2.1 English

In English, ever since the seminal work of Ross (1967) came out, two types of topic preposing mechanisms have been identified in the literature: the left-dislocation (or hanging-topic) construction and the topicalization construction.³ The phenomenon of left-dislocation is illustrated in (7).

The example in (7), which happens to be an instance of embedded left-dislocation, comprises a preposed DP constituent, my father. This is followed by a prosodic boundary (customarily represented in writing by a comma) and is resumed by a (strong) subject pronoun in the subordinate clause (he). Note that the real subject of the embedded clause is actually he, coreferential with the dislocate my father. This pattern has been dubbed the Strong Pronoun Left Dislocation construction (e.g., Casielles-Suárez 2004) or the Hanging Topic Left Dislocation (HTLD) construction (e.g., López 2009), a term that I will adopt in what follows. As we will see, an important property of HTLDs is that they are base-generated in their superficial position, and their actual connection with the clause where they are interpreted is effected via the resumptive element, so for instance, they are impervious to locality-of-movement constraints (Radford 2018: Ch. 2). Moreover, hanging topics display default case, which in English is accusative. This is shown by the following example, where the dislocated pronominal is accusative, despite its obligatory association with the (nominative) subject of the embedded clause.

(8) unveils another property of the HTLD construction that sets it apart from its topicalized counterpart, a construction to which I turn immediately: the possibility of having an epithet (that guy) in the position that the dislocate would occupy in an SVO sentence.⁴

The second preposing mechanism identified in the literature is the topicalization construction, exemplified in (9).

In (9), the boldfaced object this book appears in a left-peripheral position above the subject. Its canonical position as the object of the verb is not filled overtly, contrary to what happens in (7); there is a gap (___) in the position that would be filled by this book in the non-preposed version (I believe that you should read this book). Although an intonational break is possible in (9), it is not as common as with left-dislocations/HTLDs. Since constituents like this book in (9) lack a(n obvious) resumptive element and instead leave a gap, they have generally been analyzed as involving movement to the preposed position, and thus they have been likened to focus, rather than to left-dislocation (Cinque 1990; Casielles-Suárez 2004; Radford 2018; among others).

I now concentrate on the behavior of left-dislocations and topicalizations in the context of interest (i.e., in between complementizers). As shown by the novel contrast in (10), whereas native speakers of English accept hanging topics in recomplementation contexts, they reject topicalized phrases. Incidentally, in his corpus, Radford (2018) reports hanging topics/left dislocations in between thats, but not topicalizations.⁵

My claim in Section 4.5 will be that the contrast in (10) is attributable to locality constraints on movement, with only phrases that can be externally merged in-between thats being able to be recomplementized (i.e., hanging topics/left dislocations). Accordingly, the claim made at the beginning of this section to the effect that only constituents that are capable of being dislocated can occur in-between thats/ques in English (and Spanish) recomplementation needs to be qualified to include those constituents that can be dislocated and which are derived by direct merge/base-generation in their superficial position between complementizers.⁶

A third type of constituent that can be preposed (and which, as shown by examples such as (4)a, b, can appear flanked between thats) includes adjuncts like local adverbial/prepositional adjuncts (cf. (11)a, b and (11)c, d) and local adverbial clauses (cf. (11)e, f).

If we follow Cinque (1990) and Villa-García (2015), among others, in taking HTLDs to be elements of the DP/NP category, then the preposed phrases in (11) cannot be regarded as hanging topics (see Radford 2018 for much relevant discussion). This possibility is further weakened by the fact that PPs and adverbials can iterate, whereas hanging topics have been claimed to be very high elements that are limited to one per clause (though see Emonds 2004 for a dissenting view). Ostensibly, sentences like Despite the lack of funds, we will throw the party ___ could be interpreted as leaving a gap, much like their topicalization counterparts in (9). However, dislocated PPs/adverbials do not show the same distribution as topicalized phrases (in that the former, but not the latter, can occur in recomplementation constructions; cf. (4)a, b and (10)b); my claim will be that those constituents that can only be derived by movement cannot be hosted in between thats/ques due to locality restrictions. Similarly, following the work of Uriagereka (1988), Rizzi (1990), Law (1994), Bošković (2016), and Rizzi & Bocci (2017), adjuncts can be directly merged on the edge of the clause that they modify (i.e., such constituents need not be derived by movement; see Section 4.5 for evidence).⁷

What is clear is that a preposed PP/adverbial bears a certain informational-structural load that its non-preposed counterpart lacks. Therefore, it is scarcely surprising that such constituents were analyzed by Rizzi (1997) as a sub-type of topic (i.e., dislocation). They can in fact precede left-dislocations, as shown by the following piece of data from Radford (2018: 107):

However, the PP/adverbial – left-dislocation configuration is by no means the only one we find in spoken English. Instances of the left-dislocation – PP/adverbial word order are also attested, as shown by the following example from Radford (2018: 107), which also confirms the recursive nature of the projection occupied by adverbial/PP phrases:

What the data above suggest is that whatever projection dislocated PPs and adverbials occupy, it must be able to precede and follow the category hosting bona fide dislocated constituents. For Villa-García (2015), matters are simplified by assuming that the relevant phrases all target TopicP (perhaps in a split-TopicP framework wherein TopicP can be divided into specialized projections, namely those for hanging topics, PPs/adverbials, etc.). By contrast, Radford (2018) contends that PPs and adverbials that are in the left periphery should be analyzed as being specifiers of ModifierP. Given the existing evidence, exemplified above, this category could in principle precede or follow TopicP (ModifierP > TopicP; TopicP > ModifierP; etc.). Radford (2018) argues that English recomplementation features adjunct PPs/adverbials/clauses or hanging topics, as shown in (4), so for this author recomplementation constructions involve the lexical realization of the head of TopicP or of the head of ModifierP. Anticipating the upcoming discussion, I will use the category TopicP for both dislocated modifiers and arguments functioning as left-peripheral topical elements, noting that a more sophisticated analysis where the topic domain is expanded to include a specific category housing dislocated modifiers is not incompatible with the proposal made here. In fact, there is crosslinguistic evidence militating in favor of a TopicP analysis of preposed adjunct clauses. Aboh (2015: Ch. 6) shows that the Gbe language Gungbe has a specific marker yà specialized for topics, illustrated in (14). (Incidentally, this variety also manifests a distinct focus marker wἑ, found with focal phrases and wh-phrases alike, which occur below topical constituents; Aboh argues that yà is in fact the spell-out of Topicº, with the focal marker heading Focusº).⁸

As far as adjunct clauses are concerned, when the protasis (i.e., an if-clause) appears to the left of the apodosis (i.e., the main clause of a conditional sentence), then the if-clause is also accompanied by the topic marker yà, suggesting that we are dealing with a topic here, as shown in (15).

In light of this fact, I argue that even if dislocated adjunct clauses do not constitute canonical cases of HTLD/left-dislocation per se (see Section 4.5.1), there is evidence from typologically unrelated languages that adjunct clauses are topical when dislocated, hence amenable to a TopicP analysis. Put differently, the above provides support for the proposal that left-occurring adjunct clauses are topics. I now turn to the Spanish case with special reference to recomplementation environments.

2.2 Spanish

In Spanish (and in Romance languages such as Asturian, Catalan, and Italian), two constructions are reported to deploy topics/dislocations, namely HTLDs and Clitic-Left Dislocations (CLLDs). The first one is equivalent to its English counterpart in (7), and is exemplified for Spanish in (16).

Note that in (16), in much the same way as in its English equivalent, the (dislocated) hanging topic must be a DP/NP bearing default case, a function discharged by nominative case in Spanish.

In marked contrast to the HTLD construction, CLLDed phrases are not limited to nominals, can bear structural case, can iterate, and are not necessarily set apart from the rest of the clause by an intonational boundary. (17), which involves a dislocated dative object, illustrates the phenomenon of CLLD in Spanish. Note that since a mi hermana is a nominal group, the corresponding pronominal clitic le occurs in the sentence.⁹

As is known, there is a debate regarding the derivation of CLLDs in languages like Spanish. Authors such as Barbosa (2009) for Portuguese and Fernández-Rubiera (2009) for Asturian claim that CLLDs are the result of base-generation where they surface; López (2009) submits that CLLDs in Catalan and Spanish result from movement from a VP-internal position (see also Villalba 2000); and Martín-González (2000), Bošković (2001), Villa-García (2012; 2015), and Martínez-Vera (in press) make the proposal that CLLDs can be the result of either base-generation (i.e., external/direct merge) or movement (internal merge) (see Section 4.5.1 for further details).¹⁰

Dislocated PPs can be misleading in Spanish, since they lack a(n overt) clitic — due to the impoverished clitic system of the language (Casielles-Suárez 1997) — and therefore they may in principle be taken to instantiate a phenomenon other than CLLD. An example of a dislocated argumental PP is provided in (18).

Nevertheless, Catalan provides evidence for the CLLDed nature of such phrases. As shown by (19), dislocated locatives (PPs) such as en aquesta casa in Catalan are accompanied by an attending resumptive clitic pronoun (hi) (Bonet 1991: 23).

The Catalan facts exemplified by cases like (19) provide crosslinguistic plausibility for the CLLDed analysis of dislocated PPs (Casielles-Suárez 2004; Villa-García 2015).¹¹ However, it is still possible that PPs — in particular those which are non-argumental — constitute adjuncts that are base-generated in the left periphery, much like their English equivalents in (11) (Uriagereka 1988, inter alia). This could of course also be the analysis of preposed adverbial adjuncts and clauses in Spanish, exemplified in (20)b (see also the data in (5)b, c).

Whatever the ultimate analysis of adjuncts and adverbial clauses in Spanish (and in English) turns out to be, what matters for our current purposes is that such phrases can be directly-merged dislocated constituents. This derivational possibility allows them to be featured in recomplementation contexts. As will be argued in Section 4.5, there is a locality issue that arises in the presence of secondary complementizers: movement operations across recomplementation that/que are banned, which is why only base-generated constituents can appear between complementiers (see the relevant section for evidence that adjunct clauses/PPs can be the result of merge or movement in both languages, with the former derivation being the one found in recomplementation cases).

Summarizing, topical, dislocated phrases in English come in three types: left dislocations/hanging topics, topicalizations (though see fn. 6), and adjunct adverbials/PPs and clausal adverbials. In Spanish, dislocations come in the form of left dislocations/hanging topics, CLLDs (which include — but may not be limited to — DPs and PPs), and adjunct adverbials and clausal adverbials. Except for topicalizations in English (Cinque 1990, inter alia), all the other constructions can be derived by external merge in the position where they are pronounced. As will be shown and as hinted at above, only those dislocates that can be derived by base-generation in their surface position can appear between thats/ques in recomplementation configurations, since movement cannot cross secondary complementizers.¹²

2.3 What about foci?

A final question which arises is whether recomplementation configurations can host other types of left-peripheral specifiers such as foci. The answer to this question turns out to be negative, as shown by the English and Spanish examples in (21), which indicate that foci, much like English topicalizations, cannot occur in flanked position.¹³

A plausible question to raise is why this should be the case. One obvious answer is that foci are simply incompatible with TopicP, which can easily be explained under a feature-driven framework (e.g., Uriagereka 1995): the features of the focal phrase do not match those of Topicº. There is yet another reason why focal phrases are not compatible with recomplementation that/que. Focal phrases are standardly assumed to be the result of movement. As will be argued in Section 4.5, there is a locality effect with the secondary instances of complementizers this paper is concerned with. As a consequence, movement across doubled complementizers is impeded, which prevents focal phrases from occurring in the context of interest.

Having established that recomplementation is associated with topical/dislocated elements — and not focal ones — in the next section I set out a proposal which takes as its basis the dislocated, topical nature of recomplementation configurations in both English and Spanish.

4.1 Iterative thats/ques

As is well known, topical or dislocated phrases such as adjunct PPs/adverbials, adjunct clauses, and CLLDs can be iterative, that is, there can be multiple occurrences of topical elements in languages like English and Spanish (Rizzi 1997 et seq.). This is illustrated for English in (26) and for Spanish in (27), where instances of (embedded) dislocations are boldfaced.

Under Rizzi’s approach, the boldfaced phrases in these examples sit in different recursive TopicP projections. If Rizzi is correct in assuming that TopicP is a recursive phrase, and that each topical constituent sits in the specifier of its own TopicP projection, then the prediction is that in principle we should be able to find examples of multiple-complementizer sentences involving an instance of that/que below each topic, on the assumption that secondary that/que is the spell-out of Topicº. Examples (28) and (29), the latter of which is inspired by Escribano (1991), bear out this prediction.

Under the analysis outlined in Section 3, the English and Spanish examples in (28) and (29) would receive the account in (30), simplified by showing only the relevant constituents. (Note that the subscripted consecutive numbers are only meant to serve as visual aids to show the various complementizers in the sentences).

That recomplementation that and que can be recursive confirms the topical nature of the phenomenon and is thus compatible with the TopicP analysis of recomplementation patterns pursued here. The data also argue against competing proposals such as the FinitenessP analysis, which assumes that recomplementation that/que spells-out Finitenessº. This is so because for one thing, FinitenessP is not related to topichood, and also, because FinitenessP does not manifest recursion (Rizzi 1997 et seq.).

4.2 Multiple dislocates in sandwiched position

Much like every embedded topic can in principle be followed by an instance of that/que, it is also possible to find various contiguous dislocated topics sandwiched by overt complementizers, as shown in (31), from Radford (2018: 130) and (32).

The availability of (31) and (32) is not surprising, given the possibility of not lexicalizing the heads of the intermediate TopicPs, as shown abstractly in (33), which again assumes the recursiveness of TopicP.

A different approach here would be to assume stacked, multiple specifiers of TopicP, with a shared head, spelled out as that/que. However, the existence of such data as those in the previous subsection, with an overt head after every dislocate, and those in Section 4.3 below, which involve a that/que-less dislocate after an instance of recomplementation that/que, seem to favor the analysis pursued here (though see Section 4.5 for additional discussion).¹⁷

4.3 Dislocates below recomplementation that/que

If the projection housing flanked dislocated constituents is recursive and that/que (optionally) spells out its head position, then we should be able to find instances of that/que + dislocate + that/que + dislocate sequences. This is confirmed by the English data in (34) (examples a and b are reported in Radford 2018: 183). Note that (34)a is an example of a low left-dislocated phrase (followed by a coreferential pronominal acting as the subject), (34)b comprises a dislocated prepositional phrase, and (34)c is an example of a dislocated adverbial phrase.

For Spanish, the same situation can be replicated, as shown by the spontaneous data in (35), heard in the Spanish region of Asturias.

The facts just reviewed are captured straightforwardly within the analysis currently pursued, as shown by the simplified abstract derivation in (36).

In the next subsection, I investigate the possibility of having focal phrases right below recomplementation that/que.

4.4 Flanked dislocates followed by foci

Consider again Rizzi’s proposal, shown in (37):

An interesting prediction can be derived from this analysis of recomplementation. In principle, a focal phrase cannot precede an instance of that₂/que₂, as we have seen. However, a focal phrase should be able to occur in the position below the secondary complementizer in both languages, given that FocusP is structurally preceded by TopicP in (37). This prediction is fulfilled by (38) and (39), where capitalization signals focus. Note that the preposed negative constituent AT NO TIME triggers inversion of the auxiliary will, which confirms its focal character (in Spanish, although the subject is null, if it were overt, it would have to occur postverbally, exactly as expected if the relevant left-peripheral constituent is focal; see Torrego 1984 for details).

The data in (38) and (39) argue in favor of the abstract analysis in (37)/(40) and in turn confirm the dislocated character of recomplementation configurations in English and Spanish, since topics routinely precede foci; it is then not surprising that …XP + that/que sequences, which I argue belong in TopicP, are able to precede focal phrases (cf. (40)).

The English and Spanish recomplementation data reviewed so far can be accounted for — and in fact receive a unitary explanation — under a TopicP analysis. In so doing, the crosslinguistic evidence adduced above also allows us to draw a more detailed mapping of the architecture of the left periphery in both languages. I now investigate the locality of recomplementation that/que and its consequences for the derivation of flanked topical constituents.

4.5 The locality of secondary that/que

Villa-García (2010; 2012; 2015) observes that secondary que complementizers in Spanish induce a locality-of-movement problem. This is illustrated for long-distance extraction in (41), where strikethrough marks deleted copies of the relevant moved items:

The contrast in (41) indicates that when que₂ is present, long-distance extraction of qué ‘what’ is impossible, as in (41)a, whereas in the absence of que₂, extraction can occur without difficulty, as shown in (41)b.¹⁸ Note that what creates a locality issue is the low complementizer, not the high one, as indicated by (41)b.

As far as movement to the specifier of TopicP is concerned, that is, movement to the position flanked between complementizers, Villa-García goes on to observe that in the presence of recomplementation, reconstruction effects are not exhibited by the flanked dislocate; reconstruction effects are only displayed by the version of embedded dislocation without recomplementation, as shown by the contrasting examples in (42).

In (42)a, which contains recomplementation que, the bound variable interpretation does not obtain: the only interpretation of the sentence is that in which the dative argument a su jefe makes reference to an extrasentential individual (e.g., the speaker’s boss, or somebody else’s boss); the interpretation in which the quantified expression todo Dios binds the possessive su in the dislocated phrase is not available. The version without recomplementation que in (42)b stands in stark contrast with (42)a in that in the absence of que₂ the bound reading is available, i.e., todo Dios can bind a su jefe. Villa-García (2015) takes this contrast to indicate that the dislocate flanked by ques is externally merged in its surface position in (42)a. The dislocate cannot cross que₂ because this crossing operation would create a locality effect along the lines of what happens in (41)a. In other words, the position flanked between ques is a merger site, rather than a landing site. This is why the base-generated dislocate does not exhibit reconstruction effects in (42)a (i.e., there is no low position below todo Dios from which a su hijo could be c-commanded under reconstruction at LF; in a Copy-Theory-of-Movement approach, the low copy would simply not exist, thus making it unavailable for purposes of LF interpretation). On the other hand, in (42)b, without que₂, a su hijo has undergone movement to the left periphery, and consequently the low position below todo Dios is available for reconstruction for purposes of the bound variable interpretation (or, alternatively, a low copy of a su hijo). This is precisely what Villalba (2000) and López (2009) claim for run-of-the-mill CLLD (without recomplementation): that such constituents are derived by movement. In short, the derivation of each element would proceed as in (43).

This analysis assumes that CLLDs in Spanish can in principle be derived by (i) base-generation/direct, external merge in their superficial position or by (ii) internal merge/movement to the left periphery (Martín-González 2002; Villa-García 2010; 2012; 2015; Martínez Vera in press), as mentioned in passing in Section 2.2. The only derivational possibility available in the case of recomplementation is that in which the flanked constituent is externally merged in the position where it surfaces, that is to say, between complementizers (cf. (i)/(43)a).

A similar situation can be found in relation to adjunct clauses. The examples in (44) involve anaphors, which, as is well known, need to be bound (i.e., c-commanded) by a local antecedent, in compliance with Condition A of the Binding Theory. Chomsky (1995) shows through famous examples such as Which picture of himself did Bill say that John bought? that the anaphor can refer either to Bill or to John, suggesting that there is a low copy of which picture of himself below John at some point in the derivation. This is exactly what happens in (44)b: a sí mismo can be bound by either el abuelo or Juan. This suggests that the adjunct clause containing the anaphor has undergone movement to the left-periphery (cf. (43)b). However, in the corresponding sentence with recomplementation que, (44)a, only the interpretation in which the anaphor is bound by the highest DP, el abuelo, obtains; the anaphor cannot be bound by Juan, which follows from the claim that sandwiched dislocates are base-generated in between ques (hence there is no lower position to which they can reconstruct for purposes of anaphor binding, in breach of Condition A) (cf. (43)a).

As shown by the English paraphrases of the relevant Spanish examples in (44), the same situation obtains in English, with binding by the embedded subject being possible only in the absence of recomplementation that.¹⁹ I return to the locality effect with English recomplementation that momentarily. (The data at hand also show that adjunct clauses can be derived either by base-generation or by movement, as noted in Section 2.1, a claim that will be further substantiated in Section 4.5.1).

The locality effect with recomplementation que has commanded considerable attention in the recent literature (see, e.g., Villa-García 2010; 2012; 2015; González i Planas 2014; MacDonald 2015; Martínez Vera 2017; in press; Cerrudo & Gallego 2018; Torrego in preparation; see Radford 2018 for recomplementation that in English). The actual implementation of the locality issue remains a point of contention in the literature, with accounts that range from Comp-t effect violations, barrier/island violations, and CED (Constraint on Extraction Domains) violations to issues arising in relation to phase theory, antilocality, and the labeling algorithm.

Be that as it may, since our goal is to outline the similarities and differences between recomplementation in English and recomplementation in Spanish, what is important for our current purposes is that recomplementation that in English also displays island-creating properties. This is shown by the contrasting examples in (45), reported in Radford (2018: 125) and based on an implicit claim made by Haegeman (2012). This highlights yet another similarity between recomplementation in Spanish and its English counterpart: the locality effect with secondary complementizers in recomplementation (see also the reconstruction examples involving anaphors above).

In (45)a, the boldfaced fronted argument what their parents did for them can only appear on condition that that₂ is absent. Radford (2018) takes this piece of data to indicate that a moved element cannot appear above that₂. Again, much like in Spanish, crossing a high instance of that is not a problem, as indicated by (45)b; it is only recomplementation that that induces a locality issue (though see Radford 2018 for a potential counterexample).²⁰

Similarly, only elements that are derived by base-generation in the position where they surface can be featured in English recomplementation contexts (see the English counterparts furnished in the paraphrases of the Spanish examples in (44)). As noted in passing in Section 2, left-dislocations (aka hanging topics), which are resumed by a pronoun in the argumental position corresponding to the dislocate, can appear in flanked position. Throughout his discussion of prior analyses of the derivation of left-dislocations (i.e., HTLDs), Radford (2018: Ch. 2) provides extensive empirical evidence from English that such dislocates are derived by direct merge in their surface position.

In contrast, topicalizations, which leave a gap standardly assumed to stem from movement, cannot occur in the position flanked by thats. This is shown by the novel contrast presented in (10), repeated here in (46).

This should come as no surprise, since that₂ induces a locality problem, as shown by (45)a, and consequently only dislocates that can be merged in their surface position are able to occur in recomplementation environments (cf. (46)a), much like in Spanish (cf. (42)a/(44)a).

In light of the locality of recomplementation, a note is in order regarding the lexicalization of secondary that/que. The claim so far is that that/que-XP-that/que dislocates are externally merged in the flanked position where they surface, and that overt secondary that/que induces a locality-of-movement problem. Villa-García (2015) makes the proposal that base-generation and que realization go hand in hand: sandwiched dislocates are directly merged where they surface; in contrast, when recomplementation does not occur, this is because the moving XP crosses que and que is deleted, since it displays island-creating properties and unless it is PF-removed, it creates a locality problem. Note that on this view the last-resort operation of secondary-que deletion is an instantiation of the widely held view that ellipsis/deletion ameliorates locality violations (see Villa-García 2015 and references therein). This is schematized in (47). (47)a shows the base-generation derivation, and (47)b shows the movement derivation, with concomitant *-marking of the offending element que, which marks it for deletion in PF.

(47)	a.	… que	XP,	que			(cf. recomplementation)
	b.	… que	XP		XP		(cf. regular embedded CLLD)

Under this approach, que₂ is always present in the case of embedded dislocations, its absence being the result of a PF-deletion operation motivated by movement of the XP to the specifier of the projection headed by que₂ and the need to avoid a locality violation caused by movement across the troublemaking secondary complementizer. Thus, the optionality of recomplementation que is only apparent under this approach, as the underlying derivations are different for each case (cf. (47)). Appealing as this approach is, it has come up against criticism (see, e.g., MacDonald 2015). In this connection, the presence/absence of that₂/que₂ is linked to the more general and non-trivial question of whether the overt realization of Cºs is purely the result of a superficial, PF Spell-Out (i.e., pronunciation) parameter (Rizzi 2013; Rizzi & Bocci 2017) (see Villa-García in preparation).

There is indeed a recent approach which is in keeping with Villa-García’s claim about base-generation vs. movement, but which does not rely on (non-)PF realization of secondary que. Martínez-Vera (in press) has recently argued that when que₂ is present, base-generation of the dislocate takes place à la Villa-García (2015). Setting aside technical details of Martínez-Vera’s proposal, the difference between the two analyses concerns recomplementationless CLLDs, which for Martínez-Vera are also moved to the specifier of TopicP, but in such cases que is not present and instead verb movement to Topicº occurs, as in (48).

In sum, Martínez-Vera’s account does not put a premium on the PF realization/deletion of que in those varieties featuring recomplementation: que₂ only occurs with base-generated recomplementation dislocates.²¹

4.6 Further similarities between English and Spanish: Other reduplicative complementizers

Villa-García (2015) notes that on occasion, instances of que in Spanish precede and follow parentheticals and other expressions, as in (54). Interestingly, Radford (2018: 124) furnishes similar examples in English, such as that in (53).

The examples above show that an element that is not a modifier or an argument of the embedded clause can occur in pre-secondary-that/que position. Such cases may not be instances of recomplementation per se, and instead constitute restarts, afterthoughts or simply concatenated discourse markers (see Section 6.2). Whatever the case may be, the existence of such examples highlights another parallelism between English and Spanish in the domain of multiple-that/que clauses. In fact, looking at the phenomenon of complementizer reduplication beyond that/que more generally, both Radford (2018: 175) and Villa-García (2015: 217) offer examples like the following, where the interrogative complementizer if/si is reduplicated:

The nature of the data in (55)/(56) is hitherto poorly understood, but again we are dealing with another clear parallelism between English and Spanish multiple-complementizer sentences. I now turn to some differences between the two languages, with the aim of determining whether the unified analysis of recomplementation pursued so far can be retained.

5.1 Different force in the embedded complement

One aspect in which English and Spanish recomplementation differ concerns the force of the complement clause that follows the recomplementized sequence, depending on the interpretation. The upshot of this section is that since recomplementation occurs most frequently with verbs of communication, the nature of the embedded clause will depend on the embedding possibilities of the selecting predicate. As we will see, Spanish verbs of speaking are far more permissive than their English equivalents, and indeed allow not just declarative, but also desiderative/exhortative, interrogative, and exclamative complements. However, my claim will be that this has little to do with the phenomenon under scrutiny, and instead is owing to independent properties related to the types of verbs embedding subordinate clauses in the two languages.

5.2 High that/que

English and Spanish stand in glaring contrast to one another in relation to the possibility of having a null complementizer in complement clauses. Whereas English generally allows the complementizer to be absent, Spanish does not, as shown by (70). (Rodríguez-Ricelli 2018 notes that this construction is possible with thinking/judgment predicates in Los Angeles and Mexican Spanish; RAE-ASALE 2009: 3232 cites some examples from 17^th-century Spanish where que is absent in cases akin to (70)b).

Prescriptively, that cannot be omitted in English if left-peripheral constituents such as dislocates occur in the embedded clause (see Bošković & Lasnik 2003). However, Radford (2018: 131) reports numerous examples to the contrary. For instance, one such case is data consisting of recomplementized constituents preceded by a null (Ø) complementizer, as in (71).

Given (70)b, it follows that a Spanish sentence like (71) should be ungrammatical. This is borne out by the data in (72).

The data in (71) and (72) have important consequences for the analysis of recomplementation. Demonte & Fernández-Soriano (2014) claim that while they endorse Villa-García’s claim that recomplementation is dependent on a dislocated phrase in Spanish (see Section 2), it is also true that recomplementation can only occur in Spanish in the presence of a high que. The English data in (71) suggest that the overt realization of the highest head in the periphery (by assumption, Forceº) is not a sine qua non condition for secondary that to surface. Put another way, the lexicalization of the highest left-peripheral head in the clausal spine is not a requirement for lower that heads to be lexically realized. With respect to root clauses, Radford (2018) reports recomplementized dislocates as well, although these are much less frequent than embedded ones. This construction is exemplified in (73), from Radford (2018: 157) (see Campos 1992 for a similar pattern in Béarn Gascon). It is of note that this structure is rather odd and unacceptable by native speakers (incidentally, all the examples of this structure Radford notes come from the same broadcaster).

Now, let us see what happens in root clauses in Spanish. As shown by the contrast in (74), only when a high que occurs can recomplementation que surface. Again, the presence of a high que is crucial for recomplementation to occur. In this case, we are dealing with a high quotative que (Etxepare 2010; 2013; among others), interpreted as a report of somebody’s speech.

The generalization which emerges from the English and the Spanish data just presented is that in English, a low that complementizer below a dislocate does not depend on the overtness of a high that (i.e., (that) XP that) whereas in Spanish, the (overt) presence of the second complementizer relies on the first one (i.e., *(que) XP que).

This parametric difference regarding the licensing of recomplementation that/que could be accounted for as follows. It would be possible to assume that Spanish recomplementation is a non-root phenomenon, and that it can only be licensed in subordinate contexts, which are generally headed by que. Note that quotative clauses heralded by que, (74)b, albeit appearing in matrix contexts, display features of embedding, in the sense that said sentences are sentential complements of an abstract quotative predicate; these include the use of the interrogative complementizer si ‘if’ and tense transposition. In this sense, therefore, they can be considered to be embedded clauses. If recomplementation is an embedded phenomenon, and embedding is usually signaled by an overt complementizer, then it follows that recomplementation que will appear under a (subordinating) high que (see also Villa-García 2015 for the prospect of head-to-head selection across left-peripheral heads). It may be the case that the (rather scarce) putative phenomenon of root recomplementation in English is an instantiation of a different construction altogether, which would explain its rarity (though see Radford 2018, who notes that (73) could be analyzed in parallel fashion to the embedded data in (71)). Thus, English recomplementation too may require a higher Forceº-like licensing head, which may remain silent, a PF option that is not available in Spanish (cf. (70)b).

Whatever the right analysis of the asymmetry brought to light here regarding the obligatoriness of high que in Spanish (but not of high that in English) and the correct account of its potential role in the licensing of recomplementation turn out to be, the TopicP account advocated here for both English and Spanish can be maintained. The proviso is that perhaps more general aspects of the left periphery determine the obligatory PF realization of the highest complementizer in Spanish embedded clauses, a condition that does not seem to be active in English (see Radford 2018 for the nature of that spell-out in English). In fact, as Radford (2018) rightly points out, the impossibility of having that as the first word in a root sentence in English (*That it’s snowing) is indicative that secondary complementizers in such contexts as recomplementation must be licensed by a superordinate constituent (i.e., a dislocate), as argued for in Section 2, which further substantiates the TopicP analysis of recomplementation pursued here. Villa-García (2015) postulates the existence of a similar constraint, dubbed the Filled-Spec Requirement, which posits that the specifier position of TopicP needs to be overtly lexicalized, thus preventing the realization of recomplementation que in the absence of a dislocate.³³ In a more general sense, and in connection with the working mechanisms underlying the left periphery, Luigi Rizzi (p.c.) notes that the fact that Spanish recomplementation requires a high que demonstrates that a dependency must be expressed between the high complementizer and the non-high one, probably in the spirit of feature “circulation” across left-peripheral heads à la Rizzi (2013).

5.3 Ellipsis

Villa-García (2012; 2015; 2016) observes that ellipsis after recomplementation que is possible in TP-ellipsis-with-polarity-particle constructions in Spanish. This is exemplified in (75):

This pattern involves a contrastive topic acting as the remnant of ellipsis, optionally followed by recomplementation que, and a (focal) licensor such as sí/ no/ también/ tampoco/ de ninguna manera… (‘yes/ no/ also/ (n)either/ on no account…’), contingent on the polarity of the antecedent clause (Villa-García 2016). As is well known, English lacks TP-ellipsis with a polarity particle, so replicating (75) is not viable, as shown by (76).

The closest equivalent to Spanish TP ellipsis in English would be VP ellipsis, as in (77).

Contrived as (77) is, it is still less deviant than (76). As an anonymous reviewer notes, left-dislocation in English does not tend to be employed for the expression of contrast. In this example, the contrastive reading comes from the presence of but. This may at least be one factor influencing the degradation of (77), whose counterpart without that₂ (Ø) has the same acceptability status. The contrast between (76) and (77) points out that the differences observed between elliptical clauses in the context of recomplementation in English and Spanish are not due to recomplementation per se; rather, such differences are rooted in the different ellipsis operations available in each language (TP ellipsis in Spanish and VP ellipsis in English), irrespective of recomplementation. In this connection, Andrew Radford (p.c.) notes that non-recomplementized preposed DPs yield better results in VP ellipsis contexts, as shown by (78). The fact that (78) is more acceptable than (77) suggests that speakers may find recomplementation sentences generally wordier, which also happens with respect to the Spanish examples.³⁴

The overall conclusion arising from the preceding discussion is that the availability of ellipsis is not dependent on the presence vs. absence of recomplementation.

5.4 Sí-que sequences

Villa-García & González-Rodríguez (in preparation) show that the emphatic sí que ‘yes that’ construction (Hernanz 2007; Batllori & Hernanz 2013; among others) can be preceded by a recomplementized topic, as in (79). The function of the sí-que sequence seems to be that of underpinning the speaker’s commitment to the truth of the proposition.

Clearly, English lacks a direct equivalent:

Nevertheless, English has the means of expressing the same idea in a different fashion (e.g., by means of adverbs), as indicated by (81).

Again, this apparent asymmetry between English and Spanish is unrelated to recomplementation; it is caused by the availability (or lack thereof) of independent grammatical mechanisms.³⁵

5.5 The issue of selection

A perennial issue in discussions on the structural geometry of the left periphery as well as in the literature on complementation is the problem of selection. After all, the configuration of the left periphery of an embedded clause is to a large extent determined by the selectional properties of the embedding predicate. As noted in Section 5.1, whereas a verb like decir in Spanish can select virtually any complement clause that may in turn display a varied left periphery, a factive verb is more limited in its selection properties, disallowing, for instance, recomplementation (Demonte & Fernández-Soriano 2009; de Cuba & MacDonald 2013; Villa-García 2015; among others), as (82) shows. According to de Cuba & MacDonald (2013), predicates including factives select a CP that lacks the full inventory of projections necessary to permit recomplementation.³⁶

English factive clauses featuring recomplementation are exemplified by the examples in (83), kindly provided by Jonathan Bobaljik (p.c.), who finds no contrast between the sentences in (83) and their non-factive counterparts (the judgment is shared by Andrew Radford):

The conclusion to be drawn from the data presented here is that whereas recomplementation is not readily available under factives in Spanish, it is possible in English; as a result, it would be far from unreasonable to make the assumption that factives in English embed a complement with a left periphery that differs from that of the equivalent verb types in Spanish, with only the former freely allowing recomplementation.

In light of our discussion regarding verbs of communication in 5.1, which indicates that the same verb in English and Spanish (i.e., a verb that is generally considered to be a translation equivalent) may display different selection properties, it is scarcely surprising that a verb that is apparently the same in English and Spanish manifests distinct selectional constraints in each language, thus allowing recomplementation in some cases but not in others (e.g., factive complements in English are compatible with recomplementation that, but not in Spanish). Pending new advances and empirical evidence bearing on the types of predicates that can subcategorize for a clause containing a periphery that can include recomplementized dislocates, I conclude this subsection by noting that in principle English and Spanish recomplementation constructions use the same grammatical tools, potential differences being due to factors other than the design of the left periphery (e.g., the idiosyncratic selectional properties of verbs in each language). Similarly, it is of note that such selectional restrictions are subject to much inter-speaker variation (Radford 2016; 2018). Whatever the case may be, this is not to underestimate the relevance of selection in left-peripheral matters, an issue that future research should address (see González i Planas 2014: 44–47 for much relevant discussion).

More generally, the asymmetries manifested by English and Spanish in the context of recomplementation configurations appear to be attributable to independent factors. This means that these differences do not pose a threat to the unified TopicP analysis pursued here, and in actuality our review of the foregoing asymmetries has revealed additional parallelisms between the two languages with regard to double-complementizer constructions, such as the existence of Finitenessº that/que in both English and Spanish. In the following section, I turn to an investigation of the reasons why the phenomenon of recomplementation exists.

6.1 Previously identified functions

Recomplementation has been claimed to be a reinforcement of Force (Martín-González 2002; Demonte & Fernández-Soriano 2014), a topic marker (Paoli 2006; Villa-García 2012; 2015), a quotative marker (González i Planas 2014), and a processing marker (Casasanto & Sag 2008). In what follows, I explore the possibility that secondary that/que can also perform the role of discourse marker.

6.2 Discourse marker

A proposal that I would like to introduce here is that one of the roles of recomplementation that/que is that of being able to function as a discourse particle. As noted by Wiltschko & Heim (2014), discourse markers signal a relationship between the element they introduce (i.e., the embedded clause) and the prior segment (i.e., the dislocated material). Interestingly, the dislocate featured in recomplementation patterns is somewhat external to the clause, as argued above. As a discourse marker, the recomplementation complementizer is not propositional (i.e., it does not change the truth value of the clause) and resembles a parenthetical. In the words of Brinton (1996: 267), discourse markers serve to “reveal or make explicit those connections already operating in texts.” Discourse markers are also said to be multifunctional (Hansen 1998, inter alia), a feature that can be ascribed to recomplementation complementizers, as I claim in this paper. Hansen (1998: 236) points out that “semantically, markers are best seen as processing instructions intended to aid the hearer in integrating the unit hosting the marker into a coherent mental representation of the unfolding discourse.” The fact that recomplementation que is preceded by a prosodic boundary, as noted in Section 4.5.1, is wholly compatible with this role, and, as noted, recomplementation has been claimed to aid processing.³⁷ Although many definitions of discourse markers have been put forward, authors including Maj-Britt Mosegaard Hansen (p.c.) define them as elements that have a connective function, which secondary that/que can be said to perform.³⁸ What is more, secondary complementizers can co-occur with unambiguous discourse markers such as pues ‘then/thus’ in Spanish (Villa-García 2015), as shown in (84) (note that discourse markers are known to cluster in many languages).

Consider the naturally-occurring English sentence in (85). In this example, the dislocated if-clause is followed by an unambiguous discourse marker —then. If spoken, then would most likely be preceded by an intonational boundary.

Now, it would be possible to manipulate (85) and create a similar sentence, with equivalent meaning, but with a reduplicative instance of that. As (86) shows, the meaning stays basically the same, and that performs a connective function, in much the same way as then. Put differently, if recomplementation that can fulfill a discourse-marking function, then it should be able to be replaced by a bona fide discourse marker (cf. (85) and (86)).

Likewise, it is important to note that since discourse markers can co-occur in many languages, we should in principle be able to see different discourse-marking elements together. This is confirmed by (87), which resembles what happens in the Spanish case (cf. (84)).

Lastly, as expected, a genuine discourse particle can precede secondary that. In other words, discourse markers can be flanked by thats. This is shown by (88).

Spanish also allows this possibility, as the data in (89) indicate ((89)b has been taken from González i Planas 2014: 47; see also the examples in Section 4.6).

Discourse markers tend to mark the resumption of a topic after a digression (Hansen 1998). In the context of recomplementation clauses, it could be argued that secondary complementizers resume a sentence after topical material has been introduced. This seems particularly useful if the topical element is long or internally complex, consonant with the proposal noted above that such complementizers can be processing markers (see Radford 2018 for relevant discussion regarding the length of the flanked dislocate). Furthermore, it is well known that discourse particles are usually drawn from categories such as conjunctions (Fraser 1999) and that they are grammatically optional (Hansen 1998). Recomplementation that and recomplementation que manifest both characteristics. Another property attributed to discourse markers is that they tend to be oral rather than written (Brinton 1996: 33–35), which is precisely the case of recomplementation that/que (although not invariably so, as has been seen). All in all, recomplementation que embodies various features pointing to its status as a discourse marker.³⁹ Regardless, the analysis of discourse markers remains debated in the literature, with some discourse markers being less prototypical than others (Jucker 2002).

Although a full categorization of recomplementation that/que as markers of discourse needs to be carried out and the current proposal remains speculative, the ability of reduplicative complementizers to function as discourse markers gains plausibility in light of the evidence adduced here.⁴⁰

6.3 A polyvalent element

The different functions identified by existing research on the topic of recomplementation — and the additional one presented in the previous section — are not necessarily mutually exclusive. For instance, while a doubled that/que may be a “reminder” of the force of the embedded clause — and thus a subordinator, as it were, at the same time it can delimit the topic field (or just signal a topical element) and indicate that the upcoming elements form part of the comment (in the topic-comment articulation established by the topical phrase and the material below it), serve as a marker of quotation, and connect the discourse.

The prospect that a single complementizer may serve more than one function in the same realization is independently corroborated by the following data, where the same que complementizer realizes more than one single function. Suppose that speaker A utters (90):

Now, speaker B, who is A’s interlocutor, has not heard or understood (90) properly. Speaker B then says something equivalent to Sorry? or What?. Interlocutor A, trying to clarify, exclaims (91).

(91) instantiates root jussive/optative clauses (see Section 5.1.1). Interestingly, it also exemplifies the quotative construction of Etxepare (2010), since the sentence in (91) is interpreted as a command/wish on the part of speaker A. Critically, however, que is obligatory for most speakers, as would be expected if we are dealing here with jussive/optative que. That que needs to be overt in this case is shown by (92), which is the ill-formed que-less counterpart of (91).

Note that regular quotative sentences which do not involve jussive/optative interpretations are still grammatical in the absence of the quotative element, as shown by (93)c.

This example demonstrates that whereas the presence of que would be the most natural response to speaker B’s question, it is still possible to drop it. This is not possible in (92), however, indicating that one of the functions performed by this que, namely that of a jussive/optative marker serving to lexicalize the subjunctive (or jussive/optative) mood, requires the overt realization of que.

The existence of polyvalent ques lends plausibility to the analysis of peripheral heads as conflated projections, which started with Rizzi’s (1997) contention that in the absence of left-peripheral phrases such as topics and foci, ForceP and FinitenessP could effectively be conflated into a single projection, CP or FFP (ForceFinitenessP), presumably for reasons of economy of projections (see Section 3). In the particular case illustrating the polyvalence of complementizers in Spanish just reviewed, both QuotativeP and FinitenessP appear in a single projection, which bears features relevant to both quotation and mood, with the head of this projection being ultimately spelled out as que (Villa-García 2015). These facts are also relevant to the aforementioned proposal of feature transmission across CP heads.