The Syntax of Basque Allocutive Clitics ∗

A key piece of evidence in favor of recent proposals that speech act roles are encoded in the syntax comes from languages with “allocutivity”—agreement with non-thematic addressees. This paper analyzes the syntax of allocutive morphemes in the best studied allocutive language, Basque. It is shown that properties of these morphemes, including morpheme order and the way allocutive morphemes condition exponence of neighboring heads can be modeled using standard assumptions about cliticization, head-adjunction and case in Basque. The analysis lends support to a central idea in recent literature on the syntax of speech act roles, namely that addressees are encoded by a nominal element introduced in the left periphery. This paper proposes that, in Basque, this nominal element is overt, with a distribution that obeys principles similar to those for thematic addressees.


Introduction
Following Speas & Tenny (2003) and Sigurðsson (2004), much recent generative work has taken new interest Ross' (1970) performative hypothesis-in its essence, the proposal that speech act roles corresponding to speaker and/or addressee are encoded in the syntax. In its updated form, this approach holds that syntactic objects corresponding to speaker and addressee speech act roles are introduced in designated left peripheral positions, as schematized in (1) (Hill 2007;Baker 2008;Zanuttini 2008;Miyagawa 2012;Haegeman & Hill 2013;Hill 2013;Zu 2013;Slocum 2016;Miyagawa 2017;Zu 2017;Portner et al. to appear). (1) The

neo-performative hypothesis [Speaker … [Addressee … [CP [TP ]]]]
An expectation raised by (1) is that parts of the postulated Speaker and Addressee-related projections will be phonetically realized in some languages. One phenomenon frequently cited as evidence to this effect are languages with agreement with non-thematic addressees-"allocutive agreement" (Miyagawa 2012;. We illustrate this phenomenon in (2), from Basque, where the -k and -n morphemes on the right edge of the auxiliary agree with familiar masculine and feminine addressees, respectively, when these are not arguments of the verb.
(4) Hi-k egin-go d-u-k/-n. you-erg do-irr expl-root-2sg.fam.erg.masc/fem 'You (familiar) will do it.' (5) Hi-ri eman-go d-i-a/na-t. you-dat give-irr expl-root-2sg.fam.masc/fem.dat-1sg.erg 'I will give it to you.' The gender agreement shown in (2)-(5) occurs only with familiar addressees and only in some conservative dialects. In other dialects, the formal/familiar distinction has been lost and the gender-invariant form historically used exclusively with formal addressees is used for all second person addressees (Alberdi 1995). 1 In addition, some Northern and Eastern dialects also mark agreement with formal, non-thematic addressees, which are invariant for gender as in (6) (Oyharçabal 1993;Alberdi 1995). The following discussion will not consider these forms and focus instead on informal agreement which is attested across a broader set of dialects. In addition, in many contemporary dialects, agreement with masculine interlocutors is more broadly used than feminine forms. Many younger speakers with intuitions about masculine agreement forms do not have intuitions about feminine forms. The examples provided below will therefore principally involve masculine forms.

(6)
Northern dialects (adapted from Oyharçabal 1993) Pette-k lan egin d-i-zü. Peter-erg work do.perf expl-root-2sg.form 'Peter has worked.' Two properties of Basque allocutive morphemes suggest that they should be understood as species of vocative expression. First, in addition to addressee morphemes on the auxiliary, some Basque dialects maintain vocative pronouns. These morphemes share with allocutive clitics the property that a gender distinction is made in informal contexts. This is illustrated in (7), where to and no agree with informal masculine and feminine interlocutors, respectively. These pronouns and the addressee agreement morphemes discussed above are the only two contexts in which Basque has grammatical gender. 2 These morphemes are obligatorily separated from clausal material to the right by an intonational break, suggesting that these pronouns may be extra-sentential. As Oyharçabal (1993) notes, unlike clitics cross-referencing arguments, allocutive clitics never co-occur with overt doubles in the absence of such an intonational break.

(7)
To/no,# etorri-ko d-u-k/n, ala? 2sg.fam.masc/fem come-irr expl-root-2sg.fam.masc/fem or 'Hey you, is he/she/it going to come?' 1 Unless otherwise indicated examples are taken from the standard dialect. 2 A reviewer points out that to/no can appear in contexts in which they are not plausibly vocative/allocutive pronouns but rather something closer a presentative marker (Zanuttini 2017). In such cases, it need not co-occur with allocutive agreement. In (i), the finite verb in the root clause, takes no allocutive agreement.
(i) (Larzabal 1968) To! Gure gizonek geldiarazi die plazarat sartu den zaldi-karrosa bat. look our men stop.caus aux plazaabl enter aux horse-carriage one 'Look! Our men have stopped a horse carriage that has entered the plaza.' Many thanks to a reviewer for a discussion of these facts. I set aside the nature of these uses here.
A second property of Basque allocutive morphemes characteristic of vocative expressions cross-linguistically is their sensitivity to clause type (Oyharçabal 1993;Hill 2013). For most speakers, allocutive marking is sharply out in all embedded contexts, as in (8), a property common among vocative expressions cross-linguistically (Hill 2013). 3 (8) Jon-ek esan d-i-k etorri-ko Jon-erg say.perf expl-root-2sg.fam.masc come-irr d-u-[%k]-ela. expl-root-2sg.fam.masc-c 'Jon has said that he will come.' In the analysis below, allocutive morphemes will be viewed as akin to vocative forms, drawing on similar approaches to vocatives cross-linguistically (Hill 2007;Espinal 2013;Haegeman & Hill 2013). (For convenience, this paper follows the convention in the Basque literature of referring to such morphemes (2) as "allocutives".) Specifically, allocutive morphemes will be analyzed as clitics merged in a designated left peripheral Addressee projection. This approach will aide in modeling three recalcitrant sets of facts related to the behavior of these morphemes, which are introduced in the following section.

Problem 1: Morpheme order
The basic morpheme order of Basque auxiliaries is traditionally described as following the template in (9) (Cheng & Demirdache 1993;Laka 1993;Albizu 1997;2002;Arregi & Nevins 2012). 4 This template has no formal status in the analysis to follow, but rather is used here as a descriptive tool, intended to represent the fact that these morphemes, when present, must appear in the order given. The brackets marking elements in first position are intended to represent the fact that this position may be filled either by an absolutive clitic, or, when no absolutive clitic is available, an expletive morpheme (Albizu 2002;Arregi & Nevins 2012). These facts are spelled out in section four. Allocutive morphemes may occupy one of two slots in the template in (9)-either to the right of the ergative number morpheme, as in (10), or to the left of the ergative person morpheme, as in (11) and (12).
(10) Egin-go d-i-te-k. do-irr expl-root-pl.erg-2sg.fam.masc 'They will do it.' (11) Egin-go d-i-a-gu. do-irr expl-root-2sg.fam.masc-1pl.erg 'We will do it.' 3 Hualde et al. (2003) notes that some younger speakers do accept allocutive agreement in embedded clauses. For these speakers, this availability doesn't appear sensitive to assertion/MPU interpretation (as in Mainland Scandinavian embedded V2) (Hooper & Thompson 1973;Simons 2007). For these speakers the availability of embedded allocutive doesn't appear to be conditioned by the complementizer type as described by Elordieta & Haddican (2018) for ba-insertion and VP-Neg-Aux word orders. I set aside this set of speakers in the following discussion. 4 Our discussion will set aside absolutive and dative number morphemes and also *edin forms.

(12)
Egin-go d-i-a-t. do-irr expl-root-2sg.fam.masc-1sg.erg 'I will do it.' Importantly, in 2pl ergative contexts like (13), the ergative person morpheme appears to the left of the ergative number morpheme. Assuming unique merged positions for the allocutive and ergative morphemes in (10)-(13), then the morpheme orders in (10)-(12) seem to require movement of either the allocutive morpheme, the ergative number morpheme or the ergative person morpheme. (The only ergative person/number combination in which both person and number are overt is 2pl, with which allocutive clitics do not co-occur.) (13) Egin-go d-u-zu-e. do-irr expl-root-2.erg-pl.erg 'You all will do it.' Indeed, there is another set of contexts, called "ergative displacement" constructions in the Basque literature, in which morphemes marking ergative person may move independently of the ergative number marker -(t)e-as illustrated in (14). In ergative displacement, the leftmost slot in the auxiliary may, under conditions to be described later, be occupied by ergative person morphemes rather than an absolutive morpheme. In (14), the ergative person allomorph zen-appears in first position, while the ergative plural marker, -(t)e-, appears in its usual position, further to the right in the auxiliary.
(14) Egin zen-u-te-n. do.perf 2pl.erg-root-pl.erg-pst 'We will do it.' While ergative displacement appears in all Basque dialects, dialects with allocutive displacement are much more restricted. 5 In the general case, then, parsimony favors an analysis by which the position of the allocutive morpheme in (10)-(12) is invariant, and the position of the ergative person morphemes in (11) and (12) suggests an additional context in which these morphemes may undergo displacement. A partially unified analysis of these two ergative morpheme displacement contexts will be presented later.
An additional challenge posed by the morpheme order of allocutive clitics concerns the fact that they always appears to the left of the auxiliary-final -(e)n morpheme traditionally taken to be a tense marker (Laka 1993).
(15) Egin-go z-i-te-a-n. do-irr expl-root-pl.erg-2sg.fam.masc-pst 'They were going to do it.' A well-described property of Basque auxiliaries is that they exhibit a Mirror Principle effect (Baker 1985), whereby the canonical linear order of morphemes is the inverse of their (bottom-up) merged order, on standard assumptions about argument structure and 5 I'm grateful to an anonymous reviewer for a discussion of these facts. An example of allocutive displacement to the left edge of the auxiliary appears in (i) from the Donibane-Lohizane dialect.
(i) (de Yrizar 1997: 284) ∅-u-en 2sg.fam-root-past 'He/she/it was.' the functional sequence (Laka 1993). From the perspective of Speas & Tenny's (2003) proposal, where morphemes marking non-thematic addressees are merged in a C-field position, and if, indeed, the -(e)n morpheme truly corresponds to a T head, then the ordering of the allocutive morpheme to the left of -en is surprising in that it suggests a merged position below T. An account of these facts is developed below.

Problem 2: Effects on root vowels
Basque is traditionally described as having a have/be alternation (Ortiz de Urbina 1989;Arregi 2004). In unaccusative contexts without allocutive morphology, the auxiliary takes a form of the copular verb izan, 'be', and in transitive non-allocutive contexts takes a form of possessive 'have', edun, as illustrated in (16) and (17). In (16), the auxiliary root vowel is -a-, marking izan, 'be', while in (17) The foregoing description holds for contexts without allocutive clitics. In unaccusative contexts with allocutive clitics, however, the root vowel -u-appears, as shown in (2), repeated here (Rebuschi 1981;Oyharçabal 1993;Albizu 2002;Arregi 2004;Arregi & Nevins 2012;Torrego 2013;Torrego & Fernández 2016). 6 (18) Kotxea garestia izan-go d-u-k/n. car.abs expensive cop-irr expl-root-2sg.fam.masc/fem 'The car is going to be expensive.' The allocutive clitics in (18) are reminiscent of ergative clitics both in their exponence (see (4)), and in the fact that they co-occur with auxiliary switch. Nevertheless, as Albizu (2002) notes, allocutive clitics in intransitive contexts are unlike thematic ergative clitics in that, in most dialects, the allocutive clitic does not undergo ergative displacement, as Plausibly related to the above set of facts is an additional root vowel alternation triggered in monotransitive contexts. We saw previously that monotransitive contexts without an allocutive clitic determine 'have', i.e. with an auxiliary root vowel -u-, as in (17). In the presence of an allocutive clitic, however, the root vowel in monotransitive contexts is not -u-but rather -i-, as in (21).
(21) Egin-go d-i-a-t. do-irr expl-root-2sg.fam.masc-1sg.erg 'I'm going to do it.' The -i-morpheme in contexts like (21) has traditionally been taken to be related to the -i-that appears in the root in ditransitive applicatives like (22) (Rebuschi 1981;. These facts, alongside the fact that exponence of allocutive forms is similar to that of thematic dative forms (see (5)) suggest the possibility that allocutive clitics may be related to dative clitics.
give-irr expl-root-2sg.dat-1sg.erg 'I'm going to give it to him/her/it.' The nature of -i-in these contexts will be addressed shortly. For the moment, let us observe that allocutive clitics differ from recipient dative clitics in several ways. First, as noted above, allocutive clitics appear to the right of ergative plural morphemes in third person ergative contexts, while clitics marking thematic datives appear to the left, as illustrated in (23).

(27)
Adapted from Albizu (1997) Peru-k ni kale-an ikus-i n-a-i-k. Peru-erg 1sg.abs street-in see-perf 1sg.abs-bv-root-2sg.fam.masc 'Peru has seen me in the street.' To summarize, allocutive clitics are superficially similar to ergative and dative clitics in their exponence and in the ways they condition exponence of the 'have' root vowel -u-and the -i-morpheme of dative contexts. Albizu's (1997;2002) observations, however, suggest that allocutive clitics must be analyzed as partially distinct from "true" ergatives and recipient datives at least at certain derivational levels.

Problem 3: Effects on the first position expletive morpheme
A third property of allocutive clitics to be accounted for is their effect on the exponence of a morpheme appearing on the left edge of the finite verb. As noted briefly in section 3.1, the first position slot may be occupied by one of two kinds of elements. One possible denizen of this position are first and second person clitics, generally absolutive clitics.
(We return to the special case of ergative clitics in this position later.) Ikusi-ko n-a-u-zu-e. see-irr 1sg.abs-bv-root-2.erg-pl.erg 'You all will see me.' In third person absolutive contexts, such as (29), this slot is occupied by one of five forms-{d-, z-, l-, b-, ∅}, whose selection is determined by a complex set of factors including tense and mood.
give-irr expl-root-3sg.dat-pl.erg-pst 'They were going to give it to him/her/it.' In the Basque literature, there are two main approaches to these facts. One approach, by Lafon (1943: 379) and Ortiz de Urbina (1989), is that this latter set of forms-{d-, z-, l-, b-, ∅}-are third person absolutive markers with allomorphs conditioned by tense, mood etc. A second approach, which has been the consensus in recent literature, is that third person absolutive markers-like third person ergative-are ∅ or absent entirely (Trask 1977;Laka 1993;Albizu & Eguren 2000;Fernández & Albizu 2000;Albizu 2002;Arregi & Nevins 2012). On this approach, {d-, z-, l-, b-, ∅} are typically treated as expletive elements to fill a phonologically empty position (Laka 1993;Gómez & Sainz 1995;Arregi & Nevins 2012). Section four develops an account of these morphemes close in spirit to this latter group of proposals. The contexts determining selection among these expletive form-{d-, z-, l-, b-, ∅}are often said to form a natural class of tense and modality (Lafon 1943;Trask 1981;Laka 1993). In realis contexts, the morpheme surfaces as d-in present tense contexts and z-in past contexts. (30)-(32) show that this alternation is constant across unaccusative, monotransitive and ditransitive contexts (the relevance of which will be made clear shortly). 8 b. Egin-go z-u-te-n. do-irr expl-root-pl.erg-pst 'They were going to do it.' [past] give-irr expl-root-3sg.dat 'He/she/it will give it to him/her/it.' [present] b. Eman-go z-i-o-n.
give-irr expl-root-3sg.dat-pst 'He/she/it was going to give it to him/her/it.' [past] In irrealis contexts, the expletive morpheme appears as l-in present tense contexts and z-in the standard dialect as in (33) The importance of these facts for our analysis of allocutive morphemes relates to a fact less commonly discussed in the formal literature on allocutives, namely that exponence of the expletive morpheme is conditioned not just by tense and modality, but also by the presence of an allocutive clitic (Rebuschi 1984;Albizu 2002). This pattern, described first by Rebuschi (1984: 621) as far as I am aware, is as in (34). 9 (34) Rebuschi's generalization In present tense realis contexts, the allocutive clitic determines expletive z-rather than d-if and only if it does not trigger a change in the verb root. 9 Rebuschi's generalization describes the behavior of expletives in three kinds of contexts.
The first is in ditransitive constructions. In section 3.2, it was noted that in intransitive and monotransitive contexts, the presence of an allocutive clitic triggers the appearance of the root vowels -u-and -i-respectively. In present tense contexts, the expletive morpheme is unchanged-d-. This is illustrated in (35) and (36). In ditransitive contexts with analytic verbs, the root vowel is -i-in non-allocutive contexts, and no further root-vowel change operation is applicable. Here, the addition of an allocutive clitic has no effect on the auxiliary root, but rather changes the auxiliary-initial expletive morpheme from d-to z-. (Our discussion will henceforth follow the convention in the Basque literature of referring to the d-to-z-change as "spirantization".) 10 give-irr expl-root-3sg.dat-1sg.erg 'I will give it to him/her/it.' give-irr expl-root-3sg.dat-2sg.fam.masc-1sg.erg 'I will give it to him/her/it.' [with allocutive clitic] A second context in which spirantization applies in present tense allocutive forms is with synthetic verbs-the closed class of verbs on which tense and agreement are marked on the verb rather than a separate auxiliary. In such contexts, no auxiliary is present and hence, no auxiliary root vowel change is applicable. As shown in (38) and (39), in the presence of an allocutive clitic, no change in the root applies, and the expletive morpheme is z-as in past tense contexts. In addition, in at least some dialects, the presence of an allocutive clitic does not trigger spirantization in the absence of a root alternation (Albizu 2002). Such is the case in the standard dialect for forms of edin the intransitive possibility modal root. The root undergoes no change in allocutive clitic contexts, and spirantization fails to apply as illustrated in (i). At least in some dialects, therefore, Rebuschi's generalization is not properly biconditional, but rather a simple conditional, such that if spirantization applies, a root change does not apply, but not vice-versa.
As (Rebuschi 1984: 621) notes, an additional context in which no root change applies are dative intransitive contexts like (41). Rebuschi (1984: 621) suggests that in such cases, spirantization applies vacuously since there is no initial d-for the change to affect. These facts are considered below.
A question that arises at this point is what the relation is between the aux-initial z-triggered by allocutive clitics in Rebuschi's generalization contexts, and the homophonous element triggered in past tense contexts. One possibility is that the homophony is accidental, i.e. that z-spells out different sets of features in the two contexts. A disadvantage to this solution, however, is that, in many dialects, the presence of an allocutive morpheme co-occurs with another phenomenon found in past tense forms. Examples (38) and (39), show that with synthetic verbs like etorri, 'come' and ibili, 'walk', the presence of an allocutive clitic triggers both spirantization and a change in the "buffer vowel" between the verb root and expletive morpheme (Albizu 2002). 12 Whatever the nature of this buffer vowel alternation, which will not be considered in detail here, a parsimonious approach to these two alternations-spirantization and -a-→ -e-buffer vowel change-will require us to assume that the homophony is not accidental and that a more general property is responsible for both of these alternations. An analysis of these facts is developed in the next section.

A model of morpheme order
In recent work, Miyagawa (2012; analyzes Basque allocutive morphemes as the reflex of an Agree relation between an allocutive probe and a silent Hearer morpheme, merged in the specifier of higher speech act-related projection, S(peech)A(ct)P. Based on a structure initially proposed by Haegeman & Hill (2013), Miyagawa likens the structure in which speech act participants are introduced to a high applicative structure, in which the speaker is parallel to the agent, the addressee to the applied recipient and CP to the theme. The allocutive probe is merged on C and, after raising to the SA head c-commanding the Hearer morpheme, probes and agrees with the latter. This proposal is illustrated in (42) From the perspective of Miyagawa's proposal, allocutive morphemes in Basque might be thought of as a species of "vocative agreement". That is, given the presence of vocative case in natural language, and from the perspective of approaches to case and agreement as different morphological manifestations of an Agree operation, one imagines that vocative agreement might also be overtly manifested in human language. Such an approach may turn out to be correct for other allocutive languages, but it is not well suited to modeling the Basque facts introduced above for reasons to be made clear shortly. (See Portner et al., to appear, for such an approach to a class of honorific morphemes in Korean.) Miyagawa's model is within a tradition of work analyzing the person morphemes on Basque verb forms as the reflex of an agree operation (Etxepare 2006;Rezac 2008;Béjar & Rezac 2009). A competing approach to these morphemes, to be adopted in the analysis to follow, is that they are instead obligatory clitics, doubling a possibly silent argument (Laka 1993;Oyharçabal 1993;Preminger 2009;Arregi & Nevins 2012). The main advantage of a clitic analysis of person morphemes has to do with the absence of Head Movement Constraint (HMC) violations. Let us assume, more or less standardly, the structure in (43) showing the merged positions of Basque ergative, dative and absolutive arguments and the three sources of case for thematic arguments. As discussed in section 3, morphemes marking ergative, absolutive and dative arguments appear on the auxiliary verb. On an approach that takes these morphemes to reflect exponence of Agree+feature valuation, the morphemes corresponding to dative person and absolutive person will presumably be the probes Appl and v, which, after probing, will raise via head movement to the position of the auxiliary. Importantly, modals and aspectual morphemes do not appear adjoined to the auxilary. Modals are free morphemes that, in affirmative declarative sentences like (44), appear between the main verb and the auxiliary. Aspectual markers appear affixed to the main verb, as in (44), repeated here.
give want expl-root-3sg.dat-erg.pl 'They want to give it to him/her/it.' Kotxeak garestiak iza-ten d-it-u-k. car.abs.pl expensive cop-imperf expl-abs.pl-root-2sg.fam.masc 'Cars are often expensive.' Assuming the structure in (43), for the absolutive and dative morphemes to raise to the position of the auxiliary, in T or higher, they will need to skip over the intervening modal and aspectual heads in apparent violation of the HMC. If, on the other hand, dative and absolutive morphemes on the auxiliary raise as clitic XPs, then the absence of HMC effects is expected.
The evidence from morpheme ordering just presented supports a clitic approach to absolutive and dative person morphemes, specifically. In the analysis to follow, allocutive morphemes will also be treated as clitics, in view of the similar behavior of these two classes of forms in terms of exponence and conditions on allomorphy, as discussed in section 2. Specifically, let us assume that clitics are elements of category D, merged in a "KP" structure with the arguments they double, as in (46) (Nevins 2011 Following Matushansky (2006), let us assume that clitics raise as XPs to the spec of their hosts and then undergo m-merger-post-syntactic adjunction to their host heads-as in (47) Let us assume, further, that clitics "tuck in" (Halle & Marantz 1994). That is, for a given clitic host X, m-merger will adjoin the clitic below the node formed by adjunction of the head of X's sister, as in (48).
Let us further assume the sequence of heads in (49), which, along with their associated clitics, will form auxiliaries through successive head adjunction, with adjuncts uniformly linearized to the left of their hosts. Here, Fin is the merged position of complementizers. Following Bianchi (2003), we take Fin to be the locus of several speech act-deictic properties including speech act time and participants. The Speaker and Addressee heads will be the positions in which morphemes encoding speaker and addressee speech act roles are introduced. This structure, then, is akin to Speas and Tenny's and Miyagawa's sequence of speech act-related heads described above, but no position is taken here on whether this configuration should be understood as parallel to the thematic structure in which agents, themes and recipients/applied objects are introduced. In addition, unlike in Miyagawa's analysis, these heads will be merged below the position of the complementizer, Fin, for reasons to be made clear shortly. 14 13 As noted above, allocutive clitics never co-occur with an overt double (Oyharçabal 1993). One possible approach to this fact is that these clitics differ from argumental clitics in not being merged in a "big DP" structure like (i), but rather in being merged as DPs. Why these clitics should be exceptional in not being able to embed within a KP structure as in (46), would call for some further explanation. Another possibility is that the clitic doubles a silent Addressee operator (Baker 2008;Zanuttini 2008). We set aside these issues in the remaining discussion. 14 The label "Addressee" is used here instead of Speas and Tenny's "Hearer" label to make clear that the relevant distribution of morphemes is conditioned by properties of addressees, specifically, rather than ratified or unratified "hearers" in a discourse context.)

(49) Heads involved in Basque auxiliary formation [Fin [Speaker [Addresee [T [Mod [Aux … ]]]]]]
In addition, following Laka (1993), Arregi & Nevins (2012) and Gómez & Sainz (1995), the foregoing discussion has assumed that, in the absence of a suitable person clitic, the first position slot is filled with an expletive morpheme. Let us assume that this reflects an EPP property of Fin, akin to standard approaches to V2 (Chomsky 2000;Roberts 2004;Holmberg 2015;Leu 2015). Let us state this requirement as in (50) for the time being. In particular, (50) will be applicable in third person absolutive contexts, where there are no third person clitics to occupy the first position slot (Arregi & Nevins 2012 A question that arises is whether (50) applies before or after m-merger in the phonological component. The fact that absolutive first and second person clitics can satisfy (50), as in (28) (repeated here, as (51)), thereby blocking expletive insertion, suggests that either (50) applies before m-merger or that absolutive clitics do not undergo m-merger in this context. A consequence of these assumptions is that the "first position" slotspec, FinP-will necessarily be filled by either an absolutive morpheme or an expletive element.
(51) Ikusi-ko n-a-u-zu-te. see-irr 1sg.abs-bv-root-2.erg-pl.erg 'You all will see me.' We saw previously that allocutive clitics appear to the left of the auxiliary-final -(e)n morpheme traditionally taken to be a tense marker in T, as shown in (15), repeated here (Laka 1993). Again, if allocutive morphemes are merged in a high Addressee position, then the order of -(e)n to the right of the allocutive clitic is surprising given that auxiliary morphemes otherwise mirror their merged orders quite consistently (Laka 1993). Let us instead follow Arregi & Nevins (2012) in taking auxiliary-final -(e)n to be not an element in T, but a past tense complementizer, in Fin. As Arregi & Nevins (2012) note, this proposal is supported by the fact that past tense -(e)n and aux-final declarative complementizers are in complementary distribution. Let us assume, further, that the presence of a past tense feature on Fin also determines spell out of the z-expletive morpheme in spec, Fin.
(52) Egin-go z-i-te-a-n. do-irr expl-root-3pl.erg-2sg.fam.masc-pst 'They were going to do it.' Basque clitics only appear in finite contexts-never in non-finite embeddings. Let us follow Arregi & Nevins (2012) in taking his property to reflect a requirement that they be licensed by adjunction to Fin, which will be discussed in greater detail shortly. The fact that absolutive clitics appear only in first position suggests that Fin is their landing site. Ergative clitics will arrive in Fin by first adjoining to T-their source of case-and subsequently to Fin through successive head adjunction. Following Laka (1993), let us assume that dative clitics target a modal head, silent in some contexts but overt as -ke in possibility modal contexts like (53 Let us now consider how these assumptions fare in modeling morpheme order in relevant contexts. (56) illustrates the proposed structure for the auxiliary in (55), with a 3pl.erg clitic. The structure represents the application of successive cyclic adjunction of the heads in (54), and adjunction of clitics to their respective landing sites. Again, adjoined heads are uniformly linearized to the left of their hosts, and clitics "tuck in", forming the adjunction node closest to their hosts. The arrow in (56)  Recall from Section 3.1 that first person ergative clitics appear unexpectedly to the right of allocutive clitics as in (11), repeated here. Exploiting further the neo-performative assumption of designated positions for Speaker and Addressee speech act roles, let us assume that contexts with first person ergative clitics differ minimally from the representation in (56) in that the ergative clitic raises to Speaker. If this involves cyclic movement of the clitic (stopping off in T), then m-merger must be allowed to fail to apply, such that the ergative clitic can raise from Spec, T. (58) shows the proposed structure for the auxiliary in (57) with a first person ergative clitic. It represents a sequence of movement steps identical to those in (56), with difference that the ergative clitic -gu has raised to Speaker. A question raised immediately by this approach is whether clitics cross-referencing second person thematic arguments raise to Addressee parallel to the way that first person ergative clitics target Speaker. More generally, the proposal that person morphemes move to designated Speaker and Addressee projections leads to the expectation that the person hierarchy in Basque should map to linear order, at least for the relative order of first and second person morphemes (for which Basque unambiguously has person clitics).
In the standard dialect, first and second person dative morphemes do not undergo further displacement to Speaker/Addressee. That is, in terms of present proposal, they always appear adjoined to the Aux-internal Mod head. Evidence that first position dative clitics do not raise to Speaker comes from modal contexts like (59), with the verb ezan, where all dative clitics, regardless of person, appear to the left of the modal particle, -ke, as in (59), rather than to the right, where they would appear if they had raised to Speaker. Dative clitics, therefore, cannot be used to examine person effects on morpheme placement.
give expl-bv-root-1.sg.dat-mod-2.sg.erg 'You can give it to him/her/it.' In addition, absolutive clitics, regardless of person, must occupy the first position slotwhat has been taken here to be spec, Fin. The surface position of absolutive clitics, like dative clitics, is therefore insensitive to person hierarchy effects and cannot be used to infer evidence of movement of first and second person morphemes to Speaker and Addressee respectively. The only person clitics left, therefore, that are suitable for inferring movement to Speaker/Addressee are ergative clitics, and the only head that intervenes overtly between the position of the third person clitic and the first person clitic, to distinguish these two positions on the surface, are allocutive morphemes, as discussed above. There is therefore no way to test whether thematic second person ergative clitics raise from T to Addressee, since this movement will necessarily be string vacuous. 16 A question unaddressed so far is why ergative but not dative clitics are able to raise to Addressee. One possibility is that the difference between these two classes of morphemes has to do with the nature of the way dative and ergative arguments receive case. Let us follow the consensus in the Basque literature in taking dative case in Basque to be inherent (Oyharçabal 1992;Rezac 2008). Ergative case in Basque, however, appears to be structural, assigned by T (Ortiz de Urbina 1989;Artiagoitia 2001;Rezac et al. 2014). Suppose, further, that T inherits its person features from Speaker/Addressee in the spirit of Chomsky (2008). From this perspective, one possible understanding of the movement of first person ergative clitics to Speaker is that, in such contexts, Speaker withholds its person feature, and probes the first person ergative clitic directly. (See Legate, 2011, for a likeminded proposal.) The fact that dative and absolutive clitics, which are not probed by "T-C", do not raise to Speaker/Addressee is accounted for under this approach.

Auxiliary root alternations and Rebuschi's generalization
We saw previously that, in unaccusative contexts, the presence of an allocutive clitic triggers a change in the auxiliary root, -a-→ -u-.

(60)
Allocutive clitics trigger -u-roots in unaccusatives auxiliaries a. Kotxea garestia izan-go d-a. car.abs expensive cop-fut expl-root 'The car is going to be expensive.' (without allocutive clitic) b. Kotxea garestia izan-go d-u-k. car.abs expensive cop-fut expl-root-2sg.fam.masc 'The car is going to be expensive.' (with allocutive clitic) The fact that 'have' may appear in intransitive contexts suggests that the have/be alternation is not conditioned directly by argument structure (Arregi 2004;Arregi & Nevins 2012). Rather, the distribution of Basque have/be is reminiscent of person-sensitive have/be systems in some Romance varieties (D'Alessandro & Roberts 2010). For Basque, the Obligatory Case Parameter (Bobaljik 1993;Laka 1993) provides a useful way of implementing this idea. For ergative languages, this holds (in updated parlance) that if one source of structural case is needed, this will be the lower probe "v". If a second source of structural case is needed, this will be the higher probe "T". In allocutive clitic contexts, this means that T will be the source of case for the allocutive clitic. On this approach, T will have to probe the allocutive clitic after raising to the Addressee projection. 16 A set of facts that is problematic for the proposal that second person ergative clitics raise to Addressee concerns the behavior of second person plural ergative contexts like (i), repeated here.
(i) Egin-go d-u-zu-e. do-irr expl-root-2.erg-pl.erg 'You all will do it.' If the second person ergative morpheme can raise independently of the ergative plural morpheme, -e as proposed in section 3.1, and if the person morpheme raises to Addressee, then, under the assumptions outlined above, the opposite linear order of these two morphemes is expected, that is, -e > -zu, rather than the attested -zu > -e. An alternative understanding of these facts is that both of these ergative forms raise to Addressee together in the order -zu > -e. This would leave unexplained, however, why both ergative person and ergative number morphemes cannot raise together in ergative displacement contexts, to be discussed later. (62) have/be in Basque Edun 'have' appears when T* is a probe; izan, 'be' appears otherwise.
Something more is needed to account for the fact that allocutive clitics trigger the root vowel -i-in transitive contexts.
give-irr expl-root-3sg.dat-1sg.erg 'I'm going to give it to him.' This suggests that Addressee is a species of applicative head capable of supplying an extra source of case when needed (Miyagawa 2012;Haegeman & Hill 2013). Unfortunately, for our purposes, it won't do to say that -i-spells out an applicative head directly (pace Elordieta 2001; Fernández 2015b; and Fernández 2015a), for two reasons. The first concerns a morpheme order problem: if Addressee could spell out as -i-, this morpheme should show up to the right of the allocutive and pl.erg morphemes rather than to the left, contrary to fact. Second, as noted by Arregi & Nevins (2012), for (non-allocutive) applicative contexts like (64), if the -i-element spells out an applicative head merged TPinternally, and if Basque auxiliaries are assembled via head-movement, then, a movement step adjoining the -i-morpheme to auxiliary seems inexorably to require a Head Movement Constraint violation (Travis 1984). In particular, (65) and (66) show that both modals and aspectual morphemes linearly intervene between the auxiliary containing the -i-element and the verbal shell. (For the moment, the -i-element is agnostically glossed as dat.) give want expl-dat-3.sg.dat-1.sg.erg 'I want to give it to him/her/it.' give-irr expl-dat-3.sg.dat-1.sg.erg 'I will give it to him/her/it.' These facts, then, suggest that -i-does not spell out an applicative head directly, but instead is determined by a vocabulary insertion rule similar to that proposed in (62), inserting -ionly in the presence of case-assigning T and Applicative heads. In other words, extending Arregi's (2004) and Arregi and Nevins' (2012) approach to Basque have/be, a plausible approach to -i-roots is that exponence of auxiliary roots is conditioned by phi-agreement more generally (D'Alessandro & Roberts 2010). Let us state the rule for -i-as in (67)  Still to be explained is the relationship between the allocutive clitic and spirantization of the expletive morpheme. Recall Rebuschi's generalization, repeated here. (68)

Rebuschi's generalization
In present tense contexts, the allocutive clitic determines z-rather than d-(spirantization) if and only if it does not trigger a change in the verb root (Rebuschi 1984: 621).
The contexts where spirantization (z-) is triggered in the presence of an allocutive clitic are those where no root vowel change is possible, i.e. no form exists. A plausible understanding of these facts from the perspective of our proposal is that probing of the allocutive clitic by T* or Addressee/Applicative* is lexically blocked in these contexts, that is, there is no vocabulary item mapping to this feature set. Let us assume that in these cases, the allocutive morpheme is instead probed by Fin itself (cf. Rizzi 1997;Carstens 2003;Chomsky 2008;Haegeman & van Koppen 2012). That is, Fin is a source of case for the allocutive clitic only when other (lower) sources of case are not available, in the spirit of the Obligatory Case Parameter. Importantly, unlike the Addressee head, Fin is not applicative-like in the sense that it does not introduce an extra argument/participant. Let us recall, now, that the other context in which expletive z-appears-past tense forms with the past tense specified complementizer -(e)n-also presumably involves probing by Fin, which must agree in its tense value with T, the locus of tense. 18 This agreement is illustrated in (69). In present tense contexts, the past tense specified Fin head, -(e)n is not merged and no agreement between Fin and T applies. 19 (i) a. …erori d-ela. …fall.perf aux-c.decl '…that he/she/it has fallen.' b.
…erori z-ela. …fall.perf aux-c.decl '…that he/she/it fell.' This plausibly reflects a vocabulary insertion rule that spells out -(e)la in all embedded declarative contexts, regardless of tense specification. 19 As a reviewer notes, in past tense allocutive contexts in which -(e)n is present, Fin will need to agree with both the past tense feature on T, and the person features on the allocutive morpheme. From this perspective, what the two contexts determining expletive z--past tense forms and Rebuschi's generalization contexts-have in common, apart from the appearance of expletive z-, is that Fin is a probe. In past tense contexts, it will agree with the past tense value on T and in allocutive contexts it agrees with the allocutive clitic. 20 The distribution of expletive {d-,z-} can now be summarized as in (70). 21 (70) Vocabulary insertion rules for expletive morphemes in Basque auxiliaries a. z-↔ Expl --Fin [uF] b. d-↔ Expl --elsewhere To summarize, the proposal developed so far models three effects of allocutive morphemes on auxiliary morphology: (i) their effect on ergative clitic placement; (ii) their effect on auxiliary roots; and (iii) their effects on first-position expletives in Rebuschi's generalization contexts. Central to this approach is the proposal that allocutive morphemes like other person morphemes on the auxiliary are clitics that participate in phi-agreement. Moreover, the distribution of allocutive forms is similar to that of thematic addressees, and modelable using assumptions standardly employed in analyzing case and cliticization for thematic arguments in Basque. A remaining challenge for the analysis, however, concerns the exceptional behavior of dative intransitive auxiliary forms to be addressed in the next subsection. 20 A reviewer raises a concern about this proposal in connection with the fact allocutive morphemes are not possible in the presence of bipersonal forms. That is, in (ia), with both a first and second person arguments, it is not possible to add an allocutive clitic (ib).
(i) a. Ikus-i za-it-u-t. see-imperf 2.abs-abs.pl-root-1sg.erg 'I have seen you.' [without allocutive clitic] b. *Ikus-I za-it-i-a-t. see-imperf 2.abs-abs.pl-root-2sg.fam.masc-1sg.erg 'I have seen you.' [with allocutive clitic] The reviewer suggests that a plausible approach to this fact is that there is a single locus of person licensing, namely T, which cannot license more than two person features. If, on the other hand, Fin can the source of case in Rebuschi's generalization contexts, then some other explanation is needed for the impossibility of (ib). I take the unavailability of (ib) to reflect a more general restriction on allocutive morphemes in the context of second person thematic clitics (Oyharçabal 1993, a.o), but remain agnostic about the nature of this restriction. See Oyharçabal (1993), for an analysis of this restriction as a principle B effect. 21 Additional assumptions will be required to explain exponence of other expletive morphemes including the l-morpheme of irrealis contexts. A detailed analysis of these forms is beyond the scope of this paper.

Dative intransitives
A problematic set of facts for the analysis developed so far concerns dative intransitive constructions such as (71 In particular, the auxiliaries in (71) pose two problems. One relates to the fact that the allocutive clitic in (71b) triggers no change in the form of the auxiliary root. Specifically, the addition of the allocutive clitic in (71b) should require agreement with T*-the next source of case available following the Obligatory Case Parameter, and this, together with Appl*, as required given the presence of the dative clitic, should trigger a spell out of the auxiliary root as -i-pursuant to (67). A second problem concerns the fact that the auxiliary root is the leftmost element in the auxiliary in apparent violation of (50). Let us consider these two issues in turn.
The consensus in the Basque historical literature is that auxiliary forms like (71b) reflect a combination of the izan root -a-with an incorporated dative marker or "dative flag" -(k)i-, which precedes the dative (Lafon 1943;Trask 1981;. Palatalization of an initial d-before a high front vowel also marking dative contexts yielded the initial z-attested in the standard and most central and eastern dialects, and [x]/[j] in western dialects. This proposal is illustrated in (72). 22 (72) *d-i-a-ki-t > zait expl-dat-root-df-1sg.dat In addition, -ki-appears in a broader set of dialects, including the standard, in dative contexts with synthetic verbs, as in (73). (73) d-a-tor-ki-t expl-bv-come-df-1sg.dat 'He/she/it is coming to me.' Let us assume that the -i-element in (71) is the same element as -ki-in (73). Specifically, the analysis to follow will draw on Trask's (1981) and Fernandez's (2015a) proposal that -(k)iis a preposition that introduces some dative arguments and incorporates into the auxiliary. 23 The idea that the -i-element in contexts like (71) represents a dative-introducing head low 22 Forms close to the reconstructed form in (71) appear in the contemporary dialect of Huarte-Arakil, as in (i).
give-irr expl-root-3sg.dat-1sg.erg 'I'm going to give it to him/her/it.' in the functional sequence-either a P or Appl head-has been proposed by several authors (Trask 1981;Elordieta 2001;Fernández 2015b;a). As noted by Arregi and Nevins (2012), an important obstacle to such an analysis of this -(k)i-as a low P/Appl head concerns morpheme order. If this applicative morpheme is merged in a position below the auxiliary, then, on this approach, this element will need to head raise across intervening aspectual and modal morphemes in apparent violation of the Head Movement Constraint (Travis 1984).
(See discussion of (65) and (66) above.) Second, if-as assumed in all syntactic approaches to Basque auxiliary formation-auxiliaries are formed by successive head-adjunction, it is hard to understand why such an applicative morpheme should surface to the right rather than to the left of the (higher) auxiliary root -a-in dative intransitive contexts like (72). The fact that -(k)i-is able to raise past intervening heads suggests that it raises as a clitic, like the person morphemes described above. Let us assume, in particular, that -(k) i-reflects the presence of a PP-doubling clitic, akin to French/Italian -y/-ci (Kayne 1975;, that raises and adjoins to the auxiliary. This movement step will be independent of the step that raises the dative clitic, which will move to its host, ModP, in the usual way. Specifically, let us assume that -(k)i-undergoes m-merger and head-adjunction in the way described above for other clitics, to a position here labeled F, just above the auxiliary root, as in (74). This latter assumption is required in order to express the fact that -(k)i-surfaces to the right of the auxiliary root, (t)za-.
The distribution of the auxiliary root (t)za-, like that of its competitors -a-, -u-and -i-can now be stated in terms of the different phi-probes on the auxiliary. Specifically, (t)za-is inserted in the presence of case-assigning P* and v* heads. When an allocutive clitic is present in dative intransitive contexts (like (71b)), the allocutive clitic will be probed by T* in the usual way, i.e. with the total inventory of phi-probes {v*, P* T*}. The root will spell out as (t)za-rather than -u-, as desired, since the former is the more specific context. 24 The rules governing the spellout of auxiliary heads can now be summarized in their final form in (75).

(75)
Vocabulary insertion rules for Aux (final version) A second problem posed by dative intransitive constructions concerns the fact that the auxiliary root appears word-initially in present tense third person contexts, in apparent violation of the expletive insertion rule in (50). Let us assume, following  suggestion, that expletive d-is inserted as usual and subsequently deleted in phonology. Such a phonological process might be related to the fact that, in Basque there are no *[dts] or *[ds] strings-the sequence that would otherwise be produced by combining expletive d-with the -(t)za auxiliary root. As a reviewer notes, however, such an approach does not explain why insertion of a buffer vowel that seems to appear in past tense forms like (76a)  On the approach developed in the previous section, expletive z-is triggered in allocutive clitic contexts when Fin probes the allocutive clitic. In contexts like (41), however, the above analysis has assumed that the extra source of case for the allocutive clitic is T, as provided for by the Obligatory Case Parameter. The absence of an expletive zelement here, as in (77c), is therefore accounted for on this approach without further assumptions.

Ergative displacement
Section 3.1 proposed an account of the fact that first person ergative clitics are linearized to the right of allocutive clitics. Specifically, in the analysis above, this reflects the fact that first person clitics move to a higher position-SpeakerP-than do third-person clitics. A question that arises from the perspective of this proposal is how this person effect relates to another person hierarchy effect on ergative clitic placement, termed "ergative displacement" in the Basque literature (Ortiz de Urbina 1989;Laka 1993;Fernández & Albizu 2000;Albizu 1997;2002;Albizu & Eguren 2000;Rezac 2003;Béjar & Rezac 2009;Arregi & Nevins 2012).
Ergative displacement refers to auxiliary forms in which the first position slot is occupied by an ergative person morpheme rather than an expletive morpheme or an absolutive clitic. Examples of such forms are given in (78) and (79). (78) Ikus-i n-u-en. see-perf 1sg.erg-root-pst 'I saw him/her/it.' Ikus-i-ko n-u-ke. see-inf-irr 1sg.erg-root-mod 'I would see him/her/it.' Ergative displacement is subject to two restrictions relevant to the discussion. First, ergative displacement is blocked in the presence of an absolutive clitic: Ikus-i nind-u-zu-n. see-perf 1sg.abs-root-2sg.erg-pst 'You saw me.' Second, ergative displacement only applies in past tense or irrealis contexts. This restriction has been observed in much of the literature cited above on ergative displacement, but I am not aware of any account in the literature for why the restriction should hold. This tense/mood restriction on ergative displacement is reminiscent of conditions on exponence of the expletive morpheme in that ergative displacement is triggered only in tense/mood environments in which the expletive morpheme does not spell out with the elsewhere value, d-. The previous section proposed that exponence of the expletive morpheme is sensitive to properties of Fin. In view of the similar distributions of z-/lexpletive exponence and ergative displacement, let us consider a partially unified account of the two phenomena in terms of properties of Fin.
The proposal in section 4.1 followed the consensus in the Basque literature in taking T to be the locus of ergative case (Rezac 2008;Etxepare 2006, a.o.). In the case of first and second person ergative arguments, the analysis assumed that T inherits uninterpretable person features from Speaker/Addressee and that these features probe the external argument when present. Let us assume following Arregi & Nevins (2012) that the property of T that makes it a potential clitic host is a [+Finite] feature, which it inherits from Fin. This [+Finite] feature in collaboration with the phi-probe triggers ergative clitic movement to the position of the auxiliary. In contexts in which Fin is specified for past tense-that is, in ergative displacement and expletive z-contexts-let us assume that Fin can withhold this feature, i.e. that it does not descend to T. This property might be understood in terms of the more general fact that properties of finiteness in Basque, including clitic licensing, correlate with tense specification (as it does in other languages). The [+Finite] feature on Fin, together with the speaker/addressee features that have raised to Fin by virtue of head movement, will now make Fin a potential clitic host for first and second person ergative clitics. The EPP property of Fin is therefore satisfied by the ergative clitic. Specifically, let us assume that this is spec-to-spec movement, which applies before m-merger would otherwise adjoin the clitic to Speaker. Again, m-merger will not adjoin the clitic to Fin in order to satisfy the requirement that spec, Fin be filled (see discussion of (50)). This proposal is illustrated in (81) In present tense contexts, where ergative displacement does not apply, the past tense specified Fin head, -(e)n is not merged and ergative displacement is not triggered. In contexts like (79), however, ergative displacement does apply, despite the absence of an overt complementizer. The analysis will require that such contexts involve a silent tensespecified irrealis complementizer that also triggers ergative displacement.
In the presence of an absolutive clitic, where ergative displacement is blocked, the absolutive clitic rather than the ergative clitic will move to Fin, since Fin is the only potential host for absolutive clitics. Again, following Arregi & Nevins (2012), let us assume that all person clitics must be licensed by raising to FinP. Let us assume further that a given head can host only one clitic, which otherwise seems consistent with the facts. If the ergative clitic rather than the absolutive clitic were to raise to FinP, the absolutive would have no landing site in a local relationship with FinP, in apparent violation of the licensing condition on Basque clitics.

Conclusion
Basque allocutivity has been a star witness in recent literature on the syntactic encoding of speech act roles, but to date, little work has examined the morphosyntax of these forms in close detail. This paper presents an analysis of the fine-grained syntax of Basque allouctive morphemes. A parochial consequence of the analysis is to model three recalcitrant properties of the morphosyntax of Basque allocutive clitics. More generally, this paper provides support for the position in recent literature that speech act roles in are encoded by a nominal element introduced in a left peripheral head. I have suggested that, in Basque, this element is overt and in fact behaves in ways similar to thematic addressees in Basque.
A question unaddressed in the preceding discussion is how the syntax of Basque allocutive forms compares with that of other languages described in the literature, including Galician (Carbón Riobóo 1995;Longa & Lorenzo 2001;Huidobro 2009), Korean (Pak 2017;Portner et al. to appear), and Magahi (Verma 1991;Bhattacharya 2016). The analysis above leads straightforwardly to the expectation that allocutive forms in these languages will have properties shared by thematic addressees. Work on these issues would benefit from formal comparative work, which has not so far been extensively pursued.