Deletion of reflexive clitics with the verb custar in European Portuguese: An MTC account

The impersonal verb custar (lit. ‘cost’) in European Portuguese selects for a dative experiencer argument and an infinitival clause, which may be preceded by the preposition a. Interestingly, a reflexive clitic co-referential with the experiencer argument can be deleted (under conditions to be specified) if it is within the prepositional infinitival complement, but not within its prepositionless counterpart. We argue that the presence of the preposition a preceding the infinitival complement of custar correlates with obligatory control and show how deletion of reflexive clitics within the prepositional infinitivals can be captured under the movement theory of control (MTC; see e.g. Hornstein 1991, 2001; Boeckx, Hornstein and Nunes 2010). More specifically, we show that if the infinitival subject is a deleted copy left by the “controller”, this copy can trigger deletion of the reflexive clitic as a way to satisfy a superficial ban on morphologically identical clitics in a local domain.


Introduction
The impersonal verb custar 'cost' in European Portuguese presents us with an interesting puzzle. 1 At first sight, all that needs to be said about it is that its infinitival complement may be optionally preceded by the preposition a 'to', with a corresponding subtle difference in meaning, as illustrated in (1) below. That is, custar may select for either a prepositionless or a prepositional infinitival as a matter of lexical subcategorization.
cost-me cl.dat write-inf the report 'Writing the report was hard on me.' b. Custou-me a escrever o relatório.
cost-me cl.dat to write-inf the report 'It was hard for me to succeed in writing the report.' Verb Custar in European Portuguese Art. 2, page 2 of 25 Curiously, however, a reflexive clitic in the infinitival clause co-referring with the matrix experiencer must sometimes be deleted in the prepositional version but not in its prepositionless counterpart, as exemplified by the contrast in (2).
cost-me cl.dat sit-inf-cl 1sg on-the ground 'To sit on the ground pained me.' b. Custou-me a sentar-(*me) no chão.
cost-me cl.dat to sit-inf-cl 1sg on-the ground 'It was hard for me to succeed in sitting on the ground.' The sentences in (3) below present additional attested examples of this unexpected availability of deletion of the reflexive clitic in the prepositional complement of custar.
In each of the examples, the infinitival verb should in principle be associated with a reflexive clitic (levantar-me, adaptar-me, integrar-me and convencer-me, respectively).
( homing-pigeon but had of myself render to-the evidence 'I very much resisted to recognize that it would not be successful as homing pigeon but I eventually had to accept it.' (Google search, 03-07-2016) This process of reflexive deletion in European Portuguese involves a plethora of complexities. First of all, the availability of deletion is sensitive to the type of verb involved in the embedded clause. Putting aside lexical idiosyncrasies which may affect speakers' judgements, three general patterns can be identified, as outlined in (4). 2 2 In the following discussion we will abstract away from these independent lexical restrictions, for they do not interfere with the distinction between the two types of infinitival complements associated with custar.
That is, lexical conditioning may derive the pattern in (4) for the prepositional complement, but deletion is uniformly ruled out in the case of the prepositionless complement, regardless of the lexical items involved. For concreteness, we will henceforth focus on examples of the type described in (4c). For relevant discussion and further refinements on types of reflexive verbs, see e.g. Burzio (1986); Cinque (1988); Vilela (1992); Brito, Duarte and Matos (2003); Duarte (2013); Gonçalves and Raposo (2013); Mendikoetxea (1999); Peregrín Otero (1999); Sánchez López (2002). Verb Custar in European Portuguese Art. 2, page 3 of 25 (4) • Deletion is impossible if the reflexive clitic can alternate with a non-reflexive argument with no significant changes in the meaning of the verb: a. Custou-nos a ver-*(nos) na fotografia.
cost-us cl.dat to see-refl 1pl in-the picture 'It was hard for us to succeed in spotting ourselves in the picture.' a'. Custou-nos a vê-los na fotografia.
cost-us cl.dat to see-them cl.acc in-the picture 'It was hard for us to succeed in spotting them in the picture.' • Deletion is optional with verbs that always require a reflexive clitic, regardless of whether or not they are semantically reflexive (that is, verbs that are traditionally classified as intrinsically "pronominal"): b. Custou-te a arrepender-(te) de tudo o que fizeste.
cost-you cl.dat to repent-refl 2sg of everything the what did-2sg 'It was hard for you to succeed in repenting everything you did.' b'. *Custou-te a arrependê-la de tudo o que fez.
cost-you cl.dat to repent-her cl.acc of everything the what did-3sg *'It was hard for you to make her repent everything she did.' • Deletion is obligatory for some speakers (and possible for others) if alternation between reflexive and non-reflexive arguments leads to (slight) changes with respect to thematic properties of the arguments involved: 3 c. Custou-me a levantar-(*me) da cadeira.
cost-me cl.dat to raise-refl 1sg from-the chair 'It was hard for me to succeed in rising from the chair.' c'. Custou-me a levantá-la da cadeira.
cost-me cl.dat to raise-her cl.acc from-the chair 'It was hard for me to succeed in raising her from the chair.' Second, custar is, to the best of our knowledge, the only verb in European Portuguese that allows both types of impersonal constructions illustrated in (1) and correlates deletion of the reflexive clitic in (2)/(3) with the presence or absence of the preposition. However, this deletion process is not necessarily dependent on the presence of the preposition a in the embedded clause, as the contrast between (2a) and (2b) could lead one to think. This phenomenon is also found with the (prepositionless) infinitival complement of perception and causative verbs, as shown in (5) It is worth observing that the data in (5) do instantiate biclausal bare infinitival ECM structures and should not be confused with restructuring faire-infinitive constructions, which independently exclude reflexive clitics associated with the infinitival verb (see Kayne 1975;Gonçalves 1999; among others). The preverbal vs. postverbal position of the embedded subject in (6) below, for instance, respectively signals an ECM and a faire-infinitive construction. Notice that the reflexive clitic associated with the infinitival verb is licensed in the ECM configuration in (6a), but not in the faire-infinitive counterpart in (6b). A confounding factor is that for some speakers (including the first author, but not e.g. Gonçalves 1999), suppressing the reflexive can save faire-infinitive sentences with causative verbs but, crucially, not with perception verbs. Thus, these speakers allow a reflexive reading for (7a) below, for instance, but not for (7b), which can only have a nonreflexive interpretation in which the infinitival subject is a null pronoun with arbitrary interpretation. This shows that the deletion of reflexives seen in (5), for example, is not the rescue strategy available for some speakers in faire-infinitive constructions such as (7a), for it involves not only causative (see (5c-d)), but also perception verbs (see (5a-b)). This brief survey of the complexities involving reflexive deletion in European Portuguese raises the question of what property the prepositional infinitival of custar in (1b) has that makes it bluntly contrast with (1a) regarding the phenomenon of reflexive deletion illustrated in (2b) and (3). As we have just seen, it cannot be just a matter of lexical subcategorization. Crucially, we cannot say that reflexive deletion is somehow licensed by the preposition subcategorized by custar, for it can also be licensed in the prepositionless complement of perceptual and causative verbs (see (5)). Thus, we seem to be forced to either take the preposition a and perceptual and causative verbs to form a natural class or assume a construction specific condition tied to custar a banning the co-occurrence of the relevant clitics. Needless to say, neither of these options is conceptually appealing. Our approach to this puzzle has two parts. Building on work by Martins and Nunes (2005), we first argue that the relevant difference between (1a) and (1b) involves obligatory control. Once this is established, we then proceed to show how the contrast between (2a) and (2b) may receive a natural account under the movement theory of control (MTC; see e.g. Hornstein 1999Hornstein , 2001Boeckx, Hornstein, and Nunes 2010). We should point out at the outset that it is not our goal to undertake a comparative evaluation of different theories of control with respect to the data presented here. Our main reason for framing the discussion in terms of the MTC is that one of its key ingredients -namely, the assumption that obligatorily controlled PRO is a deleted copy of the "controller"provides a straightforward way to handle reflexive deletion within the prepositional complement of custar. We leave for another opportunity an adequate investigation of whether and how non-movement approaches to control can replicate the results obtained under the MTC (see Martins and Nunes forthcoming).
Before we move to the discussion proper, some clarification remarks are in order. What matters for the argument to be developed below is the existence of contrasts like (2), relative to the availability of deletion of the embedded reflexive clitic. Whether deletion may be forbidden, optional, or mandatory for different speakers is irrelevant for the ensuing discussion. What is important is that whenever deletion is possible (for a given speaker), it takes place in the prepositional, but not in the prepositionless, infinitival complement of custar. It is also worth noting that the relevant contrast only arises if the infinitival verb does not allow an intransitive use. Speakers that independently allow the verbs sentar 'sit' or levantar 'to raise', for example, to be intransitive do not identify the contrast in (2) or see anything especial about (4c), but do so in analogous sentences with verbs that disallow an intransitive option in their grammars.
The got-used 'It was hard on me to get used to Tom but eventually I did.' Speakers that allow sentences like (9) may simply not have the relevant ban on identical clitics in their grammars and deletion is not an option. Thus, (9) contrasts with (10), where the reflexive of the first conjunct has been deleted. That this indeed involves a case of reflexive deletion and not an intransitive use of habituar is shown by the fact that the reflexive clitic is present in the second conjunct of (10). Furthermore, notice that the infinitival complement in (10) is prepositional, which conforms with the generalization that deletion is only licensed when the preposition is present (see (2)). In turn, (11) apparently contradicts what we have just said, for there is no reflexive in the infinitival complement and the preposition is not present either. However, when we examine the second conjunct of (11), we see that this speaker independently treats habituar as intransitive. Hence, (11) is not at odds with (2), for the first conjunct does not involve deletion, but an intransitive use of habituar. Finally, the second conjunct of (12) shows that this other speaker takes habituar to be reflexive, but the reflexive in the first conjunct cannot be deleted because the infinitival is not prepositional. The grammar that we will be discussing throughout the paper is the one illustrated in (2), (3), and (10), that is, the grammar where deletion of reflexive clitics is enforced in the prepositional complement of custar. The paper is organized as follows. In section 2 we make some brief remarks regarding the deletion process illustrated in (2b)/(3)/(4b')/(4c')/(5)/(10). It should be pointed out that our goal is not to account for the deletion process itself, but to use it as an independent criterion of empirical adequacy to test structures assigned to (1a) and (1b). In section 3 we show that the two infinitival complements in (1) sharply contrast with respect to obligatory control diagnostics and that the type of deletion seen in (2b) is limited to the obligatory control structure. Given this result, in section 4 we show that this correlation between obligatory control and deletion can receive a straightforward account if the obligatorily controlled infinitival subject is analyzed as copy of the "controller", as postulated by the MTC. Section 5 presents some concluding remarks.

Some remarks on the deletion of reflexive clitics with custar a
Bearing in mind that only the prepositional complement of custar may allow for deletion (see (2a) vs. (2b)), let us consider the data in (13) and (14).
cost-me cl.dat to sit-refl 1sg on-the ground 'It was hard for me to succeed in sitting on the ground.' b. Custou-te a sentar-(*te) no chão.
cost-you cl.dat to sit-refl 2sg on-the ground 'It was hard for you to succeed in sitting on the ground.' At first sight, the contrast between the sentences in (13), on the one hand, and (14a), on the other, simply indicates that deletion is triggered when the clitic in the embedded clause is identical to the clitic attached to custar; hence, deletion takes place in (13a) and (13b), but not in (14a). However, the contrast between the two sentences of (13) and (14b) shows that phonological identity is not sufficient, for the two clitics in (14b) are identical, but deletion is not triggered. Upon close inspection, we can see that deletion targets reflexive clitics; hence, it is possible in (13), but not in (14). 4 For the sake of completeness, it should be noted that some speakers may allow deletion of reflexive se in the presence of a third person dative clitic, as illustrated in (15)  than it looks if one takes into consideration that in Spanish, the reflexive se may be a suppletive form of the dative le in "spurious"-se constructions, as shown in (17) (see e.g. Perlmutter 1971;Bonet 1995).
se/him cl.dat.3sg it cl.acc gave.2sg 'You gave it to him/her.' What matters for the purposes of our discussion is that for speakers who allow deletion of se in sentences such as (15), they only permit it in the prepositional version. In the following discussion we will abstract away from the variation regarding sentences like (15), for it does not interfere with the distinction between the two types of infinitival complements associated with custar.

Differences between the two infinitival complements of custar
Let us now return to the intriguing puzzle in (2), repeated below in (18), and discuss differences between the two types of complement of custar that may provide a basis for us to account for why deletion is triggered in (18b) and blocked in (18a).
cost-me cl.dat sit-refl 1sg on-the ground 'To sit on the ground pained me.' b.
cost-me cl.dat to sit-refl 1sg on-the ground 'It was hard for me to succeed in sitting on the ground.'

Some interpretive differences
Despite their similarities, the two infinitival complements of custar contrast in many aspects (see Martins and Nunes 2005). Although the experiencer argument of custar is interpreted as affected by the state of affairs described in the infinitival clause of both types of complements, in the prepositional version it is also interpreted as being more actively engaged in carrying out the events described in the infinitival. This is very clear in the pair of sentences in (19)  cost-them cl.dat much to win the game 'It took them a lot of continued effort to succeed in winning the game.' Also telling is the pragmatic oddness of b-sentences in (21) and (22)  This engagement by the experiencer in the prepositional version may also correlate with duration, as illustrated by the different interpretations of (23a) and (23b) below. It is also behind the oddness of (24a) against the felicity of (24b arrive-inf on time to-the work 'Due to the strike in public transportation, it took me a lot of effort and time to get to work on time.'

Some differences in structural complexity
The two infinitival complements also contrast in terms of structural complexity, as the prepositional version does not license auxiliaries (see (25)), modals (see (26)), or independent time adverbials (see (27)):

Standard Control diagnostics
The interpretive and structural differences reported above suggest that the prepositional infinitival but not its prepositionless counterpart may instantiate obligatory control. This is further confirmed when unequivocal diagnostics of obligatory control are examined, as we will show below.

Licensing of independent subjects
The two infinitivals differ in their ability to license an overt subject. The attested prepositionless example in (28a), for instance, sharply contrast with its prepositional counterpart If the prepositionless infinitival can license an independent subject within its clause, as the combination of the data in (28) and (29) clearly shows, we are led to expect that in sentences such as (30), the null subject of the prepositionless infinitival need not correfer with the matrix experiencer, whereas the null subject of the prepositional version (as an instance of obligatory control) must. cost-me cl.dat to fail that student 'It was hard for me to succeed in failing that student.'/*'It was hard for me to succeed in having other people fail that student.' We take this prediction to be essentially correct, but it should be observed that there is a very strong bias for the embedded null subject to take the experiencer of custar as its antecedent, which leads some speakers (including one of the anonymous reviewers) to reject (30a) with index k and other analogous structures. We do not have an account of why this bias is stronger for some speakers and not others, but we would like to mention two points that support our description of the data. First, the non-correferential interpretation in out-of-the-blue sentences such as (30a) may become more salient if an appropriate pragmatic context is provided. This is illustrated in (31) below, for instance, where the context set by the question in (31A) pragmatically precludes the correferential reading for the null subject of the infinitival in (31B). Crucially, no pragmatic context is able to license the non-correferential reading with prepositioned infinitivals. 9 cost-me cl.dat have-inf such talent and not it profit-inf 'It pains me that he is so talented and does not take advantage of it.' More importantly, the bias towards the correferential interpretation can be turned around via contrastively focused pronouns. In a subject control construction such as (32) below, for instance, the postverbal pronoun in the infinitival is interpreted as imposing a contrastive focus on the embedded subject (for relevant discussion, see e.g. Costa 2004;Barbosa 2009;Szabolcsi 2009). When a contrastively focused pronoun is added to (30), as shown in (33) the Rui thinks that me cost to write-inf he the report

On the c-command condition and the nature of the antecedent
The argument that the null subject in (34b)/(35b) requires a local antecedent presupposes that the dative argument of custar sits in a c-commanding position. That this holds true is shown by the Principle C effect illustrated in (36). That being so, one predicts that a DP within the experiencer should not count as a proper antecedent for the null subject of a prepositional infinitival, due to lack of c-command. Unfortunately, this prediction cannot be tested because the experiencer argument of custar can be realized by a dative clitic, but not by a full DP, as illustrated in (37). We speculate that this idiosyncrasy is related to the inherent nature of the Case assigned by custar to its experiencer.
cost to-the João to write the report 'It was hard for João to succeed in writing the report.' The fact that custar assigns (inherent) dative Case to its specifier yields an additional contrast between the two infinitival complements of custar. As shown in (38) below, the indefinite clitic se is not licensed in the prepositionless infinitival. However, some speakers (including the first author) allow it with the prepositional infinitival, as illustrated by (39a) and the attested example in (39b). Interestingly, these speakers also allow constructions such as (40), where custar functions as a raising verb.
(38) *Custa-se acreditar numa coisa dessas. Under the standard assumption that indefinite se is intrinsically nominative, the ungrammaticality of (38) (for all speakers) and (39) for speakers who do not allow (40) is due to a feature clash, as the verb custar assigns inherent dative Case to its specifier (see section 3.3.2). For speakers who also admit a raising construction for custar, (39) is to be derived on a par with (40)

Only-DP antecedents, VP-ellipsis, and de se readings
Additional confirmation for distinguishing the two infinitival complements of custar in terms of obligatory control is provided by the interpretive properties of sentences such as the ones in (41)-(43) below. The null subject of the prepositionless infinitival allows coreferential and bound readings when anteceded by an only-DP (see (41a)), permits strict and sloppy readings under ellipsis (see (42a)), and is compatible with a non-de se reading in contexts of lack of self-awareness (see (43a)). By contrast, the null subject of the prepositional infinitival displays the opposite behavior: it enforces a bound reading when anteceded by an only-DP (see (41b)), triggers a sloppy interpretation under ellipsis (see (42b)), and only allows de se readings, thus being pragmatically infelicitous in the context provided in (43)  cost-me cl.dat give-inf the news and to-the João cost-him too 'It pained both me and João that I had to deliver the news.' (strict reading) or 'It pained me that I had to deliver the news and it also pained João that he had to deliver the news.' (sloppy reading). cost-him to lay-down-inf the weapons 'It was hard for him to succeed in laying down his weapons.'

Differences regarding inflection
Finally, the two infinitival complements also contrast in terms of inflection. The prepositionless infinitival allows subject agreement morphology, but not the prepositional one: cost-us cl.dat to fail-inf-1pl that student 'It was hard for us to succeed in failing that student.' The availability of agreement morphology in (44a) is actually not surprising, for the prepositionless infinitival can license an independent subject, as discussed in section 3.3.1. As for (44b), there was not an a priori expectation, for in European Portuguese subject control verbs generally do not license overt agreement morphology, whereas object control optionally do so, as illustrated in (45). In this regard, what (44b) shows is that it patterns like subject rather than object control. There is actually interesting indirect evidence that shows that it is not the case that the prepositional infinitival in (44b) has ф-features that do not get morphologically realized, but rather that it simply has no ф-features. The evidence is based on Raposo's (1987) observation that only uninflected infinitives license tough-movement, as shown in (46).
(46) Esses livros são difíceis de ler(*mos) these books are hard of read-inf-1pl 'These books are hard to read.' Interestingly, only the prepositional complement of custar allows a tough-like construction, as illustrated in (47)  should in principle expect both of them to allow tough-movement in (47) and (48). The fact that this is not what happens may be taken to show that the prepositionless infinitival has ф-features, which may be morphologically realized or not, whereas the prepositional infinitival has no ф-features whatsoever.
those students cost-past-3pl to fail 'Those students were hard to fail.'

Summary
In sum, the prepositional infinitival complement of custar involves obligatory control, whereas its prepositionless infinitival complement does not, as sketched in (49) (49a) involves a personal infinitive which may license a pro in the subject position; pro may − but need not − be bound by the experiencer argument in the specifier of the VP headed by custar (see section 3.3.1). By contrast, the embedded subject of (49b), as an instance of obligatorily controlled PRO, must be bound by the experiencer argument of custar, as it is the most local c-commanding antecedent.
Having characterized the infinitival complements of custar, we may now go back to the puzzle of why prepositional infinitivals may trigger the deletion of a reflexive clitic.

Back to the deletion puzzle
Once the fundamental control difference between the two types of infinitival complements selected by custar has been identified in section 3, the data in (2), repeated below in (50), can be taken to show that deletion of reflexives locally bound by an (identical) clitic (see section 2) is only operative within the obligatory control structure (i.e. the prepositional infinitival complement). 'A magical moment that led the Greeks to ecstasy, as they were proud to see the Olympic Games, which were so hard to organize, finally begin.' (Jornal de Notícias, 13-08-2004. http://www.jn.pt/arquivo/200404/07/2016)  to raise and to ask apologies to-the instructor for-the mistake 'After falling on his back in a swamp, he hastened to get back to his feet and apologize to the instructor for the mistake.' It is worth noting another parallel. Recall that standard object control in European Portuguese allows the infinitival to optionally carry overt agreement morphology, whereas the prepositional complement of custar bans such inflection (see section 3.3.5). Interestingly, only the non-agreeing version licenses deletion of reflexive clitics, as illustrated by the contrast between the uninflected infinitives of (51) and their inflected counterparts in (53).
they forced-us to get-away-1pl-refl 1pl from-that path 'They forced us to get away from that path.' b. O médico vai forçar-te a sentares-*(te) de outra maneira.
the doctor goes force-you to sit-2sg-refl 2sg of other manner 'The doctor will force you to sit in another way.' However, deletion of reflexives when bound by an (identical) local clitic is not restricted to object control configurations. Recall that this phenomenon is also found with the infinitival complement of perception and causative verbs (see section 1), as illustrated in (54)  The contrast between sentences such as the ones in (54), where deletion is available, and (55) below (= (8)), where deletion is blocked, suggests that the co-occurrence restriction that triggers reflexive deletion is clause bound. Hence, the two clitics in (55)  Importantly, what matters is not exactly where the higher clitic ends up, but where it is generated. In (54) the upper clitic is generated in the embedded clause and later cliticizes to the matrix verb (see e.g. Gonçalves 1999;Martins 2000), as becomes clearer with sentences such as (56) below, which involves proclisis to the matrix verb. To put it in different terms, the higher clitic of (54) and (56) has a chance to interact with the reflexive, triggering the deletion of the latter, before it moves to the matrix clause. In (55), on the other hand, there is no derivational step where the two clitics appear in the same clausal domain. As discussed in section 3, the empty category in the embedded subject position is pro in the case of the prepositionless infinitival complement of custar (see (57a)), but an obligatorily controlled PRO in the prepositional version (see (57b)). However, this difference does not seem to be of much help. Given that deletion of the reflexive is triggered in (57b), but blocked in (57a), one would like to treat (57a) on a par with (55), which also blocks deletion, and (57b) with (54) and (56), which trigger deletion. In this sense, (57a) and (55) may be taken to form a natural class in the sense that its clitics are generated and surface in different clauses. Hence, it may be expected that they do not interact -a correct result. However, (57b) and (54)/(56) do not seem to form a natural class, for the higher clitic is generated in the embedded clause in the latter, but in the higher clause in the representation in (57b). In other words, if the co-occurrence restriction under discussion is indeed clause bound, the representation in (57b) leads to the incorrect prediction that the two clitics do not interact and the reflexive cannot be deleted. The intriguing contrast between (57a) and (57b) ceases to be puzzling, though, if the obligatorily controlled PRO in (57b) is actually a deleted copy of the "controller", as postulated by the MTC (see e.g. Hornstein 1999Hornstein , 2001Boeckx, Hornstein and Nunes 2010). We have seen that custar assigns inherent dative Case to its specifier (see section 3.3.3) and selects for either a personal or an impersonal infinitival complement (see section 3.3.5). If the embedded infinitival is personal, it assigns nominative to its Spec, regardless of whether or not its ф-features are morphologically realized, i.e., whether or not it is inflected (see section 3.3.5). The embedded subject then gets frozen in the embedded clause and custar must assign its other θ-role to a new element selected from the numeration. By contrast, if the infinitival is impersonal, it does not assign Case to its subject. Under the MTC, the subject may then move to [Spec, custar], where it gets an additional θ-role and is assigned inherent dative Case. Thus, the MTC analyzes the sentences in (50) along the lines of (58), where the embedded subject position of (50b) involves a copy of the "controller" in the embedded subject position. In (58b) -but not in (58a) -there are two instances of the same clitic within the embedded clause. However, this is usually not tolerated in European Portuguese, as seen with ECM constructions in (54)/(56). Given that the ungrammaticality of the two clitics in (50b) parallels that of the sentences in (54)/(56), the exceptional deletion of the reflexive clitic in (50b) can be viewed as a way to comply with the superficial filter ruling out morphologically identical clitics within the same clause (see (58b)). As for (50a), no problem arises as the subject of the infinitival (pro) is not a clitic (see (58a)); hence, it is morphologically distinct from the reflexive clitic and no deletion is triggered. To sum up, the major assumption of the MTC -namely, that obligatorily controlled PRO is a (deleted) copy of its antecedent -provides a straightforward account of why reflexive deletion within the infinitival complement of custar can only take place if custar is used as an obligatory control verb. 13

Concluding remarks
The MTC has all the ingredients to provide an account for the curious problem brought up by (59) (= (2)), which shows that reflexive clitics may be deleted in the infinitival complement of custar when it is used as an obligatory control verb.
cost-me cl.dat to sit-inf-refl 1sg on-the ground 'It was hard for me to succeed in sitting on the ground.' Under the MTC, we expect obligatorily controlled PRO to behave like a regular copy of its antecedent. More precisely, we expect obligatorily controlled PRO to be subject to whatever computations and restrictions apply to its antecedent in the post-syntactic components of grammar. We have seen that the co-occurrence restriction under discussion computes obligatorily controlled PRO but not a co-referential pro. From the perspective of the MTC, this is not at all surprising. If PRO is a copy of its antecedent (in this particular case, a copy of the experiencer clitic associated with custar), it may be computed with respect to the ban on morphologically identical clitics in a local domain and trigger reflexive deletion.
The discussion above has focused on the apparently erratic behavior of a single lexical item in European Poruguese (custar), but should also be considered under a broader (even if speculative) perspective. It is very likely that the lexical idiosyncrasy of custar regarding the optionality of the preposition in its infinitival complement illustrated in (1), repeated here in (60), is something that can be acquired based on primary linguistic data. That is, a child exposed to data parallel to (60a) and (60b) may reach the reasonable conclusion that the optionality in (60) is a matter of lexical subcategorization: custar may select for either a prepositionless or a prepositional infinitival. 13 A reviewer asks whether one could not get the same results under a PRO-based account by assuming that ECM and controlled infinitivals are not phases. From this perspective, a phase intervenes between the matrix experiencer and the reflexive in (ia), but not in (ib). Thus, deletion could be triggered in (ib) but blocked in (ia). The reviewer's point is well taken. In fact, in Martins and Nunes (forthcoming) we assume the phase-based framework and make a detailed comparison between non-movement approaches to control and the MTC with respect to two types of co-occurrence restrictions in European Portuguese: the one discussed here, which leads to deletion of reflexives, and ungrammatical cases of indefinite se co-occurring with reflexive se. Our conclusion is that when reflexive deletion is considered in isolation, there is no clear basis for distinguishing between movement and non-movement approaches to control under a phase-based analysis. However, when reflexive deletion is computed together with the co-occurrence restriction involving indefinite and reflexive se, only the movement approach is able to provide a unified account of the two phenomena. For our current purposes, it is worth noting that our main point here -namely, to show that the different behavior displayed by each of the infinitival complements of custar with respect to reflexive deletion is linked to obligatory control and cannot be a simple matter of lexical subcategorization -still remains valid if we frame our MTC account in terms of the phase approach suggested by the reviewer.