Abstract

Harris and Halle (2005) present a framework (Generalized Reduplication) that unites the treatment of phonological reduplication and metathesis with similar phenomena in morphology, thereby accounting for the apparently spurious placement of the imperative plural -n in mesoclitic Spanish forms such as hága-lo-n ‘Do it!’, in which clitic lo is sandwiched between the verbal stem and the plural suffix. Kayne (2010) has challenged their analysis, arguing that such cases should be treated purely within the syntax. We reassess some of Kayne’s arguments, agreeing with his conclusion that the most important desideratum of any general analysis of such phenomena is restrictiveness. However, we contend that greater restrictiveness can be achieved through morphotactic constraints and repairs in the Generalized Reduplication formalism, triggered by a Noninitiality constraint on the positioning of the plural affix, and we develop constraints on these operations that situate interspeaker variation within the postsyntactic component.

1 Mesoclisis in Spanish and Other Languages

Our focus in this article is mesoclisis in Spanish, a pattern of displacement, and its relation with doubling of the agreement suffix before and after the clitic, in forms like those in (1)–(2), where the agreement suffix is plural -n and the clitic is reflexive se.1

(1)

  • Standard form: Enclisis

  • Siénte -n -se!

  • sit.IMP -PL -CL.REFL

  • ‘Sit down! (imperative plural)’

(2) Alternation between displacement and suffix doubling

  • Siénte -se -n!

    sit.IMP -CL.REFL -PL

  • Siénte -n -se -n!

    sit.IMP -PL -CL.REFL -PL

    ‘Sit down! (imperative plural)’

In what follows, we use the cover term mesoclisis for both displacement as in (2a) and doubling as in (2b). This phenomenon has been the focus of much discussion (see, in particular, Harris and Halle 2005 and Kayne 2010).

Our point of departure is the extent to which these nonstandard forms share a diachronic and synchronic relation. While many approaches might find it straightforward to treat them using separate machinery, our goal is to unify them and understand the coherence of their relation. In exactly this connection, Haspelmath (1993) discusses the “externalization of inflection”; when inflection gets trapped inside of derivation, eventually it would like to move out, as in (3).

(3)

  • Externalization of inflection in compounds

  • sister-s-in-law > sister-s-in-law-s > sister-in-law-s

  • (Haspelmath 1993:288)

This example involves compounding, a derivational word-formation process, and the data are most likely accessible to readers familiar with such forms in English. However, examples can be found in various other languages with derivational suffixes, as Haspelmath (1993:280) demonstrates for the case-marked form of the Georgian pronoun for ‘anything’, which involves the interaction of indefiniteness and case marking. This combination, Haspelmath shows, displays a three-way alternation between its original etymological sequence ra-s-me, a hybrid form ra-s-me-s, and its reordered form ra-me-s, with the dative case marker -s now outside of the derivational suffix -me (whose function is to derive an indefinite pronoun from an interrogative one).

(4)

  • Externalization of inflection in Georgian indefinite formation

  • ra -s -me > ra -s -me -s > ra -me -s

  • what -DAT -INDF

  • ‘anything’

The tendency to reorder configurations in which inflectional suffixes (e.g., plurality marking, case marking) end up occurring inside of derivational morphemes that have been added outside of them is argued to follow from Greenberg’s (1963) Universal 28, later elaborated by Bybee (1985) and Dressler et al. (1987).

(5)

  • Derivation inside Inflection

  • Derivation should be linearly between the root and inflection.

For Bybee, this morphotactic constraint has a motivation in that derivational affixes are iconically closer to the stem’s meaning, while for Dressler et al., the idea is that inflectional formatives are outwardly indexical toward other sentential elements. Whatever the functional or processing motivation may be, let us take Derivation inside Inflection as a morphotactic constraint that holds at all synchronic stages of a grammar.

From this perspective, Haspelmath (1993) notes that historically, one might reexamine the Spanish examples above in this light, on the assumption that the clitic, originally a weak pronominal and a separate element, is outside of the plural agreement. However, once the verbal reflexive marker comes to be its own derivational marker through a kind of process of univerbation (namely, becoming a single phonological word with its stem), the plural inflection becomes trapped inside of derivation, resulting in enclisis.

(6)

  • Siénte -n -se!

  • sit.IMP -PL -CL.REFL

  • ‘Sit down! (imperative plural)’

Let us suppose that reflexive se, as a reducer of argument-structural valency, can be considered a derivational morpheme. According to the morphotactic constraint in (5), this should become the displacement form in (7), where the inflectional plural marker follows the clitic.

(7)

  • Siénte -se -n!

  • sit.IMP -CL.REFL -PL

Indeed, such forms are synchronically attested. But Haspelmath points out that “[l]anguage change must be gradual, otherwise innovating speakers would not be understood by conservative speakers” (p. 302). Thus, as part of the transition from siénte-n-se to siénte-se-n, “speakers have no choice but to create hybrid forms” like siénte-n-se-n in (8), as “innovations can take only one step at a time, so hybrid forms are necessary” (p. 302).

(8)

  • Siénte -n -se -n!

  • sit.IMP -PL -CL.REFL -PL

Note that in fact the hybrid form in (8) would allow (5) to be existentially satisfied, as there is at least one instance of a derivational morpheme -se before an accompanying inflectional morpheme -n.

We could thus frame (5) as “For any derivational morpheme d and inflectional morpheme i, at least one surface occurrence of d must be closer to the root than a surface occurrence of i.” The intuition, therefore, is that hybrid forms allow one to have one’s morphotactic cake and eat it too: there is one instance of the -n that se precedes in (8). However, Haspelmath (1993:303) leaves the following question open: “How do speakers get rid of the residual, nonfunctional internal inflection? . . . Some details of the final cleaning up remain to be accounted for.” In this article, therefore, we seek an explanation that can account for the following three properties, which we view as intimately related and to be mechanistically unified:

(9) Explananda for the innovation of displacement and hybrid doubling forms

  • The morphotactic that is violated by the old forms (6)

  • The creation of hybrid doubling forms as a response to the morphotactic (8)

  • The one-step innovation that leads to eventual displacement (7)

As (9) makes clear, we view displacement and doubling as morphological phenomena—and specifically as postsyntactic morphological phenomena, motivated by morphotactic constraints such as (5) and others that we will develop. Not all models of this phenomenon treat it as morphological; in particular, Kayne (2010) and Manzini and Savoia (2011) present a syntactic view. We contend that once all of the generalizations and restrictions on this phenomenon are considered, however, the appeal of a purely syntactic approach is weakened.

We thus turn in greater detail to the Spanish case at hand. Importantly, the mesoclisis phenomenon in question occurs not only with the reflexive marker se, but also with a host of other pronominal clitics in the inventory, such as third person singular masculine accusative lo and singular dative le. A great deal of dialectological and descriptive work has looked at the existence and limits of mesoclisis in Spanish cases such as the following interpretation of the paradigm in (6)–(8), where > indicates a postulated diachronic development and, correspondingly, a degree of divergence from the standard, prescriptive variety:

(10) siénte-n-se > siénte-n-se-n > siénte-se-n

Note, however, that the mesoclisis in question jumps over, or involves, only the plural -n, and not other stem-final instances of this segment that are, say, part of the verb root (Harris and Halle 2005:202).

(11)

  • De -n -le eso! > {De-le-n/De-n-le-n} eso!

    give.IMP -PL -CL.3SG.DAT that

    ‘Give that to him! (imperative plural)’

  • Ten -le eso! > *{Te-le-n/Ten-le-n} eso!

    hold.IMP -CL.3SG.DAT that

    ‘Hold that for him! (imperative singular)’

Moreover, while the pronominal clitic lo can be affected, other phonologically identical instances are unaffected (Harris and Halle 2005:202).

(12)

  • Hága -n -lo mejor! > {Hága-lo-n/Hága-n-lo-n} mejor!

    do.IMP -PL -CL.3SG.M.ACC better

    ‘Do it better! (imperative plural)’

  • Hága -n lo mejor! > *{Hága lo-n/Hága-n lo-n} mejor!

    do.IMP -PL the.N best

    ‘Do the best thing! (imperative plural)’

Finally, the mesoclisis in question occurs with positive imperatives (which are enclisis environments), but not with negative imperatives, which are proclitic—hence, the clitic and -n are never contiguous to begin with.

(13)

  • No lo haga -n! > *No {lo-n haga/lo-n haga-n}!

  • not CL.3SG.M.ACC do.IMP -PL

  • ‘Don’t do it! (imperative plural)’

However, as we will discuss in section 4, the phenomenon exhibits many dialectal (and more likely, idiolectal) differences, particularly with respect to which clitics are involved. We should make clear here that geographic distribution, while important for fieldwork and description, particularly in knowing where one is most likely to elicit idiolects of the relevant type, is largely orthogonal from the point of view adopted here: grammar enables and limits a speaker’s possibilities, and the distribution of possible and impossible is determined by grammar; it is a matter of convention whether and which of the restricted possibilities might be more frequently accepted in a given community. Following Harris and Halle (2005) and Kayne (2010), we do not report on the geographical distribution of particular grammatical features (e.g., which geolects have mesoclisis and which do not, or which geolects have which dialectal variants of constraints on mesoclisis); rather, we report on the clustering of these features within particular dialects or idiolects. Nonetheless, we underscore the importance of geographically based description as a way of potentially testing the clustering of patterns within specific mesoclisis dialects.

The empirical generalizations that are of central concern here are the following:

(14) Generalizations over all Spanish mesoclisis varieties

  • It involves pronominal clitics only.

  • It involves the plural agreement marker -n only.

  • It occurs only in enclisis environments.

Harris and Halle (2005) (and related work preceding theirs, such as Minkoff 1993, Halle and Marantz 1994, and Harris 1998) brought the theoretical interest of a postsyntactic approach to Spanish mesoclisis to the attention of researchers debating the existence (and nature) of a postsyntactic component. Subsequently, Manzini and Savoia (2011) documented a number of important parallels in different varieties of Romance and Albanian that are spoken in Italy, exhibiting related phenomena. This research shows that mesoclisis in the imperative is far from being a quirk of Spanish. Data from the S. Marzano dialect of Arbëresh (a group of Albanian varieties spoken in southern Italy), where plural agreement has a very different phonotactic form (namely, -ni-), exhibit the same alternation between regular enclisis (15) and mesoclisis (16).

(15)

(16)

The restrictions that generate dialectal and idiolectal variation among Spanish speakers are at the heart of section 6, where the person features of the pronominal clitic may determine whether it is eligible for mesoclisis or not. Similarly, as the contrast in (15)–(16) shows, person features can determine whether mesoclisis applies in Italian dialects. While we focus mainly on Spanish, we return periodically to the Albanian and Italian dialects discussed by Manzini and Savoia (2011) where they help to constrain the analytic space of options in explaining restrictions on mesoclisis more generally.2

The structure of this article is as follows. Section 2 introduces the Generalized Reduplication formalism, which has the property of unifying doubling and displacement as alternative repairs to the same morphotactic. Section 3 provides background on the Spanish clitic and agreement system, which is relevant for understanding constraints on where mesoclisis occurs, and section 4 includes three explananda that govern mesoclisis: the person and case hierarchies, the plural number constraint, and the two-clitic hierarchy. Section 5 discusses Kayne’s (2010) syntactic analysis of mesoclisis (and how it fares with respect to these three explananda). Section 6 presents a revised morphotactic analysis based on Harris and Halle’s (2005) original proposal, with new constraints to address the explananda of section 4, and it integrates our account of mesoclisis within a more general analysis of the postsyntactic formation of clitic clusters in Spanish, which we argue is a cyclic process. Sections 78 conclude and outline avenues for future research.

2 The Harris-Halle Formalism and Its Properties

Harris and Halle’s (2005) formalism, which we called Generalized Reduplication (GR) in Arregi and Nevins 2012, was developed to account for kinds of partial reduplication found in phonology and morphology. Some of the specific initial motivation came from exceptions to Marantz’s (1982) generalization that suffixing partial reduplication copies from the right edge, while prefixing partial reduplication copies from the left edge. The known counterexamples include the following:

(17)

  • Absolutive singular reduplication in Chukchee

    nute-nut ‘earth, absolutive singular’

    (Marantz 1982:439)

  • Plural reduplication in Madurese

    wáq-búwáq-án ‘fruits’

    (Marantz 1982:451)

In the Chukchee case, reduplication is to the right, but skips a segment (e) at the right edge. In Madurese, reduplication is to the left, but skips segments ( ) at the left edge. In the GR formalism, these are treated in terms of special brackets that are interpreted in specific ways by the phonology.3

(18) Partial reduplication in the GR formalism

  • Repeat all material inside [ . . . ]:

    [A B] → ABAB

  • Delete the material after > in the second copy:

    [A > B] → ABAB → ABA

  • Delete the material before < in the first copy:

    [A < B] → ABAB → BAB

The notation in (18a) describes the standard cases falling under Marantz’s generalization, with no skipping. In Chukchee (17a), the representation [nut > e] for absolutive singular derives reduplication to the right, but skipping the rightmost segment: nute-nutenute-nut. In Madurese (17b), plural [ < wáq]-án derives reduplication to the left, but skipping the leftmost segmental material: wáq-búwáq-ánwáq-búwáq-án. Note that these angled brackets can be thought of as arrows, in the sense that their effect is also to indicate that [A > B] will teleologically “move” A to a rightward position, while [A < B] will teleologically “move” B to a leftward position (e.g., nut moves to the right in Chukchee, and wáq moves to the left in Madurese). This is the intuition behind having these arrows point in these directions, in fact.

One of the interesting properties of this specific formalism is what happens when the two kinds of arrows are combined. The effect of deleting A in the first copy and B in the second is wholesale metathesis.

(19)

  • Metathesis in the GR formalism

  • [A > < B] → ABAB → BA

Therefore, this formalism very closely links partial reduplication (henceforth, doubling) and metathesis. These processes in fact differ in only one angled bracket, as represented in the following statement of the Spanish mesoclisis rule, which we will call the GR rule, adapted from Harris and Halle 2005:4

(20) Generalized Reduplication in Spanish imperatives (to be revised in section 6)

  • SD: X Agr Cl Y, where Agr is [−participant, −singular].5

  • SC: Insert

    • [ to the immediate left of Agr

    • ] to the immediate right of Cl

    • > < to the immediate right of Agr (Displacement)

      or

      > to the immediate right of Agr (Doubling)

(21) Derivation of mesoclisis in (7)–(8)

  • siénte [n > < se] = siénte se n

  • siénte [n > se] = siénte n se n

Morphological metathesis and doubling rules, with a clearly postsyntactic nature, have been amply documented in the literature (see, e.g., Arregi and Nevins 2012: chap. 5, Myler 2013, Calabrese and Pescarini 2014, Smith 2014). Interestingly, although metathesis and reduplication are virtually never linked in phonology, numerous phenomena of this type are found in morphosyntax (e.g., the many cases of hybrid forms documented in Haspelmath 1993). While the GR formalism is closely related in spirit to Embick and Noyer’s (2001) operation of Local Dislocation, it has the advantage of tying together displacement/metathesis with doubling. Importantly, the GR formalism as shown above applies after Vocabulary Insertion, or, at the very least, one cannot tell if it is applying before. However, in Arregi and Nevins 2012:chap. 5 we document the existence of ergative metathesis alongside ergative doubling in Basque, and importantly, in the case of the latter the two sites can be occupied by different allomorphs, thereby demonstrating that metathesis applies before Vocabulary Insertion in this case.

While Harris and Halle (2005) do not comment on the morphotactic motivations for the mesoclisis phenomenon in (21), it is clear that they unknowingly have provided an implementation for exactly Haspelmath’s diachronic scenario and his desiderata for showing how the intermediate copy is eventually deleted. The GR formalism predicts that whenever morpheme displacement is found, doubling should also be found close by in time (i.e., diachronically) or space (dialectally). This seems confirmed in cases such as Lithuanian si mesoclisis, which shows doubling in addition (Nevis and Joseph 1992); movement of the English comparative morphemes -er/-est (which show doubling as in more better, most unkindest cut); and potentially a number of others as well. However, the extension of the phenomenon in (21) to Spanish clitics that seem to have no derivational character, yet cause inflection to move “outside” of them, suggests a closer look at the details and restrictions on this phenomenon, to which we turn in section 4. First, however, a close description of Spanish clitic and agreement morphology is in order.

3 Exponence and Syncretism in Spanish Clitics and Agreement

Crucial in understanding Spanish imperative mesoclisis are a number of facts concerning the exponence of pronominal clitics and agreement in Spanish. Both plural agreement -n and several instances of the clitic se involved in mesoclisis are second person morphemes that are syncretic with third person. Also relevant is the fact that certain clitics are polymorphemic, but others are monomorphemic. An explicit account of these facts and others is, we believe, an important step toward understanding the nature of the restrictions on mesoclisis discussed in the next section. This section provides an analysis of the postsyntactic exponence of clitic and agreement morphemes in non-Iberian Spanish,6 focusing on the details relevant to mesoclisis. A more thorough analysis of a wider range of dialects, including an account of the paradigm of second person strong pronouns, is to be found in online appendices A and B (http://www.mitpressjournals.org/doi/suppl/10.1162/ling_a_00286).

The analysis offered here is postsyntactic. It is, however, conceivable that syntactic accounts of the second-third person syncretism and the polymorphemic character of some clitics described here would be compatible with our postsyntactic account of mesoclisis in later sections. The main purpose of this section is thus not to claim that a postsyntactic analysis of these facts of exponence is superior to a syntactic one; rather, it is to give an explicit account that can provide a solid basis for our postsyntactic analysis of mesoclisis (though see appendix B for discussion of a potential syntactic account of the second-third person syncretism).

The paradigms of (non-Iberian) Spanish pronominal clitics and agreement, shown in table 1,7 display a number of syncretisms affecting distinctions across all relevant features of person, number, gender, case, and reflexivity. Furthermore, while some clitics are polymorphemic (e.g., accusative third person singular feminine l-a, third person plural dative le-s), others are not (third person singular dative le). Within the framework of Distributed Morphology adopted here (Halle and Marantz 1993), we assume that both clitics and agreement are syntactically specified for these features, and that the different syncretisms are due to the standard operation of underspecification in vocabulary entries (and the Elsewhere Principle) and to impoverishment rules. The focus of this section is syncretism due to the latter, and we provide a more complete account, including vocabulary entries, in appendix A. We furthermore assume that all clitics are syntactic atoms of category D, though some surface as polymorphemic, due to the effect of postsyntactic fission, discussed at the end of this section.8

Table 1
Pronominal clitics and finite agreement affixes in Spanish (non-Iberian dialects). (a) Nonreflexive clitics and agreement (in third person accusative, -o- forms are masculine, and -a- forms feminine). (b) Reflexive clitics.

The syncretisms most directly relevant to mesoclisis are these:

(22) Syncretisms in Spanish agreement and pronominal clitics

  • First and second person clitics are syncretic for reflexivity, case, and gender.

  • All second person plural forms are identical with third person plural.9

  • Third person dative clitics (unlike accusatives) are syncretic for gender.

  • The third person reflexive clitic is syncretic for case, gender, and number.

Certain parts of the paradigm covered by more than one of these generalizations make the interaction between these neutralizations nontrivial. In particular, second person plural clitics should be subject to (22a), but because of (22b), they do in fact manifest distinctions in reflexivity, case, and gender (table 1). Furthermore, (22b) and (22d) together cause second person plural reflexives to be realized by a number-neutral exponent, unlike other first and second person clitics (table 1b). This section provides an account of these neutralizations and interactions.

The postsyntactic operations responsible for these neutralizations are impoverishment rules that act on clitic and agreement formatives containing the following φ-features:

(23) Person features (Halle 1997, Harbour 2016)

  • First person: [+participant, +author]

  • Second person: [+participant, −author]

  • Third person: [−participant, −author]

(24) Number feature (Harbour 2003)

  • Singular: [+singular]

  • Plural: [−singular]

(25) Gender feature in clitics

  • Feminine: [+feminine]

  • Masculine: [−feminine]

Within clitics, datives are [+peripheral], and accusatives are [−peripheral] (features due to Calabrese 2008), and reflexivity is encoded by the feature [±anaphoric]. Clitics and agreement morphemes are distinguished by their category features, D and Agr, respectively. In addition, clitics are specified as [−strong], which distinguishes them from their strong pronominal counterparts, which are [+strong]. The feature [−strong] is thus responsible for defining the class of D elements that undergo syntactic cliticization, as well as any postsyntactic operations particular to pronominal clitics. In the remainder of this article, we will use the subscript Cl for elements specified as [−strong], so that pronominal clitics are DCl.

The metasyncretism observed in (22a) for first and second person clitics is due to the following impoverishment rule:10

(26) Participant Impoverishment

  • SD: [D, −strong, +participant, ±author, ±anaphoric, ±peripheral, ±feminine]

  • SC: Delete [±anaphoric, ±peripheral, ±feminine]

Second person plural clitics and agreement are syncretic with the third person (22b) due to the effect of a different impoverishment rule.11

(27) 2Pl Impoverishment

  • SD: [+participant, −author, −singular]

  • SC: [+participant] → [−participant]

By changing the value of the feature [±participant] from positive to negative, second person plural is exponed in the same way as third person plural. As shown in table 1, these impoverished second person morphemes behave like third person not only in the sense that they have the same form as third person, but also in that they express featural distinctions in gender, case, and reflexivity available only to the third person. This entails that 2Pl Impoverishment precedes Participant Impoverishment, as the former rule bleeds neutralization of reflexivity, case, and gender features in the participant morphemes it applies to.

(28)

  • Order of postsyntactic rules (to be revised)

  • 2Pl Impoverishment > Participant Impoverishment

The exponence of second person plural morphemes is discussed below, together with the third person forms they are syncretic with.12 These syncretisms figure prominently in our description and analysis of mesoclisis in Spanish second person plural imperatives in sections 4 and 6; in particular, second person plural clitics and agreement (i.e., -n) are treated as nonparticipant (third person) by mesoclisis, a consequence of the independently motivated rule of 2Pl Impoverishment.

Also relevant to mesoclisis is the fact that, although third person clitic forms make distinctions in reflexivity, case, number, and gender, they do undergo some neutralizations, described in (22c) and (22d) and accounted for by two separate impoverishment rules.

(29) Dative Impoverishment

  • SD: [D, −strong, −anaphoric, −participant, −author, +peripheral, ±singular, ±feminine]

  • SC: Delete [±feminine]

(30) Reflexive Impoverishment

  • SD: [D, −strong, +anaphoric, −participant, −author, ±peripheral, ±singular, ±feminine]

  • SC: Delete [−participant, −author, ±peripheral, ±singular, ±feminine]

Dative Impoverishment accounts for the fact that gender contrasts available in accusative clitics are absent in dative clitics (see table 1a). In reflexive clitics, feature neutralization is more extensive, as (30) deletes person, case, number, and gender (cf. tables 1a and 1b). As a result of this rule, third person reflexives are realized by the default clitic exponent se (e.g., Bonet 1991, 1995, Halle and Marantz 1994, Harris 1995, Nevins 2007), as shown in more detail in appendix A.

A different type of neutralization can be observed in clitic clusters. In the context of a third person accusative clitic, third person dative clitics are realized as se, a phenomenon known as spurious se (Perlmutter 1971). As illustrated in (31), when nonreflexive third person dative clitics (singular le, plural les) are not clustered with another clitic, their form is distinct from that of their reflexive counterpart se.

(31) Third person dative clitics

  • María se dio un libro.

    María CL.REFL gave a book

    ‘María gave a book to herself.’ (Ungrammatical in the nonreflexive reading)

  • María {le / les} dio un libro.

    María {CL.3SG.DAT / CL.3PL.DAT} gave a book

    ‘María gave a book to him/her/them.’ (Ungrammatical in the reflexive reading)

When they are clustered with an accusative clitic, the distinction is neutralized in favor of the reflexive.

(32) Spurious se

  • *María {le / les} lo dio.

    María {CL.3SG.DAT / CL.3PL.DAT} CL.3SG.M.ACC gave

    ‘María gave it to him/her/them.’

  • María se lo dio.

    María CL.3.{REFL/DAT} CL.3SG.M.ACC gave

    ‘María gave it to herself/him/her/them.’

This is due to the following impoverishment rule, inspired by Bonet 1991, 1995, Halle and Marantz 1994, and Nevins 2007:

(33) Spurious se Impoverishment

  • SD: Cl1 specified as [D, −strong, +peripheral, −participant, −author, ±singular] and Cl2 specified as [D, −strong, −peripheral, −participant, −author]

  • SC: Delete [−participant, −author, ±singular] in Cl1

Similar to the impoverishment of third person reflexives (whether in clusters or not), this impoverishment results in all dative clitics being realized with the default exponent se when clustered with accusative clitics.

Due to the overall neutralization between second and third person in the plural, second person plural clitics are also subject to these neutralizations that are otherwise particular to the third person. This entails that 2Pl Impoverishment is in a feeding relation with Dative, Reflexive, and Spurious se Impoverishment:

(34)

  • Order of postsyntactic rules (to be revised)

  • 2Pl Impoverishment > Participant, Dative, Reflexive, and Spurious se Impoverishment

The absence of gender in dative clitics, as well as of person and number in reflexives and spurious datives, due to the joint action of these impoverishment rules, is an important ingredient of our account of constraints on mesoclisis and variation thereof, as described in the following section.

Finally, as is evident in table 1, all plural clitics contrast with their singular counterparts and with number-neutral clitics (e.g., reflexive se), in that they are polymorphemic; that is, their [−singular] feature is realized by means of a separate plural exponent -s. Similarly, accusative third person and second person plural clitics not otherwise affected by Reflexive Impoverishment are the only ones that maintain gender features after impoverishment, and these gender features are expressed by separate gender-specific exponents (masculine -o and feminine -a). We propose that this agglutinative effect is due to fission, triggered by the following feature cooccurrence restrictions:13

(35) Constraints on joint exponence of φ-features

  • *[±participant, ±feminine] (no joint exponence of person and gender)

  • *[±participant, −singular] (no joint exponence of person and plural)

  • *[±feminine, −singular] (no joint exponence of gender and plural)

These constraints ensure mutual separate exponence of person, gender, and plural number, by triggering fission of morphemes in which they cooccur into separate nodes. For instance, like other plural clitics lacking gender, nonreflexive dative plural le-s undergoes fission splitting [−participant] and [−singular], triggered by constraint (35b) (the clitic lacks gender, due to Dative Impoverishment).

(36)

graphic

In contrast, singular le remains monomorphemic, since it is not [−singular] (and also lacks gender). In addition, nonreflexive accusative clitics retain gender features, and their plural counterparts thus violate all constraints in (35), triggering fission into three separate nodes, as illustrated in (37) in the feminine.14

(37)

graphic

As is evident in the examples above, we follow Arregi and Nevins 2012:132–136 in assuming that copies of all features not affected by the triggering constraint are present in each of the morphemes in the outputs of fission. In particular, the categorial D feature present in the input, as well as others (e.g., [−strong], represented with the subscript Cl above), is present in all the morphemes that result from fission.

Fission applies after impoverishment.15

(38)

  • Order of postsyntactic rules (to be revised in appendix A)

  • 2Pl Impoverishment >

  • Participant, Dative, Reflexive, and Spurious se Impoverishment > Fission

This ordering accounts for the fact that the number of exponents in a clitic correlates with the number of featural distinctions it makes. This is illustrated in the following examples (for details of Vocabulary Insertion, see appendix A):

(39)

graphic

(40)

graphic

Crucially for our discussion below, clitics differ in the amount of internal structure they have, since fission applies to them differently depending on their feature content. Thus, it partitions the set of clitics in two different ways. First, all postsyntactically plural clitics are polymorphemic, while others are not necessarily so (e.g., dative plural le-s vs. dative singular le and number-neutral reflexive se). Second, third person (nonreflexive) accusatives are the only clitics whose gender features are not deleted by impoverishment, so they are always polymorphemic (l-o(-s), l-a(-s)), contrasting with all others (e.g., dative le(-s)). Mesoclisis is variably sensitive to this internal structure, which accounts for some of the variation in the constraints on the phenomenon discussed in the next section.

In addition to the clitics’ internal structure, the postsyntactic account above derives featural properties of certain morphemes that are relevant for mesoclisis triggered by plural agreement -n in imperatives. As part of the more general neutralization between second and third person plural, second person plural agreement -n is third person plural at the point of the derivation at which it is displaced/doubled. In addition, reflexive and spurious se, regardless of their semantically motivated feature content (i.e., second or third person, singular or plural), have no person or number features postsyntactically. These syncretisms, and the impoverishment rules that account for them, are crucial in understanding some of the constraints on mesoclisis, to which we now turn.

4 Three Crucial Explananda in Spanish Mesoclisis

The literature describes several restrictions on mesoclisis in Spanish. Here, we concentrate on three such restrictions, which we analyze in later sections. We begin with constraints involved in some of the variation displayed by mesoclisis: while some dialects allow it only with nonthird person clitics, others also allow it with third person; in addition, some make a further cut among third persons, between dative and accusative. The second restriction, which applies to all dialects, bans mesoclisis of plural clitics. Finally, we discuss a restriction on mesoclisis in clusters with more than one clitic.

4.1 The Person and Case Hierarchies

As discussed in the dialectal literature reviewed by Harris and Halle (2005), in particular Rosenblat 1946:229–232 and Kany 1951:112–114, not all clitics pattern equally in terms of their ability to undergo mesoclisis. In what follows, we concentrate more on the morphosyntactic generalizations themselves than on the specific geographic distributions, noting however that in general, Iberian Spanish is more restrictive than Latin American varieties. The generalizations are as follows. While some speakers allow mesoclisis for all clitics, including third person singular lo, la (accusative masculine and feminine, respectively), and le (dative, syncretic for gender), the more restrictive speakers allow it only for the first person singular me and for (reflexive or spurious) se. This variation is illustrated in the following paradigm, where “%” indicates that the mesoclitic form is not allowed by the more restrictive speakers:

(41)

  • Siénte -(n) -se -n!

    sit.IMP -PL -CL.REFL -PL

    ‘Sit down! (imperative plural)’

  • Vénda -(n) -se -n -lo!

    sell.IMP -PL -CL.DAT -PL -CL.3SG.M.ACC

    ‘Sell it to them! (imperative plural)’

  • Vénda -(n) -me -n eso!

    sell.IMP -PL -CL.1SG.DAT -PL that

    ‘Sell me that! (imperative plural)’

  • %Vénda -(n) {-lo / -la} -n!

    sell.IMP -PL {-CL.3SG.M.ACC / -CL.3SG.F.ACC} -PL

    ‘Sell it! (imperative plural)’

  • %Vénda -(n) -le -n eso!

    sell.IMP -PL -CL.3SG.DAT -PL that

    ‘Sell him/her that! (imperative plural)’

This dialectal split can be described naturally in terms of a person hierarchy: l- clitics are third person, but first person singular me is not, and neither is se: in any of its uses, a clitic that will surface as se is subject to either Reflexive or Spurious se Impoverishment, which deletes its person features (among others). Thus, nonthird person clitics outrank third person clitics in the hierarchy of accessibility to mesoclisis.

(42)

  • Mesoclitic person hierarchy: Nonthird person over third person

  • se, me > lo, la, le

Any speaker who allows mesoclisis with the lower part of the hierarchy also allows it with the higher part. The phenomenon to be understood is how to capture the restriction that, for some speakers, third person clitics cannot undergo mesoclisis. Clearly, nothing in the GR formalism as it stands would block this, which was one of Kayne’s (2010) criticisms of Harris and Halle 2005. We agree that any adequate account of the phenomenon must cover this, and in section 6.5 we demonstrate how such restrictions can be built into the formulation of mesoclisis within the GR formalism.

In addition to the basic cut in terms of nonthird person vs. third person, there is a hierarchy within the third person clitics, related to case. It seems that some speakers allow mesoclisis only with reflexive se, and, among third person clitics, some speakers can mesocliticize dative le, but not accusative lo, la. The datives, of course, do not show gender distinctions and, as shown in section 3, this means that they are structurally less complex. (Regarding our implementation in structural terms, see section 6.5.) For such speakers, a more articulated hierarchy would be characterized as follows:

(43)

  • Mesoclitic case hierarchy: Dative third person over accusative third person

  • le > lo, la

While this hierarchy does not cut across persons, it does cut within persons, specifically between dative and accusative, as the latter clitics show additional complexity in bearing a gender distinction.

Two further observations on these hierarchies are in order before we turn to other restrictions on mesoclisis. First, the person hierarchy does not include second person singular te. The reason is simple: the imperative forms with mesoclisis have a second person plural subject, but second person singular clitics are universally banned from cooccurring with second person plural subjects (a ban we assume is due to Condition B of binding theory). This restriction applies independently of mesoclisis and is found in both both enclitic and proclitic environments.

(44)

  • *Vénda -n -te eso!

    sell.IMP -PL -CL.2SG.DAT that

    ‘Sell yourself that! (imperative plural)’

  • *Te vendieron eso.

    CL.2SG.DAT sell.PST.2PL that

    ‘You (plural) sold yourself that.’

Also absent from the person hierarchy is Iberian Spanish second person plural colloquial os, described in appendix A. Two properties of Spanish grammar conspire to make it impossible to test the mesoclitic properties of os. First, it cannot be combined with plural imperative -n.

(45)

  • *Vénda -n -os eso!

  • sell.IMP -PL -CL.2PL.DAT that

  • ‘Sell us that! (imperative plural)’

Since in Iberian Spanish -n in its second person plural use is restricted to formal environments, it clashes with the use of colloquial os (quite generally, the addressee must be referred to consistently with formal or colloquial forms, at least within a sentence). Second, the Iberian second person plural colloquial counterpart of -n in imperatives, -d, does not participate in mesoclisis, regardless of what clitic is present.

(46)

  • Vende -d -me eso!

    sell.IMP -2PL.COL -CL.1SG.DAT that

    ‘Sell me that! (imperative colloquial plural; Iberian Spanish)’

  • *Vende -(d) -me -d eso!

    sell.IMP -2PL.COL -CL.1SG.DAT -2PL.COL that

    ‘Sell me that! (imperative colloquial plural; Iberian Spanish)’

Therefore, os is banned from mesoclisis environments for reasons that are independent from the clitic hierarchy.

4.2 The Ban on Plural Mesoclitics

The next restriction, universal across mesoclitic varieties, is that plural clitics (first person plural nos, third person dative les, and third person accusative los, las) uniformly do not undergo mesoclisis.16

(47)

  • Sírva -n -nos eso!

    serve.IMP -PL -CL.1PL.DAT that

    ‘Serve us that! (imperative plural)’

  • *Sírva -(n) -nos -n eso!

    serve.IMP -PL -CL.1PL.DAT -PL that

(48)

  • Sírva -n -les eso!

    serve.IMP -PL -CL.3PL.DAT that

    ‘Serve them that! (imperative plural)’

  • *Sírva -(n) -les -n eso!

    serve.IMP -PL -CL.3PL.DAT -PL that

(49)

  • Sírva -n -los!

    serve.IMP -PL -CL.3PL.M.ACC

    ‘Serve ’em! (imperative plural)’

  • *Sírva -(n) -los -n!

    serve.IMP -PL -CL.3PL.M.ACC -PL

Harris and Halle (2005) claim that the ban on plural mesoclisis is phonotactic in nature, since the resulting consonant cluster sn# is banned by Spanish phonology. We do not think this is a valid explanation of the facts.

First, existing phonotactic repairs do not step in to fix the banned mesoclitic examples. Spanish does not allow homosyllabic consonant clusters starting with s, and such underlying clusters are typically repaired by epenthesis, as in stop [estop] ‘stop sign’ and escribir ‘write’, from underlying [skribir] (cf. in-scribir ‘engrave’, syllabified as [ins.kri.βir]). However, *sírva(n)losen, *sírva(n)losne are just as ungrammatical as *sírva(n)losn as mesoclitic variants of sírvanlos. Deletion and metathesis would result in n#, s#, or ns#, all of which are allowed by Spanish phonotactics (ns# is possible at least in the plural of some borrowed words, such as yens), but neither helps in repairing *sírva(n)losn: *sírva(n)lon, *sírvalos, and *sírva(n)lons are not possible as mesoclitic variants of sírvanlos (the latter is of course grammatical, but it is simply the enclitic form, not the result of mesoclisis with deletion of word-final n).

Second, the ban on plural mesoclisis holds even in examples that do not result in sn#.

(50)

  • Vénda -n -nos -lo!

    sell.IMP -PL -CL.1PL.DAT -CL.3SG.M.ACC

    ‘Sell us it! (imperative plural)’

  • *Vénda -(n) -nos -lo -n!

    sell.IMP -PL -CL.1PL.DAT -CL.3SG.M.ACC -PL

As illustrated in this example (due to Rafael Nuñez-Cedeño, pers. comm.), a plural clitic makes mesoclisis ungrammatical even when plural -n would not be adjacent to it in a clitic cluster. This suggests that a morphotactic rather than phonotactic explanation is in order.

Finally, note that the Italian and Albanian dialects have -ni and other nonsigmatic plural agreement exponents that may also show, if they are disallowed, that phonology alone is not at stake. As Manzini and Savoia (2007:234) remark, “The first person plural presents some asymmetries with respect to the first person singular. In particular, mesoclisis of ne (us) is not attested by the data and is given as dubious or impossible by native speakers” of S. Marzano (the question mark in (52), repeated from the source, indicates this judgment).

(51)

  • {siɟˈɟɔ / ˈŋgrigə} (m)mə ni

  • {wake.up / pick.up} CL.1SG.ACC 2PL

  • ‘Wake/Pick me up! (imperative plural)’

  • (S. Marzano Arbëresh; Manzini and Savoia 2007:235)

(52)

(53)

  • {siɟˈɟɔ / ˈŋgrig} (n)ni {ɛ / i}

  • {wake.up / pick.up} 2PL {CL.3SG.ACC / CL.3PL.ACC}

  • ‘Wake/Pick him/them up! (imperative plural)’

  • (S. Marzano Arbëresh; Manzini and Savoia 2007:235)

As (53) shows, the normal position of clitics in imperatives is enclitic. Yet mesoclisis is possible with first person singular (51) but not first person plural (52) clitics, even though there is no potential phonological problem with (52), as the plural clitic in question ends in a vowel. We therefore seek a more principled reason why the GR formalism should be blocked from effecting mesoclisis with plural clitics.

Before we continue, it is important to note that the ban on plural mesoclisis does not extend to reflexive se, even though in plural imperatives it is syntactically and semantically plural.

(54)

  • Siénte -(n) -se -n!

  • sit.IMP -PL -CL.REFL -PL

  • ‘Sit down! (imperative plural)’

However, as was shown in section 3, se is not plural when GR applies; as Reflexive Impoverishment has applied earlier, se is syncretic for number.

4.3 The Two-Clitic melon Hierarchy

According to Harris and Halle (2005), dialectal variation is also found when there is more than one clitic, in terms of the placement of plural -n with respect to each clitic. Some speakers allow -n to appear after either clitic within the cluster.

(55)

  • Dé -n -me -lo!

    give.IMP -PL -CL.1SG.DAT -CL.3SG.M.ACC

    ‘Give me it! (plural imperative)’

  • dé(-n)-me-n-lo, dé(-n)-me-lo-n

Other speakers allow -n only after the first clitic.

(56) dé(-n)-me-n-lo, *dé(-n)-me-lo-n

However, no speakers allow -n only after the second clitic.

(57) Unattested: dé(-n)-me-lo-n, *dé(-n)-me-n-lo

In other words, a grammar that allows -n after the second clitic in a cluster implies that it allows it after the first one as well. Our aim in section 6.3, therefore, will be to develop an account of the restrictions on mesoclisis that derive this particular implicational generalization.

Before proceeding, we note that the above data from Harris and Halle are not sufficient to establish the empirical generalization. In particular, the speakers who do not allow mesoclisis of both clitics in the cluster me lo might simply be speakers who do not allow mesoclisis with lo in the first place, due to the person hierarchy. That is, the patterns of variation shown above might simply be due to the person hierarchy, not to a restriction specifically applying to clusters. The generalization could be confirmed by speakers who do allow mesoclisis with lo, but not in clusters in which lo is the second clitic. Unfortunately, we do not have direct access to a diverse enough range of speakers who allow mesoclisis with lo, so we have not been able to confirm this. Although we tentatively take the restriction on clitic clusters as a valid generalization, we note that more fieldwork is needed in this particular area.17

5 Kayne’s (2010) Approach and the Three Explananda

Kayne (2010) offers a syntactic reinterpretation of Harris and Halle 2005, which attempts to do away with any postsyntactic mechanisms and to account for displacement (and doubling) entirely with syntactic structures and processes. At the outset, Kayne points out a number of places where Harris and Halle’s account overgenerates; he argues that it could potentially metathesize any segment, or any morpheme. As we will make clear in section 6, our morphotactic formalization of mesoclisis as being specifically triggered by a Noninitiality constraint on plural -n provides a motivating force for these operations. Syntactic movement, too, is a “free” operation, but in practice is limited to cases where its application is motivated. Once a motivation for the application of GR rules yielding mesoclisis is in place, these particular arguments of Kayne’s therefore dissolve.

More generally speaking, there are two basic types of arguments that an entirely syntax-based approach to displacement and doubling makes against a postsyntactic analysis. The first turns on the apparent redundancy between syntactic and morphological movement. While indeed morphological displacement may look like syntactic movement, we would argue that the two operate on different kinds of structures, for different reasons: syntactic movement is upward and involves the derivational requirements of either the mover or the target in creating a new hierarchical relation, while morphological displacement (and doubling) is due to the language-specific requirements on the linear order of elements. The second type of argument involves the apparent unconstrainedness of the reduplication formalism, but we hope to show that in fact the kinds of restrictions one needs to place on the syntactic account look even stranger within the syntax than they do in the morphology.

To further develop the latter point, we turn to the basic outline of Kayne’s analysis. It begins by analogizing mesoclisis with multiple agreement, of the kind where more than one verbal element shows evidence of agreement with the same DP, as in (58) (see also Alcázar and Saltarelli 2010).

(58)

  • Maria è stata lodata.

  • Maria is been.F.SG praised.F.SG

  • ‘Maria has been praised.’

  • (Italian; Kayne 2010:148)

Kayne’s argument, therefore, is that the multiple instances of -n found in Spanish doubling cases like vénda-n-lo-n (or those in (55b)) involve multiple, independent projections of Agr within the clausal spine. But what about the placement of the clitic itself, sandwiched between the verbal stem and the Agr projection? Kayne argues that this is the result of a sort of clitic climbing, to a position higher than the lowest Agr. Kayne’s empirical motivation comes from the observation that the person and case hierarchies, discussed in section 4.1, parallel the linear order of clitics in a clitic sequence, as se and me appear further to the left (and by hypothesis, higher) than third person clitics. The parallel can be seen in the following Ligurian example, in which first person me precedes the negative element n, whereas third person le follows it:

(59)

  • U me n le darà nent.

  • CL.3SG.NOMCL.1SG.DATNEGCL.3SG.ACC give.FUT.3SGNEG

  • ‘He won’t give it to me.’

  • (Ligurian; Parry 1997:244)

Kayne proposes that as in Ligurian, where the first person singular clitic can move above negation but the third person one cannot (at least not the accusative one; subject clitics are typically in a higher domain to begin with), in the relevant Spanish varieties first person clitics can move above plural Agr, yielding mesoclisis, while third person ones cannot. He makes the parallel explicit: “[(59)] is very much like what we see in [(60)], modulo the difference between the plural morpheme -n and the negative morpheme n” (2010:156).

(60)

  • De -me -n -lo!

  • give.IMP -CL.1SG.DAT -PL -CL.3SG.M.ACC

  • ‘Give me it! (imperative plural)’

The potential unification of clitic ordering within mesoclisis and enclisis is a promising aspect of the analysis, though we will return to it in section 7.1. (Our own explanation of the person hierarchy is found in section 6.5.) For the second aspect of Spanish mesoclisis—its restriction to singular clitics—Kayne appeals to phonological factors (as do Harris and Halle (2005)), an appeal we have argued to be insufficient (see section 4.2). Perhaps more telling is the fact that a purely syntactic analysis of this restriction would be hard pressed to find consistent examples of singular clitics moving higher than their plural counterparts within Romance clitic sequences.

Another parallel that Kayne draws (p. 147) to develop the syntactic analysis involves the suggestion that mesoclisis is like clitic climbing, which can also exhibit doubling in nonstandard varieties, as in (61).

(61)

  • Juan lo quiere hacer -lo.

  • Juan CL.3SG.M.ACC wants do.INF -CL.3SG.M.ACC

  • ‘Juan wants to do it.’

However, an important difference between the two configurations is that clitics always have to climb together in Spanish (62), but they don’t necessarily mesocliticize together (cf. (60)).

(62)

  • Juan me lo quiere dar.

    Juan CL.1SG.DATCL.3SG.M.ACC wants give.INF

    ‘Juan wants to give it to me.’

  • *Juan me quiere dar -lo.

    Juan CL.1SG.DAT wants give.INF -CL.3SG.M.ACC

Kayne in fact acknowledges this difference (p. 164n38). Similarly, there are no dialects of Spanish that exhibit restrictions—in terms of person or number—such that only a subset of the clitics can climb. The potential appeal to mesoclisis as being parallel to clitic climbing is thereby weakened, as arguably the latter is more directly syntactic.

Of perhaps more immediate concern is the relationship between displacement and doubling within Kayne’s approach. Displacement results when the clitics move above Agr, whereas doubling involves two Agr projections. According to Kayne, “[E]ach -n is merged as an independent morpheme in the ordinary syntax” (p. 160). For the derivation of cases such as vénda-n-lo-n, the relevant steps, as culled from Kayne’s paper, are as follows:

(63)

  • The lower -n induces movement of vénda to its immediate left.

  • The higher -n is merged subsequently and induces movement of the object clitic.

  • The higher -n then triggers movement of the verb phrase to its left, yielding vénda-n-lo-n.

In an analysis with two separate Agr (-n) morphemes, certain problems arise in trying to account for dependencies between the two Agr positions. Given the fact that both vénda-n-lo and vénda-lo-n are possible, it must be the case that either Agr position can be optionally absent. However, vénda-lo as a plural imperative is ungrammatical, a fact that goes unexplained if the optionality of the two Agr morphemes is not linked. Moreover, as we have already noted, there can be cases with three instances of -n: for example, vénda-n-me-n-lo-n.

In fact, under a purely syntactic account, the plural restriction (e.g., *vénda-n-nos-lo-n in (50b)) is problematic, because to rule it out, the account would have to ban generation of the higher (i.e., rightmost) -n only when the clitic moving above it is plural. However, the lowergenerated Agr does not seem to mind what the clitic’s number is, as cases of enclisis demonstrate (e.g., vénda-n-nos-lo in (50a)).

The dependency (and lack of independence) of these Agrs is found precisely in that they reflect the plurality of the same argument. Kayne’s parallel between Spanish n-doubling and Romance languages with multiply agreeing auxiliary-participle constructions such as (58) founders because the latter often exhibit agreement with different arguments as shown in (64) for French.

(64)

  • Paul les a repeintes.

  • Paul CL.3PL.ACC have.3SG repainted.PL

  • ‘Paul has repainted them.’

  • (French; Kayne 1989:85)

In this example, the auxiliary agrees with the singular subject argument, while the participle agrees with the plural object clitic. The Spanish cases, however, always show agreement only with the subject; the example in (58) is a passive, and not fully representative. The multiple instances of Spanish -n, we contend, show interdependence because they reflect one single syntactic element, multiply copied only in the postsyntax.

For a final argument that the Agr morphemes are not independent, consider yet another dependency that arises, namely, between the number of -n affixes and the number of clitics. There can be at most x + 1 instances of -n in a sequence with x clitics (e.g., no dialect has *vénda-n-me-n-lo-n-n). Similarly, every instance of -n must be right-adjacent to the verb or to a clitic; *vénda-n-me-lo-n-n is not possible. Though it may be tempting to dismiss such cases as being ruled out by phonology, we note that, to our knowledge, their would-be S. Marzano counterparts with -ni are equally unattested. In sum, the free generation of Agr morphemes to account for Mmultiple instances of -n in Kayne’s syntactic account is much more unrestricted than it may seem.

While the purely syntactic analysis is unsatisfactory to account for the range of restrictions on Spanish mesoclisis, Kayne’s paper nonetheless makes an important contribution to the discussion on this topic as it calls for a more developed version of Harris and Halle 2005 that provides motivation for and insight into the patterns of attested and unattested interspeaker variation. We now turn to providing such a version. Our account is formulated within the postsyntax and based on the intuition that the multiple instances of -n and their placement reflect specific aspects of the GR rule that acts during morphotactic repair.

6 A Morphotactic Formulation of Constraints on Mesoclisis

We agree with Kayne (2010) that Harris and Halle’s (2005) GR as such is just a formalism that does not account for the restrictions described above. However, we argue that, when supplemented with morphotactic constraints, a postsyntactic analysis based on GR can provide an insightful analysis of all the relevant facts of mesoclisis. We propose that mesoclisis arises for the following reason:18

(65)

  • Mesoclisis as a second-position effect

  • -n is a second-position clitic within the post-stem clitic domain. Displacement or doubling of -n occurs in order to put a clitic to its left.

While a Derivation inside Inflection constraint may have been the original source of mesoclisis (recall Haspelmath’s (1993) proposal in section 1), we propose that (65) is what is synchronically active and that it is more accurate than Derivation inside Inflection, since mesoclisis affects clitics such as first person plural me, which are clearly not derivational.

In what follows, we implement this idea in more formal terms, as part of a more general analysis of the postsyntax of clitic clusters, and we account for the restrictions on mesoclisis in terms of morphotactic constraints. Section 6.1 formalizes our approach to the syntax of clitic clustering, independently of mesoclisis. Section 6.2 implements (65) in terms of a Noninitiality constraint that triggers mesoclisis, and demonstrates how displacement and doubling are repairs to the constraint. Sections 6.36.5 provide restrictions on mesoclisis that derive the three explananda in section 4.

6.1 Clitic Clustering

Following Kayne 1991 and much subsequent work, we assume that Romance pronominal clitics surface adjoined to certain functional heads in the clausal spine. We implement here a version of Terzi’s (1999) specific proposal, according to which clitics adjoin to a head generated immediately above functional projections associated with verbal inflectional material, notated here as Cl.19

(66)

graphic

In finite contexts, verb movement to Agr derives proclisis. In imperatives (as well as all nonfinite contexts in Spanish), proclisis is the result of verb movement to a high-peripheral position above the clitic-hosting head (e.g., Kayne 1991, Rivero 1994). In Terzi’s (1999:93–99) specific implementation, which we adopt here, the verbal complex first adjoins to the clitic in Cl; then, the resulting complex head including the clitic moves to its higher surface position.20 In the case of imperatives, the complex head in Cl has the structure in (67b).21

(67)

graphic

In the rest of the article, we ignore the internal structure of the constituent dominated by the highest T node and omit the Cl projections. Clitic clusters are formed by the adjunction of the dative clitic to the accusative clitic; thus, a form with multiple cliticization has the structure in (68b) (Terzi 1999:99–108).22

(68)

graphic

In these syntactic representations, the clitic cluster does not form a constituent separate from its host. In the postsyntactic component, however, Clitic Clustering makes a constituent out of adjacent clitics.

(69)

  • Clitic Clustering

  • A node specified as [−strong] (a clitic) is rebracketed to form a constituent with a node specified as [−strong] that it c-commands and that is to its immediate left.

This rule performs the rebracketing that leads to new sisterhood relations, whereby clitics become a cluster (see Williams 2003:chap. 8 for a related set of reassociating operations). It includes the specific statement about c-command so that in a complex clitic node (such as the one constituting l-o), the entire clitic subtree l-o undergoes rebracketing with me, as opposed to just the single clitic node l. We assume that the operation of rebracketing is cyclic, meaning that it potentially applies multiple times, starting with the two innermost (i.e., most deeply embedded) clitics.23 Though examples with multiple clitics meet the structural description of the rule, examples with a single clitic do not.

(70)

graphic

Thus, our specific proposal is that, after Clitic Clustering, the clitics and their verbal host form a constituent (rooted at the top DCl node in (70)), but that within this higher constituent, the clitics form a subconstituent (rooted at the boxed DCl node in (70)), which we will refer to as the clitic cluster.

A number of stress-related facts provide evidence that clitics form a separate postsyntactic constituent within the larger constituent that contains both the clitics and their verbal host. First, proclitics are not part of the domain over which stress is computed. This can be seen most clearly in the present tense, where stress is always on the penultimate syllable.

(71)

  • Vende. [bénde]

  • sells

  • ‘She/He sells (things).’

As might be expected, the only exception to this generalization is monosyllabic forms, in which stress is on the only syllable in the verb.

(72)

  • Va de copas. [bá ðekópas]

  • goes of glasses

  • ‘She/He goes out drinking.’

This is true even in cases in which proclitics could in principle provide enough syllabic space for penultimate stress.

(73) Se va de copas. CL.REFL goes of glasses ‘She/He goes out drinking.’

  • [seβá ðekópas]

  • *[séβa ðekópas]

This is expected if proclitics are not part of the domain over which word-level stress is assigned in verbs. Stress-related evidence for treating enclitic clusters as a separate constituent can be seen in their interaction with the three-syllable window restriction on stress in Spanish, according to which stress cannot be further to the left than the antepenultimate syllable.24 In second person singular imperatives, stress is on the penultimate syllable of the verb.

(74)

  • Venda eso! [bénda éso]

  • sell.IMP that

  • ‘Sell that! (formal singular imperative)’

Stress remains on that syllable in the presence of enclitics, even in cases where the result violates the three-syllable window restriction.

(75)

  • Vénda -me eso! [béndame éso]

    sell.IMP -CL.1SG.DAT that

    ‘Sell me that! (singular imperative)’

  • Vénda -me -lo! [béndamelo]

    sell.IMP -CL.1SG.DAT -CL.3SG.M.ACC

    ‘Sell me it! (singular imperative)’

In addition, for many speakers the clitic cluster (even in cases with a single clitic) has its own (secondary) stress (Roca 1986, Harris 1991).

(76)

  • Alternative pronunciation for (75a)

    [béndamé éso]

  • Alternative pronunciation for (75b)

    [béndameló]

All these facts provide evidence that, within the verb-enclitic constituent, the clitic cluster forms a subconstituent separate from the verbal host postsyntactically.

6.2 Noninitiality

Our idea stated in (65) that mesoclisis is the result of a second-position clitic requirement on -n requires two more additions to the analysis. First, we propose that -n is subject to the following rule, which applies optionally in the postsyntactic component:25

(77)

  • n-Extradition

  • In a structure in which plural Agr -n is immediately c-commanded by DCl, insert

  • [−strong] in Agr (notated as AgrCl).

The intuition behind this rule is that it transforms the agreement suffix into a clitic-like morpheme in terms of its participation as a mobile element in the domain following the verb stem. This change from affix to clitic is relevant postsyntactically and reflects the more general reanalysis that yields the fluidity between agreement affixes and clitics (Fuß 2005); in the specific case at hand, it causes plural -n to participate in Clitic Clustering. Thus, as an effect of this rule, plural -n is optionally a [−strong] element—a clitic—in enclitic contexts, and once it participates in Clitic Clustering, it is subject to morphotactic constraints that hold of such sequences.

Second, the second-position requirement itself is formalized in terms of a Noninitiality constraint that governs the order of clitic -n within clusters.

(78)

  • Noninitiality

  • Not all instances of -n are initial in a clitic cluster.

This constraint is active throughout the postsyntactic component and automatically triggers mesoclisis whenever it is violated. In particular, it triggers the GR rule, which we reformulate as follows:26

(79) GR rule

  • SD: X AgrCl DClY, where AgrCl is [−participant, −singular] and AgrCl and DCl are sisters.

  • SC: Insert

    • [ to the immediate left of AgrCl

    • ] to the immediate right of DCl

    • > < to the immediate right of AgrCl (Displacement),

      or

      > to the immediate right of AgrCl (Doubling)

In the rest of this section, we illustrate the combined effect of n-Extradition, Clitic Clustering, Noninitiality, and the GR rule in deriving mesoclisis in both simple and complex clitic clusters.

Consider first clusters with a single pronominal clitic such as (67). If n-Extradition (77) does not apply, the syntactic structure derived in (67) is not altered postsyntactically, and the result is enclisis (vénda-n-lo). If n-Extradition does apply, clitic -n is clustered with lo.

(80)

graphic

The output, however, violates Noninitiality (78),27 which is repaired with either displacement or doubling by applying the GR rule (79).28

(81)

graphic

(82)

graphic

Thus, the optionality of mesoclisis is ultimately rooted in the optionality of n-Extradition, which makes -n pattern together with clitics. These examples, with a branching clitic l-o, also highlight the need for the sisterhood condition in the structural description of the GR rule (79), which requires the agreement and the pronominal clitics it applies to to be sisters. In the structures above, this ensures that the rule applies to the entire clitic l-o and not to its subconstituent clitic l- (which, being right-adjacent to -n, would otherwise meet the structural condition of the rule).

At this point, it is important to spell out three assumptions that are crucial in understanding the way in which the doubling structure in (82) satisfies Noninitiality (78). First, the constraint is crucially stated with the negative operator outscoping universal quantification over instances of -n, so that (82) is a good repair to Noninitiality because of cluster-final -n, despite the presence of cluster-initial -n. Second, clitic cluster in (78) simply refers to a branching clitic (i.e., [−strong]) node. Thus, a terminal clitic node does not violate Noninitiality, even if it contains -n. Third, the instances of -n that are quantified over by the universal in the statement of the constraint include all instances of -n that are dominated by the branching clitic node, even if they are not immediately dominated by that node. Hence, the boxed DCl node in doubling (82) does not violate Noninitiality, due to the presence of the embedded instance of -n that is sister to lo.29

In configurations with more than one pronominal clitic in the cluster, the attested variants can be descriptively classified into three cases.

(83) Attested variants in clusters with more than one pronominal clitic

  • Enclisis, with in-situ -n

    vénda-n-me-lo

  • Mesoclisis with no in-situ -n

    vénda-me-n-lo, vénda-me-lo-n, vénda-me-n-lo-n

  • Mesoclisis with in-situ -n

    vénda-n-me-n-lo, vénda-n-me-lo-n, vénda-n-me-n-lo-n

The basic question that the analysis must address is the following. If mesoclisis is a second-position (Noninitiality) effect on -n, why are there variants such as vénda-(n-)me-lo-n and others in (83b–c), in which -n is in apparent third (or higher) position within the cluster? Our analysis would seem to be able to derive mesoclisis with respect to the first pronominal clitic (vénda- (n-)me-n-lo), in which -n is clearly in second position. The question is thus what triggers displacement or copying further to the right.

The answer provided by our implementation of the second-position analysis, illustrated in detail below, is that this is a cyclic second-position effect afforded by the more complex structure present in clusters with more than one pronominal clitic. For instance, consider vénda-me-lo-n, with -n in apparent third position in the cluster. One can think of this form as being derived by displacement from mesoclitic vénda-me-n-lo, which is itself derived by a prior application of displacement.

(84)

graphic

The first application of displacement yields a clear second position for -n: due to Clitic Clustering, -n and me form a cluster, within which the former must be in second position. Since Clitic Clustering is cyclic, the displacement of -n to the right of lo from that displaced position is due to further clustering with this outermost clitic, which can result in the Noninitiality-violating cluster -n-lo and thus trigger one more application of the GR rule. Thus, the six attested mesoclitic variants can be derived by cyclic iterations of displacement and/or doubling.30

(85)

graphic

(86)

graphic

We now illustrate each of these possibilities, beginning with simple enclisis (83a).

If the optional rule of n-Extradition does not apply, the result is enclisis (83a). As a result, -n does not participate in Clitic Clustering, and Noninitiality is not violated. The resulting structure for vénda-n-me-lo is shown in (68b), repeated here.

(87)

graphic

If n-Extradition does apply, the result is a structure in which -n is a clitic.

(88)

graphic

As in cases of clusters with a single pronominal clitic, n-Extradition feeds application of Clitic Clustering to -n, with ensuing violations of Noninitiality that trigger mesoclisis. However, the more complex structure of these clitic clusters yields more derivational options, which result in the six attested mesoclitic variants in (85)–(86), as we now show.

The first step after n-Extradition (88) is cyclic application of Clitic Clustering to the two lowest clitics, -n and me.

(89)

graphic

The output violates Noninitiality. The repair can be either displacement, deriving forms with no in-situ -n (the cases in (85)), or doubling, resulting in forms with in-situ -n (the cases in (86)).

(90)

graphic

(91)

graphic

At this stage in the derivation, further applications of Clitic Clustering, displacement, or doubling yield the six different forms in (85)–(86), as we now show. The next step involves clustering of the top clitic l-o. At this point, the structural description of Clitic Clustering, repeated here from (69), is met in two different ways by the structures above.

(92)

  • Clitic Clustering

  • A node specified as [−strong] (a clitic) is rebracketed to form a constituent with a node specified as [−strong] that it c-commands and that is to its immediate left.

This is because in (90)–(91) there are (at least) two clitic nodes in the c-command domain of l-o that are to its immediate left: the top one dominating the lower clitic cluster (n) me n, and the rightmost instance of clitic n. Clitic Clustering can thus apply in two different ways, which we will refer to as high attachment and low attachment.31 As we will show, this optionality results in optional application of displacement/doubling of -n with respect to l-o.

Under high attachment, the output does not violate Noninitiality, and neither displacement nor doubling applies again. We illustrate this first for the case in which high attachment is fed by initial displacement.

(93)

graphic

Crucially, this structure with high attachment does not violate Noninitiality, since the new application of Clitic Clustering does not disrupt the structure resulting from displacement. Within the lower clitic cluster containing the lower pronominal clitic me and -n (rooted at the boxed DCl node), the latter remains in second position, and no repair is needed. A similar result occurs when high attachment is preceded by doubling instead of displacement.

(94)

graphic

Thus, low attachment derives the mesoclitic variants in which the rightmost copy of -n surfaces to the immediate right of the first pronominal clitic.

On the other hand, the output of low attachment does violate Noninitiality, which we illustrate first for the case in which low attachment is preceded by displacement.

(95)

graphic

This low attachment structure leads to further Noninitiality repairs, since within the newly formed cluster rooted at the boxed DCl node, -n is initial. These repairs yield mesoclitic variants in which a copy of -n surfaces to the right of the outermost pronominal clitic l-o. The same applies to a low attachment structure when it follows initial doubling instead of displacement.

(96)

graphic

The ensuing displacement or doubling repair only apparently violates the second-position requirement: within the immediate cluster containing -n, a copy of it is in second, not higher, position. The derivations for the four variants fitting this description follow.

First, the output of low attachment applied to a displacement structure (95) can be repaired by either displacement or doubling.

(97)

graphic

(98)

graphic

Similarly, the output of low attachment applied to a doubling structure (96) can also be repaired by either displacement or doubling.

(99)

graphic
graphic

(100)

graphic

This exhausts all the analytical possibilities—and, as desired, exhausts all the attested forms of mesoclisis in clitic clusters.

To summarize, n-Extradition, followed by cyclic application of Clitic Clustering and displacement or doubling repair to clusters violating Noninitiality, derives all the mesoclitic variants, including those found in complex clitic clusters. The availability of all these variants (at least in the least restrictive dialects) is governed by two derivational forks in the road. First, n-Extradition is optional; if it does not apply, -n is not a clitic and therefore not subject to Noninitiality, resulting in enclisis. Application of n-Extradition and subsequent Clitic Clustering violates Noninitiality, which is repaired by displacement or copying to the right of a pronominal clitic (i.e., mesoclisis). In cases of clusters with more than one pronominal clitic, this leads to a second derivational choice afforded by the two ways in which the outermost clitic can cluster with the lower clitics. Under high attachment, -n remains to the left of the outermost clitic, since Noninitiality is not violated; under low attachment, further displacement or doubling to the right occurs, since the newly formed cluster of -n plus outer clitic violates Noninitiality.32 Crucially, Noninitiality is an obligatory constraint, and apparent surface violations of the second-position requirement in some of the mesoclitic variants are due to other factors that influence the course of the derivation.

In the following sections, we provide an account of the different explananda discussed in section 4, some of which place dialect-specific restrictions on the derivational choices discussed above.

6.3 Deriving the melon Hierarchy

Recall the melon hierarchy from section 4.3, under which no speaker allows mesoclisis only to the second pronominal clitic in a cluster. Given the derivations provided in the previous section, a natural explanation emerges. The cyclic character of these derivations is such that -n will always be clustered with the closer clitic first, hence will have an opportunity to repair Noninitiality immediately. Displacement of -n to the position after the first clitic is thus available in all dialects that initiate n-Extradition.

After the first step of displacement, clustering of the higher clitic to an adjacent clitic node can proceed either by high attachment (i.e., to the clitic constituent formed by -n and the inner clitic, as in (93)–(94)) or by low attachment (i.e., just to -n, as in (95)–(96)). Only low attachment will result in a configuration that freshly violates Noninitiality and thereby trigger further displacement to the right of the outer clitic.

Therefore, what prevents some speakers from displacing -n all the way to the end of the clitic cluster is simply a restriction on high attachment. Those who allow both high and low attachment will allow either -me-n-lo or -me-lo-n. Given the optionality between high and low attachment, the question that arises is why no speakers allow only low attachment. The answer to this question can be grounded in locality: there is no clear grammatical bias that would restrict speakers to the hierarchically more distant node satisfying a structural description.

6.4 The Number Restriction

We now turn to the observation that mesoclisis never occurs with plural clitics, even in cases where one might expect that a coda restriction is not at play, such as *vénda-nos-lon, which would be phonotactically well-formed (see section 4.2). Our explanation is cast in terms of an intervention constraint on the GR rule.

(101) A [−singular] element may not move across another [−singular] element to satisfy Noninitiality of -n.

Returning to the structural description of the basic GR rule in (79), repeated in (102), the intervention constraint can be incorporated into the additional constraint in (103).

(102) GR rule

  • SD: X AgrCl DClY, where AgrCl is [−participant, −singular] and AgrCl and DCl are sisters.

  • SC: Insert

    • [ to the immediate left of AgrCl

    • ] to the immediate right of DCl

    • > < to the immediate right of AgrCl (Displacement)

      or

      > to the immediate right of AgrCl (Doubling)

(103)

  • Plural Intervention Constraint on (102) (all dialects)

  • DCl is not [−singular].

Although the term intervention here may recall Rizzi’s (1990) Relativized Minimality, we emphasize that plural -n is not undergoing syntactic movement (in fact, even on Kayne’s (2010) analysis, where it is the clitics that are moving). We take the existence of such parallel though distinct constraints on displacement operations across distinct modules to reflect an organization of the grammar where particular computational mechanisms may be reused, which is of course distinct from viewing both operations as resulting from applications of the same process within a single module. The fact that this constraint holds across all dialects means that the rule of n-Extradition (which is optional throughout all the relevant varieties) cannot take place in the presence of plural clitics—if it did, a crash would result.

However, importantly—and this is a property of our general morphotactic approach to microvariation—we contend that there may be a second implementation that learners internalize to yield the number restriction. This one is stated specifically in terms of movement across internally complex clitics.

(104) A clitic may not move over an internally complex clitic to satisfy Noninitiality on -n.

This can be incorporated into the structural description of the mesoclisis rule in (102) as follows:

(105)

  • Nonterminal Constraint on (102) (dialectal)

  • DCl is a terminal node (i.e., is not internally complex).

The consequence of this, given the operations of fission triggered within all plural clitics by the constraints on joint exponence in (35b–c) and resulting complex subtrees such as (36)–(37), is that no displacement is possible for any plural clitic, because any plural clitic in Spanish will necessarily involve a branching DCl sister node to the erstwhile metathesizing -n.

Of course, the specific restriction in (105) causes blocking only in the case of clitics that have undergone fission. Therefore, there may be language varieties in which fission-triggering constraints such as (35b–c) do not apply; hence, (105) will not be at issue. Nonetheless, recall that there is an additional, distinct representational constraint on plural clitics (103), which holds for all speakers. Now, given that the number restriction empirically holds across all dialects and that there are two ways to derive it, how do we know which implementation—(103) or (105)—a given speaker has internalized? In fact, we will show that there are consequences of choosing one or the other. Specifically, choosing the implementation in terms of complexity will also rule out mesoclisis with third person accusatives, as these undergo gender fission, hence are internally complex even in the singular. We return to this in the next section.

6.5 The Person and Case Hierarchies

Another explanandum, pointed out in section 4, is the person hierarchy, repeated here from (42).

(106)

  • Mesoclitic person hierarchy: Nonthird person over third person

  • se, me > lo, la, le

According to this hierarchy, some speakers allow all clitics to participate in mesoclisis, while others allow only nonthird person clitics to do so. We will focus on the latter group, where the restriction in question can be viewed in terms of an intervention constraint on the structural description of the GR rule. Specifically, since the -n ending in the imperatives in question is postsyntactically [−participant] (recall from section 3 that -n is the result of 2Pl Impoverishment or Formal Impoverishment, which yield neutralization of second and third person forms), the relevant restriction bans displacement operations from moving it across third person clitics.

(107) A [−participant] element may not move across another [−participant] element to satisfy Noninitiality of -n.

Returning to the structural description of the basic GR rule in (79), repeated in (108), the intervention constraint can be incorporated into the additional constraint in (109).

(108) GR rule

  • SD: X AgrCl DClY, where AgrCl is [−participant, −singular] and AgrCl and DCl are sisters.

  • SC: Insert

    • [ to the immediate left of AgrCl

    • ] to the immediate right of DCl

    • > < to the immediate right of AgrCl (Displacement)

      or

      > to the immediate right of AgrCl (Doubling)

(109)

  • Nonparticipant Intervention Constraint on (108) (dialectal)

  • DCl is not [−participant].

We leave open the question of why (109) is limited to some speakers, whereas the Plural Intervention Constraint (103) is found across all mesoclisis varieties. It may be that as learners generalize from the observed patterns, -n is more clearly recoverable as [−singular], whereas its identity as [−participant] is less direct, given that it is syntactically second person and undergoes subsequent impoverishment.

This particular implementation in terms of an intervention constraint on GR operations stated in terms of [−participant] features makes an important prediction for the Italian and Albanian varieties discussed in Manzini and Savoia 2007, 2011, namely, that when the plural imperative suffix is specific to second person (and not syncretic with third), any intervention constraints that are imposed on the person features of the clitics it crosses will yield a hierarchy different from the one in (106).

We now turn to the case hierarchy, repeated from (43).

(110)

  • Mesoclitic case hierarchy: Dative third person over accusative third person

  • le > lo, la

How are we to derive those dialects that allow mesoclisis with all singular clitics except third person accusatives? We argue that such speakers have extended their implementation of the number-complexity constraint in (105) to all complex clitics, automatically. The ban on displacing a clitic over an internally complex clitic for cases resulting from number fission can thereby extend to those resulting from gender fission. Returning to the structural description of the basic GR rule in (108), the intervention constraint applicable to gender-bearing clitics results from the following constraint, repeated from (105):

(111)

  • Nonterminal Constraint on (108) (dialectal)

  • DCl is a terminal node (i.e., is not internally complex).

A consequence of the interaction between (111) and the structures that result from the fission process described in section 3 (see (37)) is that because any accusative clitic will have gender fission, no displacement will be possible over such internally complex clitics. Since datives are not subject to gender fission, they will allow displacement to occur even in dialects that have adopted (111).

To summarize the patterns of variation within the case hierarchy, dialects in which accusative, gender-bearing lo/la cannot mesocliticize have the Nonterminal Condition in (111)—which also derives no mesoclisis with plurals. By contrast, in dialects in which lo/la can in fact mesocliticize, it is the Plural Intervention Condition (103) that prevents plural mesoclisis. However, the absence of any nonterminal constraint—whether for complex D subtrees that have undergone number fission or for those that have undergone gender fission—means that any singular third person clitic can undergo metathesis.

6.6 Interim Summary: Interspeaker Variation Derives from Additional Morphotactic Constraints

To summarize sections 6.36.5, we have proposed additional morphotactic restrictions on the GR formalism that account for the three explananda outlined in section 4. We have accounted for the melon hierarchy in terms of constraints on attachment preferences in rebracketing, claiming that no dialect would specifically prefer low but not high attachment when identifying a clitic node for rebracketing. In terms of φ-featural restrictions, we have proposed that the person hierarchy and the number restriction are due to additional morphotactic intervention constraints imposed upon the GR rule; whether speakers and dialects include these constraints varies. Finally, also in terms of variation, there is another way to derive the number restriction, based on a constraint on the complexity of the metathesizing clitic; this in turn rules out mesoclisis of both number-bearing and gender-bearing clitics for the relevant speakers.

7 Open Problems for Further Analysis

In this article, we have developed a morphotactic analysis combined with the GR formalism to provide advances in modeling mesoclisis. More broadly, we contend that our account demonstrates the benefits of a division of labor between syntactic and postsyntactic mechanisms for morpheme order. The claim is that there is no fundamental syntactic difference between grammars that allow mesoclisis and those that do not, and that the variation found among the former is to be localized within morphotactically grounded constraints. This obviates the need for appeal to ill-understood syntactic constraints on the generation of Agr heads and movements around them. Put differently, the temptation to do everything within one module and avoid the apparent “redundancy” of having two similar but distinct modules is a blunt application of Occam’s razor. While one could in fact type an entire novel using only a smartphone (and thus throw away a redundant laptop), arguably there are other tools at hand that remain better-suited for certain labors.

The task of understanding mesoclisis in all of its details is not finished, however. In this section, we outline a few challenges and areas for further research that face all approaches to this phenomenon, however they may be cast.

7.1 Mesoclitic Orders vs. Other Cluster Orders

In our analysis, mesoclisis is derived from enclisis, by metathesis of the relevant clitics (cyclically in the case of more than one) with the plural -n that ends up flanking them. In Kayne’s (2010) analysis, the relation between enclisis and mesoclisis is even more direct, to the point where he specifically emphasizes the parallels with clitic ordering in a cluster as found in proclisis/enclisis. Thus, the properties of the relevant clitics that determine leftward (i.e., higher, for Kayne) order in clusters will in turn determine their relative order with respect to -n and indirectly derive the person hierarchy.

Given this perspective, we do not expect differences between the linear order of clitics in enclitic (or proclitic) environments and their relative order with respect to displaced inflection in mesoclisis: if clitic x can mesocliticize but clitic y must remain enclitic in the same context, then x should precede y in purely enclitic (or proclitic) environments. Divergences can be handled within our framework, by additional morphotactic constraints imposed on mesoclitic orders, but are certainly not expected. On Kayne’s analysis, by contrast, they are essentially impossible to derive, given that the explanation for the person hierarchy as it stands crucially depends on this isomorphism.

In Spanish, since the relative order of clitics in enclisis and proclisis correlates with their ability to undergo mesoclisis, little can be made of this theoretical possibility. Some provocative data, however, come from the Arbëresh varieties discussed by Manzini and Savoia (2011:1114). They note that throughout Arbëresh, third person dative clitics precede the middle-passive clitic [u] in nonmesoclitic environments, as illustrated with proclitics in Portocannone in (112), but that in mesoclitic contexts in S. Marzano (a different dialect, but with the same morphemes), the middle-passive clitic precedes inflection (113a), but third person dative clitics follow it (113b) (repeated from (15)).

(112)

(113)

The general point that Manzini and Savoia (2011) wish to make with this contrast is that if the relative order of clitics with respect to inflection in mesoclitic contexts does not correlate with the order of clitics in purely enclitic or proclitic environments, then a climbing explanation for the hierarchy is not in order—certainly a worthwhile point to explore. However, to make this point more forceful, it would be ideal to find paradigms like (112)–(113) with comparable data from within the same dialect, as well as further illustration of the contrast between clitics that mesocliticize and those that do not, as in (113), but within the same cluster. We hope that further work on mesoclisis, wherever it may be found, can shed light on this question, given its theoretical relevance for a syntactic or a morphotactic analysis.

7.2 Instances of -n with No Original Source

We have argued that -n, when reanalyzed as part of the postverbal domain by n-Extradition, undergoes displacement or doubling to satisfy a second-position requirement. While this exhausts the description of imperatives, intriguing cases are found in nonstandard Spanish, where the verb is infinitival (114b) or gerundive (115b) and an -n follows clitics (data from Harris and Halle 2005:213; cf. the standard enclitic forms in (114a), (115a)). In these cases, there is no original source for the -n, as these nonfinite forms do not otherwise show agreement.

(114)

  • Quiere -n ver -me.

    want.PRS.IND -PL see.INF -CL.1SG.ACC

    ‘They want to see me.’

  • Quiere -n ver -me -n.

    want.PRS.IND -PL see.INF -CL.1SG.ACC -PL

(115)

  • Está -n besándo -se.

    be.PRS.IND -PL kissing -CL.REFL

    ‘They are kissing each other.’

  • Está -n besándo -se -n.

    be.PRS.IND -PL kissing -CL.REFL -PL

We will refer to speakers with these specific types of mesoclisis as vermen speakers.

It may be tempting to view these instances of postclitic -n as being copies of plural agreement in the matrix finite verb (quiere-n and está-n in the examples above).33 However, it is straightforward to find examples in which this is not a possible analysis, as they involve postclitic -n in a nonfinite clause with no possible source in a finite verb. For instance:

(116)

  • Escuchaba el ruido de la gente, de los esferos moviéndo -se -n . . .

    listen.IPFV.3SG the noise of the people of the pens moving -CL.REFL -PL

    ‘He listened to the noise of the people, of the pens moving, . . . ’

  • . . . con motivo de cumplir -se -n los cincuenta años de aquel viaje . . .

    with reason of turn.INF -CL.REFL -PL the fifty years of that trip

    ‘ . . . because it’s been fifty years since that trip . . . ’

    (Mare to appear:6)

Postclitic -n occurs on a gerund in (116a)34 and on an infinitive in (116b)35. Neither example has another occurrence of -n on a finite verb. Furthermore, unlike in (114)–(115), the subject of the nonfinite verb is overt (los esferos ‘the pens’ and los cinquenta años ‘the fifty years’, respectively), not a PRO whose controller could potentially trigger overt plural agreement on a finite verb. We take this as evidence that postclitic -n on nonfinite forms has no original source on a finite verb, even though examples such as those in (114)–(115) might lead one to conclude that it does.

The infinitival and gerundive forms have in common with the imperatives on which we focus that all three are enclitic forms in the standard variety (and, in our analysis, at the output of the syntax). Under Kayne’s (2010) analysis, one might say that the appearance of a final -n on all three is expected; there is simply a high -n, to which the entire gerund or infinitive phrase moves. However, this movement is obligatory only in the presence of a clitic (Harris and Halle 2005:213).

(117)

  • *Está -n besando -n.

  • be.PRS.IND -PL kissing -PL

  • ‘They are kissing.’

The apparent simplicity of the movement account lapses, and about (117), Kayne suggests it is illicit because the high -n (the one following the gerundive) “requires a (certain kind of) filled specifier” (p. 165).

Let us seek a more principled reason why this -n sprouts only in the presence of a clitic—given that the infinitival and gerundive morphology “shouldn’t” have the -n in the first place. We would like to suggest that the speakers in question have inverted the figure-ground relation between -n and clitics: rather than formulating a noninitiality requirement on -n, speakers who are exposed to metathetic forms such as véndalon might instead formulate a nonfinal requirement on the clitics. If so, then (114) and (115) can be understood as follows:

(118) For vermen idiolects, the clitic has a nonfinal requirement (“Nonfinality”).

The intuition behind (118) is that forms like (114b) and (115b) involve the sprouting of an -n item exactly to shield the clitic from final position. What however, is the source of this -n, given that it is not visible on the verbal form itself? If pursued to its consequences, this proposal would mean that the plural feature is morphosyntatically present on the infinitive or gerundive, though realized with zero.36 However, this plural feature can undergo displacement, of the kind witnessed throughout the derivations in section 6, but prior to Vocabulary Insertion. The morphotactic motivation for this displacement from the nonfinite form to the position after the clitic is to provide a rightward shield for the clitic in these dialects. Given that this displacement occurs prior to Vocabulary Insertion, it enables this displaced Agr to be realized as -n in its postclitic form, while its original copy that is right-adjacent to the nonfinite form is realized as zero.

As discussed by Minkoff (1993), there is suggestive further evidence for a Nonfinality requirement on clitics that can be satisfied by displacement of plural material to their right.37 Specifically, if plural clitics such as first person plural nos can in fact be decomposed into bimorphemic no-s, as in our analysis in section 3, then displacement of the plural -s portion could indeed provide a buffer against Nonfinality for a clitic that is in turn to its right. Such cases are in fact found in clitic clusters, as in (119b), from Harris and Halle 2005:196 (cf. standard placement of -s in (119a)).

(119)

  • Vénda -no -s -lo!

    sell.IMP -CL.1PL.DAT -PL -CL.3SG.M.ACC

    ‘Sell it to us! (imperative singular)’

  • Vénda -no -lo -s!

    sell.IMP -CL.1PL.DAT -CL.3SG.M.ACC -PL

Considering (119), and the fact that plural morphemes may displace to the right of clitics even when they themselves originate in an earlier clitic, the logical possibility arises that such displacement could occur prior to Vocabulary Insertion. This would have the startling result that a plural morpheme originating on a first person plural verb could receive an elsewhere realization as -n if it found itself displaced to the rightmost edge of a clitic cluster. Doubling forms of the type in (120b), observed by Mare (to appear:9–12) confirm this prediction, as the final morpheme -n seems to be originating within the plural hortative verb where its features have been realized as -s (cf. the standard enclitic form in (120a)).38

(120)

  • Dé -mo -s -le ánimos.

    give.PRS.SBJV -1PL -PL -CL.3SG.DAT courages

    ‘Let’s encourage him/her!’

  • Dé -mo -s -le -n ánimos.

    give.PRS.SBJV -1PL -PL -CL.3SG.DAT -PL courages

    ‘Let’s encourage him/her!’

Research on such forms is still incipient, but we hope that increased formalization of the GR formalism within a cascaded derivational model of postsyntax will provide tools for understanding variants of this shape and how they may reflect a reinterpretation of the relevant morphotactic constraint as targeting the clitic itself.

7.3 Prosodic Factors Conditioning Optionality and Variability

As emphasized from the outset, mesoclisis is largely an optional phenomenon, and we have not gone into great detail in discussing the factors that might condition its greater rates of application. Here, we outline a few that appear relevant. One is the greater application of -n metathesis in monosyllabic first conjugation verbs such as dar ‘give’, yielding de-me-n, de-le-n. This tendency may find motivation in increasing the weight of the final syllable. Indeed, metathesis would satisfy well-formedness across multiple domains given that a cluster-final clitic that benefits from plural displacement to its right not only becomes morphotactically nonfinal in its domain but also gains a syllable coda in the phonology as a result, thereby adding to its syllable weight.

This points toward a more general idea: that -n metathesis occurs more often specifically for metrical reasons. In fact, Colantoni and Cuervo (2013) investigate the extent to which -n metathesis occurs alongside stressed enclisis in Argentinian Spanish, the idea being that the Stress-to-Weight Principle would prefer a heavy syllable, made possible by -n metathesis, as the final stressed syllable.

If this analysis is on the right track, it would explain why the imperative marker -d in Iberian Spanish never undergoes metathesis, as Maria Luisa Zubizarreta points out (pers. comm.): as a stop consonant, it is nonmoraic, and so would have no syllable weight to contribute toward this putative goal.

Similarly, given the local vs. long-distance hierarchy discussed above, whereby some speakers allow vénda-me-n-lo but not vénda-me-lo-n, one could clearly appeal to rhythmic factors of heavy and light syllables: vénda-me-n-lo forms an alternating HLHL sequence, whereas vénda-me-lo-n forms an HLLH trough. Further explorations of this analysis would naturally need to examine the relative rates of application and preferences with a variety of verb-stem types.

8 Conclusion: Properties of a Morphotactic Analysis

We have proposed a set of morphotactic constraints that govern the applicability of the doubling and displacement rules proposed in Harris and Halle 2005. Specifically, we have argued that mesoclisis occurs in order to satisfy a second-position requirement on the imperative plural suffix -n operative within the clitic domain. We have linked restrictions on mesoclisis to the internal structure of clitics—specifically, with intervention constraints on certain featural values undergoing metathesis with -n and in terms of a constraint on metathesis with internally complex (i.e., gender- or number-bearing) clitics. Such restrictions would not be derivable from attempts to reduce mesoclisis ordering to the ordering of clitics within clusters more generally. We have also capitalized on the role of cyclicity in capturing attested and unattested variants of mesoclisis with clusters consisting of more than one clitic. The overall ingredients of a morphotactic analysis such as the one proposed here (reflecting the research strategy found in our earlier work on microcomparison between Basque dialects in Arregi and Nevins 2012) are combined to account for interspeaker variation in terms of restricted differences in the morphotactic constraints that speakers posit and fine-grained differences in the inventory of repair operations and their timing, an issue of recent discussion (Arregi and Nevins 2017, Kiparsky 2017, Weisser and Guseva 2018).

More generally, the approach to mesoclisis proposed here is embedded in a theory of grammar according to which crosslinguistic morphosyntactic phenomena are to be accounted for not in terms of a (purely lexical or purely syntactic) monolithic system, but in terms of an explanatory division of labor among separate but interrelated grammatical modules, each with its own principles and operations. In this, we follow a tradition in generative research going back at least to Chomsky’s (1970:185) observation that “it is to be expected that enrichment of one component of the grammar will permit simplification in other parts”; indeed, treating the microvariation between doubling and metathesis and its feature-based restrictions and interactions with morphotactics and phonology within a postsyntactic component leads to a much more principled syntax. As Chomsky continues, “The proper balance between various components of the grammar is entirely an empirical issue. . . . There are no general considerations that settle this matter” (p. 185). In the case at hand, we argue that while the bulk of clitic placement in Romance is the purview of syntax, mesoclisis and its special relation with enclisis result from the interaction between the syntactic and morphotactic components.

Notes

1 Unless otherwise noted, all numbered examples are from Spanish. Deviating from orthographic convention, we represent enclitics and mesoclitics (and occasionally, suffixes) with hyphens in Spanish examples. We also omit the standard opening exclamation mark from imperatives (¡), to avoid confusion with grammaticality/acceptability judgment marks. We follow the basic spirit of the Leipzig Glossing Rules in representing all examples, with the addition of the following abbreviations: CL (clitic), COL (colloquial), FRM (formal), and MP (middle-passive).

2 We remain agnostic, however, on whether the mesoclisis found in the future and conditional tenses of European Portuguese (e.g., dar-lhe-ei (give-CL.3SG.DAT-FUT.1SG) ‘I will give to him’) or the so-called allocutive cases found in Romanian imperatives discussed by Hill (2015) are to be united with those under discussion here.

3 We reverse the direction of the angled brackets from the way they are defined in Harris and Halle 2005 for reasons described in the text.

4 Throughout this article, we state all postsyntactic rules of impoverishment and mesoclisis in terms of a structural description (SD) and a structural change (SC), following Arregi and Nevins 2012.

5 Although the plural agreement morpheme agrees with a (pro-dropped) semantically second person subject, the agreement morpheme itself is postsyntactically third person. See section 3, where we spell out in some detail our assumptions about the features of clitics and agreement morphemes in Spanish, as well as their postsyntactic derivation.

6 We concentrate on non-Iberian Spanish for simplicity, as this allows us to abstract away from certain contrasts that are available only in Iberian dialects and that are not directly relevant to our account of mesoclisis.

7Table 1a only shows agreement affixes in tenses other than the present indicative and the perfective past (aorist), which have certain tense/aspect/mood-specific allomorphs (e.g., -o for first person singular in the present indicative, -ste for second person singular in the perfective). These paradigms also have the agreement syncretisms discussed in the text. In this section, we also gloss over the colloquial/formal distinction, whose relevance is only orthogonal to mesoclisis, and we focus on non-Iberian dialects. For more complete paradigms, see appendix A.

8 For other Distributed Morphology accounts of the postsyntax of Spanish pronominal clitics, see Bonet 1991, 1995, Halle and Marantz 1994, and Harris 1995.

9 In addition, second person formal forms are syncretic with third person in both numbers. See appendix A.

10 On the general notion of metasyncretism, see Williams 1994, Bobaljik 2002, and Harley 2008. The latter two argue for impoverishment-based accounts of metasyncretism.

11 In appendix B, we argue against a potential syntax-based alternative analysis to our postsyntactic account of this syncretism.

12 Second person singular morphemes, which are not affected by 2Pl Impoverishment, are realized by exponents specific to second person (te, -s), as shown in table 1. See appendix A.

13 Although we state fission as it applies specifically to pronominal clitics, it should be noted that agglutinative exponence of gender and number is a more general property of Spanish inflection, present in all other parts of speech that jointly inflect for these features (nouns, adjectives, determiners, etc.). The particular claim made here that person ([±participant]) features are in part responsible for fission in clitics cannot be extended to other parts of speech such as nouns and adjectives, which do not inflect for person. In addition, although this is not directly observable in pronominal clitics, inflectional class is a relevant category in describing Spanish inflection. We leave a more comprehensive account of the exponence of Spanish inflection for future work. See Harris 1991, 1999.

14 Arguably, clitics in other Romance languages are subject to only a subset of the constraints in (35) and thus display a higher degree of fusional inflectional morphology. For instance, Italian is not subject to constraint (35c) banning the joint exponence of gender and plural number, as illustrated by the accusative clitic paradigm l-o (masculine singular), l-a (feminine singular), l-i (masculine plural), l-e (feminine plural).

15 We leave open here the more general possibility that some of this ordering is derived from intrinsic properties of the rules themselves (as we proposed in Arregi and Nevins 2012:chap. 4, where paradigmatic impoverishment rules such as 2Pl Impoverishment precede all syntagmatic impoverishment rules such as Spurious se Impoverishment).

16 On Iberian second person plural os, see section 4.1.

17 Unfortunately, the generalization cannot be checked with speakers who restrict mesoclisis to the higher part of the person hierarchy, that is, speakers who allow mesoclisis only with me and se. The cluster se me, although grammatical, is possible only in cases in which se is interpreted as a third person reflexive. It is not grammatical with se as a second person plural reflexive, which we take as related to the fact that Iberian colloquial second person plural os cannot cluster with me, whether it is interpreted as reflexive or not: *me os, *os me. Since reflexive se in imperatives is necessarily second person plural, it cannot cluster with me in imperatives, which makes testing mesoclisis with this cluster impossible.

18 See Minkoff 1993 and Halle and Marantz 1994:285–288 for analyses along the same lines, but stated as linearization requirements imposed on the pronominal clitics, not on -n. We agree that such requirements are needed, but in order to account for a different type of mesoclisis in Spanish, discussed in section 7.2.

19 The structure in (66) omits any specifier or complement positions associated with arguments (including subjects). Terzi’s (1999) own label for the clitic-hosting head is F instead of Cl. We use the latter label following a suggestion by Terzi (1999:93). We also reverse the order of T and (subject) Agr with respect to Terzi’s practice, to make the structure directly compatible with more explicit accounts of the morphophonology of Spanish verbal inflection (Oltra-Massuet and Arregi 2005). The verb stem is always (at least) bimorphemic (Marantz 1997), containing a categoryless root (√) and a verbal category-fixing head, which we label with a capital letter following Arregi and Nevins 2014.

20Terzi (1999) proposes a further difference between finite and nonfinite contexts: while the clitic is hosted by Cl (labeled F in her account) in the former, it is hosted by T in the latter. This presupposes a structure in which T is higher than Agr (see Terzi 1999:92–99 for details), the reverse of the order adopted here (see footnote 19 for discussion). Our implementation preserves the insight that the verb remains low in proclitic contexts, but moves to a high-peripheral position (picking up the clitic on the way) in enclitic environments.

21 The structure in (67) also includes a theme position (Th; see Oltra-Massuet 1999, Oltra-Massuet and Arregi 2005) and a complex internal structure for the accusative clitic lo (section 3). Both are the result of postsyntactic operations.

22 Unlike in Romance, clitics in clusters can adjoin to separate clitic-hosting heads in Greek, which accounts for the variable order of clitics attested in enclitic position in this language (Terzi 1999:99–108).

23 Although the effects of cyclicity are not visible in cases of enclisis, they derive the attested variation in mesoclitic possibilities, as shown below.

24 This can be seen, for instance, in English borrowings that have primary stress on the preantepenultimate syllable in the source: Terminator is pronounced [terminˈator] or (for speakers more aware of the English pronunciation) [terminˈejtor].

25 Interestingly, there is an additional suffix used in Iberian Spanish imperatives, -d, for second person plural colloquial, mentioned in section 4.1. This suffix does not allow mesoclisis to its left (e.g., there is no *Canta-me-d! ‘Sing it to us!’ alongside grammatical Canta-d-me!). We assume that this is because this morpheme is not subject to n-Extradition.

26 Our statement of the GR rule in (79) differs from that in (20), adapted from Harris and Halle 2005, in two ways. First, the agreement morpheme -n and the clitic to its right are referred to as AgrCl and DCl respectively, reflecting our claim in this section that mesoclisis is the result of reordering specifically within clitic clusters (in which agreement is a clitic). Second, these two elements must be sisters. The relevance of the latter difference is discussed below.

27 Note that Noninitiality is a triggering constraint, and not a blocking constraint. Thus, Clitic Clustering is not blocked in (80), though it creates an environment that violates Noninitiality (which in turn triggers displacement or doubling). We assume, following previous work (Arregi and Nevins 2012:chap. 5), that being blocking or triggering (or both) is a property of each postsyntactic constraint. On the use of triggering and blocking constraints in phonology, see Paradis 1988.

28 Following previous work (Arregi and Nevins 2012:322–326), we assume that the hierarchical structure is affected by the application of mesoclisis operations in such a way that it minimally satisfies the trigger of the rule (in this case, Noninitiality), as well as general constraints on hierarchical structure, such as the bans against branch-crossing (McCawley 1968) and against ternary (and higher) branching (Kayne 1981). In the case of doubling (82), this results in the projection of additional structure to accommodate the added copy of plural -n, in order to avoid ternary branching.

29 A more precise formulation of Noninitiality that takes all these assumptions into account is the following:

  • It is not the case that for all instances of x and y such that (a) xy, (b) x is plural -n, and (c) x and y are dominated by a clitic node, x precedes y.

30 A similarly cyclic analysis relating enclisis to some of the mesoclitic variants discussed here is proposed in Minkoff 1993:187–190, but in the context of a different postsyntactic analysis of mesoclisis. See footnote 37.

31 In the case of doubling (91), an additional clitic node is present dominating me and the rightmost instance of -n. As the reader can verify, allowing for this intermediate attachment does not result in overgeneration, as it yields the same outputs as high attachment. As discussed below, only low attachment results in further applications of displacement or doubling.

32 Interestingly, optionality in mesoclisis could be entirely reduced to the way Clitic Clustering applies. In particular, the very first cyclic step of Clitic Clustering depicted in structures such as (80) could also be applied in two ways, by rebracketing the pronominal clitic with either the top or bottom AgrCl node. As in more complex cases, the latter option (which is the one assumed in (80)) violates Noninitiality for -n and leads to mesoclisis by displacement or doubling. However, the former option does not violate the constraint, resulting in enclisis. If this analytical optionality in terms of the rebracketing site proves empirically correct, the rule of n-Extradition would then be obligatory in the dialects that have mesoclisis. Not having explored all the consequences of this hypothesis, we leave it as a matter for future research.

33 We thank an anonymous reviewer for bringing this up as a possible analysis of sentences of this type.

34 The original example cited by Mare is found at http://www.calameo.com/books/002937235545338fb2644. The rest of the sentence includes further instances of postclitic -n on a gerund with no possible source on a higher finite verb: los pupitres arrastrándo-se-n, hojas rompiéndo-se-n, solo ruido, voces chillonas y ásperas de las cuales no daban gusto oir ‘the desks being dragged, sheets being torn, only noise, loud and rough voices that weren’t pleasant to hear’.

35 The full sentence (from http://www.diariovasco.com/pg060716/prensa/noticias/AltoUrola/200607/16/DVA-ALU-028.html) does not have a possible source for -n on a finite verb: Miguel confiesa que está pinchando para animar a sus compañeros y confía en que se pueda hacer, por lo menos, una cena con motivo de cumplir-se-n los cincuenta años de aquel viaje que les marcó en su infancia ‘Miguel admits that he’s pushing and encouraging his friends, and is confident that at least a dinner can be held because it’s been fifty years since that trip that marked their childhood’.

36 Under this analysis, the relevant nonfinite forms in Spanish are like inflected infinitives in Portuguese (e.g., Raposo 1987, Pires 2006).

37 In fact, Minkoff (1993) claims that data such as (119b) and the n-displacement forms that are the focus of this article are both due to postsyntactic displacement of clitics to the left of the plural suffix (-s or -n), in response to a second-position (from the end) requirement on clitics. Although such an analysis accounts for multiple-mesoclisis variants in which plural -n outflanks more than one clitic (e.g., vénda-me-lo-n ‘Sell it to me!’), it cannot be straightforwardly extended to variants in which -n is sandwiched between the two clitics (e.g., vénda-me-n-lo), as the second clitic in the sequence is final in the word. Similar comments apply to Halle and Marantz’s (1994) postsyntactic analysis of mesoclisis.

38 In appendix A, we analyze -mos as an atomic exponent of first person plural agreement. Examples such as (119b) strongly indicate a polymorphemic analysis in which -s in -mo-s is in fact a separate realization of number, as suggested by our parse and glosses in these examples.

Acknowledgments

We thank María Cristina Cuervo, Elena Jaime Jiménez, María Mare, Liliana Sánchez, Cristina Schmitt, Maria Luisa Zubizarreta, audiences at the 45th Linguistic Symposium on Romance Languages and the VII Encuentro de Gramática Generativa, and colloquium audiences at MIT, the University of Utah, and the University of Geneva for helpful discussion. We owe an incalculable debt of gratitude to the late Morris Halle for his mentorship for two decades and for his contributions to the study of morphology, without which the work reported here would not have been possible.

References

Alcázar,
Asier
, and
Mario
Saltarelli
.
2010
. In support of a syntactic analysis of double agreement phenomena in Spanish. In
Romance Languages and Linguistic Theory 2008: Selected papers from ‘Going Romance’, Groningen 2008
, ed. by
Reineke
Bok-Bennema
,
Brigitte
Kampers-Manhe
, and
Bart
Hollebrandse
,
1
15
.
Amsterdam
:
John Benjamins
.
Arregi,
Karlos
, and
Andrew
Nevins
.
2012
.
Morphotactics: Basque auxiliaries and the structure of Spellout
.
Dordrecht
:
Springer
.
Arregi,
Karlos
, and
Andrew
Nevins
.
2014
.
A monoradical approach to some cases of disuppletion
.
Theoretical Linguistics
40
:
311
330
.
Arregi,
Karlos
, and
Andrew
Nevins
.
2017
. Presyntactic morphology or postsyntactic morphology and explanatoriness in the Basque auxiliary. In
The morphosyntax-phonology connection: Locality and directionality at the interface
, ed. by
Vera
Gribanova
and
Stephanie S.
Shih
,
401
418
.
Oxford
:
Oxford University Press
.
Bobaljik,
Jonathan David
.
2002
. Syncretism without paradigms: Remarks on Williams 1981, 1994. In
Yearbook of morphology 2001
, ed. by
Geert
Booij
and
Jaap
van Marle
,
53
85
.
Dordrecht
:
Kluwer
.
Bonet,
Eulàlia
.
1991
.
Morphology after syntax: Pronominal clitics in Romance
.
Doctoral dissertation, MIT, Cambridge, MA
.
Bonet,
Eulàlia
.
1995
.
Feature structure of Romance clitics
.
Natural Language and Linguistic Theory
13
:
607
647
.
Bybee,
Joan L
.
1985
.
Morphology: A study of the relation between meaning and form
.
Amsterdam
:
John Benjamins
.
Calabrese,
Andrea
.
2008
. On absolute and contextual syncretism: Remarks on the structure of paradigms and how to derive them. In
Inflectional identity
, ed. by
Asaf
Bachrach
and
Andrew
Nevins
,
156
205
.
Oxford
:
Oxford University Press
.
Calabrese,
Andrea
, and
Diego
Pescarini
.
2014
.
Clitic metathesis in the Friulian dialect of Forni di Sotto
.
Probus
26
:
275
308
.
Chomsky,
Noam
.
1970
. Remarks on nominalization. In
Readings in English transformational grammar
, ed. by
Roderick A.
Jacobs
and
Peter S.
Rosenbaum
,
184
221
.
Waltham, MA
:
Ginn
.
Colantoni,
Laura
, and
María Cristina
Cuervo
.
2013
. Clíticos acentuados. In
Perspectivas teóricas y metodológicas del español de la Argentina
, ed. by
Laura
Colantoni
and
Celeste Rodríguez
Louro
,
143
157
.
Berlin
:
Verveurt
.
Dressler,
Wolfgang U.
,
Willi
Mayerthaler
,
Oswald
Panagl
, and
Wolfgang Ullrich
Wurzel
.
1987
.
Leitmotifs in natural morphology
.
Amsterdam
:
John Benjamins
.
Embick,
David
, and
Rolf
Noyer
.
2001
.
Movement operations after syntax
.
Linguistic Inquiry
32
:
555
595
.
Fuß,
Eric
.
2005
.
The rise of agreement: A formal approach to the syntax and grammaticalization of verbal inflection
.
Amsterdam
:
John Benjamins
.
Greenberg,
Joseph H
.
1963
. Some universals of grammar with particular reference to the order of meaningful elements. In
Universals of language: Report of a conference held at Dobbs Ferry, New York, April 13–15, 1961
, ed. by
Joseph H.
Greenberg
,
73
113
.
Cambridge, MA
:
MIT Press
.
Halle,
Morris
.
1997
. Distributed Morphology: Impoverishment and fission. In
PF: Papers at the interface
, ed. by
Benjamin
Bruening
,
Yoonjung
Kang
, and
Martha
McGinnis
,
425
450
.
MIT Working Papers in Linguistics 30. Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Halle,
Morris
, and
Alec
Marantz
.
1993
. Distributed Morphology and the pieces of inflection. In
The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger
, ed. by
Kenneth
Hale
and
Samuel Jay
Keyser
,
111
176
.
Cambridge, MA
:
MIT Press
.
Halle,
Morris
, and
Alec
Marantz
.
1994
. Some key features of Distributed Morphology. In
Papers on phonology and morphology
, ed. by
Andrew
Carnie
and
Heidi
Harley
,
275
288
.
MIT Working Papers in Linguistics 21. Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Harbour,
Daniel
.
2003
.
Elements of number theory
.
Doctoral dissertation, MIT, Cambridge, MA
.
Harbour,
Daniel
.
2016
.
Impossible persons
.
Cambridge, MA
:
MIT Press
.
Harley,
Heidi
.
2008
. When is a syncretism more than a syncretism? Impoverishment, metasyncretism, and underspecification. In
Phi theory: Phi-features across modules and interfaces
, ed. by
Daniel
Harbour
,
David
Adger
, and
Susana
Béjar
,
251
294
.
Oxford
:
Oxford University Press
.
Harris,
James
.
1991
.
The exponence of gender in Spanish
.
Linguistic Inquiry
22
:
27
62
.
Harris,
James
.
1995
. The morphology and phonology of Spanish clitics. In
Evolution and revolution in linguistic theory: Studies in honor of Carlos P. Otero
, ed. by
Héctor
Campos
and
Paula
Kempchinsky
,
168
197
.
Washington, DC
:
Georgetown University Press
.
Harris,
James
.
1998
. Enclitic -n in Spanish. In
Romance linguistics: Theoretical perspectives: Selected papers from the 27th Linguistic Symposium on Romance Languages (LSRL XXVII), Irvine, 20–22 February 1997
, ed. by
Armin
Schwegler
,
Bernard
Tranel
, and
Myriam
Uribe-Etxebarria
,
111
127
.
Amsterdam
:
John Benjamins
.
Harris,
James
.
1999
. Nasal depalatalization no, morphological wellformedness : The structure of Spanish word classes. In
Papers on morphology and syntax, cycle one
, ed. by
Karlos
Arregi
,
Benjamin
Bruening
,
Cornelia
Krause
, and
Vivian
Lin
,
47
82
.
MIT Working Papers in Linguistics 33. Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Harris,
James
, and
Morris
Halle
.
2005
.
Unexpected plural inflections in Spanish: Reduplication and metathesis
.
Linguistic Inquiry
36
:
195
222
.
Haspelmath,
Martin
.
1993
.
The diachronic externalization of inflection
.
Linguistics
31
:
279
309
.
Hill,
Virginia
.
2015
.
Allocutive agreement through mesoclisis
.
Paper presented at the 45th Linguistic Symposium on Romance Languages, University of Campinas, 6–9 May
.
Kany,
Charles E
.
1951
.
American-Spanish syntax
.
Chicago
:
University of Chicago Press
.
Kayne,
Richard S
.
1981
. Unambiguous paths. In
Levels of syntactic representation
, ed. by
Robert
May
and
Jan
Koster
,
143
183
.
Dordrecht
:
Foris
.
Kayne,
Richard S
.
1989
. Facets of Romance past participle agreement. In
Dialect variation and the theory of grammar
, ed. by
Paola
Benincà
,
85
103
.
Dordrecht
:
Foris
.
Kayne,
Richard S
.
1991
.
Romance clitics, verb movement, and PRO
.
Linguistic Inquiry
22
:
647
686
.
Kayne,
Richard S
.
2010
. Toward a syntactic reinterpretation of Harris&Halle (2005). In
Romance Languages and Linguistic Theory 2008: Selected papers from ‘Going Romance’, Groningen 2008
, ed. by
Reineke
Bok-Bennema
,
Brigitte
Kampers-Manhe
, and
Bart
Hollebrandse
,
145
170
.
Amsterdam
:
John Benjamins
.
Kiparsky,
Paul
.
2017
. The morphology of the Basque auxiliary: Thoughts on Arregi & Nevins 2012. In
The morphosyntax-phonology connection: Locality and directionality at the interface
, ed. by
Vera
Gribanova
and
Stephanie S.
Shih
,
361
400
.
Oxford
:
Oxford University Press
.
Manzini,
M. Rita
, and
Leonardo M.
Savoia
.
2007
.
A unification of morphology and syntax: Investigations into Romance and Albanian dialects
.
London
:
Routledge
.
Manzini,
M. Rita
, and
Leonardo M.
Savoia
.
2011
.
Mesoclisis in the imperative: Phonology, morphology or syntax?
Lingua
121
:
1101
1120
.
Marantz,
Alec
.
1982
.
Re reduplication
.
Linguistic Inquiry
13
:
435
482
.
Marantz,
Alec
.
1997
. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In
Proceedings of the 21st Annual Penn Linguistics Colloquium
, ed. by
Alexis
Dimitriadis
,
Laura
Siegel
,
Clarissa
Surek-Clark
, and
Alexander
Williams
,
201
225
.
University of Pennsylvania Working Papers in Linguistics 4.2. Philadelphia
:
University of Pennsylvania, Penn Linguistics Club
.
Mare,
María
.
To appear
.
Una nueva mirada sobre la concordancia inesperada en español
.
Revista de Filología Española
.
McCawley,
James
.
1968
.
Concerning the base component of a transformational grammar
.
Foundations of Language
4
:
55
81
.
Minkoff,
Seth A
.
1993
. Plurality, clitics, and morphological merger in Caribbean Spanish. In
Proceedings of the 5th Student Conference in Linguistics (SCIL 5)
, ed. by
V. M.
Lindblad
and
M.
Gamon
,
177
192
.
MIT Working Papers in Linguistics 20. Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Myler,
Neil
.
2013
. Linearization and post-syntactic operations in the Quechua DP. In
Challenges to linearization
, ed. by
Theresa
Biberauer
and
Ian
Roberts
,
171
209
.
Berlin
:
Mouton de Gruyter
.
Nevins,
Andrew
.
2007
.
The representation of third person and its consequences for person-case effects
.
Natural Language and Linguistic Theory
25
:
273
313
.
Nevis,
Joel A.
, and
Brian D.
Joseph
.
1992
. Wackernagel affixes: Evidence from Balto-Slavic. In
Yearbook of morphology 1992
, ed. by
Geert
Booij
and
Jaap
van Marle
,
93
111
.
Dordrecht
:
Kluwer
.
Oltra-Massuet,
Isabel
.
1999
.
On the notion of “theme vowel”: A new approach to Catalan verbal morphology
.
Master’s thesis, MIT, Cambridge, MA
.
Oltra-Massuet,
Isabel
, and
Karlos
Arregi
.
2005
.
Stress-by-structure in Spanish
.
Linguistic Inquiry
36
:
43
84
.
Paradis,
Carole
.
1988
.
On constraints and repair strategies
.
The Linguistic Review
6
:
71
97
.
Parry,
M. Mair
.
1997
.
Preverbal negation and clitic ordering, with particular reference to a group of North-West Italian dialects
.
Zeitschrift für romanische Philologie
113
:
243
270
.
Perlmutter,
David M
.
1971
.
Deep and surface structure constraints in syntax
.
New York
:
Holt, Rinehart and Winston
.
Pires,
Acrisio
.
2006
.
The Minimalist syntax of defective domains: Gerunds and infinitives
.
Amsterdam
:
John Benjamins
.
Raposo,
Eduardo
.
1987
.
Case theory and Infl-to-Comp: The inflected infinitive in European Portuguese
.
Linguistic Inquiry
18
:
85
109
.
Rivero,
María Luisa
.
1994
.
Clause structure and V-movement in the languages of the Balkans
.
Natural Language and Linguistic Theory
12
:
63
120
.
Rizzi,
Luigi
.
1990
.
Relativized Minimality
.
Cambridge, MA
:
MIT Press
.
Roca,
Iggy
.
1986
.
Secondary stress and metrical rhythm
.
Phonology Yearbook
3
:
341
370
.
Rosenblat,
Ángel
.
1946
. Notas de morfología dialectal. In
Estudios sobre el español de Nuevo Méjico. Parte II, Morfología
, ed. by
Aurelio M.
Espinosa
,
105
316
.
Buenos Aires
:
Impr. de la Universidad de Buenos Aires
.
Smith,
Peter W
.
2014
.
Non-peripheral cliticization and second position in Udi and Sorani Kurdish
. .
Terzi,
Arhonto
.
1999
.
Clitic combinations, their hosts and their ordering
.
Natural Language and Linguistic Theory
17
:
85
121
.
Weisser,
Philipp
, and
Elina
Guseva
.
2018
.
Postsyntactic reordering in the Mari nominal domain: Evidence from suspended affixation
.
Natural Language and Linguistic Theory
. 10.1007/s11049-018-9403-6.
Williams,
Edwin
.
1994
.
Remarks on lexical knowledge
.
Lingua
92
:
7
34
.
Williams,
Edwin
.
2003
.
Representation Theory
.
Cambridge, MA
:
MIT Press
.

Supplementary data