2. Prior analyses of HTLDs and of recomplementation constructions

Since the seminal work of Rizzi (Reference Rizzi and Haegeman1997 et seq.), the uppermost part of the clause, the traditional CP domain, has been split into several dedicated projections devoted to hosting different left-peripheral phenomena (i.e. Force > TopicP > FocusP > FinitenessP). For our purposes, the most relevant category is the TopicP projection (or topic field, if the periphery is further split, as in Benincà & Poletto Reference Benincà, Poletto and Rizzi2004), responsible for hosting left-dislocated material, such as left dislocations (see, e.g. Radford [Reference Radford2018: Ch. 2] and references therein). Under this analysis, an embedded sentence containing an HTLD like (2a) would be analyzed thus:

An equivalent sentence displaying recomplementation would receive the following analysis à la Villa-García (Reference Villa-García2012, Reference Villa-García2015) and Radford (Reference Radford2018), among many others:Footnote ¹

Appealing though this account appears to be at first sight, it faces a number of issues, especially in relation to the contrast between English and Spanish brought to light here (cf. (3) and (4)) and the status of HTLDs.

Recall that, as indicated by (3) and (4), embedded HTLDs require no double complementizer in English, but they do in Spanish. The impossibility of HTLDs in Spanish embedded environments is again highlighted by the following data:

On the assumption that the secondary complementizer is the spellout of a Topic-like head (e.g. Topicº), it follows that the difference between the two languages in the relevant respects is related to the lexicalization possibilities of Topicº:Footnote ²

Villa-García (Reference Villa-García2015) has advanced an analysis of the obligatoriness of the complementizer in Spanish by appealing to the lack of movement of HTLDs: que is always present and is only deleted when movement operations cross it (in the spirit of the rescue-by-PF-deletion approach that started with the seminal work of Ross [Reference Ross, Binnick, Davison, Green and Morgan1969]). Since HTLDs do not exhibit movement properties and are base-generated directly where they surface (though see below for a radically different view), no movement operation occurs past the low que in Topicº; therefore, the complementizer is not PF-removed. As should be obvious, this analysis fails to explain the English case, since HTLDs in English do not move either; yet the secondary instance of that is not mandatory. Needless to say, such an account of the contrast in (3) and (4) would be, at best, stipulative.

What is more, a monosentential account like that in (9) and (10) would pose the question of why a topic head would have to be overt in one language (e.g. Spanish) but not in another (e.g. English). Likewise, it is not obvious why a topic marker would have to have the same shape as the higher complementizer (que – que; that – that).

An even more pressing issue is that HTLDs have been shown to be syntactically disconnected from the host sentence with which they occur, with different works by authors of different theoretical persuasions providing arguments in favor of the conclusion that HTs are not part of the left periphery of the clause they occur with (e.g. Dik Reference Dik1978, Reference Dik1989; Cinque [1983] Reference Cinque1997; Ziv Reference Ziv1994; Acuña Fariña Reference Fariña and Carlos1995; Ott Reference Ott2015; Fernández-Sánchez & Ott Reference Fernández-Sánchez and Ott2020; Villa-García Reference Villa-García2023a). Though I will not review the vast body of evidence militating for this position, I will adopt the view that HTLDs neither move nor are base-generated where they surface: they are simply not part of the left periphery, and indeed the sentence in whose vicinity the HT occurs is syntactically complete without it.Footnote ³ Thus, the fact that the HTLDed constituent is outside the sentence to which it is contiguous further undermines an intrasentential analysis wherein the HTLDed phrase is in a left-peripheral projection like TopicP, as claimed by intrasentential analyses like (6), (7), (9), and (10).

For all the above, I contend that the account of the novel contrast in (3) and (4) cannot be that in (9) and (10); hence, the explanation must be sought somewhere else.

3. Recomplementation, bisententiality, and extrasentential HTLDs

In glaring contrast to monosentential analyses of double-complementizer constructions in languages like English and Spanish, Villa-García & Ott (Reference Villa-García and Ott2024) have proposed that recomplemention actually masks two separate sentences linked paratactically. Recomplementation is exemplified again in (11):

The authors provide a range of empirical arguments in favor of a bisentential account of data like (11), as follows (see the paper for evidence):Footnote ⁴

This analysis assumes two complete CPs that are independently generated root clauses and invokes ellipsis à la Ott (Reference Ott2014, Reference Ott2015).Footnote ⁵ On this view, the (boldfaced) secondary occurrence of the complementizer constitutes a restart in discourse reprising CP1.

Now, it was claimed in the previous subsection that HTLDs are extrasentential elements that are not generated on the left edge of their host clause. In this connection, Villa-García (Reference Villa-García2023a) has argued that HTLDs are also derived bisententially and furnishes an analysis that also assumes ellipsis, drawing on Ott’s work:

Analyzing HTLDs paratactically has a number of advantages, including an explanation for the absence of Principle B and C effects (Nicolás and al pobre are in different sentences in (13a) and thus no problem arises; see fn. 3), as well as the fact that hanging topics do not typically sit well in embedded contexts; they are indeed often classed as main-clause phenomena. I will then pursue the view that the embedded HTLDs this paper concerns itself with are inserted parenthetically, on the assumption that parentheticals are merged late in the derivation or that they exhibit transparency for purposes of selection (see, e.g. Ott [Reference Ott2015] and Radford [Reference Radford2018]).

To illustrate how this account works in practice, let us consider further examples of purportedly embedded HTLDs in Spanish, including both hanging topics with a resumptive pronoun/epithetic correlate and orphaned/unliked topics, as in (14d). Note that example (14c) features two restarts (on which, see Villa-García & Ott [Reference Villa-García and Ott2024]):

Under Villa-García & Ott’s bisentential proposal for double-complementizer clauses, an example like (14a) would involve CP1 and CP2, on a par with the parenthetically inserted/late-merged HTLD (I adopt a simplified version of Villa-García’s analysis in CP3 without making a commitment to its technical implementation):

In (15), the actual complement clause in CP1 restarts in the embedded site of CP2. In fact, Villa-García & Ott (Reference Villa-García and Ott2024) contend that the secondary instance of que serves to signal the restart explicitly. Under this approach, CP1 and CP2 are parallel to each other, which is why the same complementizer surfaces in both cases (que – que; see below for additional evidence from interrogative sentences to this effect). The HTLDed phrase un coche is merged parenthetically (i.e. it is not selected as the complement of decir; indeed, the presence of the high que indicates that it is a subordinate clause that follows). The fact that (subordinate) HTs are often perceived as anacolutha, or – especially intonationally – as planned sequences with interpolated material, is therefore not surprising.Footnote ⁷ Consequently, for CP2 to properly resume CP1 (recall that under this account, CP2 is a mere restart), the complementizer needs to be lexical, as it serves to overtly signpost the presence of the sentential complement of decir ‘to say;’ a null complementizer is ungrammatical in Spanish in this context (cf. (3a)).Footnote ⁸

At this juncture, the question arises as to why embedded HTLDs do not require that in English, as shown again by (16) (see also (2) and (4) above):

In order to see how the current account explains the Spanish–English contrast under consideration, let us look at the analysis of an example like (16) with overt secondary that under parataxis:

As they stand, there is in principle no difference between Spanish (15) and English (17). However, it is important to draw attention to one of the most obvious and well-known (but poorly understood) differences between English and Spanish: complementizer optionality in regular, non-recomplementation contexts, as shown by the contrast in (18):

If Villa-García & Ott are correct in their characterization of recomplementation configurations as restarts (i.e. CP2 basically reprises CP1), then it follows that the Spanish restart in cases like (15) will require the presence of que, since the complementizer cannot be absent in general in this language. This is not the case in English, however, where the complementizer is often dropped:

All in all, the difference between the Spanish and the English case regarding the (non-)obligatoriness of the secondary complementizer in embedded HTLDs boils down to the possibility of omitting the complementizer in English, but not in Spanish, in the regular case (cf. (18)), since the sentential constituent heralded by the second instance of que/that (CP2) is basically an overt marker of the restart of – and thus identical to – the initial sentence (CP1), as assumed in (19).

An immediate question posed by this state of affairs is what happens to the high complementizer in English, which should in principle be optional as well (since in fact the high and the low complementizer are one and the same element in two different, juxtaposed sentences). Data from spoken English bear out this prediction and indicate that there are different options, which by the way contravenes traditional claims made in the literature that embedded left-peripheral material forces the lexicalization of the high that (on which see, among many others, Grimshaw [Reference Grimshaw1997]):

The situation in (20) adds to the examples furnished so far and shows exactly what the restart/paratactic proposal predicts, since the higher and lower complementizers behave alike. The four logical possibilities obtain:Footnote ⁹

By contrast, Spanish does not tolerate sentences akin to those in (20), which is what we expect, given how rigid Spanish is in terms of complementizer omission (cf. (18a)):

It is of note that complementizer doubling is also attested with interrogative complementizers (Haegeman Reference Haegeman2012; Villa-García Reference Villa-García2015; Radford Reference Radford2018), as shown by the following examples:

Three native speakers, two from American English and one from Canadian English, observe that examples along the lines of (23b), with the reduplicative interrogative complementizer, feel much more natural than their counterparts without it (cf. ^??I wonder whether Kyle Quentin Wolfk, Ø we can count on him_k/that bastard_k), which is wholly compatible with the analysis pursued here. Moreover, Haegeman (Reference Haegeman2012: 85) provides a real, written example of multiple (and in fact distinct) [+interrogative] complementizers in embedded indirect questions, which the author herself claims ‘decidedly deserves further study.’ Note that this example features an adverbial in sandwiched position, not a genuine HT, though:

Although the ‘repeated’ complementizers are semantically equivalent, they differ from one another (if – whether), which further reinforces the restart nature of the construction (Villa-García & Ott Reference Villa-García and Ott2024): [_CP1 I wondered [if …]] … [_CP2 I wondered [whether …]].Footnote ¹¹

Lastly, cross-linguistic evidence from languages such as spoken Dutch, which behaves like Spanish in the relevant respects, further substantiates the analysis proposed here. As noted by an anonymous reviewer and as indicated by (25), allegedly subordinate hanging topics in Dutch also require double dat ‘that’, which cannot be left out in non-recomplementation contexts, in much the same way as its Spanish counterpart:

Overall, the paratactic approach to recomplementation adopted here sheds new light on several properties of double-complementizer constructions and, crucially, offers a less–ad hoc–than–competing–proposals account of the obligatoriness of secondary que and the optionality of secondary that with embedded HTLDs in Spanish and English, respectively.Footnote ¹²