Clitic Left-Dislocations with Epithets in Rioplatense Spanish (CLLD+ep; 1) are transitive sentences with a clitic-left-dislocated topic DP (DP-LD), a clitic (CL) and a post-verbal epithet (DP-ep).1

All three components are identifiable as direct objects. The object CL is accusative.2 Both DPs appear with a iff they are animate (“personal a”, glossed A, an instance of Differential Object Marking, DOM).3 The three components apparently agree in phi-features and are co-referential, although this squib will challenge this generalization by presenting data with number mismatches between a plural epithet and a singular DP-LD and CL.

Suñer (2006) introduced CLLD+ep to support base-generation of CLLD. Her licensing mechanism combined Clitic Left-Dislocation (CLLD) and Clitic Doubling (CLD). Noting that the DP-ep is within the scope of the Nuclear Stress Rule (therefore, in situ), she licensed the CL/DP-ep chain in a BigDP. The clitic’s obligatoriness and the epithet’s definiteness and a-marking follow directly from independent requirements of CLD.

She also argued that DOM shows the DP-LD to be clitic-left-dislocated and not a hanging topic (connectivity, Cinque 1990). However, the DP-LD cannot arrive in the left periphery via movement, since its putative original position is occupied by the DP-ep. Hence, it is licensed via Long-Distance Agreement. Finally, since the in-situ position is bound by the DP-LD, only epithets, which can be bound by a c-commanding DP, can occur there, but not other R-expressions.

López (2009: 230) challenged this proposal. A major theoretical concern is that probes for Agree are usually heads with unvalued phi-features. Contrary to this, in CLLD+ep the DP-LD is an XP with valued phi-features. Empirically, López noted that the DP-ep is in a relation of predication with the DP-LD, not of coreference. Crucially, predication is generally assumed to be licensed in a local configuration, not via Long-Distance Agreement. Hence, he proposed instead to license both DPs locally in a clitic-doubled small clause. That way, Suñer’s example (2a) below would be licensed via the underlying small clause in (2b), followed by leftward movement of the DP-LD.

Whereas this proposal successfully accounts for the predication relation and the agreement between the DPs, López recognized that special stipulations would need to be added to account for DOM of the DP-ep, since predicative DPs in Spanish copular and small clauses appear without a.4

Even though both accounts have merits, the data I present below are not easily accommodated by either proposal. Example (3) shows a DP-LD that is a negative indefinite quantifier, morphosyntactically singular. In this case, the CL agrees in the singular with the DP-LD. However, the DP-ep is plural, creating a number agreement mismatch between the components. Moreover, as (3b) and (3c) show, it is not possible for the CL to agree in the plural with the DP-ep, or for the DP-ep to be singular, thus agreeing across the board. That is, the number mismatch is obligatory.5

This datum is problematic for both Suñer’s and López’s accounts. First, it strongly argues against the idea that CLLD+ep is an instance of Clitic Doubling, since this construction does not allow number mismatches (4), casting doubt on Suñer’s mechanism for jointly licensing the CL and DP-ep.

As for López’s proposal, number mismatches are generally impossible in copular sentences and small clauses (5).

Careful consideration of the semantics of the predication relation in CLLD+ep also argues against a small clause analysis. Note that in (3a) (repeated below for convenience) the DP-ep is not predicated of the overt (singular) DP-LD, but of the covert (plural) restriction set of the quantifier, ‘the students’ (that is, it is the students who are lucky, not ‘no student’ o ‘some student’).

Trying to license (3a) with López’s mechanism is also problematic. The possibilities in (6) match the surface DPs, but their small clause is ungrammatical and would make ningún estudiante the subject of predication.

These mismatch data force us to continue looking for a viable licensing mechanism for CLLD+ep.

Notably, number mismatches like these are also found in Intersentential Anaphora. Such mismatches are familiar from studies of the reference of they across sentences in English (7a shows the pronoun they; 7b a plural epithet). Similar examples can be constructed for Spanish (8).

The configuration of intersentential anaphoric relations allows a plural pronoun (or epithet) to refer to the plural (covert) restriction of a singular (overt) quantified antecedent. Note that in (7), not only do they/the poor things take a plural antecedent that is not itself overtly present in the preceding sentence, but also were well protected/are so sensitive are predicated of ‘professors’, that is, of the restriction of the quantified no professor. This reference to the restrictor set is often called maximal set anaphora (Nouwen 2003: 43), and is generally available in intersentential contexts, provided the restrictor set is guaranteed to be non-empty (that is, with strong determiners; see Nouwen 2003: 50). Likewise, both in (3a) and in (8) habían estudiado mucho/suertudos are predicated of ‘estudiantes’, the restriction of the quantified ningún estudiante. Therefore, the singular/plural mismatches occur similarly in Intersentential Anaphora and in CLLD+ep.

In light of these parallels, I propose that number mismatches can be explained if CLLD+ep is derived from a structure that is underlyingly biclausal, allowing for anaphoric relations where a subsequent plural epithet can refer to the plural maximal set (restriction) of the singular quantified antecedent. Biclausal structures have been assumed in analyses of dislocations (Ott 2014; Ott & de Vries 2016), ellipsis (Arregi 2010), emphatic replies (Poletto & Zanuttini 2013), and clefts (Zubizarreta 2014). The two underlying clauses involved are reformulations that support deletion of identical parts and phonological integration, thus deriving a CLLD+ep that is monoclausal on the surface.

Let us assume for simplicity the CLLD+ep in (9a). The underlying structure in (9b) is composed of a CP1 with a VP that contains the DP-LD, and a CP2 with a VP that contains the DP-ep. In (9c), Clitic-Left Dislocation applies in CP1 for information-structural reasons (this is independently needed to obtain a CLLD sentence). In (9d), Bare Argument Ellipsis (or Gapping, more generally, since more than one constituent can remain; see (16) below) deletes the redundant odio in CP2, leaving the epithet as remnant.6

In the underlying structure in (9b), CP1 and CP2 are reformulations in the sense of Ott (2016), that is, the CP2 is fully redundant with the CP1 modulo the differing DPs. In this case, ellipsis allows a CP2 remnant to be integrated prosodically in the CP1. Hence, the reformulative character of CP2 allows the DP-ep to be within the scope of the Nuclear Stress Rule.

Because my proposal depends on the underlying structures being syntactically parallel, I assume that CLLD in CP1 is obtained by movement. Evidence in support of this is the fact that a lower quantified DP can bind DP-LDs (10) (see Villa-García 2019). Another supporting fact is that a prosodic boundary is not necessarily present after the DP-LD (Labastía 2018), which would otherwise suggest base-generation.7

Let us see now how this proposal accounts for all known properties of CLLD+ep.

To summarize, a biclausal analysis accounts for the morphological shape of all the components, their surface positions, and the way the DP-ep finds a subject of predication in either the reference or the restrictor set of the DP-LD. Key components of this proposal are that CLLD is derived by movement in CP1, that both forward and backward ellipsis must be allowed to apply, and that the CP2 epithet must be allowed to integrate prosodically into the CP1. Yet, a couple of observations suggest the link between Intersentential Anaphora and CLLD+ep is not unproblematic.

First, since Intersentential Anaphora involving non-epithet DPs is acceptable (20a), we should expect non-epithet DPs to occur in CLLD+ep (20b), contrary to fact.14

One possibility worth exploring is to follow Suñer in exploiting Condition C of classical Binding Theory. For that, we need to assume that in CLLD+ep there is, in fact, some mechanism that syntactically integrates the DP-ep into the CP1, and that this full syntactic integration is responsible for this restriction. Biclausal analyses of fragment answers and sluicing (Merchant 2004), split questions, and right- and left-dislocations (Arregi 2010; Ott 2014; Ott & de Vries 2016) yield structures where the remnants of ellipsis are not fully integrated in the host clause. Contrary to this, CLLD+ep is monoclausal on the surface. Allowing full syntactic integration of the DP-ep would make it subject to Condition C, thus preventing R-expressions from occurring. This would require making the deletion site in CP1 available for this kind of “extreme” merging. For example, this could be a kind of external remerge à la de Vries (2012). But, prima facie, it looks like this solution risks overgenerating in a different way, since now all traces of movement could in principle be vacated to allow remergings.

Another possibility is that in (20) there is no real predication relation between the DP-LD a Méndez ‘Méndez’ and the DP-ep al peor presidente que tuvimos ‘the worst president we’ve had’. The latter being an R-expression, it is freer to find a referent than are anaphoric expressions like pronouns or epithets. It could be that there is a strong pragmatic preference to interpret R-expressions as referentially disjoint, and that this interferes with a hearer’s ability to appropriately parse (20b) to avoid a valency violation. This would mean that (20b) is indeed grammatical, but extremely dispreferred for audience-design/parsing reasons. Experimental evidence could presumably be brought to bear on this issue.

Another issue is raised by the behavior of cada ‘each’.16 It allows expected mismatches intersententially (21), but apparently not in CLLD+ep (22).

First, note that a singular epithet is disallowed, which is consistent with the ungrammaticality of examples involving ningún + singular epithet, and consistent with the Intersentential Anaphora patterns.

Now, I think the reason that (22) also sounds unacceptable is because of the complicated expectations set up by the semantics of cada. It is known that, unlike todo(s) ‘every’, cada ‘each’ seems to require distributivity, and a focus on individuals (e.g. Vendler 1967; Tunstall 1998).

Cada, then, requires an interpretation at the level of events predicated of single individuals. Example (22) is odd because one does not expect a predication about the whole set if you use cada. Something is needed that reflects the distributivity of cada. Note that if one adds the postverbal subject sus padres, which “satisfies” the distributivity of cada, in my opinion acceptability improves:

Therefore, I believe the issue in (22) is how to satisfy the expectations that the use of cada sets up. This is not a problem in Intersentential Anaphora because the first sentence’s interpretation is unproblematically distributive, and the quantifier cada still introduces its restriction set as a possible antecedent for future reference.

To resolve the issues raised in this squib satisfactorily, it is crucial to determine experimentally, in future work, the range of grammatical quantifier/DP-ep combinations and their degree of acceptability. This will help narrow down the possible analyses of Romance Clitic Left-Dislocations. Moreover, if the analysis presented here is correct, it can have important consequences for recent proposals about “unconventional” merging operations, and more generally for the infrastructure of the syntax-semantics interface.