Clitic climbing does not always exhibit the type of free variation illustrated in (2). In what follows, I will make reference to various contexts in which CC is impeded, and finish this subsection by concentrating on the main data point of this squib.

First, it is well known that there is a great deal of dialectal variation with regard to the phenomenon in question. In the Romance languages spoken in the North of Italy, for instance, clitic climbing is generally banned, in contrast to what happens in the South. Hence, while the (standard) Italian spoken in Northern Italy favors non-clitic climbing, the Italian spoken in the South typically opts for clitic climbing instead (see, among others, Kayne 1992; 2013 for empirical evidence). For Spanish, RAE-ASALE (2009: 1235) notes that clitic climbing is more robust in speech and in informal registers (on the issue of variation, see Cinque 2006 and Sitaridou et al. 2015).

Leaving aside variation for the time being, it is also the case that certain intervening elements, such as adverbs, complementizers, negation, and most prepositions, tend to block climbing (see Ordóñez 2013 and Gallego 2016 for evidence; see RAE-ASALE 2009: 1239 for exceptions). One illustrative example is the well-known contrast in (3), which includes the verb querer ‘want’ plus an infinitive –a quintessential CC environment. Here, the presence of the negation between the inflected verb and the infinitive causes the clitic lo to stay low.

This type of evidence has been brought to bear on different issues, depending on the theoretical perspective adopted. On the one hand, for those advocating that clitic climbing is the result of restructuring in a clause-union environment, the presence of the negation would disrupt the necessary relationship between the inflected verb and the infinitive, thus breaking the verbal complex characteristic of monoclausal sentences and, in turn, disallowing transparency effects such as clitic climbing. Alternatively, for those who assume biclausality in cases like (3), the presence of no impedes movement of the clitic to the inflectional head Tº. Thus, such data have been taken to be an argument in favor of the movement analysis of clitics, as they display sensitivity to locality constraints on movement.

Similarly, with certain predicates, such as causatives, clitic climbing in Spanish is subject to a poorly understood but well-documented animacy restriction, due to which animate clitics cannot appear with the high verb (Fernández-Soriano 1999; 2016; RAE-ASALE 2009; Ordóñez 2013).5 This is exemplified in (4)a, where the low clitic is an animate (more specifically, [+human]) direct object. This example stands in glaring contrast to (4)b, with an inanimate/[–human] one. Note that the dative clitic is an argument of the higher predicate and therefore does not originate within the lower VP, and that the non-CC versions of both sentences are perfectly acceptable (i.e., Te hizo invitarlo/comprarlo).

Lastly, the novel case that this squib is concerned with, exemplified in (1) and illustrated again by the examples in (5), is particularly interesting because contrary to what happens in (4), it displays no animacy restriction.6 This is straightforwardly corroborated by (6)a, where the CCed direct object clitic lo is coreferential with a preverbal [+human] direct object. Thus, although a problem clearly arises in (5)a, the issue cannot be animacy, unlike in (4)a.

The periphrastic case of quedarse + -ing verb is considered by RAE-ASALE (2009: 2211) to be a semi-lexicalized construction, since quedarse in this context strongly tends to co-occur with the -ing verbal forms mirando ‘looking’ and viendo ‘seeing.’7 The configuration is said to be frequent in areas such as Mexico, Central America, and the Caribbean, especially with the accusative forms le/les (see below), although it is also found in other parts of the Spanish-speaking world, such as Spain.89

In the next section I introduce in some detail the ingredients necessary for the analysis of the contrasting examples in (5)a and (5)b to be pursued here.

Despite the general preference to privilege the highest copy created by movement in PF, a sizeable body of research has shown that on occasion, a non-high copy in a movement chain has to be pronounced in order for the derivation to be convergent. In other words, there are cases when phonologically realizing the leftmost copy causes a problem in PF; nonetheless, this issue can be sidestepped by favoring a different, lower copy (i.e., by pronouncing the element in question in a different position), while keeping movement in the (narrow) syntax.

A number of works have offered a host of empirical arguments from different constructions and from different languages to this effect, including cases of A-movement, A-bar movement, head movement, and remnant movement (e.g., Bobaljik 1994; 2002; Pesetsky 1997; Franks 1998; Bošković 2001 et seq.; Nunes 2004; 2011; Reglero 2004; Bošković & Nunes 2007; Kandybowicz 2008; Saab 2008; Villa-García 2013; 2015; in press; see also the collection of papers in Corver & Nunes 2007). The operation in question has been dubbed Pronounce Lower Copy (PLC) or Lower Copy Pronunciation (LCP). The strong version of PLC (Franks 1998) would have it that a low copy in a movement chain can only be pronounced if convergence so demands (i.e., if pronouncing the head of the chain causes a problem). Thus, the requirement to realize the highest copy of a non-trivial chain can be overridden iff the derivation fails in the event that the PF realization of the highest copy takes place. On this view, PLC is a last-resort operation effected to save a derivation that would otherwise crash.

One of the most compelling arguments for PLC comes from the behavior of wh-phrases in Romanian, a multiple-wh-fronting language (Bošković 2001; 2002; Bošković & Nunes 2007). In questions involving more than one question word, all the wh-phrases are normally fronted to the preverbal field. This is shown by (9).

By the same token, if the requirement that all wh-phrases in multiple-wh-questions be fronted in this language holds, one would expect all of the wh-phrases in a question to appear at the beginning. For instance, if we replace cine ‘who’ for ce ‘what’ to form the Romanian equivalent of What precedes what?, we would in principle obtain (10).

Nevertheless, in contrast to (9), the question in (10) is unacceptable for most native speakers, as it creates an issue vis-à-vis phonologically identical forms, which are barred (*ce ce).11 Can the meaning of (10) be expressed at all in Romanian? In other words, how does the system avoid the problem induced by (10)? Bošković (2002) shows that in cases like this, the object wh-phrase, ce, must appear postverbally. Put differently, ce seems to stay postverbally in a low position (i.e., as if it had not undergone movement to the preverbal field). This is indicated by (11), which is the grammatical counterpart of (10).

There are in principle two options for the derivation of (11). One is to assume that the relevant wh-constituent does not move to the left of the clause, despite evidence that wh-phrases ordinarily front in Romanian, as shown by (9) above:

In contrast, the reason for the putative absence of movement in this case seems to be morphophonological in nature, as has been noted, which begs the question of why the syntax would care about factors that pertain to PF. The analysis in (12), simplified by only showing the derivation of the constituents that are directly relevant to our discussion, would suffer from an acute ‘look-ahead’ problem, as ce₂ would not move so as to avoid a problem arising only once the derivation reaches PF.

Following Bošković’s (2001) lead, the other derivational possibility is to assume instead that syntactic movement in (11) actually takes place, in compliance with the requirement that all wh-phrases move in Romanian, as in (10)/(13). Critically, however, a problem ensues in the PF component if the highest copies of the moved wh-phrases are pronounced together (cf. (10)), as represented schematically in (14).12

Instead, this is one instance in which the system can –and in fact must– choose a low copy of a moved element in order to bypass a violation. This PLC-based analysis is shown in (15), which illustrates the simplified derivation of (11).

Therefore, (11) is an illustration of the intricate interaction between syntax and morphophonology, with a PF requirement taking precedence over a syntactic one. Contra the previously outlined account (cf. (12)), the PLC analysis solves the issue posed by (10) without making changes to the syntax of multiple-wh questions, without appealing to PF movement, and importantly without incurring a rather theoretically unappealing ‘look-ahead’ problem. As a result, the PLC alternative in (15) is preferable to the account in (12).

With special reference to Spanish and other Romance languages, Villa-García (2013; 2015) has extended this type of analysis to account for the notorious contrast in (16)a and (16)c, assuming that subjects are also amenable to a CTM account (Stjepanović 1999; Ortega-Santos 2006; Boeckx & Gallego 2008).

On this view, pronouncing a high copy of the subject, as in (16)b, would lead to an ill-formed outcome in PF (Zubizarreta 2012 actually attributes the ungrammaticality of (16)a to phonological reasons). This problem vanishes if a low copy of the subject is PF-privileged, as in (16)d, which is the derivation of the grammatical counterpart of (16)a, given in (16)c.13

In the next section, I argue that the novel facts presented in this squib regarding CC constitute further evidence for the CTM, and in particular for PLC.

A first assumption that seems natural and well justified on independent grounds is that Romance CC involves movement of at least one clitic from its base position in the VP (i.e., its canonical postverbal object position) to the TP domain (Kayne 1975; 1989; Roberts 2010, inter alia. Ordóñez 2013 offers various arguments militating in favor of the movement of clitics with particular reference to Spanish clitic climbing; see also Boeckx & Gallego 2007 and Gallego 2016). In this connection, Spanish provides direct evidence that there are various positions for the clitic in CC environments, for the same clitic can occur concurrently in different positions in some speakers’ colloquial speech, as well as in varieties such as Chilean Spanish (RAE-ASALE 2009: 1234; Mann 2012; González-Urzúa 2016). This is illustrated in (19); (19)b is of particular interest, since it involves the pattern investigated here:16

The standard assumption is that clitics that appear in proclitic positions with inflected verbs are heads that incorporate (or head adjoin) to the left of the host (Roberts 2010, inter alia).17 A run-of-the-mill case of proclisis would be analyzed as follows, where I take Spanish-style V-to-T movement for granted (see, e.g., Ordóñez 2015). Note that the analysis applies to both regular proclitics and CC proclitics:18

For the particular CC case this paper is concerned with, which involves the pronominal verb quedarse plus an -ing form of the lower verb, recall that the highest/leftmost clitic (CL₁) is part of the high verb and therefore does not undergo CC, and the second or low clitic (CL₂) is initially merged/construed with the lowest predicate.

It goes without saying that one of the main conundrums that the literature on cliticization phenomena has faced is how to explain the (apparent) optionality of CC in cases like (2). An authoritative line of research which began with the seminal work of Rizzi (1982) contends that CC is the result of a complex-predicate-formation operation that effectively creates a single clause (cf. monoclausality) (see Section 2). This process is generally referred to in the literature as restructuring (see, e.g., Wurmbrand & Lohninger 2019). Given that clitic placement is clause-bound, the fact that CC can occur across two predicates indicates that the two verbs occur in a restructured/monoclausal environment. Under this analysis, CC is obligatory because restructuring has occurred, as shown abstractly in (21). As noted above, clitics move to functional projections in the TP layer (presumably for feature-checking purposes, as in Rizzi 2000 and Roberts 2010, inter alia, though the actual motivation for clitic movement remains a topic of intensive research, as discussed by Gallego 2016, among many others). If the two verbal predicates in (21) form a unique complex predicate under clause union/monoclausality, it follows that the clitic will climb to the inflectional domain.

On this view, in cases where the clitic does not climb, as in (2)a above and (23)b below, restructuring has not taken place, as schematized in (22).

Interestingly, although in the case at issue, which features the verb quedarse + an -ing verb, both the CC and the non-CC options seem to be readily available (cf. (23)), some speakers actually seem to favor the CC option. More specifically, two Spanish speakers, one from Valladolid and the other from Zamora, prefer the version with CC, exemplified in (23)a. (Note that in general they prefer the le version of the clitic (cf. leísmo), but speakers of other dialects tend to use lo). This strong preference for CC in the structure of interest also holds for a Venezuelan consultant, who also employs le.19

But even if both (23)a and (23)b are available to most speakers, the underlying derivations of the two sentences are distinct, with only the restructuring analysis triggering climbing of the clitic (cf. (21)). The alternative order in which the clitic is low (i.e., in which CC has not occurred) is thus symptomatic that restructuring has not obtained (as claimed by the representation in (22)). In other words, in non-CC contexts, the two verbs remain independent from each other, hence preventing transparency effects like CC. This is substantiated by cases like the following, where the presence of an adverb disrupts the necessary adjacency between the two verbs for complex-predicate formation to occur. As a consequence, CC is impossible and the clitic associated with the lowest predicate mandatorily remains low and surfaces as an enclitic, in contrast to what happens in cases like (23), without the intervening adverbial.

Significantly, in the cases that this paper focuses on, illustrated again in (25) for convenience, there is no a priori reason to assume that restructuring does not occur, since no intervening material appears to break the relationship between quedarse and mirando (cf. (25)a vs. (24)a). Recall that in principle both the restructuring and the non-restructuring derivations are possible for most speakers, in light of (23) (cf. the derivations in (21) and (22)).

One plausible derivation here would be to simply assume a non-restructuring context in (25), as in (22). This would automatically yield the desired outcome, (25)b, which poses no problem at all, as the clitics would never be adjacent to one another under this readily available derivational option.

Let us now look at the more interesting case, however, under the reasonable assumption that the restructuring derivation is in principle freely available in (25), as in (21). Suppose for a moment that in (25), the accusative clitic needs to remain to the right of its predicate, as in (25)b, and that the reason is that in this case, restructuring is barred. This approach poses the question of what would prevent the restructuring derivation in (25). Clearly it would be far-fetched to claim that a morphophonological issue regarding the impossibility of certain clitic combinations such as *me me would impede complex-predicate formation, in particular because one would expect restructuring to occur before clitic combinations are assessed, arguably in the PF component; this account thus also faces a ‘look-ahead’ problem, in analogous fashion to the account of the Romanian facts sketched in (12). Moreover, it is not at all clear why illicit clitic combinations would have a bearing on whether predicates restructure or not. Therefore, deriving (25)b by stipulating that restructuring simply cannot take place in this specific case appears to be ad hoc and does not seem to be a promising account of the absence of CC in the data at hand, notwithstanding that in principle both the non-restructuring and restructuring derivations should be available.

Let us assume then that in theory the clitic is able to move up in the syntax in (25), at least as far as the available restructured context is concerned. The reason why some clitic combinations are not legitimate has been much debated in the literature, with accounts that range from Agree-based analyses to morphological approaches.20 Suppose then that the clitic needs to move for purposes of φ-feature satisfaction à la Roberts (2010). One approach would be to think that (i) this movement does occur, but then something happens which precludes the appearance of the clitic as a proclitic of the highest verb. An alternative would be to propose that (ii) the clitic does not move at all and stays low (i.e., it does not climb). Motivating (ii) is not obvious, since in principle the low clitic would move in order to satisfy its own features (see Bošković 2007). Furthermore, given that in such cases restructuring is in principle in place (i.e., there is no way to block clause-union syntactically), how to prevent transparency effects like clitic climbing is not apparent at all.21 Similarly, if it is PF that rules out certain clitic combinations, as seems reasonable, the familiar ‘look-ahead’ problem from the Romanian case in (10)/(12) would become ostensible here as well: “an apparently syntactic operation [i.e., CC] appears to be blocked just in case it would interfere with a post-syntactic operation” (Bobaljik 2002: 200; my addition in square brackets, JVG).

Thus, let us assume that in (25)b, movement of the low clitic does take place normally, as in (25)a (cf. (i)). As soon as the syntax does its job, the structure is sent to the interfaces. Once the structure reaches PF, the system assesses whether the output of syntactic operations adheres to the well-formedness conditions of the PF component. Because certain combinations of contiguous clitics such as *me me in (25)a and *me te in (18)a are ruled out in PF, then the only procedure that the system can resort to so as to rescue the derivation at this point is to privilege (i.e., pronounce) the low copy of the clitic. This is precisely why the clitic must occur low in (18) and (25). This derivation is illustrated in (26), with (26)a showing the particular case of (25)b and (26)b representing the simplified abstract derivation of the CTM/PLC system. (I deliberately ignore verb movement here to avoid visual clutter).

Recall that in this case restructuring is available, but it does not produce a legitimate output. In consequence, the derivation crashes in PF unless a salvaging strategy is employed, namely PLC. The Spanish non-CC examples in (1)b/(5)b/(18)b/(25)b are thus now cast as an environment where in principle CC is allowed to proceed syntactically, but crucially a low copy needs to be PF-favored in order to save a derivation, in much the same way as in the familiar Romanian case (cf. (11)/(15)). Incidentally, the copy that is privileged in PF is the same copy that is privileged in L(ogical)F(orm), as this is the clitic’s base position as the object of the lowest predicate.22

The CTM/PLC system offers a principled account of the contrast in (25) and has a number of added advantages, including the avoidance of a severe ‘look-ahead’ problem and the need to assume that the restructuring derivation is banned owing to putatively irrelevant (PF-related) illicit clitic clusters. The current analysis overcomes these issues and also dispenses with movement being effected in PF, the major theoretical significance of the CTM/PLC mechanism being that it enables PF to affect word order (by privileging a copy that is independently available) without having to invoke PF movement. In Section 4.4 I provide additional data from other Romance varieties in support of this approach. Before undertaking this task, however, I turn to the compatibility of the PLC system outlined here with alternative analyses of CC.

It is important to bear in mind that the approach just sketched actually does not commit us to assuming a restructuring/clause-union account, and would in fact also be compatible with a biclausal analysis of CC à la Kayne (1975) (see also Ordóñez 2013, among others). Under this type of account, sentences where intervening material prevents the clitic from surfacing as a proclitic, as in (3)/(24), are not seen as cases where restructuring is blocked, but rather as regular structures involving two CPs where intervening material (e.g., negº no) may be a hindrance to movement operations. Accordingly, pronouncing the lowest copy of the clitic due to the presence of offending material (such as negation) would be an instantiation of the more general rescue-by-resumption approach. On this view, phonologically realizing a resumptive pronoun has an ameliorating effect with respect to locality-of-movement violations, which implies that locality is to a certain extent PF-related. Put another way, locality violations can be fixed by phonologically realizing a copy within the barrier/island as a resumptive pronoun. The availability of this rescuing strategy has been independently argued for by Shlonsky (1992), Pesetsky (1997), Bošković (2001; 2011), and Bošković & Nunes (2007). The approach is illustrated in (27), with the negation barring CC in cases like (3)a, whose grammatical counterpart requires the clitic to surface low (cf. (3)b).

In the particular configuration that this paper is concerned with, (26), where the issue is not triggered by a locality problem, the CTM/PLC approach would be applied in exactly the same way as under the monoclausal analysis outlined in the preceding subsection, with lower-copy pronunciation of the clitic saving a derivation that would crash on account of an ill-formed clitic cluster in PF.23

The construction consisting of quedarse + an -ing verb is not the only clitic-related pattern where a different copy of a moved clitic may be chosen for purposes of PF compliance. Other CC environments that can in principle be analyzed as involving PLC include perception verbs like those in (28).

Again, *la la clusters are ungrammatical, a problem that fades away if the low clitic is pronounced in its base position as the object of cantar (cf. (28)b; V-la_ila_k V-la_k), in parallel fashion to what happens in (26).24 For reasons of space, I will not delve further into these potential cases here.

Before concluding this section, reference to other (Romance) languages is in order. First, it is of note that Catalan manifests a construction akin to the main basic paradigm this squib centers upon, as shown in (29), kindly provided to me by Josep Ausensi-Jiménez, Carolina González García, and Clàudia Pons-Moll.

This contrast, which is in principle also amenable to a PLC account, reveals that the pattern investigated here is not limited to Spanish.

Departing from the specific construction at issue, but still within the realm of CC, Bellinzonese, a Romance language spoken in Northern Italy, disallows two identical sa sa clitics in CC environments featuring a must-like modal, as (30)a shows. Importantly, however, the construction improves dramatically if the reflexive clitic is pronounced (i.e., PF-realized, in our terms) in its base position, as an enclitic of the lexical verb, as indicated by (30)b.25

Such cases lend further cross-linguistic plausibility to the PLC/CTM account of the contrasts with which this paper is concerned.

Lastly, as noted by an anonymous reviewer and by Simone De Cia (p.c.), it is important to note that contiguous identical clitic forms such as se se in languages like Sovramontino, another Romance language of the North of Italy, are legitimate, as shown in (31), which is a regular clitic cluster in a non-CC environment.

Data like (31) are compatible with the claim that the constraints that ban certain clitic sequences differ from language to language, with Catalan and Spanish observing the restriction in question, but not Sovramontino.26