Control is typically associated with non-inflected infinitives in European Portuguese. The starting point of the present paper is the observation of the pervasiveness of inflected infinitives as complements of object control verbs in European Portuguese (EP).

European Portuguese displays inflected infinitives, which are infinitival forms morphologically marked for person and number features. Inflected infinitives are typically non-control contexts, co-occurring with nominative overt subjects or pro subjects (Raposo 1987). However, the existence in EP of inflected infinitives which are apparently controlled has been noticed in the case of object control in standard dialects (Raposo 1989) and in the case of subject control under transitive verbs (Gonçalves & Santos & Duarte 2014, who associate these occurrences to colloquial/substandard varieties; Sheehan 2018). In this paper, we are concerned with the case of inflected infinitives in the complements of object control verbs. Moreover, we discuss the interpretation of these inflected infinitives in European Portuguese, even though studies on Brazilian Portuguese are mentioned when relevant.

Raposo (1989) seems to assume that the inflected infinitive under object control verbs corresponds to a control context. The same has been argued by Sheehan (2018) and Modesto (2010; 2018), the latter discussing Brazilian Portuguese. However, recently Barbosa (2021), discussing the behaviour of inflected infinitives under attitude verbs, including object control verbs, questions the conclusions reached by Sheehan and Modesto. When discussing object control verbs, she focuses on the cases of attitude verbs such as convencer ‘convince’ and persuadir ‘persuade’ (the same verbs that are considered by Sheehan and Modesto) and argues that inflected infinitives under these verbs do not show the behaviour of obligatory control contexts. Based on the facts that she enumerates, Barbosa argues that the impossibility of obligatorily controlled inflected infinitives under these object control verbs is expected if agreement blocks Control under attitude verbs, as argued by Landau (2015). Aiming at discussing attitude verbs, Barbosa (2021) only explores the interpretation of subjects of inflected infinitives under the object control verbs convencer ‘convince’ and persuadir ‘persuade’.1 The present paper broadens the types of object control verbs under discussion; by doing this, it confirms the intuitions of Barbosa, but it also asserts the existence of inflected infinitives under other object control verbs which maintain obligatory control readings.

It is therefore a major goal of this paper to make an empirical claim: inflected infinitives under a subclass of object control verbs create obligatory control (OC) contexts. By observing a split between what we can call resilient object control verbs and non-resilient object control verbs, we will draw a parallel with Landau’s (2015) divide between verbs inducing predicative and logophoric control. To the best of my knowledge, the present paper is the first discussing this type of difference in the interpretation of inflected infinitives under object control verbs. An explanation for these facts will also be suggested – the explanation provided can work as an additional argument for theories of Control that assume a heavy contribution of semantics in determining control readings (Jackendoff & Culicover, 2003; Landau 2015, to a different extent; also Stiebels 2007). This view contradicts some of the main assumptions underlying the Movement Theory of Control (Hornstein, 1999).

The Portuguese inflected infinitives are characterized by overt inflection marking person and number (with the exception of the 1^st and the 3^rd singular forms) and typically occur either with lexical subjects or with overt/null pronominal subjects (pro), i.e. they correspond to contexts in which Control is not observed. Therefore, whereas non-inflected infinitives in complement clauses exhibit the properties of Obligatory Control (OC), as it is the case of the non-inflected infinitival clause in the complement of lamentar ‘regret’ in (1a), inflected infinitives generally block control, as it is the case of the inflected infinitival clause under the same verb lamentar ‘regret’ in (1b).

This is the general position taken in the literature with respect to the contrast produced by the presence of inflection in infinitives in Portuguese under transitive verbs (see Raposo 1987; also Pires 2006): when in the complement of a transitive (subject) control verb, inflected infinitives create non-control environments.

However, the distribution of inflected infinitives has been itself an issue. Even though inflected infinitives generally occur only in embedded clauses (a distribution that they share with non-inflected infinitives), their distribution in embedded clauses is also restricted (see the initial discussion by Raposo 1987).2 According to the description of Raposo (1987), apart from subject and adjunct clauses (non-obligatory control contexts), inflected infinitives can occur as complements of epistemic, declarative and factive verbs. Gonçalves & Santos & Duarte (2014) argue that only verbs that do not specify the temporal orientation of its complement (as it is the case of lamentar ‘regret’ in (1)) can take an inflected infinitive clause as a complement. Barbosa (2021) instead argues that inflected infinitives are not available under restructuring verbs and in interrogative complements; in contrast, they are available in the complement of attitude non-restructuring predicates.

The specific case of inflected infinitives under object control verbs has been subject to a more limited discussion. The availability of inflected infinitives under object control verbs has been noticed by Raposo (1989) and by Madeira (1994); even though not discussing this, Raposo (1989) seems to assume that inflected infinitives under object control verbs are (obligatory) control structures, therefore with properties similar to controlled non-inflected infinitives. Both Modesto (2010; 2018) for Brazilian Portuguese and Sheehan (2018) for EP consider inflected infinitives under object control verbs on a pair with inflected infinitives under transitive (and ditransitive) subject control verbs. Sheehan (2018: 34) points out that inflected infinitives under object control verbs are widely accepted with an exhaustive reading (the reference of the controlled subject is exhausted by the controller), in contrast with contexts of subject control. Modesto (2010: 85) suggests that in BP the inflected infinitive under object control verbs (as in other contexts in BP) is required to signal a partial control reading – see (2).

The author shows the effect of the inflected infinitive in sentences that do not have an embedded collective predicate (such as reunir ‘meet’) and does not explicitly distinguish partial and split readings: a split reading (1+2 in example (2)) corresponds to the case in which the reference of the controlled subject is exhausted by the matrix arguments, but split between them (see Landau 2013: 172); the partial reading (2+ in example (2)) corresponds to a case in which the interpretation is less restricted: the controller is included in the reference of the controlled subject but does not exhaust it (Landau 2013: 157). Modesto (2018: 61), who quotes Negrão (1986) on this, suggests that speakers of BP who maintain verbal inflection (it is a well-known fact that verbal inflection is reduced in some varieties of BP), tend to obligatorily use inflected infinitives in object control structures.

This type of context for inflected infinitives deserves further attention. If it is proven that there are indeed controlled inflected infinitives under object control verbs, it would be interesting to understand why agreement would not block obligatory control in this context.

Recently, Barbosa (2021) has argued against the idea that inflected infinitive subjects are controlled in complements of convencer ‘convince’ and persuadir persuade’, the verbs that Modesto (2010; 2018) and Sheehan (2018) use to generally discuss this type of context. The argumentation of Barbosa is restricted to the complements of attitude verbs (subject and object control verbs) and the object control verbs convencer ‘convince’ and persuadir persuade’ fall in this class. She explicitly leaves the issue open in what concerns the status of inflected infinitives under non-attitude verbs. In this paper, we only consider object control verbs and it is the aim of the next section to contrast the attitude object control verbs that Barbosa discussed with other verbs, representative of another subclass of object control verbs.

The present section aims at establishing the main properties of inflected infinitives under different object control verbs in EP. To do this, we will test inflected infinitives under the object control verb obrigar ‘force’ (as well as a reduced number of other verbs with similar behaviour) and compare it to convencer ‘convince’, the verb systematically used by Barbosa (2021).3 We take these verbs to be representative of two types of verbs, an issue to which we will return to.

The starting point to establish the relevant criteria is what Landau (2013) calls the Obligatory Control (OC) signature. To this extent, we should be interested in determining if the antecedent (the controller) of the null subject of an inflected infinitive under the relevant object control verbs must be local and c-command the subject position (long-distance and non-c-commanding antecedents are excluded from OC). These are central criteria which come from the set of initial criteria for OC defined by Williams (1980) – at least the idea that there must be an antecedent and that it must c-command the controlled subject -, and these criteria are accepted independently of the analysis of Control which is assumed, either a non-movement (Landau 2013) or a movement approach (Hornstein 1999) – see Sheehan (2018) for a recent synthesis of diagnostic criteria taken from different trends in the literature.

We start here by testing the possibility of a long-distance controller. The example in (3) shows a contrast between an inflected and a non-inflected infinitive under convencer ‘convince’ (in accordance with Barbosa’s 2021 description). The subject of the inflected infinitive under convencer in (3b) does not behave as obligatorily controlled: it allows for a long-distance antecedent (the subject of the matrix clause as crianças ‘the children’).

However, this contrasts with what is found in (4), under the object control verbs obrigar ‘force’ and ajudar ‘help’ – in this case, a non-local antecedent is precluded, with both inflected and non-inflected infinitives (4a, b). In (4b), with an inflected infinitive, and since there is no 3^rd plural DP in a local position available as a controller, the use of the 3^rd plural form of the inflected infinitive can only be interpreted with a split control reading, a case in which its reference is exhausted by the arguments of the control verb in the superordinate clause; in the same case, a partial control reading (for instance a case in which one could understand that Maria and the children go to bed early) does not seem to be equally available (in the absence of a collective predicate). The object control verb impedir ‘prevent’, which selects for an infinitive complement introduced by the preposition de behaves as obrigar ‘force’ (5); the verb forçar ‘force’ also behaves in the same way. If a 3^rd plural controller is available, the sentence with the inflected infinitive is possible with an exhaustive control reading, as shown in (6); in this case, the split control reading seems to be available but not preferred (or equally salient).

The next set of sentences tests for the possibility of identifying a non-c-commanding antecedent of subjects of inflected infinitives under the same group of object control verbs. In the following sentences, the DP [a ama das crianças] c-commands the embedded subject; in contrast, the embedded DP [as crianças] does not c-command the embedded subject. Moreover, our knowledge of the world and the natural authority relations established between the different entities mentioned in the sentences may favour a reading in which children are going to bed early. We therefore test the extent to which syntax (the syntax of control) can block that reading. The sentences in (7) test this reading under convencer ‘convince’ with inflected and non-inflected infinitives. Whereas the subject of the non-inflected infinitive is controlled as expected – the controller is necessarily identified as the c-commanding DP [a ama das crianças] -, we confirm Barbosa’s (2001) insight in the interpretation of the subject of an inflected infinitive: in this case, the non c-commanding DP [as crianças] can get interpreted as the antecedent of the embedded subject; as an alternative, and given the 3^rd plural features of the inflected infinitive, we can get a partial control interpretation, a case in which the nanny and someone else (the children or other entity) will go to bed early.

However, an interesting contrast is obtained when we compare (7b) with (8b), with the matrix verb obrigar ‘force’ or the matrix verb ajudar ‘help’ and (9b), with the matrix object control verb impedir ‘prevent’.

The interpretation of (8b) and (9b) is more restricted and the subject of the inflected infinitive patterns with the subject of the non-inflected infinitive (8a, 9a) to the extent that, in both cases, taking the non-c-commanding DP [as crianças] as an antecedent of the embedded subject is not an option; the embedded subject must be controlled and the controller must be, as expected, local and in a c-commanding position. Since obrigar ‘force’ and ajudar ‘help’, as well as impedir ‘prevent’ are object control verbs, only the object [a ama das crianças] qualifies as a possible controller. The difference in the reading of the non-inflected vs. inflected infinitive only concerns the interpretation of the plural feature of the 3^rd plural inflected infinitive: in the latter case, the interpretation of the embedded subject is compatible with split control, being exhausted by the two matrix arguments (the nanny and the doctor), even though pragmatics does not favour this interpretation. In a more neutral context, the interpretation is easier to obtain (see 10); a true partial control interpretation seems more difficult to obtain (a similar example could be built with impedir ‘prevent’ or of course with forçar ‘force’).

As (11) also shows, if the controller is plural ([as primas das amigas]), the interpretation of the inflected and the non-inflected infinitive can be the same and the direct object controller exhausts the reference of the controlled subject – in this case, a split control reading is not excluded but also not made necessary. The sentence in (11), which presents a plural controller, also allows to confirm the interpretation of the subject of a non-inflected infinitive, since contrary to what happens in (8a), the form ir in this case could not be ambiguous between a non-inflected infinitive and a 3^rd singular inflected infinitive (a 3^rd singular inflected infinitive does not bear an overt morpheme).

As a last type of test, let us consider sloppy readings under ellipsis. Assuming that the controller must be local, we derive the obligatoriness of sloppy readings under ellipsis when a control structure gets elided (Landau 2013: 30). An obligatory sloppy reading under ellipsis has been accepted as a mark of obligatory control and Pires (2006) has used this criterion to show that inflected infinitives under subject control verbs do not maintain an obligatory control reading, i.e. they block control. The following examples are taken from Pires (2006: 94), who builds this test using a pseudo-stripping structure (I use here the term as defined by Depiante 2000):

Sheehan (2018) equally uses ellipsis, in this case VP ellipsis, to test the possibility of maintaining an obligatory control reading with an inflected infinitive. Barbosa (2021: 50) uses an example in which only an internal argument is omitted, whose ellipsis status is unclear and which involves a passive (example 43 in Barbosa 2021: 50) to argue that the subject of an inflected infinitive under convencer ‘convince’ gets a strict interpretation and therefore is not controlled. In what follows, I use pseudo-stripping, whose elliptical status is arguably clearer, to show that Barbosa’s intuition concerning the interpretation of the complement of convencer ‘convince’ is correct. Pseudo-stripping, which was already used by Pires (2006) in simpler contexts, is interesting for the present purposes since in this case the remnant is not necessarily a subject (see Depiante 2000; also Matos 1992 for EP), it can also be an object – something which is useful when we want to test the interpretation of the clausal embedded argument of convencer, which contains the infinitive, but we want to contrast the controller, which is another internal argument, an object DP. Also, according to Depiante (2000: 61), pseudo-stripping behaves as surface anaphora, i.e. a case which corresponds to syntactic structure projected but not pronounced (in the perspective of Merchant, 2001). In the case of pseudo-stripping, the remnant is interpreted contrastively and is possibly extracted to the left periphery. Therefore, the following example updates Barbosa (2021)’s example, now using pseudo-stripping. As we can see, taking [a mãe delas], the remnant of ellipsis, to be an object, the interpretation of the sentence is ambiguous, as it includes the possibility of a sloppy reading (paraphrase in a), as well as a strict reading (paraphrase in b).4

The same readings are maintained with stripping (Depiante 2000; see Matos 1992 for Portuguese) – in this case, the linear order of the DP and the negation is inverted (see 15).

In contrast, as expected, the counterpart with a non-inflected infinitive only allows the sloppy reading:

However, again the interesting contrast is the contrast with what happens when the matrix verb is an object control verb such as obrigar ‘force’ (see 17). In the case of pseudo-stripping, we should maintain the interpretation of [a mãe delas] as an object. In agreement with our previous observations, the strict reading is not available independently of the type of embedded infinitive, which means that the subject of the inflected infinitive must maintain a sloppy reading, i.e. it must have a local controller. The same pattern is found with ajudar ‘help’ (in 18) or with impedir ‘prevent’ (19).

This indeed confirms that the subject of an inflected infinitive under convencer ‘convince’ does not behave as controlled, as shown by Barbosa (2001). However, it also shows that there are subjects of inflected infinitives under object control verbs which behave as obligatorily controlled (when under obrigar ‘force’, ajudar ‘help’ or impedir ‘prevent’). Accordingly, and as a simple extension of what was already shown, we can observe that the subject of an inflected infinitive under convencer ‘convince’ can be disjoint in reference from any DP in the clause, contrary to what happens when the matrix verb is obrigar ‘force’ or ajudar ‘help’. This explains the possibility of the sentence in (20), with an inflected infinitive marked as 1^st plural, as well as the rejection of the correspondent case in (21), either with the matrix verb obrigar ‘force’ or the matrix verb ajudar ‘help’ (impedir ‘prevent’ patterns with the latter verbs).

The behaviour of the subject of the inflected infinitive in (20), under convencer ‘convince’, is suggestive of its pronominal nature. We should therefore test the possibility of alternation with an overt pronoun or a DP. As shown in (22), it is indeed possible to have the clausal complement position of convencer filled by an inflected infinitive clause with an overt pronoun or a full DP as subject. On the contrary, the same is not possible under obrigar or ajudar, as shown in (23) or under impedir ‘prevent’. The fact that the embedded subject is (and must be) postverbal in (22) is common to other classes of verbs selecting inflected infinitival clauses as complements (Raposo 1987).

The impossibility of alternation with an overt DP has been seen as a diagnostic criterion for OC (Williams 1980), and was equally explored by Modesto (2010) and Sheehan (2018), when discussing the status of the subjects of inflected infinitives in different varieties of Portuguese. The different behaviour of the inflected infinitives in (22) and (23) suggests that under convencer ‘convince’ we have a non-controlled inflected infinitive (with all the expected properties, including the ability to license a subject) but that under obrigar ‘force’ and ajudar ‘help’, we have a controlled inflected infinitive. Notice that the only possibility of having an overt pronoun in the subject position of the inflected infinitive under obrigar or ajudar is a case in which it is interpreted as a bound variable, as shown in (24). This is a possibility in OC structures in Portuguese, as shown by Barbosa (2009). As noticed by Barbosa (2018: 134, footnote 4), a paper centred on this type of subjects, the inflected infinitive is highly preferred in sentences such as (24) – in my judgement, in (24) it is obligatory.

The evidence presented until this point suggests that obrigar/forçar ‘force’ and ajudar ‘help’, as well as impedir ‘prevent’ are what we can call consistent (resilient) object control verbs, which maintain an OC interpretation of their infinitival complements, either non-inflected or inflected, whereas convencer does not show the same consistency. Along the same lines, it can be shown that a null subject of an inflected infinitive under convencer can be interpreted as having a discourse antecedent:

The same happens with a very common object control verb, the directive verb dizer (dizer para ‘tell to’), a case of indirect object control. As shown in (26), when occurring with an unambiguous form of non-inflected infinitive (as it is the case in (26a), given the absence of morphology in the verb and the 3^rd plural features of the controller), this verb behaves as an object control verb. A similar interpretation seems to be obtained when an inflected infinitive is used (see (26b)). However, given the right context, a non-controlled interpretation of the null subject of the inflected infinitive under this verb can be obtained (27). As shown in (28), an overt subject of the inflected infinitive is also possible under the directive dizer ‘tell’.5

It seems that dizer para (the directive dizer) is not a consistent object control verb, similarly to convencer ‘convince’. In this case, OC is not maintained with inflected infinitives.

Returning now to convencer ‘convince’, there is another dimension in which convencer does not show a consistent behaviour as an object control verb: given a relevant context, its interpretation can be coerced into a subject control interpretation, i.e. it allows control shift. The sentences in (29) and (30) exemplify this possibility – (30) was extracted from the public database CETEMPúblico6 and seems to favour a control shift reading:

In addition, another ditransitive verb which is recognized as easily allowing control shift, pedir ‘ask’, is also a verb which does not consistently behave as an object control verb in the relevant contexts. Control shift with pedir is exemplified in (31), a case in which both object and subject control interpretations are available. In (32), the preference for what looks like an object control reading is explained by the 3^rd plural morphology of the infinitive, only compatible with the features of the indirect object. The sentences in (33) and (34) show that the inflected infinitive under pedir is not obligatorily controlled.

In sum, it seems that looking at object control verbs in the light of their behaviour when taking inflected infinitive complements allows to identify two subclasses: verbs that generate a “resilient” object control environment and verbs which create less resilient contexts. The first group (the “resilient” class) seems to be a small group including obrigar ‘force’, forçar ‘force’, ajudar ‘help’ and impedir ‘prevent’, with the latter predicate selecting the preposition de instead of a. The second group (the “non-resilient” class) includes convencer ‘convince’, persuadir ‘persuade’, as well as dizer (para) ‘tell to’ and pedir (para) ‘ask’ and is probably less homogeneous, an issue we will return to. At least a subset of the non-resilient object control verbs allows control shift; the resilient control verbs do not. We will try to suggest an explanation for these facts in the following section.

The first relevant observation concerning the split in inflected infinitives under object control verbs identified in the preceding section relates this split with two different types of control defined in Landau (2015). The two verbs identified as paradigmatic in the preceding section (obrigar ‘force’ and convencer ‘convince’) are included by Landau (2015) in two different classes of verbs, associated to two different types of control: force is treated by Landau (2015) as part of a small set of object control verbs giving rise to predicative object control – this is a small subclass of verbs described as implicative and causative and also as non-attitude predicates (other elements identified as belonging to the class are compel, coerce and impose); in contrast, convince is classified by Landau (2015) as an attitude predicate giving rise to logophoric control – other object control verbs which Landau includes in the subgroup of attitude predicates, which occur in logophoric control structures, are persuade or dissuade (also psychological verbs) and communicative verbs, including tell and ask. It is not the aim of the present paper to fully discuss the attitude/non-attitude status of the different predicates placed by Landau (2015) in two different groups. Instead, I highlight the fact that Landau argues that there is a contrast in the effects of inflection on the two types of control, associated to the two different subgroups of predicates: predicative control holds in the presence of inflection, whereas logophoric control only holds when [+Agr] is absent (Landau’s OC-NC Generalization).

To this extent, the split found in EP data concerning the interpretation of inflected infinitives under obrigar ‘force’ and convencer ‘convince’ (which Landau associates to predicative control – the former – and to logophoric control – the latter) matches well with the predictions of Landau (2015). According to the author, the OC-NC generalization mentioned above is a formal constraint of grammar and there is no deep semantic reason that justifies the selective effect of agreement; the source of the explanation for the different effects of agreement in the different subsets of verbs is syntactic and comes from the different derivations of predicative vs. logophoric control. Landau’s assertions concern the full set of control verbs, namely, subject and object control verbs, and here we are zooming in on the restricted set of object control verbs. In what follows, I will try to establish a relation between the two subsets of object control verbs, their semantics and the availability of non-controlled inflected infinitives under these verbs. In doing this, the focus will be on the set of “resilient” object control verbs, i.e. those that maintain an obligatory control interpretation of their inflected infinitival complements.7

Landau (2015) classifies force as a non-attitude predicate, but causative and implicative – this verb corresponds to forçar or obrigar in EP. Even though we cannot deny that the interpretation of the sentence in (35a) might imply considering the mental state of the participant corresponding to the matrix subject, in (35b) no mental state is involved and it would be impossible to assume that this is an attitude context.8

However, the interpretation of obrigar or forçar ‘force’ is causative in both contexts and I will concentrate here on the causative interpretation. Discussing the expression of causation, Shibatani & Pardeshi (2002) argue for a causative continuum, from direct causation to indirect causation. Citing preceding work by Shibatani, the authors assume that indirect causation is associated to directive situations, namely situations in which the causer gives an instruction/direction to the causee. In general terms, the authors define “direct causation as a situation involving an agentive causer and a patientive causee” and “indirect causation as one involving two agentive participants, one an agentive causer and the other an agentive causee” (Shibatani & Pardeshi 2002: 140). Considering the verb deixar ‘let’ in EP, which is traditionally considered a syntactic causative, we could suggest that it can work both as a direct causative (36a) and as an indirect causative (36b) (in (36b), deixar ‘let’ also exemplifies what Shibatani & Pardeshi 2002 call permissive causation).

Given this type of definition, it is natural to include obrigar ‘force’, as well as convencer ‘convince’ or persuadir ‘persuadir’, in the set of causative verbs – in fact, Shibatani & Pardeshi (2002) treat cause, persuade and force as causatives, but causatives which retain their literal meaning, in contrast with the semantically bleached get, make or have. The idea of a causative continuum is akin to the position maintained by Nadathur & Lauer (2020) that different causatives express different causal relations and therefore that no unitary view of causation is desirable: “causation, at least as it is encoded in language, cannot be captured by a single type of dependency relationship, but instead reflects an umbrella notion, comprising a set of contrasting ‘bringing-about’ relations” (Nadathur & Lauer, 2020: 30).

Let us now consider in more detail the case of obrigar ‘force’. At a certain degree, we can identify similarities between the interpretation of obrigar and certain properties of the English causative make, as defined by Nadathur & Lauer (2020). First, there is a coercive implication associated with the meaning of obrigar: the causee did not make a free decision when bringing about the event in the embedded clause. Second, as the same authors suggest, there is a direct relation between the coercive implication and an interpretation of causal sufficiency: sentences with obrigar, such as (37) denote a causal dependence which can be defined as causal sufficiency, i.e. given the occurrence of the cause, the occurrence of the effect was guaranteed. Karttunen (1971: 340) states that “an implicative main verb carries a presupposition of necessary and sufficient condition which alone determines whether the event described in the complement took place”. In the case of obrigar ‘force’, only a sufficient condition is presupposed, as noticed by Kartunnen (1971: 357). According to Karttunen’s (1971) definition, this interpretation of sentences with obrigar explains the fact that obrigar entails the truth of its complement under positive polarity, as shown in (37a) – since Rui forcing Ana to eat the bean stew is a sufficient condition for Ana to eat the bean stew, the truth of the embedded proposition is entailed under positive polarity. However, Rui forcing Ana to eat the bean stew is not a necessary condition for Ana to eat the bean stew, explaining that obrigar does not entail the falsity of the embedded proposition under negative polarity, as shown in (37b) (but see Nadathur 2016 for a more complex discussion of causal sufficiency and implicativity).

This means that obrigar ‘force’ cannot be defined as an implicative in the stricter sense of Karttunen (1971), since under negative polarity, obrigar ‘force’ does not entail that the embedded proposition is false (37b). This places obrigar in the class of one-way implicatives, which are defined by Karttunen (2012) as yielding an entailment only under one polarity.9 Force is given by the author as an example of a one-way implicative, of a subtype corresponding to a causative, namely an implicative which yields an entailment under one polarity.10

Both ajudar ‘help’ and impedir ‘prevent’ can also be understood as part of a large set of verbs expressing cause and cause-like concepts, according to different models of causation (see Wolff & Song & Driscoll 2002). Specifically, in a “Force dynamics” framework, “causing” is analyzed into primitives which include both “helping” or “hindering” (Talmy 1988). For what is relevant for the present paper, it is also possible to show that ajudar ‘help’ and impedir ‘prevent’ behave as (one-way) implicatives, as it happens with obrigar ‘force’. This is shown in (38) for ajudar and in (39) for impedir, in the latter case with the relevant modification for a negative implicative (Kartunnen 1971; 2012).11

Now let us look at the case of convencer ‘convince’ and persuadir ‘persuade’. In the case of these predicates, a notion of causal influence is maintained, but the causal link between Rui’s action in (40) and the event of Ana eating the bean stew is more indirect: no causal sufficiency relation is expressed in sentences with convencer or persuade. The sentence in (40a) must mean that as a result of Rui’s action, Ana at some point made the decision of eating the bean stew, but this does not entail that Ana did eat the bean stew. In any case, and contrary to what happens under obrigar ‘force’, the causee, even if under the influence of the causer, makes a free decision when initiating the event described in the embedded clause. If nothing else is said, and if Ana made this decision, we can infer that she did eat the stew, but it might be that something else happened or that she changed her mind before completing the event and the stew was not eaten. In the case of convencer ‘convince’ or persuadir ‘persuade’, and again contrary to what happens with obrigar ‘force’, there is no coercive implication and there is not an interpretation of causal sufficiency, to the extent that given the occurrence of the cause, the occurrence of the effect is not guaranteed. As a result, the inference that the embedded proposition is true under positive polarity seems to be a Gricean implicature, which can be cancelled (see 40a); no entailment is found also under negative polarity (see 40b).

The difference between obrigar ‘force’ and convencer ‘convince’/persuadir ‘persuade’ is also in line with the analysis of Jackendoff & Culicover (2003), which is followed by Barbosa (2021) when discussing the behaviour of persuadir ‘persuade’. According to Jackendoff & Culicover (2003: 537), a verb such as persuade contains the predicate INTEND as part of its meaning and corresponds to ‘cause to come to intend’. This can be seen in (41). This contrasts with force, which would correspond to ‘cause to act’ (42).

According to Jackendoff & Culicover (2003), the predicate INTEND is a two-place function whose arguments are the intender and an action and, according to the representation in (41), a verb of intending should not allow a reading in which the intender is not the subject of the embedded action. However, Jackendoff & Culicover (2003:544), following and adapting Sag & Pollard (1991), suggest a mechanism of coercion which allows to reinterpret the complement as the action ‘bring about situation’. In this case, the possibility of a sentence such as (43) would be explained by the mechanism of coercion represented in (44).

This solution is adopted by Barbosa (2021) to explain cases in which the complement of convencer ‘convince’/persuadir ‘persuade’ is a non-controlled inflected infinitive. Therefore, this type of coercion would explain the possibility of non-controlled inflected infinitives under convencer ‘convince’: in (25), repeated here as (45), as well as in (46), the causee is not the agent in the embedded event.12

In (45) and (46), there still is a causal dependence between the matrix situation and the embedded situation, however the causee (the matrix direct object) is not the agent of the embedded event (and this is why it is not a controller, and no control structure is at stake here). According to the approach by Barbosa (2021), which uses the idea of bring about coercion of Jackendoff & Culicover (2003), the interpretation of (46) could result from the coercion mechanism represented in (47):

If we consider again the case of obrigar ‘force’, we understand that the same type of coercion is not available. This specific type of coercion is indeed not expected in this case if it is associated with the complement of INTEND, which is not present in the meaning of obrigar ‘force’ and other implicative object control verbs. We would add that an additional participant (associated to the interpretation of non-controlled inflected infinitives) is incompatible with the implicative entailment associated to the interpretation of this verb and this might be captured by an incompatibility between an implicative interpretation and the mechanism of coercion that derives the possibility of the interpretation corresponding to the non-controlled inflected infinitive complement.13 In general, the facts presented here support the idea that there is a semantic explanation for the resilience of obligatory control in inflected infinitives under obrigar ‘force’, which is not itself incompatible with the subclasses of control verbs identified by Landau (2015). The generalization would be that the implicative meaning of the verb would be correlated with the impossibility of the coercion which makes available a non-controlled reading. As for non-implicative verbs, more research would be needed to determine to what extent this type of coercion mechanism is possible.

Landau (2021), investigating gerund complements of prepositions, also more clearly insists on the relevance of the implicative/non-implicative contrast: namely, implicative gerund complements of P resist lexical subjects or partial control, whereas non-implicative do not. In that case, Landau (2021) considers that the implicative entailment, the ban on partial control and on lexical subjects result from the fact that in these implicative gerund complements control corresponds to direct predication. In the present paper, I am suggesting that the implicative meaning itself correlates with the impossibility of the semantic operation that would justify the occurrence of a non-controlled inflected infinitive. As it will be shown at the end of this section, the facts described in the present paper are not totally predicted by Landau’s analysis of predicative control.

A related issue is control shift, which we have shown to be available with convencer ‘convince’, but not with obrigar ‘force’. The sentence in (48), which repeats (29), exemplifies a case of control shift with convencer ‘convince’ (an equivalent sentence could be obtained with persuadir ‘persuade’). For control shift to obtain with these verbs and without an embedded passive, there must be some authority relation between the participant corresponding to the direct object (the mother) and the participant corresponding to the matrix subject (the children).

Farkas (1988: 49) signals the acceptance of control shift with persuade and convince in English when the infinitival complement is passive and presents sentences in which permission is explicitly involved:14

The same type of interpretation seems to be associated to control shift with pedir (ask) – see subject control in (50b), in contrast with (indirect) object control in (50a).15

Along the lines of Sag & Pollard (1991) and Jackendoff & Culicover (2003), the control shift interpretation can also result from semantic coercion – and Barbosa (2021: 65) observes that “the contexts that license an inflected infinitive with independent reference are strongly reminiscent of the contexts that license control shift”. Sag & Pollard treat the case of control shift with ask as a case of bring about coercion, whereas Jackendoff & Culicover (2003) argue that another type of coercion (a someone allow, ‘allow/enable X to’) is involved in the interpretation of these sentences. It is beyond the goals of the present paper to discuss the type of coercion involved in control shift with different types of predicates. The point here would rather be to signal that the same type of mechanism would be able to derive the control shift interpretations and the non-controlled interpretations of convencer ‘convince’ and persuadir ‘persuade’. In that case, the fact that implicative object control verbs such as obrigar ‘force’ do not allow control shift would be a consequence of the unavailability of the same type of coercion mechanism in the case of these verbs.

Another relevant result of the preceding section was the contrast between convencer ‘convince’ and obrigar ‘force’ with respect to the availability of partial and split control readings: whereas partial control, defined as a case in which the controller is included in the reference of the controlled subject but does not exhaust it (Landau 2013: 157), is available under non-implicative control verbs, the verbs in the subclass of obrigar ‘force’ only allow for split control, i.e. a more restrictive reading, in which the reference of the controlled subject is exhausted by (and split between) the matrix arguments, which c-command the embedded subject.

The possibility of partial control readings in the complement of non-implicative object control verbs is expected, under the hypothesis put forward by Landau (2015): predicates licensing logophoric control complements allow for partial control readings. In addition, in the case of the inflected infinitive complements of convencer ‘convince’ and persuadir ‘persuade’, if we accept that these are not instances of control, we can assume that what looks like partial control in these contexts is not a real case of partial control, if this is a case of an inflected infinitive which licenses a pro subject – this is the line of argumentation of Barbosa (2021), who suggests that semantic coercion, which explains the availability of non-controlled inflected infinitives under convencer ‘convince’, is also able to explain readings of the subject of embedded inflected infinitives which are apparent cases of partial control.

Implicative control verbs do not allow a partial control reading, but they induce split readings of a plural inflected infinitive, when the controller is singular. The fact that these are true split readings is shown by the fact that they license a reciprocal coindexed with the matrix arguments (see Jackendoff & Culicover 2003: 548): the following examples illustrate these facts with both obrigar ‘force’ and impedir ‘prevent’.

Although we have argued that inflected infinitives under the small class of implicative object control verbs do not block control, which is predicted by the hypothesis developed in Landau (2015), the fact that we find split control in this context is unexpected under Landau’s (2015) analysis of control: according to the author, control under these verbs is predicative control, it is derived as predication, and predicative control excludes split readings, since predication also does not allow split readings (as shown in 52).

We must finally consider another aspect in which we find a clear split between implicative object control verbs and other object control verbs: the behaviour of these verbs when combined with a finite complement. Convencer ‘convince’ occurs with a subjunctive finite complement which may take a null subject or an overt subject (53 a,b). In addition, the null subject has a free interpretation, as shown in (53c). Notice that we are still dealing here with the reading of convencer associated to Intent; the epistemic convencer takes indicative.

The sentences in (53) contrast with those presented in (54). First, with the matrix verb obrigar ‘force’, subjunctive complements with null subjects in (54a) and (54b), parallel to (53a), are felt as not natural. Another interesting contrast is found between (54c) and (54d): even though a lexical DP is possible in the subjunctive complement of obrigar (54d), this is only possible if the verb does not occur in a ditransitive structure (see the contrast between (54d) and (54c)). An example of the same type found in the press is presented in (54e).

Let us consider the contrast between (53a) and (54a,b). One could consider to justify the rejection of these sentences as an effect of obviation, understood as competition between infinitive and subjunctive, in the sense of Farkas (1992) (or, only to some extent, also Hornstein & San Martin 2001): the interpretation of the null subject in the subjunctive complement of obrigar is not distinct from a controlled subject in an infinitival complement of the same verb; this is not exactly the case of the subjunctive complement of convencer ‘convince’, given the right context, as shown in (53c) above.16 However, an explanation centered on the contrast between subjunctive and infinitive would not explain the following contrast:

(55a) repeats (54b) and suggests the rejection of the subjunctive complement with a null subject; (55b) repeats (54d) and shows acceptance of the subjunctive complement with an overt subject but in a structure which is not ditransitive, i.e. there is no direct object (there is even no space for an implicit argument in the interpretation of 55b). The paradigm is completed with (55 c): in this case, we show that the subjunctive complement with a null subject is accepted in the transitive structure (i.e. when no direct object is projected, in a structure parallel to 55b). This contrast is explained again if we resort to the semantics of the verb. If we adopt the view of Jackendoff & Culicover (2003), the same type of coercion mechanism allowing the occurrence of non-controlled inflected infinitives under convencer ‘convince’ would explain the availability of the finite complement. The fact that the semantics of obrigar ‘force’, linked to its interpretation as an (one-way) implicative, precludes the possibility of the same type of coercion explains the unavailability of (54a, c) or (55a).

As for the possibility of a finite complement to obrigar ‘force’ in a transitive structure, it is possible that it corresponds to a different configuration, which Jackendoff & Culicover (2003: 539) argue that some force-dynamic predicates can display: X CS [EVENT]. In this case, according to the authors, these verbs would look like raising-to-object/ECM verbs. Given the current discussion concerning the nature of Control and the suggestion that Control might result of movement (Hornstein, 1999), therefore not distinct from raising, it could be tempting to see here an argument for the analysis of control under implicative verbs as derived by raising. However, the availability of split control readings for obligatorily controlled inflected infinitives under these verbs precludes that line of argumentation. A movement analysis of control makes it difficult to explain the availability of split readings.17

A final consequence of what has been observed is that the behaviour of non-implicative object control verbs might be less homogeneous, if their acceptance of non-controlled inflected infinitives depends on (different types of) semantic coercion, as well as other conditions yet to be investigated. This is beyond the scope of the present paper, which focuses on the particular behaviour of the smaller group of implicative object control verbs.

In the present paper, we developed a view of a set of object control verbs as verbs occupying different places in a causative continuum and we linked this to different semantic properties of these verbs and to their behaviour in control structures. The idea that semantics is involved in controller choice is a relevant trend in the literature (see Farkas 1988; Stiebels 2007 for a review). According to the view of object control verbs sketched here, the lexical semantics of the verb must be involved in controller choice: if the object control verb is seen as a causative of the types described here, this must result in a control reading described as object control. This by itself would be a view deviating from the association of controller choice to strict syntactic locality, in the terms defined by the Movement Theory of Control (Hornstein 1999). However, if lexical semantics is also linked to the availability of controlled inflected infinitives, this is also an argument for a larger weight of semantics than the Movement Theory of Control allows. The interpretation of these structures must be explained by a model with a strong syntax-semantics interface, where coercion mechanisms can also be explained, as a result of different conditions, in certain cases also pragmatic conditions.

Finally, a note should be added concerning learnability. How can the fine-grained contrasts defined here between object control verbs be acquired? I would like to suggest that the mapping established between the number of lexically overt arguments which occur in a sentence and the semantic interpretation of the verb might be at the core of (or at least contribute to) an explanation for the acquisition of the semantic structure of these verbs (and consequently for the acquisition of the fine-grained semantic and syntactic restrictions that we have discussed). An important assumption of the syntactic bootstrapping hypothesis is that children are biased to link nouns in sentences to participant-roles in events (Gleitman 1990; see also a recent synthesis in Fisher & Jin & Scott 2020). It is possible that by hearing direct object control verbs such as obrigar ‘force’ and convencer ‘convince’ in different subcategorization frames (specifically in ditransitive vs. transitive frames with finite embedded complements), children can be guided towards having access to the subtle meaning differences between these verbs. A prediction might also follow: since children can only rely on positive evidence, it is possible that the default interpretation of an object control verb falls into the obrigar ‘force’ category and only upon accessing positive evidence (namely, sentences which describe situations with three participants) children assume that a particular verb may fall into the convencer ‘convince’ type.

This paper argues that not all object control verbs behave similarly with respect to the interpretation of inflected infinitives. In particular, we have shown that implicative object control verbs create resilient obligatory control contexts. In the resilient class, obligatory control is maintained even in the case of inflected infinitives and a plural inflected infinitive with a singular controller can be interpreted with a split control reading, but not a partial control reading. As for the non-implicative verbs analysed in the present paper, they belong to the non-resilient class, to the extent that inflected infinitives block obligatory control in their complements and both non-controlled and partial control readings are available; however, more research is probably needed to determine the behaviour of different non--implicative control verbs. To the extent of the argument developed here, the entailment works in one direction: if an object control verb is an implicative, its inflected infinitival complement resists non-controlled readings.

In general, a correlation was found between the semantics of implicative object control verbs and their behaviour with respect to the interpretation of inflected infinitive complements. The two types of verbs compared in this paper also fall in the two different classes defined by Landau (2015) for control verbs, corresponding to different types of control: according to Landau, the implicative obrigar ‘force’ induces predicative control and the non-implicative convencer ‘convince’ logophoric control. However, we have shown, contrary to Landau (2015), who suggests that split control is not possible with predicative control, but who does not discuss inflected infinitives, that split control is available as the interpretation of a plural inflected infinitive under an implicative object control verb in EP, which means that the analysis of predicative control and the hypothesis that the implicative meaning is correlated with control as direct predication cannot be accepted without further discussion. A more detailed discussion of this issue is left for future research.