Address inversion in southern Italian dialects

This article proposes a ‘topological’ reinterpretation of the extended nominal architecture in relation to southern Italo-Romance vocatives with and without allocuzione inversa (‘address inversion’, Renzi 1968), a phenomenon involving the ‘inverse’ lexical indexation of the speaker-addressee relationship (reg.It. Mangia, papà ! ‘Eat up, little one !’, father to child). Topological Mapping Theory (Longobardi 2005; Martín & Hinzen 2014) posits a unified model of grammatical structure and nominal reference denotation in argumental constituents, where a hierarchy of referentiality (from predicativity to deixis) emerges through the expansion of the functional architecture. Contributing to a growing theoretical consensus favouring extra ‘vocative’ structure in the nominal left periphery, I argue that Italo-Romance vocatives with and without address inversion involve a part-whole expansion of structure, yielding a necessarily tripartite nominal architecture (VocP-DP-NP) in line with topological principles. The non-literal interpretation of N observed in the ‘lexical flip’ of address inversion vocatives is argued to be the surface manifestation of movement into VocP, a functional space whose internal articulation serves to construe


Introduction
Many southern Italian dialects exhibit allocuzione inversa ('address inversion', Renzi 1968), a typologically widespread phenomenon in which the lexical material used in a vocative expression 'inverts' the expected pattern indexing the speaker-addressee relationship: 1 (1) San Benedetto del Tronto (Savoia 1984:180, cit. Mazzoleni 1997 Mäñña, mamma mmi! eat.IMP mummy my 'Eat up, my darling!' (mother to child) In (1), the utterance-final vocative does not function as a self-address but instead expresses affection from caregiver to child, whence the idiomatic gloss of the vocative mamma mmi (literally, 'mummy my') as 'my darling'. Whilst a modest body of studies has described address inversion cross-linguistically, 2 the phenomenon has mostly been overlooked in formal theory, with some notable exceptions (e.g., Hill 2014Hill , 2017Akkuş & Hill 2017).
This article examines the grammatical properties of Italo-Romance address inversion as a means by which to (re-)consider more broadly how structure is used to convey meaning in vocatives and, by extension, the nominal domain. The present study profits from two major lines of enquiry: firstly, the resurgent interest in the syntacticization of discourse, in particular the body of work which proposes to extend the functional structure to capture aspects of the context of the utterance (Speas & Tenny 2003;Haegeman & Hill 2013;Wiltschko 2021); and, secondly, proposals for a reconceptualization of the grammatical architecture as a unified system for generating referential units of language (Longobardi 2005;Sheehan & Hinzen 2011;Hinzen & Sheehan 2013).
Through comparative examination of the empirical patterns of Italo-Romance vocatives with and without allocuzione inversa, I argue that the structural and interpretative properties of address inversion in southern Italian dialects are systematically configured in the nominal domain by a 'topological' architecture of grammar. Empirically, I claim that address inversion is a non-argumental phenomenon occurring in fully-fledged, autonomous VocPs which, like 'regular' vocatives, incur an ostensive-deictic (Espinal 2013:114) interpretation, and where the kinship reversal, or lexical 'flip' (i.e. the non-literal interpretation of N), is a surface manifestation of movement of the lexical noun into VocP.
Theoretically, I argue that, despite the explicit exclusion of vocatives from earlier iterations of the framework (e.g., Longobardi 1994Longobardi , 2005, the theory of grammatical reference predicts that the referential 'strength' of vocatives should incur a commensurate expansion of the nominal edge. I argue that this prediction is borne out in the internal syntax of southern Italo-Romance vocatives with and without address inversion. I thus propose to recast the internal mapping of the extended nominal left periphery in terms of grammatical reference and, in so doing, make the case for a conceptual rethinking of how we model vocatives and, by extension, the grammar-discourse interface in the nominal domain.
The article's argumentation is organized as follows. Section 2 introduces the theoretical background and the framework adopted in the study. Section 3 outlines the key empirical observations, with special reference to relevant contrasts between the properties of Italo-Romance vocatives with and without address inversion. Section 4 undertakes a critical review of the theoretical foundation for a new formal analysis of the internal syntax of Italo-Romance vocatives and address inversion. Section 5 presents the analysis itself. Section 6 concludes.

Vocatives and functional structure
Theoretical consideration of the internal syntax of vocatives has conventionally departed from the assumption that vocatives are structurally reduced nominal items, with much of the debate in the formal syntactic literature centred on whether vocatives constitute NPs or DPs (Longobardi 1994;Crisma 1997;Moro 2003;Coene & D'hulst 2003;D'hulst et al 2007;Bernstein 2008a,b). The proposal for further structure in the nominal domain (Hill 2007(Hill , 2013(Hill , 2014Espinal 2013;Stavrou 2013;Ritter & Wiltschko 2019, 2020; González López 2020) has nonetheless been gaining ground in recent years, in large part due to renewed formal theoretical interest in the so-called syntax of speech acts (Speas & Tenny 2003), a line of enquiry which seeks to incorporate aspects of interaction-oriented pragmatics into the functional structure.
In the formal literature on vocative syntax, the functional layer beyond D has most commonly been labelled VocP. I follow this convention, with the proviso that my operationalization of this nomenclature does not imply that I consider vocatives to be the only nominal expression which has a direct interface with, or otherwise acts upon, the extensional world (cf. Ritter & Wiltschko 2020:15 for comparable arguments motivating their rejection of VocP as a label for the leftmost nominal layer). Indeed, in Corr (2022), I explore the possibility that the grammatical and interpretative properties of nominal interjections may also necessitate expansion of the nominal architecture. 4 Espinal (2013) distinguishes between 'true' deictic vocatives (e.g., Tu! 'you.2SG' and 'fake' vocatives in complex nominal expressions. These are proposed to constitute the nominal parallel of Higgin's (1979) typology of copular sentences, and can be subdivided into 'identity' (e.g., Cat. Tu! Joan! 'You! Joan!'), 'identificational' (Tu! El noi de la camisa blava! 'You! The boy in the blue shirt!'), and 'predicational' (Tu! Noi! 'You! Boy!') types. 5 In light of their exclusion from the taxonomy of address inversion (Akkuş & Hill 2021), the present article omits discussion of the formal analysis of these mutable address forms. contextually given set (Schaden 2010:181) within the speaker's deictic frame. In this sense, their referent corresponds to individuals who are not (only) "spoken of" but, more importantly, "spoken to" (cf. Göksel & Pöchtrager 2013:89;Anderson 2007:215). The utterance of a vocative in this way identifies an extensional referent in the speaker's immediate non-linguistic environment, where this extensional referent is necessarily specific, familiar and unique. Furthermore, vocatives are 'rigid designators', since they "designate the same object in all possible worlds" (Kripke 1971:145). Additionally, they pattern with 'essential indexicals' (Perry 1979), in that they admit substitution salva veritate without their reference failing (cf. §5.4).
However, beyond establishing their extensional referent, vocatives do not (only) 'describe' (or predicate some property onto) their referent but do things in relation to that individual (d' Avis & Meibauer 2013:192;Corr 2022). In particular, vocatives are used to activate, confirm, or uphold the addressee status of their referent (Schegloff 1968;Zwicky 1974;Schaden 2010:182;d'Avis & Meibauer 2013:197), and to anchor, maintain and modulate the social dynamic between speaker and addressee (Zwicky 1974:795;Haegeman & Hill 2013;Paul & Pan 2017;Ritter & Wiltschko 2019, 2020. In light of these properties, I consider vocatives to be referentially strong (in the sense of Martín & Hinzen 2014;cf. §2.3), individualdenoting constituents which mediate between the linguistic system and the exophoric world to which they point (i.e. they are ostensive) and in which they are anchored (i.e. they are deictic).

Theoretical framework
The analysis proposed in the present article is theoretically formalized within the Topological Mapping Theorem (henceforth, TMT) of the grammar of reference (Longobardi 2005;Sheehan & Hinzen 2011;Hinzen & Sheehan 2013;Martín & Hinzen 2014). Under the TMT, words do not have a fixed semantic type, but instead their mode of reference is configurationally determined by the geometry of their grammatical architecture. Specifically, lexical (or 'descriptive') content enters the derivational structure by merging in the interior of the following template: (2) The Phasal Template (Sheehan & Hinzen 2011;Hinzen & Sheehan 2013) [ Edge [Interior]] The referential specification of the constituent, however, is determined by expansion of, and movement into, the grammatical 'edge' of (2), wherein the grammatical complexity of a constituent directly maps its referential complexity in line with the following hypothesis: (3) The grammar-reference link hypothesis (Martín & Hinzen 2014:102) Referential strength (from predicativity to deixis) is not an intrinsic property of lexical items, but rather of certain grammatical configurations.
This hypothesis predicts that the more grammatically complex a structure becomes through expansion of its edge, the more 'referential' the expression becomes. Thus, the larger the grammatical structure, the more reliant that structure is on the grammar itself to produce its meaning and, by similar reasoning, the less reliant on the lexicoconceptual content of its interior.
The template in (2) applies cyclically via a part-whole cycle which is proposed to yield the major units of computation, viz. 'phases', of recent formal syntactic theory: Crucially, the phases yielded via (2), i.e. DP, vP and CP, are proposed to correspond to key semantic types: respectively, individuals/objects, events, and propositions. The theoretical implication of this is that syntactic computation is reenvisaged as generating referentially significant units of language, rather than semantically autonomous ones (Sheehan & Hinzen 2011;Hinzen & Sheehan 2013). Finally, three key mechanisms are proposed to give rise to the variation observed in the surface patterns of given semantico-syntactic units across languages: overt (5a) and covert (5b) movement of the interior to the edge, or, alternatively, expletive associate chains (5c) formed through the insertion of morpholexical material in the corresponding referential layer (examples adapted from Longobardi 1994:623, andHinzen 2011:8, 10 On this model, whilst the surface manifestation of a certain linguistic object varies from language to language (as illustrated above for an argumental DP with 1SG possession in English and Italian), the grammar will produce the same structural topology to yield the appropriate semantics for that constituent. These mechanisms find correlates in, and can explain, even the most fine-grained microvariation (cf. Longobardi 1994:622;Sheehan & Hinzen 2011:9). However, as I will discuss in §4.1, the theory of grammatical reference as articulated by Longobardi (2005) for the nominal domain is explicitly built on the distinction between argumental and non-argumental nouns, such that vocatives are presently excluded from a topological analysis. This article contests this conclusion. My contention is that if indeed grammatical complexity co-varies with referential strength as the TMT principle predicts, then vocatives, qua referentially 'strong' nominal constituents ( §2.2), should correspond to grammatically complex constituents with an expanded edge proportional to their referential potential. Italo-Romance 6 Note that the extent to which (5a) is available for argumental DPs varies across Italian dialects (Longobardi 1994:623). vocatives with (and without) address inversion, I propose, provide empirical corroboration of this hypothesis.

Empirical preliminaries
Amongst Romance languages, address inversion is found in Daco-Romance and, dialectally, in southern Italo-Romance. 7 It is also attested in numerous languages outside these branches in the wider geographic region (i.e. beyond the corresponding territories of the Romània), irrespective of typological proximity. The data presented here are mostly taken from the address inversion literature (cf. fn1), although some have been elicited from L1 speakers or are spontaneously produced tokens.

Functions
First, as outlined in §1, address inversion involves a lexical 'flip' of the expected term for the addressee, incurring the expression of an affective stance between interlocutors. In (Italo-)Romance, this usage is asymmetric-from senior to junior-within the interpersonal dynamic (6a), such that a parent or caregiver can use address inversion when speaking to a child, but not vice versa (6b). 8 (6) a. Abruzesse (Rohlfs 1969:32) nəm plagnə, la mamma not cry.IMP the mummy 'Don't cry, my darling' (mother to son) b. Southern regional Italian (Sgroi 2017) Ascolta, {papà/*figlia} listen.IMP daddy/daughter 'Listen, {daddy/*dearest daddy [lit. daughter]}' (from daughter to father) That address inversion expresses affect in an asymmetric power relation sets these vocatives apart from 'regular' vocatives insofar as the latter permit, but do not require, the expression of social deixis in this way.
'Regular' vocatives and address inversion vocatives are further differentiated in terms of their interpretation insofar as the former can be used either as calls or addresses (Zwicky 1974), whereas the latter are restricted to the address function in Italo-Romance (Rohlfs 1925:440;Mazzoleni 1995  The morpholexical item a that introduces the lexical N in (7) is a reflex of the Latin preposition AD 'to, towards' and may have originated as a dative expression (i.e., It. veni a mama ('come to mummy', reanalysed as 'come, my dear'). Given its synchronic function is to select a non-thematic (vocative) XP and not to mark dative case, as well as the absence of prepositional marking of address inversion cross-linguistically, I treat the item as a prenominal particle which is homophonous with the modern Italo-Romance preposition of the same (7) Southern regional Italian (Iovino & Rossi 2014:220) Context: A mother calls from the balcony to her child who is playing with other children below. Torna a casa, (a) {Giovanni / # mamma}! return.IMP to home PTC Giovanni mummy 'Come home, {Giovanni / #my dear}!'

Forms
In many cases, Romance address inversion vocatives have the same surface form (i.e. are undifferentiated from) 'regular' vocatives. Unlike 'regular' vocatives, though, their productivity is mostly or exclusively restricted to kinship terms (8a), with the caveat that figurative extensions beyond this semantic field are nonetheless amply attested in southern Italian dialects (8b). Some L1 speakers even admit personal names (8c).
(8) a. Southern regional Italian (Bazzanella & Gili Fivela 2004) ma non è vero papà but not be.3SG true daddy 'But it's not true, my dear [child]' b. Sicilian (Sorrento 1915:113ff) Veni cca, {a signura / u mastru} come.IMP here the lady the master 'Come here, maid/boy' c. Southern regional Italian (Iovino & Rossi 2014:223) Mancia la minestra, % {a Guido} eat.IMP the soup PTC Guido 'Eat up your soup, my dear' (parent/Guido to child) Whilst (8c) is not felicitous for all Italo-Romance speakers who otherwise admit address inversion, Iovino & Rossi (2014:223) report that the usage of personal names becomes increasingly acceptable for a wider number of speakers as a function of the degree of affect expressed via, e.g., hypocoristic and/or diminutive forms.
Amongst southern Italo-Romance address inversion vocatives whose surface form is distinct from 'regular' vocatives, the most common differentiation strategies are i) the co-occurrence of the vocative with a prenominal definite article (8b), violating the generalization that prenominal articles are disallowed with regular vocative XPs (cf. §4.2), 10 or ii) the insertion of a prenominal particle (9a) (cf. fn9) where a 'regular' vocative would use an alternative form (9b). Nescisti, a matre? leave.2SG PTC mother etymology (see Sgroi 1983 for reflections on the putative 'prepositional/dative' status of address inversion vocatives in Sicilian). In the present article, I use the label PTC (for 'particle') to reflect the ambiguity of its grammatical status, thereby differentiating it from morphemes with an unambiguously dedicated vocative function (cf. the contrasting glossing of the prenominal particles in the vocative expressions of (10a-b)). 10 Though see Corr (2022:70-3, 87-8) for discussion of the numerous Romance counterexamples to the generalization that vocatives cannot co-occur with definite articles. Furthermore, in regional Italo-Romance, vocatives with and without address inversion exhibit contrasts in i) the licensing of syntactic gemination (raddoppiamiento fonosintattico, RF); and ii) truncation: (10) Southern regional Italian (Iovino & Rossi 2014:222) Me lo fai un caffè… 'Can you make me a coffee…' Specifically, in varieties allowing both address inversion and RF, gemination is admitted only in address inversion vocatives (10a), whereas the opposite pattern is true for truncation, i.e. the address inversion reading is ruled out with truncated vocatives (10b).

Contexts
Insofar as address inversion involves a lexical form which 'masks' the actual identity of its 2SG referent, these vocatives look at first glance like 'imposters', viz. a subset of third-person nominal expressions which denote the speaker or addressee (Collins & Postal 2012;Collins 2014 In (11a), the proper name Emilio, used by the mother to address her child, cannot be substituted by an address inversion expression (e.g., the DP la mamma, idiomatically translated here with the non-literal meaning of 'my darling child'). Likewise, in (11b), the DP la mamma 'the mummy' can only be used by the mother/speaker to refer to herself in the third person when addressing her child, in which environment the address inversion interpretation is ruled out.
Regarding their external syntax, address inversion vocatives in southern Italian dialects can occur as the sole vocative in an utterance, in which context they typically appear utterance-finally (12a). However, they can also co-occur either adjacently (12b) or discontinuously (12c) with another vocative in the sentence.

Interim summary
The fundamental referential function of vocatives with address inversion is to i) identify an object within the speaker's hic-et-nunc as the addressee of the utterance, and ii) undertake an action-viz. an expression of the speaker's affective stance-in relation to the extensional referent of that object. In Italo-Romance, these vocatives diverge systematically from those without address inversion across a range of diagnostics in terms of their function, form and the environments in which they are licensed, as summarized in Table 1. Altogether, the data outlined in the preceding subsections provide persuasive evidence that address inversion expressions can be distinguished from-or, more precisely, converge as a coherent (sub)group within-the wider class of vocative XPs in southern Italian dialects.

Theoretical foundations for a topological (re-)modelling of vocatives
The empirical observations of §3 lead us to the conclusion that southern Italo-Romance address inversion is a phenomenon operative at the grammar-discourse interface and, accordingly, merits analysis in these terms. This section details the theoretical foundations for our formal analysis.

Grammatical reference and the nominal domain
To recap, the TMT approach posits that words gain reference through insertion into the grammatical structure in line with a basic configurational 'phasal' template composed of a descriptive 'interior' and a grammatical 'edge' (2) whose structure is expanded through a part-whole cycle (Sheehan & Hinzen 2011;Hinzen & Sheehan 2013). As that grammatical structure increases in size by extending the structure at its edge-as configured via the formal mechanisms of overt/covert movement of its interior content or expletive-associate chains-so too does its referential strength, in line with Martín & Hinzen's (2014) grammar-reference link hypothesis (3). Let us examine how this model is implemented in the nominal domain. The lexicalization of N alone (i.e. a nominal structure with an empty edge) in an argument position yields an object with no presupposed existence (15a), whilst lexicalization of D (15b) and N-to-D movement (15c) respectively yield an indefinite/definite and maximally-specific 'rigid' (in Kripke's sense) DP. Deictically anchored expressionswhich, as we will see in §5.1, involve further topological complexity in the D-layer Crucially, with deictically-anchored expressions, the meaning of the nominal expression can be configured entirely at the edge, as demonstrated by the optional realization in (15d) of the demonstrative in D without the lexical N, and in the case of personal pronouns, these merge directly in D, leaving their interior present but empty.
In this sense, grammatical reference (on this framework) is an edge phenomenon, conceived as the point at which language-as-grammar 'touches' the world (Hinzen & Sheehan 2013:319). However, Longobardi's (2005) initial proposal for a unified grammar of reference is explicitly formulated in relation to determinerless arguments based on the complementary syntax-semantics of argumental versus non-argumental nominals. His proposal departs from the generalization in Longobardi (1994) that bare common nouns cannot be referential and do not undergo N-to-D (as in (15a)), whereas proper names are (object-)referential and involve N-to-D (as in (15c)). 11 Building on this observation, Longobardi (1994Longobardi ( , 2005 proposes a biconditionality between the projection and lexicalization of D, on the one hand, and object reference-defined as constant denotation to particular individuals-on the other. Longobardi (2005) further notes that non-argumental nominals escape these constraints (e.g., It. Presidente, venga qui!, 'President, come here!', Maldetto maiale! 'Damn pig!', ibid.:24). This leads him to conclude that the projection and lexicalization of D is a condition for argumenthood (Longobardi 2005:27, citing Szabolcsi 1987Stowell 1989Stowell , 1991Crisma 1999), and vice versa.
As non-arguments, vocatives are thus explicitly excluded from the TMT mechanisms that yield (grammatical) reference, which is understood as "the property of certain expressions of necessarily designating one and only one fixed individual entity" (Longobardi 2005:13, fn12). However, this exclusion comes with the reservation that vocatives may instead be exempted "from the argument-position constraints via a different mechanism" (Longobardi 2005:25, citing Crisma 1997my emphasis). Longobardi (2008:199) returns to this suggestion, mooting that vocatives, given their individual-denoting character, may have "a peculiar way to license an 11 Recall that, under the TMT, N-to-D does not require overt movement but can be accomplished via covert movement or an expletive-associate chain. empty D". 12 Ultimately, then, Longobardi (2005Longobardi ( , 2008 acknowledges the need to identify a mechanism which can capture both the grammatical behaviour and the referential properties of vocatives, thereby implicitly paving the way for a topological reinterpretation of the extended nominal structure to account for vocative XPs.

Previous approaches
The hypothesis that (bare) vocatives remain low in the structure (i.e. do not undergo N-to-D/Voc) adheres to the theoretical assumption that vocative XPs are structurally 'small' (cf. §2.1). The characterization of vocatives as structurally reduced nominal items-articulated, crucially, in opposition to argumental constituents (Zwicky 1974;Longobardi 1994Longobardi , 2005Stavrou 2013)-is seemingly premised on the frequency of bare nominal expressions and absence of definiteness marking observed across vocative XPs cross-linguistically: (16)  The ostensible ban on article realization with vocatives has been analysed in the formal literature as reflecting an empty (Szabolsci 1994;Longbardi 2008) or absent (Longobardi 1994;Stavrou 2013;Hill 2013:149) D projection in vocative phrases.
However, as observed by Moro (2003:255), Italo-Romance provides plentiful empirical evidence that challenges the hypothesis that vocative phrases do not project D: (17) Italian (Moro 2003:255) a Thus we observe within the vocative expression the realization of pronouns (17a) and prenominal demonstratives (17b), both of which, following standard assumptions, involve the lexicalization of D. In (17c), we observe (again, interior to the vocative expression) N-to-D raising over a possessive adjective. 12 See also Longobardi (1994:626) for early speculation that vocative particles may lexicalize D.
These data lead Moro (2003:255) to conclude that the non-realization of the article with vocatives cannot be attributed to the absence of D but may be "possibly related to the referential capacities of the noun phrase involved", an intuition which motivates many of the theoretical explanations for the empirical patterns of vocative syntax cross-linguistically. For example, Longobardi (2005:24) ventures that vocatives have a "special status" due to their "highly individualising function […] creating a direct relation between the speaker and the hearer" which leads them to be "semantically incompatible with indefiniteness".
Indeed, recent expansions of the nominal structure are premised, at least in part, on the need to capture the extensional semantics of vocatives as specific, familiar and unique. Thus Hill (2013Hill ( , 2014Hill ( , 2017 (2013:336), on the other hand, motivates the presence of D in vocative expressions on account of the observation that vocatives are always specific, which, for her, is "a corollary of the fact that [the vocative noun] denotes an addressee-i.e., has a second person feature". Similarly, Espinal's (2013) distinction between "true" and "fake" vocatives (cf. §2.1) is motivated precisely on divergences in the referential properties between different types of vocative constituents in complex nominal expressions.
Such intuitions also underpin the formal analysis of Daco-Romance address inversion offered in Hill (2013Hill ( , 2014Hill ( , 2017 and related joint work (Akkuş & Hill 2017). In their analysis, Romanian address inversion vocatives merge as kinship DPs within a larger, complex VocP (18b), where Voc 'splits' into separate heads to accommodate what is framed as a semantic mismatch between second-person marking (as in Hill's other work, a formal feature which requires checking on Voc) and an uninterpretable feature [KIN] (introduced, they propose, in cases of address inversion). For these authors, [KIN] is incompatible with the "inherent" (Hill 2014:58, et passim) second-PERSON specification of vocatives because "[kin] is intrinsically marked for 1 st person [1] (that is, the kin person who intends to exert authority)" (Akkuş & Hill 2017:61).
In their work, the vocative expression in an utterance with address inversion such as the Daco-Romance example in (18a) is proposed to have the structural representation represented in (18b). On their analysis, a dedicated kinship DP (viz. the address inversion vocative) merges in the lower SpecVocP and 'checks' the [INTERPERSONAL/KINSHIP] feature. The 'regular' vocative merges in N and undergoes head movement to a lower Voc, yielding the kinship relation by entering into a local Spec-Head relation with the kinship DP, before undergoing further movement to a higher Voc to check the second-person feature [2] to obtain the required addressee reading. (The latter step is reflected schematically in (18b) by the affixation in Voc1 of the vocative morpheme -e on Dane.) Whilst the derived hypothesis for address inversion is not theoretically implausible per se, the postulation of bespoke formal features such as [FAMILIARITY] or [KINSHIP] runs the risk of being stipulative. Moreover, its formal implementation as proposed by these authors runs into some empirical problems. Notably, the derived hypothesis as articulated in Hill and Akkuş's work is reliant on the co-occurrence of both the 'regular' and the address inversion vocative i) in the same sentence, ii) in linear adjacency, and iii) within the same prosodic unit. However, the widespread attestation of address inversion vocatives in the absence of a co-occurring 'regular' vocative (observed in examples throughout this paper) as well as the attestation of vocatives co-occurring discontinuously in a single sentence (as in (12c)) undermines this analysis. Conversely, if co-occurring vocatives are in fact a single complex VocP, the impossibility of two utterance-final vocatives reported by L1 speakers (as in (13c)) remains unaccounted for. 13 Moreover, Hill (2017:348) argues that the merging of a possessive modifier in 'regular' VocPs prevents the co-occurrence of address inversion in such strings (Romanian Dănuţa {mea/mamă} 'Dănuţa {my/mummy}', said by mother to child), since the two are in complementary distribution, a proposal which does not hold for Italo-Romance (Calabrese Nicola meu, a nonna 'Nicola my.MSG, the granny'; Abbate 2010:152). 13 See also the co-occurrence of three vocatives in reg. It. Licia, papá, suona, papá 'Licia, darling, play, darling' (Chieti, Abruzzo;Rohlfs 1925:29 Another issue for their analysis is that address inversion is predicted to occur only in languages that admit both head movement from N to Voc and the merging of a vocative particle in SpecVocP (Akkuş & Hill 2017:62). This restriction is proposed to account for the absence of address inversion in languages without both parametric options (e.g., English, Greek), yet address inversion does indeed occur in at least some such languages (e.g., 'What's the matter, mama?', said to young son by monolingual English L1 speaker).

The nominal edge revisited
So far, we have observed that the TMT framework provides a formal mechanism and explanatory principle for capturing the interface between grammar and the nonlinguistic world, yet the TMT has not, to date, been utilized to capture the grammatical and referential properties of vocatives. On the other hand, recent theoretical work on the internal syntax of vocatives has motivated the expansion of the nominal architecture to satisfactorily accommodate the syntactic and semantic properties of vocatives. However, despite many such accounts relying on semantic insights to motivate their structural proposals, work in this area arguably does not yet offer a principled explanation of the empirical behaviours of vocatives more generally. 14 I propose that the TMT offers such a framework (cf. Corr 2022:1-5, 41-8 for further discussion). Building on the groundwork laid in Martín & Hinzen's (2014) expansion of the nominal TMT template to capture deictic and person reference within an internally-articulated 'double' DP structure, the remainder of this article sets out the principal arguments for a topological recasting of the extended nominal edge of Italo-Romance vocatives and allocuzione inversa, including the division of labour between the DP and the 'utterance' layer-labelled here as Voc[ative]P following convention (cf. fn3)-above it. In particular, I hold that a TMT analysis has the following advantages: i) the properties of Italo-Romance address inversion vocatives follow from the predictions of the TMT; moreover, a TMT analysis ii) captures a number of key insights from existing formal analyses that propose an expansion of the nominal architecture to include VocP; iii) avoids the stipulation of bespoke or post-14 An anonymous reviewer suggests that Ritter & Wiltschko's (2020) analysis of vocatives-formulated under Wiltschko's (2014Wiltschko's ( , 2021 Universal Spine Hypothesis, USH (latterly ISH, for 'interactional')-is "also not ad hoc" given that their nominal interactional structure is motivated by i) the existence of an interactional structure at the height of the clausal architecture, and ii) the parallelism between nominal and clausal projections more generally (the same case is made in Ritter & Wiltschko 2019). Ritter & Wiltschko's arguments are, to my mind, convincing. Indeed, the motivation is fundamental to the approach reported here, although it has not been explicitly stated as such, and the reviewer is right to flag its importance. Note, however, that these authors are not the only ones (nor the first) to make this point in motivating an expansion of the nominal left periphery (e.g., Hill 2017:338). Moreover, the parallelism argument, unlike the TMT and its accompanying grammar-reference link hypothesis (3), does not motivate why the leftmost space should be the locus of the grammardiscourse interface in the first place. In other words, the parallelism argument, although theoretically principled, is not principled from an explanatory point-of-view, whence my contention that a topological analysis of vocatives has an advantage vis-à-vis alternative theoretical approaches precisely in that the TMT-as argued at (book-)length by Hinzen & Sheehan (2013)-offers an account of the empirical facts that carries explanatory significance. hoc features to account for the 'special' behaviour of vocatives; and iv) can account for the empirical peculiarities of (Italo-Romance) vocatives more generally beyond address inversion, where these properties are not 'anomalous' but predicted by TMT principles.

The role of the DP
Our starting point is the empirical observation that Romance vocative expressions can be formed with personal names and pronouns, and that these can co-occur with a dedicated vocative particle: The facts of the internal composition of (19a-b) necessarily implicate the projection of D in such expressions, since under the TMT framework, personal names and pronouns respectively involve movement into, and direct merge of, morpholexical material in the nominal edge (cf. §4.1). More precisely, Martín & Hinzen (2014) propose that the internal articulation of the DP-layer is responsible for topologically mapping finegrained distinctions in the referential capacities of argumental XPs (a hypothesis in line with a healthy literature on the 'split DP', and analogous ideas that favour the structural decomposition of previously unitary syntactic objects, e.g. . 15 DeixP is the locus of deictic reference (in the broader, rather than narrowly exophoric, sense) 16 and supplies a referential index to the nominal (Jayaseelan & Hariprasad 2001:140). This layer, whilst often silent, shows up overtly in the pronominal systems of some languages (e.g., the Catalan third-person dative clitic els hi /əlzi/, cf. Martín 2012:ch3). Adherence to the grammar-reference link hypothesis (3) mandates that the internal syntax of (19a-b) must contain as subparts the (grammatically configured) personal name Gianni and the personal pronoun tu/te ('you.2SG'). For the personal pronoun merged inside (19b), I propose that we can adopt wholesale the TMT proposal of Martín & Hinzen (2014), on which the pronoun merges directly in the lower Dhead, 17 before undergoing head movement to the higher D-node where it receives its formal specification as [+PERSON], picking up its referential index from DeixP along the way. In the case of personal names (19a), however, I propose a slight adjustment 15 Where capitalized, Person refers to syntactic Person, as opposed to the descriptive category of person (which is left uncapitalized). 16 Cf. Martín (2012:56-61) for a cogent explanation of this point. 17 On their account, the lower-D is the locus of gender and number features.
to Martín & Hinzen's (2014) configurational mapping of proper names. Their argument for extra D-structure is premised, in part, on the need to distinguish between "merely referential 3 rd Person nominals, including proper names", and deictic forms of reference (which on their account include, among others, Romance third-person dative clitics and differentially-marked 'strong' accusative clitics), whose properties align these elements more closely with essential indexicals (Martín & Hinzen 2014:110). Yet proper names are not a uniform class, as demonstrated (amongst others) by Longobardi (1994Longobardi ( , 2005Longobardi ( , 2008 for Romance versus Germanic. For example, Ledgeway et al. (2019) have recently shown that dialectal microvariation in the (non-)licensing of differential marking with direct object proper names in Calabrese is correlated with the ability to value [+PERSON] through the lexicalization of D° (2019:17).
Combining this insight with Longobardi's (2008:201) proposal that personal names and pronouns involve movement to "the Person position" on D, I claim that at least some proper names, following N-to-D substitution (cf. (5a), (15c)), should also undergo D-to-D movement (with the further implication that their structures must also project DeixP). An obvious candidate for such DPs are personal names that form (part of) a vocative expression. On this view, N-to-D-to-D movement allows N to acquire the properties that give rise to the "intrinsically" specific and unique nature of vocatives (cf. e.g., Stavrou 2013:336). I assume that the initial N-to-D movement imposes a uniqueness condition in the lower D, syntactically expressed as [+DEFINITE]. 18 Projection of DeixP supplies the expression with a referential index, enabling it to 'point' to the wider linguistic context, although-crucially-not necessarily the non-linguistic one. Head movement to the higher D activates [+PERSON], a (primitive) syntactic feature ordinarily associated with the morphological exponence of grammatical Person, but conceptualized here, following Longobardi (2008) and Bernstein (2008a,b), as a necessary formal condition for type-shifting from property to individual denotation. This minimal [+PERSON] specification is understood here to be a necessary, but not sufficient, condition for exophoric reference (i.e. allowing a DP to single out an individual in the non-linguistic context). Moreover, the involvement of the higher D-head and Person in the licensing of VocPs captures the essence of various theoretical proposals which hold this position and/or feature to be more widely responsible for various 'specificity', 'individuation', 'particularized', and 'humanness' readings observed for 'highly referential' lexical DP arguments (e.g. Bernstein et al. 2018;Ledgeway et al. 2019). Namely, if these readings are corollaries of the structural presence of the higher D-node and/or [PERSON], then it follows that VocPs-whose internal configuration contains a [+PERSON] DP-should necessarily incur those interpretations. 18 The suggestion, then, is that vocatives do involve definiteness, at least syntactically. A neat advantage of this hypothesis is that it parsimoniously explains why definiteness marking, contra to common understanding, does regularly show up in vocative expressions (cf. fn10). Whilst definiteness marking has been associated with pragmatic-interpretative effects (cf. Corr 2022:71ff and references therein), the implication here is that lexicalizatio n of D by definiteness markers incurs such pragmatic-interpretative effects as well as (as opposed to rather than) expressing definiteness in vocatives. My hunch is that affective interpretations associated with D-lexicalization arise as a conversational implicature, rather than being syntactically encoded per se (pace e.g. Hill 2014:65).
Note that there is no (potential) feature mismatch in terms of Person with respect to an interlocutor (pace e.g. Bernstein 2008a:227;Hill 2014:60ff, and passim), since I am not proposing that so-called 'third-person' DPs are formally specified as such (cf. Harley & Ritter 2002:486-8, and references therein). That the definite article can also show up in vocative expressions (pace, e.g. Bernstein 2008a,b;Ritter & Wiltschko 2020) suggests that the proposed absence of formal [PERSON:3] can be extended to DPs with apparent third-person marking (in which case, such marking can be theoretically interpreted as 'default ', i.e. non-specified, morphological exponence, cf. Benveniste 1966:256). A corollary advantage of this position is that it is consistent with the empirical observation that second-person forms (i.e. formally [PERSON:2]) and third-person forms (i.e. [+PERSON] only) are systematically compatible with respect to an addressee (in argumental and nonargumental contexts alike), but first-person forms ([PERSON:1], which would incur a potential feature mismatch) are systematically incompatible in vocatives.

Motivating a tripartite template
The preceding section made the case that 'regular ' vocatives involving particle-N combinations (19a-b) have in their interior a maximally-expanded DP with a [+PERSON] value and a referential index, structural properties which are a precondition (though not a sufficient one) for exophoric reference. Collapsing the DP complex for expository convenience, the empirical evidence from such expressions requires us to posit extra structure at the phasal edge to accommodate the overt morpholexical material in these XPs (i.e. Gianni/tu/te, which must be at least in D, plus the prenominal particle above it): Crucially, expansion of the functional structure beyond D (viz. the locus of proper names and pronouns in their argumental function) is not only motivated by the empirical data (i.e. to produce the prenominal placement of the vocative particle). It is also mandated by the TMT requirement that the topological architecture "carve out" (Arsenijević & Hinzen 2012:432) the interpretative divergence between personal names and pronouns in their argumental function (15c,e) and when, in this case, they merge in the interior of an expression with 'stronger' referential-deictic potential (19ab). (Notably, the argumental personal names and pronouns in (15c,e) can have endophoric or exophoric reference, whereas in vocatives, their reference is necessarily exophoric.) Our framework thus predicts an increase in complexity in the internal build of vocatives vis-à-vis argumental DPs. The key idea is that vocative expressions are formed of an internally-complex DP layer but must also project further functional structure beyond D, differentiating them from argumental DPs (i.e. the wider class of linguistic objects with individual denotation). Once the structure of DP has been configured ( §5.1), the 'utterance' layer above it-identified here as VocP (cf. fn3)is responsible for narrowing down the referential specification of the individual with respect to the non-linguistic world, 19 constraining aspects of that individual's relation to, and interaction within, the utterance context.
This yields an expanded edge as in (21a), which, I suggest, can be captured as a tripartition of the basic template for the nominal functional structure (21b).
The crucial implication of (21a-b) is that, since TMT principles require the same configurational typology for linguistic expressions with the same fundamental deicticreferential potential, vocative XPs should also involve the projection of Voc even in the absence of an overt vocative morpheme (i.e. (22a) = (22a')): In such cases, I suggest that further movement of the (bare) N directly into Voc (22b) is required in order to designate the individual-which has been grammatically configured in DP-as the addressee. Whilst I remain agnostic on what, precisely, this corresponds to in featural terms, I follow Espinal (2011Espinal ( , 2013 in understanding that the projection of, and movement into, Voc must involve a syntactic mechanism that designates the nominal expression as the addressee. The motivation for this requirement-and a suggestion which is hardly novel-is that bare lexical DPs have no relevant properties in of themselves (unlike, e.g., second-person pronouns) that would enable them to grammatically pick out an addressee. Given I have identified [+PERSON] as a necessary condition for this task, I assume that this feature is also somehow implicated in elevating an otherwise 'third-person' nominal to addresseedesignator status. (I am not, however, suggesting that such DPs are specified as second-person within the DP itself. On the other hand, I do not rule out the possibility, as defended by Hill (2014Hill ( , 2017, that such DPs value [PERSON:2] in/via Voc, although I have reservations about this proposal.) 20 The foregoing schema for VocPs has three central advantages. First, the configurational derivation of the extensional semantics of vocatives via N-to-D-to-Voc movement (as also proposed by, e.g., Espinal 2011, 2013) makes vocatives not 'inherently' (that is to say, arbitrarily) referential but obligatorily referential. On this view, their referential status is simply a structural consequence of the key TMT principle that reference is induced through expansion of and movement into the phasal edge via a part-whole cycle ( §2.3), which, in this case, mandates that N passes through D before projecting, and merging in, Voc. In this way, though the templatic schema in 19 See also Coene et al. (2019:105) on VocP as the layer which provides "object-referring expressions with additional information with respect to the addressee". 20 My agnosticism on the syntactic mechanism(s) involved in the grammatical configuration of addressee reference is not for lack of options, given the flurry of recent works on the syntacticization of the utterance and its participants in the nominal and clausal domains. Rather, it is because the issue merits careful consideration, and the enumeration and discussion of the various theoretical possibilities and their consequences requires a lengthier word count than the present format allows.
(21a-b) may at first blush seem merely to reiterate earlier proposals for an expanded nominal structure, the TMT tenet of the part-whole cycle requires us to reject for principled reasons proposals (such as those in Hill 2013 et seq.;Stavrou 2013;Eckardt 2014, i.a.) that allow for the possibility of an architecture for a nominal VocP which bypasses D. The proposal also enables us to dispense with some of the theoretical machinery of recent studies (e.g. Hill's (2014) [+SPECIFIC] uninterpretable feature, since specificity on our account comes 'for free' as it is subsumed under [PERSON], a necessary property of VocPs).
Secondly, the generalization of the schema in (21a-b) to all (Romance) vocatives with the same referential potential illustrates how VocP can be configured by the three mechanisms identified by Longobardi (1994Longobardi ( , 2005 for capturing crosslinguistic variation (viz. overt and covert movement of the interior to the edge, expletive associate chains). Thus, whilst bare lexical VocPs are proposed to undergo N-D-Voc movement (22b), the internal syntax of the Italian vocative O Gianni (19a) involves an expletive-associate chain configured by merge of the vocative particle (Voc PTC ) in Voc plus N-to-D movement. Direct merge of Voc PTC and Voc N in Voc and D,respectively,yields It. O tu/te (19b).
Finally, given the widely accepted hypothesis that kinship names exhibiting the behaviour of proper names also involve N-to-D movement (Longobardi 1994:625, fn19;1996;2005:16-18), we can extend the above-outlined proposal for proper names ((21b), (22b)) to VocPs involving kinship terms, including address inversion vocatives, without further stipulation. (Note that I will make some further amendments to the proposal in §5.3-5.Crucially, however, the underlying thrust of the analysis remains the same.)

Modifying the template
Whilst the baseline hypothesis for a tripartition of the nominal architecture remains unchanged, evidence from complex VocPs involving a particle-noun combination elsewhere in Romance requires us to make a further adjustment to the proposed template. Namely, we need to account for 'regular' vocatives with a dedicated nonthematic form (i.e. a surface form that is only felicitous in the vocative function) that co-occurs with a nominal vocative particle (e.g., Lat. O amice! 'O friend.VOC'; EuPt. Ó pá 'mate!'; Ast. Á ne 'kiddo!'). Observing that the lexical noun in such cases (e.g., EuPt. pá < rapaz 'boy'; cf. de Carvalho 2013:53) cannot be used predicatively (23a) nor argumentally (23b), I interpret these restrictions as evidence (pace, e.g., Ritter & Wiltschko 2020; cf. fnError! Bookmark not defined.) that N in such cases has moved out of the nominal core and into its edge beyond D (23c In (23c), argumental rapaz ('boy') is used as shorthand for the observation that pá retains lexico-conceptual (i.e. descriptive) properties, yet cannot be used as an argument.
On this reckoning, the distributional restriction of address forms such as (23c) to vocative environments is simply a surface manifestation of compulsory head movement into VocP. Theoretically, and in line with our general (i.e. exoskeletal) approach, I intuit a parallel here with the N-to-D movement that allows kinship names to exhibit the behaviour of proper names (cf. §5.2). Specifically, I suggest that N-to-D movement in such cases enables otherwise common nouns to undergo property-toindividual denotation shift en route to Voc (cf. also Espinal (2011Espinal ( :20, 2013:116f) on the extension of N-raising to common nouns in vocatives). As with the vocatives of (19a-b), further expansion of the nominal edge is required to accommodate dedicated vocative constituents, with the difference that, in the case of (23c), we have two dedicated vocative forms (N and the vocative particle) needing to be accommodated within the vocative layer itself (see Corr 2022:85ff for additional empirical evidence motivating the relative ordering of Voc PTC > Voc N ).
Further support for an expanded VocP comes from the 'ungrammatical' linearization of complex nominal constituents in Italian vocatives, which permit a linear order for adjective-noun combinations that is ruled out in argumental DPs. The linearization constraints on N-Adj placement in Italian are illustrated by the argumental DP in (24) Note, too, that the linearization constraints of (24) are not simply 'relaxed' in the vocative context, since the felicitous linear order of constituents within the argumental DP is judged marginal when the same order (i.e. Art-Poss-Adj-N) occurs in vocative XPs (26).
(26) ? O il mio caro ragazzo! o the my dear boy (Moro 2003:255) These apparent anomalies can be straightforwardly accounted for by the proposed expansion of the phasal template in (23c). Namely, the presence of the prenominal definite article il (24-25) requires us to assume that D is projected in both the argumental (27a) and the non-argumental (i.e. vocative) XPs (27b). That the leftward placement of the affective adjective vis-à-vis the definite article is ungrammatical in argumental constituents but licit in vocative constituents suggests that its landing site must be external to the argumental layer. Thus, as in (23c), the site of the highest two constituents must be within the vocative layer itself, as schematized in (27b) Assuming that the adjective caro undergoes phrasal movement to its landing site in VocP, it follows that a higher functional projection should be present in order to accommodate the merging of the vocative particle to the left of the specifier position to which the AdjP moves. Thus we arrive at an expanded nominal template which projects an internally articulated vocative layer beyond D: For us, the internal expansion of the vocative layer beyond D is not only empirically motivated, but has conceptual significance. Our point of departure in this regard is that, under Sheehan & Hinzen's (2011) formulation of the TMT hypothesis, D represents the structural locus of 'rigid' object reference and the point at which "contact with the extensional world is made" (Sheehan & Hinzen 2011:4). My proposal is that expansion of the nominal architecture beyond D yields a structural layer at the nominal 'edge' that acts as the locus of mediation between the nonlinguistic world and the individual 'carved out' by the topological architecture in D. That the 'mediating' role of this functional space should be conceptualized in terms of grammatical referential distinctions is necessarily entailed by the layer's internal complexity (as motivated above on empirical grounds). Accordingly, and as already outlined in §5.2, I understand that this layer serves to configure the ostensive-deictic possibilities of the object and, as such, provides instruction on how the object should be discursively interpreted on entering the extensional world (such as the various interpretative possibilities for vocatives summarized in §2.2).

Modelling the lexical 'flip'
For our purposes, however, the key advantage of the proposed template is that it simultaneously provides a means of modelling the lexical 'flip' observed in address inversion as a function of the grammatical encoding of the argumental vs. nonargumental distinction within the internal build of the vocative XP. Thus, in the same way that the dedicated vocative form pá in (23c) requires a landing site in Voc to license its non-argumental (and non-literal) function, I propose that the lexical 'flip' in Italo-Romance address inversion is derived by N-D-Voc movement to yield its nonargumental reading.
On A question that arises at this juncture is why further movement into the expanded edge should correspond to the possibility of yielding a non-literal meaning of the lexical material in vocatives with and without address inversion. Recall that a central prediction of the TMT is that the greater the grammatical complexity, the more the grammar itself is required to 'carve out' the meaning of an expression, with the result that the descriptive content of its lexical interior becomes increasingly redundant (and in some instances, e.g., demonstrative proforms, is dispensed with altogether).
In the case of vocatives, their extensional reference holds regardless of the descriptive content. Substitution salva veritate is permitted, in TMT terms, because the extensional reference of vocatives is induced at the grammatical root: i.e. the maximally-expanded (here, unembedded) structure which does not rely on (and can thus do away with) the descriptive content for its interpretation. As observed by Hinzen et al. (2014:320), that I can get the descriptive content of a vocative 'wrong' without the vocative's reference failing supports this. For example, I can address Bianca by mistakenly calling her Maria, but it is nonetheless still Bianca who is being referenced, i.e. identified as the addressee, through my utterance (cf. also Hinzen & Sheehan 2013:124).
With this in mind, the fact that the kinship term undergoes the 'lexical flip' becomes relevant to our analysis only inasmuch as i) it offers empirical confirmation that the descriptive properties of N are orthogonal to the referential specification of the VocP, and ii) its non-literal reading provides a key clue that N is no longer in the nominal core. That is not to say that the descriptive properties of N play no role in the vocative's meaning-clearly, the overt material of N continues to provide lexicoconceptual information, an observation which provides further motivation for the proposal that the item originates in N before moving upwards in the structure 22 -but that those descriptive properties have no special theoretical consequence on our current analysis. In this sense, there is nothing exceptional about address inversion relative to other non-literal and/or opposition-based uses of lexical material in vocatives. That is, the fact that a 'lexical flip' takes place is no different from any other (animate) common noun that ends up used non-literally in vocative expressions (cf. §5.3); nor (pace Akkuş & Hill 2017) does the kinship reversal, in of itself, 23 tell us anything about the putative syntactic representation of SPEAKER or proxies thereof, e.g., KIN (recall that KIN is "intrinsically marked for 1 st person", Akkuş & Hill 2017:61), in the nominal architecture. 22 On this point, my analysis diverges from Ritter & Wiltschko (2020:8-9), who propose that, in vocatives, names and terms of endearment (must) merge directly in the specifier of their 'interactional' structure above a pro-DP. 23 To be clear, this assertion does not exclude the possibility that SPEAKER has syntactic representation independent of formal [PERSON:1], only that such a conclusion is not necessitated by the lexical reversal characteristic of address inversion.

Differentiating vocatives with and without address inversion
The contention that the lexical flip itself is largely unremarkable (relative to analogous vocative expressions) from a grammatical perspective does not mean that there is nothing of theoretical relevance in the grammar of address inversion vocatives themselves. Indeed, our comparative exposition of the Italo-Romance data in §3 led us to conclude that vocatives with and without address inversion-in southern Italo-Romance dialects at least-diverge from one another in terms of both grammatical and interpretative properties. This raises the question of whether these empirical divergences are codified in the structural mapping of Italo-Romance vocatives.
First, it is important to underscore that the hypothesis that vocatives with and without address inversion project and lexicalize VocP corresponds to a theoretical instantiation of the key intuition that both types of vocative are fundamentally alike in their (ostensive-deictic) referential potential. This does not, however, preclude the possibility that these vocatives might diverge in their grammatical configuration at finer granularities. With these caveats in mind, evidence in favour of a structural distinction in the nominal syntax itself comes from southern Italian dialects with RF which, as we saw in §3.3, exhibit minimal pair contrasts in the licensing of the sandhi phenomenon according to the presence/absence of address inversion. Namely, (at least some) dialects permitting both address inversion and RF license the gemination process in address inversion VocPs (e.g., A ppapà! 'my son!') but disallow it in 'regular' vocatives (A (*p)papà! 'daddy!').
Crucially, formal investigation of Italo-Romance RF in other environments has shown that the two constituents on which the sandhi process acts are required to be in a local configuration (Fanciullo 1986:88, cit. Ledgeway 2018  In the above examples, RF obtains between the head and its complement (30a), specifier (30b), and a postnominal modifier (30c); and between specifiers (30d) in the left periphery of Cosentino nouns. What these operations all have in common is that they take place within the phasal DP. 24

24
For a complete exposition of the theoretical and empirical details, including corroborating evidence of the phasal application of RF in the clausal domain, see Ledgeway (2018:289ff The importance of these results, for us, is that it lends support to the hypothesis that the distinction between vocatives with and without address inversion is not merely interpretative but (can be) syntactic, and that, in at least some southern Italian dialects, the syntactic distinction manifests itself at the phasal level. Given that a phasal distinction between constituents entails a divergence in their respective grammatical structures, the requirement for structural distance between the particle and the noun in vocatives which ban RF (28b) introduces the possibility that these VocPs involve a larger internal structure vis-à-vis those vocatives permitting RF. The TMT prediction in such an instance is that these vocatives should also differ referentially. Recognizing that their necessarily affective interpretation and infelicity with a 'call' function differentiates address inversion expressions from the wider class of vocatives in southern Italian dialects, in Corr (2022:91-3) I argued in favour of the hypothesis that 'regular' vs. 'address inversion' vocatives correspond to a grammatically-configured contrast between call and address VocPs. 26 This proposal comes with a further (TMT) prediction: the call vs. address distinction should be configurationally mapped in the internal syntax of vocatives more generally. Why this distinction should have a phasal expression in the vocatives of southern Italian dialects-or, indeed, why linguistic objects that are otherwise alike in their deicticreferential potential should incur a phasal distinction in the first place-remains to be seen. 27 In view of my already-expressed agnosticism on the matter of the formal syntactic properties of VocP (cf. fn20), however, I abstain from committing to a particular position on these issues for the time being, pending more compelling evidence.
Provisos aside, support for the conclusion that 'regular' vocatives involve extra structure vis-à-vis vocatives with address inversion comes from the minimal pair contrasts in terms of truncation illustrated in (10a-b), repeated here as (32a-b). As illustrated in (32a-b), truncation is permitted in the 'regular' vocative, but disallowed with address inversion.
Returning to the TMT hypothesis that omission of morpholexical material is a function of the heaviness of the edge, I speculate that the minimal pair contrast in (32ab) may have a syntactic explanation. If the omission, or non-pronunciation, of phonetic material correlates with the expansion of the phasal template at its leftmost edge, then we would predict that constituents permitting truncation would be those with further expansion at their edge, since an expanded edge permits them to dispense with the descriptive content of their lexical interior (recall that salva veritate substitution is possible with vocatives, an illustration of which is that the addressee can be misnamed without the vocative's reference failing). Given that only Italo-Romance vocatives without address inversion admit truncation, we would expect these to involve extra structure (or movement thereto) in their internal build vis-à-vis vocatives with address inversion, consistent with our theoretical observations regarding the RF data. 28 By interpreting these divergences at the syntax-phonology interface in favour of the hypothesis that vocatives with and without address inversion may diverge in their internal syntax, my account finds commonality with previous theoretical work on address inversion vocatives (e.g. Akkuş & Hill 2017;Hill 2014Hill , 2017 in our shared contention that the answer to the puzzle of address inversion lies in microvariation that obtains in the extended functional structure of the nominal spine. Nonetheless, our respective studies have led us to theoretically divergent conclusions. Notably, I have argued that address inversion vocatives are fully-fledged VocPs in their own right, rather than DPs that merge within, and modify the head of, a larger VocP. Another key point of discrepancy is that, on my account, the kinship reversal observed in address inversion is crucially not implicated in the syntactic derivation of these vocatives (cf. e.g., the kinship reversal as motivation for the postulation of an uninterpretable feature [KIN]). Rather, the lexical 'flip' and its associated non-literal reading is interpreted as evidence in favour of the TMT thesis-and exoskeletal assumptions more generallythat the descriptive properties of the lexical noun have no involvement in the formal/referential specification of (here) VocPs.

Conclusions
This article has offered a topological mapping of the extended nominal left periphery, drawing predominantly on data from southern Italo-Romance address inversion vocatives. Using these data, coupled with conceptual motivation from the TMT, I have proposed that Italo-Romance vocatives with and without address inversion, which are basically alike in their referential potential, both involve the projection of a necessarily tripartite nominal structure (viz.  (2016) and Hill (2014:105ff) for syntactic treatments of truncation in vocatives-although note that, for Hill (2014), truncated vocatives are structurally reduced forms for which the designation 'vocative' is "misleading" (Hill 2014:106). height of the nominal edge beyond D (labelled here as VocP) is conceived as the locus where language-as-grammar 'makes contact' with the extensional world, and whose internal articulation serves to construe the ostensive-deictic possibilities of the objectreferring expression at the exophoric level. Dialectal evidence of Italo-Romance vocatives with minimal pairs at the syntax-phonology interface is nonetheless consistent with the hypothesis that address inversion vocatives are-or, more tentatively, can be-grammatically differentiated in their internal syntax, albeit at a finely granular level. That said, I have also urged caution in how these data should be interpreted theoretically.
By contrast, the ostensible curiosity of the lexical 'flip' is taken as further evidence for the crucial TMT thesis that the descriptive properties of the lexical noun play no role in the referential specification of (here) VocPs. As such, address inversion vocatives are assimilated to a general (sub)class of common nouns that undergo a shift from property-to-individual denotation as a (referential-interpretative) consequence of N-to-D substitution, where subsequent movement into Voc-whose externalization in the case of address inversion happens to show up in the lexical 'flip'-enables a nonargumental, non-literal interpretation of the lexical N. More broadly, I have taken the case of address inversion in southern Italian dialects as a springboard for demonstrating how topological mapping principles can be fruitfully applied beyond the DP to account for the properties of non-argumental constituents operative at the grammar-discourse interface in the nominal domain. In turn, my study underscores the feasibility and utility of expanding the TMT to offer a unified theory for capturing the syntax-semantics of natural language across structural environments.