Deconstructing SE-Constructions: Number Agreement and Postsyntactic Variation

Traditionally, nonparadigmatic SE in Spanish has been analyzed as belonging to two constructions, referred to as passive (1a), where the verb agrees with its complement as in analytic passives (P-SECs), and impersonal (1b), where the verb shows default agreement (I-SECs).¹

Distributionally, I-SECs (1b) are used in all contexts except in transitive configurations in which the object does not receive differential object marking (DOM) (1a). Theoretical approaches overwhelmingly assume this state of affairs and complete (1) with the opposite pattern in (2), generally considered ungrammatical or dialectal in the literature.

This divergence is derived by assuming the existence of two constructions with their own structural and case assignment properties (3) (a few authors analyze the divergence differently, most explicitly Oca 1914 and Otero 1972, 1973; also see references in ^{footnote 1}).

However, the paradigm in (1)–(2) simplifies the facts (MacKenzie 2013). Our first goal, in section 2, is to show that variation on agreement in SECs is far more extensive than assumed in previous analyses, which makes most theoretical approaches to SECs empirically unsound. Furthermore, we uncover a structural distinction that has gone unnoticed in the literature so far:² while number agreement with postverbal objects exhibits a high and random degree of variation internal to every dialect, variation with preverbal and null arguments is systematic and mostly dialectically determined. Agreement with preverbal and null arguments splits dialects between A(greement)-varieties, where the argument covaries with number agreement (4b), and C(litic)-varieties, where both object clitic (4a) and subject number agreement (4b) are possible outputs (see section 2.2 for details).

Our proposal is that (4a) and (4b) are two instances of the same clitic left-dislocation (CLLD) structure and that in (4b) postsyntactic morphological operations reshape the realization of the object clitic as number agreement. As we show in section 4.2, our analysis is supported by the behavior of SECs in clitic climbing and in idioms. We further argue that the change of the clitic into number agreement with preverbal and null objects (4) (Clitic Mutation) and the phenomenon of number agreement with postverbal objects (1)–(2) (Number Harmony) are two independent postsyntactic processes, neither of which constitutes a genuine subject agreement relation.

Our second goal is to present a unified analysis of SECs as regular sentences that do not require any construction-specific provision: properties attributed to SECs follow from the lexical features of se, an argumental pronoun in subject position, together with the application of postsyntactic processes at the interface levels. These processes are triggered when the subject is not specified for number, including se.

An advantage of our proposal is that it does not attribute to these structures emerging properties such as the removal of accusative case required to make the agreement pattern distinctions between [+DOM] and [–DOM] objects in Spanish. Furthermore, getting rid of the distinction between P-SEC (1a), I-SEC (1b), and clitic SEC (4a) is consistent with the absence of meaning differences between them (Mendikoetxea 1999) despite the fact that, with varying distribution, the three possibilities exist in all dialects. Evidence in section 2 shows that the patterns that allegedly support a passive approach are deprived of empirical motivation. An immediate consequence is that agreement cannot be used in these structures as a reliable test to uncover underlying syntactic relations. This raises a general issue about the meaning of overt evidence, and questions the foundations of approaches that require construction-specific P-SECs. In contrast, a purely derivational approach is perfectly equipped to explain the complete absence of semantic differences and provides a syntax able to relate to the whole range of PF representations.

The structure of the article is as follows. In section 2, we describe number agreement in SECs and show that it does not behave like a genuine subject agreement relation. In section 3, we summarize the logic and arguments in Ormazabal and Romero (O&R) 2019a to analyze SECs as regular active sentences where V’s complement is also its syntactic object and se is a pronominal subject. We detail the ϕ-features specified in the lexical entry of se, which agree with T and ultimately determine the overt agreement patterns. In section 4, we present the syntactic derivation common to all SECs. In a nutshell, se merges as the highest argument and behaves like a subject pronoun, maintaining ϕ-agreement and nominative-case-checking relations with T. We also briefly discuss our proposal regarding Agree as applied to SECs. In section 5, we derive the asymmetry between preverbal or null arguments and postverbal ones. While preverbal or null arguments are CLLD structures in which the direct object clitics lo(s)/la(s) ‘him/her/it/them’ in most cases mutate into number agreement in the morphology, for postverbal complements agreement is a postsyntactic, extragrammatical phenomenon not subject to syntactic conditions; instead, it is amenable to a sociolinguistic approach (Otero 1973) whose domain of application is tightly circumscribed by the syntactic derivation. In section 6, we close with some general conclusions on crosslinguistic variation and on the properties of agreement in syntax and natural language processing.

The regularity of number agreement in SECs has been grossly overestimated, possibly for reasons of normative pressure (MacKenzie 2013). In this section, we review all the attested agreement patterns in transitive SE structures and we conclude that number agreement is not a reliable criterion for distinguishing between P-SECs and I-SECs. In addition, we describe a new asymmetry between postverbal objects (section 2.1) and preverbal/null ones (section 2.2). Finally, in section 2.3 we briefly analyze agreement patterns with 1st and 2nd person.

For each agreement type discussed in this section, we have conducted an analysis of the data in CORPES and undertaken Google searches that corroborate our point.³ Given the nature of the database, our samples are mostly from written sources, which are more prone than speech to be conservative and match normative standards. We expect spoken language to depart from the norm to a larger extent, further supporting our conclusions, but the variation observed in the written samples is rich enough for our purposes.

According to traditional descriptions, only 3rd person objects not marked for DOM (1)–(2) trigger number agreement. Here, we show that this description is not accurate and that number agreement with postverbal elements cannot be accounted for in syntactic terms (see Gallego 2016, Planells 2017, and Pujalte 2018 for observations in the same direction).

Apart from some scattered observations pointing out cases where agreement between the verb and [–DOM] objects fails in SECs (Lemus 2014, Ordóñez and Treviño 2011, 2016, Pujalte 2018), the only systematic description of the range of variation is DeMello 1995. Our analysis of CORPES suggests that central and northern South American dialects show a stronger tendency than southern American and peninsular Spanish to exhibit a default 3rd person singular form, with no subject agreement whatsoever (5)–(6).

Lemus (2014), from El Salvador, observes that in his own dialect both options, with and without agreement, are acceptable and are used interchangeably, underlining the complete absence of semantic effects in this alternation. It is also well-known that, independently of dialect, bare plural objects quite commonly do not trigger agreement (Sánchez López 2002 and references).⁴ But this agreement alternation is much more general and affects all kinds of objects (see the online appendix). To understand its distribution, in addition to dialectal trends, we must take into account a more general sociolinguistic condition (DeMello 1995): when the pressure exerted by formal education diminishes, lack of agreement increases.⁵

Following analyses for Italian (Cinque 1988, D’Alessandro 2007), two LI reviewers suggest that these facts might correspond to an alternation between a P-SEC and an I-SEC (see Ordóñez and Treviño 2016). However, this analysis is inconsistent with A-varieties. First, according to standard analyses, in I-SECs accusative case is not suppressed; instead, it is assigned to the object, as the presence of object clitics clearly indicates.

If nonagreeing forms were impersonal, we would expect parallel examples with clitics to be fully available. That prediction is not borne out. Examples like (8b) with an object clitic are not acceptable in general. Thus, contrasting with postverbal arguments (8a), preverbal ones always show number agreement (8c).⁶

Furthermore, if P-SECs were passives in that DOM is not assigned to the object, animate arguments would agree with the verb. However, that option is not attested, either with 3rd person animate arguments or with 1st/2nd person ones.

The only way to allow case assignment in (8a) and to block P-SEC in (9) is to resort to some mechanism where DOM objects are excluded a priori. We cannot see how that could be captured derivationally in a principled way, and in fact we know of no attempt in the literature to explain this distribution of facts. In that sense, traditional analyses, including generative ones, are constructionist in nature.

To compound matters, finding a coherent proposal for the cases discussed so far would only cover one corner of the problem since, as we show in section 2.1.2, the rule also fails to apply on the opposite side of the equation. In many cases, impossible syntactic agreement holds between the verb and any noun phrase in the sentence (long-distance agreement (section 2.1.3), complements of prepositions, and even nominal adjuncts (section 2.1.4)).

According to traditional descriptions, [+DOM] objects do not trigger agreement. For 3rd person objects, the description does not accord with the data in these contexts either, and the sentence marked as ungrammatical in (2b), repeated here, is much more common than usually assumed (see DeMello 1995, Planells 2017).

As a matter of fact, examples in written texts are readily available in every Spanish dialect, and much more frequently in oral Spanish.

MacKenzie (2013) studied frequencies of singular/plural in ‘se+V+a+plural object’ sequences in the CREA corpus for elegir ‘choose’, invitar ‘invite’, nombrar ‘name’, and matar ‘kill’. Agreement results ranged from 9.65% with invitar to 38.89% with elegir.

As these observations show, the distribution of agreement with [±DOM] objects hardly reflects anything more than a choice prescribed by the Academia through the school system. The weaker the influence of the Academia, the greater the asystematicity of the agreement patterns in SECs: agreement between V and DOM objects permeates the Spanish language, and it is only consistently rejected by speakers who show a highly prescriptive behavior.

A context where this dysfunctional agreement shows up frequently is long-distance agreement relations, where the complement of an infinitival clause agrees with the matrix V (see, e.g., Boeckx 2004, Bhatt 2005, Etxepare 2006).

CORPES yields many such results for all geographical areas, and a Google search for se requieren hacer ‘it is required to make’ (5 June 2018) yielded 17,400 hits. Among the examples we observe long-distance agreement not only with [−DOM] objects (13a–b) (Gallego 2019) but also with verbs such as as nombrar ‘to appoint’ and with [+DOM] verbs (13c).

Note that these structures disallow analytic passives (14a) and clitic climbing (14b).

Therefore, agreement between the matrix verb and the complement of the embedded infinitive is completely unexpected in this syntactic context as well. Moreover, irregular agreement is found even in hyperraising configurations.

In (15), the matrix verb agrees with the subject of the finite subordinate clause. The number of Google hits for this structure rises to 8,260,000 (6 June 2021), including cases of number agreement with embedded 1st and 2nd person subjects (see the online appendix).

Agreement in SECs is even messier than described so far. The verb may agree with complements of lexical prepositions (Gallego 2016, examples from peninsular Spanish).

Although marginal compared with previous phenomena, examples like (17a–b) are not isolated cases: a Google search for se hablaron de temas ‘issues were talked about’ (8 March 2018) produced 6,350 instances of plural agreement between hablar and P’s complement temas (18), many of them from formal sources.

This kind of agreement is found even in long-distance relations: for se deben recurrir ‘it must be resorted to’ Google returned 26,900 hits (5 June 2018).⁷

Finally, it is surprisingly common to find examples where agreement is not triggered by arguments, but in fact by temporal DP-modifiers.

For instance, a Google search for se abre los domingos ‘it is open on Sundays’ (5 June 2018) returned 27,400 hits for the irregular plural agreement pattern se abren los domingos and 27,300 for the expected singular one, se abre los domingos. Although these data do not fit with standard P-SECs (the verbs are not transitive), their morphological behavior mimics it. In section 2.1.5, we argue that these agreement patterns cannot be understood as subject agreement in any syntactically coherent way.

If we consider the whole range of data, subject agreement behavior is completely unexpected. Numbers are big enough to dismiss them as performance errors. True subject agreement is much more regular and predictable: it neither appears with unexpected elements nor suddenly disappears with agreeing ones. Thus, in contrast to (21), temporal nominal adjuncts never trigger agreement in other contexts, not even with impersonal verbs.

And true subject agreement is never affected, not even in most favorable contexts. For instance, while bare plurals easily fail to agree in SECs (Mendikoetxea 1999, Sánchez López 2002), bare plural subjects of un accusative sentences can never avoid agreement.

Moreover, as in SECs, in inchoative sentences (a) the verb ends up agreeing with its complement, and (b) there is a se intransitivizing the verb. Despite the similarity between the two structures, agreement failure is unattested in inchoative sentences.

Finally, in contrast to data presented in section 2.1.1, in analytic passive sentences lack of agreement between the verb and its complement is clearly ungrammatical.

The conclusion is clear: agreement facts in SECs do not work as predicted by theories based on syntactic agreement. In order to explain their agreement patterns, we cannot simply treat them as subjects, because this is not the way subject agreement behaves. As shown in (22), we cannot resort to a default syntactic agreement either, because, in clear contrast to subject agreement behavior in SECs, default agreement is systematic (D’Alessandro 2007, López 2020). Furthermore, in the next section we show that there is a clear asymmetry in agreement behavior between postverbal and preverbal arguments, an asymmetry that, again, is not present in regular subject agreement relations.

When the internal argument is null or appears preverbally,⁸ it covaries with an object clitic in some cases and with a verbal plural agreement in others. Both structures in (26) exist in all dialects of Spanish, but their distribution diverges, yielding what we have called A(greement)- and C(litic)-varieties.

Preverbal [+DOM] arguments, including 1st and 2nd person ones, exhibit a completely uniform pattern cross-dialectally: they are always doubled by a pronominal clitic (27a) and never trigger subject agreement (27b) (cf. (10)).⁹

Verbal agreement is strictly circumscribed to preverbal/null [−DOM] arguments in all varieties; no significant exceptions are found either in corpora or in oral elicitations in which any other preverbal element might trigger number agreement. (We will come back to the special case of relative clauses and Ā-dependencies in section 5.2.)

In C-varieties, the clitic strategy extends to [−DOM] objects, so that the agreement option illustrated in (28) coexists with the clitic option in (29).

With preverbal/null [−DOM] objects, agreement is mandatory (28)–(29). There are no relevant cases lacking either agreement or an object clitic (cf. (8)). Variation is strictly established on the agreement/clitic strategy distinction; it does not affect number.

The clitic strategy is reported in the literature (Santiago 1975, Sánchez López 2002, and references), but it has not been exhaustively described, and standard theories on SECs tend to ignore it. The clitic strategy coexists with the agreement strategy in the southern half of South America (NGLE, García Negroni 2002). Both are commonly found in CORPES examples from that area, and all our (linguist) consultants from the Buenos Aires area agree that (a) they are equally available and (b) there is no semantic difference between them.¹⁰

In clear contrast, the clitic is impossible with [−DOM] postverbal objects in both varieties.

In the rest of the dialects, object clitics with preposed and null [−DOM] objects are sometimes attested, but in general terms subject number agreement (28) is the sole strategy for all speakers in A-varieties,¹² including dialects where postverbal arguments tend not to agree.¹³ The result is that SECs may lack agreement with postverbal arguments but they obligatorily agree with preverbal ones.

Standard descriptions fall short regarding the significance of the clitic structure. Insofar as our description is right, approaches based on constructions are forced to postulate three different constructions—passive, impersonal, and clitic SECs—that show no meaning or use differences and, quite tellingly, extend to all dialects in one way or another.

To complete the picture, 1st and 2nd person objects show no dialectal or idiolectal variation whatsoever. They never allow number or person subject agreement, and they always manifest themselves through an obligatory object clitic, as in any transitive clause.

The behavior of number agreement in SECs is exceptional with respect to Spanish subject agreement in several respects:

• 1.

  With preverbal/null arguments, subject agreement alternates with object clitics in some dialects, systematically occurs with [−DOM] objects only, and never fails.

• 2.

  With 1st and 2nd person as well as with preverbal 3rd person [+DOM] objects, no agreement is possible; they are uniformly represented as object clitics.

• 3.

  With postverbal elements, agreement is erratic. It frequently fails with [−DOM] objects and unexpectedly occurs with other DPs: [+DOM] direct objects, long-distance DPs, or complements of Ps and adjuncts. This distribution sharply contrasts with that of analytic passives, where subject agreement is robust and systematic.

In what follows, we present an integral analysis of SECs that derives distributions 1 and 2. We also argue that although distribution 3 should be analyzed by extragrammatical mechanisms—thus, it is not predictable in morphosyntactic terms—the properties of the derivation condition the domain in which the unsystematic agreement alternations may appear. With that goal in mind, in section 3 we summarize the gist of the argumentation in O&R 2019a supporting a unified analysis of I-SECs and P-SECs, and we present the lexical properties of se that make it possible. In sections 4 and 5, we detail the derivation and discuss the postsyntactic operations responsible for “subject” number agreement with the object, and the distribution of facts presented so far.

The primary focus of most analyses of SECs has been to justify the status of P-SECs as “selective passives” where accusative case is removed, but only for [−DOM] complements.¹⁴ These analyses not only require ad hoc mechanisms, but also are based on false empirical claims. First, [−DOM] objects may be realized as object clitics, which dismantles the idea that they may not receive accusative case.¹⁵ Second, preverbal and postverbal arguments show radically different behavior, which questions their subjecthood. More generally, their behavior in SECs invalidates number agreement as evidence for nominative case or subjecthood. This alone indicates that the passive approach is not on the right track, but there are independent positive arguments that confirm the same conclusion.

The proposal in O&R 2019a,¹⁶ developing a suggestion by Oca (1914),¹⁷ is that P-SECs are just active transitive sentences, where the internal argument of V is a regular direct object and the subject is the clitic se. The peculiarities of these constructions—especially those regarding agreement—are due to the lexical properties of se, with no difference between passive and impersonal SECs.

A. P-SECs are active sentences morphologically and syntactically (Raposo and Uriagereka 1996, Rivero 2001). First, P-SECs may be formed with all transitive predicates allowing [−DOM] complements (with only the restrictions imposed in point C). That includes, crucially, predicates completely resistant to analytic passivization, where it is not even clear that the alleged accusative case existed in the first place (e.g., measure complements; Mendikoetxea 1999). Second, they do not accept by-phrases (Pujalte 2013, Saab 2014, NGLE 41.6.1, and references). Third, they do not “passivize” [+DOM] objects.

B. Consequently, contrary to what number agreement might suggest, the internal argument is the syntactic object. That is shown, among other things, by the fact that it can be modified by pseudorelatives (Aldama 2016, O&R 2019a) and that, unlike subjects of analytic passives, it may not be controlled from outside (MacKenzie 2013). Furthermore, idiomatic readings of fixed object idioms are not lost in P-SECs, showing that grammatical relations have not been altered (O&R 2019a; also see section 5.1).

C. The subject of the sentence is the clitic se (Oca 1914). As such, se shows properties typical of overt pronominal subjects in Spanish, including its animacy interpretation: the structure is only possible when the subject is argumental, and it is disallowed by verbs that reject animate subjects (Mendikoetxea 1999). Se receives nominative case; it is therefore incompatible with infinitives that do not license nominative subjects, but allowed in tensed clauses in general as well as in infinitival constructions licensing overt subjects (Cinque 1988, Mendikoetxea 1999). It also raises in raising-to-subject contexts (see Mendikoetxea 1999, Martins and Nunes 2016, O&R 2019a:sec. 3.2 for extensive discussion).

D. Se is syntactically active. It may host secondary predication (Demonte 1986, Rivero 2001, Martins and Nunes 2016, O&R 2019a, and references). (See Pujalte and Saab 2012 for different judgments; also see Collins 2017, 2021 for a related discussion regarding short passives in English and, more generally, implicit arguments.) SECs also show the same obligatory control/disjoint reference distribution in infinitive/subjunctive alternations as any other active sentence, no matter whether se is interpreted as a generic or an indefinite, and it may bind anaphoric elements vague enough not to conflict in features with it, as we argue next.

The idea that se is the missing argument has already been proposed for I-SECs (Cinque 1988, Mendikoetxea 1999, D’Alessandro 2007, and references). The reasons to discard this idea for P-SECs are theory-internal pre-Minimalist arguments that no longer hold (see O&R 2019a: sec. 3 for discussion). In sum, a derivation of I-SECs and P-SECs where se itself is the sentential subject and receives nominative case from T straightforwardly derives all the structural properties of nonparadigmatic SECs listed above in a natural and unified way. In the next section, we present the details of se’s lexical entry. Extending MacDonald 2017, we consider se’s binding properties as particularly indicative of its syntactic status, giving a clear indication about its formal feature specifications.

Saab (2014) observes that unlike other impersonal elements such as uno ‘one’ or la gente ‘people’, the subject of SECs does not allow pronominal binding (35); from this, he concludes that there is no syntactic subject.¹⁸

However, the inability to bind the pronoun may also reflect a feature mismatch between su and its antecedent se. In Romance, determiners may act as bound variables, but unlike the 3rd person pronoun su, this variable gets its person from the binder.¹⁹ As observed by MacDonald (2017: 364), unlike in (35) se can bind the less specified determiner variable in (36).

If the divergence arises from the feature specifications of su and the determiner,²⁰ two conclusions follow. First, se is a syntactically active subject and, as such, has the ability to bind a c-commanded anaphoric element; in that sense, nonparadigmatic se clearly contrasts with other se-clitics (Alcaraz 2021). Second, given the contrast between minimally specified determiners and fully specified 3rd person pronouns, we must conclude that se is not 3rd person. This conclusion is supported by the fact that SEC interpretation is not confined to generic or existential 3rd person, but se may refer to any animate argument, including 1st (37a) and 2nd (37b) person (see, e.g., Cinque 1988, Mendikoetxea 1999, Menuzzi 1999, D’Alessandro and Alexiadou 2003, O&R 2019a).

The specific interpretation se adopts is mostly determined on pragmatic grounds. Notice, also, that 1st and 2nd person subjects in the next clause may corefer with the subject of the impersonal construction in these contexts.²¹

Since se is always animate, following Richards (2014) we assume that animacy in pronouns is encoded as [person]. We thus interpret se’s ability to bind anaphoric determiners but not 3rd person pronouns as indicating that the person feature is underspecified, in the sense that it lacks a specific 1st, 2nd, or 3rd value. Let us see how these results may be put together.

ϕ-features, as well as their internal organization, play an important role at different grammatical levels and their interfaces (also see Ackema and Neeleman 2018 and references for extensive, though slightly different, discussion and implementation):

• •

  At the computational level, interpretable features determine the behavior of lexical items and the corresponding feature values of the functional projections they agree with.

• •

  At the semantic level, the feature structure of these syntactic objects—in the cases we are discussing, the pronouns—must be interpreted.

• •

  The morphology component operates on the features of the pronoun as well as on the noninterpretable features valued in the agreeing functional heads.

Focusing on its person features, semantically se is always animate (i.e., [person]), but truly devoid of any person value and interpreted contextually. It must therefore be distinguished from 3rd person animate pronouns (ella/él ‘she/he’) and from no-person ones, such as sentential subjects or object clitics lo(s)/la(s) ‘he/she/it/them’, as we discuss below (also see Trommer 2008). Most feature hierarchies stemming from Harley and Ritter’s (2002) are morphology-oriented and do not leave enough room for such fine-grained distinctions in the syntactic derivation, far less in the semantic component. In particular, both their hierarchy in (39) and their conception of underspecification are too narrow (see, e.g., D’Alessandro and Alexiadou 2003, McGinnis 2005, Bianchi 2006, D’Alessandro 2007).

Assigning a 3rd person value by default to underspecified [person], as in Harley and Ritter’s morphological hierarchy—or Béjar’s (2003) syntactic version—makes the wrong interpretive predictions even if an overruling mechanism is allowed. In order to capture the observed distinction, we need a more articulated [person] geometry that incorporates Harley and Ritter’s unmarked values (see, e.g., McGinnis 2005; also see D’Alessandro 2007:26ff. for a similar discussion and a different solution).²²

The resulting feature specifications of the relevant forms are distinguished in (41); we also include a 1st person pronoun for illustration.

To complete the picture, there is independent evidence and general consensus that se lacks number feature specifications (Mendikoetxea 2012, Harris and Halle 2005, and references).²³ In sum, its lexical entry contains the feature information in (42).

With these ingredients, in the next section we present a formal analysis of SECs as regular sentences that derives the agreement distribution in section 2 as well as the structural properties summarized in section 3.1 and developed in detail in O&R 2019a.

From the point of view of the syntactic derivation, this article’s epigraph, from Raposo and Uriagereka 1996, may be extended to all nonparadigmatic SECs.

That is, no selective case absorption or argument suppression operation is involved in P-SECs, and both constructions, P-SECs and I-SECs, follow the same derivation. Se is introduced by external Merge in the argument position determined by the predicate: the external argument position for transitive active sentences (44) (= (1)) and unergatives, and the internal argument position for un accusatives and passive sentences. We illustrate the derivation with transitive structures because they cover all the relevant structural properties under discussion.

The derivational difference between [+DOM] and [−DOM] objects is equally observed in any transitive structure; it is therefore independent of the properties of se. Following proposals in Torrego 1998, López 2012, and O&R 2013a,b, 2016, 2019b, we assume that [+DOM] and [−DOM] objects are structurally different and that object agreement is triggered only by [+DOM] objects, the ones encoding person (Richards 2014). [−DOM] objects (44a) lack [person] and do not enter into an agreement relation with v.²⁴ In conclusion, the DP in (46a)—(a) los oponentes políticos ‘the political opponents’—has the following ϕ-feature representation:

Simplifying things a bit, assume that Agree is realized in Spec,vP and that v enters the derivation as [uϕ] and is valued via Agree, which copies the feature specifications of its controller, the DOM object (see section 4.3 for details, discussion, and references). The resulting structure after the application of Agree between the probe v and the animate argument los oponentes is shown in (47).²⁵

In (47), accusative case has been assigned to the object in both constructions. From that point on, the derivation is the same for (44a) and (44b). T is merged and triggers agreement with se, as with any other subject. T enters the derivation specified as [uϕ] and copies the feature values of the closest argument, se (see section 4.3). Finally, se moves to Spec,TP, where it checks the EPP feature in T and receives nominative case.

This part of the derivation is the same independently of whether se is generated in the external or internal argument position, as in passives.

In sum, no ad hoc mechanism is postulated. The derivation is restricted by exactly the same general conditions constraining any other syntactic derivation where the subject has a more articulated ϕ-feature structure, and by nothing else. Its characteristic features are determined by those of the lexical elements involved, as in any other case. The only remaining issue is the agreement pattern, to which we will return in section 5.

Consider the derivation of sentences containing a preverbal/null argument, as in (50a–b). Remember that, as discussed in section 2.2, this structure is subject to dialectal variation.

Raposo and Uriagereka (1996) and Martins and Nunes (2016) argue that preverbal arguments in sentences like (50a) are not in subject position, nor associated to it, but left-dislocated.²⁶ according to our approach, the structure of (51b) is the structure found in CLLD in Romance languages (51a). We assume that the object DP is base-generated in a nonargument position (Van Riemsdijk 1997)²⁷ and that D-clitics lo(s)/la(s) are merged in the internal argument position, from where they cliticize to the verb (see, e.g., Roca 1996, Uriagereka 1997, O&R 2013a). The derivation is represented in (52).

As we show in section 5.1, there are strong reasons to assume that (52) is the structural description for both (50b), with an overt clitic, and (50a), where the clitic shows up as number agreement. The derivation extends to nonovert arguments, where a null topic stands for the dislocated element, and to animate and 1st/2nd person objects. In this case, however, there is no variation: all dialects and speakers follow the object clitic strategy.

We will come back to the ungrammaticality of (53c) in section 5.1. At this point, it is important to stress that the main difference between (50) and (53) is that 1st/2nd person pronouns are fully specified for person and number, while the clitic lo(s) has no [person] (Roca 1996, O&R 2013a,b, Alcaraz 2021; in (54), we abstract away from issues concerning the feature geometry of number and gender).

From this point on, the derivation continues as in (48), where the subject se agrees with T.

We have concluded that se is a subject pronoun specified for person and lacking number features. This is not the only context where the subject is not fully specified for agreement features (see section 5.2), and any approach to T-agreement must account for them. The Agree mechanism in Chomsky 2000, 2001 is subject to a completeness condition on Match, under feature identity; however, contexts where the goal is not ϕ-complete and the derivation converges show that it cannot be a general condition for agreement. As an illustration, consider SECs: even if our analysis of P-SECs (55b) (= (1a)) were not on the right track, I-SECs (55a) still constitute one case among a myriad of cases where the subject—be it se, a null pronoun, and so on—could never qualify to completely satisfy subject agreement ϕ-features.

The mere existence of default verbal agreement in all impersonal contexts, where no possible number agreement checker is available, may be taken as evidence that number agreement does not need to be valued. The same extends to [person]; see, for example, Béjar’s (2008:148) analysis of Subject [3rd person]-Object [3rd person] combinations in Mordvinian. By the same token, given that (55a) is grammatical, a second-cycle analysis for (55b) would be stipulative; in fact, as shown in section 2, if there were a second cycle (e.g., Anagnostopoulou 2003, 2005, Rezac 2003, Béjar and Rezac 2009), we would expect number features that remain unvalued in the first cycle to be available for Agree to be established with the object, showing a consistent pattern, contrary to fact. If, as we have argued, number agreement in transitive contexts is not syntactic, not only can second-cycle agreement not be imposed over the derivation—in fact, it must be prevented.²⁸

As Baker (2008) observes, Match lacks explanatory power. The idea of a set of prearranged, arbitrarily defined feature slots to be valued by the goal might be appropriate to feed the morphological component. However, in the syntactic component it looks like a conspiratorial prerequisite that fails to capture the systematic properties of agreement in natural language. It is, in some sense, a very short-sighted “lookahead” from the lexicon into the computation.

Notice, in addition, that when combined with a hierarchy where features do not have intrinsic values but depend on their place in the geometry, Match, as standardly established, is hard to codify from a strictly formal point of view. As Béjar (2003, 2008) observes, each node entails the presence of all the superordinate nodes, so that when a pronoun is specified for [1st person], that specification entails that all the nodes up to the root of the hierarchy are also specified. Given those entailment relations and assuming that the hierarchy of features in T is identical to that of pronouns (the null hypothesis), a [uperson] probe would be general enough to target all goals that have person features. Following our proposal in section 3, that would include se [person], as well as [3rd person] and all [participant] goals. Generalizing even further, the simplest system would be one where the probe in T is specified as [uϕ] (see Deal 2021 for a related proposal).

The Agree relation copies the entire hierarchical structure of the goal into the probe.²⁹ Since T ϕ-features are uninterpretable, the idea that Agree copies whatever the goal brings to the picture is both suitable for its morphological analysis and in compliance with standard views on agreement in natural languages. The resulting complex head will be different depending on the set of features the subject contributes to the agreement relation, as desired.

Under these natural assumptions, agreement in SECs is regular by all criteria. Consequently, T’s [uϕ] is completely valued in the syntax. However, the number slot in the resulting matrix of features is empty, since it is absent in the feature matrix of se. In other words, the agreement problem is not syntactic, but morphological.

Given the discussion so far, the agreement asymmetry between postverbal and preverbal positions corresponds to the difference between in-situ and CLLD objects in regular transitive clauses. As we discuss next, the two resulting structures will be exposed to different fates after syntax.

We have shown that agreement behaves differently depending on the position of the NP. With preverbal topics, its morphological manifestation is obligatory, either as an object clitic or as subject agreement. In contrast, with postverbal NPs the only morphological device used is subject number agreement, but in a semirandom fashion. As a consequence, we conclude that each type of agreement follows a different path. For preverbal/null topics, the A-position is occupied by a clitic. In some dialects, it mutates postsyntactically into number subject agreement (section 5.1). For postverbal NPs, the verb may show up marked with a default singular number or it may harmonize postsyntactically with a plural NP (section 5.2).

The pronoun se is the only subject clitic in Spanish. It cliticizes onto T the same way French il ‘he’ and impersonal on do. Similarly, the determiner head lo(s)/la(s) ‘him/her/it/them’ in object position cliticizes onto the verbal complex. The result of the computational component’s inner workings is a morphological word that includes the two clitics, the verbal root, and all the T features.

The presence of the object clitic in C-varieties is the straightforward manifestation of the structure in (58), while in A-varieties it is spelled out as subject number agreement. Subject agreement is therefore an alternative morphological expression of the same structural description expressed in C-varieties by the clitic. We term this particular morphological process Clitic Mutation (ClMut), an instance of a family of agreement effects including clitic cluster phenomena, agreement displacement, and eccentric or omnivorous agreement (e.g., Bonet 1991, Hale 2001, Rezac 2008). In other words, there is no P-SEC, but a regular active structure whose object clitic is morphologically camouflaged as subject agreement.

In what follows, we present direct evidence supporting our analysis (section 5.1.1), and we detail the conditions under which ClMut applies in the morphology (section 5.1.2). Finally, we provide independent crosslinguistic evidence for the proposal (section 5.1.3).

There are three different contexts that constitute strong evidence for ClMut. First, consider idioms containing a nonreferential clitic, such as liarla parda ‘to make a complete mess’, matarlas callando ‘to go about things slyly’, pasarlas canutas ‘to have a rough time’, and verlas venir ‘to see them coming’. The clitic is part of the idiom and must be present under all circumstances (59) (García-Page 2010 and references). As expected, passivization is completely impossible (60).

Nonparadigmatic SECs are the only exception where the clitic disappears. In those cases, which sound completely natural in A-varieties,³⁰ if the clitic of the idiom is singular, the verbal form shows up in the singular (61a), and if it is plural, the verbal form is obligatorily plural (61b).

Next, consider clitic climbing. In transitive structures (62), including SECs (63), all dialects show enclisis to the embedded infinitive ((62a), (63a)). The clitic may also climb to the finite verb; there it appears as a proclitic (62b). However, in SECs there is a split: in C-varieties the clitic is preserved (63b), while in A-varieties it shows up as subject agreement (63c).

The parallelism is maintained across the board. In long-distance clitic climbing, A-varieties retain the clitic in all the intermediate positions (64a–b) and mutate it only when it reaches the se-clause (64c), exactly as ClMut predicts.³¹

Finally, ClMut is also subject to clitic-clustering effects. When there is more than one clitic, the entire cluster must climb together ((65a–b) vs. (65c–d)). As expected under ClMut, SECs’ subject agreement behaves in this respect like a clitic (66). If it were true subject agreement, the ungrammaticality of (66c) would be completely unexpected, since subject agreement does not interfere with clitic clusters in other contexts (65a–b).

We know of no previous discussion of these facts in the literature, but we see no possible explanation for them in traditional terms. In contrast, this is precisely what ClMut predicts, since subject number agreement is structurally an object clitic, and it only surfaces as number agreement after morphological manipulation.

In this section, we make explicit the properties of the syntax-morphology interface. We do not pretend to lay out a morphological analysis in such an intricate area as clitic-agreement interactions; instead, we thoroughly describe the conditions that syntax imposes on ClMut. According to our analysis, the syntax supplies the morphology with a T head valued for [person], but not for [number]. Since the exponent of the T head in Spanish encodes number, the morphology needs to assign it a value. A priori, there are two options: (a) the value may be assigned by default, or (b) some element is blended into the morphological word to provide a value for it. C-varieties generally have both options available. In (67a) object features are realized as a clitic and subject agreement is set by default, while in (67b) object clitic features blend with subject agreement. In A-varieties, only ClMut is available (67b).³²

Following Distributed Morphology postulates, for instance, we can assume that an impoverishment rule is applied. This rule removes any conflicting feature—in this case, gender and accusative—paving the way for ClMut. A strikingly similar effect is found in the interaction of clitics. Bonet (1991, 1994) observes that in many clitic clusters in Romance the output is opaque with respect to the information encoded by the same clitics in isolation. According to her, in those contexts the clitics suffer an impoverishment process, as a consequence of which their representation corresponds to the PF realization of another clitic.

Consider now the distribution of the clitic/agreement strategies for [+DOM] objects, including pronouns (53a–c), vs. the distribution of strategies for [−DOM] objects. ((68b–c) repeat (27a–b).)

Remember that, unlike number, person agreement is valued between se and T (section 4.3). The difference between these contexts and those where ClMut applies is that the clitics in (68), [+DOM] object clitics, contain [person]. We may thus assume that these person features conflict with the person feature in T and, in consequence, the clitic cannot mutate.³³ Similarly, when T fully specifies [person] and [number] (e.g., (62)), the clitic has no room in T’s resulting morphological exponent to mutate into subject number agreement, not even with lo(s)/la(s) (68c). In sum, ClMut is possible when lack of [number] in the subject and lack of [person] in the object cooccur.

With ClMut, Spanish morphology deals with the fact that subject agreement is morphologically mandatory at the expense of object clitics. As we discuss next, with some interesting differences, that is a property that ClMut shares with other morphological phenomena crosslinguistically.

Crosslinguistically, ClMut belongs to the family of effects known as agreement displacement. To illustrate, in Basque, under certain conditions, ergative agreement takes the form and position of absolutive agreement. In order for this Ergative Displacement (ED) to take place, a necessary condition is that the absolutive argument be 3rd person, an agreement that happens to lack a morphological exponent ((69c–d) vs. (69a–b)).

Most analyses interpret this distribution as indicating that absolutive agreement is defective for [participant], but we have extensively argued elsewhere that no agreement relation holds with 3rd person objects in many languages including Basque (O&R 2007). If this is correct, the ergative element colonizes the absolutive slot in (69c–d) when it lacks person features altogether. This description is not very different from diachronic accounts (Gómez and Sáinz 1995, Lakarra 2005, Ariztimuño 2017), where the subject pronoun is argued to occupy the first position precisely in those cases where no object pronoun may occupy it (historically, Basque lacks 3rd person pronouns).

Current analyses of ED mostly propose syntactic solutions (e.g., Ortiz de Urbina 1989, Rezac 2008, Béjar and Rezac 2009); however, there is also a long tradition of morphology-based explanations (e.g., Laka 1993, Albizu and Eguren 2000, Fernández and Albizu 2000, Arregi and Nevins 2012). The clear similarities between ClMut and ED argues for a morphological approach. The main difference between ED and ClMut is that while in the latter [person] has been checked between T and se, in the case of Basque there is no person agreement in v. That explains why the entire ϕ-feature set of the ergative can take over the object agreement slot in Basque while only number is involved in ClMut, and why ClMut does not apply to animate clitics. The differences then follow from the different morphological specifications of the host exponents in each case, as desired.

Agreement in SECs with postverbal DPs exhibits radically different properties. In addition, its structural description does not include an object clitic; as a result, it cannot be an instance of ClMut. In this section, we argue that this agreement is the reflex of a post-PF procedure that we call Number Harmony (NH).

The only role syntax plays in NH is in providing a structure where the set of ϕ-features in T includes a [person] value supplied by se but no [number], and where there is a DP nearby in postverbal position (70). Unlike in ClMut, in NH the V-object relation is not represented in V’s morphological word.

Agreement follows two different paths in that context: either it takes a default value (71a) or it adopts the value of the closest nominal (71b). When the closest nominal is singular, as expected the verb never shows up in plural (71c).

In all cases, “closest” is computed in pure linear proximity terms, as shown by the battery of long-distance cases discussed in section 2.1.

In sum, the distribution of agreement in these contexts does not show any of the properties of a syntactic checking relation. Additionally, it lacks a morphological motivation: there is no possible source for the plural morpheme within the morphological word in (71) and, unlike in ClMut, the agreement patterns in unpredictable ways. The same conclusion may be reached concerning PF: the application of the plural number mechanism is larger by far than the domain of regular phonological processes, as clearly shown by the hyperraising case (15) from section 2.1, repeated here.

In (72), the nominal triggering plural marking in the matrix verb is an argument of the embedded clause, far from any conceivable prosodic unit with it.

As far as we can see, the only option in these contexts is that agreement patterns are determined by extragrammatical factors. For speakers who invariably resort to number harmony with [−DOM] direct objects (P-SECs) and to a singular default in the rest of the cases (I-SECs), we must assume that this additional regularization process is a learned strategy imposed by means of socioeducative pressure.³⁴

In the next section, we briefly review other instances of subjects lacking number specification in Spanish that follow a similar pattern. We show that in some cases number agreement works essentially as it does in SECs, while in other cases it is fully determined, sometimes in completely unexpected ways.

In addition to contexts of gender agreement that we cannot consider here (Benmamoun, Bhatia, and Polinsky 2009, Demonte and Pérez Jiménez 2012), the same uncertainty also appears in other so-called impersonal contexts (see also Feliu to appear for verbs taking an infinitive clause as subject).

In Standard Spanish, agreement between the verb and the nominal phrase is not recommended in existential haber and hacer contexts (NGLE 41.6b), and for many speakers the only option is a default 3rd person form. However, the use of plural agreement, where the verbal form reflects the number of its nominal argument, is increasing considerably, and in some areas it is becoming general.

Agreement in these contexts is remarkably similar to what we find with SECs, including a high degree of dialect-internal and idiolectal variation. In the case of haber, while the impersonal present form, hay, is invariant, the past form is regular 3rd person with a corresponding plural form habían. Consequently, NH appears in the past but almost never shows up in the present, reinforcing the conclusion that we are not dealing with syntactic or morphological agreement (DeMello 1991, Gràcia i Solé and Roca 2017, O&R 2022).

Regarding hacer, when the complement includes a measure phrase, for some speakers it often triggers number agreement on the verb (74b).

As in SECs, NH with haber/hacer is idiolectal and arguably subject to cultural pressure.

The situation is different in copulative sentences whose subject is the neuter deictic pronoun eso or a propositional subject.

In (75), the pronoun in subject position does not encode number, and the predicate, tonterías, determines agreement. Agreement is obligatory; default singular is rejected. Similarly, with propositional subjects it is determined by the predicate.

Note that in these cases where the predicate is replaced by the invariant predicative clitic lo, plural agreement is still obligatory (para mí lo son).

When the complement of V Ā-moves to a preposed position, clitic doubling is not possible (77).

In SECs, these structures exhibit the lack of systematicity typical of NH (78). In fact, this is the only case where even speakers who otherwise always adhere to the norm show some degree of variation.

This exceptional behavior is expected in our analysis: while arguments in CLLD are base-generated in the left periphery, the Ā-chain tail in (78) includes a postverbal copy of the object in its VP-internal position. Clitic doubling is thus not possible, and consequently ClMut lacks the structural conditions to apply (58). In contrast, this Ā-chain configuration is a potential target for NH to apply, and the lack of systematicity reappears: (79a) illustrates the lack of number with plural objects, (79b) the case of plural agreement with DOM objects, and (79c) long-distance number agreement.³⁵

In sum, different pieces of evidence converge and support our proposal that subject number agreement in SECs conceals two different phenomena with very distinctive properties, both occurring postsyntactically, namely, ClMut and NH.

ClMut is a rather regular and systematic morphological process by means of which an argument D-clitic shows up morphologically as number agreement. It coexists with a clitic strategy whose distribution is also systematic and predictable both dialectally and in terms of the configurations where it applies. The dialectal cut lies between its optionality in C-varieties and its obligatoriness in A-varieties.

On the other hand, NH is a less systematic, extragrammatical phenomenon with a vague distribution. NH in SECs shares many properties with similar agreement “failures” sketched in section 5.2.2. Moreover, its distribution seems to be largely determined by sociolinguistic and stylistic factors, such as social acceptability and acquisition of the Academia’s norm, that escape the domain of grammar.

If the results in this article are on the right track, they open new questions about and ways of researching old issues regarding the syntax and morphology, as well as processing, of agreement. They also raise questions about crosslinguistic variation and diachronic change.

An important issue that often arises concerning our proposal is how cross-Romance variation fits into the picture. Our analysis of Spanish se illustrates that a fine-grained investigation is required before deciding the situation in each language; consequently, our observations here must perforce be considered highly speculative.

An important contribution of Cinque’s (1988) seminal analysis of se/si is that it links the differences between P-SECs and I-SECs to the lexical properties of the different se items involved. From our point of view, putting the burden of explanation on the lexical properties, as Cinque does, seems the right strategy to account for the behavior of nonparadigmatic SECs and other SECs language-internally. As shown, the effects of a minimal lexical distinction may be considerable at different levels of the grammar and beyond. Similarly, divergences in diachronic evolution may be responsible for important synchronic differences in the Romance family.

The diachronic diversification of Latin SĒ’s attributes and cliticization patterns is not uniform. In the case of Iberian languages, although some authors (e.g., Maddox 2021) present a unified approach to the evolution of all ses, some facts suggest a split at least since the 16th century (Monge 1954 for Spanish, Martins 2005 for Portuguese). If that is correct, as the Latin reflexive SĒ evolved, a variety of lexical clitics with particular specifications emerged in Romance languages, resulting in different subsystems both language-internally and crosslinguistically. Taking Spanish as an arbitrary reference point, different possibilities arise for the evolution of non-paradigmatic se.

First, some languages have taken a completely different path, most characteristically French and many dialects of Portuguese, which developed alternative impersonal structures with on (Wolfsgruber 2017) and a gente (Martins 2005 and references), respectively.

Second, Romanian might illustrate the case of a language that presents a more conservative evolution. Following MacDonald and Maddox (2018) and Maddox (2021), some of its properties suggest that diachronically se is in a different stage—a different lexical item in synchronic terms: though a clitic pronoun in Spanish and Italian, se is a voice marker in Romanian. Consequently, agreement facts must be derived very differently, which is consistent with the observation (Cornilescu 1998, Dobrovie Sorin 1998, Giurgea 2019) that Romanian does not show agreement mismatches of the kind exemplified here.

Third, other languages may have gone further in diversifying lexical se items. For instance, Italian has developed a 1st person plural se—similar to French on or Portuguese a gente—that does not exist in Spanish and Catalan (Cinque 1988, Chierchia 1995, D’Alessandro 2007, Ordóñez 2021).

If lexical diversification constitutes an important source of crosslinguistic variation, a priori we will not expect to be able to determine the direction of the changes. Consequently, we have no expectations about what the output will be in each system. However, what we could expect is that the resulting lexical item will be internally coherent relative to the general properties of syntactic derivations, as we have argued for se in Spanish, with clear empirical predictions about its syntax and morphology.

A different area where much work may be advanced is in determining more precisely the syntax and morphology of “defective agreement” contexts and the limits of variation at the postsyntactic level. Once we reconsider number agreement facts in those new terms, the possible contexts of defectiveness and its potential to reanalyze many constructions (quirky subjects, long-distance agreement, expletives, etc.) are quite broad. Our proposal also has consequences for the analysis of classical agreement restrictions and their extensions. In particular, if number agreement in many of these contexts is treated as a postsyntactic effect, what Baker (2008) calls the “two-and-a-half agreement” factor becomes an “at-most-two-agreement” factor, reducing the playing field of possible explanations for agreement restrictions considerably. That is particularly clear if, as argued here, syntactic agreement applies in a single cycle.

Related to the previous point is a question regarding the organization of agreement among the different components in the faculty of language in the broad sense (Hauser, Chomsky, and Fitch 2002, Fitch, Hauser, and Chomsky 2005). The issue is how to capture the syntactic and processing differences reported in the literature, such as the ones concerning person vs. number vs. gender. Cartographic approaches (e.g., Shlonsky 1989, Sigurðsson and Holmberg 2008, Mancini et al. 2014) interpret these asymmetries as reflecting the projection of independent functional heads in syntax. Other proposals (e.g., Chomsky 1995, 2000) argue for an “all at once” checking operation, combined with some processes in other components of the grammar. It is worth noting that the two approaches are not incompatible, and in fact it might be the case that we are dealing with a nonuniform system (see, e.g., Dillon et al. 2013, Zawiszewski, Santesteban, and Laka 2016 for discussion). However, if the line of analysis we propose can be effectively extended, there is room to reinterpret the evidence for separated probes in a more compact syntactic system (Ackema and Neeleman 2019, O&R 2022, and references). Fine-grained theoretical and experimental work on the syntax of agreement restrictions as well as on language processing would help to clarify some of these issues.

Finally, as mentioned several times above, our results on SECs intermingle in a complex way with the analysis of DOM, clitic doubling, and the syntax and morphology of Romance clitics. As we have suggested, most of the issues involved are independent of SECs, but these structures provide a tool particularly well-suited to reveal new properties of the system of clitics and agreement.

This work was financed in part by the Junta de Extremadura’s grant GRI5163 (Grants to Research Groups Program) and Research Project IB18080 to the second author, and by the institutions supporting the research activities of the Basque Group of Theoretical Linguistics (HiTT): the Basque Government grants IT1396-19 and IT1537-22 (Euskal Unibertsitate Sistemako Ikerketa-taldeak); the Ministry of Science, Research and Universities (MCIU) grant PROLE PGC2018-093464-b-100, financed partially by the European Union’s FEDER program; and the University of the Basque Country (UPV/EHU) grant GIU18/221.

The two authors, listed in alphabetical order, are equally responsible for the entire content of the article. We are grateful to Pablo Albizu, Alejo Alcaraz, Olga Fernández Soriano, Ángel Gallego, Silvia Gumiel, Georg Kayser, Jonathan MacDonald, Jairo Nunes, Paco Ordóñez, Andrés Saab, Michelle Sheehan, and Myriam Uribe-Etxebarria. To Andrés, as well as to Luigi Andriani, Carlos Muñoz, Mercedes Pujalte, and Pablo Zdrojewski, very special thanks for their patient discussion and important clarifications on their dialects. We also are thankful to two anonymous LI reviewers and to Eric Reuland, the LI European editor, for many valuable comments. Very special thanks to one of the reviewers, who contributed extremely accurate observations and suggested many new ideas for further research. Previous versions of this article were presented at the II Encuentros sobre dialectos del español (Universidad de Castilla-La Mancha, Ciudad Real), the Spanish Dialects Meeting-V Encuentro de sintaxis de los dialectos del español (Universidad de Extremadura, Cáceres), the Romance Linguistic Circle (University of Cambridge), and the Basque Group of Theoretical Linguistics (HiTT) 2018–2019 Seminar Sessions. We are thankful to the participants for discussion and suggestions.