3.1 Calls vs. addresses, selecting vs. non-selecting calls

As is well-known, the basic function of vocatives is to refer to the addressee. But there are two possible situations to consider in this respect: (i) the addressee has been established prior to the uttering of the vocative, or, alternatively, (ii) the vocative is itself used to establish the addressee. In this latter use, labeled call by Zwicky (1974), the vocative is employed to attract the attention of some individual and to invite that individual to participate in a further communication exchange (see d’Avis & Meibauer 2013: 192), see (9)a. Vocatives used as calls are an instantiation of the communicative act described by Schegloff (1968) under the label summons. When the addressee is already established, a vocative can be used “to maintain or emphasize the contact between speaker and addressee” (Zwicky 1974: 787), see (9)b. This use is labeled address by Zwicky.⁸

Schaden (2010) claims that the distinction between call and address is purely pragmatic, falling outside the scope of context-change-potential semantics, and considers more relevant for semantics a distinction between an identificational use, in which the vocative is used to single out an addressee from a bigger group of potential addressees, and two non-identificational uses: the activational use, which keeps the addressee active, and the predicational use, which assigns some property to the addressee. We have seen in (1) that the existence of a special type of identificational vocative is supported by a morphological contrast found in Romanian: in case the vocative is used to single out an addressee from a group, it must be marked as [+def] and cannot be marked as [+voc]. When there is no previous discourse, in what we may call out-of-the-blue calls, the use of a [+def –voc] form is possible, but not required. Compare (1) with (10):

This does not mean, however, that Zwicky’s call vs. address distinction does not have formal correlates – for instance, calls cannot occur in the middle of a sentence, addresses can (Zwicky 1974); for other differences, see Zwicky (1974), Hill (2014: 107–110, 204–205), Giorgi (2023). Therefore, we do not want to abandon this distinction, but prefer to consider identificational vocatives as a subtype of calls, which may be labelled selecting calls. A call that draws a person’s attention but is not used to select out of several possible addressees may be called non-selecting.⁹

3.2 Evaluative vocatives vs. addresses

Among vocatives that are not used as calls, i.e., when there is an established addressee, the Romanian data presented in (2) above and in section 5 below, as well as data from other languages, show that we need to distinguish between evaluative and non-evaluative vocatives (for the special formal properties of the evaluative type in other languages, see Corver 2008; d’Avis & Meibauer 2013; Gutzmann 2019). Non-evaluative addresses can contain proper names or introduce descriptive content, see (11), which led Schaden to distinguish between ‘activational’ (see John in (11)) and ‘predicational’ (see my friend in (11)). As in this article we do not analyze proper names, which do not show all the patterns in section 2 (there is no +/–def contrast), this distinction is not relevant. On the other hand, among those vocatives that qualify as ‘predicational’, there is a significant distinction between the non-evaluative type in (11) and the evaluative type in (12).

An important difference between (12) and (11) is that in (12) the nominal predicate assigned to the addressee is new information, it does not belong to the common ground: whereas in (11) the friendship relation between the speaker and the hearer is part of the common ground, in (12) the hearer is not presupposed to characterize himself as a fool. The fact that the predicate is foregrounded in the evaluative use becomes clearer when we consider the use of vocatives in isolation. A non-evaluative vocative as in (13)a is felt as incomplete, whereas an evaluative vocative is not and can constitute a speech act by itself, see (13)b:

This speech act comes close to an assertion (‘You are a fool!’) but cannot be negated in the same way as assertions (see Gutzmann 2019: 243) and should rather be treated as a type of exclamative (see Gutzmann 2019 and Welte 1980, who uses the term “exclamational vocative”). Note indeed that (13)b seems to be the same type of speech act as the one conveyed by the exclamatives in (14), where the subject is a third or first person (see (14)a and (14)b, respectively):

Some terms of address are similar to evaluatives in that they express the speaker’s feelings towards the addressee, e.g. dear friends, my dear, beloved guests. However, such expressions differ from evaluatives in that they would be felt as incomplete if used in isolation, just as my friends in (13)a, and the assignment of the NP-property to the addressee is not foregrounded. In Romanian, such expressions do not require Voc-marking (see (15)a, as opposed to the evaluative in (15)b, which requires Voc-marking), and in English they are not used in the Pronoun+NP construction of the type you fool (see (15)c, compared to (12)a, (13)b).

For these reasons, we do not include expressions of endearment such as my dear or beloved guests in the evaluative type.

Summing up, we adopt the classification in (16) in our treatment of vocatives, distinguishing in the class of vocatives that are not used to establish an addressee the sub-classes of evaluatives and non-evaluatives. For the latter, we will use Zwicky’s term ‘address’.

4.1 The data

Diagnostic contexts for the identificational use are situations in which there are several persons that can hear the speaker and the speaker uses the vocative to choose one or several of them (but not all) as addressee(s), see Schaden (2010). As we have seen in section 1, for Romanian identificational vocatives based on common nouns, out of the four options listed in section 2 ([–def –voc], [–def +voc], [+def –voc], [+def +voc], see ex. (3)), only the [+def, –voc] option can be chosen – i.e., the noun cannot be vocative-marked and cannot be bare.¹⁰ Here are other examples illustrating this contrast, which show that the facts remain the same if the noun is not modified ((1) is built with a modified noun):

The variants in (17)b–(18)b would be appropriate only if the intended addressee is already clear in the conversational situation – for instance, the teacher has approached the groups of boys or the brothers and is facing them while speaking.¹¹ In this case, the vocative is not used as an identificational, but as a non-selecting call.

Another important observation is that the verb of a subject relative clause inside an identificational vocative can appear in the 3^rd person.

This is impossible for the other types of vocatives, which require 2^nd person inside a subject relative clause embedded in the vocative phrase, see (20). The vocatives illustrated in (20) are not calls. As they are not evaluative either, they belong to the (non-evaluative) address type.

In addresses and non-selecting calls complex modification is less common than in identificationals, because it does not help to identify the addressee, as it does in examples such as (19). Relative clauses may nevertheless appear, if the information they introduce about the addressee is relevant for other purposes, as is the case in (20), where they provide a reason for the assertion in (20)a and for the advice in (20)b, respectively.¹²

The obligatory 2^nd person feature in subject relatives embedded in address vocatives is expected given the fact that with this type of vocatives the addressee is already established. The use of 3^rd person in the descriptive part of identificational vocatives (inside relative clauses) is not unexpected either, since the addressee is not yet established at the point when the sentence is uttered. In the following sub-section, we will propose a line of formal analysis for these intuitions.

The Romanian data presented so far point to the following generalizations:

One might be tempted to explain these generalizations by the fact that in the identificational use, the nominal describes a referent that is not already presupposed to be the addressee. The impossibility of specific addressee markers such as vocative morphology or the special bare use would follow from the fact that such marking presupposes that the nominal refers to an already established addressee. The fact that embedded subject relative clauses show 3^rd person in selecting calls (see (19)) but second person in addresses (see (20)) may receive a similar account. However, the generalizations in (21) only hold for selecting calls. In what we may call ‘out-of-the-blue calls’, where there is no common ground prior to the initiated conversation, [-def] and [+voc] vocatives can be used (see (22)–(23)), although the addressee is not already established:

For some nouns, the use of the [+def –voc] form in an out-of-the-blue call is even impossible – for instance, the [+def –voc] variant is not acceptable in the context in (22), see (24)a. This is not because chelner ‘waiter’ would disallow a vocative use of the [+def –voc] form: in a selecting call, the [+def –voc] form is allowed, see (24)b. A comparable contrast is illustrated in (25) for the plural copii ‘children’.

We conclude that the special construction described by (21) is restricted to selecting calls.

Note now that there is an exception to identificational vocatives not being marked as addressees. If the identification of the proposed addressee is achieved via deixis, the form that is used is not a demonstrative, but rather a 2^nd person pronoun, see (26)a vs b (the symbol ‘☞’ indicates a pointing gesture).¹³ Pointing is also used if the pronominal DP contains a relative clause that contributes to the identification, see (26)c:

This is one of the few situations when pointing is essential in establishing reference and can therefore be considered to contribute directly to the denotation of an expression. This also characterizes deixis found with demonstratives and what we call deictic items, more generally – deictic adverbs and deictic uses of some personal pronouns (those referring to humans: he, she, they, but not it, see Wolter 2006; Ahn 2019; 2022) – see (27), which shows that a demonstrative with a pointing gesture can be used to identify various referents (characterized by the same nominal property) via their location, while a definite DP, even accompanied by pointing, cannot do that:

Ahn (2019; 2022) proposes a semantics of deictic demonstratives in which pointing directly contributes to the denotation of the DP – it translates as a property, which places the referent at the pointed location. This property instantiates the second argument of the demonstrative D – in her analysis, demonstratives introduce a supremum operator, like the definite article, but differ from the definite article in that D takes two arguments, the NP and a second argument; this second argument may be an index, a relative clause or the property contributed by pointing; the operator in D yields the supremum of the intersection of the two properties denoted by the NP and the additional argument, respectively.¹⁴

In (26)a, the property that identifies the addressee is, obviously, not provided by the person feature of the pronoun, but by the pointing gesture and by the [+singular] and [+human] features of the pronoun (in (26)c, the property introduced by the relative is also included in the description of the intended addressee, but a sort of deictic gesture still seems to be needed). This means that the obligatory 2^nd person feature is a formal property that signals the vocative use of the DP. This feature is imposed on the DP by the configuration in which it occurs, a configuration that is responsible for the identificational vocative reading.

Modulo this (important) exception, identificational vocatives behave as regular definite DPs. For instance, DPs without an overt N but with overt descriptive material display the strong form of the definite article, cel (in Romanian, the definite article takes a ‘strong’, i.e. non-affixal form, in the context _[_NØ]):¹⁵

Other phrases that license an empty N in definite DPs, such as DP-initial ordinals, are also allowed:

The use of definite DPs in selecting calls is probably a more general property of languages with articles – see Espinal (2013), who analyzes this pattern as a subtype of what she calls ‘fake vocatives’.¹⁶ Here are examples of vocatives with the definite article, usually involving modified nouns, in languages which are claimed to disallow [+def] vocatives – English, Italian, Greek, German:

These examples seem to be instances of identificational vocatives, as is clear at least for (30)g. This strongly suggests that the existence of a special structure for identificational vocatives, involving a definite DP, is not a mere peculiarity of Romanian. Note however that these examples involve modification, and it may be the case that modification is required for [+def] identificationals in these languages (as reported for Greek by Anastasia Giannakidou, p. c.), which is not the case in Romanian (see the unmodified [+def –voc] nouns in (17)a, (18)a, (24)b, (25)b).

4.2 An analysis of identificational vocatives

The syntactic approach to vocatives in recent generative studies characterizes the vocative use of nominals by postulating special syntactic positions, introduced via functional categories. A special position on the clausal spine, above Rizzi’s (1997) ForceP, is proposed by Moro (2003), who calls the relevant head Voc⁰ and places vocative nominals in SpecVocP. Zanuttini (2004) and Portner (2007) label this projection AddrP (the ‘addressee phrase’), see (31):

Hill (2007; 2014) argues that vocative nominals are also characterized by a nominal functional head above D, initially labeled Role⁰ (Hill 2007), then relabeled Voc⁰ (Haegeman & Hill 2013; Hill 2014; 2022; see also Stavrou 2014, who uses the label VoP), see (32):

Adopting Speas & Tenny’s (2003) proposal of a Speech Act functional layer above ForceP, Hill (2007; 2014), followed by Haegeman & Hill (2013) and Stavrou (2014), places VocP in the specifier of the lowest projection in the speech act domain, labeled SAP in Haegeman & Hill (2013), SAhP in Hill (2014) (from ‘Speech Act Hearer-Phrase’, see (33)) and PR_addressee in Stavrou (2014):

A VocP on top of DP in vocative nominals is also assumed in Espinal (2013), Roehrs & Julien (2014), Julien (2016), and Gutzmann (2019). Espinal (2013) does not use a dedicated sentential position for VocPs, treating them as specifiers of CP or parentheticals.

The evidence for assuming both a nominal projection (VocP) and a special sentential position (such as SpecSAP) for the syntax of vocatives comes from distributional facts which mainly involve word order with respect to vocative particles and speech act particles (see Haegeman & Hill 2013; Hill 2014). Giorgi (2023), who argues that vocatives in general express separate speech acts, proposes different configurations for addresses and calls: addresses, which can occupy parenthetical positions, are in the specifier of a prosody-oriented head projected above the left-periphery, which assigns the comma intonation (sentence-internal positions are derived via movement); calls are analyzed as sentence-external, situated in the specifier of a vocative discourse projection which is integrated with the following sentences only at a level higher than sentences, called micro-discourse.

In our analysis, we adopt the proposal of a VocP in the extended nominal projection, leaving open the issue of how VocP is integrated in the sentence (i.e., whether a dedicated projection is necessary or not). There are two reasons for which the data we discuss favor an analysis involving a nominal VocP, as in (32), rather than just placing the DP in the specifier of a sentential projection, as, for instance, in (31).

First, we have seen that identificational vocatives must be marked 2^nd person when pronominal (see (26)) and the types that cannot function as identificationals (i.e., [–def –voc] and [+/–def +voc]) require 2^nd person in embedded subject relative clauses, see (20). This indicates that there must be a source for the 2^nd person feature specific to vocatives, in particular for identificationals, where this feature plays no role in the description of the intended addressee (see the discussion around (26)). Hill (2014) proposes that Voc is the head that introduces a valued 2^nd Person feature in the derivation.

Secondly, an interesting contrast involving coordination supports the proposal that vocative-marked nouns in Romanian (the types [–def +voc] and [+def +voc]) raise to Voc⁰ (as proposed by Espinal 2013 and Hill 2014). As shown in (34), marked vocatives differ from unmarked non-definite vocatives in being infelicitous in coordination¹⁷ (this contrast was observed, but not analyzed, in Croitor & Hill 2013):

Under the assumption that morphologically marked vocatives raise to Voc⁰, the contrast can be explained as follows: in (34)b,d, since each vocative marker corresponds to a distinct Voc⁰, we have a coordination of VocPs. If the semantic contribution of Voc⁰ is to (exhaustively) specify the addressee of a speech act, a coordination of VocPs should be ruled out, because it would result in contradictory specification of the addressee. One may wonder why a group interpretation is not available. Note that coordinations of referential expressions (e.g. John and the girl) or NPs (e.g. every [boy and girl]) typically yield group readings. Unmarked vocatives, both bare and definite (see (34)a,c), do allow an interpretation by group formation. The unavailability of this interpretation in (34)b,d indicates that the coordinated constituents cannot be analyzed as mere NPs or DPs. We propose that they are VocPs.¹⁸ In the analysis in (31), there is no way of representing the contrast between coordinations of marked and unmarked nominals: in both cases, we would have a DP coordination in SpecAddrP. In order to account for (34)b,d, one would have to assume that using vocative morphology forces the projection of several AddrPs, but it is not clear why this should be so. One could assume that vocative-marked nouns raise to addr⁰, but movement from a specifier to a head position is not accepted in the current generative syntactic framework.¹⁹

Note now that coordinated [–def] vocatives are allowed, see (34)a,c. This leads to the conclusion that N⁰ does not raise to Voc⁰ in this type (contra Espinal 2013; Hill 2014). The structure of these examples involves a single Voc⁰ and a coordination of NPs or DPs with a null D in its complement:

The existence of a D in [–def] vocatives is debated: some studies propose that a D must be projected (Szabolcsi 1994; Espinal 2013; Coene et al. 2019), while others allow for Voc⁰ to directly select NP (Hill 2014; Stavrou 2014). For Romanian, we do not know of any compelling evidence in favor of a null D in this type.²⁰ Therefore, we will use the [Voc NP] configuration for simplicity, but our analysis can be made compatible with the other alternative with minimal modifications, as we will indicate when necessary.

The proposal that vocative types in Romanian can be distinguished at the level of a nominal functional projection VocP does not exclude differences in the way VocP is integrated into larger structures. In particular, we are sympathetic with Giorgi’s (2023) view that calls are sentence-external, involving separate discourse units. But the distinction between selecting calls, which use the identificational structure, and non-selecting calls, is done at the level of the VocP. For addresses, our analysis is compatible both with the hypothesis of a dedicated position in the Speech Act domain (Hill 2007; 2014) and with the hypothesis of a specifier position of a prosody-oriented head not specialized for addressees (Giorgi 2023).

Summarizing, the general assumptions about the structure of vocatives that appear to be best fit for describing the facts of Romanian are the following:²¹

Under these assumptions, the formal differences between identificational vocatives and the other types, presented in the previous sub-section (4.1), indicate a difference at the level of Voc. We propose the following:

The obligatory referentiality and definiteness observed for identificational vocatives can thus be viewed as being due to the fact that Voc_ident introduces the identification of the Addressee. In order to explain the limitation of Voc_ident to selecting calls and to clarify the notion of context involved in (37)b, we follow the suggestion of an anonymous reviewer to represent the potential addressees from which Voc_ident selects the addressee by using the notion of ratified recipient (see Goffman 1981; McCawley 1999). Ratified recipients include all the accepted participants to the conversation, who can hear the message and are accepted by the speaker as recipients of the message, but are not necessarily addressees; they include the addressee but also third parties. In the context in (17), all the choir members are ratified recipients; likewise, in the class contexts in (18) and (19), all the students in the class are ratified recipients. McCawley (1999: 605) characterizes the shared knowledge involved in the notion of common ground as consisting ‘of the propositions that the ratified participants in a discourse jointly take as already established at the given point in the discourse and take for granted in the immediately following discourse’. Using a notion of context parameterized to the set of all the ratified participants allows a conversational move by which the Addressee is changed, as in (37)b. We conclude that it is this notion of context that must be assumed for c and c′ in (37)b. For the restriction of Voc_ident to selecting calls, we propose that Voc_ident introduces the presupposition that the proposed Addressee is properly included in a set of Ratified Recipients – see (38), which is formulated so as to cover plural vocatives as well as singulars; RR(c) notates the property of being a ratified recipient in the context c, Σz:RR(c)(z) is the maximal sum of the ratified recipients in the context c, ‘<’ notates the proper part relation.

In out-of-the-blue calls, there is no such previously established set from which the vocative selects an addressee. Hence the impossibility of Voc_ident.

On the hypothesis that in [–def] vocatives Voc⁰ directly selects NP, (37) excludes [–def] vocatives (because Voc_ident selects DPs).²²

In order to explain the fact that [+def +voc] vocatives are excluded in the identificational construction, we propose that Voc_ident does not have the attracting feature responsible for Voc-marking (see (36)(ii) and the coordination facts in (34)). In the following, we will use the label Voc* for the Voc that triggers movement. Using the subscripts N and D for the Voc that select NP or DP, respectively, we conclude that addresses and non-selecting calls can use the heads Voc_N*, Voc_N Voc_D* and Voc_D, all distinct from Voc_ident. The choice between these heads is strongly influenced by the nominal lexeme (see (22)–(25) above and also 4.4 below), a matter which lies outside the scope of this article.

An alternative to using Voc_ident for identificational vocatives would be to directly merge the definite DP in the projection of a discourse head that characterizes calls – a special head of this type was proposed by Giorgi (2023), who calls it Dis-Voc and places the vocative in its specifier (in her system, the complement of a discourse head is the following discourse projection). But we would still need to assume a dedicated head for selecting calls, because, as we have seen above, not all calls are selecting (out-of-the-blue calls do not use the identificational structure, see (21)–(25)): for instance, a Dis-Voc_ident with a DP-specifier, distinct from a Dis-Voc head used for non-selecting calls, with a VocP-specifier. Moreover, Dis-Voc_ident will have to be endowed with a valued 2_nd person feature, responsible for the Person agreement facts illustrated in (26) above and discussed in the following sub-section. As we do not wish to take a stance on the external syntax of vocatives, we will consider Voc_ident as part of the nominal functional structure.

4.3 The Person feature in identificational and non-identificational vocatives

Let us now see how this analysis can account for the facts concerning Person noticed in section 4.1. Let us recall the generalizations, starting with identificationals:

As we have already mentioned, (39)b is evidence for the fact that 2^nd person may come from Voc: if the interpretive contribution of 2^nd person is the presupposition that the referent of the constituent it attaches to is the addressee (see Heim & Kratzer 1998; Kratzer 2009), we must conclude that the 2^nd person on the DP of identificationals cannot be interpreted, because the fact that the referent of the DP is the addressee is not yet established at that point of the discourse, but is achieved by the context update operated by Voc_ident. We conclude that the person feature comes from Voc⁰, via upward agreement.²³ The necessary use of the 2^nd person (see (26)b) is due to the fact that pronouns always carry a Person feature. In the particular case of identificational vocatives, this feature is uninterpretable and receives a value from Voc_ident. This φ-feature matching requirement can be seen as a consequence of the fact that Voc belongs to the same extended nominal projection as the pronoun’s D. Independent evidence for the existence of uninterpretable person features on pronouns, which are valued via Agree, comes from the so-called ‘fake indexicals’ discussed in Kratzer (1998; 2009).

Note now that in Kratzer’s proposal, the Person feature is interpretable on the member of the agreement chain where it is inserted as valued. This is prima facie at variance with our proposal, which assumes valued Person on Voc_ident.²⁴ A possible solution is to decompose Voc_ident into several features, among which Person, which is what Kratzer (2009) proposes, in fact, for pronouns. Note that the semantics in (38) makes reference to the Addressee, but Voc_ident, in addition, updates the context so that the Addressee of the updated context is identified with the referent of the complement DP. These two meaning components can be represented as two features that are bundled together in Voc_ident:

Ident-Update updates the context to a context in which an identity relation holds between the referent of its first argument (the person feature inside Voc_ident) and the referent of the second argument (the DP), and introduces the presupposition on the second line in (38). The 2^nd person feature introduces an entity that refers to the Addressee.

If the DP is not pronominal, an embedded subject relative may have a 3^rd person verb (see (39)a), which means that the person feature is not necessarily transmitted to the complement of Voc_ident. We propose that in this case D, represented by the definite article, lacks a Person feature altogether.²⁵

Let us now observe that subject relatives embedded in identificational vocatives may also show 2^nd person on the verb:

As we do not see any meaning difference between the two variants in (41), we propose that the difference relies in formal features: the definite D may come with unvalued Person, which must then be shared with the rest of the nominal spine, or can lack Person altogether:

Moving on to the other types of vocatives, we have seen in section 4.1 that embedded subject relatives must show 2^nd person (see ex. (20), resumed in (44) below):

An explanation based on agreement with Voc⁰ may carry over to these cases, in particular for marked vocatives, which have been shown to involve N-(to-D)-to-Voc. It is reasonable to assume that raising N to Voc⁰ requires sharing all the φ-features between Voc⁰ and N. Since Voc⁰ bears 2^nd person (see (31)(i)), feature sharing results in N having 2^nd person. This feature must be matched by the subject of the relative clause (via agreement or any other mechanism involved in matching phenomena in relative clauses, see Cinque 2020). Because in identificational vocatives there is no raising to Voc⁰, there is no requirement that N should share all Voc⁰’s features, including Person.

For unmarked non-definite vocatives ([–def –voc]), on the hypothesis that there is a null D⁰, it can be assumed that this D⁰, which is selected by Voc_N, carries 2^nd Person; on the hypothesis that there is no D-layer, we may assume that Voc_N acts like D with respect to φ-features, requiring full feature sharing with N.

But there is also a possible semantic explanation for (43). Note that all the examples of relatives in these types involve addresses. We have not been able to find or imagine examples of relatives embedded in non-selecting calls (see footnote 12). In addresses, the addressee is already established. We may then assume that these types involve an interpretable 2^nd person on D or N. Admittedly, this explanation must still be supplemented with a formal component: the Voc of these types must have a selectional feature requiring a complement bearing Person, because, as we have seen in (39)a/(41)/(42)b, D in Romanian does not necessarily bear Person.

4.4 Other uses of unmarked definite vocatives

While Voc_ident only allows [+def –voc] vocatives, the reverse is not true: [+def –voc] vocatives may also occur in non-selecting calls and addresses, but in that case, they are subject to strong lexical restrictions.

Examples of [+def –voc] in non-selecting calls are given in (45).

This means that the [+def –voc] pattern is not restricted to Voc_ident. But when it is used with a non-identificational Voc⁰, it is subject to strong lexical restrictions. We have seen in (24)a that profession nouns such as chelneru’ ‘the waiter’ are not felicitous. Even the feminine counterpart of the noun in (45)a, fată ‘girl’, does not occur in the [+def –voc] form:

Such a lexically restricted use is also found for [+def –voc] in addresses. The existence of stronger lexical restrictions on addresses than on calls was noticed as early as Zwicky (1974: 790–791) and is expected since in the case of addresses the content of the noun phrase does not play any role in the identification of the addressee, which is already established. Portner (2007) proposes that the meaning contribution of addresses consists, at least in part, in describing the relationship between speaker and addressee (see the use of nominals such as dear friend, my lord, honey). This proposal is supported by Romanian, where kinship terms and relational nouns used in addresses normally take the speaker as their implicit argument (e.g. prietene ‘friend.voc’ = ‘my friend’, soră ‘sister’ etc.)²⁶: under the assumption that the noun is used to describe the relationship between Speaker and Addressee, if the noun is inherently relational, its implicit argument will have to be set to the Speaker. This type of interpretation may be encoded in the i(nter)-p(ersonal) feature assumed by Hill (2014; 2022) for Voc heads. [+def –voc] vocatives are allowed with relational nouns and kinship terms only when they have an explicit possessor, see (47)a and (48)a (the sentence-internal placement indicates that these vocatives are not calls). In the b examples, we show that these nouns do allow a [+def –voc] version without a possessor, but only in selecting calls (identificationals). (48)b also shows that identificational [+def –voc] vocatives do not have to take the speaker as the implicit argument.

Within the analysis proposed in 4.2, the instances of [+def –voc] in addresses (see fraţii mei ‘brothers-the my’ in (47)a) and non-selecting calls (see (45)) should be accounted for by a Voc head different from Voc_ident. We conclude that, among the Voc heads that are not identificational, there is a head Voc_D that takes a definite DP without triggering movement, besides the movement-triggering Voc* (for marked vocatives) and Voc_N (for [–def +voc] vocatives).

4.5 On the type of DP used with Voc_ident

So far, we have only considered definite DPs as identificational vocatives. English and German also allow DPs with universal quantifiers (see Portner 2008; Schaden 2010; Gutzmann 2019: 179–180); Schaden (2010) provides the English example (49)a, and Gutzmann (2019: 179) provides its German equivalent in (49)b:

For these examples, Portner (2007: 411), followed by Gutzmann (2019), assumes that the universally quantified DP is shifted to type e, referring to an entire group. In Romanian, quantified DPs with singular number (with fiecare ‘each’, oricine free-choice ‘any’, ‘whoever’), which are obligatorily interpreted distributively, are disallowed. Toţi ‘all’ is allowed, see (50), but this quantifier combines with definite DPs and arguably yields a plurality-referring expression, its semantic contribution being the removal of the exceptions to the maximality condition that plural definites typically allow (see Brisson 1998; Lasersohn 1999; Križ 2016 for details).

The pattern illustrated in (51), which is also found in English (see (52)), is only an apparent exception to the definiteness of the DP. First, notice that it is restricted to subjects of imperatives, see (51) (the same holds for English according to Portner 2007). Moreover, as Portner (2007) notices, the addressee is the entire group and the requirement is that some members of the group should undertake the desired action. This means that, unlike in selecting calls, no actual change in the Addressee takes place. We believe, with Portner, that these DPs are subjects of imperatives or peripheral DPs that bind the subject of imperatives, rather than vocatives.