Recomplementation as a paratactic phenomenon: Evidence from Spanish and English

We provide a variety of empirical arguments in favor of a paratactic account of recomplementation constructions, in which a left-dislocated element appears in between two complementizers. Contrary to integrated analyses assuming Complementizer Phrase (CP) recursion or Rizzi’s split periphery, we assume that the dislocated phrase is structurally independent from the embedded clause it precedes, which in turn is an elliptical sentence fragment. The juxtaposed fragmentary sentences are linked by the doubled complementizer, which serves to overtly flag a ‘restart’ in discourse. We show that this account makes a range of welcome predictions while sidestepping non-trivial problems that arise for integrated/cartographic analyses, which assume that dislocated XPs are in left-peripheral positions (such as Spec-TopicP) and that the doubled complementizer spells out Topic0. A further advantage of the approach is that it provides a handle on recomplementation constructions beyond the core cases involving left-dislocation, which reduce to a mere subcase of the general phenomenon of elliptical ‘restarts’ in discourse.

Although previous research has revealed many empirical properties of the construction (such as its iterative character, the inability of focal phrases to be flanked by complementizers, the possibility of multiple dislocates appearing between complementizers, and the island-inducing effect of the secondary complementizer), many questions surrounding the syntactic behavior of such constructions remain open and will be taken up here.Most extant analyses are framed within a cartographic or CP-recursion-based approach to the left periphery.Departing from such monosentential accounts, we argue that the issues raised by prior analyses can be circumvented on the assumption that recomplementation actually instantiates a discursive arrangement of two juxtaposed root clauses, which stand in a paratactic rather than hypotactic configuration.The analysis thus furthers our understanding of left-dislocation constructions more generally.
The remainder of the paper is organized as follows.In Section 2 we provide a brief overview of previous (monosentential/cartographic) accounts; we then move on, in Section 3, to present our paratactic account of recomplementation and its consequences.Finally, Section 4 offers some concluding remarks.

PREVIOUS ACCOUNTS
Accounts of recomplementation patterns in the generative tradition so far have assumed a multiclausal monosentential structure, where both complementizers and the sandwiched dislocate XP (… que/that XP que/that …) are bona fide constituents of an embedded clause.There are various versions of this general approach on the market, illustrated below for (1a): CP recursion (3a); Rizzian split CP and a TopicP position (3b); and Uriagereka's (1995)  A number of analyses of R-ClLD draw on Rizzi's cartographic approach, varying mainly with respect to the left-peripheral head assumed to be occupied by the second complementizer.For instance, for López (2009), the first complementizer sits in Force 0 , while the second one occupies Finiteness 0 .Martín-González (2002) makes a proposal that departs slightly from the traditional Rizzian approach, in that the first complementizer also occupies Force 0 , whereas the second one heads a (Doubled)ForceP situated below TopicP in Rizzi's periphery (i.e., ForceP > TopicP > (Doubled)ForceP > FocusP …); the sandwiched dislocate is in Spec-TopicP under this account.According to Rodríguez-Ramalle (2003), Villa-García (2012), and Martínez Vera (2019), the second complementizer occupies the head of TopicP, whose specifier hosts the dislocated XP, as shown in (3b).For reasons of space, we will not review each proposal separately here; the reader is referred to Villa-García (2015: Ch. 2) for a critical review of existing accounts.
The difference between the accounts in (3) is mostly terminological (a matter of labels); they are summarized by the abstract tree in (4): (4) V … ForceP/CP For'/C' que/that TopicP/CP/FP XP For'/C'/F' … que/that TP … Empirically, an analysis that assumes this kind of complex CP structure has several advantages.For the sake of illustration, we will concentrate on the TopicP account shown schematically in (3b).First, this analysis provides an elegant account of why phrases that can be left-dislocated can occupy the pre-secondary-complementizer position, but not foci, which do not undergo ClLD: (5) Dice que A NADIE (*, que) invitaron __. says that ACC nobody that invited 'S/He says that they didn't invite anybody.'If focal phrases like A NADIE in (5) occupy FocusP, then it follows from the peripheral template that they will not be able to occur before the secondary instance of que, which occupies Topic 0 (i.e., ForceP > TopicP > FocusP > FinitenessP …).
Moreover, the possibility of multiple ques after each dislocated phrase (Escribano 1991;Rodríguez-Ramalle 2003;Villa-García 2010, 2012, 2015;Radford 2018) is straightforwardly accommodated once TopicP is assumed to be recursive, as argued independently by Rizzi (1997), among many others (though see Benincà & Poletto 2004 for counterarguments): (6) Dice que al final (, que) a la niña, (, que) says that ATþTHE end that ACC the girl that la van a llamar Noelia.CL.ACC go to call Noelia 'S/he says that in the end, they are going to call the girl Noelia.' English also allows for such iterative complementation: (7) I don't think that for the sake of your own well-being (, that) if you are in a bilingual classroom (, that) once you have completed the homework in one language (, that) you should have to do it all over again in the second one.(Radford 2018: 126, attributed to Jim McCloskey) Adopting the structure in (3b), Spanish (6) would be analyzed thus: Despite their appeal, however, integrated/monosentential analyses of R-ClLD configurations such as those in (3) face major challenges.We will briefly outline these challenges below; they will be revisited and discussed more thoroughly in the course of the discussion of our paratactic-bisentential analysis in the following section: (i) intonation; (ii) connectivity (in relation to case/theta-roles and binding) and anticonnectivity (including lack of Condition B effects and seemingly doubled subjects); (iii) opacity for extraction (i.e., islandhood); (iv) dependence on first/high complementizer; (v) non-distinctness of complementizers; (vi) selectional restrictions.
As the examples provided so far show, a comma typically appears before the secondary complementizer (i).This comma is not arbitrary, but an orthographic rendition of the fact that the dislocated XP and the secondary complementizer are separated by a salient intonational boundary, typically represented by a pause before the second complementizer, which tends to be indicated in writing by means of a comma before que/that in the examples (Villa-García 2012, 2015, 2019;Frank 2016Frank , 2020;;Radford 2018;Martínez Vera 2019).On an integrated, multiclausalmonosentential account, it is not obvious how to derive the presence of this intonational break without stipulation, in particular under those accounts that assume that the dislocate and the second complementizer occupy the specifier and head of the same projection, respectively (Kempchinsky 2013;Radford 2018) As far as connectivity and anticonnectivity (ii) are concerned, dislocates in R-ClLD exhibit properties that seemingly diagnose movement and base-generation simultaneously (a paradox first noted for ClLD generally in Cinque 1990).Regarding the former, contra the judgments reported in Villa-García (2012), local anaphors contained in a dislocate can be bound by an element occurring after the secondary complementizer: (10) Me dicen que esa foto de sí mismo i , que al CL.DAT say that that photo of himself that DATþTHE duque i no le gusta nada.duke not CL.DAT pleases nothing 'They tell me that the duke detests that picture of himself.' Villa-García (2012, 2015) has argued that the dislocate in R-ClLD is base-generated in the sandwiched position in between complementizers, since the secondary complementizer blocks extraction from the lower clause (iii).However, data such as (10) challenge this conclusion, suggesting that reconstruction of the dislocate is not systematically absent. 2 In addition, note that the dislocate is case-marked, as shown in (1a), repeated here for convenience: (11) Dice que a tu madre (, que) la invitaron.(= (1a)) says that ACC your mother that CL.ACC invited 'S/He says that your mother, that they invited her.' If the dislocate is directly merged in the specifier of a left-peripheral projection headed by the second complementizer, then how the dislocate receives its case and theta-role is mysterious under standard assumptions, since the relevant case/theta assigner is lower in the structure and does not c-command the dislocate at any point in the derivation. 3 At the same time, invoking movement of the XP to a left-peripheral specifier is at variance with the non-movement properties of R-ClLD and ClLD in general (see Ott 2014Ott , 2015 for detailed discussion); furthermore, it is unclear why such movement could not feed further, successive-cyclic movement to the higher clause.We turn to these properties of R-ClLD directly.
Regarding anticonnectivity (non-movement properties), two major issues arise for traditional monosentential analyses of R-ClLD.First of all, the dislocated phrase flanked by complementizers co-occurs with a coreferring lower pronominal/clitic correlate (i.e., a tu madre k 'your mother'la k 'her' in (1a)/ (11) above).On a monosentential analysis, it remains unclear why this configuration does not incur a Condition B violation (at least where the correlate is not further embedded).Put differently, no Condition B effect arises that would be expected if the dislocate locally c-commanded its pronominal correlate, as shown below for (1a)/(11): (12)  Similarly, as illustrated in ( 13), R-ClLD can seemingly lead to doubling of the subject, which is generally impossible in null-subject languages such as Spanish (Barbosa 2009), but common in non-null-subject languages like English (cf.MARY K , SHE K IS COOL): (13) Dice que Hugo k , que él k no es así.
says that Hugo that he not is thus 'S/He says that Hugo k , that he k is not like that.' (spontaneous speech, Asturias, Spain, reported in Villa-García 2019) Monoclausal analyses like ( 14) are forced to tolerate doubled subjects in such cases.
Note that Condition B is again not violated, as we already saw above for cases such as (11).The above problems are special cases of the general problems posed by ClLD for monosentential analyses, viz. the simultaneous occurrence of a dislocated XP and a pronominal correlate in what is hypothesized to be a single, integrated structure.Overall, the fact that the dislocate displays connectivity and anticonnectivity simultaneously yields a paradox that monosentential analyses are unfit to resolve (Ott 2014(Ott , 2015)).
We now turn to problems posed specifically by the recomplementation variety of ClLD.Regular ClLD (without recomplementation) generally permits extraction across the dislocate into the main clause (Uribe-Extebarria 1991; In any case, under a recursive-CP or TopicP-based account, for instance, the assumption that the lexicalization of C 0 /Topic 0 creates a locality violation that is suspended in its absence is necessarily stipulative (though see Villa-García 2012 for an analysis of this kind).For this reason, the island-creating effect of secondary complementizers has eluded principled analysis.
As a further problem for monosentential accounts of R-ClLD, Demonte & Fernández-Soriano (2014) point out that the secondary complementizer is contingent on the first one (iv), i.e., it can only appear when the preceding clause contains a complementizer as well: This criticism is rooted in the fact that if the secondary que is a topic-marker lexicalizing the head of TopicP, then it is unclear why it would depend on the lexicalization of a higher left-peripheral head (Force 0 , by hypothesis).Demonte & Fernández-Soriano solve this issue by appealing to Martín-González's (2002) (Doubled)ForceP (i.e., ForceP > TopicP > (Doubled)ForceP > FocusP…), but independent motivation for the existence of such a projection is not easy to come by, and thus the analysis remains stipulative.The issue is aggravated by the fact that TopicP, responsible for hosting the dislocate, hierarchically separates the postulated ForcePs. 6Furthermore, as observed by João Costa (pers.comm.2019), R-ClLD cross-linguistically employs form-identical complementizers (e.g., queque; thatthat) (v; see Section 3.7 for further evidence).This is mysterious under monosentential analyses, where each occurrence of the complementizer occurs in a different left-peripheral head (e.g., Force 0 for que 1 and Topic 0 for que 2 ).Note in addition that if que 2 is a bona fide exponent of a Top(ic) head, as held by proponents of the TopicP account (Rodríguez-Ramalle 2003; Villa-García 2012), we would naturally expect a distinct morphological realization (rather than a default complementizer). 7The verbatim repetition of the complementizer in R-ClLD is thus unaccounted for under monosentential analyses, barring additional stipulations.
Finally, Demonte & Fernández-Soriano (2009: 47)  [6] Furthermore, as pointed out by Peter Svenonius (pers.comm.2022), the dependence of the second que on the higher one is at odds with a model of incremental, bottom-up structure-generation: the linearly first que would be merged later than its lower counterpart, the appearance of which would nevertheless be conditional on that of the former.As indicated, regular ClLD is possible in these cases, but R-ClLD is infelicitous.How can this state of affairs be accounted for on monosentential analyses?Again, the answer is not evident: for example, on a TopicP account, the difference between R-ClLD vs. regular ClLD reduces to the lexicalization of Topic 0 (or C 0 , under CP-recursion).Thus, to account for the incompatibility of certain predicates such as factives and R-ClLD complements, one would have to assume that the higher predicate conditions the (non)lexicalization of a lower, structurally remote complementizer: One could, of course, assume a different, more limited left-peripheral structure for factives (in the spirit of de Cuba & MacDonald 2013), but we would still need to account for why regular ClLD is legitimate in the same environment.
Overall, then, a number of non-trivial problems call into question traditional multiclausal-monosentential approaches to recomplementation patterns, despite their initial appeal.In what follows, we develop a proposal that radically deviates from standard assumptions about ClLD and R-ClLD, building on the idea that both configurations involve multiple root clauses in a paratactic configuration, albeit in different ways.As we show, the proposal not only solves the issues raised above but also makes additional welcome predictions.

A PARATACTIC ANALYSIS
The proposal advocated here draws on work by Ott (2014Ott ( , 2015Ott ( , 2017)), who argues that left-dislocated XPs are remnants of elliptical sentences (i.e., fragments) that are juxtaposed to their host sentences in discourse. 8That is to say, dislocated XPs are not intra-sentential constituents moving to or base-generated in a left-peripheral position (e.g., TopicP); instead, they are bona fide extra-sentential constituents.The analysis is illustrated below for (19a): The dislocate in CP1 and its host CP2, each a root clause, are thus paratactically ordered but endophorically related by ellipsis and cross-sentential anaphora.Ott shows that this approach can resolve the paradox initially discovered by Cinque (1990), i.e., the concurrent movement and non-movement (connectivity/anticonnectivity) properties of left-dislocation constructions.
One major advantage of this approach is that the obligatory co-occurrence of the clitic in CILD is no longer mysterious: the clitic is a pronoun in CP2 that anaphorically resumes the XP ese coche in the first clause (CP1), as would be the case across independently generated sentences: (20) Me gusta ese coche.Lo van a comprar.CL.DAT like that car CL.ACC go to buy 'I like that car.They are going to buy it.' The presence of the clitic in (R-)ClLD (and of a correlate more generally in leftdislocation) is thus a result of the fact that the clause containing it is indeed an entire sentence, which consequently must be syntactically complete.The following contrast in English (from Villa-García 2019) highlights this point: (21) (a) They told me that Peter, that they like him.
(b) They told me that Peter, (*that) they like.
The dislocation in (21a) permits recomplementation since the second, structurally autonomous clause is syntactically complete, unlike in the fronting case (21b), where recomplementation forces a parse of the second clause as an independent, and consequently syntactically incomplete, root-clause fragment. 9 In the same vein, the paratactic analysis explains why left-dislocated XPs are typically intonationally separated from their hosts (because they are independent, elliptical root clauses), precede entire V2 configurations in languages like German, and can mismatch their correlates in phi-features in a precisely delimited range of cases; see Ott (2014Ott ( , 2015Ott ( , 2017) ) for detailed discussion and references.Ott (2015) touches on embedded ClLD, exemplified in (22a), and adumbrates an analysis which assumes interpolation or intercalation of a parenthetical fragment, as in (22b) (cf.[9] For the same reason, the locative adverbial of locative-inversion constructions cannot appear in between thats, as shown in (i)a; as expected, rendering CP2 a syntactically complete sentence by addition of an expletive subject improves the result (i)b.
(i) (a) *They told me that in that corner, that stood a lamp.(b) They told me that in that corner, that there stood a lamp.
As noted by Jonathan Bobaljik (pers.comm.2022), there is a register clash in the examples in (i), since locative inversion is much more formal than recomplementation, which is a phenomenon of spoken language; the relevant contrast holds regardless.
Building on Ott's analysis of matrix and embedded ClLD, we propose to analyze R-ClLD cases as paratactic configurations, but involving a somewhat different arrangement compared to regular ClLD on Ott's approach.
In order to illustrate the proposal, let us focus again on (1a), repeated here for convenience: (23) Dice que a tu madre, que la invitaron.(= (1a)) says that ACC your mother that CL.ACC invited 'S/He says that your mother, that they invited her.' Unlike in the case of regular ClLDs just discussed, our claim is that R-ClLD instantiates a sequence CP1 ≺ CP2 of two elliptical root clauses, such that CP1 realizes an embedded fragment, while CP2 constitutes a 'restart' spelling out a parallel embedded clause in full but leaving the redundant main clause unpronounced: This paratactic analysis captures rather directly the intuitive character of recomplementation as a bona fide 'repetition' or 'reduplication' (see, e.g., Vincent, Bentley & Samu 2004;Ledgeway 2005;Kemchinsky 2013;Villa-García 2012, 2015, 2019).
The juxtaposed root clauses CP1 and CP2 in ( 24) are syntactically parallel, modulo the difference between the dislocated XP in CP1 and its clitic correlate in CP2.In this sense, again, coreference between the two is ensured by text/discourse coherence, analogous to sequences of non-elliptical sentences (see Ott 2017; Onea & Ott 2022): (25) Dice que invitaron a tu madre k .Dice que says that invited ACC your mother says that la k invitaron.CL.ACC invited 'S/He says that they invited your mother k .S/He says that they invited her k .'On this view, recomplementation as in ( 24) is thus a mere surface variant of the sequence of sentences in (25). 10  [10] In this connection, it should be noted that not all dislocates in Spanish occur with a correlate/ resumptive element, as observed by Adolfo Ausín (pers.comm.2021).For instance, PP arguments in Spanish do not have a clitic counterpart (Casielles-Suárez 2004; Villa-García 2015): (i) Te contará que con ella, que no cuentan.CL.DAT will-tell that with her that not count 'S/He will tell you that they don't count on her.' Despite the lack of an overt correlate following the second que, the second sentence is syntactically complete, and the construction is a proper instance of (R-)ClLD (see Ott 2015: It is important to note that the type of ellipsis assumed in ( 24) is independently available in Spanish (and other languages).CP1 instantiates an embedded fragment analogous to B's response in ( 26 Such embedded fragments must involve abstract structure and deletion (see Merchant 2004;Temmerman 2013), as a María is case-marked and the complementizer que is not a case assigner (i.e., there must be an underlying transitive verb in (26B)).We return to this type of fragment when discussing the issue of selection in R-ClLD (Section 3.8). 11241, fn.22).An argument may not be overtly realized in CP2, but this does not make the second sentence syntactically incomplete; text/discourse coherence ensures that the sentence in CP2 is properly interpreted based on the preceding context (i.e., CP1).The same holds for incontrovertibly independent elliptical sentences, as in (ii): (ii) A: Dirá que no cuentan con ella.will-say that not count with her 'S/He will say that they don't count on her.' B: Sí, sí.No cuentan.yes, yes not count 'Yes, absolutely.They don't (count on her).' Since (ii)B occurs in the context of (ii)A, B can only mean 'counting on her,' not 'counting numbers,' for example.The same applies to conditional sentences, as in (iii), where the conditional meaning is understood even if not overtly expressed (cf. (iv)B): (iii) Dice que si llueve, que no vienen.
says that if rains that not come 'S/He says that they won't come here if it rains.'(iv) A: The second sentence in the recomplementation sequence (i.e., … [ CP2 dice [que la invitaron]] in the above example) involves Phonetic Form (PF) deletion of the material above the complementizer, including the subject (if present) and the verb, leaving only the lower, embedded clause as a remnant.This type of ellipsis, too, is generally available, as evidenced by clausal fragment responses (see (30) below and Merchant 2004 for discussion).
Ellipsis resolution in CP1 can only occur once the postcedent (CP2) has been uttered, as is generally the case in 'backward' ellipsis: (27) I don't know when they will crack the code, but they will crack the code.
Note that for the embedded clause in CP2 to be a licit fragment, it must contain an instance of (possibly broad) focus, in the way remnants of ellipsis are generally distinguished from the informational background.A corollary of this basic requirement is that the XP sandwiched between complementizers in R-ClLD cannot be itself focal-if it were, the remainder of the embedded clause would invariably be backgrounded, undermining its ability to function as a 'restart' fragment.In this way, the ellipsis-based analysis derives the fact that sandwiched XPs in R-ClLD will be contrastive topics or independent interjections (see below), but never foci. 12 Ellipsis in both clauses is certainly favored, though not in any grammatical sense obligatory: non-elliptical counterparts are naturally overly redundant but not unacceptable (see also (25) above): (28) Dice que invitaron a tu madre; dice que la says that invited ACC your mother says that CL.ACC invitaron.invited 'S/He says that they invited your mother; s/he says that they invited her.' Importantly, however, in cases involving quotative complementizers à la Etxepare (2010) (see, e.g., (16b) above) and lacking an (overt) embedding predicate in CP1, if the material preceding the second instance of the complementizer does not undergo ellipsis in CP2, the resulting sentence improves significantly vis-à-vis (28): 13   [12] As far as we can see, and as one would expect given this explanation, the only way to render such configurations acceptable is by making the restarted embedded clause an emphatic repetition: (i) Pienso que NUNCA, <interjections>, que NUNCA vendrá.think that never … that never will-come 'I think that never… that never will s/he come.' In such cases, the rhetorical demand for emphasis overrides considerations of focus-background structuring and redundancy avoidance.
[13] Note that we are not claiming that there is an elided verb of communication above quotative que, although clearly the quotative character of this pattern makes it compatible with bona fide verbs of communication like decir 'to say' for purposes of parallelism, for instance.See Etxepare (2010) and Demonte & Fernández-Soriano (2014) on matrix que with quotative/evidential meaning.
(29) Que a tu madre dicen que la invitaron.that ACC your mother say that CL.ACC invited 'They say/it is said that they invited your mother.'Lastly, one might wonder why the second que survives ellipsis in R-ClLD.Our claim is that this que serves to mark the 'restart' in discourse (see Section 3.6); its PF realization indicates that the point of restart is the embedded domain.A parallel situation can be found in clausal fragments, which require the presence of the complementizer que, as shown by ( 30). that the recomplementation is intriguing 'That recomplementation is intriguing.'After all, a that-type complementizer is the hallmark of a finite embedded clause, the simplest way of signaling the presence of a clause (i.e., … [ CP2 dice [que …]]).In fact, as shown below in Section 3.7, complementizers other than the default declarative complementizer que/that can appear in R-ClLD, which is fully consistent with our claim that reduplicative complementizers are overt manifestations of a 'restart' in discourse.
Overall, we conclude that there is nothing special about the types of ellipsis instantiated in (24): the composition of R-ClLD draws entirely on familiar and independently attested forms of phonological reduction.We take this to be an advantage over a recent proposal by Echeverría (2022), which, while similar in spirit, relies on multiple syntactic planes within a unitary representation to implement the intuition of R-ClLD as a restart (Villa-García 2019).Since Echeverría's proposal remains rather vague on the formal mechanisms involved, we merely note here that our proposal relieves the grammar of the burden of generating multidominance structures.
In the next subsections, we turn to a variety of empirical arguments in favor of a bisentential, paratactic account of R-ClLD configurations.

Argument 1: Intonation
The dislocated XP and secondary que are typically separated by a salient intonational boundary, represented by a comma in orthography, as shown again in (31) (Villa-García 2012, 2015, 2019;Radford 2018; Martínez Vera 2019): (31) (a) Me cuentan que a ese tipo, que no lo tragan.CL.DAT tell that ACC that guy that not CL.ACC swallow 'I am told that they can't stand that guy.' (b) They said that if I keep writing this email, that I will miss my plane.This is wholly compatible with our claim that the relation between fragment and host is paratactic, each sentence forming a separate intonational phrase (cf.Nespor &Vogel 1986 andDehé 2007 on parentheticals and intonational phrasing): (32) ( IntonP dice que/says that XP) fragment ( IntonP … que/that … ) host Fragment and host thus exhibit 'comma intonation' (intonational isolation/separation), exactly as expected if the sequence is composed of linearly juxtaposed root clauses in a paratactic rather than hypotactic arrangement.

Argument 2: connectivity and anticonnectivity
As we saw above, the dislocated XP displays a range of connectivity effects, showing that it cannot be base-generated in its surface position sandwiched between the two complementizers (as claimed by some integrated, monosentential accounts such as Villa-García's 2012 TopicP analysis shown in (3b)).
By contrast, on the paratactic approach advocated here, these properties follow directly from the fact that ellipsis in CP1 is licensed under identity with CP2, so that the dislocate and the clitic share theta/case-properties (see Ott 2015and Merchant 2001, 2004 on ellipsis-mediated connectivity more generally).In an example such as (1a), a tu madre receives case and a theta-role from the elided verb invitaron in CP1 in a run-of-the-mill fashion; analogously, the accusative clitic la in CP2 receives its case and theta-role from overt invitaron in CP2: (33)  At the same time, the paratactic approach handles anticonnectivity effects which reveal that the dislocated XP is syntactically disjoint from the following clause headed by the doubled complementizer.Coreference between the dislocate and the pronominal correlate, on this approach, is an instance of cross-sentential anaphora, analogous to the following: (35) Conocieron a tu madre.La invitaron.met ACC your mother CL.ACC invited 'They met your mother.They invited her.' Therefore, the dislocated element sandwiched between complementizers and its correlate are never in a (symmetric or asymmetric) c-command relation at any stage of the derivation; the dislocate is simply not a constituent of the host clause/CP2, which accounts for why the simultaneous presence of dislocate and correlate does not give rise to a Condition B violation.
Furthermore, the apparent doubling of a preverbal subject in cases such as (13) above dissolves on the analysis pursued here, as dislocate and correlate are in separate sentences: the double can appear simply because CP2 is a syntactically complete sentence in its own right: (36) [ CP1 dice [que no es así Hugo]] [ CP2 dice [que él no es así]] (= (13)) In sum, the connectivity and anticonnectivity effects found in R-ClLD constructions are accommodated straightforwardly by our approach, while posing an irresolvable paradox for integrated, multiclausal-monosentential analyses (as expounded in Section 2).

Argument 3: Non-clitic correlates
On the paratactic approach advocated here, recomplementation is a bona fide restart in discourse.Accordingly, we expect this restart to formally manifest itself in various ways beyond the 'classical' R-ClLD pattern, where some topical XP is resumed by a clitic correlate.As we will see presently, this expectation is borne out.
In addition to pronominal clitics, R-ClLD readily permits epithetic correlates: (37) (a) Cuentan que a María, que no le dan nada tell that DAT Mary that not CL.DAT give nothing a la muy malvada.DAT the very evil 'They say that they don't give anything to Mary, that evil woman.'(b) Dijo que con Juanito, que no contaban con él.
said that with John that not count with him 'They said that they don't count on John' (inspired by data in Villa-García 2019) (Note that no pause occurs between the epithetic correlate and the rest of the sentence).The ability of epithetic correlates to occur in CP2 is not surprising under a paratactic approach to R-ClLD, where the dislocate and its correlate belong to separate root clauses and are anaphorically related cross-sententially: (38) [ CP1 cuentan [que no le dan nada a María]] [ CP2 cuentan [que no le dan nada a la muy malvada]] This is exactly as in an analogous sequence of non-elliptical sentences: (39) Odian a María.No le dan nada a la muy malvada.hate ACC Mary not CL.DAT give nothing DAT the very evil 'They hate Mary; they don't give anything to that evil woman.' In contrast, traditional multiclausal-monosentential analyses including the TopicP account are at odds with data such as those in (37): one and the same predicate is required to assign identical cases (and theta-roles) to two separate elements.The problem does not vanish on the assumption that the dislocated phrase is externally merged in between ques (Villa-García 2012), since there would still be only one single predicate (i.e., dan 'give' in (37a)) in the overall structure: This non-trivial issue does not arise under the paratactic approach espoused here, as seen in ( 38).Epithetic correlates, in short, are a natural by-product of our paratactic analysis of recomplementation. 14 Non-clitic correlates are furthermore found when the XP sandwiched between complementizers is non-topical/non-referential, as is the case with quantifiers which do not undergo regular ClLD, since they cannot be resumed crosssententially by clitics (Ott 2015); in this case, the quantifier itself is repeated verbatim in CP2: (41) (a) Dice que nada, que no hizo *(nada).says that nothing that not did nothing.'S/He says that s/he didn't do anything.'(spontaneous WhatsApp conversation, Spain, April 2022) [14] Regarding epithetic correlates in regular ClLD, there seems to be considerable inter-speaker variation.For some speakers, a full (i.e., non-clitic) epithetic correlate is not possible; for others, it is possible as long as a pause occurs between the sentence and the epithetic correlate; finally, there are speakers for whom an epithetic correlate is unobjectionable, much like in the recomplementation cases in ( 37 For cases of dialectal variation regarding the availability of epithets with ClLDs, see, e.g., Estigarribia (2020), where a biclausal analysis à la Ott (2014)  Note that in both cases the phrase is repeated in the embedded clause because the fragmentary clause must be syntactically complete; at the same time, no such function could be performed by a clitic (or epithetic) correlate.
Examples such as the above (as well as others to be discussed in Section 3.6 below) militate decisively against analyses of recomplementation that derive its occurrence directly from the activation of some dedicated topic position; they point instead to the conclusion that recomplementation is a phenomenon that occurs independently of ClLD, exactly as expected on our approach.

Argument 4: Clausal negation
A further argument for our paratactic approach derives from recomplementation patterns involving dislocated negatively quantified phrases.In the presence of such a negative dislocate, the host clause must obligatorily contain a negation (Martín-González 2002;Villa-García 2012, 2015): 15   (42) (a) Dice que a ninguno de ellos, que no lo(s) llamó.says tha ACC none of them that not CL.ACC called 'S/He says that s/he didn't call any of them.'(b) *Dice que a ninguno de ellos, que lo(s) llamó.says that ACC none of them that CL.ACC called On a monosentential analysis, the obligatory appearance of negation no 'not' in ( 42) is puzzling and requires elaborate auxiliary assumptions.To illustrate, Villa-García (2015: 149-157) rationalizes the mandatory presence of negation in (42a) as follows: since the dislocate a ninguno de ellos 'none of them' is base generated where it surfaces (that is, in the position between complementizers), it never transits through Spec-NegP on its way to the left periphery (that is, to Spec-TopicP); since no spec-head relation is ever established between the dislocated negative quantifier and Neg 0 , the head no needs to be inserted as a last resort in Neg 0 , following the logic of Bošković (2001).The reason is that Bošković treats negation as an affix that requires an n-phrase to be PF-adjacent to it at some point in the derivation.Since a base-generated dislocate in the sandwiched position is never PF-adjacent to Neg 0 , inserting no is necessary to circumvent a violation that would ensue if the negative affix were left stranded.
On our paratactic analysis, by contrast, the mandatory presence of negation naturally follows from the fact that the host clause must have the same polarity as the elliptical clause to satisfy ellipsis parallelism.The dislocated negative XP surviving ellipsis in CP1 does not negate CP2: no 'not' is thus required in CP2 for CP1 and CP2 to be semantically parallel, as shown schematically in (43).In other words, the otherwise puzzling interaction of R-ClLD and negation in these cases emerges as a direct consequence of the 'restart' nature of the construction, implemented here as a paratactic sequence of elliptical root clauses. 16The behavior of negation in a parallel sequence of non-elliptical sentences is exactly analogous, as expected: (44) Dice que no llamó a ninguno de ellos.Dice que *(no) says that not called ACC none of them says that not lo(s) llamó.CL.ACC called.'S/He says that s/he didn't call any of them; she says that she didn't (call him/them).' We turn next to the by-now longstanding issue of the impossibility of extraction across secondary complementizers.

Argument 5: An extraction asymmetry
As illustrated in (15) above, unlike regular ClLD, R-ClLD prohibits extraction from the clause following the dislocate.On our terms, this follows from the structural disjointedness of the sentences involved, each a separately generated root clause.For purposes of illustration, compare the paratactic configurations in regular ClLD according to Ott (2015), in (45), and R-ClLD, in (46), on our approach: (45) [ CP2 qué i me contaste que… [ … a tu novia …] … le habías regalado t i ] (= (15a), no recomplementation) [16] Regarding non-recomplementation ClLD, in the regular case negation is impossible, as in (i): (i) Dice que a ninguno de ellos lo(s) llamó.says that ACC none of them CL.ACC called However, if a prolonged pause (#) occurs after the ClLDed phrase a ninguno de ellos, then negation is again required: (ii) Dice que a ninguno de ellos # *(no) lo(s) llamó.says that ACC  ClLD in ( 45) permits extraction from the lower clause into the main clause because the former is structurally subordinate to the latter (hypotaxis); the dislocate a tu novia is a separate fragmentary expression interpolated in production.
By contrast, the attempted extraction in the R-ClLD case is cross-sentential (46) due to the paratactic nature of the configuration (CP1 and CP2 are separately generated expressions), hence ruled out on principled and general grounds as illicit extraction across SENTENCES.Recall that no such explanation is available on monosentential analyses, which treat ClLD and R-ClLD as phonological variants and consequently struggle to account for asymmetries of this kind.
Note that our explanation building on the structural disconnect of CP1 and CP2 is not undermined by apparent cross-sentential binding, as in the following: (47) Ninguno de los niños i pensó que, por navidad, none of the boys thought that for Christmas que darle a su i madre un libro era bueno.that GIVE-CL.DAT DAT his mother a good was good 'None of the boys i thought that it would be a good idea to give his i mother a book for Christmas.' The binding dependency here is computed internally to the elliptical CP2, which, due to general parallelism requirements, contains a silent instance of the binder: (48) [ CP1 …] [ CP2 ninguno de los niños i pensó [que darle a su i madre un libro era bueno]] An anonymous reviewer wonders why the ungrammatical extraction in (15b) could not alternatively be derived as involving a wh-dependency in each of the juxtaposed sentences CP1 and CP2: This would indeed be a plausible alternative if the required elliptical CP1 were independently licensed.But embedded fragments with concomitant wh-extraction appear to be illicit quite generally, and regardless of context:  2013) for some preliminary observations.Whatever rules out fragmentary expressions such as (50), they are plainly not available as building blocks for R-ClLD configurations, ruling out the alternative suggested by the reviewer.
Overall, the extraction asymmetry observed above furnishes a strong argument in favor of the differential analysis of regular ClLD and R-ClLD advocated here: while the former involves bona fide subordination/hypotaxis (plus linear interpolation of the fragment dislocate), the latter instantiates a genuine restart in which no syntactic connection exists between the first sentence containing the embedded dislocate (CP1) and the second sentence introduced by the secondary complementizer (CP2).

Argument 6: Interdependence of complementizers
Villa-García (2019) contends that one of the functions of the additional occurrence of que is that of a discourse marker; in a similar vein, Casasanto & Sag (2008) argue that doubled complementizers aid processing.This general notion meshes well with our approach to R-ClLD, where the dislocated XP is syntactically disjoint from the subsequent clause, and as such a separate unit of discourse.While we cannot fully develop this idea within the confines of the present paper, we suggest that the discourse/processing-related function of the secondary complementizer in R-ClLD is to overtly signal a restart by identifying the following material as an embedded clause and thus necessarily a subsentential fragment.This is wholly consistent with our analysis, where the dislocate is part of the initial clause, and the two sentences involved in R-ClLD are cataphorically linked by ellipsis under identity.R-ClLD is not the only case where a that-like element takes on this function.Clausal fragment responses generally require a complementizer, even when it can be omitted in the non-elliptical source sentence (see Merchant 2004 for discussion and sources, and (30) above for Spanish data): (51) A: What does no one believe?B: #(That) I'm taller than I really am.
(52) No one believes (that) I'm taller than I really am.
As in the case of R-ClLD, the complementizer forces a parse of the clause as a subsentential fragment rather than a non-elliptical root clause, anaphorically anchoring it in the immediate linguistic context.
Recall from the discussion in Section 2 that the question of why the second que is contingent on the first one (i.e., the occurrence of second que depends on the occurrence of a higher que) is an unresolved issue on monosentential accounts.By contrast, the paratactic analysis advocated here offers a straightforward rationalization of this otherwise puzzling dependence: occurrence of the second que is contingent on that of the first simply because a 'restart' of an embedded clause is possible only where there is one to begin with.
What is more, the second complementizer can co-occur with an unambiguous discourse marker such as pues 'then/thus,' as in (53).Note that discourse-related particles tend to cluster together in many languages (Hansen 1998). 17(53) Dice que como no hay dinero, pues que no va.
says that as not have money then that not goes 'S/He says that since there is no money, that then s/he is not going.' In fact, the paratactic approach naturally extends to instances of restarts in the absence of a dislocate, where instead a mere pause (54a) or some interjection (54b,c, d) separates the two fragmentary sentences (Villa-García 2015, 2019), as is widely attested in spontaneous speech (recall also the facts in (41)  Echeverría (2021, 2022: 10n4), who pursues an analysis of recomplementation in terms of multiple nonlinear planes, claims that the absence of a dislocate in such cases poses a problem for the paratactic analysis pursued here.On the contrary, we submit that this approach offers precisely the leeway required to accommodate such restarts more generally, permitting reflective pauses, interjections and the like to be intercalated in between both sentences: As previously noted in Section 3.3, cases such as those in (54) show that recomplementation is not a corollary of ClLD but a more general production [17] Altering the order pues > que is not possible.Similarly, dropping que in such contexts and leaving pues alone does not lead to an acceptable outcome at least in Iberian Spanish (see Martínez Vera 2019 for Latin American Spanish recomplementation): (i) ?? Dice que como no hay dinero, pues no va.says that as not THERE.IS money then not goes We would like to thank an anonymous reviewer for raising this point.
phenomenon. 18It would therefore be inaccurate to model the observed complementizer doubling as a direct effect of dislocation, however this intuition is implemented.
3.7.Argument 7: Form-identity of complementizers On the paratactic analysis, the fact that the doubled complementizer is formidentical to the first one is a straightforward corollary of the parallelism required of the sentences involved for purposes of ellipsis resolution: says/tells/thinks that says/tells/thinks that By the same token, the paratactic approach to recomplementation can easily accommodate cases of multiple reduplicative complementizers (i.e., que/that XP, que/that XP, que/that …), as illustrated in (6) in Section 2. Recall that such iterative cases are analyzed on the TopicP account as cases of TopicP recursion (cf.( 8)), argued for on independent grounds by Rizzi (1997).On the paratactic account, these are simply cases of sequential elliptical sentences, as shown in (57); nothing else needs to be said.(Tim Vickery, BBC Radio 5, reported in Radford 2018: 175) [18] An important question for future research is just how generally such an analysis can be applied to restarts and other 'disfluencies' conditioned by grammatical factors (Ferreira et al. 2004).We thank an anonymous reviewer for bringing up this issue, which, however, we cannot take up in the present paper.
[19] English non-finite complementizer for also exhibits reduplication, as noted by the following example, kindly provided by Andrew Radford (pers. comm. 2022): (i) It's important for, at some point, for a crisis intervention team to come in and step in.
This naturally follows from the paratactic analysis, whereby the repeated interrogative complementizer is selected by the predicate preguntar 'ask'/wonder in CP2: Further support for this line of reasoning derives from Plann's (1982) discovery of a contrast in Spanish embedded questions under verbs that do not inherently take an interrogative clausal complement.As shown by the translations of the minimal pair in (60), the presence of que turns the embedded clause into a reported interrogative (i.e., an indirect question) (see RAE-ASALE 2009 for further discussion and examples): (60) (a) Dijo si venían.
said if would-come 'S/He said whether they would come.'(b) Dijo que si venían.
said that if would-come 'S/He asked/wondered whether they would come.' With recomplementation, however, the presence of a second occurrence of the que-si sequence becomes obligatory (Villa-García 2015: Ch. 5): 20 (61) Dijo que si a la fiesta, *(que) si venían.
said that if to the party that if would-come 'S/He asked/wondered whether they would come.'For monosentential analyses which assume a complex left periphery, accounting for this contrast is not straightforward: if, for instance, the interrogative complementizer heads Int(errogative)P (as in Rizzi's 2001 analysis), the only possibility to analyze (61) would be to invoke (discontinuous) recursion of IntP: On this type of analysis, the high que would by hypothesis be the realization of Force 0 and the second one the realization of Topic 0 .As noted, two occurrences of IntP would have to be postulated, one above and one below TopicP.Note that if we were to dispense with TopicP and locate the dislocate in the specifier of the lower IntP, this would in effect deprive us of the head position for secondary que, yielding the wrong result: All in all, the above considerations strongly suggest that an account of this kind is untenable.
On our approach, on the other hand, the repetition of the que-si sequence in cases like (61) is predicted, simply because CP2 is an elliptical restart of the embedded indirect question: Having only si without que would violate parallelism between CP1 and CP2, as it would not allow the surface remnant of CP2 to be interpreted as a reported question.Overall, the facts reviewed above provide further support for the paratactic account of recomplementation patterns advocated here.
It is worth stressing again that our reasoning vindicates the intuition behind the label re-complementation originally given by Higgins (1988): the second que heralds the presence of a duplicated complement clause.Recomplementation truly is recomplementation.

Argument 8: Selection
As noted in Section 2 in relation to the data in (17), only certain predicates allow for R-ClLD in their complement domain.More specifically, in Spanish, factives such as lamentar 'lament' cannot take recomplementized clausal complements, as noted by Demonte & Fernández-Soriano (2009), Villa-García (2012, 2015), and González i Planas (2014).In principle, this is handled straightforwardly by the paratactic approach, given that the kind of embedded fragments argued here to feature in recomplementation is likewise restricted to verbs such as decir 'to say,' but unavailable with factive predicates such as lamentar 'to lament:' (65) A: ¿Quién ha ganado?
who has won 'Who has won?' B: Me dijo que ha ganado tu niña.CL.DAT said that has won your girl 'S/He told me that your daughter won.' C: *Lamento que ha ganado tu niña.lament that has won your girl 'I lament that your daughter won.' The same applies to other predicates, such as volitional querer 'want,' which are also incompatible with both recomplementation (66) and embedded fragments (67), much like factives.
(66) Quiero que a mi boda (*, que) vengas.want that to my wedding that come 'I want you to come to my wedding.' (67) A: ¿A dónde tenéis que ir?
to where have that go 'Where do you have to go?' B: *Quiere que vayamos a tu boda.want that go to your wedding 'S/He wants us to go to your wedding.' Although the issue of selection should be further investigated in more detail in future work, the above observation adds to the repertoire of arguments in favor of a paratactic account of recomplementation.
3.9.Argument 9: No recomplementation with ClRD Demonte & Fernández-Soriano (2009) observe that embedded CLITIC RIGHT-DISLOCATION (ClRD) cannot be accompanied by a doubled complementizer: (68) *Dice que las i tiraron, que esas casas i .says that CL.ACC demolished that those houses 'S/He says that they demolished them, (*that) those houses.'Under a TopicP-based account of ClRD, this is unexpected: everything else being equal, if a doubled que is the realization of Topic 0 , this topic marker ought to be licensed for right-peripheral topics as much as for left-peripheral ones.
Can the paratactic approach deal with this asymmetry between ClLD and ClRD?The answer to this question turns out to be positive. 21 Note, first, that in (69a) there is an incongruence between the nominal cataphoric clitic in the first clause and the subsequent (illicit) fragment, identified as clausal by the complementizer; (68) thus violates the expectation of a nominal fragment based on the rhetorical connection between the sentences (Ott 2017;Onea & Ott 2022).Furthermore, the kind of fragment that (68) attempts to employ is simply not licensed in general; we can see this by considering (70B), which is a direct equivalent of CP2 in (69a) on our analysis: This contrasts markedly with the kind of sequence comprising a verb such as decir 'say' þ que þ XP that appears in R-ClLD (as in CP1 in (69b) above) and that is independently available as a fragment response: The incompatibility of recomplementation and ClRD thus follows straightforwardly on the bisentential paratactic analysis.

CONCLUSION
We have shown in this paper that a version of the paratactic account proposed by Ott (2014Ott ( , 2015) ) for regular ClLD adapted here to the recomplementation variety overcomes many of the problems raised for monosentential accounts of R-ClLD configurations (such as cartographic analyses adopting a Rizzian left periphery) and makes several welcome predictions.
The analysis has also helped unveil previously unnoticed properties of recomplementation, such as the fact that not only dislocated constituents, but also other kinds of intercalated expressions can occur in between the two complementizers.As we have argued, this is just what we expect if the secondary complementizer signals a 'restart' in discourse.
As noted by an anonymous reviewer, our analysis undermines the common idea (implicit, for instance, in the cartographic works referenced above) that Spanish and English differ in relevant respects regarding the composition of the left periphery; on our analysis, so-called recomplementation is the result of entirely general mechanisms.This is not to say that left-peripheral restarts will be uniform crosslinguistically; ellipsis options vary across languages, and this will affect the form of fragmentary expressions involved, among other factors.Needless to say, this paper merely scratches the surface of the phenomenon, and we hope that it can inspire deeper investigations into the interrelation of ellipsis and restarts across languages.
Overall, we have shown within the expressly narrow focus of this paper that a paratactic analysis of recomplementation for languages like Spanish and English not only directly implements Higgins' (1988) original intuition about the phenomenon but also provides elegant and principled answers to a host of perennial questions that have proven difficult to resolve in a principled fashion on previous approaches.
FP projection (3c) (see Villa-García 2015):(3) (a) …[ C' que [ CP a tu madre [ C' que [ TP la invitaron]]]] (b) …[ Force' que [ TopicP a tu madre [ Topic' que [ FinP … la invitaron]]]] (c) …[ C' que [ FP a tu madre [ F' que [ TP la invitaron]]]] [7]  As noted by MichelleSheehan (pers.comm.2021), there are varieties of Romance where the complementizers in what at first sight looks like recomplementation are morphologically distinct.See, in particular, the work ofLedgeway (2005), who observes that in addition to the frequent cases of complementizer repetition with dislocated phrases in certain dialects of southern Italy (i.e., cheche), along the lines of what we observe in Spanish and English in the main text, there are che and ca complementizers (and other variants), ca-style complementizers being used for subjunctive clauses (cf.[ ForceP che [ TopicP [ FocucP [ FinitenessP ca [ TP …]]]]], based on Ledgeway 2005: 364; see also Villa-García 2015 on the low que complementizer in Spanish subjunctive clauses).Ledgeway's cases also include complementizers below foci, which are not possible in Spanish.We leave to future work the question of whether these examples constitute cases of R-ClLD proper or a different phenomenon altogether.

(
50) context: I know you said you gave the bike to your brother, but… *girlfriend Intended: 'What did you tell me that you gave to your girlfriend?' (Compare English *WHAT DID YOU TELL ME THAT TO YOUR GIRLFRIEND? in the same context).The rather tight restrictions on embedded fragments are generally not well understood; see Temmerman (

(
57) [ CP1 dice [que van a llamar Noelia a la niña al final]] [ CP2 dice [que van a llamar Noelia a la niña]] [ CP3 dice [que la van a llamar Noelia]] (= (6)) As a further argument, consider reduplicated interrogative complementizers in both Spanish and English (Villa-García 2015, 2019): 19 (58) (a) Preguntó si a la fiesta, si vais a ir.asked if to the party if go to go 'S/He asked if you guys are going to the party.'(b) I wonder if, given time, if Ramírez can fulfil that sort of role for the club.
[CP1 dice [que invitaron a tu madre ]] [CP2 dice [que la invitaron ]] is advocated.We will leave a detailed comparison of epithetic correlates in ClLD vs. R-ClLD to future work.
none of them not CL.ACC called (46) *[ CP1 qué i me contaste [que … a tu novia]] [ CP2 … [que le habías regalado t i ]] (= (15b), recomplementation) He says that … that they invited her.' (b) Dice que bueno, que la invitaron.says that well/ok that CL.ACC invited 'S/He says that, well, that they invited her.' (c) Dice que oye, que ha suspendido.says that LISTEN.IMPERATIVE that has failed 'S/He says that s/he has failed.'(inspired by Etxepare 2010) (d) It just shows that, you know, that they have to pass the time.(Ian Carter, BBC, Radio 5, cited in Radford 2018: 124)