2.1 Infinitival questions

Overt wh-phrases in English are hosted at the edge of CP, where they satisfy an Edge Feature of the C phase head. Therefore, the presence of a wh-phrase implies the presence of a phasal C. Landau (2024a) argues that in (8) it is impossible to read the embedded accusative pronoun as referring to the matrix subject. In his terms, this is because control by the matrix subject instantiates obligatory control and disjoint reference is triggered. If so, obligatory control is possible across a phasal CP boundary, creating a problem for FC theory (or the MTC, cf. Landau 2003).

Now, arbitrary control (i.e. the possibility for the empty subject of infinitives to have generic reference) is widely assumed to be allowed in infinitival questions. Examples used to establish this take the form in (9), in which the embedded anaphor (a generic) is mismatched with respect to the potential controller (here, the matrix subject), as illustrated in (9) (Landau 2013: 159).

Since arbitrary control is generally accepted as a diagnostic of NOC, (8) and (9) seem to contradict one another. Landau (2013: 159–160) argues that (9) is not a genuine instance or arbitrary control (i.e. generic interpretation of the empty subject). He supports this conclusion with additional data, in particular the impossibility for the empty subject of infinitival relatives to be non-locally bound, as in (10). For Landau (2013), instead, the well-formedness of the oneself-examples requires the reference of the embedded subject to include the matrix subject. Landau concludes that interrogative complements are “a species” of partial control (see section 5 below) and not NOC.

Let me observe preliminarily that, though partial control is attested under attitude verbs, it is not reported to license oneself-anaphora, cf. (11). It is true that Landau (2013), in the quote reported above, mentions that what we find in interrogatives is “a species” of PC. However, in the absence of a more precise characterization, the discussion cannot be taken any further.

Unification theories face the question whether the idea that interrogatives are instances of NOC can be maintained face to the facts in (8) or (10). Recent accounts of NOC analyze it as logophoric binding (McFadden & Sundaresan 2018 in a minimal pronoun framework, Manzini & Roussou 2025 for a FC-compatible analysis).¹ According to Charnavel (2020), logophors require an antecedent that is an attitude holder or empathy focus (cf. also Sundaresan 2021). This means that logophors can only have human antecedents. By default, if no attitude or psych predicate is present, the logophoric center is taken to be the speaker. Turning to infinitival questions, in the Italian examples in (12) the verbs establish, determine etc. take an infinitival interrogative complement and a non-animate subject DP. The latter is not a possible antecedent for logophoric binding. In that case, the embedded subject (which must be human) is disjoint in reference with the matrix subject. The same disjoint reading is possible with the same matrix verbs even if their subject is animate, as in (13).

The data in (10) remains problematic, since in case more than one attitude holder is present in the sentence, all of them are predicted to be possible logophoric binders (Charnavel 2020), not just the nearest one. In my judgement intervention by a non-attitude holder (an inanimate or subject of a non-attitude predicate) improves long-distance binding, as in (14), which would mean at least that long-distance binding per se is not a problem.

Summarizing so far, Landau (2013) advocates treating infinitival questions as instances of obligatory control and oneself-anaphora as a sort of partial control. Since questions are CPs, this would mean that obligatory control is found inside phasal nodes, contrary to what FC theory predicts. However, I have argued that oneself-anaphora does not have the same properties as partial control. On the contrary, infinitival questions can be analyzed in terms of NOC, itself understood as logophoric binding.² If so, infinitival questions are compatible with the conclusion that there is no obligatory control inside phasal CPs.

2.2 Overt complementizers

A second set of facts mentioned by Landau is that in several languages control complements are embedded under the same overt complementizers as finite clauses, which are uncontroversially phasal. In the two examples in (15)-(16) quoted by Landau (2024a: 18) one can easily recognize the demonstrative finite complementizer of Germanic languages (att) and the wh-complementizer (ke), also found in Romance languages.

Landau acknowledges that in cartographic conceptions of C, the same complementizer may instantiate distinct positions. For instance, one may hypothesize that the att/ke sentential introducers in (15)-(16) are exponents of a non-phasal Fin head, while they are exponents of a phasal Force head (Rizzi 1997) in non-control environments.³

Landau fails to mention that as long as a given complementizer also introduces raising sentences, it cannot be an exponent of phasal C, since raising complements must be non-phasal in any minimalist analysis. To illustrate this point, let us consider the two languages in (15)-(16). According to Karimi (2018: 194), Persian does not have raising, so that we cannot verify whether it also presents ke or not.⁴ In Swedish, however, according to Thrainsson (1993: 1853–186), att precedes both control complements (to the exclusion of modal verbs) and raising complement. These data are confirmed by Wiklund (2007), whose raising verbs include predicates classically taken to be ambiguous between raising and control (see section 3.2 below), but also vara på väg lit. ‘be on way’ (i.e. ‘be about to’) which contains a copula and can only be a raising structure (Moro 1997). Swedish att then embeds raising sentences and therefore cannot be an exponent of phasal C.

For the same reason, Swedish is also a counterexample to Landau’s (2003: 488) generalization that “control complements may be introduced by complementizers; raising complements are never introduced by complementizers”, offered as a presumptive universal. Finite control languages also are a counterexample, for instance Greek where so-called subjunctive particles embed raising (Iatridou 1993 on Greek) as well as control (Landau 2004). Within Romance, a counterexample comes from Salento varieties of Italo-Romance (Calabrese 1992; Manzini & Savoia 2005). These are languages with Balkan-style finite control and raising (hyperraising). They have two different wh-complementizers, ka and ku, as in (17), one of which, ka, embeds CP declaratives, as in (17a), while ku turns up in control environments, as in (17b). As expected, the control ku complementizer also appears in the progressive, as in (18), which must be a raising structure, given that the copula is a raising verb. This confirms that the ku complementizer is an exponent of non-phasal C.

Data supporting Landau’s (2003) generalization come from French sembler ‘seem’, or the comparable Italian examples in (19). The verb seem has both a raising and a control construal. The latter must be embedded under a prepositional complementizer (19a), the former cannot (19b).

The problem here is that languages like Italian and French have a certain number of verbs that admit both a raising and a control reading (cf. section 3.2 below). The double reading is most clearly detectable with some attitude verbs (Ruwet 1972 and subsequent literature). Under the control reading in (a), the matrix subject is an attitude holder. This alternates with a raising reading in (b)-(c), as defined by the usual diagnostics (expletive subjects, idioms).

The point of interest for the present discussion is that verbs allowing raising and control have the same prepositional complementizer in both construals. The alternation between the presence or absence of a complementizer with ‘seem’ therefore is may be the exception rather than the rule.

Landau also observes that in some languages, OC complements count as propositional in Wurmbrand & Lohninger’s (2023) sense. Some examples are in (22); in the Romance languages the repertory of these propositional OC examples is quite large. Practically any attitude or communication verb that can take a finite complement (especially a subjunctive one) can take an infinitival control complement, cf. (23).

Landau argues that “whatever cartographic baggage is required for believe that… would also be required for believe to…” (2024a: 18). But this depends on endorsing a tenet of cartography which Culicover & Jackendoff (2005) call Uniformity – namely that the same meaning always corresponds to the same syntax. In the absence of a commitment to Uniformity, the evidence falls short of introducing a problem for FC theory (or the MTC).

In short, arguing against the FC theory of control on the basis of the phasal nature of the control complement requires more than pointing to the presence of (finite) complementizers. One must show that the same complementizers must be missing from raising contexts – and this is not verified in the languages quoted by Landau (2024a).⁵ On the contrary, as pointed out by Boeckx et al. (2010: 79), unification theories predict that a structure available for control should in principle be available for raising and vice versa – hence for instance that there is a parallel between the availability of finite control and of hyper-raising. This prediction does not follow under the TTC. The empirical discussion appears to be open at this point. Unification theories face a potential problem if we find a bona fide complementizer heading phasal and control CPs but excluded from raising complements. Vice versa, non-unification approaches must explain why structural correspondences between raising and control appear quite systematic, rather than merely coincidental.

3.1 The attitude and non-attitude classes of control verbs

A challenge to FC theory concerns the distribution of lexical subjects in non-finite sentences – i.e. whether they alternate with control or not. The classical treatment of control/lexical subject alternations in generative syntax (Chomsky 1981; 1995) is in terms of Case. Therefore, what is being questioned is the interaction between the Copy relation and Case theory – or in the terms of Chomsky (2021) Transitivity (TR). The crucial examples are those in (24).

In order to account for the grammaticality of (24b-c), Chomsky assumes that the Copy relation applies optionally at Transfer.⁶ If it applies, control is obtained, as in (24c), otherwise the ECM/raising to object sentence in (24b) is derived. In turn, the contrast between (24a) and (24b) can be accounted for in terms of Case, assuming that expect assigns accusative Case (i.e. it is associated with a φ-probe entering Agree with the embedded subject) – while this is not so for try. Chomsky (2021) in fact seems to want to avoid Case, imputing to expect, but not to try, a Transitivity property. I will treat the two as essentially equivalent.

The first objection raised by Landau (2024a: 12) is that verbs like try do occur with overt accusative complements, as in Did you try our new infusion? More to the point, many of the relevant verbs, such as aspectuals, take event nominalizations, requiring syntactic transitivity, see (25b) vs. (25c).

In terms of Case, the contrast between (25c) and (25a-b) can be described as implying an alternation between a v including a φ-probe in (25c) and a v not including a φ-probe in (25a-b). Many such alternations are known: causatives/anticausative alternations, psych-verb alternations (The situation worries John vs. John worries at the situation), alternations between accusative and oblique objects (I forgot the milk/about the milk, cf. *I forgot the milk about the groceries). It is not unnatural to assume that verbs of the try/begin class are construed as transitive with DP complements, but are construed as intransitive, i.e. lacking accusative Case/a φ-probe, with sentential complements.

The possibility of ECM in complements of verb like expect in (24b-c) is attributed to the fact that the matrix verb keeps transitive/accusative properties (a probe on v) when embedding an IP. We may further assume that some verbs, in addition to IP complements, take for-to infinitives, i.e. full CP complements, as seen in (26b) with hope.

The assumptions needed to account for English control/lexical subject alternations in English are schematized in Table 1. The non-transitive/non-accusative nature of try (begin, etc.) or hope (decide, etc.) accounts for the obligatoriness of control in their IP complements. In turn, try and hope differ because hope but not try takes CP complements, namely for-to ones. Both these classes contrast with ECM verbs (expect, want etc.) several of which also select for-to complements.

Even if the assumptions in Table 1 can be made to hold, they are rather stipulative and complex. Landau argues that a simpler, more explanatory account is provided by the TTC. As outlined in section 1, the TTC divides the domain of control predicates into attitude and non-attitude predicates. According to Landau, the correct generalization concerning lexical subjects in non-finite clauses is based on these two classes, as in (27).

In Landau’s terms the second part of (27) follows from semantic compositionality, since a lexical subject would turn the complement into a proposition, while a non-attitude verb requires a property as a complement (Landau 2024a: 15). In other words, a semantic constraint excludes (24a). Vice versa, attitude verbs, which undergo logophoric control, not predication control, are free to take lexical subjects, as stated in the first part of the generalization.

Furthermore, Landau points out that the complex picture presented by English infinitives is considerably simplified, in the same language, if instead of infinitival complements we consider gerunds – which freely alternate between empty and overt subjects. The assumption is that lexical subjects are licensed internally to the gerund. The data in (28) show that attitude verbs alternate between control and lexical subjects, while non-attitude verbs only allow control.

From the point of view of explanatory adequacy, Landau argues that in his account “we do understand the obligatoriness of control with the entire class of non-attitude predicates” (p. 16). He is referring to the fact that in his account there is a one-to-one mapping between the semantic class of non-attitude verbs and predication control. Unification theories cannot mimic the TTC’s results in this respect. In unification theories, obligatory control corresponds to IP complementation and lack of accusative Case both for attitude and non-attitude predicate – and in fact also for raising to subject. What is more, from an empirical point of view, gerunds are a good argument that (27), and in particular the impossibility of lexical subjects with non-attitude verbs, holds independently of Case/transitivity. Under the TTC gerunds and infinitives are predicted to pattern alike under (27). Under FC theory gerunds cannot be accounted for, given the irrelevance of Transitivity/accusativity in the sense of Table 1.

Now, the Romance languages lack gerunds of the English type. Rather, infinitival sentences alternate with event nominals with those control verbs that take a DP object. We observe that with non-attitude predicates, event nominals require coreference of their subject with the matrix subject, as in (29). This is not true with attitude predicates, as in (30).

As far as I can tell, English follows the same patterns as Italian (i.e. one can attempt one’s own escape but not somebody else’s – and one cannot force John to my breaking the law. In any event, Italian suffices for my argument.⁷

The fact that the subject of the embedded event DP in (29) must be interpreted as controlled by the subject of the non-attitude verb does not follow under the TTC any more than under FC theory. For, a DP complement does not denote a proposition (a truth-value) independently of whether it has a controlled subject or not. In other words, an embedded DP satisfies non-propositional complementation in any case (as acknowledged by Landau 2024a: n. 9). Hence non-propositional complementation is not sufficient to ensure the correct construal of event nominalizations under non-attitude verbs. It follows that the effects in (29) must have a different source from non-propositional selection also in the TTC. An alternative is a meaning postulate, in the sense of Chierchia (1984: 308). Following ideas of Montague, Chierchia proposes that the proposition John tries to run away entails that John runs away (if John’s attempt succeeds). This entailment is expressed by a meaning postulate, which as Chierchia says, is part of the metaphysics (or underlying ontology) of a language.⁸

Going back to FC theory, the weakness of the system outlined in Table 1 is that no ECM or for-to is possible just with those matrix verbs that do not allow that-clauses or gerunds with overt subjects. The meaning postulate provides an independent reason why this is so. In Table 2, I set the Transitivity parameter at the negative value for the try-class, because the meaning postulate does not block the case in which an embedded lexical subject is anaphoric to a matrix controller (e.g. *John attempted himself to be the winner vs. John expected himself to be the winner).⁹

In short, I propose that a semantic constraint on non-attitude verbs (a meaning postulate, or whatever turns out to be appropriate), forces obligatory coreference of the embedded subject with a designated matrix argument, even in the absence of syntactic control. This constraint is motivated in particular by nominalization data. I argue that in this respect, the TTC faces the same issues as FC theory, since nominals are outside the TTC’s predicate/proposition selection mechanism.¹⁰

3.2 Dual control/raising verbs

As we saw in the previous section, a core conceptual difference between unification and non-unification approaches to control is that in non-unification analyses, and specifically under the TTC, raising and control verbs have two different structures. By contrast, for unification theories, raising and control have the same derived structure (and for the MTC even the same derivation) modulo the distribution of theta-roles. Specifically, for Chomsky (2021: 20) the Markovian property of derivations means that “the [derivational] history is not preserved in the current state” – in other words the Copy relation is indifferent to the derivation and only the theta-calculus distinguishes raising and control interpretations of Copy pairs.

Various verbs are classically taken by the literature to alternate control and raising readings, as illustrated with threaten and promise in (20)-(21) above (cf. Ruwet 1972). Another example are modal verbs, which in the Romance languages are main verbs, taking infinitival complements. It may be assumed that the root meaning corresponds to a control structure (two theta-roles), and the epistemic meaning to a raising structure (one theta-role), cf. (31).

If we take Landau’s (2024b) list of predication control verbs in (2) at face value, the double construal (control and raising) equally affects aspectual verbs, since they all uncontroversially have raising contruals (embedded weather verbs, existentials, etc). If anything, the question may be raised whether they also have control (see below). Some implicative verbs, e.g. manage, pass the test of synonymy under passive with inanimate subject (raising construal), as in (33), though not necessarily with animate subjects (control construal), as in (32).¹¹

A control verb that does not seem to admit a raising reading, i.e. stubbornly imposes selection restrictions on subjects, is want as in example (34) from Cinque (2006). However, examples like (35a) are perfectly idiomatic, even though stains do not have desires, intentions or other mental states implied by control ‘want’. The meaning of (35a) is a sort of future ‘This stain will not go away’, best accessed in the presence of a negation.¹² The contrast between (34) and (35a) may be due to the fact that (34) embeds a stative verb, making the reading just indicated for (35a) difficult to coerce. As for other raising tests, meteorological verbs are acceptable. Idioms are already degraded. Under a reasonably conservative construal of these facts, then, a raising reading can be coerced for Italian want, to a limited extent.

I take the data in (20)-(21) and in (31)-(35) to show that there is a partially productive phenomenon of raising and control coexisting as complements of the same verb. The literature on this phenomenon is mostly interested in restructuring predicates (in the sense of Rizzi 1982); restructuring is however orthogonal to the present discussion. Zubizarreta (1982) considers the root/epistemic ambiguity of modals, concluding that all modals are raising verbs which assign no theta-role to their subject (p. 164). She argues that the root meaning comes about because the subject is assigned what Zubizarreta calls an adjunct theta-role (p. 133), akin to the roles assigned by secondary predicates. She also takes the adjunct theta-role to characterize the control reading of other alternating verbs like menacer ‘threaten’, promettre ‘promise’ (p. 84). Cinque (2006) treats all verbs that display any raising properties, including modals, aspectuals and implicatives, as pure raising verbs. Nevertheless, he still resorts to the adjunct theta-roles of Zubizarreta or some equivalent notion (Cinque 2006: 29), to characterize the stubborn control properties of want and also of motion verbs.¹³ The fact that Cinque (2006) is invoked by subsequent literature, e.g. Amato (2024), to justify a uniform raising treatment of all restructuring verbs, therefore overlooks a part of his discussion.¹⁴

Under unification theories we actually expect to find that raising and control coexist as complements of the same verb – or in any event we can easily accommodate this fact, without introducing new constructs into the theory. In particular, with attitude verbs the possibility or not of a control/raising alternation depends simply on whether the attitude holder role can be cancelled out or not, allowing or preventing raising readings. In general, the possibility of projecting an External Argument or not yields the control/raising alternation – crucially without any other structural changes. A class of pure raising verbs (happen) resist any alternation, because they are lexically characterized by the lack of a theta-role. However, control verbs are potentially ambiguous with raising. Though not all control verbs admit a raising frame, or can be coerced into it to some extent, the phenomenon is productive enough not to be dismissed as mere lexical idiosyncrasy.

The same alternations raise questions for the TTC, specifically as they apply to non-attitude verbs. Suppose we take Landau’s (2024b) list of predication control verbs at face value, and therefore modal/aspectual/implicative verbs have a predication frame. The availability of these control predicates to raising forces them to have an alternative selection frame triggering raising. But this in turn means that non-attitude control verbs (in the absence of restructuring, e.g. in English) must take a propositional complement, since raising has a propositional source in current theories. Hence non-propositional embedding cannot in fact be forced for the class of non-attitude verbs – and the generalization in (27) can no longer be derived from the fact that non-attitude predicates take non-propositional complement.

An obvious alternative route is offered by the Cinquean line of literature briefly reviewed above, under which any verb that has some raising behavior is a raising verb (and therefore the list of non-attitude control verbs in (2) must be radically revised). However, the control/raising divide, that the Cinquean approch eliminates in favor of raising, seems to me to be somehow reintroduced in the form of adjunct theta-roles or equivalent semantic constraints. In unification theories, as we just saw, control and raising can be simply differentiated by assigning or removing an ordinary theta-role.

5.1 The state of the question

A typical example of partial control is (44a) where the matrix controller is singular, but the embedded verb requires a plural subject. The sentence is well-formed, if the embedded subject is interpreted as a set including the matrix subject as well as the speaker or other individuals. Attitude verbs allow this reading, but non-attitude verbs yield degraded results as in (44b).

Apart from the usual attitude/non-attitude divide, on which I simply refer the reader to section 3.1, a more important reason why Landau raises the partial control matter is that, as he puts it, there is no partial raising, cf. (45). Under FC theory, Transfer sees a Copy pair, not a history of derivation. So we need to clarify why the lower member of the Copy pair can be understood as a plural (the subject plus some associated individual) in control but not in raising.

The interpretation involved in PC (i.e. the subject plus a set of associates) is not available in main/finite clauses either, cf. (46a). This is subject to crosslinguistic variation. In (46b) I illustrate associative agreement with an Italo-Romance variety (from Silvestri forthcoming). In (47) the singular verb agrees with a plural verb via the associative interpretation (Rita + Rita’s set).¹⁹ Standard Italian patterns with English (see (58) below).

Landau (2016) further argues that in English the mismatch involved in PC is semantic (singular controller, semantically plural predicate); the presence of a morphological plural makes the sentence ungrammatical as in (47a). A fortiori, a person mismatch is impossible, as in (47b).

Nevertheless, grammatical mismatches are possible in other languages, in particular in the Romance languages (Sheehan 2014). In Italian, for instance, overtly plural floating quantifiers can be embedded in PC complements, cf. (48a). Furthermore, person mismatch is also possible, cf. (48b), though somewhat marginal, or perhaps just colloquial. The acceptability of (48b) tends to exclude that person mismatch have anything to do with inflected infinitives, with which it has previously been observed in Portuguese (Sheehan 2018).

An account of partial control that has been proposed within the MTC, the Covert Comitative analysis (Boeckx et al. 2010: 185; Sheehan 2014 for the Romance languages), holds that examples like (49) are well-formed to the extent that they contain a silent comitative.²⁰

The main problem with this account is that it must be constrained to apply only in control clauses embedded under attitude verbs. Apart from the fact that this is stipulative, it is not entirely clear what the scope of the stipulation is. Certainly, in order to allow (49) but not (46) we must assume that for some reason comitatives can be silent in control but not elsewhere. It turns out however that (46) does not improve if the comitative is overt. For instance, in Italian it is impossible to have associative agreement of the Calabrian type and introducing an overt comitative is no help, cf. (50).

Let me then present some new data, which I do not find in the literature on associative agreement (or Romance ‘un-agreement’). Italian allows a singular subject with an added comitative (also in the singular) to agree in the plural in examples like (51). The examples are colloquial, but strongly contrast with (50). Examples like (51) show that in Italian associative agreement is licensed by a comitative construed with the sentential subject, whose interpretation is akin to that of a coordination.

The comitative forms a constituent with the DP, as shown for instance by (52a). The comitative inside the DP can furthermore cooccur with a comitative adjunct, as in (52b).

In short, there is some support for the licencing of grammatical plurals via overt comitative modifiers independently of control. The stipulation as to the fact that covert comitatives are restricted to controlled subjects is uneliminable.

Landau’s (2016) proposal is to reduce PC to associative agreement. His analysis assumes the existence of an Associative Morpheme (AM). In English partial control, he assumes that AM is adjoined to v where the function of this morpheme is to expand the subject argument of the predicate to a group (Landau 2016: 97). This licenses semantically plural predicates like meet on the LF branch of grammar. In other languages, like Italian here, PC is compatible with syntactic agreement of AM, given that semantic and morphological agreement are matched. As far as I can tell, the fact that associative agreement does not apply in main or tensed sentences in English must be stipulated.

Landau (2015: 77) further argues that the TTC provides a straightforward answer as to why partial control is blocked with non-attitude OC verbs. Those take predicative complements, hence “predicative control cannot be partial because predication cannot be so”. However, I am not clear whether this can be derived in terms of the AM – in other words how to block the possibility an AM could be added to the vP of a predication control structure yielding PC with non-attitude verbs. Similarly, as for the impossibility of partial raising, I am unable to compute why an AM could not be associated with a vP whose subject undergoes raising.

Finally, Chomsky et al. (2023) mention an account for partial control based on deletion of an embedded for-DP phrase – possible in (53a) but not in (53b).

Following this line of explanation, the challenge as to why there is partial control but not partial raising would disappear, since of course there is no source comparable to (53b) for a raising sentence. However, Landau (2024a) gives various arguments against the approach in (53). In particular, he points out that any reasonable construal of Recoverability of Deletion, i.e. deletion under Identity, will disallow deriving (44) from (50).

Other problems are that it is not necessary for a verb to take a for-to complement for it to allow PC – and anyway for-DP subjects are a feature of English not found in many languages, which nevertheless have partial control with the same properties as English. The Romance languages are a case in point, see (48). The third point made by Landau is that languages where the PC subject triggers grammatical plurality, e.g. Italian in (48), may not be problematic for a deletion account; however the licensing of semantic and not of morphological plurality in English (47) requires an explanation. Fourth, one must ensure that the pronominal subject being deleted includes the controller in its reference, since subjects of for-to clauses can be disjoint in reference with the matrix subject. Lastly, Landau points out that partial control shares with OC the bound variable reading of the controllee. Thus, John arranged to meet at noon and Peter did too only has the sloppy reading of the second conjunct. If deletion of a pronoun was involved, we would expect a different semantics, in particular strict identity in the second conjunct. These various points are addressed (as potentially relevant to any account of PC) at the end of section 5.2.

5.2 PC under FC theory

I will adopt silent comitatives which are the best developed tool within a unification framework. The main challenge for unification theories is to derive that while there is partial control, there is no partial raising. Under FC theory, Transfer sees a Copy pair, not a history of derivation. So we need to clarify why the lower member of the pair can be construed with a silent comitative in control but not in raising. The only possible answer lies precisely in derivation history, since under FC theory, raising undergoes an IM derivation, but not control.

Examples like (51) suggest that the silent comitative is to be construed with the sentential subject, whose interpretation is akin to that of a coordination. Now, raising out of a coordination is impossible because of the Coordinate Structure Constraint (CSC), see for instance (54a). Similarly, in (54b) if the comitative is an NP adjunct the target of IM is a non-constituent, cf. (54c). For practical reasons, I run the derivations below on English examples, though (54b) is really an English rendering of Italian.

I suggest that the raising example in (55) including a covert comitative is impossible for the same reason as (54), namely a violation of the CSC or structure dependency, as applied to IM.

If raising in (55) is blocked by a derivational constraint, holding of IM, we do not expect it to hold of Copy environments not created by IM, such as control. We may therefore expect partial control to be possible, even though partial raising is not. Let’s consider how this may work concretely. We immediately hit the problem that it is impossible to strand an overt comitative under Copy, for example in (56). Recall that control depends on independent EM of the controller and controllee – therefore nothing prevents the comitative adjunct to be merged in the lower and not in higher copy. Nevertheless, deleting the lower copy in the pair <the chair, the chair> is impossible. This means that deletion must also affect constituents (see for instance the analysis of fragments by Merchant 2004), blocking deletion of the non-constituent and accounting for (56).

Now, we get to the crucial point, namely why a silent/deleted comitative would make a difference with respect to the ungrammatical (56). By hypothesis, in this case the comitative is a pro, i.e. a pronoun that does undergo spell-out, i.e. it is technically deleted (see especially Saab 2009 for deletion as No Spell-Out).

I propose that if deletion is a transfer operation, what the computation sees in (57) is a deleted constituent, namely the whole DP. The violation imputed to (62) therefore disappears since the whole constituent is not spell-out.

An obvious problem remains, namely why the empty comitative could not be moved to the matrix subject position or EM-ed there as in (58), voiding my preceding discussion of the grammaticality contrast between raising and control. The problem takes us back to the unresolved question why the associative or silent comitative reading is only compatible with infinitival subjects in the languages we are considering.

Evidently, silent comitatives are only licensed in unprobed position. This restriction must be stipulated (for languages like English or Italian) – but this is true of alternative technology (namely the AM) as well, as far as I can see.

Let me turn to the other properties of PC mentioned at the end of section 5.1, as listed in Landau’s criticisms to Chomsky et al. (2023). First, there is the matter of PC being possible only with attitude verbs. On this I have nothing to add to the discussion of section 3. The impossibility of PC under non-attitude verbs must fall under the same meaning postulate as the impossibility of lexical subjects. Next, there are the semantic issues, basically ensuring that PC has a bound variable reading (of the controlled subset) and not a pronominal reading. As far as I can tell, as long as Copy applies, as it does in the derivation of (57), the bound variable reading of the controlled element is predicted, including obligatory sloppy identity.

We must then consider the English/Italian variation, with respect to whether overt plural agreement is possible, e.g. with floating quantifiers or with reflexives, as in Italian (48), or not, as in English (47). This variation seems to be connected with the fact the in Italian, comitative modification of the DP is sufficient to establish plural agreement with a finite verb, while this is not possible in English (cf. *I with my wife are going to the sea).²¹ In any event, in Italian comitative modification of DP yields a syntactic plural, covertly as in (48) or overtly as in (51). In English, comitative modification is licensed only as an interpretive phenomenon, hence only covertly and as reflected by semantic plurality phenomena. A different question is why this description would hold. Landau (2016) has an explanation, namely that Agree (as reflected by plural inflection) is a PF phenomenon and for this reason it is insensitive to the Associative Morpheme yielding PC in his terms. However, if we say that Agree at PF holds universally, then we are left with no principled reason why Italian is different. We could also postulate a major parameter (Agree in syntax in Italian and at PF in English). But in order to claim an explanatory advantage we would have to motivate this parameter independently (which has not been done, to my knowledge).

In conclusion, while generally the MTC and FC theory yield equivalent results, PC is a phenomenon with respect to which Landau may be right about the MTC – I do not see how to avoid partial raising if partial control is also movement. FC theory differentiates control and raising sufficiently to be able to see how the PC problem could be circumvented at least in principle. Important problems remain open for all accounts proposed in the literature, in particular the fact that associative agreement is not available (in many languages) with subjects of finite sentences.