Overt speakers in syntax

This paper focuses on a type of direct address in Sason Arabic that fails to behave like a regular vocative phrase. In particular, unlike regular vocative phrases, this form of address spells out both the speaker and the addressee, and it is conditioned by the speaker’s expression of affection towards the addressee. A DP and its adjacent invariable clitic (i.e., a frozen 3rd person singular form of a possessive pronoun) are the constituents of this form of direct address, and either of them can equally spell out the speaker or the addressee; hence, the label mutable direct address (MDA) for this construction. Coordination, constituency, pluralization and substi-tution tests indicate that the derivation of MDAs does not involve the left periphery of DPs (as vocative phrases do) but the left periphery of clauses. In this respect, we adopt the cartographic representation of speech act phrases (SAP) (e.g., as in Haegeman & Hill 2013) and argue that MDAs instantiate a possible variation within SAP structures. That is, while a regular SAP field is split (over speaker and hearer heads), the SAP underlying MDAs remerges the two heads, so the speech act functional features are bundled and associated with a single SA head. Thus, MDAs provide new evidence for the extent of derivational variations that may arise on the basis of the feature inventory currently accepted for SAP domains.


Introduction
An increasing number of studies argue for a syntactic representation of discourse participants, as pragmatic roles (speaker and hearer/addressee) and their mapping in the derivation of the sentence (Speas & Tenny 2003; see also Hill 2007;Baker 2008;Giorgi 2010;Sigurðsson 2011;Miyagawa 2012;Haegeman & Hill 2013;Zu 2013;Haegeman 2014;Heim et al. 2014;Slocum 2016;Haddican 2018;Portner et al. 2019, a.o.). All these studies make the case for a syntactic representation of the speaker, despite the absence of lexical material that would spell it out. That is, these studies propose a null pronominal (pro) to encode the speaker, which implies that lexical DPs are incompatible with this task. Accordingly, the prediction would be that we cannot find languages with structures akin to those containing vocatives but with reference to the speaker (i.e., vocatives can be complex DPs encoding the addressee but not the speaker). Cross-linguistic variation in the spell-out of the expressions denoting discourse participants was shown to occur only insofar as morphological allocutive agreement may or may not occur in a language, and when it does, it spells out an expression that refers to either the speaker or the addressee on the verb (e.g., agreement with the speaker in Thai or with the addressee in Basque, as in Ross 1970). 1 However, these are affixes or clitics, as discussed in McFadden (2020), not regular nouns.
In this paper, we examine a construction from Sason Arabic that challenges the expectation for a silent speaker. More precisely, this is a direct address where a common or a proper noun can spell out either the speaker or the addressee as long as it is related to a clitic that takes the form of a quasi-invariable possessive clitic pronoun. That is, the possessive clitic in Sason Arabic is routinely inflected for person, number, and partially, gender (3rd person singular forms). However, in the context of the direct address discussed here, the same clitic is obligatorily invariable for person and number (i.e., 3rd person singular) while still being inflected for gender.
What is interesting about this construction is that the clitic, despite its 3rd person form, is understood to be either the speaker or the addressee, and so is the related DP. That is, the DP and the clitic take turns for spelling, each, one pragmatic role. More precisely, if the DP refers to the addressee, the clitic refers to the speaker; vice versa, if the DP refers to the speaker, the clitic refers to the addressee. Consider the examples in (1): In (1a), the noun 'brother' refers to the speaker, whereas the adjacent 3rd person clitic stands for the addressee, his sister, as indicated by the feminine inflection of the clitic and the following sentence. On the other hand, in (1b) the noun stands for the addressee, whereas the 3rd person clitic stands for the speaker (the sister), as indicated, again, by its feminine form. This type of construction depends on affectivity; that is, the speaker intends to convey his/her affection for the addressee while making a request or a comment.
(1) a. [Context: The elder brother addresses his little female sibling] Āx-a, tıtix-e tıcib-e lastiy-ad-i? brother-3f.cl can.prs-2f bring.prs-2f shoe-pl-1sg.poss 'Speaking as your brother, sister, can you fetch my shoes?' 2 b. [Context: The elder sister addresses her little male sibling] Āx-a, tıtix tadi-ni bēs-i? brother-3f.cl can.prs.2m give.prs.2m-me skirt-1sg.poss 'Speaking as your sister, brother, can you give me my skirt?' The same phenomenon is illustrated in (2) for proper nouns. In (2a), the noun Layla denotes the addressee, whereas the 3rd person clitic stands for the speaker, i.e., the husband, as indicated by its masculine inflection. The reverse is shown in (2b): Layla is the speaker, whereas the 3rd person clitic spells out the addressee, i.e., the husband, as the masculine form of the clitic. Therefore, the noun and the clitic can equally stand for either discourse participant. If the 3rd person clitic form is replaced with a regular possessive clitic, inflected, e.g., for the 1st person, as in (2c), the entire construction (i.e., DP and clitic) stands for the addressee, counting as a regular vocative phrase. Therefore, in order for a noun to spell out the speaker, the 3rd person clitic is obligatory to spell out the addressee, and vice-versa. So the concurrent speaker and addressee readings arise from mutual dependency between a DP and a frozen 3rd person clitic. discourse participant made visible through allocutive agreement may or may not be coreferent with an argument of the verb, but, crucially, it is independent of the argument structure of the verb. For example, an assertion such as There are many people in the street may display an extra-morpheme in some languages (either as an affix on the verb or as a clitic at C) to indicate whether the addressee is female or male, in a familiar or official relation to the speaker.
(2) a. ?[Layla zıll-u] nısit ıçax cawwıj-na? Layla gen-3m.cl forget.2m when married-1pl 'Speaking as your wife, honey, did you forget when we got married?' c. [Layla zıll-i] nısit-e ıçax cawwıj-na? Layla gen-1sg.poss forget-2f when married-1pl 'My Leyla, did you forget when we got married?' Crucially, the two items can never concurrently denote the same discourse participant. For example, in (3), the context forces the DP and the clitic to pick out the same referent, because of the person/gender agreement on the verb, and the construction is ruled out.
[Context: mother talking to her son] *Imm-a şıme kıt? mother-3f.cl how be.2m Intended: 'Speaking as your mother, how are you?' b. [Context: a male patient talking to their female doctor] *Hekim-a, şıme kıt-te? doctor-3f.cl how be-2f Intended: 'Speaking as your patient, doctor, how are you?' In light of these observations, we argue that the invariable 3rd person singular clitic in these examples indicates a discourse participant in direct addresses that involve the speaker's affection towards the addressee. So a contrast arises between vocative phrases, as in (2c), with no bias for the affection contexts, and where the entire DP denotes the addressee, versus the constructions in (1) and (2a)-(2b), which obligatorily entail the speaker's affection and spell out both discourse participants involved in the affectionate relation. Note that the speaker's intention to convey affection is obligatory for this form of address to materialize, but it does not predict whether the DP or the clitic will spell out the addressee.
Formally, we argue that such constructions involve the projection of a Speech Act Phrase (SAP), such as proposed for other contexts where conversational pragmatics is mapped to syntax. The peculiarity of these forms of direct address consists in the way they exploit the derivational options available for the projection of SAP: while most addresses involve split SAP configurations, these particular forms of address involve the remerging of SAP into a collapsed field, where all the relevant pragmatic features are bundled on one head. The justification is the following: With split SAP structures, there is an underspecified [familiar] feature, uninterpretable and unvalued, which maps the social relation between speaker and addressee (as argued in Portner et al. 2019), and whose value is established according to the semantics of the particle that merges to check it (e.g., honorifics or politeness particles that reflect the speaker's standing and attitude towards the addressee). With MDAs, the familiar feature is specified for affectivity, and it is mapped as an uninterpretable but valued [affect] feature that triggers the overt identification of both the source and the recipient of that affection. A syntactic peculiarity ensues by which lexical DPs, normally invisible to the [speaker] probe in regular split SAPs (i.e., because they do not have a 1st person feature on D), become licit in the MDA, where the affection/familiar feature is bundled with the p-role features. The proposal is that [speaker] and [affect] are checked as a bundle, and that a lexical DP becomes visible to this bundle of features because it qualifies to check [affect], while [speaker] is checked through free-ride.
Along these lines, the analysis proposed in this paper supports the former studies that represent the syntactization of the discourse as a SAP field in clausal derivations (especially Haegeman & Hill 2013), it expands the finding of these studies by integrating the [familiar] feature of Portner et al. 2019, in the guise of [affect], and it shows that, while maintaining the same inventory of functional features and semantic justification for them, the construction discussed here also increases the empirical coverage of these studies. Furthermore, this paper brings evidence that the SAP domain allows for the same variation in the implementation of feature checking that was pointed out for other domains. More precisely, the variation between splitting and remerging of heads within a phrasal domain is a universal option with functional fields, shown in the literature to equally apply to CPs (Henderson 2006, a.o.), DPs (e.g. Choi 2014 following Giusti 2005) and Voice/vP (Pylkkänen 2002;.
The paper is organized as follows: Section 2 presents the problem and the information on data collection. Section 3 discusses the empirical data from Sason Arabic that indicate the pragmatic marker status of the invariable clitic. Section 4 introduces the theoretical background that allows us to provide a formal analysis for these constructions in section 5. The conclusions follow in section 6.

Terminology and formal approach
Constructions along the lines of (1) and (2a)-(2b) are traditionally included under the umbrella of inverse vocatives (e.g. Boeder 1989). This label is misleading, first, because it brings biases for the analysis, entailing that the clitic is within a vocative phrase, which we argue is not the case. Second, vocative phrases, with or without a clitic pronoun, spell out one discourse participant, i.e., the addressee, whereas the DP+clitic constructs in (1) and (2a)-(2b) systematically spell out two discourse participants, in free alternation. That is why we label the strings in (1) and (2a)-(2b) as a mutable direct address, instead of vocative, where the term mutable captures the free alternation between the spellout of speaker and addressee by the same item (be it the noun or the clitic).
Mutable direct addresses (MDAs) are puzzling for two reasons: (i) the speaker can be spelled out through a DP in a direct address, which is cross-linguistically unusual; (ii) the clitic has an invariable 3rd person form but it receives set discourse values upon interpretation, for either speaker or addressee, instead of being associated with a non-discourse participant, as its inflection may suggest. Theoretically, the question revolves around the derivational mechanism that converges to an MDA rather than the garden varieties of direct addresses, which consist of vocative phrases, imperatives or injunctive particles, none of which allow for DPs that spell out the speaker. In this respect, we argue that an analysis in terms of Speech Act Phrase as in Haegeman & Hill (2013) (following Speas & Tenny 2003), where the speaker and the addressee pragmatic roles are syntactically mapped at the left periphery of clauses, can account for both the MDA construction and the fact that this construction is rare. So far, the theory predicts that a VocP merges with SA in order to check the pragmatic role of [addressee], whereas the pragmatic role of [speaker] is checked in relation to little sa by a null DP (i.e., the speaker is silent; Sigurðsson 2011, andalso Portner et al. 2019), as shown later in the diagram (20). The analysis we propose here extends the empirical domain to which saP/SAP structures apply, by including further cross-linguistic variation. These are the main lines of our analysis: (i) The derivation of MDAs concerns the clause structure (i.e. saP/SAP), not the internal structure of the DP. That is, the clitic is outside (i.e., in sa/SA) not inside the DP (as a possessive pronoun would be), and we treat it as a form of allocutive marker whose values for speaker or addressee are obtained from an imposter operator. The obligatory adjacency between the DP and the clitic of MDAs arises from their separate merge in a Spec-head configuration within saP/SAP. (ii) The MDA is cross-linguistically rare because it involves a different feature distribution, namely one that triggers a remerged SAP structure versus the default articulated SAP structure. More precisely, default SAP structures have the pragmatic roles of speaker and addressee mapped to separate heads (i.e., sa for speaker vs. SA for addressee), whereas in MDAs these features are bundled on the same speech act head. The frozen clitic qualifies as an imposter in the collapsed SAP, so imposter operators are obligatory in this configuration in order to ensure full interpretation.

Data collection and elicitation technique
Sason Arabic is a peripheral, vernacular Arabic variety spoken in eastern Turkey ( Jastrow 1978;Akkuş 2016;2017b). It lacks a writing system. Faruk Akkuş is a bilingual native speaker of Sason Arabic and Turkish as well as a heritage speaker of Mutki Zazaki. Our data were mostly collected in a series of elicitation sessions with eight bi-or multi-lingual Sason Arabic speakers, in addition to confirmation of judgements and contexts, constructed by Faruk Akkuş. The consultants range in age from late-20s to late 50s; five are from Bitlis, and three are from Batman. All the consultants live in Istanbul, and seven of them are illiterate. The only literate consultant has a middle-school degree. Four of the consultants are bilinguals between Sason Arabic and Turkish. These are the younger speakers who moved to Istanbul in their early childhood. The fifth consultant is multilingual between Sason Arabic, Mutki Zazaki and Kurdish. However, he is also a near-native speaker of the local Armenian dialect, and quite proficient in Turkish since he learned Turkish in his late-20s. The other three consultants are bilinguals in Sason Arabic and Mutki Zazaki, but also proficient in Turkish. All these speakers regularly use Sason Arabic in their homes although code-switching with other languages is also quite common.
Elicitation work was mostly held over Skype or WhatsApp over weekends with several of the consultants together since they live in the same street or neighborhood in Istanbul. In addition to elicitation work, Faruk Akkuş spent the summer of 2018 in Istanbul, in which period he was able to ask for clarification or testing of predictions. We also have a couple of naturally occurring data, such as (10a) and (10c) from conversations.
MDAs are also present in Turkish, as mentioned in the footnote to this page. 4 Although the Turkish data are not discussed in this paper, we verified that the constructions do not show deviation from the pattern we identify for Sason Arabic. In this respect, we checked the equivalent Turkish examples on a handful of native speakers (usually two or three), via in-person conversations, or in conferences where other Turkish linguists were also present. Moreover, the majority of the Turkish data consist of attested examples, found via Google searches or in published literary works. Finally the un-cited Basque data were elicited by Karlos Arregi on our behalf from a native Basque speaker.

The status of the invariable clitic
In the first part of this section, we show that Sason Arabic uses regular vocative phrases, which spell out only one pragmatic role, i.e., the addressee. These constructions are cross-linguistically unexceptional and may contain a possessive clitic, which displays an inflected form and has a possessive meaning. In the second part, we return to MDAs, where the clitic is invariable and bleached of its possessive meaning. The main argument is that, in MDAs, the possessive is grammaticalized and reanalyzed as a pragmatic marker; that is, the MDA clitic is homophonous to the 3rd person singular form of the possessive clitic in DPs, but these two items belong to different grammatical categories (i.e., allocutive marker vs. pronoun).

Vocative phrases in Sason Arabic
This section provides data that indicate the contrast between vocative phrases and MDAs. The point is that although MDAs are direct addresses built around a noun and a clitic (originally pronominal), their structure is different from a vocative phrase.
We start by clarifying what a vocative phrase is. In line with current formal analyses (Longobardi 1994;Moro 2003;Hill 2007, a.o.), we assume that vocative phrases are DPs/NPs with an extra layer at their left periphery (i.e., VocP > DP > NP). The extra layer maps a functional head that grants a 2nd person reading to a noun which otherwise would have a 3rd person feature by default (Bernstein 2008;Stavrou 2014). That is, there is a 2nd person functional feature in the vocative layer that trumps the default 3rd person of D. While this hierarchy is cross-linguistically stable, variations arise from the feature checking implementation within this structure (e.g., head-to-head movement up to Voc; phrasal movement to Spec,VocP; or long distance Agree with the noun in situ or in D/DP). VocPs can be used to convey direct addresses in a variety of pragmatic contexts (e.g., as calls, negotiations, attention drawing, etc.; Zwicky 1974), including affectionate addresses. This derivational pattern for vocative phrases extends to Arabic dialects (for tests and arguments see Soltan 2015 a.o.). Predictably, Sason Arabic displays vocative phrases, as in (4).
(4) a. In (4), the vocative phrase indicates the addressee, on the same pattern as its English counterpart 'my beautiful daughter(s)' or 'my doctor'. The entire phrase stands for one pragmatic role (i.e., addressee), and the noun can come in singular or plural. This vocative phrase can contain an attributive adjective and a possessive clitic. The possessive agrees in person and number with the possessor, but grammatical gender agreement is with the possessee DP and it is marked on the adjective. The point is that the possessive clitic is a regular possessive inflected for φ-feature agreement, and it is interpreted in the standard way, conveying a possessive reading. As in many languages with post-nominal adjectives, the possessive clitic merges with the noun, not with the adjective.
The facts in (4) conform to any standard analysis of DPs. That is, the possessive merges within the DP structure, irrespective of whether this DP is used as a verb argument or as a vocative. Agreement relations within the DP force the possessive to spell out the person and number features of the possessor, and possibly the gender of the possessee as well, depending on the language (Kayne 1994;Longobardi 1994 a.o.). Sason Arabic DPs are no exception in this respect (for agreement within DP in Arabic see Benmamoun 2000;Shlonsky 2004;Hoyt 2006; Fassi Fehri 2017 for both free and construct state DPs). Accordingly, the Sason Arabic possessive adjective in (4) displays different inflectional forms for number (singular or plural).
Crucially, the examples in (4) are fundamentally different from those in (1) and (2a)-(2b), since in (4) the address is fixed, (i.e., the reference for the DP is the addressee only), not mutable (i.e., the reference for the DP cannot be the speaker, in alternation with the addressee). Furthermore, in (4), the possessive clitic is inflected for number (and person), whereas the clitic in (1) and (2a)-(2b) is invariable in this respect. However, the latter is inflected in a way that reflects the natural gender of one of the discourse participants, but, crucially, not the gender of the adjacent DP (no nounpossessive agreement). From these observations, we conclude that the direct addresses in (1) and (2a)-(2b) are fundamentally different from vocative phrases, in a way that allows the direct address to not only spell out two discourse participants instead of one, but also to by-pass the obligatory number and person agreement relations to which a possessive clitic would be submitted within a nominal phrase (be it vocative or not).

MDAs versus vocatives
This section sums up the contrasts between the possessive clitic in DPs and the clitic used to derive an MDA, as well as the contrast between MDAs and vocatives as forms of address. The points of disparities are: • The context of use: Vocative Phrases may convey any type of address. On the other hand, MDAs necessarily entail an affectionate address. The affection is crucial for justifying the option for this construction over a regular vocative, and it arises in familiar types of social relations, within the family, with friends or within professional groups. In these contexts, MDAs are relatively unrestricted in their applicability to novel items, as indicated by their use with profession terms. Thus, an MDA is inappropriate in a context where the speaker loathes the addressee, as in (5b); only a vocative phrase is felicitous in this context, as in (5a).
(5) a. Leyla zıll-i ımbahaj-tu mı-nki! Leyla gen-1sg.poss be.fed.up-1sg from-2f 'Leyla, I'm fed up with you!' (spoken by a co-worker) b. #Layla zıll-u, ımbahaj-tu mı-nki! Leyla gen-3m.cl be.fed.up-1sg from-2f 'Spoken by me, your co-worker, Leyla, I'm fed up with you.' Therefore, the pragmatic context in which an MDA may be used is much more restricted than the one for VocPs, where an affection reading is possible but not obligatory.
• Agreement: When the direct address is a vocative phrase, the possessive clitic spells out the φ-feature agreement with the possessor for person and number. When the direct address is an MDA, these φ-features are not spelled out on the clitic. This suggests that the clitic of the MDA does not belong to the DP internal configuration, since there would be no escape from morphological agreement within the DP (in Arabic DPs, agreement obtains irrespective of whether there is movement or long distance Agree; Shlonsky 2004; Fassi Fehri 2017). In (6a), this construction yields a regular vocative, with a masculine possessee DP and the clitic in its 1st person form, and the entire VocP denotes the addressee. The same sequence rules out the sentence in (6b), where the clitic changes its inflection to 2nd person: the possessor and the possessee pick up the same reference, which is cross-linguistically anomalous. In (6c), the clitic takes the 3rd person form, and the masculine form is ruled out (since both the speaker and the addressee are feminine), whereas the feminine form of the clitic, which indicates the speaker, not the possessee, is ruled in. Note that possessive phrases containing possessive clitics in 3rd person may occur, cross-linguistically, as regular vocatives. 5 However, in Sason Arabic this is not an option: the direct address is grammatical only if the clitic ceases to stand for the possessor and to agree in gender 5 E.g., Portuguese: seu filho da puta. restart.2sg.imp his son of.the whore 'Start again, you idiot.' with the possessee DP, and switches, instead, to a discourse marking function. 6 Although the clitic still displays gender agreement (i.e., feminine or masculine), this agreement is allocutive (i.e., with the natural gender of a discourse participant), not grammatical (i.e., not with the gender of the DP it attaches to).
From a formal perspective, the examples in (6) indicate that as long as the clitic is inside DP/VocP, it functions as a regular possessive clitic. The fact that the same clitic switches to a discourse marker in MDAs indicates that MDAs cannot be configurationally equated to VocPs: although they are both forms of direct address, VocP has a nominal structure, as an extension of the DP, whereas the MDA is not derived on the nominal spine (or else, the clitic could continue to act as a possessive clitic).

Clitic versus affix
This section verifies the clitic status of the allocutive marker in MDA. That is, considering the adjacency between the DP and the marker, the question is whether the latter is a suffix on the noun base or whether it encliticizes on the preceding lexical material. Etymologically, the marker is derived from the (possessive) pronominal markers through grammaticalization (e.g. Jastrow 1978). 7 Hence, depending on the stage of grammaticalization, the result can be either a clitic or an affix. This distinction is important for the formal analysis of MDAs. The first piece of evidence for a clitic status of the MDA marker comes in (7), which resumes (2a). This example makes use of the possessive morpheme zıll, which is a free morpheme and intervenes between the noun and the marker. This morpheme is obligatory with proper nouns but optional with common nouns (i.e., when it appears on common nouns it triggers a focus reading; for the use of zıll in other Arabic varieties see Jastrow 1978 andTalay 2001, for Mardin and Hasköy varieties, respectively). The point is that, although the linearization shows the clitic marker adjacent to the noun in MDAs, as in (1), it is not affixed directly to the noun.

(7)
[Context: the husband addresses his wife, Leyla] Layla zıll-u ıbın-na nihane mou. Layla gen-3m.cl son-1pl.poss here neg.cop.3sg 'Speaking as your husband, my Leyla, our son is not here.' The independence of the noun and the marker in relation to each other within the MDA, as in (7), indicates that the marker cannot be classified as an inflectional suffix on the noun base. Hence, we continue to refer to it as a clitic.
Further confirmation for the clitic status of the marker comes from word order, as in (8), where the MDA contains a coordination phrase.
doctor and nurse-3m.cl, where remained-2pl 'Speaking as a patient, doctor and nurse, where have you been?' First, we have to mention that the possessive clitic systematically attaches to nouns, even in N-Adj sequences. That is, it does not attach to just any type of leftward host but only to nouns or nominal inflection. An example is available in (4a) above, where the clitic occurs between the noun and the post-nominal adjective.
Keeping this in mind, in (8) we look at the coordination of two DPs, where the clitic follows the entire coordination phrase. Irrespective of the level at which the coordination takes place (i.e., either DP+DP or NP+NP), if this item were a suffix on the noun root, it should have been obligatorily repeated with each noun. 9 One may object to this test by assuming that the clitic in (8) is a regular possessive that modifies the entire coordination phrase (i.e., [DP [NP+NP]+possessive]). The problem is that under such an assumption, the possessive clitic must display number agreement, that is, it should take a plural form. The non-agreeing form in (8) indicates that this clitic belongs to the MDA, not to the DP.
To conclude, the allocutive marker in MDAs is a clitic. Its adjacency to the preceding DP must be accounted for by locality within an underlying configuration in which the clitic is outside the DP.

The clitic is outside DP
So far, the data have provided clues that the MDA clitic is merged outside the DP/VocP, by showing lack of φ-feature agreement and attachment to entire coordination phrases. This section brings further evidence in this respect from restrictions on pluralizing and from pronoun cooccurrence. Consider the distribution of the regular plural suffix -ad in (9).
(9) a. Ma-adaş-tu axt-ad-u. neg-saw-1sg sister-pl-3m i: 'I didn't see his sisters.' ii: 'I didn't see their sister.' iii: 'I didn't see their sisters.' b. Axt-ad-u, sımo le-ni. sister-pl-3m listen.2pl of-me i: 'Speaking as your brother, sisters, listen to me.' ii: *'Speaking as your brothers, sister, listen to me.' In (9), the phrase axtadu is maintained identical in its use as a direct object in (9a) and as an MDA in (9b). In (9a), the plural suffix -ad-can be associated either with the noun or with the possessive clitic; that is, it can be interpreted to its left, to its right, or both. This structural ambiguity gives rise to three readings, as indicated: one in which the noun is plural (i.e., 'sisters'), whereas the possessive is singular (i.e., 'his'); one in which the noun is singular (i.e., 'sister') but the possessive is plural (i.e. 'their'); and one in which both the noun and the possessive are pluralized (i.e., 'their sisters'). On the other hand, in (9), the plural suffix can only be interpreted with the noun, not with the clitic, so there is only one possible reading, in which the DP is plural (i.e., 'sisters'), whereas the clitic is singular and denotes a discourse participant, not a 3rd person.
The restriction on pluralizing is verified in (10), where the examples switch between plural speakers paired with singular addressee, and singular speaker paired with plural addressees.
(10) a. [Context: two uncles addressing their only nephew] Amm-ad-u, n-ımme barra? uncle-pl-3m.cl 1pl-go outside Lit: 'His uncles, shall we go outside?' *'Their uncle (s), …' 'Speaking as your uncles, nephew, shall we go outside?' b. [Context: an uncle addressing his nephews] *Amm-ad-u, n-ımme barra? uncle-pl-3m.cl 1pl-go outside Lit: '*Their uncle, shall we go outside?' Intended: 'Speaking as your uncle, nephews, shall we go outside?' c. Amm-u, n-ımme barra? uncle-3m.cl 1pl-go outside Lit: 'Speaking as your uncle, nephews, shall we go outside?' In (10a), the DP stands for the two uncles, whereas the clitic stands for the nephew they address. The DP displays the plural marker. In (10b), the speaker is a singular noun, whereas the addressees are more than one. Hence, the expectation is that the clitic should be pluralized, not the noun, but that is ungrammatical. The MDA clitic can only display gender agreement with the addressee, not number agreement. The MDA in (10b) is rescued when the plural marker -ad-is eliminated, as in (10c), since the context does not allow it to be analyzed with the DP.
To conclude, when we see a plural marker in an MDA, it is because it inflects the noun, not the clitic, as summarized in (11). The ungrammatical use of this marker, that is, when it inflects the clitic, is shown in (12) Therefore, the data indicate a sharp contrast between the use of the plural marker with nouns and possessive clitics on the one hand, as opposed to the MDA clitic on the other hand. That is, the former can associate with the pluralizing morpheme alternatively or may share it, whereas the MDA clitic does not have access to it. Formally, possessive clitics are in the D domain, so the plural suffix can modify them on a par with the noun (Valois 1991;Kayne 1994). By this reasoning, the MDA clitic does not pluralize because it is not in the D domain (no matter what the exact structure of the D domain is for Sason Arabic). 10 Thus, the only item the suffix can pluralize with MDAs, as in (9b) and (10), is the noun.
This conclusion predicts that the MDA clitic can cooccur with a DP that contains a possessive clitic. That is, the MDA clitic is not a possessive. It does not interfere with the DP internal elements, so it is merged outside the DP as a whole. This prediction is born out in examples as in (13) In (13), the DP contains a noun or nominalized adjective and a possessive clitic, and the DP as a whole is treated in relation to the MDA clitic. The possessive is inflected for person and number, whereas the MDA clitic remains invariable in this respect. The cooccurrence of the two clitics cannot be justified by an analysis where both merge within DP. The domain independence is further confirmed by the fact that switching the order of these clitics, as in (13), yields ungrammaticality: the clitic denoting the discourse participant cannot intervene between the DP and the related possessive. These data and observations indicate that the MDA clitic is merged outside the DP.

The clitic is outside VocP
MDAs are forms of address, and so are the VocPs. Although the two forms of address were shown to differ with respect to interpretation, agreement and context of use, one may still suppose that the clitic is merged in the VocP layer (which is above DP, but not outside it) and that the MDA is a different type of VocP (i.e., that the MDA still belongs to the extended domain of N). We test this hypothesis in this section and conclude that the MDA clitic is outside the VocP as well. First, we summarize the observations that indicate the pragmatization of the possessive clitic in MDAs: 12 • semantic bleaching, by which the clitic lost the possessive lexical feature; instead, the clitic acquired an affection feature; • morphological bleaching, by which the pronoun lost its φ-feature inflection; • syntactic displacement to a position from which the clitic cannot access agreement relations with the cooccurring DP. In this respect, one may suppose that the MDA clitic may map a relational feature (in the spirit of Portner et al. 2019), but within VocP (not in cP). This is plausible, considering that affection is the defining semantic ingredient of MDAs, and that VocPs encode the speaker-addressee relation in a way that may capture this affection. However, the MDA clitic behaves differently from the regular vocative particles. In Sason Arabic, on par with other Arabic dialects, the dedicated vocative particle is yaa, which is a deictic for the addressee shown to be merged in VocP (Hill 2014;Soltan 2015). Consider the examples in (14), where regular vocatives occur in (14a)-(14b), whereas (14c) shows an MDA.  Soltan (2015) brings evidence (on the basis of Egyptian Arabic) that yaa is obligatorily adjacent to the vocative noun (in constituent tests) and it precedes the noun (in word order tests), which has consequences for the position assigned to this particle in the hierarchy of the VocP. A test along these lines is provided for Sason Arabic in (14b), to confirm the extension of Soltan's analysis to this dialect, without further technical discussion. 13 In (14b), ey 'hey' is a cross-linguistic attention-drawing particle, occurring independently of whether vocatives or other forms of address are present. Basically, this particle has to precede the yaa-vocative, and reversing the order of particles yields ungrammaticality. Hence, the particles merge in different domains and/or convey semantic features that clash with their reversed distribution. Whichever the case, the test indicates that the particles are syntactically processed (versus free linearization outside syntax) and that yaa shares syntactic tasks with the vocative noun. Crucially, yaa-vocatives may occur in MDA, as shown in (14c). The observations are as follows: (i) Yaa does not clash with the MDA clitic, which means that they check different features. As yaa checks the [i-p] feature of the VocP, whereas the noun checks its [person] feature, there is no feature left within the VocP to probe for an additional clitic marker. (ii) In the presence of a VocP (signalled by yaa), the MDA restricts the discourse participant readings: the noun may only denote the addressee, which means that the clitic may only denote the speaker.
Further examples on this restriction are given in (15).
[Context: little son calling his mother for help in the bathroom] Yaa ımm-u, amma baxit-e? prt mother-3m.cl where remained-2f 'Speaking as your son, mother, where have you been?' b. [Context: a mother addressing her little son] *Yaa ımm-u, atey mı-tılle mı barra? prt mother-3m.cl why neg-2m-come from toilet Intended: 'Speaking as your mother, son, why don't you come out from the toilet?' In (15), the vocative marker yaa forces the addressee reading on the noun, which means that the MDA clitic can only denote the speaker. This is not how MDAs work when the noun is analyzed as a DP (i.e., without a VocP layer). That is, removing yaa in (15) (i.e., the clue that forces a VocP versus DP analysis of the noun) would allow for the same alternation in discourse participant denotation as in (16) The alternations between the two readings on the MDA in (16) indicate that a VocP is not obligatory for the MDA, but there is alternation between VocP and DP. When VocPs occur in MDAs, they restrict the reading because of the interpretable 2nd person feature they provide. Note, however, that the absence of yaa does not mean the absence of a VocP layer, as this layer may contain only the nominal item without vocative particles.
Finally, considering the mutable values of the clitic as either speaker or addressee, if it merged within VocP it should rule in the constructions in (17) The construction in (17a) should be grammatical under the assumption that the speaker can be encoded as a DP within VocP, and the clitic can display φ-feature agreement with it. However, the encoding of the speaker as a DP within VocP is cross-linguistically ruled out, and so it is in Sason Arabic, hence the ungrammaticality of (17a) is unsurprising. Furthermore, even when the DP stands for the addressee, which is the default situation with VocPs, as in (17b), the clitic cannot display φ-feature agreement with it. Both (17a) and (17b) disprove the hypothesis that the clitic might merge within VocP.
This information can now be combined with previous observations regarding the lack of person and number agreement on the MDA clitic, which indicated that this clitic does not have access to morpho-syntactic agreement with the DP. In the previous section, we concluded that the MDA clitic is merged outside the DP. Now, giving the alternation of VocPs and DPs, both of which are phases, in relation to the MDA clitic, we can further infer that the clitic is merged outside the VocP as well.
Formally, there are two left peripheric fields that map direct addresses and in which the reanalyzed clitic may merge directly as a marker: either above a DP (Longobardi 1994;Moro 2003) or above a ForceP (Speas & Tenny 2003;Hill 2007;Miyagawa 2012;Haegeman & Hill 2013;Zu 2013;Haegeman 2014;Slocum 2016;Portner et al. 2019). The former is identified as the VocP layer discussed so far, the latter as SAP (Speas & Tenny 2003). Since we excluded VocP as the merging domain for the MDA clitic, the remaining alternative is SAP. Indeed, a SAP configuration can accommodate the encoding of both addressee and speaker, which would cover, for example, the reading alternations in (16).
Before proceeding to a formal analysis along these lines, we also mention the hypothesis that the reading alternations in MDAs may arise only in semantics/pragmatics, without involving the syntax. That is, they entail a shift of perspective, and should be treated as a matter of expressivity, whereby the speaker is taking the perspective of the hearer (e.g., parental self-address in Georgian reverse role vocatives; Apridonidze 1991). This would entail the speaker's empathy with the addressee.
Let us explore this approach for (18), by presuming that the mother is simply putting herself in the shoes of her son.
(18) a. [Context: mother addresses her son] Imm-u, i-xlo ifi beyt-ma ıjdid. mother-3m.cl 3pl-say there.is house-a new 'Speaking as your mother, son, they say there is a new house.' b. [Context: mother addresses her sons] *Imm-ad-u, i-xlo ifi beyt-ma ıjdid. mother-pl-3m.cl 3pl-say there.is house-a new 'Speaking as your mother, sons, they say there is a new house.' The problem in (18b) is that marking the plurality of the addressees on the clitic rules out the sentence, as discussed in the previous sections. This should not happen if only a shift of perspective were at work: the mother could put herself in the shoes of two sons as well as of one. The contrast of grammaticality in (17a)-(17b) forces us to look at the morphosyntax, to explain the restrictions on agreement. There is no doubt that a shift of perspective occurs in MDAs: with VocPs, the speaker's point of view dominates the reading with no clue about the addressee's interest in the matter, whereas with MDA, the speaker's point of view caters to the addressee insofar as the affection signals the speaker's willingness to take a back stage as needed. However, identifying this pragmatic switch is not sufficient to explain the morphosyntactic restrictions that make these constructions so different from vocative phrases within the same language, and also so rare cross-linguistically, although affection and empathy are universal.

Theoretical Framework
There are three main issues for which we borrow analyses from current literature: (i) the clause structure that reflects the mapping of pragmatic roles (p-roles), as in SAP proposals; (ii) the concept of allocutive agreement within SAP; and (iii) the concept of syntactic imposters.

SAP
The syntactic mapping of speech acts has been debated since Ross (1970). Advances in the formalization of syntactic derivations (in minimalism) or representations (in cartography) allowed for a revival of Ross' performative hypothesis in analyses that propose an extra-layer to the left periphery of clauses, above the level of complementizers. The exact configuration may differ from one analysis to another, but the underlying hypothesis is the same: features of conversational pragmatics act as formal features in syntactic configurations (Cinque 1999;Speas & Tenny 2003;Giorgi 2010;Sigurðsson 2011;Wiltschko & Heim 2016;Portner et al. 2019, a.o.).
For the purpose of this paper, we adopt the cartographic implementation proposed in Speas & Tenny (2003) and further refined in Hill (2007;, Miyagawa (2012;, Haegeman & Hill (2013), Zu (2013), Haegeman (2014). To this representation, we integrate the findings in Portner et al. 2019, namely, the mapping of the social relation between speaker and addressee, which has visible syntactic effects in a number of languages where the word order and/or the spellout of arguments and of the C-related items vary in the presence of honorifics or other politeness particles. 14,15 The common highlights of these analyses are listed below, with no regard to MDAs: • The conversational set-up is mapped to a speech act phrase (SAP), where the SA head has pragmatic role features, i.e. [p-roles], in the way lexical categories (e.g. verbs) introduce theta-role features in syntactic structures. • SAP is an extension of the C domain (i.e., it belongs to the left periphery of clauses).
In  (20). • On a par with theta-roles, p-roles trigger the projection of argument A-positions.
• The arguments of a speech act cannot also be arguments of the verb (e.g., vocatives), although they may be coreferent with the arguments of the verb (intra-deictic addresses). • Optionally, SA is associated with discourse oriented features (in addition to p-roles), which trigger the projection of non-argumental A'-position; these features map the speaker's point of view as a [p.o.v.] functional feature, and this has consequences for the word order of the sentence.  , the speech act projects a double layered structure (i.e., saP > SAP), similar to any argument structure (i.e., vP > VP; nP > NP, etc.). Thus, the hierarchy in (20) abides by the general derivational patterns, whereby internal arguments (indirect/VocP and direct object/CP) are mapped lower than the external argument (i.e., speaker). The specifiers in which DP/VocP arguments merge are A-positions, checking, each, an uninterpretable p-role feature: DP-pro checks [speaker] in saP; VocP (discussed in section 3) checks [addressee] in SAP, whereas CP checks [theme] when merged as sister to SA. In addition, in (19)/(20), the discourse feature [pov] is also mapped to the sa head and triggers the projection of a non-argumental position, when activated. The [pov] feature probes for an XP relevant to the purpose of the conversation (i.e., about which the speaker proffers her point of view). In this sense, the interpretation of this XP is akin to that of a topic. In this particular configuration, the XP that checks [pov] is the DP subject in (19), which moves to A'-Spec,saP and heads a non-quantificational chain, similar to the chains arising from movement to Topic positions lower than CP/ForceP in Rizzi (2004). The exact values of [pov] (e.g., worry, bother, command etc) are obtained from the context and/or through the principle of compositionality. 16 Portner et al. (2019) argue that, in sentences with a specific interlocutor, the mapping of speech acts must also include a functional feature that encodes the type of social relation between speaker and addressee. In their system, this feature is identified as [status]/[familiar] and its values determine the presence or absence of honorific and politeness particles or the form of 2nd person personal pronouns in certain languages (e.g., the option between tu or vous in French). We introduce this feature in (20) as part of the sa feature bundle, since it is the speaker's prerogative to decide on the degree of familiarity in his/her address. In fact, when the sa head contains a lexical item (which is the case in many languages but not in English), that item systematically comes with inherent (un)familiar features which translate configurationally, into a selectional feature: i.e., its SAP complement must spell out the addressee in a way that matches the (in)formality of the sa particle. For example, when the injunctive particle hayde (common to Arabic dialects, Turkish and Balkan languages) merges in sa (as in Haegeman & Hill 2013), the address excludes formal relations and restricts the range of addressees to those that qualify for some degree of familiarity.
While [familiar] requires feature checking on a par with any other formal feature, its valuation may vary cross-and intra-linguistically. That is, the value for [familiar] may arise from the syntactic configurations (e.g., in the presence of lexical items that come with inherent specification for familiarity) or from the pragmatic context. For example,in (19) there is no lexical item that would provide a clue about the degree of familiarity, so this relation is interpreted according to intonation and pragmatic context. Hence, in (20), [familiar] is mapped as an underspecified uninterpretable feature. SAP checks this feature (hence the change to interpretable) but does not value it. Notably, whether this feature receives pragmatic valuation (as in English) or morpho-syntactic valuation (as in Korean or Japanese; see also Miyagawa 2012) is a matter of cross-linguistic variation.
In (20), the projection of multiple specifiers is not free or interchangeable: the Spec from which p-roles are checked is an argumental position and it is the closest to the head, whereas the higher Spec displaying the [pov] item is non-argumental. Crosslinguistic surveys within the same body of literature indicate that the inventory of features within saP/SAP is stable, but variation may arise in the feature checking options, which may trigger a further splitting of the hierarchy. For example, while Hill (2007) shows that the hierarchy in (20) accounts for speech act syntax in a variety of languages, including English, Haegeman (2014) argues that West Flemish displays a further articulation of saP/SAP, since in this language, the formal features shown in (20) are mapped each to a separate functional head, which amounts to four layers of SAP structure (instead of two). The cross-linguistic variation can be further compounded by the merge options, involving direct merge, move or long distance Agree.
In this paper, we adopt the saP/SAP hierarchy in (20), and pay attention to the parameters for cross-linguistic variation this hierarchy allows for. In particular, we notice that so far the cross-linguistic variation discussed for SAP involves the further splitting of this field on the basis of the same feature inventory (i.e., for West Flemish). Hypothetically, the alternative would be that the same features may be bundled instead of being individually mapped (by remerging the split structure). The alternation between split and remerged options has been shown to apply to CPs (Henderson 2006, a.o.), DPs (e. g. Choi 2014 following Giusti 2005); hence, why not to SAPs? We shall argue that MDAs arise from the exploitation of this option. 17

Allocutive agreement
This section summarizes the information on allocutive agreement markers available in the literature. The rationale is that the MDA clitic resembles an allocutive agreement morpheme insofar as it belongs to SAP and displays gender inflection that matches the natural gender of one of the discourse participants. We pay particular attention to Miyagawa's (2012) study of honorifics in Japanese on the basis of previous studies that attest to the presence of allocutive agreement in Basque.
As shown in (21), Basque has a dedicated verb suffix that spells out the addressee's gender and entails familiarity.

(21)
Allocutive agreement in Basque (Oyharçabal 1993) a. Pette-k lan egin di-k. Peter-erg worked 3.erg-m 'Peter worked.' (said to a male friend) b. Pette-k lan egin di-n. Peter-erg worked 3.erg-f 'Peter worked.' (said to a female friend) The main point in Miyagawa's study is that the allocutive agreement should be treated on par with morpho-syntactic agreement: the feature responsible for this type of agreement is located high in the left periphery of clauses, but the agreement suffix surfaces on the verb stem, as regular φ-agreement morphemes do. Miyagawa adopts the hierarchy in (20) and attributes the allocutive agreement phenomenon to the addressee p-role feature in Basque (see McFadden 2020 for allocutive agreement in Tamil as well). Similar facts are reported for Jingpo (Tibet-Burman, in Zu 2013;, where the allocutive agreement spells out the speaker. Zu (2015) points out that the presence of the speaker agreement establishes an intimate relation between the speaker and the subject. For instance, (22) is used in a scenario where a teacher reports to a principal about her students, and both sentences are truth-conditionally equivalent. What differs, however, is that (22b) indicates that the teacher and her students are on good terms, whereas (22a) has no such indication.

(22)
Allocutive agreement in Jingpo (Dai 2010: 5;cited in Zu 2015) a. Jongma du hkum ma-s-ai. students arrive complete pl-prf-3.decl 'The students have all arrived.' (subject agreement, neutral) b. Jongma du hkum sa-ga-ai. students arrive complete prf-1pl-decl 'The students have all arrived.' (speaker agreement, affection) In light of this literature, we may say that the MDAs of Sason Arabic display allocutive agreement, insofar as the clitic is an element that spells out the gender of the addressee. However, there are also important contrasts: the p-role feature the allocutive agreement reflects has a pre-set value in Basque, for the addressee, and in Jingpo, for the speaker, but not in Sason Arabic, where it can be either speaker or addressee. Also, the location of the marker is different: in Basque/Jingpo, it surfaces as a suffix on the verb root, 18 whereas in Sason Arabic it is an enclitic merged at a higher location, which remains to be determined, but which is independent of the location for verbs. Accordingly, for the MDAs in Sason Arabic, we have to establish: (i) whether the clitic can be considered an allocutive agreement morpheme; (ii) why it encliticizes on a DP that refers to a different discourse participant; (iii) how it obtains the discourse participant reading.

Imposters
The term imposter refers to a noun or a pronoun whose person interpretation differs from what is expected from their morpho-syntactic form. For example, in an address like Would her majesty need anything else?, the pronoun her displays a 3rd person form but indicates the addressee, that is, somehow this item becomes associated with a 2nd person feature (Collins & Postal 2012;Podobryaev 2014;. Collins & Postal (2012: 5) define imposters as "a notionally X person DP which is grammatically Y person, X ≠ Y"; e.g. the present authors, the undersigned, Madam. For instance, Turkish, Moroccan Arabic and Albanian permit φ-feature alternations that do not correlate with differences in meaning or truth conditions (Akkuş 2017a). The mechanism by which an imposter achieves a different person interpretation than the one indicated through its form was identified, in the same studies, as involving the merging of a null operator, that is, an imposter operator. In a nutshell, imposter operators are computational devices used in syntax to manipulate the person reading of a noun or a 3rd person pronoun. That is achieved because the imposter operator introduces indexicality with the pragmatic context in a way that overrides the inflectional information of the nominal item.
Examples of imposters with either speaker or addressee readings are shown in (23), for English (23a) and Albanian (23b).
(23) a. Little Jonnie looks tired today. (addressee-imposter) b. Albanian (Kallulli 2014: 72) Mami shkon / shkoj tani. mommy.the go.prs.3 / go.prs.1 now 'Mommy is going (away) now.' (speaker-imposter) Importantly, imposters do not necessarily concern direct addresses, as they may also occur in statements (e.g., for the present authors or the undersigned). Podobryaev (2014; provides an account of DP interpretation under which imposters are semantic binders, and φ-features are semantically interpreted. Imposters can be syntactically ordinary 3rd person DPs that refer to the speaker or the addressee because they license silent assignment-function-manipulating operators in syntax which endow them with 1st or 2nd person features, and thus they can become speaker-imposters or hearer-imposters. Imposter operators are in complementary distribution with 1st or 2nd person pronouns. Podobryaev (2014) shows that imposter operators exhibit syntactic effects, in that the relevant domain for the imposter interpretation is constrained by the grammar (see also Collins & Postal 2012). The domain is a syntactic constituent that contains both the imposter and the coreferent pronoun (e.g. a DP with possessive clitic). So the imposter operator is not purely semantic, but it is also syntactically triggered. A technical illustration of imposter operator syntax is provided in (24), with the analysis in (25). 19 19 Several details in the structure are not central to our discussion. For instance, Podobryaev (2014; follows Minor's (2011) and Sudo's (2012) theory, in which the assignment function interprets complex indices that are essentially pairs of a natural number and a person feature, e.g. ⟨5, ①⟩ → Sasha or ⟨87, ②⟩ → Mary (where ① stands for the first person, ② stands for the second person, and ③ for third person) See Podobryaev (2014; for the semantic details of how imposter operators work, which we leave aside since it is not the focus of this paper. In (25), the speaker-operator, indicated as ①, scopes over the imposter 'the present author' and the 3rd person marked pronoun 'his', thus allowing them to be interpreted as the speaker (see Frank 2016 andAkkuş 2017a for discussions regarding the sensitivity of imposter operators to the φ-feature content of the imposter and bound pronouns). The derivational pattern in (25) remains the same if it has to apply in constructions where a hearer-operator is at work. The relevant point for our data from this theory is that MDAs also display a pronominal element with a frozen inflection (i.e., 3rd person singular) whose interpretation is that of a discourse participant. Therefore, we take it that imposter operators apply in these constructions as well. Abstracting away from various aspects of this theory, which are not central to our discussion, we adopt this aspect of the imposter operators.

Formal Proposal
In section 5.1 we apply the saP/SAP hierarchy in (20) to the direct addresses that contain regular VocPs in Sason Arabic and show their compatibility. Section 5.2 repeats the exercise with MDA constructions, and finds that the saP/SAP hierarchy is inadequate for these constructions in its split pattern. In section 5.3 we remerge the two speech act heads, and find that such a configuration can rescue the SAP analysis while capturing all the properties of MDAs, including the imposter property of the invariable clitic. Before we present our analysis, a summary is in order to list the properties of MDAs pointed out so far throughout the paper: (i) The option for MDAs depends on the encoding of speaker's affection. (ii) MDAs display an obligatory cooccurrence of a DP and a clitic of pronominal origin, but the clitic is invariable for person and number. (iii) The clitic qualifies as an allocutive marker merged outside the DP/VocP. (iv) The DP and the clitic check a different p-role, and the checking task is interchangeable.
For instance, the exact domain over which an imposter operator functions, dubbed as 'impostrous domain' in Podobryaev (2014;, requires future work. It is likely that different clausal domains (e.g. the Left Periphery versus clausal domain), each have different operators, an option Podobryaev mentions. We remain agnostic as to the specific constraints on the imposter operators.

saP/SAP for direct address
First, we verify that the hierarchy in (20) can be extended to direct addresses that contain regular VocPs in Sason Arabic. Consider the example in (26).
(26) Hayde Layla, çax le maşu ye! prt Layla time of going cop.3sg 'C'mon Layla, it is time to go.' In (25), Layla is a VocP that follows the injunctive particle hayde, present in all the languages that underwent contact with Turkish (see Tchizmarova 2005 for Balkan languages). The particle has been shown (by tests of word order and constituency; Haegeman & Hill 2013) to merge directly in the sa head of (20). A speaker opts for this particle to signal his/her state of mind (here, impatience) and a degree of informality in the social relation with the addressee. Hence, unlike (19), (26) forces a familiar relation reading because this is part of the intrinsic meaning of the sa particle hayde. Hence, this particle restricts the range of possible addresses that can be mapped to its SAP complement. Accordingly, the Sason Arabic direct address in (26) can be configured as in (27) ForceP çax le maşu ye The linearization arising from (27) captures the word order in (26), where the intonation is fluent, without breaks and pauses.

saP/SAP for MDAs
Now we apply the same hierarchy to the MDA in (1), repeated as (28). The linearization in (28a), where the DP is the speaker, could be derived from (29), if no lexical material intervenes between Spec,saP and SA.

tıtix-e tıcib-e lastiy-ad-i
In (28), āx 'brother' is a DP, where D is associated with a 3rd person feature. How could such D become visible for the [speaker] feature of sa, since it does not come with a 1st person feature that would make it visible to this probe? The answer could be that this DP behaves like an imposter, which introduces an imposter operator that takes scope over the saP domain and assigns an interpretable speaker feature to the DP. The problem is that the merging of an imposter operator for the speaker would block the merging of the invariable clitic. That is, the clitic -a in SA has an invariable 3rd person form but merges in SA to check the [addressee] feature. This is a typical behavior for an imposter. Hence, there is a second imposter that introduces another imposter operator, this time with addressee semantics (i.e., equivalent to 2nd person feature), so the clitic qualifies to check the feature of SA. In other words, we end up with two operators in the same domain, each with a different feature, and both take scope over [saP > SAP]. Such configuration would trigger intervention effects: [ saP OP 1 OP 2 āx 1 [ SAP -a 2 ]]. A further problem arises from the manipulation of word order: Configurations as in (26) optionally allow for the intervention of lexical material between hayde and the vocative, as in (30a). This is not allowed in MDAs, as shown in (30b)-(30c).
(30) a. Hayde mışi daha xıfef ax-i. prt go.2m more quickly brother-1sg.poss 'C'mon, go faster brother.' b. *Ax mışi daha xıfef-a. brother go.2m more quickly-3f.cl Intended: 'Speaking as your brother, go more quickly.' c. *(Ax) hayde-a (ax) … brother prt-3f.cl brother … Intended: 'Speaking as your brother, c'mon, go more quickly.' // 'Speaking as your sister, c'mon brother,…' In (30a), which conforms to (20), the items distributed in saP and SAP, respectively, can occur with a different linearization. The exact operation responsible for this variation is beyond the scope of this analysis (it could be CP adjunction to SAP or any sequence of phrasal and remnant movement that linearizes the vocative in clause final position). The important point is that MDAs do not allow for such word order variations, as shown in (30b)-(30c). If the only reason for the adjacency between DP and the clitic is the requirement for a lexical host for the clitic, then (30c) should be ruled in, since CP provides the lexical host as well as a DP. The same can be said about (30c): hayde is a free morpheme that could support -a merged to check [addressee] in SA. In other words, (30c) should be ruled in, contrary to the fact. The ungrammaticality of (30b)-(30c) indicates that the obligatory adjacency between the DP and the clitic in MDAs is not a matter of morphophonology but of morpho-syntax, since it constrains the word order and the type of categorial host for the clitic.
The challenges extend to (28b), where -a stands for the speaker, whereas the noun stands for the addressee. According to (20), -a should merge in sa, whereas the DP merges in Spec, SAP,as in (31). In this configuration, DP comes necessarily in a VocP shell, since it may allow for the vocative particle yaa to precede the noun, as mentioned in the previous section.
For the flexibility of the clitic as the spellout of either the addressee, as in (28a), or the speaker, as in (28b), let us assume that the invariable form of this item allows it to freely merge in sa or SA. Then, in (31), āx 'brother' merges in Spec,SAP since it is a VocP with a 2nd person feature that qualifies as goal for the [addressee] in SA. Hence, the 3sg feminine clitic -a is forced to merge in sa to check [speaker], which means it also introduces an imposter operator with scope over saP. The word order DP-clitic may be justified as a post-syntactic operation that responds to the morpho-phonological requirement of clitic for lexical hosts (Embick & Noyer 2001). The questions arising from this analysis resume those pointed out for the configuration in (29): (i) why does the clitic behave as enclitic instead of proclitic; and (ii) if encliticization is obligatory because of some independent property of this language, why can't it encliticize on a different XP or lexical head that precedes it. In this respect, Sason Arabic does not observe a rule on second position clitics; however, clitic pronouns are systematically enclitic within DP. Let us assume that encliticization is obligatory for these elements even when they occur outside the DP, as is the case with MDAs. Then if there is XP movement to a discourse Spec, saP in response to a [pov] probe, which is possible with MDA as well as with regular direct addresses, this XP should be able to support the clitic. Indeed, the example in (32a) shows an MDAs where XP movement applies to the A' Spec, saP, in the same way it takes place in (20). However, the XP must precede the entire MDA, and cannot substitute for the DP as a host for the clitic, even when the clitic is in sa, above VocP, as shown in (32b). The ungrammaticality of (32b) cannot be justified on the basis of the hierarchy in (31).
(32) a. Cam Layla zıll-u te-i-bqa muqawwej! window Layla gen-3m.cl sbjv-3m-remain closed 'The window, Layla must stay closed.' b. *Cam-u Layla zıll te-i-bqa muqawwej! window-3m.cl Layla gen sbjv-3m-remain closed Intended: 'Speaking as your brother, the window, Layla, must stay closed.' Again, (32) shows that the obligatory adjacency between the DP and the clitic of MDAs is not a morpho-phonological matter, but it rather concerns a syntactic constraint that chains the two items together and prevents the clitic from taking a different host. Along these lines, (31) is not the adequate representation for (28b), as it cannot capture the sequential order, and reducing the reordering of these elements to a post-syntactic process is not warranted.
To sum up this section, split saP/SAP configurations successfully capture the word order and distribution of lexical elements within a regular direct address in Sason Arabic, where the speaker is silent but the addressee may be spelled out through a vocative phrase. However, when it comes to MDAs, the split saP/SAP configuration fails to account for the word order and for the free mutability of the two elements that spell out the speaker and the addressee. The tests showed that the clitic is restricted to a DP host (versus any XP or lexical head), and this restriction cannot be accounted for within a split saP/SAP hierarchy or on phonological grounds. Crucially, the functional features involved in the derivation remain the same for saP/SAPs and for MDAs. However, we presume that the way these features are distributed must differ. The next section aims to develop this hypothesis and to capture the points of variation in the SAP domain that result in an MDA.

Remerged speech act heads
In this section we propose that the configurational variation between a regular address, as in (26), and an MDA as in (28), can be captured by remerging the split saP/SAP configuration. That is, all the functional features pertaining to the speech act domain are bundled on a single head. This reconfiguration arises from the semantic specification that the [familiar] feature brings to the structure, namely, affectivity. In this respect, we relabel [familiar] as [affect], keeping in mind that we do not introduce a new formal feature; we only emphasize that, only in these constructions, [+familiar] maps not only an intimate social relation, but also the psychological state that supports that relation. The effect of this specific value for [familiar] is that speech act particles like hayde, orthogonal to the affection semantics, do not qualify to merge in the sa head of an MDA; instead, the mapping of affectivity requires that both the source and the recipient of affection be overtly mapped in a local domain.
The analysis of MDAs on the basis of a split saP/SAP configuration produced the following result: Although all the formal features listed for saP/SAP are identifiable in MDAs, the split hierarchy fails to account for the strict DP-clitic adjacency (versus XP-clitic) and for the mutability of these two lexical items when it comes to spelling out the speaker and the addressee. Furthermore, the obligatory affectivity in MDAs escapes the syntactic representation. That is, the direct address may optionally have an affective reading when it is derived by the regular saP/SAP pattern, whereas for MDAs the affective reading is obligatory.
Technically, we propose that the configuration in (33) underlies an MDA as in (28). This configuration remerges the speech act heads, and this process is related to the specification of [familiar] as [affect]. This precise semantic specification requires the overt identification of both the source and the recipient of the affection, and this requirement is grammaticized.  (33) provides a structural account for the mutable spellout for speaker and addressee insofar as the items in the head and the Spec positions are equidistant in relation to the feature bundle. The analysis in (33) helps us to better understand the status of the invariable clitic against the theoretical background for allocutive agreement as a cross-linguistic phenomenon. It also explains why some distributional predictions based on a split saP/SAP pattern do not obtain in MDAs.
Starting with the clitic, at first glance, it seems to qualify as an allocutive agreement marker, such as seen cross-linguistically: the clitic indicates the natural gender of a discourse participant. The fact that the Sason Arabic clitic does not surface on the verb, as in Basque, would be predicted by cross-linguistic variation. That is, a uniform cross-linguistic analysis can still be maintained under Miyagawa's (2012) hypothesis that the features spelled out by these markers belong to the set of features at C that may or may not be transferred to T, depending on the language. Along these lines, the allocutive agreement remains at C in Sason Arabic but is transferred to the T domain in Basque.
Nevertheless, there is an important contrast in the syntactic behavior of the MDA clitic and the allocutive agreement morpheme in Basque: the latter operates as an agreement marker in relation to a VocP (irrespective of whether the VocP is lexical or not), whereas the MDA clitic spells the addressee directly and cannot be used as an agreeing morpheme at all. Compare (34a) in Basque and (34b)  In (34a), the Basque allocutive agreement morpheme agrees with the clause initial VocP. This is not possible in (34b), where the Sason Arabic clitic does not allow for coreference with the vocative noun (the vocative is appositive here, which should be allowed), neither does it agree with the DP it is adjacent to, nor does it allow for an agreeing ending on the verb. This is an indication that the MDA clitic is not an agreement morpheme but a thematic morpheme (i.e., it spells out a p-role and thus substitutes for a VocP/DP), in addition to its inherent affection feature. So we consider it an allocutive discourse marker, rather than an allocutive agreement marker.
The redefinition of the clitic as a pragmatic (allocutive) argument rather than inflectional allocutive agreement, together with the obligatory remerging of speech act heads, makes certain predictions on distributional constraints, which are borne out by the data, as follows: (i) If MDAs were derived through split configurations on the pattern in (20), then (35) should be possible, contrary to the fact. In (35), a VocP replaces the clitic that would have spelled out the addressee within the same prosodic unit.
(35) *Ax Layla, cib-e-ni lastiyad-i. brother Layla bring-2f-me.dat shoes-1sg.poss Intended: 'Speaking as your brother, Layla, bring me my shoes.' The address in (35) should be possible if the underlying derivation were split: āx 'brother' is in Spec,saP for [speaker] (although stipulations are in order as to why this DP is visible to the [speaker] probe), whereas Layla could merge in Spec, SAP to check [addressee]. However, that sequence is actually ungrammatical. This indicates that MDAs, that is, the only direct addresses where the [speaker] feature is spelled out through a common or proper name, do not have two argumental Spec positions available, one for the speaker and one for the addressee. Instead, what we see is obligatory alternation between the nouns that refer to the speaker or to the addressee, which means that only one argumental Spec is available in the underlying configuration. The other participant is then spelled out by an element that qualifies for merging in the related head.
(ii) The observations on the distribution of the clitic that we illustrated in (32b) also find a natural explanation in the pattern in (33). More precisely, (32b) shows that XP-clitic sequences are ungrammatical. A split hierarchy as in (20) cannot explain this restriction: in (20), XP moves to the highest, non-argumental Spec of saP to check [pov] and thus precedes the clitic merged in sa to check [speaker]. There is no reason why XP cannot serve as lexical host for the clitic. However, (33) naturally accounts for the ungrammaticality of XP-clitic sequences: since there is only one argumental Spec, which is always filled with a DP (either for speaker or addressee), an XP moved above this Spec cannot be adjacent to the clitic merged in the head of that structure. The same explanation applies to (30b). (iii) The ungrammaticality of (30c), where the speech act particle hayde is disallowed in an MDA, is also unsurprising in (33). Since [familiar] comes with a set value for affection in this configuration, hayde (or other particle in this language) does not qualify to merge in the s/SA head since its semantics is not exclusive to the affective reading.
In sum, the remerged configuration in (33) captures all the properties we listed for MDAs throughout the paper, and it provides the most plausible representation of the structure that underlies these forms of address. In particular, (33) centers on the mapping difference that occurs in speech act fields when the [familiar] feature comes specified for affection: instead of treating [affect] as a selectional feature, in the way [familiar] behaves in split configurations, MDAs resort to a Spec-head structural agreement that overtly identify both sides concerned by the affective relation. This variation is minimal and predictable under the feature checking theory, since it switches from head-Complement (in split derivations) to Spec-head (in remerged derivations). All the illicit combinations pointed out in the tests on MDA are amenable to principled consequences arising from the contrast between split and remerged derivations. Along these lines, the MDA configuration in (33) is consistent with observations on variability in feature mapping noticed for other types of phrasal domains, especially in cartographic analyses. Thus, split and remerged representation were proposed for TopP (Frascarelli & Hinterhölzl 2007;Frascarelli 2007), ForceP (Cognilio & Zegrean 2012) or FinP (Hill & Alboiu 2016), to mention only a few relevant areas. Beyond cartography, the bundling/remerging of heads has been proposed to account for tense, aspect, mood (Giorgi & Pianesi 1997), tense and agreement (Bobaljik 1995;Thráinsson 1996), complementizer systems (Bianchi 1999), causatives (Pylkkänen 2008), V2 requirements (Hsu 2017), Austronesian voice systems (Erlewine et al. 2018;Erlewine 2019).

Conclusions
This paper focused on mutable direct addresses (MDAs) in Sason Arabic, where both the speaker and the addressee are spelled out, interchangeably, through a DP/VocP that is obligatorily adjacent to a pronominal enclitic. It was shown that these constructions occur in the language side by side with regular speech act structures (saP/SAP), where only the addressee is spelled out through a VocP, whereas the speaker is checked by a null pronominal. Unlike these default configurations, MDAs display a concurrent and overt representation of both the speaker and the addressee, either through a DP or an invariable clitic. The main aim of this paper was to understand the mechanism that allows for this contrast between obligatory silent speakers in regular direct addresses and obligatory lexical speakers in MDAs. The tests led us to the following findings: • The underlying structure of an MDA involves a remerged (versus split) speech act field. • Head remerging arises from the set specification of the [familiar] feature for [affect]. • DPs qualify as goals for [affect], while [speaker] is checked as a free ride on [affect]. • The clitic is an imposter that introduces imposter operators, which allow it to check a p-role feature; i.e., the invariable clitic is a p-argument. • The DP and the invariable clitic merge independently in the remerged speech act phrase: DP in Spec, s/SAP, and the clitic in the head s/SA. Obligatory adjacency follows from this local configuration.
• The mutable pairing of pragmatic roles with either DP or the clitic is possible because of their equidistance to the probes, and it is constrained by the type of operator that binds the imposter clitic: when DP merges for [affect], the clitic is bound by an imposter operator with an addressee feature; when DP has a VocP shell, it merges for [addressee], whereas the clitic is bound by an imposter operator with a speaker feature, which is checked by free-riding on [affect].
The remerging structure underlying the MDA provided us with the opportunity to confirm that SAP projections behave on a par with any other phrasal domain insofar as the relevant formal features can be mapped either separately or bundled, and when bundling applies, it can be partial (for some features but not for others) or complete (i.e. all the features belong to the same bundle). So far SAP was shown to entail partial bundling and a double layered structure by default, cross-linguistically (Speas & Tenny 2003;Hill 2014 a.o.), and the only variation pointed out (for West Flemish in Haegeman 2014) entailed further splitting, to a one-to-one mapping system (i.e., one feature per head). It stands to reason that the opposite option (i.e., total remerging) must also be possible, and MDAs could confirm that prediction.