Unexpected agreement in Spanish revisited

Verbal agreement in Spanish is materialized as the rightmost morpheme on a verb. When enclitics attach, the expected order is: verbal base + agr(eement) + cl(itic): hága-n agr -lo cl ‘Do it!’. Nevertheless, in non-standard Spanish, verbal agreement can appear at the right of the clitic, giving rise to what has been named unexpected agreement: haga-n agr -lo cl -n agr or haga-lo cl -n agr ‘Do it!’. This squib discusses different data related to unexpected agreement in order to provide some generalizations which previous approaches to this topic fail to address. So as to revisit the operations involved in this broad phenomenon, it is pivotal to make a distinction between unexpected exponents and unexpected locus . We appeal to heads and hands younger and stronger than ours to extend what we have gotten right in this study and to correct what we have gotten wrong. (Harris & Halle 2005: 219)


Introduction
In standard Spanish, enclitics attach to special verb forms (imperatives, infinitives and gerunds), giving rise to the following linear order: verb base + verb morphology + cl(itic).
The main goal of this squib is twofold: to show that this phenomenon is much broader than the examples considered by the analyses aforementioned, 4 and, consequently, to derive some generalizations from the data hereby presented. A detailed description is provided together with relevant empirical and theoretical statements about the occurrence of unexpected morphology as the result of internal or external merge.
This squib is structured as follows. Section 2 summarizes two previous analyses of the data in (2) and (3), whereas §3 presents other significant data in an endeavor to develop the most relevant generalizations about unexpected agreement in Spanish. In §4, I discuss the distribution of plural morphemes. Section 5 concludes with some final remarks.

A post-syntactic analysis
A&N2018 revisit Harris & Halle's (2005) approach and argue that (2) and (3) are the result of a second position effect. They claim that in non-standard Spanish the morpheme -n is a second position clitic within the post-stem clitic domain. Doubling (2) or displacement (3) of -n occurs in order to place a clitic to its left. That is to say, the agr(eement) node is reanalyzed as a cl(itic), and consequently it is subject to clitic constraints, such as Non-initiality. Their Generalized Reduplication rule is formulated as follows. 3 In some varieties (i.e. Latin American varieties and some southern Iberian varieties, as well as Canarian Spanish), the morpheme -n corresponds both to the 2pl (ustedes canta-n 'You pl sing'), and to the 3pl (Ellos canta-n 'They sing'). 4 Some of the proposals also mention the existence of unexpected agreement with infinitives (hacer-lo-n) and gerunds (haciendo-lo-n). Arregi & Nevins (2018) also include unexpected agreement with the first person plural (haga-mos-lo-n In the Generalized Reduplication rule, unexpected positions are treated in terms of special brackets that are interpreted by the phonology. Square brackets ([ ]) give the instruction of repeating all the material inside them, while angle brackets (><) specify which material to delete. The implementation of (4) in (2) and (3)  This Generalized Reduplication rule is accompanied by both general and dialectal conditions and constraints, one of them being the Plural Intervention Condition (PIC), which blocks -n movement when a plural clitic is involved. This condition will be explored in §3.3.

A syntactic analysis
M&S2011 revisit the data in (2) and (3), and argue against post-syntactic approaches. They relate unexpected agreement in Spanish to the behavior of clitics in Italo-Romance and Albanian dialects in order to show that the grammars with mesoclisis can be characterized on the basis of a property of the D inflection (associated with the Inflection node or the C domain). The structure they propose for imperative sentences with enclitics is shown in (7).
[CI(rrealis) [cl* [C [cl* [I(nfl) Under the structure in (7) and assuming the analysis the authors offer for other dialects, our data in (2) and (3)  The structures in (8) and (9) show that movement is also relevant to their proposal; however, in this case, the moved syntactic object is the verb with or without -n, not the inflectional morpheme alone. As can be observed, there exist two places for agreement: the Inflectional domain and the C domain. In the case of displacement (8) -simple mesoclisis, following the authors' terminology-the agreement morpheme appears only within the Inflectional domain, whereas in the case of doubling (9), this element is placed in Inflection as well as in CIrrealis. In Inflection it behaves as a clitic (which is the reason why the label is D), while in CIrrealis it is part of the verb (see M&S2011's paper for detailed discussion).

Summary
The main features of the analyses presented in this section can be summarized as follows: Feature 1: unexpected morphology is the result of movement. This means that the unexpected exponent (-n) is the "expected" one (-n), but in a different position. Feature 2: regarding movement, unexpected morphology depends on locality, which is to say that only clitics can intervene between the expected and the unexpected locus. Feature 3: there is a general condition (Plural Intervention Condition) that blocks movement (A&N2018).

How unexpected is unexpected agreement?
Both A&N2018 and M&S2011 manage to solve the overgeneration and the unpredictability of previous approaches and account for the behavior of clitics and agreement morphology in non-standard dialects. Putting the syntactic and post-syntactic debate aside, it must be acknowledged that both analyses provide an explanation for (2) and (3). The main problem is that (2) and (3) are just a subset of a much more complex phenomenon that includes the cases I present in this section. Most of the data outlined here are adapted from Mare (2017;2018) and from other web searches. All the data were confirmed with native speakers and some of them belong to my own variety of Spanish (see footnote 8 for details).

Discussing Feature 1
As mentioned in §1, this kind of unexpected agreement is broader than the data in (2) and (3). 7 Regarding imperatives, the phenomenon is not restricted to 2pl, but it is also found with 1pl morphology (-mos) and with Iberian 2pl morphology in colloquial imperatives (-d). The Pers(on)/Num(ber) morphology I concentrate on is presented in Table 1.
In the two cases just aforementioned, the unexpected exponent is not -mos (10) or -d (11), as could be expected (Feature 1), but -n. 8 7 It is necessary to mention that there is significant dialectal variation and that a typological study on this topic must be developed in the future. See Rosenblat (1946) and Kany (1994Kany ( [1945) for a relevant description concerning this widespread phenomenon. 8 The following data are adapted from Mare (2017;2018) and from other web searches, in the case of 2pl in colloquial Iberian Spanish. Some relevant examples of 2pl are presented below: (i) Observa-d y deci-d-me-n qué opinais lood.imp-2pl and tell.imp-2pl-cl-N what think.2pl 'Take a look and tell me what you think about it' https://charhadas.com/articulo/2850542-puertas-que-invitan-a-sonar (10) a. haga-mo(s)-lo-n / *haga-mos-lo-mos The dialects that admit unexpected agreement with infinitives (12) and gerunds (13) show this behavior as well. The unexpected morpheme is always -n. The infinitive morpheme -r or the gerund morpheme -ndo can never be in different loci.
These examples pose a number of problems for the analyses developed in §2, in terms of what I call Feature 1, for not all grammatical forms are predicted and not all ungrammatical ones are avoided. For instance, some problems arise when one tries to apply these analyses to cases which involve 1pl agreement. All other things being equal, the result of movement is always ungrammatical, because the unexpected exponent must be -mos, but, as the data have shown, it is -n.
Adapting A&N2018's proposal 10 help.imp-2pl-cl-N tell.imp-2pl-cl-N 'Please, help me. Tell me…' https://www.tuexperto.com/2012/05/31/sony-xperia-acro-s-analisis-a-fondo/ In Mare's work it is possible to find real examples from the web (with their URLs), regarding unexpected agreement with 1pl, infinitives and gerunds, and the combination of unexpected agreement with plural clitics (see below). All these data were confirmed with native speakers. In fact, unexpected agreement with 2/3pl, 1pl and plural clitics belongs to my own variety. I have adapted the examples simplifying them and using just two or three verbs for convenience. 9 Kany (1994Kany ( [1945: 146) mentions some examples of displacement with materialization of -n in Costa Rica: quieren casa-se-n (casa-r-se) 'they want to get married'. On the other hand, Kayne (2010: 166) calls attention to cases in which two clitics are separeted by the infinitival morpheme -r: dá-se-r-lo (dá-r-se-lo) 'give it to her/him/them'. These cases will be discussed in further research on this topic. 10 I will return to A&N2018's proposal in §4. As illustrated above, the morpheme materialized as -mos cannot be doubled or displaced. Moreover, while the co-occurrence of -mos and -n is possible (haga-mos-lo-n) -which could be described as partial doubling -the absence of -mos with -n in unexpected agreementpartial displacement -is ungrammatical with the relevant meaning (*hagalon 'Let's do it!'). What I exemplify with the 1pl can be extended to the examples presented in (11)-(13).
Considering the grammatical forms, two questions arise: (a) why is -n the only agr exponent that can appear in unexpected positions?, and (b) why -n, but not -mos, -d, -r or -ndo, can be displaced (separated from the verbal base)? While the first question is essential to establish the right generalization and to cast light on a proper analysis of the data, the second one shows, to the detriment of M&S2011's claim, that Spanish data are slightly different from the Italo-Romance and Albanian examples, because the only exponent that can be displaced or doubled is -n. Specifically, the authors present cases in which mesoclisis does not distinguish person information in the agr node (M&S2011: 1104). In Senise, for instance, a clitic (d'd) can intervene and the 1/2 person morpheme (-imǝ) can split from the verb base. Nevertheless, as the examples in this section indicate, this kind of intervention is impossible in Spanish when the verbal exponent is -mos, -d, -r or -ndo. In fact, the agreement morpheme in unexpected positions is the expected one only when the subject is 2/3pl (-n): ¡Hagalon! (haga-lo cl -n 2pl ). This restriction does not work in (22). On the whole, verbal forms that admit enclitics, i.e. imperatives, infinitives and gerunds, can present unexpected inflection. The added morphology is unexpected in three ways: (a) It appears in a non-prototypical position VERB imp -(Num/Pers)-cl-n. (b) It is not the expected exponent: -n instead of -mos/-d. (c) It appears in absence of Num/Pers morphology in the verb with the enclitic: infinitive/gerund-cl-n.

Discussing Feature 2
The analyses discussed in 2 explain mesoclisis as the result of movement, this operation depending on locality (Feature 2). In order to revisit this feature, it is necessary to focus on infinitive and gerund forms, which can be the surface materialization of different constructions. As Mare (2018) points out, it is noteworthy that unexpected agreement is not restricted to a particular construction, but to every infinitive and gerund form followed by a clitic (23-27).
[subject/object control constructions] swear.2/3pl do.inf-cl-N 'They/You swore to do it.' b. Obligó a los trabajadores a hacer-lo-n. compel.3sg the workers to do.inf-cl-N 'They/You obliged the workers to do it.' (25) Se hieren con la verdad [infinitive with prepositions] cl hurt.3pl with the truth para no matar-se-n con la mentira. to not kill.inf-cl-N with the lie 'They hurt each other with the truth so as not to kill each other with lies.' Ellos reaccionan pegándo-le-n.
[adverbial gerund constructions] they react.prs.3pl kick.grnd-cl-N 'They react kicking them.' A los niños les [infinitive as subject] to the boy.pl cl.dat.3pl gusta bañar-se-n. like.prs swim.inf-cl-N 'The boys like swimming' There are two aspects related to locality that deserve further discussion. On the one hand, it is not totally clear where to find the plural information that triggers the presence of -n after the enclitic. In some cases like (23a-b), (24a) and (26), this information is in the matrix verb, while in (24b) and (27) it is in a different constituent. In (25) the constituent that presents the plural feature seems to be even further beyond: hiere-n (hurt-3pl). On the other hand, regardless of the presence of Num/Pers morphology in any constituent, some clarification on the type of locality involved is needed. The superficial locality observed with imperatives seems to be quite different from the one involved in these examples. In the former, Num/Pers morphology is present in the verb to which the enclitic attaches; while in the latter, neither Num nor Pers morphology is recognized in the verb.
To sum up, unexpected morphology in infinitives and gerunds is not restricted by the structure in which these forms appear, all of which seems to trigger theoretical consequences regarding locality and the properties of non-finite forms.

Discussing Feature 3
Finally, there are some data that could be relevant to discuss the distribution of plural morphemes in Spanish and represent some (minor) problems for A&N2018's Plural Intervention Condition (PIC). Before going on, I will develop a brief description of Spanish clitics, following Harris's (1995) segmentation. It is useful to mention that the plural morpheme -s in (28b), (29c) and (30b, d) is the one that materializes plurality in the nominal domain (libro 'book'/libro-s 'books'). The PIC is proposed as a general condition which describes the empirical fact that the plural exponent -s cannot appear with unexpected -n (*-sn/*-ns). Specifically, A&N2018 argue that plural clitics block mesoclisis. 12

(31) Plural intervention condition (PIC) D cl is not [+pl]
According to these authors, this general condition predicts that examples such as (32) are ungrammatical, contrary to fact. 13 However, it does not block unexpected agreement with 11 Third person accusative clitics distinguish gender l-o/l-o-s for [-fem] and l-a/l-a-s for [+fem] (-o-as the default item and -a-as the class II item in Harris' terminology). 12 The phonotactic constraints like /*sn/ or /*ns/ can be solved by means of Spanish epenthetic /e/, so this does not seem to be a phonological problem. 13 Mare (2017) shows that examples like (32) are not only grammatical but also very productive. Accordingly, (32) is not given with a star.
(i) Diga-n-le pl -n a [los niños] pl . say.imp-2pl-cl.pl-N to the boys 'Say it to the boys!' the clitic se (33), which could be interpreted as plural (30e), but in the Morphological Structure loses this feature by an impoverishment rule (see Bonet 1995, among others Accordingly, the PIC predicts that it would not be possible to find the 1pl nos (28b) in these constructions, because it is a plural clitic with plural morphology -crucially, plural nominal morphology (-s). Notwithstanding, the data in (34)  give.imp-cl.1pl-cl.acc.pl-N 'If you don´t want them, give them to us!' All in all, clitics can be plural and their plurality does not block unexpected agreement when possible. In spite of this, the nominal plural morpheme -s and the verbal plural morpheme -n cannot be adjacent. In fact, they seem to be in complementary distribution: vamo(s)no-n 'let's go'/vamo(s)no-s/*vamo(s)no-s-n or *vamo(s)no-n-s. I return to these facts in the following section.

Further remarks on unexpected agreement
According to the data discussed in the previous section, it is possible to highlight three facts on unexpected agreement: (a) the presence of -n in unexpected positions, irrespective of the properties of the expected morpheme; (b) the locus of the features that trigger the presence of -n, especially in infinitives and gerunds; (c) the distribution of the exponents -n and -s.
A&N2018 offer a solution for (a) and (b) under their Generalized Reduplication rule. They propose that in the varieties that accept unexpected agreement in infinitives and gerunds there is not a non-initiality requirement for -n, but a non-final requirement for clitics. The -n item prevents the clitic from occupying final position. Regarding the locus, the feature that triggers the insertion of -n ([+pl]) would be morphosyntactically present on the infinitive or gerund. 15 Although this seems to explain the facts just mentioned above, the non-final requirement for the clitic predicts that a morpheme must follow it, regardless of (ii) Diga-le pl -n a [los cordobeses] pl . say.imp-cl.pl-N to the people-from-Córdoba 'Say it to people from Córdoba!' 14 As far as I am concerned, there are no examples of 2sg imperatives in a configuration like the one represented by the examples in (34): deci-nos-lo-n '(You sg ) say it to us' or da-nos-lo-n '(You sg ) give it to us'. Moreover, there is no alternation between -n and -s regarding the clitic plurality in different constructions: Juan quiere comprar-lo-s/*comprar-lo-n 'Juan wants to buy them'. For that reason, I assume that the presence of -n in (34) is related to a plural subject. 15 The authors consider Minkoff's (1993) and Mare's (2018) data regarding 1pl imperatives as independent evidence for this requirement.
plurality. If this were the case, it would be necessary to add some restrictions to avoid the insertion of undesirable Num/Pers exponents. 16 Despite this partial solution, there are still some remarks to make on the distribution of -n and -s. Minkoff (1993) discusses examples with the first person plural in Caribbean Spanish and the data he recovers are slightly different from the cases that have been presented in the previous sections of this paper: the item that appears after the enclitic is -s, not -n.
(35) a. Standard Spanish De-mos-le! give.imp-1pl-cl 'Let's give her/him (something)!' b. Caribbean Spanish (adapted from Minkoff 1993: 179) De-mo-le-s! give.imp.1pl-cl-S 'Let's give her/him (something)!' When comparing Minkoff's and Mare's data, the discussion on the distribution of -n and -s ensues again. The piece of data taken into consideration in the previous section leads to the empirical generalization sketched in (36). However, this generalization is not enough to cover all dialectal types.
(36) Empirical generalization on unexpected agreement in Spanish When Agr [+pl] merges in a non-canonical position, it is materialized as -n.
According to Minkoff's data, when the imperative presents 2pl inflection in Caribbean Spanish, the expected exponent -n moves to an unexpected position (displacement of -n). However, against (36), when the imperative presents 1pl inflection, the expected exponent -s of -mos is the one that moves to an unexpected position, as shown in (35b) (displacement of -s). In Mare's data, the phonological exponent that appears linearly after the clitic (i.e., in an unexpected position) is not the same as the expected one (i.e., -n instead of -mos or instead of the -s related to -mo-s). The difference between Minkoff's and Mare's 1pl examples can be summarized as follows: Caribbean Spanish presents only displacement and the exponent is the expected one (-s with 1pl and -n with 2pl); other non-standard Spanish varieties present displacement and/or doubling and the exponent is always -n, which can or cannot be the expected form. When the imperative is 2pl, -n is expected, while when it is 1pl the exponent should be -mos or -s, but it also happens to be -n.
In addition, Minkoff (1993) observes the same behavior for the -s of the clitic nos (1pl): when a clitic precedes it, -s appears after the clitic cluster (expected exponentunexpected locus) (37) a. Standard Spanish De-nos-lo! give.imp.2pl/sg-cl.1pl-cl.3sg 'Give us it!' 16 Heap & Pato (2012) and Mare (2018) present some examples of unexpected -n with infinitives and gerunds with clear singular reference. As mentioned in both papers, despite the lack of productivity of these cases, it is relevant to point out that the unexpected item is -n, no matter if the explicit Num/Pers morphology is materialized by a different item (for instance, 2sg -s in the example below).
(i) tu puede-s hacer-lo-n you can-2sg do.inf-cl-N 'you can do it' [Mallón, F. La sangre del Copihue, 2004, p. 189, Chile] (40) a. Selon and selos speakers Eso se lo-s dicen a los estudiantes. this cl.pl cl.sg-S say.2/3pl to the students 'You/They say this to the students.' b. Selon and selos speakers *Eso se lo-n dicen a los estudiantes.
As regards the occurrence of two enclitic elements, Minkoff (1993: 189) points out that -s displacement is impossible when the clitic se is involved (*de-mo-se-lo-s 'Let's give her/him it!'). According to the author, the merger of -s is blocked, because "merger targets clitics in cyclic succession from left to right" (Minkoff 1993: 190) and the clitic se is incompatible with plural morphology in any context (*den-[se-s] i -la [a ellos] i 'Give it to them'). Interestingly, plural -n does not present this restriction (39b). A&N2018's PIC can account for -n merger (consider diga-n-se-n-lo 'say it to them' in (33) Of course, these data deserve further study in order to establish a clear typology, but this description should be taken into account to understand the nature of unexpected morphology. In fact, the decomposition proposed by Minkoff (1993) as well as the comparison between Mare's and Minkoff's data seem to be useful to reconsider this phenomenon.
In the light of the data presented above, A&N2018's PIC should be revisited to cater for the different possibilities explored in this paper, but this is not enough to account for the distribution of plural morphemes. All in all, what seems to be impossible in Spanish is the adjacency/locality of two # pl nodes and the lack of Num/Pers morphology related to T(ense), regardless of the plurality of the clitics involved.

Final remarks
In this paper, I have discussed some pieces of data related to unexpected agreement in Spanish. As shown, the phenomenon is broad and a review of the complete panorama is essential to establish some generalizations otherwise missed. From a theoretical standpoint, the distinction between unexpected exponents in unexpected locus and expected exponents in unexpected locus could be relevant to discuss the kind of merge involved: internal merge (movement) or external merge, 19 Minkoff's data probably being an example of the former, while Mare's data one of the latter. The facts related to proclisis (38, 40) would support different approaches. On the other hand, the data show that Person and Number behave differently. Morphemes related to person cannot appear in unexpected loci and cannot be split from the verbal base, while morphemes related to number can move/merge freely. This difference could be in tune with approaches that split the Agr morpheme into Person and Number, the first being higher in the structure (see, for instance, Sigurðsson's 2000 Feature Uniqueness Principle). Related to this, unexpected morphology with infinitives and gerunds could provide some evidence to argue that non-finite forms contain Person and Number information. If these features are split, the presence of -n (only related to Number) would not be so unexpected.
To conclude, substantial evidence emerged in favor of the relevance of paying attention to the distribution of plural morphology in Spanish. It seems to be clear that -s and -n are in complementary distribution, and this can be accounted by A&N2018's PIC. However, this is not enough to explain, for instance, that plural -s can be omitted in many dialects, while verbal agreement/morphology must appear at least once in all varieties of Spanish. Moreover, plural -s can move in any context, while the presence of -n in unexpected loci is restricted to enclisis. Taken these observations into account could be useful to revisit the operations involved. All in all, I hope that this squib makes a relevant contribution to Harris & Halle's appeal.