ‘People are going to get up early tomorrow.’

(10)

  • A Maria viu-me desequilibrar-(*me).

  • the Maria saw-me lose.balance-REFL1SG

  • ‘Maria saw me lose my balance.’

At first sight, the contrasts between (9) and (11) (= (7)) and between (10) and (12) seem to suggest that identity avoidance computations are clause-bounded: an identity avoidance effect may arise when the relevant clitics are within a single clause (see (9) and (10)), but not when they are located in different clauses (see (11) and (12)).

(11)

  • Soube-se ter-se ele suicidado.

  • knew-SEIND have-SEREFL he committed.suicide

  • ‘It was heard that he committed suicide.’

(12)

  • Eles queixaram-se de que teriam de inscrever-*(se) no curso.

  • they complained-SEREFL of that would.have of register-SEREFL in.the course

  • ‘They complained that they would have to register for the course.’

However, upon closer inspection the situation is not so simple. It is completely clear that the identical clitics in (11) and (12) are generated and remain in different clauses. But the surface word order in (10), with enclisis of the upper clitic to the matrix verb, is somewhat misleading, for it masks the fact that the upper clitic does move to the matrix clause, as extensively argued in the literature on this type of ECM construction (see, e.g., Gonçalves 1999, Martins 2000). This becomes transparent when the matrix clause contains elements that trigger proclisis, such as negation, as illustrated in (13). Notice that in (13) the two reflexives are unmistakably in different clauses and deletion is triggered, nonetheless. Thus, the contrast between (10) and (13), on the one hand, and (12), on the other, shows that a simplistic monoclausal/multiclausal distinction will not do.

(13)

  • O João não se sentiu desequilibrar-(*se).

  • the João not SEREFL felt lose.balance-SEREFL

  • ‘João did not feel himself lose his balance.’

A more promising approach may be built on the basis of Chomsky’s (2000, 2001) notion of phase. The basic assumption in a phase-based model is that the computational system does not feed the phonological component with the whole structure at once; rather, it transfers chunks of the structure under construction at designated derivational points. For concreteness, we assume that the relevant strong phases for our discussion are vP and CP and that Transfer proceeds in consonance with the Phase Impenetrability Condition (PIC), as defined in (14). Given (14), Transfer applies to the complement of a strong phase head when another strong phase head is added to the structure.

(14)

  • Phase Impenetrability Condition

  • The domain of H [the head of the strong phase HP] is not accessible to operations at ZP [the smallest strong phase dominating HP]; only H and its edge are accessible to such operations. (Chomsky 2001:14)

Let us then consider the effects of (14) for the data in (9)–(13), bearing in mind that a crucial property of reflexive clitics in European Portuguese is that they behave like other clitics in Romance in that they do not remain in situ, but move and adjoin to a higher Infl head.7 Take, for instance, the simplified derivation of the monoclausal sentence in (9b), as sketched in (15).8

(15)

  • a.

    [vPSEIND v [VP raise SEREFL early tomorrow]]

  • b.

    [TP goes-SEIND raise-v-SEREFL [vPtSE.IND tv [VPtraise tSE.REFL early tomorrow]]]

  • c.

    [CP C [TP goes-SEIND raise-v-SEREFL [vPtSE.IND tv [VPtraise tSE.REFL early tomorrow]]]]

  • d.

    Transfer of VP: [VPtraise tSE.REFL early tomorrow] → OK

  • e.

    Transfer of CP: [CP C [TP goes-SEIND raise-v-SEREFL [vPtSE.IND tv [VP ∆]]]]→*

After the vP phase in (15a) is formed, the light verb, the main verb, the subject clitic, and the object clitic all move to the TP domain (see (15b)). When C is introduced into the derivation (see (15c)), the complement of the lowest phase head (namely, VP) is transferred, as shown in (15d). Finally, when the matrix CP is transferred, the two clitics induce a nonrepairable identity avoidance effect (see (9b)).

In turn, the derivation of (10) proceeds along the lines of (16).

(16)

  • a.

    [vPCL.1SGi v [VP lose.balance CL.1SGk]]

  • b.

    [TPCL.1SGi lose.balance-v-CL.1SGk [vPCL.1SGitv [VPtlose.balanceCL.1SGk]]]

  • c.

    [vP saw [TPCL.1SGi lose.balance-v-CL.1SGk [vPCL.1SGitv [VPtlose.balanceCL.1SGk]]]]

  • d.

    [vP v [vP saw [TPCL.1SGi lose.balance-v-CL.1SGk [vPCL.1SGitv [VPtlose.balanceCL.1SGk]]]]]

  • e.

    Transfer of lower VP: [VPtlose.balanceCL.1SGk] → OK

  • f.

    [TP Maria saw-v-CL.1SGi [vPtMaria tv [VPtsaw [TPCL.1SGi lose.balance-v-CL.1SGk [vPCL.1SGitv [VP ∆]]]]]]

  • g.

    [CP C [TP Maria saw-v-CL.1SGi [vPtMaria tv [vPtsaw [TPCL.1SGi lose.balance-v-CL.1SGk [vPCL.1SGitv [VP ∆]]]]]]]

  • h.

    Transfer of higherVP: [VPtsaw [TPCL.1SGi lose.balance-v-CL.1SGk[vPCL.1SGitv [VP ∆]]]] → *

The first application of Transfer takes place when the matrix v is introduced in the derivation, targeting the lower VP (see (16d–e)). Given that the transferred material only contains one instance of the first person singular clitic (see (16e)), identity avoidance is not at stake. In contrast, when the matrix C is added to the computation (see (16g)) and the matrix VP is transferred (see (16h)), there are three instances of the clitic: two copies of the embedded subject chain (CL.1SGi) and one copy of the embedded object chain (CL.1SGk). Even if Chain Reduction (see Nunes 1995b, 2004) applies to the embedded subject chain and deletes the lower copy, we are still left with two instances of the first person clitic, which gives rise to an identity avoidance violation unless the reflexive is deleted (see (10)).

Finally, the grammaticality of (11) and (12) despite the cooccurrence of two instances of se is accounted for, as there is no derivational step where (copies of) both clitics are simultaneously transferred to the phonological component. Take the derivation of (11), for instance. As shown in (17), the copies of reflexive se are transferred when the embedded C (see (17b–c)) and the matrix v (see (17d–e)) are introduced, whereas the copies of indefinite se are only transferred when the matrix CP is transferred (see (17i)). As in other standard instances of movement, Chain Reduction deletes the lower copy of SEIND in (17i) and identity avoidance is not an issue. In other words, there is no application of Transfer in consonance with the PIC in the derivation of (11) that could give rise to an identity avoidance effect.

(17)

  • a.

    [vP he v [VP committed.suicide SEREFL]]

  • b.

    [CP C [TP he has-SEkREFL committed.suicide-v [vPthe tv [VPtcommitted.suicideSEkREFL]]]]

  • c.

    Transfer of lower VP: [VPtcommitted.suicideSEkREFL] → OK

  • d.

    [vP v [VP knew [CP C [TP he has-SEkREFL committed.suicide-v [vPthe tv [VP ∆]]]]]]

  • e.

    Transfer of lower TP: [TP he has-SEkREFL committed.suicide-v [vPthe tv [VP ∆]]] → OK

  • f.

    [TPSEiIND knew-v [vPSEiINDtv [VPtknew [CP C [TP ∆]]]]]

  • g.

    [CP C [TPSEiIND knew-v [vPSEiINDtv [VPtknew [CP C [TP ∆]]]]]]

  • h.

    Transfer of matrix VP: [VPtknew [CP C [TP ∆]]] → OK

  • i.

    Transfer of matrix CP: [CP C [TPSEiIND knew-v [vPSEiINDtv [VP ∆]]]] → OK

Having seen that the PIC provides the right domains for the computation of identity avoidance, let us get back to the discussion of identity avoidance and control.

3 Identity Avoidance and Obligatory Control in European Portuguese

3.1 Basic Facts

As mentioned in section 1, object control constructions may give rise to identity avoidance effects with respect to the controller and a reflexive in the embedded clause, triggering deletion of the latter, as illustrated in (18) (= (4)).

(18)

  • a.

    Foi a mãe que te convenceu a pôr-(*te) em pé e tentar andar?

    was the mother that you convinced to put-REFL2SG on foot and try walk

    ‘Was it Mom that convinced you to get on your feet and try to walk?’

  • b.

    Os professores autorizaram-nos a sentar-(*nos) naquele banco.

    the teachers authorized-us to sit-REFL1PL on.that bench

    ‘The teachers allowed us to sit on that bench.’

  • c.

    O médico teve de obrigar-te a deitar-(*te) na maca.

    the doctor had of force-you to lie-REFL2SG in.the stretcher

    ‘The doctor had to force you to lie down on the stretcher.’

By contrast, textbook examples of subject control constructions do not yield identity avoidance effects involving the controller and the reflexive, for an obvious reason: there are no identical clitics in the relevant configurations, as illustrated in (19).

(19)

  • a.

    Ele tentou levantar-*(se).

    he tried raise-REFL3SG

    ‘He tried to get up.’

  • b.

    Eu quero sentar-*(me) naquele banco.

    I want sit-REFL1SG on.that bench

    ‘I want to sit on that bench.’

  • c.

    Tu paraste de levantar-*(te) tarde?

    you.SG stopped of raising-REFL2SG late

    ‘Did you stop getting up late?’

Accordingly, in the counterparts of (18) where the object control verb is passivized and the controller is licensed with nominative Case, no identity avoidance effect is observed either, as shown in (20).

(20)

  • a.

    Tu foste convencido a pôr-*(te) em pé e andar?

    you were convinced to put-REFL2SG on foot and walk

    ‘You were convinced to get on your feet and walk?’

  • b.

    Nós fomos autorizados a sentar-*(nos) naquele banco.

    we were authorized to sit-REFL1PL on.that bench

    ‘We were allowed to sit on that bench.’

  • c.

    Tu tiveste de ser obrigado a deitar-*(te) na maca.

    you had of be forced to lie-REFL2SG in.the stretcher

    ‘You had to be forced to lie down on the stretcher.’

However, there are more complex cases of subject control that could potentially give rise to an identity avoidance effect (see, e.g., (8b)). This is the case of the sentences in (21) and (22), for example, where the controller is embedded under an ECM verb. Interestingly, this is a configuration where the reflexive cannot be deleted.

(21)

  • a.

    Ele fez-me tentar levantar-*(me) mais cedo.

    he made-me try raise-REFL1SG more early

    ‘He made me try to get up earlier.’

  • b.

    A Maria mandou-nos parar de levantar-*(nos) tarde.

    the Maria ordered-us stop of raise-REFL1PL late

    ‘Maria told us to stop getting up late.’

(22)

  • A Maria viu-me cair após ter-*(me) levantado da cadeira.

  • the Maria saw-me fall after have-REFL1SG raised from.the chair

  • ‘Maria saw me fall down after having gotten up from the chair.’

As we showed in section 2, identity avoidance computations are ultimately subject to the PIC in that two elements can only give rise to an identity avoidance effect if they are part of the same constituent that undergoes Transfer. That being so, the contrast between the object control constructions in (18), where deletion of the reflexive clitic is permitted, and the subject control constructions in (21) and (22), where deletion is disallowed, should be taken to indicate that the reflexive clitics are transferred at the same derivational steps as the controller clitics in (18), but not in (21) and (22).

Having this in mind, let us examine whether the three approaches to control entertained here (see (1)) can yield this result.

3.2 To Be or Not to Be a Strong Phase: That Is a Crucial Question

Let us start by examining the derivation of the object control construction in (23) (= (18b)) in more detail.

(23)

  • Os professores autorizaram-nos a sentar-(*nos) naquele banco.

  • the teachers authorized-us to sit-REFL1PL on.that bench

  • ‘The teachers allowed us to sit on that bench.’

In order to determine the points where Transfer applies, we first need to identify the relevant strong phases of (23). That the matrix CP, the matrix vP, and the embedded vP should all count as strong phases is not controversial. By contrast, the phasehood nature of the embedded CP is less obvious and may hinge on some theory-internal assumptions. Under the null Case approach to control (see, e.g., Chomsky and Lasnik 1993, Martin 2001), for example, the Case/agreement relations involving PRO are determined CP-internally, which renders the embedded CP in (23) a strong phase. On the other hand, approaches based on Agree (see, e.g., Landau 2000, 2004), predicate attraction (Manzini and Roussou 2000), or movement (see, e.g., Hornstein 1999, 2001, Boeckx, Hornstein, and Nunes 2010) all must (tacitly) assume that the embedded CP in (23) is not a strong phase, as it is transparent to A-relations involving Agree or movement. Without getting into the merits of each option, it is worth noting that they do make different empirical predictions with respect to the identity avoidance effect observed in (23).

Consider the null Case analysis of (23), as sketched in (24).

(24) PRO-based account/Null Case implementation

  • a.

    [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VPtsitCL.1PLi on that bench]]]]

  • b.

    Transfer of lower VP: [VPtsitCL.1PLi on that bench] → OK

  • c.

    [vPCL.1PLk [ authorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VP ∆]]]]]]

  • d.

    [vP v [VPCL.1PLk [ authorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VP ∆]]]]]]]

  • e.

    Transfer of TP: [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VP ∆]]] → OK

Given that the embedded C is a strong phase head under the null Case approach, it triggers application of Transfer when it is introduced (see (24a–b)). Transfer then applies to the complement of this C when the matrix v is inserted (see (24d–e)). Crucially, each application transfers a copy of the reflexive, but neither affects the clitic that sits in the matrix Spec,VP. Hence, when copies of this clitic are later transferred in the derivation, the copies of the reflexive are long gone and identity avoidance is not an issue. Therefore, the null Case approach makes the incorrect prediction that (23) should not yield an identity avoidance effect.

By contrast, consider the Agree-based analysis of (23) given in (25), for instance.

(25) PRO-based account/Agree implementation

  • a.

    [vPCL.1PLk [ authorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VPtsitCL.1PLi on that bench]]]]]]

  • b.

    [vP v [vPCL.1PLk [ authorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VPtsitCL.1PLi on that bench]]]]]]]

  • c.

    Transfer of lower VP: [VPtsitCL.1PLi on that bench] → OK

  • d.

    [TP the teachers authorized-v-CL.1PLk [vPtthe teachers tv [VPCL.1PLk [tauthorized [CP C [TP PRO to sit-v-CL.1PLi [VPtPRO tv [VP ∆]]]]]]]]

  • e.

    [CP C [TP the teachers authorized-v-CL.1PLk [vPtthe teachers tv [vPCL.1PLk [v'tauthorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VP ∆]]]]]]]]]

  • f.

    Transfer of matrix VP: [VPCL.1PLk [tauthorized [CP C [TP PRO to sit-v-CL.1PLi [vPtPRO tv [VP ∆]]]]]] → *

Given that the embedded C is not a strong phase in the Agree implementation of the PRO-based approach to control, the first application of Transfer will only take place when the matrix v is introduced (see (25b–c)). The transferred VP in (25c) only contains one instance of the clitic and no issue of identity avoidance arises. Later, when the matrix C is inserted (see (25e)), the matrix VP is transferred (see (25f)). As it contains two instances of the clitic (not related by movement), an identity avoidance configuration arises and deletion must come to the rescue (see (23)).

In sum, the identity avoidance effect displayed by (23) allows us to empirically distinguish two different implementations of PRO-based approaches to control: the Agree-based implementation correctly accounts for the identity avoidance effect in (23), but the null Case implementation doesn’t. As for the predicate attraction and the movement approaches, they side with the Agree-based analysis in (25) in assuming that the embedded CP is not a strong phase. Thus, they also resort to applications of Transfer in a way parallel to (25), only differing in their instrinsic aspects such as the number of elements involved in a control relation and the nature of the controllee. The structure of the transferred matrix VP, for instance, is analyzed along the lines of (26).

(26)

  • a.

    Predicate attraction approach

    [vPCL.1PLk [tauthorized [CP C [TP to sit-v-CL.1PLi [vPtv [VP ∆]]]]]] → *

  • b.

    Movement approach

    [VPCL.1PLk [tauthorized [CP C [TPCL.1PLk to sit-v-CL.1PLi [vPCL.1PLktv [VP ∆]]]]]] → *

Under the predicate attraction approach in (26a), the specifiers of the embedded TP and the embedded vP remain empty, as the upper instance of the clitic is generated in the matrix Spec,VP and attracts the θ-role of the embedded predicate. In turn, under the movement approach in (26b), the upper instance of the clitic is generated in the embedded Spec,vP and successively moves to the matrix Spec,VP, leaving copies behind. These differences aside, in either approach the VP transferred contains a clitic identical to the reflexive, yielding an identity avoidance effect.

3.3 The Puzzle

We have shown that identity avoidance effects in simple object control constructions pose problems to the null Case implementation of the PRO-based approach to control, but are adequately handled by the alternative Agree-implementation, predicate attraction, and movement accounts. Let us now examine how these competing approaches fare with respect to more complex data, by considering the (simplified) structure of the first transferred constituent in the derivation of (27) (= (21a)), as shown in (28).9

(27)

  • Ele fez-me tentar levantar-*(me) mais cedo.

  • he made-me try raise-REFL1SG more early

  • ‘He made me try to get up earlier.’

(28)

  • a.

    Predicate attraction approach

    [vP v [VP fez [TP mei [vP v [VP tentar [CP levantar-mek cedo]]]]]]

    Transfer of the intermediate VP: [VP tentar [CP levantar-mek cedo]] → OK

  • b.

    PRO-based approach/Agree implementation

    [vP v [VP fez [TP mei [vP mei v [VP tentar [CP PRO levantar-mek cedo]]]]]]

    Transfer of the intermediate VP: [VP tentar [CP PRO levantar-mek cedo]] → OK

  • c.

    Movement approach

    [vP v [vP fez [TP mei [vP mei v [VP tentar [CP mei levantar-mek cedo]]]]]]

    Transfer of the intermediate VP: [VP tentar [CPmei levantar-mek cedo]] → *

Once the embedded CP does not qualify as a strong phase under these approaches, as discussed in section 3.2, the first application of Transfer takes place when the matrix v is merged, targeting the VP headed by the control verb. In the predicate attraction (see (28a)) and PRO-based (see (28b)) approaches, the transferred VP only contains the reflexive clitic, whereas in the movement approach the reflexive cooccurs with a copy of the controller (see (28c)). Thus, the movement approach predicts an identity avoidance effect in (27), contrary to fact. The other approaches, on the other hand, correctly account for the fact that deletion of the reflexive is prohibited in (27), for (the copies of) the reflexive and the controller are transferred at different derivational steps and therefore identity avoidance is not at stake.

A similar conclusion is reached with respect to the adjunct control construction in (29), as indicated by the simplified structures in (30).

(29)

  • A Maria viu-me cair após ter-*(me) levantado da cadeira.

  • the Maria saw-me fall after have-REFL1SG raised from.the chair

  • ‘Maria saw me fall down after having gotten up from the chair.’

(30)

  • a.

    Predicate attraction approach

    [a Maria viu-mei [VP [VP cair] [adjunct island após ter-mek levantado . . . ]]] → OK

  • b.

    PRO-based approach/Agree implementation

    [a Maria viu-mei [VP [VP cair] [adjunct island após PRO ter-mek levantado . . . ]]] → OK

  • c.

    Movement approach

    [a Maria viu-mei [VP [VP cair] [adjunct island após mei ter-mek levantado . . . ]]] → *

Consider the subject position of the adjunct clause in each of the structures in (30): it is left empty in the predicate attraction approach, as the controller is base-generated in the upper clause and attracts the θ-role of the adjunct predicate (see (30a)); it is occupied by PRO in the PRO-based account (see (30b)); and it is occupied by a copy of the controller in the movement approach (see (30c)).10 Again, we have one instance of me within the adjunct in the predicate attraction and PRO-based analyses, but two instances in the movement account. In other words, the movement approach sets itself apart from the other approaches in that it incorrectly predicts an identity avoidance effect in (29).11

Interestingly, we find the opposite result when we examine identity avoidance involving the indefinite clitic se. Consider for instance the (simplified) structures each approach assigns to the data in (31) and (33), as respectively shown in (32) and (34) (see footnotes 9 and 11).

(31)

  • *Não se conseguiu sentar-se num bom sítio.

  • not SEIND managed sit-SEREFL in.a good place

  • ‘One did not find a good place to sit.’

(32)

  • a.

    Predicate attraction approach

    [CP C [TP não SEIND [vP v [VP conseguiu [CP sentar-SEREFL . . . ]]]]]

    Transfer of the matrix VP: [VP conseguiu [CP sentar-SEREFL . . . ]] → OK

  • b.

    PRO-based approach/Agree implementation

    [CP C [TP não SEiIND [vP SEi IND v [VP conseguiu [CP PRO sentar-SEREFL . . . ]]]]]

    Transfer of the matrix VP: [VP conseguiu [CP PRO sentar-SEREFL . . . ]] → OK

  • c.

    Movement approach

    [CP C [TP não SEiIND [vP SEiIND v [VP conseguiu [CP SEiIND sentar-SEREFL . . . ]]]]]

    Transfer of the matrix VP: [VP conseguiu [CPSEiIND sentar-SEREFL . . . ]] → *

(33)

  • *Gritou-se muito após ter-se levantado da cama.

  • screamed-SEIND much after have-REFL3SG raised from.the bed

  • ‘One screamed a lot after getting up from bed.’

(34)

  • a.

    Predicate attraction approach

    [gritou-SEIND . . . [adjunct island após ter-SEREFL levantado da cama]] → OK

  • b.

    PRO-based approach/Agree implementation

    [gritou-SEIND . . . [adjunct island após PRO ter-SEREFL levantado da cama]] → OK

  • c.

    Movement approach

    [gritou-SEiIND . . . [adjunct island após SEiIND ter-SEREFL levantado da cama]] → *

Recall that identity avoidance involving indefinite se does not trigger deletion as a rescue strategy and its effects are identified simply via the unacceptability of the relevant examples (see (5)–(6)). Thus, the unacceptability of (31) and (33) leads us to expect that there is a step in their derivation in which (copies of) reflexive se and indefinite se are transferred together, inducing an identity avoidance effect. This is the case under the movement analysis (see (32c)/(34c)), but not under the predicate attraction (see (32a)/(34a)) or PRO-based (see (32b)/(34b)) approaches.

In short, the movement account undergenerates with respect to the first type of identity avoidance effect, failing to predict that sentences such as (27) and (29) are only acceptable without deletion (see (28c) and (30c)). On the other hand, the predicate attraction and PRO-based approaches overgenerate with respect to the second type of avoidance effect, incorrectly predicting sentences such as (31) and (33) to be grammatical (see (32a)/(34a) and (32b)/(34b)).

This undesirable state of affairs appears to indicate that a different domain should be independently postulated for each type of identity avoidance: a smaller domain for the first type and a larger one for the second type. However, such an approach would miss the point that when control is not at stake, the two types of identity avoidance do pattern alike and are subject to a uniform phase-based analysis, as shown in section 2. In particular, when the relevant clitics are clearly transferred separately, no identity avoidance effect arises for either type of construction. In (35a) (= (3)), for example, deletion of the reflexive is not triggered and (35b) (= (11)) is not ungrammatical.

(35)

  • a.

    Eu pergunto-me se vou arrepender-*(me) depois.

    I ask-REFL1SG if go repent-REFL1SG after

    ‘I wonder if I’m going to regret it later.’

  • b.

    Soube-se ter-se ele suicidado.

    knew-SEIND have-SEREFL he committed.suicide

    ‘It was heard that he committed suicide.’

The task before us is therefore to determine what independent property distinguishes the two types of identity avoidance and why this yet-to-be-determined special property interacts with control in a way that it seems to affect the size of the domain where identity avoidance is to be computed. This is the goal of the next section.

3.4 A Single Domain but Different Feature Specifications

3.4.1 On the Difference between the Two Types of Identity Avoidance Effects.

Recall that the two types of identity avoidance affecting reflexive clitics in European Portuguese differ in that when indefinite se is involved, deletion of the reflexive is not available as a repair strategy (see (2) vs. (5)–(6)). This indicates that although phonological identity may be relevant for an identity avoidance effect to obtain, it does not suffice to license deletion.

We believe that this asymmetry stems from an independent difference between indefinite se and the other European Portuguese clitics. As in other Romance languages, in European Portuguese first and second person clitics do not have distinct forms for pronouns and reflexives or for accusative and dative. This syncretism can be interpreted as showing that first and second person clitics are underspecified with respect to these features, which makes them morphologically identical regardless of whether they are used as pronouns or as reflexives or whether they are licensed in dative or accusative configurations. That being so, it is not surprising that deletion is sanctioned in (36), even though in (36a) the upper instance of me is used as a pronoun and the lower instance as a reflexive or that in (36b) the upper instance is assigned dative, whereas the lower one is assigned accusative.

(36)

  • a.

    A Maria viu-me desequilibrar-(*me).

    the Maria saw-me lose.balance-REFL1SG

    ‘Maria saw me lose my balance.’

  • b.

    Custou-me a sentar-(*me) naquele banco.

    cost-me to sit-REFL1SG on.that bench

    ‘It was hard for me to succeed in sitting on that bench.’

By contrast, the standard assumption regarding indefinite se is that it is specified as being intrinsically nominative (see, e.g., Cinque 1988, Raposo and Uriagereka 1996, and D’Alessandro 2004 for relevant discussion). This property renders it morphologically distinct from reflexive se and, we would like to suggest, this is what blocks deletion. Evidence that indefinite se is intrinsically nominative is provided by data such as (37a–d). (37a) and (37b) respectively show that indefinite se cannot be licensed with accusative Case by a selecting verb or in an ECM configuration.12 (37c–d) further show that the problem with (37a) is not that indefinite se cannot be interpreted as an internal argument, for it is compatible with the internal argument of unaccusative (see (37c)) and passivized (see (37d)) verbs, provided that nominative Case is available.

(37)

  • a.

    *Eu vi-se.

    I saw-SEIND

    ‘I saw people.’

  • b.

    *Eu não se vi comprar as ferramentas.

    I not SEIND saw buy the tools

    ‘I didn’t see people buy the tools.’

  • c.

    Chegou-se tarde à festa.

    arrived.3SG-SEIND late to.the party

    ‘People arrived late at the party.’

  • d.

    Aqui não se é admitido sem boas recomendações.

    here not SEIND is admitted without good recommendations

    ‘Here one is not admitted without good recommendations.’

As we will show, this intrinsic specification of indefinite se as nominative not only sets it apart with respect to deletion as a repair strategy to circumvent identity avoidance, but also proves crucial for establishing the specific domain where identity avoidance is to be computed when control is involved.

3.4.2 Feature Value Specification and Identity Avoidance.

The discussion of section 3.4.1 invites us to reconsider the conflicting results found in section 3.2 with respect to how the approaches to control entertained here fare with respect to identity avoidance effects. Up to now, the discussion has tacitly proceeded under a strong lexicalist perspective, according to which lexical items enter the derivation fully inflected and have their feature specification appropriately checked in certain designated configurations (see, e.g., Chomsky 1993, 2000). The distinction between indefinite se and the other clitics in European Portuguese suggests that a weak lexicalist phase-based approach such as the one outlined in Chomsky 2001 may be better suited to account for the phenomena under investigation. The unmarked case under this view is for uninterpretable features to be unvalued as they enter the derivation and gain their values via agreement with valued features. However, in marked situations such as the Case feature of indefinite se, an uninterpretable feature may be inherently valued (see Pesetsky and Torrego 2007 for extensive discussion). In this situation, the inherently valued uninterpretable feature must still enter into an Agree relation with a feature of an appropriate head in order to be licensed (deleted for LF purposes).

Bearing these considerations in mind, let us reconsider the derivation of (38) (= (27)).

(38)

  • Ele fez-me tentar levantar-*(me) mais cedo.

  • he made-me try raise-REFL1SG more early

  • ‘He made me try to get up earlier.’

Both clitics in (38) enter the derivation with their interpretable features ( person and number) valued and their Case feature unvalued. The lower instance has its Case feature valued as accusaive after agreeing with the most embedded light verb, and the upper instance, after agreeing with the matrix light verb. This means that any copies of either clitic that were eventually left behind prior to these agreement relations had their Case features unvalued, which in turn has consequences for computations of identity avoidance. After all, it makes sense to assume that if an element is not fully valued, it cannot be identical to one that has all of its features valued; hence, any two such elements cannot trigger an identity avoidance effect.

That being so, the structure of the transferred VP in (28) should actually be as in (39).

(39)

  • a.

    Predicate attraction approach

    [VP tentar [CP levantar-CLk[1SG, Case:ACC] . . . ]] → OK

  • b.

    PRO-based approach/Agree implementation

    [VP tentar [CP PRO levantar-CLk[1SG, Case:ACC] cedo]] → OK

  • c.

    Movement approach

    [VP tentar [CPCLi[1SG, Case:u] levantar-CLk[1SG, Case:ACC] cedo]] → OK

As was the case in (28), the predicate attraction and PRO-based approaches correctly predict lack of an identity avoidance effect, for only the lower clitic is transferred. The relevant change concerns the movement approach. The controller is generated as the external argument of the most embedded clause and undergoes successive-cyclic movement to the intermediate Spec,vP, where it receives the 0-role of the controller, before it finally ends in a position where it agrees with the matrix light verb and has its Case feature valued as accusative. Crucially, the copies left behind—in particular, the copies left in the most embedded clause—do not have their Case features valued. Hence, contrary to what the discussion of (28c) led us to think, the transferred VP in (39c) cannot yield an identity avoidance effect: the upper instance of the clitic does not have its Case feature valued and cannot be taken as identical to the lower instance, which is valued as accusative.

The same considerations apply to the structures assigned to (40) (= (29)), as shown in (41). Even though the spelled-out island has two instances of the clitic under the movement approach to control in (41c) (see footnotes 10 and 11), the upper copy is not fully valued and therefore cannot yield an identity avoidance effect.

(40)

  • A Maria viu-me cair após ter-*(me) levantado da cadeira.

  • the Maria saw-me fall after have-REFL1SG raised from.the chair

  • ‘Maria saw me fall down after having gotten up from the chair.’

(41)

  • a.

    Predicate attraction approach

    [adjunct island após ter-CLk[1SG, Case:ACC] levantado . . . ] → OK

  • b.

    PRO-based approach/Agree implementation

    [adjunct island após PRO ter-CLk[1SG, Case:ACC] levantado . . . ] → OK

  • c.

    Movement approach

    [adjunct island após CLi[1SG, Case:u] ter-CLk[1SG, Case:ACC] levantado . . . ] → OK

The interim conclusion is that contrary to what the discussion in section 3.2 led us to believe, the movement approach does not undergenerate with respect to the first type of identity avoidance effect, as it correctly predicts that deletion of the reflexive should not be triggered in either (38) or (40). The problem regarding empirical coverage that the movement account sketched in section 2 faced stemmed from its strong lexicalist commitments. Once these commitments are abandoned and the valuation procedure of Chomsky 2001 is assumed, the analysis of the first type of identity avoidance changes, circumventing the earlier empirical problems (see (28c)/(39c) and (30c)/ (41c)).

On the other hand, the analysis of identity avoidance involving indefinite se by and large remains the same. If indefinite se inherently has all of its features valued, it should be computed basically in the same way, be the analysis lexicalist or not. That is, under Chomsky’s (2001) system, the relevant transferred constituents in (42) (= (31)) and (44) (= (33)) are to be analyzed as in (43) and (45), respectively.

(42)

  • *Não se conseguiu sentar-se num bom sítio.

  • not SEIND managed sit-SEREFL in.a good place

  • ‘One did not find a good place to sit.’

(43)

  • a.

    Predicate attraction approach

    [VP conseguiu [CP sentar-CL[3SG, REFL, ACC] . . . ]] → OK

  • b.

    PRO-based approach/Agree implementation

    [VP conseguiu [CP PRO sentar-CL[3SG, REFL, ACC] . . . ]] → OK

  • c.

    Movement approach

    [VP conseguiu [CPCLi[3SG, IND, NOM] sentar-CLk[3SG, REFL, ACC] . . . ]] → *

(44)

  • *Gritou-se muito após ter-se levantado da cama.

  • screamed-SEIND much after have-REFL3SG raised from.the bed

  • ‘One screamed a lot after getting up from bed.’

(45)

  • a.

    Predicate attraction approach

    [adjunct island após ter-CL[3SG, REFL, ACC] levantado da cama] → OK

  • b.

    PRO-based approach/Agree implementation

    [adjunct island após PRO ter-CL[3SG, REFL, ACC] levantado da cama] → OK

  • c.

    Movement approach

    [adjunct island após CLi[3SG, IND, NOM] ter-CLk[3SG, REFL, ACC] levantado da cama] → *

The three analyses in (43) and (45) are similar in that the indefinite clitic licenses its nominative Case after agreeing with the matrix T. However, they diverge with respect to the position where this clitic is first merged: in the matrix Spec,TP (or wherever it surfaces) under the predicate attraction approach, in the matrix Spec,vP under the PRO-based approach, and in the most embedded Spec,vP under the movement approach. This small detail makes all the difference, though. Given that the indefinite clitic is lexically specified as nominative, a given occurrence of this clitic can provide the system with the information that it is to be realized as se in the morphological component, regardless of whether or not its Case feature has been checked. Put differently, any occurrence of the indefinite clitic can induce an identity avoidance effect with respect to a third person reflexive clitic if they undergo Transfer together, as is the case with the representations in (43c) and (45c).

It is easy to see that the approach sketched above also applies to more complex data such as (46a–b), which apparently involve a long-distance computation of identity avoidance with respect to the indefinite and the reflexive.

(46)

  • a.

    *Conseguiu-se evitar sentar-se na última fila.

    managed-SEIND avoid sit-SEREFL in.the last row

    ‘One managed to avoid sitting in the last row.’

  • b.

    *Tentou-se conseguir evitar sentar-se na última fila.

    tried-SEIND manage avoid sit-SEREFL in.the last row

    ‘One tried to manage to avoid sitting in the last row.’

(46a–b) involve a subject control predicate embedded under another subject control predicate, with indefinite se as the controller. The fact that they are ungrammatical indicates that in both, indefinite se and the reflexive should not be too far apart so that they can be computed for purposes of identity avoidance. More concretely, (a copy of) the reflexive must be transferred together with (a copy of) the indefinite clitic. Under the movement account presented above, there is a copy of indefinite se in the most embedded clause and this copy can induce an identity avoidance effect with respect to the reflexive.

To sum up, if a strong lexicalist view is pursued, the three approaches to control make different empirical predictions, but none of them captures all the facts, as the movement theory undergenerates with respect to the first type of identity avoidance and the competing alternatives overgenerate with respect to the second type (see section 3.2). By contrast, if a phase-based weak lexicalist approach such as Chomsky’s (2001) is assumed, the movement approach makes the same prediction as the other approaches with respect to the first type and maintains its superior coverage in the case of the second type. By achieving the same degree of success as the competing approaches in accounting for the first type of identity avoidance, the movement approach now emerges as the only approach to control entertained here that can account for all the data involving the two types of identity avoidance. Importantly, it does so without resorting to different domains for computing identity avoidance. Even complex data such as (46a–b), which at first sight seem to require an extension of the relevant domain for computing identity avoidance, are amenable to a uniform phase-based account.

3.5 Further Evidence

The movement approach also accounts for two other related sets of facts. The first one involves object control with indefinite se. An object control sentence such as (47a), for instance, is to be excluded for the same reason a simple sentence such as (47b) is to be ruled out: namely, the indefinite clitic is intrinsically nominative and it cannot be licensed in the object position of an accusative-Case-assigning verb. When the object control verb is passivized, the internal argument now becomes compatible with the indefinite clitic, as shown in (48), because se can have its nominative Case specification licensed by the matrix T.

(47)

  • a.

    *Eu forcei-se a sair da sala.

    I forced-SEIND to leave of.the room

    ‘I forced people to leave the room.’

  • b.

    *Eu vi-se.

    I saw-SEIND

    ‘I saw people.’

(48)

  • Nada funciona quando não se é convencido a contribuir para o debate.

  • nothing functions when not SEIND is convinced to contribute to the debate

  • ‘Nothing works when one does not get convinced to contribute to the debate.’

That being so, the movement account predicts that a licit object control configuration with indefinite se as the controller should exhibit an identity avoidance effect with respect to a reflexive in the embedded clause. That this prediction is correct is illustrated by (49a), for instance, whose (simplified) structure under a movement analysis is given in (49b).

(49)

  • a.

    *Aqui não se é forçado a inscrever-se em todas as disciplinas.

    here not SEIND is forced to register-REFL3SG in all the courses

    ‘Here one is not forced to register for all courses.’

  • b.

    Movement approach

    [aqui não CLi[3SG, IND, NOM] é [VPCLi[3SG, IND, NOM] [V' forçado a [CLi[3SG, IND, NOM] inscrever-CLk[3SG, REFL, ACC] em todas as disciplinas]]]] → *

Interestingly, the identity avoidance effect persists even in more complex configurations such as (50a), where a subject control structure is embedded under an object control verb with se as the controller. Under the movement approach, the indefinite clitic is generated in the lowest clause, as illustrated in (50b), and the copy it leaves behind on its way to the matrix clause is close enough to induce an identity avoidance effect with respect to the reflexive clitic.

(50)

  • a.

    *Aqui não se é forçado a tentar levantar-se cedo.

    here not SEIND is forced to try raise-REFL3SG early

    ‘Here one is not forced to try to get up early.’

  • b.

    Movement approach

    [TP aqui não CLi[3SG, IND, NOM] é [VPCLi[3SG, IND, NOM] [V' forçado a [CLi[3SG, IND, NOM]

    tentar [CLi[3SG, IND, NOM] levantar-CLk[3SG, REFL, ACC] cedo]]]]] → *

The other set of related facts accounted for by the movement approach involves raising constructions. As we point out in Martins and Nunes 2005, the acceptability of indefinite se with specific raising verbs is subject to variation among European Portuguese speakers. What is relevant for our purposes is that the raising verbs that allow indefinite se exhibit identity avoidance effects with respect to a reflexive clitic in the embedded clause, as illustrated in (51) with the raising verb demorar ‘last’.

(51)

  • a.

    *Demorou-se muito a levantar-se da cama.

    lasted-SEIND much to raise-SEREFL from.the bed

    ‘It took us a long time to get up from bed.’

  • b.

    Movement approach

    [TPCLi[3SG, IND, NOM] demorou a [CLi[3SG, IND, NOM] levantar-CLk[3SG, REFL, ACC] da cama]]→*

Interestingly, the addition of a control structure intervening between the raising verb and the clause containing the reflexive, as in (52a), does not interfere with the identity avoidance effect exhibited by (51a). Similar to what we saw in (50a), movement of indefinite se leaves a copy in the most embedded clause, triggering an identity avoidance effect with respect to the reflexive.

(52)

  • a.

    *Demorou-se muito a tentar levantar-se da cama.

    lasted-SEIND much to try raise-SEREFL from.the bed

    ‘It took us a long time to try to get up from bed.’

  • b.

    Movement approach

    [TPCLi[3SG, IND, NOM] demorou a [CLi[3SG, IND, NOM] tentar [CLi[3SG, IND, NOM] levantar-

    CLk[3SG, REFL, ACC] da cama]]] → *

4 Conclusion

In this article, we have undertaken a comparison among the PRO-based, predicate attraction, and movement approaches to control, focusing on their fitness to account for two types of identity avoidance effects in control constructions in European Portuguese. We have shown that under a strong lexicalist perspective such as that in Chomsky 2000, neither approach successfully covers all the data, as the movement account undergenerates with respect to one type of identity avoidance effect and the predicate attraction and PRO-based accounts overgenerate with respect to the other. On the other hand, when a weak lexicalist model such as that in Chomsky 2001 is adopted, only the movement approach is able to provide a unified analysis for both types of identity avoidance effects. The overall conclusion is that the movement approach to control is better equipped to account for the intricate set of data involving identity avoidance in European Portuguese.

We take this result to be of some importance in that the argument has focused on the architectural properties of each approach, highlighting their key foundational assumptions: whether the control relation requires just the controller or a controller and a controllee and whether the controllee is a lexical formative like PRO or a residue of movement. Thus, unless the architecture of the approaches examined is changed, adjustments of technical implementation in either of them should in principle not affect the gist of the argumentation explored here and the conclusion reached.

It is also worth noting that from a conceptual point of view, the empirical advantages of the movement approach to control with respect to the topic under discussion are not surprising. The types of identity avoidance effects we have examined are sensitive to phonological and morphological information and if the controllee is a copy of the controller, we expect it to be subject to whatever computations and restrictions may independently apply to the controller in the phonological component.

Acknowledgments

Earlier versions of this article have been presented at the University of the Basque Country and at the Workshop on Morphology and Interpretation held at the University of São Paulo. We would like to thank the audiences at these presentations, as well as three anonymous reviewers, for their valuable comments and suggestions. Writing of the current version has been partially supported by FCT – Fundação para a Ciência e a Tecnologia (funding UID/LIN/00214/ 2013; first author) and CNPq (grant 307730/2015-8; second author), to which we are also thankful.

Notes

1 The relevant empirical paradigms to be discussed below are not shared by Brazilian Portuguese. In this dialect, the existence of constructions that at face value involve deletion of clitics (in particular, reflexive clitics and the indefinite clitic se) is completely unrelated to identity avoidance (see, e.g., D’Albuquerque 1984, Galves 1987, Nunes 1991, 1995a).

2 Throughout the article, judgments are due to the first author. It should be noted that what matters for the following discussion is the availability of deletion of the embedded reflexive clitic in (2) and related structures. We will put aside the issue of why deletion may be mandatory or optional for different European Portuguese speakers. Deletion of reflexive clitics is sensitive to the type of reflexive verb involved, and lexical idiosyncrasies may affect speakers’ judgments. Focusing on the ECM (exceptional Case-marking) environments seen so far, we identify the following hierarchy (see Martins and Nunes 2017): deletion is prohibited when the reflexive is a true argument of the embedded verb (see (ia)), optional when the verb is inherently reflexive (see (ib)), and obligatory with verbs that change their meaning when used with a reflexive (see (ic)).

  • (i)

    • a.

      O João ouviu-me elogiar-*(me).

      the João heard-me praise-REFL1SG

      ‘João heard me praise myself.’

    • b.

      O João ouviu-me queixar-(me) ao diretor.

      the João heard-me complain-REFL1SG to.the director

      ‘João heard me complain to the director.’

    • c.

      O João ouviu-nos lamentar-(*nos).

      the João heard-us lament-REFL1PL

      ‘João heard us lamenting.’

3 Similar effects are well-known from other Romance languages. In Italian, for instance, the ban on two instances of si in the same clause is circumvented by means of a suppletive clitic, as illustrated in (i) (see, e.g., Burzio 1986, Cinque 1995). European Portuguese has no analogous repair mechanism to fix the ungrammaticality of sentences like (5).

  • (i)

    Italian

    • a.

      *Si si lava.

      IND REFL washes

    • b.

      Ci si lava.

      REFL IND washes

      ‘One washes oneself.’

4 For relevant discussion, see for example Golston 1995, Yip 1998, Bo'ković 2002, Van Riemsdijk 2008, and the collection of papers in Nasukawa and Van Riemsdijk 2014.

5 It is worth pointing out that clitic climbing of the reflexive in structures with the auxiliary ir ‘go’ is optional, as shown in (i) (see, e.g., Gonçalves 1992). Hence, the unacceptability of (9b) cannot be due to lack of clitic climbing. Furthermore, clitic climbing of the reflexive in (9b), as in (ii), yields a sentence with the two instances of se cliticized to the auxiliary and is to be excluded on a par with (9a).

  • (i)

    • a.

      O João vai levantar-se cedo amanhã.

      the João go.3SG raise-SEREFL early tomorrow

    • b.

      O João vai-se levantar cedo amanhã.

      the João go.3SG-SEREFL raise early tomorrow

      ‘João is going to get up early tomorrow.’

  • (ii)

    *Vai-se-se levantar cedo amanhã.

    go.3SG-SEREFL-SEIND raise early tomorrow

    ‘People are going to get up early tomorrow.’

6 Of course, this does not preclude other types of cooccurrence restrictions involving indefinite se from relying on adjacency. For instance, indefinite se in (Standard) European Portuguese cannot cooccur with a third person nonreflexive accusative clitic if they are in an adjacent configuration, as illustrated by the contrast in (i) (see, e.g., Naro 1976, Martins and Nunes 2016).

  • (i)

    • a.

      *Alugou-se-o ontem.

      rented-SEIND-it yesterday

      ‘One rented it yesterday.’

    • b.

      Vai-se alugá-lo amanhã.

      go.3SG-SEIND rent-it tomorrow

      ‘One is going to rent it tomorrow.’

7 On clitic placement in European Portuguese, see for example Martins 1993, 2013.

8 For expository purposes, English glosses will be used in the presentation of derivational steps, copies left by movement that are relevant to our discussion will be annotated with superscripted indices, and strong phases will be shaded.

9 For reasons of space, we will only present the representations of the Agree implementation of the PRO-based approach. As the reader can verify, the two implementations do not differ in their predictions with respect to the data discussed below.

10 As extensively discussed by Hornstein (1999, 2001) and Boeckx, Hornstein, and Nunes (2010), in adjunct control constructions the controller moves out of the clause where it is generated before this clause is adjoined and becomes an adjunct island (an instance of sideward movement in the sense of Nunes 1995b, 2001, 2004). Thus, in the case of (30c), the upper clitic moves (to the object position of cair ‘fall’) before the temporal clause is adjoined to the VP.

11 For expository purposes, we have equated adjunct islandhood with strong phasehood. Thus, the adjunct clause in (30) is taken to function as a strong phase after it becomes an island, which has consequences for PIC computations. In particular, when the matrix v is merged, the TP within the adjunct CP is transferred. Nothing specifically hinges on this assumption, though. It could well be the case that the reflexive is not accessible in the higher domain because a complex adjunct must be independently sent to Spell-Out (see, e.g., Uriagereka 1999, Nunes and Uriagereka 2000). Whatever the ultimate analysis of adjunct islandhood may be, the important point here is that it could in principle be incorporated by the three approaches to control we have been examining, and the movement approach distinguishes itself from the other approaches in including a copy of the controller within the island.

12 It should be pointed out that a construction superficially similar to (37b) is grammatical in European Portuguese, as shown in (i). However, the verb ver ‘see’ in European Portuguese may also select for an inflected infinitival clause, with a nominative subject, as illustrated in (ii). Thus, se in (i) is to be analyzed as the nominative subject of the embedded inflected infinitival.

  • (i)

    Eu não vi comprar-se as ferramentas.

    I not saw buy-SEIND the tools ‘I didn’t see people buy the tools.’

  • (ii)

    Eu não vi tu comprares as ferramentas.

    I not saw you.SG.NOM buy.INF.2SG the tools

    ‘I didn’t see you buy the tools.’

References

Boeckx,
Cedric
,
Norbert
Hornstein
, and
Jairo
Nunes
.
2010
.
Control as movement
.
Cambridge
:
Cambridge University Press
.
Bošković,
Željko
.
2002
.
On multiple wh-fronting
.
Linguistic Inquiry
33
:
351
383
.
Burzio,
Luigi
.
1986
.
Italian syntax: A Government-Binding approach
.
Dordrecht
:
Reidel
.
Chomsky,
Noam
.
1993
. A minimalist program for linguistic theory. In
The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger
, ed. by
Kenneth
Hale
and
Samuel Jay
Keyser
,
1
52
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
.
2000
. Minimalist inquiries: The framework. In
Step by step: Essays on Minimalist syntax in honor of Howard Lasnik
, ed. by
Roger
Martin
,
David
Michaels
, and
Juan
Uriagereka
,
89
155
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
.
2001
. Derivation by phase. In
Ken Hale: A life in language
, ed. by
Michael
Kenstowicz
,
1
52
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
, and
Howard
Lasnik
.
1993
. The theory of principles and parameters. In
Syntax: An international handbook of contemporary research
, ed. by
Joachim
Jacobs
,
Arnim
von Stechow
,
Wolfgang
Sternefeld
, and
Theo
Vennemann
,
506
569
.
Berlin
:
Walter de Gruyter
.
Cinque,
Guglielmo
.
1988
.
On si constructions and the theory of arb
.
Linguistic Inquiry
19
:
521
581
.
Cinque,
Guglielmo
.
1995
.
Italian syntax and Universal Grammar
.
Cambridge
:
Cambridge University Press
.
D’Albuquerque,
Alair
.
1984
.
A perda dos clíticos num dialeto mineiro
.
Tempo Brasileiro
78/79
:
97
121
.
D’Alessandro,
Roberta
.
2004
.
Impersonal si constructions: Agreement and interpretation
. Doctoral dissertation, University of Stuttgart. Available at http://elib.uni-stuttgart.de/opus/volltexte/2004/1630.
Galves,
Charlotte
.
1987
.
A sintaxe do português brasileiro
.
Ensaios de Linguística
13
:
31
50
.
Golston,
Chris
.
1995
.
Syntax outranks phonology: Evidence from Ancient Greek
.
Phonology
12
:
343
368
.
Gonçalves,
Anabela
.
1992
.
Para uma sintaxe dos verbos auxiliares em português europeu
. Master’s thesis, Universidade de Lisboa.
Gonçalves,
Anabela
.
1999
.
Predicados complexos verbais em contextos de infinitivo não preposicionado do português europeu
. Doctoral dissertation, Universidade de Lisboa.
Hornstein,
Norbert
.
1999
.
Movement and control
.
Linguistic Inquiry
30
:
69
96
.
Hornstein,
Norbert
.
2001
.
Move! A Minimalist theory of construal
.
Oxford
:
Blackwell
.
Landau,
Idan
.
2000
.
Elements of control: Structure and meaning in infinitival constructions
.
Dordrecht
:
Kluwer
.
Landau,
Idan
.
2004
.
The scale of finiteness and the calculus of control
.
Natural Language and Linguistic Theory
22
:
811
877
.
Manzini,
Maria Rita
, and
Anna
Roussou
.
2000
.
A Minimalist theory of A-movement and control
.
Lingua
110
:
409
447
.
Martin,
Roger
.
2001
.
Null case and the distribution of PRO
.
Linguistic Inquiry
32
:
141
166
.
Martins,
Ana Maria
.
1993
.
Enclisis, VP-deletion and the nature of Sigma
.
Probus
6
:
173
205
.
Martins,
Ana Maria
.
2000
. A Minimalist approach to clitic climbing. In
Portuguese syntax: New comparative studies
, ed. by
João
Costa
,
169
190
.
Oxford
:
Oxford University Press
.
Martins,
Ana Maria
.
2013
. A posição dos pronomes pessoais clíticos. In
Gramática do português
, ed. by
Eduardo Buzaglo
Paiva Raposo
,
Maria Fernanda
Bacelar do Nascimento
,
Maria Antónia
Mota
,
Luísa
Segura
, and
Amália
Mendes
,
2231
2302
.
Lisboa
:
Fundação Calouste Gulbenkian
.
Martins,
Ana Maria
, and
Jairo
Nunes
.
2005
.
Raising issues in European and Brazilian Portuguese
.
Journal of Portuguese Linguistics
4
:
53
77
.
Martins,
Ana Maria
, and
Jairo
Nunes
.
2016
. Passives and se constructions. In
The handbook of Portuguese linguistics
, ed. by
Leo
Wetzels
,
Sergio
Menuzzi
, and
João
Costa
,
318
337
.
Hoboken, NJ
:
Wiley-Blackwell
.
Martins,
Ana Maria
, and
Jairo
Nunes
.
2017
.
Deletion of reflexive clitics with the verb custar in European Portuguese: An MTC account
.
Journal of Portuguese Linguistics
16
:
1
25
.
Naro,
Anthony
.
1976
.
The genesis of the reflexive impersonal in Portuguese: A study in syntactic change as a surface phenomenon
.
Language
52
:
779
810
.
Nasukawa,
Kuniya
, and
Henk
van Riemsdijk
, eds.
2014
.
Identity relations in grammar
.
Berlin
:
Walter de Gruyter
.
Nunes,
Jairo
.
1991
.
Se apassivador e se indeterminador: O percurso diacrônico no português brasileiro
.
Cadernos de Estudos Lingüísticos
20
:
33
57
.
Nunes,
Jairo
.
1995a
.
Ainda o famigerado se
.
D.E.L.T.A.
11
:
201
240
.
Nunes,
Jairo
.
1995b
.
The copy theory of movement and linearization of chains in the Minimalist Program
. Doctoral dissertation, University of Maryland, College Park.
Nunes,
Jairo
.
2001
.
Sideward movement
.
Linguistic Inquiry
32
:
303
344
.
Nunes,
Jairo
.
2004
.
Linearization of chains and sideward movement
.
Cambridge, MA
:
MIT Press
.
Nunes,
Jairo
, and
Juan
Uriagereka
.
2000
.
Cyclicity and extraction domains
.
Syntax
3
:
20
43
.
Pesetsky,
David
, and
Esther
Torrego
.
2007
. The syntax of valuation and the interpretability of features. In
Phrasal and clausal architecture
, ed. by
Simin
Karimi
,
Vida
Samiian
, and
Wendy
Wilkins
,
262
294
.
Amsterdam
:
John Benjamins
.
Raposo,
Eduardo
, and
Juan
Uriagereka
.
1996
.
Indefinite SE
.
Natural Language and Linguistic Theory
4
:
749
810
.
Riemsdijk,
Henk van
.
2008
. Identity avoidance: OCP-effects in Swiss relatives. In
Foundational issues in linguistic theory: Essays in honor of Jean-Roger Vergnaud
, ed. by
Robert
Freidin
,
Carlos P.
Otero
, and
Maria Luisa
Zubizarreta
,
227
250
.
Cambridge, MA
:
MIT Press
.
Uriagereka,
Juan
.
1999
. Multiple Spell-Out. In
Working Minimalism
, ed. by
Samuel David
Epstein
and
Norbert
Hornstein
,
251
282
.
Cambridge, MA
:
MIT Press
.
Yip,
Moira
.
1998
.
Identity avoidance in phonology and morphology
. In
Morphology and its relation to phonology and syntax
, ed. by
Steven G.
Lapointe
,
Diane K.
Brentari
, and
Patrick M.
Farrell
,
216
246
.
Stanford, CA
:
CSLI Publications
.