Spanish modal ellipsis is not null complement anaphora

2.1 Test 1: Syntactic mismatches

Building on Hankamer and Sag (1976), Depiante (2000: 33) contends that surface anaphora is subject to a strict parallelism requirement that disallows any sort of syntactic mismatch between the elided string and its antecedent. One piece of evidence in favor of this comes from the observation that sluicing, an instance of surface anaphora involving TP-ellipsis, does not allow for voice mismatches. In the following example, from Merchant (2013b), the antecedent clause is a passive sentence (Joe was murdered), whereas the elided clause is in the active form (who murdered Joe), a clash which renders the sentence ungrammatical.

Depiante provides the following analogous example featuring modal ellipsis, where the antecedent clause is in the passive voice, while the elided string is constructed so that it is interpreted with a verb in the active voice. Despite the mismatch, the sentence is grammatical, which Depiante takes as evidence against a PF-deletion analysis.

The existence of such mismatches has been a hot topic in the literature on ellipsis. As a general argument, however, it must be stated that the absence of a mismatch is not enough to exclude a PF-deletion analysis. For example, English VP-ellipsis is generally considered a case of surface anaphora (Aelbrecht 2010), and it is well known to allow certain voice mismatches, as the following examples (from Merchant 2013b) illustrate.

In order to explain the difference between sluicing (7) and VP-ellipsis (9), Merchant (2013b) proposes that in the latter voice mismatches are tolerated because the syntactic head that encodes voice, namely VoiceP, is located above the nodes computed for identity, i.e., the VP, so the VP in the antecedent clause is essentially identical to the VP in the elided clause. Conversely, sluicing, which involves TP-ellipsis, cannot tolerate voice mismatches because VoiceP is contained within the TP. Consequently, the TP in the antecedent clause is not identical to the TP in the elided clause.

Having clarified this, let us go back to Example (8). If, as I claimed in the introduction, ME in Spanish involves TP ellipsis, and if we assume Merchant’s proposal regarding voice mismatches in elliptical phenomena, then voice mismatches should not be allowed at all, contrary to what (8) apparently reveals. It is worth noting that in this example, the antecedent clause is passive and the elided clause is active. Some research (Arregui et al. 2006; Clifton et al. 2019; Frazier 2013) has pointed out that mismatches of this type tend to be judged as more grammatical than the reverse case (i.e., active antecedent – passive ellipsis). To explain this, Arregui et al. (2006) propose that speakers are able to repair some cases involving mismatches, especially when the antecedent is passive. The reason is that passive structures are misremembered as actives more commonly than the other way around. Hence, upon hearing a passive antecedent – active ellipsis site, the speaker is likely to “repair” the passive antecedent as an active one, thereby creating a matching effect.

To avoid a potential repair issue, we should also test voice mismatches in Modal Ellipsis in a configuration where the antecedent clause is active, and the ellipsis site is passive. The following example shows that if we transform (8) into such a configuration, the example becomes completely ungrammatical.

I provide two more sets of examples which show that voice mismatches are not tolerated in Modal Ellipsis, even though the passive antecedent – active ellipsis configuration sounds slighly better than the reverse case, which can be attributed to the repair effect mentioned above.

In conclusion, ME does not allow for voice mismatches, contrary to what is often claimed in the literature (Brucart and Macdonald 2012; Depiante 2000; Saab 2008). In this respect, ME behaves like sluicing (7), and unlike English VP-ellipsis, which is a welcome result if ME involves TP-ellipsis, as I defend.

2.2 Test 2: Missing antecedents

Grinder and Postal (1971) contend that it is possible to establish an anaphoric link with an indefinite within an elided VP, which means, obviously, that the gapped VP must have internal structure and cannot be thus analysed as a null proform. This is illustrated in (13a), from Hankamer and Sag (1976: 403), where it refers to the DP a camel in the elided VP. Note that it cannot refer to the DP a camel in the leftmost clause, as one cannot establish an anaphoric link between a pronoun and an indefinite expression under the scope of negation (13b).

Compare (13a) to the following example, from Hankamer and Sag (1976: 405), apud Bresnan (1971), which features do-it anaphora instead of VP-ellipsis. In this case it cannot co-refer with the DP a knife, which suggests that do-it anaphora does not involve PF-deleted structure, but rather it consists of a null proform. Under this view, the ungrammaticality of the following example stems from the lack of a potential antecedent for the pronoun.

Depiante (2001: 208) shows that ME seems to behave like do-it anaphora, and unlike VP-ellipsis, in that it is incompatible with missing antecedents. In particular, in this example Depiante shows that the null subject pro cannot co-refer with the DP un cuchillo ‘a knife’. This ought to be possible if ∅ consisted of a fully-fledged VP containing the DP un cuchillo.

Note, however, that this example is not ungrammatical, she simply gives it a question mark. It has been noted (cf. Bresnan 1971) that judgments regarding missing antecedents are often difficult, which undermines the strength of the argument. I find the Spanish example quite okay, and so does Saab (2008: 52), who proposes, instead, the following example.

On the basis of this example, Saab, like Depiante, concludes that missing antecedents are indeed impossible in Spanish ME. In my opinion, the main problem with (16) is that it is not clear what the exact content of the ellipsis site should be. The issue, in particular, has to do with what we consider the counterpart of ‘ninguno’ in the antecedent clause to be in the elided clause. If we consider it to be alguno ‘some’ (17a), then the ungrammaticality of (16) follows from the fact that an object pronoun like lo, which is a definite expression, cannot be endophorically related to this quantifier as shown in the non-elliptical context in (17b).

An anonymous reviewer argues that there can be other counterparts to ningún ‘any’, like un ‘one’, as in (18a), which unlike alguno ‘some’ can indeed by linked anaphorically to lo, as shown in (18b). In this case, the reason why (16) is not grammatical is not really straightforward.

To avoid the problem of determining the exact composition of the ellipsis site (which is in turn a consequence of the use of quantifier ningún), I present new data regarding missing antecedents where the exact make-up of the ellipsis site is not an issue. Each of the two following examples, (19) and (20), features three subexamples (a), (b) and (c): (a) shows that an indefinite NP under the scope of negation cannot be linked anaphorically to a weak pronoun. These will therefore be ungrammatical. The (b) counterpart adds a modal ellipsis context to the (a) sentences. The result is that the pronoun can indeed bind an indefinite expression, which is expected if there is unpronounced structure. Finally, the (c) examples compare ME with proform hacerlo ‘do it’, a case of deep anaphor (Saab 2008: 117–122) like English do-it anaphora: in these cases, the pronoun cannot bind an indefinite expression, which is expected because hacerlo lacks internal structure. The clear asymmetry between the (b) and the (c) examples is important, as it strengthens the validity of the argument in general.

2.3 Test 3: Extraction

Extraction constitutes a valuable argument to support or reject the PF-deletion view of ellipsis (Aelbrecht 2010; LaCara 2016; Merchant 2013b; Winkler 2013). If a gap has internal syntactic structure, we expect extraction from this structure to be possible prior to deletion at PF. Instead, if the gap is atomic, extraction is predicted to be impossible, as there is nothing from which we can potentially extract. Compare, in this respect, VP-ellipsis and do-it anaphora in English.

Extraction will be explored in depth in Section 4, as it constitutes a complex issue which merits a more in-depth approach.

2.4 Test 4: Pragmatic control

Hankamer and Sag (1976) claimed that deep anaphora, but not surface anaphora, can be used exophorically, that is, in the absence of a linguistic antecedent. Depiante (2001: 216) shows that ME can indeed be controlled pragmatically, as evidenced by the following data ((a) and (b) are Depiante’s examples, (c) is mine).

Depiante (2000, 2001 takes these data to show that ME must involve a null pronoun, given that if they involved PF-deletion of syntactic material, they would have to be linguistically controlled. Effectively, that such cases do not involve syntactic structure is further confirmed by the impossibility to extract. In this respect, take any of the examples above and compare them with the following example.

The presence of the VP-internal PP al palacio de Wilanów (‘to Wilanów Palace’) renders the example ungrammatical. Importantly, though, in the presence of a linguistic antecedent, the string in (24) is grammatical.^[3]

The interplay between pragmatic control and extraction, which to my knowledge has not been discussed in the literature, reveals a paradox, which appears to be solved if we assume that ME can be somehow associated with two different structures. I would like to suggest, however, and in line with what Pullum (2000) argues for English VP ellipsis, that the exophoric uses of ME are fossilized expressions (see Saab 2008: 67 for similar claims). To start with, most cases of exophoric ME reported in the literature involve modals in the 1st person form. So while a child may utter (26a) while holding an apple in order to ask if she can eat it, it would be way less natural, perhaps even incomprehensible, if she were asking for permission on whether her sister Ana could eat that apple.

Note also that the relation between the form and meaning in (26a) is certainly mysterious: why should it be understood as ‘can I eat the apple’ and not as ‘can I take the apple to my room’? Related to this, most of the well-known cases of exophoric use of poder involve cases in which it is impossible to replace the complement of the modal by a particular VP. In all of the cases in (27), the utterances constitute a lexicalized way of cheering someone to help them overcome a situation which may be challenging or even painful.

Exophoric ME is certainly an interesting issue that deserves some more in-depth consideration. Consequently, in the remainder of the article I will consider linguistically controlled examples that, based on their ability to allow extractions, must involve the presence of syntactic structure.

4.1 Clitic climbing and object preposing

Clitic climbing, CC hereafter, refers to the fact that a verbal clitic may move across a clausal category (Bok-Bennema 2006). Although CC is attested in many languages, the exact predicates under which it is possible are not uniform (Wurmbrand 2014). Crucially, though, modal verbs are contexts where CC can easily occur in Spanish.

A long standing observation (Hernanz and Rigau 1984; Radford 1977; Zubizarreta 1982) is that CC cannot occur in the context of modal ellipsis (37a). Depiante (2000, 2001 takes this observation to mean that CC is impossible because the gap corresponds to an atomic particle, so there is no (silent) structure out of which the clitic could have been extracted. Compare this example to (45b), where no CC has taken place.

Depiante (2000, 2001 provides a similar argument regarding object preposing (OP) in Spanish. OP (Rizzi 1986) refers to a syntactic operation whereby the direct object in an embedded clause becomes the subject of the matrix predicate in impersonal se constructions. The example in (38) (modified from Depiante 2000: 24) illustrates this phenomenon. In particular, the object DP estas casas ‘these houses’ in (38a) is promoted to a subject position in (38b). This movement operation is not a type of topicalization, it is a bona fide object-to-subject raising operation that triggers plural agreement on the matrix verb.

However, ME bleeds the application of OP (39). The reason provided by Depiante is identical to the CC case: the ungrammaticality follows from the fact that there is no syntactic material out of which esos departamentos ‘these flats’ could have been extracted.

4.2 An alternative explanation: a ban on VP-ellipsis

In this subsection, I sketch an alternative explanation to the data reported in Section 4.1, which is compatible with a PF-deletion approach to ME. It is crucial to stress that CC and OP are not unrestricted in their distribution; mainly, they occur when the matrix verb is a restructuring predicate, a category where modal verbs belong.^[7] Compare, in this respect, CC and OP with modals (40) on the one hand and the same phenomena with non-restructuring verbs – (41) and (42) respectively.

In order to account for these facts, I assume a shrinking approach to restructuring where modals are linked to two distinct configurations: one monoclausal and one biclausal (Hernanz and Rigau 1984; Manzini 1983; Rizzi 1986; Zubizarreta 1982). In general, modals are subject raising predicates selecting a TP (Wurmbrand 2001). In restructuring contexts, modals select a VP complement, which accounts for the transparency, monoclausal effects (Cardinaletti and Shlonsky 2004, among others).

The reason why ME bleeds CC and OP is because in these contexts, the modal selects for a VP, and not a clausal node. The ungrammaticality of such cases is simply a corollary of the fact that Spanish disallows ellipsis of a VP (Dagnac 2010; López 1994), contrary to English (44a) or Portuguese (44b) (from Cyrino and Matos 2002: 178).

Take the minimal pair in (37), repeated in (45).

In the ungrammatical string, (45a), the modal needs to select for a VP. This is due to the locality restrictions imposed by clitic climbing. The problem is that in order to derive the sequence, ellipsis needs to target a VP, which is not possible in the grammar of Spanish (46a). Conversely, if CC does not apply, then the modal selects for a TP, and therefore ellipsis targets a clausal node (46b).

One piece of evidence in favor of the claim that ME involves deletion of a TP is that, as we saw, ME patterns with sluicing – and not English VP ellipsis – in disallowing voice mismatches. Another piece of evidence in this respect is shown in the following examples by Dagnac (2010), who argues that the projections above the VP must be elided as well in ME, either the passive auxiliary in VoiceP or the perfect auxiliary haber ‘have’ in AspP.

All in all, the impossibility for CC and OP to occur in ME should not be taken as evidence against the presence of silent, syntactic structure, but rather as a consequence of the independent fact that Spanish disallows VP ellipsis. This alternative explanation is better considering the data reviewed in the previous sections in this article. In the next section I provide data involving A-bar dependencies in ME contexts, which is yet further support for the argument that ME involves PF-deletion.

4.3 A-bar movement

The following data show that ME allows for various kinds of A-bar extraction from within the ellipsis site, which follows naturally from the PF-deletion approach. This is illustrated with free relatives (48a), Antecedent Contained Deletion (48b) and (48c), wh-movement (48d), (48e) and (48f) and topicalization (48h) and (48i).

These data stand in clear opposition to the NCA examples presented in (29) in Section 3. The claim against A-bar extraction in ME comes from two examples provided in Depiante (2000). Before ending this section, I would like to argue that the ungrammaticality of these two examples follows from independent reasons that have nothing to do with extraction itself. The first case involves topicalization of the indirect object a Susana ‘to Susan’. The result, as reported here, is an ungrammatical sentence.

While this ungrammaticality could potentially be a corollary of the lack of a VP out of which the indirect object could have been extracted, this explanation is clearly incompatible with all the extraction facts illustrated in (48). In my view, the problem with this example has to do with the huge amount of redundant material present in the rightmost clause: the subject Juan is the same as in the antecedent clause and the modal triggering ellipsis is also the same one. If we insert a pro instead of the overt subject Juan in the rightmost clause, and we change the modal verb as well as the polarity, we obtain a perfectly grammatical structure. In this case, the contrast is established between two prospective recipients of a book (María and Susana) on the one hand, and the contrast between a negative desire (no quiere ‘he doesn’t want to’) and a positive obligation (sí debería, ‘he should’).

The asymmetry between (49) and (50) strongly suggests that once we control for the relevant contrasts, extraction is perfectly possible. The second and final example is from Saab (2008: 57), who builds on an original example from Depiante (2000: 37).

Contrary to (49), which contained too much redundancy, (51) contains too many contrasts: (i) Juan knows versus Pedro knows; (ii) what book versus what magazine, and (iii) María versus Ana. Without a proper context, such strings may be absolutely difficult to parse. However, the biggest issue here has to do with the subject Ana in the rightmost clause. As claimed in Dagnac (2010), and as reported in the previous section, ME in Spanish involves ellipsis of complement of the modal, so the subject in (51) has no place to be. Note that it cannot be placed before the modal either, because as defended in Saab, that would violate the requirement in Spanish that wh-operators must be adjacent to the finite verb. So the only possible structural site for the subject would be right at the beginning of the second conjunct, as a case of long distance topicalization. But in this case, Ana is likely to be interpreted contrastively with the matrix subject in the leftmost clause (i.e., Juan).

This restriction regarding the position of the subject Ana may also explain why Spanish disallows pseudostripping with modal verbs. Pseudogapping, which has been mostly described in English (Gengel 2013), could be thought of a hybrid between gapping and VP ellipsis: like gapping, it deletes a verb under identity with a neighboring antecedent (with special structural restrictions) leaving a gap between (at least) two arguments and/or adjuncts, but unlike gapping, it leaves finiteness overtly realized by means of an auxiliary or a modal verb (like VP-ellipsis), as shown in the next example (from Levin 1978: 229).

Note, however, that in the equivalent example in Spanish, the internal argument must precede the modal.