No ellipsis-contained antecedents: adjunct-licensed ellipsis in Spanish

2.1 Spanish AGAC as TP-ellipsis

We characterize Spanish AGAC as TP-ellipsis, similar to what Fernández-Sánchez (2023) has proposed for modal complement ellipsis. One piece of evidence for this comes from verb movement. Spanish has obligatory V-to-T movement (e.g. Zagona 2002), making the minimal ellipsis site for verb-deleting processes TP. Additionally, auxiliaries, hosted in T, can never surface overtly when gapping (5a) or AGAC (5b) occurs, i.e., when the lexical verb is deleted and two remnants survive. Auxiliaries have to be part of the deleted phrase (Reglero 2006), see (5).

Additional evidence that AGAC deletes at least up to TP comes from the impossibility of voice mismatches between antecedent and ellipsis site. Neither gapping (6) nor AGAC (7) allow voice mismatches, as expected if the ellipsis site contains Voice.

Finally, Spanish AGAC allows non-subject remnants, (8). This confirms that the ellipsis site must be large enough to contain the derived subject position in Spec,TP, therefore supporting the conclusion that AGAC elides a constituent at least as big as TP.

As we have described the pattern, Spanish AGAC seems to closely resemble gapping in terms of what is elided. However, Spanish AGAC empirically differs from Spanish gapping with respect to the licensing of determiner sharing, i.e., the interpretation of a single overt determiner on two NPs (e.g., Johnson 2000; Arregi and Centeno 2005). In Spanish, determiner sharing is available across gapped coordinations (9), but not in the cases we have identified as AGAC (10).

We preliminarily conclude from this that there are actually two possible derivations in Spanish for constructions which on the surface look like gapping: a coordinate structure patterning mostly with English gapping (9) and the AGAC structure (10), which elides a bigger constituent than English gapping.^[1]

2.2 Premise conditionals and AGAC

As mentioned above, not all types of adverbial clauses can be antecedents for AGAC. Conditional clauses can provide an antecedent, as in (1b) – but only on a particular reading of the conditional clause. Conditionals that license AGAC are not true hypothetical conditionals, but so-called premise or factual conditionals (Castroviejo and Mayol 2024; Iatridou 1991; Schwenter 1999), illustrated in (11). Premise conditionals are used to structure the discourse by making manifest a proposition that is privileged as the context to properly assess the following clause (Haegeman 2003: 317).

For instance, in (11) the speaker is not setting a condition for Bill to leave, which the truth value of the proposition depends on. The speaker is highlighting a premise, often one which echoes a previous assertion in context, which is presented as background to assess the speaker’s claim, which is that Bill should leave.

The premise conditional structure which can provide the antecedent for AGAC has recently been thoroughly analysed by Castroviejo and Mayol (2019, 2021, 2024), who term them echoic contrastive conditionals. They are echoic, as the protasis is naturally interpreted as directly reproducing a claim made earlier by the interlocutor (rather than setting up a true conditional relationship between the truth of the protasis and apodosis). In fact, in (1b) above, repeated here as (12), the natural interpretation is that the interlocutor has just boasted or complained that he graded 50 exams.

True hypothetical conditionals can be distinguished from such echoic or factual conditionals in various ways. Premise conditionals reject focus particles like only or even as modifiers of the protasis (Bhatt and Pancheva 2006), perhaps because the protasis acts as a theme that sets a background to evaluate the consequent of the conditional. (13) is infelicitous on the ‘premise’ reading, presupposing that Bill is unhappy.

Accordingly, adding focus markers to the conditionals in Spanish is incompatible with AGAC, (14), indicating that the antecedent-providing adverbial is not a hypothetical conditional.

For the same reason, the protasis of premise conditionals cannot be focused by clefting (15a), and again focus by clefting is incompatible with AGAC in Spanish (15b).

Sequence of tense and mood also illuminate the contrast. Spanish has a contrast between indicative and subjunctive in finite verbs (Quer 1998). Indicative mood is compulsorily used in premise conditionals, while subjunctive is used in the protasis of true hypothetical conditionals, enforcing a verb in conditional in the consequent (see RAE and ASALE 2009:§47.8 for a detailed description). Again, AGAC is impossible if the protasis contains a verb in subjunctive: (16) sharply contrasts with (12) above.

While the subordinator si allows both hypothetical and non-hypothetical interpretations, other conditional expressions are specialised for hypothetical meanings, with verbs in the subjunctive and always subject to sequence of tense: this is the case of con tal de que ‘provided that’ and en caso de que ‘in case’, as well as siempre y cuando ‘just in case’. None of these subordinators allows a premise conditional interpretation, and none of them licenses AGAC in the apodosis, (17).

2.3 Other adverbial clauses and AGAC

Some non-conditional clauses can provide the antecedent for AGAC, precisely to the extent that they can have such high ‘factual’ or ‘discourse-structuring’ readings. The subordinator cuando ‘when’ allows a temporal reading or a ‘factual’, ‘discourse-structuring’ reading. The temporal reading can be controlled for by adding entonces ‘then’, and is subject to sequence of tense restrictions. (18) with entonces forces the temporal reading where María paints only during the same time period as Juan draws. (18) without entonces is ambiguous between the two readings.

AGAC is only possible in the non-temporal condition without entonces, (19).

A similar situation arises with mientras ‘while’. Normatively in modern Spanish, mientras ‘while’ only has a temporal simultaneity reading (20a) which enforces imperfective tenses, and mientras que ‘whereas’ should be used for an adversative contrast reading (20b), which allows perfective tenses. However, many speakers use mientras for both meanings. Only the non-temporal reading of mientras allows AGAC, as (1c), repeated here as (21), illustrates.

Concessive clauses also show a contrast. The Spanish literature has identified two main types of concessive clauses (see e.g. Flamenco 1999): restrictive concessives (22a), where the two clauses are temporally autonomous from each other and indicative must be used, and pure concessives (22b), where subjunctive is used and there must be a correlation, parallel to hypothetical conditionals, between the temporal forms used in each.

Only the first reading, which directly expresses opposition between two existing facts, licenses AGAC, (23).

Temporal clauses with antes de que ‘before’, después de que ‘after’, cada vez que ‘whenever’ cannot provide the antecedent for AGAC (irrespective of whether the verb is in indicative or in subjunctive), see (24).^[2]

We argue that what differentiates the possible antecedents from the impossible ones is the position in the tree to which they are adjoined, with clauses that do not license AGAC being base merged within TP.

2.4 Peripheral versus central adverbial clauses

Haegeman (2012) and Frey (2011), among many others, have divided adverbial clauses into three groups, based on their degree of interaction with the host clause: non-integrated clauses, peripheral clauses, and central clauses. We will limit discussion to the latter two groups. Central adverbial clauses (CACs) are merged low and are dependent on the host clause, for instance with respect to tense, aspect, and mood. Peripheral adverbial clauses (PACs) show a more independent behaviour, and are merged at a higher position. We demonstrate in this subsection that adverbials which can provide the antecedent for ellipsis pattern with PACs, and non-antecedent adverbials pattern with CACs.

Sequence of tense provides one diagnostic (Badan and Haegeman 2022). A configuration like (25), with perfective past in the host clause and perfective future in the adverbial, does not follow a sequence of tense. Correlatively, (25) can only be interpreted in the adversative or premise reading, i.e., the reading associated with PACs. This is the structure that can license AGAC.

True temporal (26a) or hypothetical conditionals (26b), on the other hand, respect the sequence of tense, indicating that they are in a low enough position to depend on main clause tense; such structures disallow AGAC.

For pure concessive aunque ‘although’ clauses, (27a) shows that a present subjunctive clause can receive a future interpretation dependent on the future tense in the main clause, suggesting that these clauses are CACs merged below the tense of the main clause;^[3] such configurations also fail to license AGAC (cf. (23b), repeated here as (27b)).

In addition, AGAC-licensing adjuncts cannot be in the scope of focal operators like only or even. Recall the contrast in (14): ellipsis is only possible without the focal operator.

Furthermore, temporal and adversative readings of mientras differ in binding possibilities. A subject in the host clause can bind a pronoun in the adverbial clause only if it is a CAC. In (28), only the temporal reading of mientras is available under variable binding. This indicates that the temporal adverbial is c-commanded by the subject of the main clause. The adversative reading is impossible, suggesting again that those antecedents that provide antecedents for AGAC, e.g., adversative mientras, are PACs, merged high in the left periphery.

This is also true of restrictive concessives: variable binding is impossible with an adverbial headed by aunque ‘although’ in the restrictive indicative reading with indicative mood (29a), while it is available in the pure concessive hypothetical use with subjunctive mood in the adjunct (29b).

The ellipsis-licensing adjuncts thus pattern with Haegeman and Badan’s (2022) characterization of peripheral adjuncts, while the adjuncts that cannot license AGAC show properties of CACs.^[4]

2.5 All adverbial clauses license AGAC if outside the ellipsis clause

To complete the empirical picture, the distinction between central and peripheral adverbial clauses dissolves when they are adjoined within a clause which subordinates the AGAC clause, as can be seen in (30).

In contrast with the examples presented in Section 2.4, the while-clause in (30) provides the antecedent even though it is a central temporal clause with purely temporal reading. Similarly, if the propositional content of the adverbial clause is established in a previous clause, as in (31), AGAC also becomes possible with central adverbial clauses.

The generalisation is that the adverbial clause licenses AGAC provided that it is outside the sentence that undergoes ellipsis. If this condition is met, temporal clauses, hypothetical conditionals, concessives and all other central clauses – which we do not illustrate for reasons of space – can provide the antecedent for AGAC.

3.1 No antecedents inside ellipsis sites

We adopt the ‘move-and-elide’ or ‘evacuating movement’ approach to clausal ellipsis (e.g., Merchant 2004; Sailor and Thoms 2014; Weir 2014; Broekhuis 2018; Schwarzer 2024). The two remnants of AGAC move to a left-peripheral position in the clause, with the rest of the clause being elided. (For simplicity we show the left periphery as a recursive CP here; of course it may have a more fine-grained cartographic structure à la Rizzi 1997.)^[5]

The question is under what circumstances ellipsis can take place, i.e. what can provide the antecedent and how. In the previous section, we have seen that adjunct clauses that can be antecedents for AGAC are merged outside of the elided TP, while adjuncts that cannot be antecedents for AGAC are contained within the elided TP. The generalization we therefore pursue is that adjuncts that cannot be antecedents for AGAC are those that are interpreted within the very material that is elided, while adjuncts that can be antecedents are interpreted external to the ellipsis site.

The generalization that an ellipsis site cannot contain its own antecedent seems intuitively logical and simply stated. However, we believe it is not a trivial conclusion: in particular, if it is correct, it provides evidence (a) that ellipsis identity (at least in AGAC) is calculated specifically over LFs, and (b) that ellipsis remnants move at LF.

To see this, observe that the surface (pronounced) structure of an example like (33) presumably contains a trace of the movement of the después-clause to the left periphery.

In general, clausal ellipsis sites can contain traces which their antecedents do not; for example, ‘sprouting’ cases such as (34).

So, in (33), simply copying the verb from the surface structure which is available in the adverbial clause – and ‘sprouting’ the trace of that very adverbial clause itself – should (one might imagine) suffice to license ellipsis in the main clause; but ellipsis is not possible. (The putative antecedent TP is underlined in (35b).)

The failure of ellipsis therefore seems to be semantic: the meaning of the elided TP has to contain the meaning of the central adverbial clause (because CACs are interpreted within TPs, i.e. they fully reconstruct in a structure like (33)), but the meaning of the TP inside the CAC obviously then cannot be identical to the meaning of the elided TP (because one contains the other as a part).

Even this, however, is not wholly trivial to capture. We believe, for example, that one prominent semantic identity condition on ellipsis, Merchant (2001)’s e-GIVENness, would incorrectly permit ellipsis in cases like (33).^[6] The reason is that Merchant’s e-GIVENness calculates mutual entailment ‘up to’ the existential closure of traces, variables, and focused elements in antecedent and ellipsis site. This can be seen in a contrast sluicing example such as (36).

Now consider the LF of (33) – with the CAC reconstructed to its base position. Even when reconstructed, the CAC is presumably in focus here; it is not deaccented, it is new information,^[7] and (on the surface) it is pronounced in the clausal left-periphery, as is common for information-structurally distinguished material. Yet if the CAC is in focus, e-GIVENness should – contrary to fact – license ellipsis here, as shown in (37).

We suggest that the pattern seen in AGAC can be best explained under two assumptions: (a) clausal ellipsis is licensed under identity of LFs (after reconstruction of elements such as CACs),^[8] and (b) ellipsis remnants move at LF (that is, they do not themselves reconstruct). In the following sections, we step through how this derives the observed patterns.

3.2 AGAC within the same clause

We have shown in Section 2.4 that the adverbial clauses that can provide antecedents for AGAC are PACs, specifically adverbials that, like the premise conditionals analysed by Castroviejo and Mayol (2019), are used to contrast two situations.

(39)

We assume that within the LF of the PAC si Juan compró un coche, the contrastive topics Juan and un coche move, so that the LF of CP_PAC above is as in (40):

(40)

The elided TP in (39) is therefore LF-identical to the TP inside the PAC (40), and so ellipsis is licensed. Note that the ellipsis remnants must be taken to move at LF, not just PF (contra Weir 2014); if they did not, the right identity would not be obtained (because Juan ≠ María and un coche ≠ una casa).

By contrast, adjuncts which reconstruct into the ellipsis site at LF cannot be the antecedent for ellipsis. This is the case with central adverbial clauses such as hypothetical si ‘if’, pure concessive aunque, and temporal cuando and mientras ‘when’, as well as clauses introduced by antes de que ‘before’ and después de que ‘after’. (41) represents their position within the TP, at LF.

(41)

Suppose the following (putative) LF structure in (43), which attempts to create an ellipsis-licensing configuration for (the ungrammatical) (42).

(43)

The two TPs that would have to be LF-identical are shown in boxes; but this is obviously impossible, as the higher TP contains the CAC itself, while the lower TP does not.

3.3 AGAC from a different clause

As we saw in Section 2.5, the central versus peripheral distinction dissolves when the antecedent TP is already merged outside the clause which undergoes AGAC. This is predicted by the present account. As long as the antecedent TP is not contained within the ellipsis site, it can provide the antecedent for ellipsis:

On the temporal reading, the CAC mientras Juan corregía 50 exámenes is interpreted at LF inside the superordinate clause. Because the TP in this CAC (underlined in (45)) is not contained within the ellipsis site at LF, ellipsis is possible:

The same of course goes for antecedents in preceding sentences in discourse, which will of course not be contained within the ellipsis site. The sentence in (46) is grammatical; the antecedent TP is underlined (we abbreviate by not showing the LF-movement of Juan and un coche).

Note that the empirical pattern illustrated by (44) indicates that it is not characteristics of the adjunct clauses themselves which account for the asymmetry originally noted between central and peripheral adverbial clauses as antecedents for AGAC. One could imagine arguing, for example, that peripheral adverbial clauses can affect the Question-under-Discussion (QUD, Roberts 2012) while central adverbials cannot, and that clausal ellipsis is anaphoric to the QUD or something similar (AnderBois 2014; Barros 2014; Weir 2014; Griffiths 2019 a.o.). Alternatively, one might argue that PACs, but not CACs, allow for the kind of internal movement that creates the relevant variable-binding configurations at LF (e.g. the movement of Juan and 50 exámenes in (45); see in particular Haegeman 2012 for discussion of movement inside CACs and PACs). However, it appears that any adverbial clause, including central ones, can provide the antecedent for AGAC once it is in the correct syntactic configuration, suggesting that the apparent asymmetry between the classes of adverbial clause is strictly a matter of their syntactic position, rather than their internal syntax or their semantics.^[9]