Hanging Topic Left Dislocations as extrasentential constituents: toward a paratactic account. Evidence from English and Spanish

3.1.1 Binding and bound variables

As noted, HTLDs that have a clause-internal corresponding element establish a coreference relation with said element, as suggested again by the following sentences:

Under monosententiality, it is not clear how to void the Condition B effect that would arise if the HTLDed phrases (that man and Hugo) c-commanded the pronouns with which they corefer, contravening Condition B of the Binding Theory:

The HTLDed phrase may be base-generated in the specifier of a projection like XP in (45), but this does not preclude it from c-commanding the pronoun in the VP phrase, not matter how far down in the syntactic tree the pronoun occurs. Therefore, it is not at all obvious how to prevent a Condition B violation in such cases under an integrated analysis along the lines of (45). Note that such effects certainly ensue in sentences with the subject higher than the pronominal:

The same situation holds for Condition C effects, as referential expressions should be free in the overall structure (i.e., sentence) containing them, as indicated by the contrasts in (47):

Again, monosententiality provides no convincing answer to why Condition C is not violated in the case of HTs and their correlates, as shown by (48).

Note in passing that examples (48)a, b, d actually feature cases where the HT and the correlate are the same element, the DP John/Patrick/el dinero, exactly as expected if the second sentence is an independent, syntactically complete sentence (much like what we observe in relation to unambiguously independent sentences, as in (47)c); however, this state of affairs is rather problematic under unisentential accounts.

For reasons like this and others to be explored below, in this paper, I take the characterization of HTs as ‘extra-sentential’ constituents seriously (Dik 1978, 1989; Cinque 1997 [1983]; Ziv 1994; Acuña Fariña 1995; Ott 2015; Keizer 2020, among others). In fact, my major theoretical claim is that HTs belong to a preceding, elliptical sentence juxtaposed to the host sentence, as outlined in the preceding section:

Under bisententiality, the cofererence between HT and resumptive is not established grammatically at the level of the sentence, but at the rhetorical level, in discourse grammar (see Section 3 for further details).^[14] Hence, no Condition B or C violations arise. Therefore, once analyzed as extra-sentential elements, the relevant examples do not contravene either of the principles of Binding Theory (see also Ziv [1994: 631]). The same is found, unsurprisingly, in parallel cases of unambiguously separate sentences, as in (47)c/(50), which are analogous to (49), modulo the ellipsis assumed in CP1 in (49).

All in all, under the two-sentence proposal, the potential problem of what look like anomalously coreferring nominals posed by monoclausality ceases to exist. What is more, the need for HT and resumptive to be coreferential follows from the assumptions laid out at the beginning of Section 3 without further ado.

Lastly, HTLDs stand in stark contrast with ClLDs and topicalizations in not being able to feature anaphors or reciprocals, as shown by (51):

This state of affairs follows naturally under bisententiality, since the anaphor is in an elliptical copular sentence in which it fails to be bound by a suitable antecedent:

The same can be said of the unavailability of the bound-variable interpretation with HTLDed constituents (for Spanish, see, especially, López 2009 and references therein):

The data just reviewed are fully consistent with the bisentential, paratactic analysis of HTs proposed here (as the operator, e.g., every student/todo quisqui, does not c-command the possessive pronoun, namely his/su, since operator and pronoun belong to two different sentences, CP2 and CP1, respectively, much as in the case of anaphors in (52) above).

Overall, the anti-connectivity displayed by HTs can straightforwardly be accounted for by assuming that the HT and its correlate belong to two separate sentences (CP1 and CP2), linked to one another paratactically.^[15] Needless to say, the evidence adduced here strongly militates against a movement analysis of HTs in a monosentential setting under integrated approaches (cf. (19)b); the binding and variable-binding evidence is at best marginally compatible with base-generation of the HT in a high position in the left periphery, although both derivations run up against non-trivial problems.

3.1.2 Agreement and pronouns

Another aspect that underscores the anti-connectivity evinced by HTs is agreement. If hanging topics are extra-sentential elements that are juxtaposed to the host sentence, one would expect a certain degree of agreement mismatches between the HT and the element associated with it inside the sentence to which it is adjacent, since the (pragmatic) connection between the two is rather loose and in no case intra-sentential. Examples like (54) confirm that lack of full agreement is sometimes the case:

Spanish also displays such mismatches:

These data argue strongly against a movement analysis under monosentential/integrated accounts (Radford 2018; van Riemsdijk and Zwarts 1974); clearly, the HT cannot be a mere copy of the second NP arising via movement (e.g., full details and latest transfer news – it).^[17]

Further, agreement facts in relation to collective nouns like gente ‘people’ in Spanish provide additional support for the claim being put forth here that the HT occurs in a sentence that paratactically precedes the host. Despite making reference to a group of individuals, gente is syntactically singular and feminine:

By contrast, López (2007: 81–85) notes that despite being syntactically singular, a word like gente is semantically plural. Thus, in discourse, López (2009: 15) argues, “a pronoun that refers back to gente will show up in plural form.” Note that in this case, gender is also default masculine. This is shown by (57):^[18]

This situation makes a very clear prediction regarding HTs. If HTs are in a different sentence from the host and the relationship between HT and host is discursive, rather than purely syntactic, then HTs should be related to a plural entity in the host sentence. In other words, we should observe a pattern of behavior akin to that manifested across sentences, as in (57), rather than inside sentences, as in (56). This prediction is borne out, as speakers manifest a strong preference for a masculine-plural correlate of the HT containing la gente:^[19]

The agreement facts just presented corroborate the extra-sentential character of HTs, exactly as predicted under parataxis (i.e., [_CP1 … HT] [_CP2 host sentence]).

Finally, a related point can be made on the basis of Brazilian Portuguese pronouns, which I will illustrate through the sequence a gente ‘lit. in the people.’ Brazilian Portuguese is a language that has been reported to display only a partial-null-subject system. Moreover, at present, a gente is used instead of the first-person plural form nós ‘we.’ However, a gente agrees with a third-person singular verb form:

If a second, embedded clause appears, then a null subject can occur (Jairo Nunes, pers. comm.), as shown by (60); I use pro for expository reasons, without making a commitment to the theoretical status of the coreferential null subject:

Nonetheless, across sentences, a null subject is not possible: the pronoun (e.g., a gente) must be repeated:

When HTs are brought into the picture, their pattern of behavior aligns not with what we observe in (60) for elements across dependent clauses within the same sentence, but with what we see across independent sentences, as in (61). The pronoun must be repeated:

Parataxis provides a clear answer to the pattern observed: a null subject is not possible across independent sentences in Brazilian Portuguese, and the second sentence (CP2, the host) must be syntactically complete in (62) irrespective of the occurrence of a HT. Since Brazilian Portuguese in such contexts behaves like a non-null-subject language (we are dealing with a root sentence here, not with an embedded one, as in (60)), the second occurrence of the pronoun (a gente) must be present so as to satisfy the requirement to employ an overt preverbal subject operative in this language –a null subject is not licensed (i.e., without an overt preverbal subject, CP2 would be incomplete from the syntactic point of view). Consequently, HT examples featuring pronominals like a gente in Brazilian Portuguese indicate that HTs mask two underlying root sentences juxtaposed paratactically, much like in the case of unequivocally separate sentences (cf. (61)).

3.1.3 Case

The anti-connectivity of hanging topics is substantiated by the fact that HTs across the world’s languages have been reported to appear in absolute form (without case-marking) (Hidalgo 2002), or to bear either default case (accusative case in English, as shown again in (63)a, and nominative in Spanish, as in (63)b) or, more traditionally, casus pendens (‘hanging case’):

The question which arises in relation to the case borne by HTs is how such nominals come to bear accusative or nominative case, depending on the language. Radford (2018) observes that this may be (i) default case (Schütze 2001), as the HT is outside the scope of any potential case-assigner (though see Merchant [2004a] against the notion of default case); (ii) the result of an abstract spec-head agreement relationship with a (null) topic head; or else (iii) the product of being preceded by an abstract (null) transitive topic-introducing preposition of the as-for type (e.g., as for me, they don’t trust me). The plausibility of such an account for the Spanish case is called into question by data like the following, indicating that topic-introducing expressions in Spanish typically assign accusative case (see also (31)b above):

There is one topic-introducing sequence, however, compatible with nominative case (albeit less common than en cuanto a):

Attractive as this account may seem, it is not clear that crosslinguistically all as-for constructions (or topic-introducing constructions more generally) constitute genuine HTs (on this issue, see, among others, Acuña-Fariña [1995]; Villalba [2000]; Krapova and Cinque [2008], and Stark [2022], among others). Moreover, an implicit assumption in some of the options suggested by Radford (2018) in (i)-(iii) is of course monoclausality, as the HT is regarded as an intra-sentential element, albeit left-peripheral.

Recall that the currently-pursued proposal constitutes a drastic departure from the above in that the HT belongs to an underlying copula sentence that has undergone ellipsis and which is juxtaposed to the host clause. The HTs are predicates in those copular clauses and as such receive the relevant case in each language: accusative in English (cf. (66)) and nominative in Spanish (cf. (67)) (see Ott [2016] for a similar approach to nominative case assignment in predicative non-restrictive nominal appositives in languages like German, illustrated for English above in (28)):

Consequently, the bisentential account of HTs dispenses with the need to invoke ad hoc case-marking mechanisms for hanging topics; the case of such nominals is determined in CP1 in a standard, local fashion: they receive predicative accusative (English)/nominative (Spanish) case. Other languages follow this pattern too. For instance, Sigurðsson and van de Weijer (2021) show that the case exhibited in copular sentences in Swedish is nominative, in parallel fashion to Spanish. HTs likewise bear nominative case in Swedish, as expected:^[20]

The account put forth here can also go a long way to explain why it is only DPs/NPs that can be featured as HTs. A PP, for instance, would not be possible in predicative position:^[21]

A bare NP, such as niños in (15)b above, is also tolerated in the predicative position in CP1 (i.e., [_CP1 el tema es/son niños] …), which is consistent with the observation that such phrases sit well as HTs.

Having discussed anti-connectivity in terms of binding, agreement, and case, we now turn to other properties of HTs easily accommodated under bisententiality, and which further allow us to tease apart competing proposals.

3.1.4 Comma intonation/pause potential

As noted in Section 2, one of the distinguishing features of HThood is that the HT is more often than not separated from the sentence it accompanies by a salient intonational boundary (Acuña-Fariña 1995; van Riemsdijk 1997; Emonds 2004; Feldhausen 2016, among others). This is often rendered in spelling by a comma or by suspension points, as has been noted.

This is wholly compatible with the claim made here that the relation between the fragment HT (CP1) and its host sentence (CP2) is paratactic, each sentence forming a separate intonational phrase (cf. Nespor and Vogel 1986):

Fragment and host thus typically exhibit ‘comma intonation’ (intonational isolation), exactly as expected if the sequence is composed of linearly juxtaposed root clauses in a paratactic rather than intrasentential arrangement. In Section 3.1.8, this claim is further substantiated by HTs whose illocutionary force differs from that of the host sentence, such as ‘interrogative’ HTs, which display their own interrogative intonation, unlike the falling intonation contour displayed by the non-interrogative sentence to which they are contiguous.

3.1.5 Really external, or rather, extra-sentential: complementizers, V3, and clitic directionality

Spanish provides a syntactic context which strongly confirms the outside-the-host-sentence character of HTLDs. In Spanish, the answer to a what-is-going-on or what-happened-to-you kind of question is typically heralded by the complementizer que ‘that.’ The data are inspired by those in Villa-García (2015):

Now, if we try to promote the PP de mi madre to the very front, the resulting utterance is ungrammatical:

By contrast, if the element preceding que is a genuine HTLD (not a PP), the sentence becomes fully acceptable:

That a HTLDed phrase (cf. (73)), but not a ClLDed PP (cf. (72)), can occur in pre-complementizer position confirms that the HTLDed phrase is truly external to the clause with which it appears.^[22] Incidentally, a HT would be impossible in the immediate post-que position:

Not surprisingly, a true ClLDed constituent can sit well in the position immediately following que:

Summarizing, we have the following possibilities in what-happened contexts in Spanish (though see fn. 22):

Importantly for our purposes, only a true HTLDed phrase can occur in the pre-que position in the context at issue (cf. (76)b), which corroborates the claim made here that HTLDs belong in a separate sentence and are therefore even more external than ClLDs, which may be subject to more stringent proximity requirements with respect to the host sentence (i.e., they require parallelism with CP2). Under monoclausality, an external position above the highest element in the left periphery (presumably above ForceP) would have to be invoked for HTs. Alternatively, it could be claimed that the head whose specifier is occupied by the HTLDed phrase would be lexicalized as que. However, it is not evident how this last option would be implemented, as que can occur (and normally occurs) independently of the HTLDed constituent, as indicated by (71), which severs the HTLDed element from the complementizer occurring to its right in cases like (73). Such stipulations are not necessary under bisententiality, and hence this theoretical option is to be preferred.^[23]

Still within the realm of complementizers, it is important to discuss a well-known but poorly understood asymmetry in Spanish regarding the (im)possibility of embedded HTLDs. Several authors, including Zubizarreta (1999), have shown that HTs are confined to root contexts, examples like the following being illicit:

However, authors like Grohmann and Etxepare (2003), González i Planas (2011), and Villa-García (2015) have provided data to the effect that if a second que complementizer occurs after the (embedded) HT, then the HT becomes far more acceptable:

The obligatoriness of the second que in examples of embedded HTs like those in (78) has been shrouded in mystery to date (see Villa-García [2015] for much relevant discussion). Nevertheless, recent research has shown that secondary complementizers constitute restarts in discourse, signaling a new sentence that resumes the first one (Villa-García 2019; Villa-García and Ott 2023). If the second que marks the beginning of another sentence, as claimed by, e.g., Villa-García and Ott (2023), then the external nature of HTs (even in what appear to be cases of embedding) is not surprising: the HT is not part of the embedded clause, but outside it, much like in the cases reviewed above (cf. (71)–(75)):

Since the HT in the elided CP2 is not syntactically part of the sentence heralded by dice que ‘says that’ (cf. CP1), the restart (CP3) needs to resume CP1 by means of repeating the complementizer in (79), which accounts for its mandatory occurrence in such cases.^[24] In this sense, note that (78)b would be syntactically complete without the (seemingly embedded) HT:

  Moving away from English and Spanish momentarily, as an additional, crosslinguistic piece of evidence symptomatic of the external or, more precisely, extra-sentential nature of HTLDs, it is important to mention the well-known fact that in German, a prototypical example of a V2 language, HTs lead to apparent V3 orders (Ziv 1994; Ott 2015, among others):

Under the two-sentence analysis pursued here, this is explained away: in the host sentence/CP2, the verb is canonically in the second position, as expected in German (i.e., … [_CP2 sie lobten ihn]). The seeming V3 position of the verb in (81) is just illusory; it stems from the fact that der Professor is part of CP1, an independent elliptical copular sentence, juxtaposed to CP2.

Lastly, Bulgarian cliticization facts further support this conclusion. Bulgarian is a language which normally displays enclitics (postverbal clitics), (82)a, unless a preverbal element occurs that supports the clitic, yielding proclisis (preverbal clitics), (82)c. In other words, Bulgarian obeys Wakarnegel’s law, forcing unstressed pronouns to occur in syntactic second position.

Crucially, a HT is not sufficient to support the clitic, that is, a HT does not qualify as the first syntactic element in the sentence, leading to enclisis in such cases (contra the judgments reported in Krapova and Cinque 2008):

This follows naturally from the paratactic account advocated here: the pronominal HT az is in a separate sentence (CP1), and therefore the clitic me cannot be the first element in CP2; it needs to occur postverbally, in second position in CP2. Superficially, however, in an example like (83) the clitic seemingly occurs in third position, but in reality, it is the second element in its own sentence (CP2) (i.e., … [_CP2 pokanixa me …]). Under monosentential proposals, locating the HT in a high, left-peripheral CP-related projection would lead to the expectation that the HT should serve to support the clitic, contrary to fact.^[25]

Overall, the evidence regarding complementizers, V3 configurations, as well as clitic placement point to the conclusion that HTLDs are extra-sentential elements, exactly as argued under the bisentential analysis.

3.1.6 Insensivity to islands and islanhood

As noted in Section 2 in relation to examples like (17) and (18), one of the hallmarks of HTLDs is the well-documented fact that they do not obey islands. The examples therein focused on the coordinate structure constraint. That HTs are not sensitive to islands is shown again for [adjunct islands] in (84):

Given that islands are opaque domains for extraction, any account that assumes movement of me from the [bracketed structure] in cases like (84) would predict that the HT would be marooned within the island, contrary to fact. The paratactic account handles such cases in a straightforward fashion: me is never part of the island (or of the clause preceding the island, for that matter); it is in a different sentence, and thus it can coexist with an island occurring inside the host sentence. What is more, the HT is coreferential with an element within the island. There is no movement or attempted extraction at any point; the HT establishes no syntactic dependency with the host sentence. In effect, this is analogous to what happens in a sequence of unambiguously independent root clauses where coreference also obtains despite the presence of the island, just as predicted under parataxis:

Moreover, it has been noted in the literature that HTs themselves exhibit island-creating properties (Ziv 1994: 131; Cinque and Rizzi 2010, among others); that is, no movement operations can cross HTs.^[26] (86) illustrates:

Parataxis offers a direct solution to this issue while dispensing with the need to appeal to movement versus base-generation and to relative positions in the left-peripheral spine: under parataxis, the attempted extraction in (86) is cross-sentential. In traditional monosentential analyses, a cuál de ellas moves from the indirect object position of the predicate asignar to the front of the clause; however, in (86), yo is in its own (elliptical) sentence. A cuál de ellas cannot be extracted across another sentence, which results in strong ungrammaticality. If HTs are independent root sentences, movement dependencies crossing the HT boundary simply cannot be computed. More generally, this type of extraction is now ruled out on principled and general grounds as illicit extraction across sentences, on which see, e.g., Villa-García and Ott (2023). In other words, extraction can occur across clauses, under hypotaxis, but not across sentences, under parataxis, as is the case in (86).

3.1.7 Intercalated interjections

Hanging topics contrast with their topicalization homologs in being compatible with an intercalated interjection such as man and well (see also Al Pacino’s example featuring yeah in (1)a):

Interjections are common across sentences, but not within sentences, as the following pair illustrates:

The fact that an interjection like man can freely occur between a hanging topic and the (syntactically complete) sentence it precedes, as in (87)a, c, lends further support to the paratactic account wherein the HT is in a separate root clause, juxtaposed to the host clause, much like what happens in (88)a. Under monosentential/integrated analyses, we would have to postulate that the high position hosting HTs can be followed by a position able to host interjections (which in fact display extra-sentential properties themselves). These in turn would have to be higher than the position responsible for topicalized phrases and focal phrases, which are not possible above man, as in (87)b, d. Nothing of the sort needs to be stipulated under parataxis, which handles the intruding-interjection facts with ease.

3.1.8 ‘Interrogative’ hanging topics

RAE-ASALE (2009: 2979) furnishes examples where the topical element constitutes an interrogative segment followed by a non-interrogative sentence:

As shown by the paraphrases and by the following example, English hanging topics are also possible in this context:

These data are important for several reasons. First and foremost, only the HT is interrogative, which indicates that the force of the elliptical sentence it belongs to must be +Q. On the assumption that a C-type head bears the relevant interrogative features, ¿tus lentes?, ¿yo? and her? in the above examples must be in an (elliptical) interrogative sentence; under standard generative assumptions, the nominal cannot be interrogative on its own. Since the host sentences these +Q HTs superficially occur with are -Q, then there must be a preceding underlying interrogative source where they are generated.^[27]

These facts are highly problematic for monosentential, integrated analyses, which are forced to tolerate an initial left-peripheral +Q segment in a non-interrogative sentence. Since the CP layer/left periphery is responsible for marking the interrogative force of the sentence, it is not easy to see how to reconcile the two force types (i.e., interrogative and declarative) manifested in (89) and (90) under monosententiality. Both movement and base-generation analyses fall short of explaining the divergent force specifications displayed by different elements in what at first sight look like monosentential structures.

Under parataxis, by contrast, this is not only unproblematic, but also predicted, as shown schematically in (91) for RAE-ASALE’s Spanish example in (89)a:

¿Tus lentes? survives ellipsis in a +Q sentence, which is juxtaposed to the non-interrogative (i.e., -Q) CP2, yielding (89)a (see also (90)). Again, this is exactly what we observe in two root sentences that are uncontroversially linked just by juxtaposition:

Similarly, an account like that in (91) easily captures the rising intonation with which the HT is uttered in these examples, suggesting that we are dealing with an underlying yes/no question in CP1, as claimed by (91). Accordingly, the host sentence (CP2) displays the falling intonation typically manifested by declarative sentences. Again, this is rather challenging for monosentential, integrated accounts of HTs.

Finally, it is of note that the HT may also be exclamative (i.e., +Excl CP1), which should come as no surprise given the preceding discussion. This is shown by (93):

More generally, therefore, HTs may bear a different force specification from that of their host sentence, which is in effect what the overall paratactic account adopted here predicts.

The next subsection presents HTs exhibiting no resumptive element in the host clause.

3.1.9 Orphaned hanging topics

There is a subtype of HTs that occurs with no correlate in the host clause. Radford (2018) refers to this kind of HTs as P(ragmatically)-linked topics or, more metaphorically, as orphaned topics (see Lambrecht [1994: 193] for the denomination ‘unlinked’ topics). Radford goes on to note that such HTs are looked down on prescriptively. This is perhaps due, in part, to the fact that they are widely regarded as anacolutha. The examples in (94) illustrate this phenomenon in English and in Spanish:^[28]

The connection between HTs and the host sentences with which they appear in the examples just provided is rather tenuous: the HT and the sentence are just thematically related. In fact, as far as syntax is concerned, the HT does not even have a correlate (never mind a corresponding gap) in the sentence. This immediately rules out a movement analysis which derives the HT as moving from a sentence-internal position; even for base-generation analyses under monosententiality, how to derive cases like those in (94) is not evident: why would the structurally unconnected HT be in the initial numeration in the first place? The connection with the sentence is purely pragmatic, which raises the question of why the syntax would care about the HT in the first place. Note that this applies not only to orphaned HTs, but to all HTs discussed throughout, although the case of orphaned HTs highlights this issue even more evidently. Under parataxis, these questions are not even raised, as would be the case in a sequence of uncontroversially independent root clauses, exactly as expected:

Under parataxis, therefore, the existence of orphaned topics constitutes a natural empirical expectation of the account.

3.1.10 No right-dislocated hanging topics

Authors including Rodman (1997 [1974]: 47–51), Samek-Lodovici (2009: 354) and Fernández-Sánchez and Ott (2020: 16) point out that there is no Clitic Right Dislocated (ClRDed) counterpart of HTs. This is shown by the unacceptability of the following examples:

If HTs are presentational elements about which the upcoming sentence predicates something, then it follows that they should precede, but not follow, the relevant sentence. In other words, if HTs constitute a formal way of signaling the theme in the theme-rheme dichotomy, then presenting the theme after the rheme, as in the examples in (96), would be anomalous, to say the least. This is wholly consistent with the paratactic proposal advocated here, since the elliptical copular sentence (CP1) where the HT is generated introduces the topic of conversation and hence naturally precedes the host sentence (CP2). Nothing else needs to be said.

3.1.11 Hyperdetached hanging topics

As a final argument, I discuss what I will term long-distance, hyperdetached HTLDs:

That a HT need not be immediately adjacent to the sentence containing the correlate is a natural consequence of the paratactic account, and as a result provides additional empirical support for the overall proposal put forth in this paper. Such cases are in reality not different from orphaned topics (see Section 3.1.9); the current cases just happen to involve a resumptive/correlate which occurs in a sentence which is linearly far away from the sentence immediately contiguous to the hanging topic. Put another way, the correlate need not be in the linearly adjacent, host sentence: it can be further downstream.

Spanish also provides examples of hyperdetached HTs:

‘My friend Marimé_k, the other day I was with Juanín in the Suburbia. And well, it turns out that we got there with the car and there was a neighbor of hers_k crossing that dangerous main road recklessly.’

Additionally, data like the above further undermine integrated analyses in which HTs are sentence-internal constituents on the clausal left edge.