In Brazilian Portuguese (henceforth BP), verbs that are traditionally classified as object control verbs such as convencer ‘convince’, for example, in fact display an intriguing complex pattern. When they take an infinitival complement preceded by a ‘to’, they do behave like object control verbs, as illustrated in (1a) below. On the other hand, when they are associated with a finite clause, a subject control pattern is observed, as shown in (1b) (e.g., Modesto, 2000; Rodrigues, 2004), and when the clausal complement is preceded by de ‘of’, both subject and object control are allowed, regardless of whether the clause is finite or infinitival, as respectively illustrated in (1c-d) (Bueno Dias, 2023; Nunes, 2009).1

Assuming that anaphoric null subjects in BP are to be analyzed in terms of obligatory control (e.g., Ferreira, 2000, 2009; Kato, 1999; Nunes, 2019; Rodrigues, 2004) and that obligatory control is derived by A-movement (e.g., Boeckx et al., 2010; Hornstein, 1999, 2001; Hornstein & Nunes, 2014), in this paper we show how a movement approach is able to capture the whole paradigm illustrated in (1).2 From a movement perspective, the object control reading in (1a), (1c), and (1d) is not surprising. Under a Larsonian structure (Larson, 1988), the matrix object should c-command into the embedded clause of the sentences in (1), thus counting as a proper intervener and blocking the movement of the embedded subject to the matrix subject position. That being so, the ungrammaticality of the object control reading and the grammaticality of the subject control reading in (1b) is doubly puzzling (Modesto, 2000). Even more challenging, the contrast between (1b), on the one hand, and (1c)/(1d), on the other, indicates that whatever the account for the contrast between (1a) and (1b) is (Rodrigues, 2004), it cannot be extended to (1c)/(1d). The unexpected licit subject control reading is associated with an ungrammatical object control reading in (1b), but with a grammatical object control reading in (1c)/(1d). In other words, why is the unexpected subject control reading obligatory in (1b), but optional in (1c)/(1d)?

The paper is organized as follows. Section 2 presents general properties of null subjects in BP and their interaction with the preposition de. Section 3 introduces our analysis of de as an Applicative head, showing how the object control reading of (1c) and (1d) is derived. Section 4 then shows how the puzzling subject control reading of (1c) and (1d) is accounted for. Section 5 presents our concluding remarks.

It has been a consensus in the literature that BP is not a canonical pro-drop language.3 In finite environments where topic drop is independently excluded such as matrix wh-questions, for example, the acceptability of a null subject depends on the pronoun that it arguably stands for, as illustrated in (2) (e.g., Kato et al., 2023a, 2023b, Martins & Nunes, 2021, 2024; Nunes, 2020a).

Interestingly, a null subject corresponding to one of the pronouns in the unacceptable sentence in (2d) may yield an acceptable result if it sits within an embedded clause and finds an antecedent in the subordinating clause (e.g., Figueiredo Silva, 1996; Kato, 1999). There are several restrictions on such licensing, though. For instance, the antecedent must be in a c-commanding and local position (e.g., Ferreira, 2000, 2009; Rodrigues, 2004), as shown in (3a), and if the antecedent is quantified, the null subject must be interpreted as a variable (Negrão, 1986), as shown in (3b).

The fact that the restrictions on embedded null subjects in BP mimic the ones found in standard obligatory control constructions has led researchers to propose that the paradigm illustrated in (3) involves obligatory control, as well (e.g., Ferreira, 2000, 2009; Kato, 1999; Rodrigues, 2004).4 Assuming Hornstein’s (1999, 2001) Movement Theory of Control, Ferreira (2000, 2009) and Rodrigues (2004) in fact argue that embedded null subjects in BP are traces left by a movement operation targeting an A-position in the immediately subordinating domain.5 In particular, Ferreira (2000, 2009) and Nunes (2008, 2019) have proposed that verbal agreement inflection in BP is opaque, being compatible with either a complete φ-set with person and number or an incomplete φ-set, with number only. When T is associated with an incomplete φ-set, it fails to value the Case of its subject, which must then move to a higher domain to value its Case. Importantly, this movement is in compliance with Chomsky’s (2001) Phase Impenetrability Condition, for a CP dominating a φ-incomplete T counts as a weak phase (Ferreira, 2000, 2009; Nunes, 2019).

Compelling evidence for this movement approach to null subjects in BP is provided by island effects (Ferreira, 2000, 2009; Rodrigues, 2004 for extensive discussion). The null subject of (4) below, for instance, finds a local c-commanding antecedent in the subordinating clause, but the result is deviant because the embedded clause is a relative clause island, thus preventing movement from within it.

An interesting challenge to this movement approach was posed by Modesto (2000), based on contrasts like the one in (5) below with verbs of the class of convencer ‘convince’. (5a) is a garden-variety object control structure, which under the Movement Theory of Control (Boeckx et al., 2010; Hornstein, 1999, 2001; Hornstein & Nunes, 2014) is analyzed as involving movement of a Maria from the embedded subject position to the specifier of convenceu ‘convinced’ in the matrix clause; crucially, the embedded null subject of (5a) cannot be interpreted as the matrix subject, for movement of o João from the embedded to the matrix subject position is blocked due to the intervention of a Maria. Surprisingly, the null subject of the analogous sentence in (5b), with a finite complement, must take the matrix subject for its antecedent and not the matrix object. Modesto (2000) observes that under a movement approach, the matrix object should also count as an intervener for the movement of the embedded subject in (5b) and concludes that the movement approach to null subjects in BP is not empirically adequate.

Ferreira (2000, 2009) nonetheless observes that the matrix object in (5b) does not c-command the embedded subject, as shown by the lack of Principle C effects in (6) below, for instance (Nunes, 2019 for discussion). Furthermore, Rodrigues (2004) shows that the finite clause embedded under verbs like convencer, as in (5b), actually behaves like an adjunct, for it induces island effects for wh-movement, as shown in (7).

That being so, Rodrigues (2004) argues that (5b) should be subsumed under adjunct control, which has been analyzed by Hornstein (1999, 2001) as involving sideward movement (Nunes, 2001). In the case of (5b), o João moves from the embedded subject position to the matrix [Spec,vP], as sketched in (8), before the embedded clause is adjoined to the matrix vP and becomes an adjunct island.6

Relevant to our current discussion is Nunes’s (2009) observation that if the embedded clause of (5b) is preceded by the preposition de ‘of’, subject and object control are available, as shown in (9) below.

Nunes (2009) proposes that in the object control reading in (9), de is a marker of inherent Case. By being thematically marked, the embedded clause behaves like a complement and accordingly, movement of the embedded subject must target the closest c-commanding position, namely, [Spec,VP]; hence, the object control reading is derived. As evidence for this proposal, Nunes shows that when de is present, the island effects noted by Rodrigues (2004) (see (7)) are no longer observed, as illustrated in (10).

As for the subject control reading of (9), Nunes (2009) follows Rodrigues (2004) and takes it to be derived as adjunct control, thus involving sideward movement of the embedded subject to matrix [Spec,vP] before the embedded clause is adjoined and becomes an island (as in (8)). Nunes takes the presence of de under this derivation to be spurious — a stylistic effect of the fact that the presence of de preceding finite clauses enjoys sociolinguistic prestige (which very often leads to hypercorrection). In other words, from the perspective of Nunes’s (2009) proposal, the subject control reading of finite clauses embedded under verbs of the convencer-type should always involve the same derivational steps associated with adjunct control, regardless of the presence or absence of the preposition de.

Bueno Dias (2023) points out two serious problems for taking de to be just a stylistic mark in the case of subject control reading in sentences like (9). Firstly, as a putative spurious element extraneous to the syntactic computation, de should not interfere with wh-extraction from within finite clause embedded under verbs like convencer. The contrasts in (11) below indicate that this is not the case: if de is absent, the embedded clause behaves like an adjunct island, falling under the pattern seen in (7); if de is present, wh-extraction is permitted even if the embedded null subject is controlled by the matrix subject. In other words, the presence of de actually forces a complement analysis for the embedded clause even under the subject control reading.

Secondly, with infinitival sentences embedded under verbs like convencer, the preposition de cannot be considered spurious because it cannot be deleted. Again, the presence of the preposition renders both subject and object control readings available: as illustrated in (12).

So, the question before us is why de enforces a complement interpretation for clauses embedded under verbs like convencer and how it allows both subject and object control readings. In particular, under the movement approach to embedded null subjects in BP, how can the subject control reading be obtained in structures with de without inducing a minimality violation?7

Constructions with the preposition-like element de associated with verbs like convencer have two salient properties in BP. First, de introduces an expression that looks like an “optional” argument, which may be nominal or clausal, as illustrated in (13) below. And second, if the added expression is clausal (be it finite or infinitival), it behaves like a complement rather than an adjunct with respect to extraction, as illustrated in (14).

Following Bueno Dias (2023), we take these facts to indicate that de in constructions with verbs like convencer in BP is a low applicative head (e.g., Cuervo, 2003; Pylkkänen, 2008). In particular, de looks like an Affected Applicative head in the sense of Cuervo (2003). Different from low applicatives that express possession, in the case of Affected Applicatives, the applied object is affected by the event depicted by the verb. In (15) below, for example, Cuervo argues that Valeria is generated in [Spec,ApplP], receives inherent Case from the Appl head, and is interpreted as being affected in virtue of receiving the radio broken. Similarly, the specifier of de in (16), for instance, is affected by the attempt at pregnancy (regardless of the controller for embedded null subject).

The relevant structures for (13c-d) are then as depicted in (17).

In (17), the clausal complement of de is arguably φ-complete, for its subject is licensed (with nominative Case) within it; accordingly, o Ivo is externally merged in the specifier of de. Recall, though, that regardless of whether CP is infinitival or finite, the morphological makeup of its verbal agreement inflection may be opaque in BP, being compatible with a complete or an incomplete φ-set (Ferreira, 2000, 2009; Nunes, 2008, 2019). In case the T head of the clausal complement of de is φ-incomplete, the embedded subject remains active for A-movement purposes and may move to the specifier of de, as illustrated in (18) below, yielding what may be thought of as “applicative control”. Crucially, a CP with a φ-incomplete T head does not count as a strong phase for the purposes of the PIC (see section 2).

Also following Bueno Dias (2023), we take the derivational step in (18) to underlie the object control reading of sentences like (19a) and (19b) below, for example, as respectively outlined in (20a) and (20b). Like the externally merged DP in [Spec, ApplP] in (17a) and (17b), the moved embedded subject in [Spec, ApplP] in (20a) and (20b) may have its Case valued after agreeing with the matrix v, which is later introduced in the derivation.

As can be easily seen, this analysis only provides an account for the object control reading of the sentences in (19). Under a movement approach, the subject control reading of the sentences should be obtained through movement of a Ana from the embedded subject position to the matrix [Spec, vP], as respectively sketched in (21) below. However, all things being equal, the intervening DP in [Spec, ApplP] should block such movement.

Recall that a sideward movement analysis like the one proposed by Rodrigues (2004) to account for convencer-constructions with finite clauses not preceded by de is of no use here. Sideward movement proceeds from a root syntactic object to another root syntactic object (see (8)) and this is not the case in (21). Moreover, the embedded finite CP behaves like a complement when preceded by de, but like an adjunct when lacking de (see (11)). In other words, the complementation structure in (21) allows A’-movement from within the embedded clause, thus capturing long distance wh-extraction in (11), but not the A-movement of the embedded subject depicted in (21), which should underlie the acceptable subject control reading of the sentences in (19).

To circumvent the minimality problem illustrated in (21), Bueno Dias (2023) proposes a smuggling approach in the sense of Collins (2005a, 2005b). Details aside, the derivation of the subject control reading of (19a), for instance, should proceed along the lines of (22) below. In (22a), v agrees with o Ivo, valuing its Case and rendering it inert for A-movement. In (22b), the embedded TP of (22a) moves to the matrix [Spec,vP], smuggling a Ana, which may then move to [Spec,VoiceP], without being blocked by o Ivo (see (22c)). The final steps involve no minimality violation either: VP undergoes remnant movement to an XP projection above vP, as shown in (22d), and a Ana finally moves from [Spec,VoiceP] to [Spec,TP], as shown in (22e).

Although the smuggling step in (22b) does circumvent the minimality problem of (21a), the additional movements required to derive the desired word order receive no independent motivation. Moreover, as pointed out by a reviewer, the response in (23B) below strongly suggests that TP forms a constituent with de que, which is straightforwardly captured by (21a), but not by (22e).

In sum, the structures in (21) have two virtues: their complement structure allows long distance wh-extraction (see (11)) and they are thematically parallel to the corresponding structures that underlie object control readings (see (20)). So, the challenge is to keep these welcome results with the structures postulated in (21) and find another way to circumvent their flagrant minimality problem. This is what we pursue in the next section.

Nunes (2008, 2010) has argued that elements assigned inherent Case do not induce intervention effects for A-movement (see McGuinnis 1998). Assuming that the object experiencer verb strike in English, for instance, assigns inherent Case to the experiencer in its Spec (annotated by shadow lettering), movement of the subject of the small clause in (24b) below across the experiencer does not yield a minimality effect. Boeckx et al. (2010, p. 174) suggest that a similar derivation is available for subject control with promise-type verbs in English. In a sentence such as (25a), for instance, promise assigns inherent Case to its Spec, allowing movement of the embedded subject to the matrix external argument position without inducing a minimality effect, as illustrated in (25b).

Going back to our original discussion, inherent Case can be used as the key ingredient to rescue the otherwise crashing derivations sketched in (21). As an introducer of arguments, the Appl head is arguably capable of assigning inherent Case (see (15)). Let us then suppose that the applicative head de may optionally assign inherent Case to its Spec. In the case of (21), this amounts to saying that if o Ivo receives inherent Case from de, it does not count as a proper intervener for movement of a Ana to the matrix external argument position, as illustrated in (26) below. In other words, the derivation of subject control readings with verbs like convencer in BP in (26) mirrors the derivation of subject control with verbs like promise in English sketched in (25).

This proposal makes an interesting prediction. Besides being lexically defined, inherent Case is usually associated with additional (sometimes idiosyncratic) restrictions. For instance, Larson (1991) observes that the object of subject control constructions with promise contrasts with the object of verbs like persuade in that it resists undergoing Heavy NP Shift, as shown in (27) below.

We have already seen that the object of promise can be analyzed as being inherently Case marked (see (25b)); that being so, the contrast in (27) may be interpreted as showing that expressions marked with inherent Case resist undergoing Heavy NP Shift, as opposed to expressions marked with structural Case. In turn, if this restriction also applies to the inherent Case assigned by the Appl head de, what appears to be the matrix object (i.e., the element in [Spec, ApplP]) may display a different syntactic behavior, depending on whether we have subject or object control.

Let us consider why this is so. Like what we find in other instances of inherent Case assignment (e.g., Belletti, 1988), inherent Case assignment by the Appl head de to its Spec is optional. Therefore, in the case of object control, the DP in [Spec, ApplP] actually has two possibilities for being Case licensed: it may be assigned inherent Case by the Appl head or it may be licensed with structural Case by agreeing with v. Thus, lack of inherent Case assignment does not cause the derivation to crash, for the DP in [Spec, ApplP] may still be licensed with structural Case, as sketched in (28).

By contrast, in subject control structures, de must assign inherent Case to its Spec, as sketched in (29) below; otherwise, the embedded subject will not be able to move across [Spec, ApplP] and the derivation crashes, for the Case feature of the embedded subject remains unvalued.

Thus, restrictions applying to elements marked with inherent Case are expected to apply to the “matrix object” of convencer-structures with de if we have subject control, but not necessarily if we have object control, for inherent Case assignment is a sine qua non condition for the subject control derivation, but not for the object control derivation.

This prediction is fulfilled. Bueno Dias (2023, p.75) shows that the matrix object of convencer-constructions with de can undergo Heavy NP Shift in object control structures, but not in subject control structures, as respectively illustrated in (30) and (31).

In (31) we have subject control, which entails that the “matrix object” has been assigned inherent Case by de; hence, it is not compatible with Heavy NP Shift. In (30), on the other hand, we have object control and the matrix object may or may not be assigned inherent Case; if it isn’t, it can undergo Heavy NP Shift, which is generally available to direct objects in BP.

In sum, inherent Case not only provides a way out of the minimality problem for the movement of the embedded subject across the matrix object in subject control structures, but also provides a rationale to account for the surprising contrast between (30) and (31). It should be stressed that the predicate estar grávida ‘to be pregnant-FEM.SG’ and its gender and number agreement induce a pragmatic bias towards a unique possibility for the interpretation of the embedded null subject. This indicates that the problem in (31) is not with the identification of the null subject, but rather the extraposition restriction on inherently Case-marked elements.8

At first sight, the complex paradigm in (1), repeated here in (32), can be simply captured in lexical terms: convince is an object control verb when selecting an infinitival headed by a ‘to’ (see (32a)), a subject control verb when selecting a finite clause (see (32b)), or be underspecified, allowing both subject and object control when selecting a clause preceded by de ‘of’ (see (32c-d)).

This simple what-you-see-is-what-you-get approach however fails to capture complex correlations such as island effects depending on the presence or absence of de (see (11)) or the availability of Heavy NP Shift depending on the type of control one has (see (30)/(31)). Even more relevant, data such as (11) and (30)/(31) are too complex to be acquired by children based solely on primary linguistic data. We have argued instead that a movement approach can adequately capture the intricate paradigm in (32) by couching the availability of subject or object control on independent restrictions on A-movement.