Vocatives as parenthetical adjuncts: Evidence from Arabic

Vocatives are noun phrases that are generally used as calls to attract an addressee’s attention or as addresses to maintain contact with her/him. In her 2014 book, Virginia Hill analyzes vocatives as forms of address; they are converted to syntax as nominal functional projections in the form of vocative phrases. Drawing on work by Speas and Tenny (2003) and Moro (2003), she further posits that vocatives merge as arguments of a speech act phrase in the left periphery. This merging site, Hill 2014 and others (e.g., Slocum 2016) argue, is fixed. This is so despite the fact that vocatives may appear in different positions in their host clause. Under this account, pre-vocative elements appear in that location as a result of movement. This paper presents counterevidence to this argument from Arabic. Using naturally occurring data, mainly from Twitter, the paper shows that Arabic vocatives do not merge in a fixed position in the left periphery. Rather, they are parenthetical adjuncts whose relationship with their host clause is minimally constrained. If correct, the analysis serves as an invitation for a closer and more data-driven look at vocatives in other languages, including those in which vocatives have already been explored.

phrases may stand alone, as the tweets in (1) and (2) illustrate. Alternatively, they may be part of a host clause, in which case they may appear sentence-initially, (3), sentencefinally, (4), or sentence-medially, (5  -The paper relies predominantly on tweets. Most of the Arabic tweets seem to come from the following Arabic varieties: Gulf, Levantine, and Egyptian. Some of the tweets tend to be long. When this is the case, I only include the portion that is relevant to the topic of discussion. I also include the Arabic tweets without correction. I include punctuation but not emojis. The URL of each tweet is provided for readers who would like to examine the whole text. These tweets were retrieved between November 10, 2019 and July 10, 2020. examples include mid-sentential vocatives. Unlike sentence-initial vocatives that are mainly employed to grab the hearer's attention, mid-sentential vocatives are addresses that carry more evaluative and relational information (see Slocum 2016: 12) and tend to be either endearing or disparaging. Evaluative vocatives of this type in Arabic are often only licensed with ya:. For example, terms like kalb 'dog' and ħma:r 'jackass' may not be used as vocatives without ya:. Most work on vocatives, including Arabic vocatives, focuses on their pragmatic functions, their distribution, and/or their internal syntax. As to the external syntax of vocatives, the literature falls into two camps. There are those like Levinson (1983) who consider vocatives as elements that are set apart from their host and are only inserted parenthetically. Espinal (1991) belongs to this camp; she groups vocatives with other parentheticals and analyzes them collectively as disjuncts that merge in a separate plain in a three-dimensional structure; see Section 5. On the other side of the fence, there are researchers who analyze vocatives as syntactically incorporated in the host construction; see, for example, Ashdowne (2002), Hill (2014), and Slocum (2016). Using data from Latin, Ashdowne (2002) argues that vocatives are not parentheticals. To him, parentheticals, but not (Latin or English) vocatives, are structurally minimally constrained. 2 (6) Parentheticals vs. vocatives (Ashdowne 2002: 154-155) a. Parentheticals may interrupt their host structure freely. Vocatives are more constrained; they do not have "complete freedom of placement." b. Any connection between parentheticals and their host is optional.
Vocatives, however, "do have some necessary connection with the accompanying utterance through something in the discourse context, viz. they must refer to the addressee(s) and are unacceptable if they do not." 2 c. There is no limit on the number of parentheticals that may be present in a construction; the number of vocatives, on the other hand, is not so unlimited.
Hill (2014) agrees with Ashdowne's conclusion. She provides an account that details how vocatives merge in a fixed position in the left periphery. Slocum (2016) makes a similar argument. She uses Ashdowne's criteria to argue against a parenthetical account of vocatives. She builds on Hill (2007; to argue for a syntactic account of vocatives as elements that are structurally incorporated in the left periphery of their host. In the rest of this paper, I use Ashdowne's three criteria in (6) to argue against a left-periphery account of vocatives, focusing mainly on Arabic. I present evidence, mostly from Twitter, to show that Arabic vocatives are best analyzed as parenthetical adjuncts. If the analysis is correct, it serves as an invitation for a closer and more data-driven examination of vocatives in other languages, including those in which vocatives have already been explored. Much of the work on vocatives in recent years, along with the theoretical conclusions presented in that work, relied on constructed examples and/or grammaticality judgment and elicitation. A closer look at vocatives in naturally-occurring data is bound to get researchers to rethink some of their conclusions, as this paper hopes to show. The rest of the paper is organized as follows. Sections 2, 3, and 4 address points (6a, b, & c) respectively. Together these sections show that Arabic vocatives exhibit behavior that is typical of parentheticals and thus they should be analyzed as such. Section 2 summarizes two accounts that argue that vocatives merge in a fixed location in the left periphery while other sentential elements move around them: Hill (2014) and Slocum (2016). The section provides counterevidence to these arguments; it concludes that vocatives merge in the locations where they are pronounced and that they have freedom of placement. Section 3 deals with the topic of connection. It approaches the topic from semantic and syntactic perspectives and shows that there is no necessary connection between vocatives and their host clause. It also examines the pragmatic restriction that vocatives must refer to the addressee(s) of their host clause and shows that vocatives may refer, not only to addressees, but also to unaddressed overhearers. Section 4 examines whether the number of vocatives in a given structure is limited. It shows that multiple vocatives are allowed in a host clause, both consecutively and intermittently. Section 5 briefly presents a possible derivation of vocatives as parenthetical adjuncts. Section 6 offers a few final remarks. Ashdowne (2002) states that parentheticals but not vocatives have "complete freedom of placement." That is, parentheticals like as far as I know in the English and Arabic examples in (7) through (9) may freely merge with the host clause sentence-initially, sentencemedially, or sentence-finally. Vocatives, however, are more restricted in terms of where they may merge.

Freedom of placement
(7) a. Proponents of Ashdowne's view argue that vocatives may only merge in the left periphery. Sentence-final and sentence-medial positions as in (4) and (5) are not due to the freedom of placement of the vocative. Rather, they are due to the movement of pre-vocative material to that position. In order to show that vocatives are indeed parentheticals and thus have freedom of placement, we need to show that they may appear in locations out of which movement is syntactically prohibited. This is the purpose of this section.
Two studies that argue that vocatives merge in a fixed location in the clause hierarchy are Hill (2014) and Slocum (2016). Hill (2014) analyzes vocatives as forms of address converted to syntax as nominal functional projections in the form of vocative phrases, (10). Inspired by Speas & Tenny (2003) and their revival and formalization of Ross's (1970) Performative Hypothesis, Hill (2014) argues that vocatives merge as arguments of a speech act phrase (SAP) in the left periphery. As (11) illustrates, SAP is a doublelayered projection similar to the verbal projection vP/VP; it comprises a higher segment, SAsP, identified as the speaker's field, and a lower segment, sahP>SAhP, identified as the hearer's field. The hearer is considered the goal of the speech act, while the speaker and the force phrase (ForceP) are its agent and theme respectively. Vocative phrases merge in the hearer's field. Crucially, for Hill, Vocatives are realized above CP. As (11) shows, they are part of a speech act projection that takes ForceP as an argument. To her, mid-sentential vocatives as in (5), repeated here as (12), are the outcome of movement; the pre-vocative material starts out in ForceP before it moves to Spec,sahP, as (13)  Note that movement in (13) takes the form of remnant movement. This type of movement is especially necessary if the syntactic string that ends up in a pre-vocative position is a nonconstituent. In (12), w-alla: ma: la-k bi-ʔalb-i: 'I swear that you do not have in my heart' is not a constituent. This is why, ʃari:k 'partner' moves first, allowing phrasal remnant movement of the entire ForceP to follow. Under Hill's (2014) account, movement targets the two projections of sahP. Slocum (2016) proposes a similar, though not identical, approach. She draws on intuition by Taglicht (1984) that vocatives may be used to mark the boundary between old information and new information in a sentence, and she formalizes this intuition as (14). Using Rizzi's (1997) split CP, she argues that vocative phrases merge in the left periphery above the focus phrase (FocP) but below the higher topic phrase (TopP).
(14) [ForceP [TopP [VocP [FocP [TopP [FinP [… Under Slocum's (2016) approach, mid-sentential vocatives are the outcome of topicalization. Any material that appears before a vocative phrase does so as a result of movement to TopP. In other words, (12) has the derivation in (15). The vocative starts out in the left periphery. The host clause merges below FinP before the pre-vocative part undergoes topicalization. Topicatization takes the form of remnant movement, allowing a nonconstituent to move to a pre-vocative postion. Thus, 'partner' moves to FocP before remnant movement to TopP takes place.
The following subsection presents data that challenge Hill's (2014) and Slocum's (2016) accounts of the external syntax of vocatives.

Vocatives and syntactic island
If vocatives always merge in a fixed location in the left periphery, as maintained in Hill (2014) and Slocum (2016), and if any material that precedes a vocative phrase moves to that location from a post-vocative position, then we should expect sentences in which vocatives interrupt a syntactic island to be ungrammatical. Slocum (2016) tests this prediction on English. She ran an online survey whose goal was to elicit grammaticality judgments on mid-sentential vocatives that interrupted syntactic islands versus the same type of vocatives outside of syntactic islands. The survey yielded results from 128 participants. Slocum reports that sentences with vocatives that interrupt islands, such as those in (16), were rated significantly worse than their counterparts with mid-sentential vocatives outside of an island in (17) (2016: 117).

(16)
English (Slocum 2016: 118, (Ross 1967). This is why, for example, the question in (19) is ungrammatical, and this is why, according to Slocum (2016), the sentence is judged as unacceptable or degraded by native speakers.
*What k did the farm have a goat and ________ k ?
Arabic vocatives defy the expectations presented in Hill (2014) and Slocum (2016) as they may interrupt islands freely. They have what Ashdowne calls freedom of placement. As such, they present a serious challenge for any account that argues that vocatives merge in a fixed location and that other sentential elements move around them. We start with Arabic vocatives in subject islands. Movement out of subject islands is illicit in Arabic, just as it would be in English, as (20) and (21) illustrate. The questions in (b) are ungrammatical under the designated reading.
for-whom k the-shirt that bought-it Ziad ________ k was expensive 'For whom k was the shirt that Ziad bought ________ k expensive?' (21) a. l-qara:ra:t llati: ttuxiðat ʔams / fi: ʕamma:n fi: rafʕ the-decisions that were.taken yesterday / in Amman in raising baʕdˤ ʔasʕa:r l-silaʕ wa-l-xadama:t ʔaθθarat ʕala l-muwa:tˤin. some prices the goods and-the-services affected on the-citizen 'The decisions that were taken yesterday/in Amman about raising the prices of some goods and services had an effect on citizens.' b. *mata:/ʔayn k ʔaθθarat l-qara:ra:t llati: ttuxiðat ________ k when/where k affected the-decisions that were.taken ________ k fi: rafʕ baʕdˤ ʔasʕa:r l-silaʕ wa-l-xadama:t ʕala in raising some prices the goods and-the-services on l-muwa:tˤin? the-citizen 'When/Where k did the decisions that were taken ________ k about raising the prices of some goods and services have an effect on citizens?' Although extraction out of subject islands is not allowed, the tweets in (22) and (23) show that vocatives may interrupt such islands (between square brackets) without leading to ungrammaticality. The islands in these examples also qualify as complex NP islands -or DP islands, as Adger (2003) prefers to call them.

(22)
[l-qara:ra:t llati: ttuxiðat ya: dawlat l-raʔi:s fi: rafʕ baʕdˤ the-decisions that were.taken voc Mr. Prime Minister in raising some ʔasʕa:r l-silaʕ wa-l-xadama:t] ʔaθθarat ʕala l-muwa:tˤin … prices the goods and-the-services affected on the-citizen 'The decisions that were taken, Mr. Prime Minister, about raising the prices of some goods and services had an effect on citizens.' Arabic vocatives may also interrupt adjunct islands, as (25) through (27) illustrate. This is so despite the fact that movement out of an adjunct island leads to ungrammaticality, [ʔabl ma: tʔu:l ya: ʕabʔari: ʔidda:m kamera l-ʔaxba:r ʔinno before that you.say voc genius in.front.of camera the-news that fi: musa:wa:t be:n l-marʔa wa-l-raʒul bi-l-dustu:r there equality between the-woman and-the-man in-the-constitution bi-ʔiʕtˤa:ʔ l-ʒinsiyye] ru:ħ tˤa:lib bi-ħaʔ-ha: tiftaħ ħsa:b in-granting the-citizenship go demand of-right-her open account banki: li-ʔawla:d-ha: bank for-children-her 'Before you say, genius, in front of the news camera that there is equality between women and men with regard to granting citizenship, go demand that she be granted the right to open a bank account for her children. Similarly, vocatives may interrupt coordinate structure islands, (29) through (32), although violations of the Coordinate Structure Constraint (Ross 1967) in Arabic, just as in English, lead to ungrammaticality, (28). 3 (28) a. layla: bitħibb l-sbe:ħa w-sabaʔ l-sayya:ra:t. Leila like the-swimming and-race the-cars 'Leila likes swimming and car racing.' b. *ʃu: k bitħibb layla: ________ k w-sabaʔ l-sayya:ra:t? what k like Leila ________ k and-race the-cars 'What k does Leila like ________ k and car racing?'

‫ﺍﻟﻘﺮﺍﺭﺍﺕ‬
ʔuħibb [l-ʕarab ya: ʒama:ʕa wa-ʔaqwa:la-hum] l-ħikma taʒri: I.love the-Arabs voc folks and-sayings-their the-wisdom run fi: ʕuru:qi-him ʒari:. in veins their running 'I love Arabs, folks, and their sayings. Wisdom certainly runs in their vein.'  (32) are "all expected to be good under a gapping analysis: the coordination concerns two clauses, the second one having verb ellipsis … The examples should have been of the type: 'John i , and you idiot k , Mary j are those who get married.'" I was not able to find a tweet that is similar to the reviewer's example. However, the following constructed example is readily available and judged as acceptable by native speakers.
(33) a. ziya:d nisi: le:ʃ layla: se:farit ʕa-ʔitˤa:lya:. Ziad forgot why Leila travelled to-Italy 'Ziad forgot why Leila travelled to Italy.' b. *la-we:n k ziya:d nisi: le:ʃ layla: se:farit ________ k ? to-where k Ziad forgot why Leila travelled ________ k 'Where k did Ziad forget why Leila travelled ________ k ?' It is important to note that not only do vocatives interrupt different types of islands without leading to ungrammaticality, but they also have a great amount of flexibility as to where inside the island they may appear. Take (34), for example; the vocative in this tweet is adjacent to the wh-word. This is not the only available location, however. Other locations inside the wh-island are readily available for the vocative, as (38) illustrates. 4 (38) mumkin ʔaʕrif [ki:f (ya: ma:ma:) txarraʒti: (ya: ma:ma:) possible I.know how (voc mom) you.graduated (voc mom) w-sˤurti: (ya: ma:ma:) mudarrise (ya: ma:ma:) maʕ ha-l-kutub and-became (voc mom) teacher (voc mom) with these-the-books l-maʕu:qa] the-handicapped 'May I know how, (Mom), you graduated, (Mom), and became, (Mom), a teacher, (Mom), with such handicapped books.' Examples like the ones presented in this section pose a serious challenge for accounts that argue for a fixed merging location for vocative phrases. This is so because their main premise is that vocative phrases always merge in a sentence-initial position and that any pre-vocative material moves to that location. As we mentioned earlier, an important prediction follows: it should be illicit for vocatives to interrupt islands. This section has presented naturally-occurring data from Arabic to show that vocatives may in fact occur inside different types of syntactic islands without inducing any violation. Consider (27), repeated here as (39). Under the accounts examined in this section, the derivation of this sentence would involve remnant movement of 'I will not leave before you complete t j .' As (40) demonstrates, this step is preceded by the movement of 'your exams' out of an adjunct island. We know that this type of movement is illicit and, therefore, this sentence is expected, contrary to facts, to be ungrammatical.  (41) and (42) show that vocatives may interrupt different types of islands in English and Spanish respectively. These include subject islands, (a), adjunct islands, (b), and wh-islands,(c). Recall that Slocum (2016) ran a survey to test the grammaticality of mid-sentential vocatives that interrupted syntactic islands and found they were judged as degraded. The examples in (41) and (42) cast doubt on the results of this survey. One reason Slocum's survey yielded unfavorable results may be the fact that all sentences were presented out of context, as Slocum herself points out. This may have been compounded by the fact that the vocative used in all the sentences was the proper noun Jason. While a proper noun may be used as a sentence-medial address and may carry some emotional load, it is unlikely to be interpreted as such in an out-of-context statement. I found that titles that emphasize hierarchy, such as Mr. President, as well as terms of endearment and insult, fare much better. The next subsection presents additional challenges to the account in Slocum (2016).

Vocatives and topicalization
An account that argues for a fixed merging site for vocatives faces additional challenges, especially if the argument assumes that pre-vocative material is the outcome of a special type of movement: topicatization. This is the case with Slocum's (2016) account. As we saw at the beginning of this section, according to Slocum, VocP merges above FocP in the left periphery, (43). Any material that appears in a pre-vocative position does so as a result of movement to TopP.
(43) [ForceP [TopP [VocP [FocP [TopP [FinP [… If this is correct, two predictions follow. First, we should expect wh-questions, in which a vocative phrase follows a wh-word, to be ungrammatical. This is so because, under standard assumptions, wh-words are located in FocP. This prediction is not borne out. Arabic data show that questions with vocatives to the right of a wh-word are readily acceptable, as (44) through (49) show. Note that (44) starts with w-ʔinti 'and you' as topic, followed with ʔe:mta 'when' as focus. Under Slocum's analysis, the vocative should appear between the topic and the focus, which is also possible in Arabic, but not after the focus.  (50). Examples (b) and (c) show that a direct object -'Kareem' in (b) and 'Leila' in (c) -may be focused. However, once a wh-word is involved as in (d), neither 'Kareem' nor 'Leila' may be focused. As (e) and (f) demonstrate, a fronted 'Kareem' or 'Leila' would only be grammatical if it is associated with a resumptive pronoun. A resumptive pronoun makes the fronted element a topic and not a focus. Assuming that wh-words occupy FocP, (e) and (f) show that Arabic licenses only one FocP. 5 (50) a. ʔamarit kari:m ye:xod layla: ʕa-l-madrase. I.ordered Kareem take Leila to-the-school 'I ordered Kareem to take Leila to school.' b. kari:m ʔamarit ye:xod layla: ʕa-l-madrase, miʃ sa:mi:. Kareem I.ordered take Leila to-the-school not Sami 'Kareem, I ordered to take Leila to school, not Sami.' c. layla: ʔamarit kari:m ye:xod ʕa-l-madrase, miʃ ziya:d.
Leila I.ordered Kareem take to-the-school not Ziad 'Leila, I ordered Kareem to take to school, not Ziad.' d. la-we:n ʔamarti: kari:m ye:xod layla:? to-where you.ordered Kareem take Leila 'Where did you order Kareem to take Leila?' e. kari:m la-we:n ʔamarti:-*(h) ye:xod layla:? Kareem to-where you.ordered-*(him) take Leila 'Kareem, where did you order *(him) to take Leila?' f. layla: la-we:n ʔamarti: kari:m ye:xod-*(ha:)? Leila to-where you.ordered Kareem take-*(her) 'Leila, where did you order Kareem to take *(her)?' Second, Slocum's analysis predicts that vocatives should be ungrammatical if they appear after quantifiers or if they interrupt idiomatic expressions. This is so because these elements resist topicalization. 6 This prediction is not borne out either. The tweets in (51)  The data in this subsection indicate that topicalization cannot successfully account for mid-sentential vocatives. And while the discussion focuses on Arabic, the observation is not limited to Arabic. The examples in (57) and (58) show that mid-sentential vocatives in English and Spanish may occur after wh-words, (a), may be preceded by quantifiers, (b), and may interrupt idioms, (c). The more plausible conclusion then is that vocatives have freedom of placement and may interrupt their host clause freely, which is expected if vocatives are parentheticals. Now we turn to (6b) and the topic of connection between vocatives and their host clause.

Connection with the host clause
In his discussion of vocatives, Ashdowne maintains that, unlike parentheticals, they "have some necessary connection with the accompanying utterance through something in the discourse context, viz. they must refer to the addressee(s) and are unacceptable if they do not" (2002: 155; emphasis in original). In this section, I will approach the question of connection from a broader semantic and syntactic perspective before I address Ashdowne's point. Some of the points discussed here are well-established about vocatives cross-linguistically. For a start, vocatives, including Arabic vocatives, may stand alone; for example, (1) and (2) in Section 1. They do not have to be part of a larger utterance, and thus no connection with a host clause is required at all. Ashdowne (2002: 147) identifies such vocatives as their own group and calls them 'isolative.' There is nothing syntactically special about these vocatives, at least not in Arabic, and therefore they do not warrant a special label. Even proper names, which are more likely to be used as calls to grab an addressee's attention and are normally followed with an utterance, may be used as stand-alone vocatives. For example, a man called Ziad could be addressed via (59) Even when they are part of a host clause, vocatives are cross-linguistically optional elements. They may be added or deleted from their host clause without altering its truth-conditional meaning or the conditions under which it is considered true or false. When they are added to an utterance, their contribution is pragmatic, evaluative; they may express the speaker's evaluation of the addressee and/or of the event in the host clause. They also serve to manage the relationship between the speaker/writer and the addressee; they may maintain, redefine, enhance, or challenge it (see Spencer-Oatey 2002;Haegeman & Hill 2013). This is why, vocatives are characterized in the literature as interpersonal elements (e.g., Hill 2013). The tweet in (60) is a telling example of the interpersonal, evaluative characteristics of vocatives. The tweet is a response to a post about Islam and Muslim holy sites. The post is full of factual errors. The tweeter uses the vocative 'jackass' six times, always between parentheses, to express his negative evaluation of the writer of the post. 7 If the relationship between the tweeter and the addressee was once good, this vocative is likely to challenge it and redefine it. If it was already bad, the vocative is likely to make it worse. Importantly, the insertion or deletion of the vocative does not alter the factual veracity of the host construction.
(60) To begin with, there is no mosque called Jerusalem, (you jackass), and Al-Aqsa mosque is the first of the two Qiblas, (you jackass), and from there the prophet ascended into heaven, (you jackass), and even if it were a prayer room in your house and people want to make it impure, you should defend it, (you jackass), because all places of worship are equally sacred, (you jackass), and as a Muslim, you are expected to give sacred persons, places, and things a priority, (you jackass). Still on the topic of connection, Arabic vocatives, like vocatives in other languages, do not display sensitivity to clause-typing. For example, they are compatible with questions and imperatives; e.g., (47) and (56) respectively. In this sense, they are even less connected to the host clause than some other pragmatic elements may be. For example, discourse particles in Romanian and West Flemish show sensitivity to clause-typing (Haegeman & Hill 2013). And while there is normally a distinction between sentence-initial and sentencemedial vocatives, with the former normally -though not necessarily always -serving as calls used to grab the hearer's attention and the latter as addresses employed "to establish or reestablish the relationship between the speaker and the addressee" (Slocum 2016: 12), this distribution seems to be informed by pragmatics rather than syntax (Slocum 2016: 128-129).

‫ﻣﺎﻓﻲ‬
Further evidence that vocatives are syntactically independent of the host clause comes from scope information. For example, vocatives in Arabic and cross-linguistically fall outside the scope of negation, a behavior that is typical of non-truth-conditional material, including parentheticals (e.g., see Vries 2012). Take the tweet in (61). If someone negates the statement or responds with laʔ, miʃ saħi:ħ 'no, incorrect,' the negation targets the veracity of the virus information. It does not target the vocative to indicate, for example, that the addressees are not ʃaba:b 'guys but, say, sˤaba:ya: 'gals.' The vocative may be challenged separately, however. For example, one may respond with ʃaba:b? ʃu: ʔasˤdak ʃaba:b? niħna: sˤaba:ya: 'Guys? What do you mean guys? We are gals' (see Potts 2011 Arabic Vocatives are even more syntactically independent from their host constructions than some parentheticals might be. For example, while some parentheticals, such as nonrestrictive relative clauses, may be "directly attached to an anchor/antecedent" and may "form a constituent with it" (Vries 2012: 154), Arabic vocatives never form a constituent with an element in the host clause no matter where they are linearly pronounced in it. Also, some parentheticals, such as appositions, may "take over the Case of the anchor in many languages" (Vries 2012: 155). Vocatives are different in this respect. "They are unlike other cases in that they do not mark the relation of dependents to heads" (Blake 1994: 9;cited in Ashdown 2002).
Lack of connection between vocatives and their host construction is also evident in the possible mismatch in number agreement between the vocative nominal and the second-person referent in the host clause. The focus here is on vocatives employed in what Stavrou (2013), drawing on Moro (2003), calls intradeictic address. 8 Both researchers distinguish between extradeictic and intradeictic address. In extradeictic address, the sentence does not contain a 2 nd person element (e.g., subject, object), and if a vocative is used, there is "no co-indexation between the vocative and a clausal argument" (Stavrou, 2013: 317). Many examples used in the paper so far fit in this category; e.g., (51) through (53). In intradeictic address, on the other hand, the sentence necessarily contains one or more 2 nd person elements (e.g., subject, object) and "[i]n the presence of a vocative …, any second person element in the sentence inescapably co-refers with it" (Stavrou 2013: 316). To Stavrou, this co-reference is significant because it "points to the 'engagement' of the vocative to the syntactic structure" (2013: 2016). Examples (46) and (48), repeated here as (62) and (63), fit in this category.

(62)
le:ʃ ya: ħuku:me tra:ʒaʕti: ʕan l-ʔamn l-qawi: why voc government backed.away from the-security the-strong l-kuwayti: the-Kuwaiti 'Why, government, did you back away from strong Kuwaiti security?' (63) ʃu: ʕmiltu: ya: ʕarab bi-mawdˤu:ʕ naʔl l-safa:ra ʕala l-quds ? what you.did voc Arabs with-issue moving the-embassy to Jerusalem What did you do, Arabs, with regard to the issue of moving the embassy to Jerusalem?' Now, observe the tweets in (64) and (65). The vocative in both is singular, but all reference to the addressee in the host clauses is plural. The examples show that it is possible to reference a whole group in the host clause but use a vocative that references each member individually. Of course, it may be argued that the tweeter may address one person with the vocative but make reference to the addressee and her/his cohort in the host clause. This is true; however, the tweets in (64) and (65) are marked as "replying to" two individuals each. In fact, it is not uncommon to find similar tweets in which three or more individuals are marked as addressees via @, and yet the vocative is singular. The opposite is also true. It is possible to use a vocative that references a whole group, while at the same time make reference to individual members of the group in the host clause. The tweets in (66) and (67) are examples. They show that a plural vocative may be incorporated in a host clause in which all reference to the addressee is singular.
(66) ya: ʃaba:b ʔiza: baddak tiħki: msˤi:be la-ʔabu:-k xo:d voc guys if you.sg.want tell disaster to-father-your.sg you.sg.take ha-l-nasˤi:ħa stanna-a lamma: yitɣadda: w-yiʃrab this-the-advice you.sg.wait-him when he.eat.lunch and-he.drink ʃani:na w-ʔiza: ʔakal batˤtˤi:x biku:n ʔaħsan yogurt.drink and-if he.ate watermelon it.would.be better w-ʔaħsan… and-better 'Guys, if you want to talk to your father about a major problem, take this piece of advice. Wait for him until he has had lunch and a yogurt drink, and if he has eaten watermelon, it would be even better...' Why are examples (64) through (67) significant? There is evidence that speaker and hearer information in the left periphery of a structure controls agreement lower down in the construction; see, for example, Miyagawa (2012) who shows that the allocutive marker on the verb in Basque agrees with the hearer information in the left periphery, irrespective of whether the hearer is lexical or not. Under the analysis that a vocative phrase merges as part of the speech act phrase in the left periphery, we expect the phi-features of the hearer to control the phi-features of coreferential elements in the host clause. This is the type of coreference that Stavrou (2013) seems to point to in her discussion of vocatives used in intradeictic address. The Arabic tweets in (64) through (67) show that full agreement may not always apply and that mismatches are possible. Such mismatches may be easier to explain if vocative phrases are analyzed as parenthetical adjuncts. Under this analysis, we expect elements referring to the addressee in the host clause to agree with the unpronounced hearer in the left periphery but not necessarily with the pronounced vocative. Agreement with the vocative is pragmatically determined with the possibility of some mismatches.

‫ﻳﺎ‬
Further evidence along the same lines comes from the fact that a speaker may address a group of people but use a vocative to anchor parts of the utterance as especially but not exclusively relevant to certain individuals in the group. Consider (68) from a Lebanese play called bi-l-nisbe la-bukra ʃu: 'what are the plans for tomorrow' by Ziad Al-Rahbani. In this scene, the owner of a bar in Beirut is talking to three of his employees, Rida, Najib, and Zakaria. He is not satisfied with their work; he believes that they are not trying hard enough to be friendly and to attract more clients. He characterizes Rida as clumsy, Najib as gloomy, and Zakaria as lifeless. He goes on to suggest that the situation may easily improve if the three employees try to be more agile, cheerful, and animated. The suggestions are relevant to all three employees to varying degrees (see Slocum 2016: 23 for a note on relevance). The speaker uses vocatives to anchor each suggestion to one employee based on a salient aspect of that employee's personality. However, by doing so, he does not imply that Rida needs to be agile but does not need to worry about being cheerful or animated. By the same token, he does not imply that Zakaria only needs to be animated but could otherwise be clumsy. In other words, the manager's suggestions are expected to be taken seriously by all the employees as a group. At the same time, the manager is able to use a vocative to profile a specific suggestion as especially relevant to one employee. This is only possible if there is no syntactic connection between the vocative and the host clause.
(68) tˤarru:-ha: ħaliħlu:-ha: laħilħu:-ha:. l-ʃaɣle badha: xiffe ya: you.pl.soften-it resolve-it loosen-it the-situation need agility voc ridˤa:, badha: basme ya: naʒi:b, badha: badha: ru:ħ ya: zakariyya:. Rida need smile voc Najib need need spirit voc Zakaria 'You all need to loosen up. The situation needs some agility, Rida, needs a smile, Najib, needs needs positive energy, Zakaria.' The constructed example in (69) is further illustration. In this case, the parents of three children are going on a trip for one day. The children's aunt will be taking care of them. A parent is giving them instructions to make sure they wake up early, put their rooms in order, and eat breakfast before they go to school. They are also expected to be nice to their aunt. All the instructions are relevant to all the children, as the plural agreement on the verbs indicates; yet, the parent anchors some of instructions to some of the children based on their behavioral history. For example, Samir is known to be the most resistant to eating breakfast before school, so the parent anchors the instruction of eating breakfast to him. The parent still uses the verb 'eat' with second-person plural agreement, indicating that all three kids are concerned. In this respect, (69), as well as (68), is different from (70) (from Ashdowne 2002; fn. 17-iii) in which there are three presents and three addressees involved, but each addressee receives a different present.

(70)
English Here are your presents: I'm giving you this, Mark, you this, John, and you this, Mary.
Now we return to Ashdowne's observation that vocatives show necessary connection with the host clause in that they "must refer to the addressee(s) and are unacceptable if they do not" (2002: 155). This section has shown that any connection between vocatives and the host clause is more likely to be the outcome of pragmatics rather than syntax or semantics. Pragmatically, one could expect a vocative to refer to an addressee. Still, the lack of connection raises the question whether it is possible for a clause to host a vocative that does not refer to the addressee. My data shows that this is possible only if the non-addressee serves as a targeted overhearer, while the actual addressee is not channel-linked. In other words, the utterance event must involve the following two types of participants: A. Addressed Participants: These are individuals addressed "by the speaker in a manner to suggest that his [sic.] words are particularly for them" (Goffman 1981: 9-10). They are not channel-linked, whereby channel-linkage is "the ability to receive the message" (Levinson 1988: 174) and to respond to it if they choose. The speaker's message is addressed to them in absentia, solely for the benefit of the group in (B). B. Unaddressed Participants: These are channel-linked recipients. They are not specifically addressed by the speaker, but the speaker's message is uttered for their benefit. They serve as the indirect target or the targeted overhearers of the message, and as such they are able to assess the speaker's message and respond to it if they choose to (see Levinson 1988: 194-197).
At least one scenario meets the conditions in (A) and (B). 9 First, however, some background about a conversational practice in Arabic is in order. In Arabic conversations, when reference is made to animals, body parts, and other objects that may be considered offensive, disgusting, or otherwise inappropriate, the 9 Another scenario that satisfies the conditions in (A) and (B) involves optatives like (i). In this case, God is a non-channel-linked addressee. The tweeter may believe that God receives the message, but there is no way to tell for sure. The tweeter does not expect God to assume the role of the speaker. speaker is expected to use an expression such as l-baʕi:d 'the far one' or ħa:ʃa: X 'to the exclusion of X,' where X references recipients of the message. The expressions indicate that the inappropriate terms do not reference the recipients and do not mean to offend or show disrespect. The tweets in (71) and (72) are examples. In (72), the tweeter expresses indignation toward another tweet that defends the Syrian regime. Notice that the expression ħa:ʃa:-kum ya: kira:m 'no offense, noble people' contains a vocative referencing the individuals whose presence the reader is mindful of and to whom he tries to be respectful.
(71) la: yadˤurru l-saħa:ba nabħu l-kila:bi ħa:ʃa: l-ħa:dˤiri:n neg harm the-clouds barking the-dogs exceptive.of the-attendees wa-l-ʔaħra:r and-the-free.people 'Clouds are not harmed by the barking of dogs (i.e., honorable people are not affected by the actions or words of lowlifes), with no offense to the attendees (followers reading the tweet) and free people.' Importantly, in a conversation that involves gossip about an absent third party, a speaker may use a turn or an utterance event to address the subject of the gossip directly as if s/he were present. Consider (73), for example. The tweet defends a Palestinian poet and columnist called Tamim Al-Barghouti. Al-Barghouti had recently addressed the United Nations, and a Saudi writer claimed that the address had been dictated to him by the Israeli delegation. The tweeter pretends he is addressing the writer with the comment in (73). The term l-baʕi:d 'the far one' is an explicit indication that the Saudi writer as an addressee is not channel-linked. And in fact he is not; for example, he is not atted or hashtagged. It is also an indication that the tweeter is aware of the presence of channellinked overhearers, namely, his followers on Twitter. By using l-baʕi:d 'the far one,' he states that the vocative ya: ħma:r 'voc jackass' is not meant for them and that he is not addressing any of them.  Now, consider the constructed example in (74). It could be used in the following context: the speaker learns that Ziad was late to an important meeting because he had gone out partying the night before. The speaker is gossiping about Ziad to a group of friends. At one point, she shifts her gaze away from the hearers and pretends she is talking directly to Ziad. The performance is reminiscent of the use of the present tense in story telling in English in that it makes gossip more real and more dramatic. During the performance in (74), the hearers are no longer addressees; rather, Ziad is; he is an addressed participant although he is not channel-linked. The friends who are listening to the speaker are channel-linked recipients; their role is converted into targeted overhearers.

‫ﻻ‬
(74) ʕindak ʔiʒtime:ʕ mhimm te:ne yo:m ya: ʒaħeʃ, ħaʃa: you.have meeting important following day voc mule exceptive.of l-se:mʕi:n, w-bitru:ħ btishar. the-listeners and-you.go party 'You have an important meeting the following day, you mule, no offense to the listeners, and you go party.' The tweet in (75) is a naturally occurring example. It was written in response to a retweet. The retweet shows a graffiti that reads 'God is great but the revolution is greater.' The retweeter expresses disbelief at how blasphemous the graffiti is. The tweeter in (75) agrees and goes further to offend the individual behind the graffiti. That individual is not channel-linked; e.g., s/he is not atted or hashtagged. The retweeter, on the other hand, is a channel-linked recipient; the tweet is anchored to him via @, and thus he serves as a targeted overhearer. The tweeter seems to assume that his followers are also channel-linked since they could read his tweets. In this sense, they are targeted overhearers as well. 10 (75) tˤiz fi:-kum w-bi-he:k sawra ya: kle:b (ħa:ʃa:-kum boo to-you.pl and-to-like.this revolution voc dogs (exceptive.of-you.pl ya: kira:m) tatakabbaru:n ʕala: man xalaqa-kum ya: baʒam. voc nobles) you.behave.arrogantly with who created-you voc savages 'Screw you and screw a revolution of this type, you dogs, (no offense to you, noble people). You behave arrogantly with you creator, you savages. Example (75), as well as example (i) in footnote 9, shows that Ashdowne's (2002: 155) statement that vocatives "must refer to the addressee(s) and are unacceptable if they do not" may be an oversimplification and that the term 'addressee' needs closer examination. In (75), the vocatives refer to two types of participants: an addressed participant that is not channel-linked and unaddressed overhearers that are channel-linked. The tweet could alternatively only include one or the other or neither.

‫ﻅﺰ‬
Many of the tweets we have seen so far contain multiple vocatives, which raises two questions: (i) How many vocatives are allowed in any one host clause? (ii) If more than one vocative is allowed, do they have to be listed continuously or can they be listed intermittently? These questions are addressed in the next section.
(90) l-wadˤiʕ ya: ʃabe:b, ma: baʕa: yinħamal. sˤa:r le:zim the-situation voc guys no longer bearable it.has.become necessary ya: ziya:d, hallaʔ ʔaktar min ʔay waʔit madˤa: ya: layla:, ʔinno voc Ziad now more than any time past voc Leila that nfakkir bi-tˤari:ʔa mubtakara tʃi:l-na: min ha-l-ʒu:ra ya: sami:r, we.think of-way innovative remove-us from this-the-hole voc Samir w-tħassin mustawa: l-ʔinte:ʒ bi-l-ʃirke. and-improve level the-production in-the-company 'The situation, guys, is no longer bearable. It has become necessary, Ziad, now more than any time before, Layla, that we think of an innovative way that would remove us from this hole, Samir, and that would improve the level of production in the company.' The paper set out to show that Arabic vocatives are parenthetical adjuncts. Parenthetical adjuncts are not a uniform group. They come in different forms (e.g., appositives, comment clauses) and vary a lot in terms of their internal structures. However, their title follows from their external behavior and the way they relate to their host clause. Sections 2, 3, and 4 have shown that Arabic vocatives, like parentheticals in general, are minimally constrained when it comes to their external syntax and how they relate to the host clause. They behave on a par with parentheticals with regard to three criteria put forth by Ashdowne (2002): (i) freedom of placement, (ii) lack of connection with the host clause, and (iii) no limit on the number of occurrences. Still, vocatives are pronounced, often as part of a larger construction. The question that follows is: How are they syntactically integrated with the host clause? The following section provides a possible answer.

The external syntax of vocatives
A number of accounts have been proposed to account for parenthetical adjuncts. I briefly present two here: Espinal (1991) and Vries (2012).
Drawing on work on autosegmental phonology, as well as on three-dimensional syntactic theories proposed by Goodall (1987), Haegeman & Van Riemsdijk (1986), and Haegeman (1988), Espinal accounts for the external syntax of parenthetical elements by proposing a three-dimensional approach to syntactic derivations. The approach allocates "the surface structure of a surface string into a number of independent trees that may be located in different planes" (Espinal 1991: 742). To Espinal, the host clause and the parenthetical element merge in separate planes before they intersect at the terminal string, as (91)  Morphology and morphosyntax may restrict where a vocative may adjoin as well. A vocative may not come between two elements if one of them is a bound morpheme. For example, the definite article l-in Arabic is a prefix. Therefore, a vocative may not adjoin to the NP complement of D if D is the definite article l-. Also, ʔidˤa:fa 'genitive constructions' in Arabic are resistant to intervening material. Therefore, it is unlikely for a vocative phrase, or any parenthetical element, to come between two parts of a genitive construction.
The second question has to do with sentence-initial vocatives. Do they merge in the speech act phrase (SAP) in the left periphery or do they also merge as parenthetical adjuncts? Examples (64) through (67) above seem to indicate that they merge as adjuncts. Take (67), repeated here as (98). The sentence-initial vocative is plural, while the reference to the hearers in the host clause is singular. If we assume that there is agreement between the speaker/ hearer information in the left periphery and any reference to the speaker and hearer lower in the structure, then the vocative cannot be part of that agreement relationship. Finally, it may be plausible to suggest that VocP itself does not merge as a parenthetical adjunct. Instead, VocP merges as an argument of SAP, which in turn merges as a parenthetical adjunct, as (99) and (100) demonstrate. Importantly, this SAP is different from the one in the left periphery. This proposal seems to be aligned with Starvou's (2013) brief observation that mid-sentential vocatives "display the distribution of any other parenthetical," while also adopting the view that vocatives merge in SAP (301-302). Admittedly, Starvou (2013) does not reconcile the view that SAP resides in the left periphery with the fact that vocatives may also be mid-sentential.  (99) and (100) represent a theoretically feasible proposal is a question that I defer for another occasion. Empirically, the proposal may be motivated on at least two grounds. First, we saw above that vocatives may stand alone; they do not need to be part of a host clause. As stand-alone syntactic objects, they are use-conditional elements (see Gutzmann 2015) that reflect the speaker's attitude toward the addressee and serve as tools for relationship management. Since the speaker's point of view is always present, it is reasonable to assume that a speaker projection -and thus SAP -is always present as well. Second, there is lexical evidence for the speaker's presence in the context of vocatives. Arabic varieties license what is normally referred to as inverse vocatives or reverse role vocatives. These are vocatives in which a senior person addresses a younger person by using a term of address that the younger person normally uses. For example, a mother may address her daughter or son with ya: ma:ma: 'voc mom' or ya ʔimm-i: 'voc my mother'; see Rieschild (1998) for an analysis of such vocatives in Lebanese Arabic. Importantly, reverse role vocatives may also serve as stand-alone utterances. The fact that the speaker is spelled out in such vocatives may be used as evidence for the presence of the speaker projection; see Akkuş & Hill (2018) for a recent analysis of the internal syntax of similar vocatives in Turkish.

Conclusion
Commenting on the external syntax of vocatives, Hill (2014) writes, [T]he variation in the location of the vocative (i.e., clause initial, medial or final) does not follow from the random insertion of the vocative noun in the sentence, but from constituent movement above the argument position in which VocP is merged. For example, in The door, Mary, should be closed, the DP the door moves to a hanging topic position in the functional field of SA o , above Mary in Spec, SAP. In this way, our analysis pre-empties any attempt of treating VocP as parentheticals, adjuncts or appositions in the clause. (11)(12) This paper presents evidence that vocatives may in fact be parenthetical adjuncts. Vocatives, this paper shows, exhibit a variety of behaviors that are characteristic of parenthetical elements. And while any one piece of evidence may be challenged independently (e.g., it may be argued that some idiomatic expressions may be more prone to topicalization), taken collectively, the evidence is overwhelmingly in favor of the adjunction approach to the external syntax of vocatives. The focus in the paper has been primarily on Arabic. However, Section 2 also presents preliminary evidence from English and Spanish in order to show that the conclusions arrived at here are not limited to one language and that a closer look at vocatives in other languages is warranted, with special attention to be paid to naturally occurring data.
Finally, I suggest that naturally occurring data should be given more weight in studies on theoretical linguistics in general. Reliance on constructed examples, elicitation, and grammaticality judgment may not be sufficient, especially when a study involves the examination of pragmatically loaded expressions like vocatives that crucially rely on context. I would not be surprised if native speakers judge some of the examples examined in this paper less favorably if they were presented to them in shorter forms out of context. When I presented these vocatives as parts of full tweets, however, none of the native speakers I consulted judged them as ungrammatical or degraded. It, therefore, becomes an open question whether a closer look at vocatives in naturally occurring data in other languages, including those in which vocatives have already been examined, would show that they are less constrained than originally thought.