On the syntax of addressee in imperatives: insights from allocutivity

Zanuttini (2008), and Zanuttini et al. (2012) claim that all imperatives host the Jussive head, an imperative-specific 2nd person projection which agrees with the subject to restrict it to the addressee. This paper argues against the proposed link between clause type and addressee encoding by examining imperatives with allocutivity in Punjabi which exhibit two unique properties: first, these imperatives lack 2nd person imperative agreement and appear instead with allocutivity, which is available across clause-types. Secondly, they can only be used in contexts with a specific addressee, whose (non)-honorific status vis-à-vis the speaker is well-defined. I argue that allocutivity in Punjabi does not underlie an additional person probe distinct from the subject agreement probe. Instead, there is a unique person probe in the C-T domain, labeled Pers0, which agrees with the addressee/Adr upon failing to agree with the subject. Extending this syntax of allocutivity to imperatives, the presence of obligatory allocutivity in the imperative underlies a unique Pers0, which in the absence of a co-occurring person probe, i.e. the Jussive head, mediates an agreement relation between the Adr and the minimally specified pro subject to restrict the latter to the addressee. Furthermore, I show that the allocutive head/Pers0 is distinct from the Jussive head in that it not only hosts a 2nd person feature, but also a [status] feature which encodes the speaker-addressee relation. Valuation of both these features by the Adr restricts allocutive imperatives to specific addressee contexts.


Introduction
It is well-known that in many languages, the subject of an imperative is restricted to the addressee in that it must refer to, overlap with, or quantify over the addressee or the set of addressees (Mauck et al. 2005)). 1 The interpretive requirement of the imperative subject to be associated with the addressee amounts to unique properties of imperative subjects, which are not available for subjects of declaratives or interrogatives (Bolinger 1967;Schmerling 1982;Platzack & Rosengren 1997;Potsdam 1998;Rupp 2003;Zanuttini 2008). For instance, languages such as English, which are not pro-drop, allow null subjects in imperatives. Moreover, these null subjects in imperatives can bind 2nd person possessive pronouns, as shown in (1) from Zanuttini (2008: 187). Similarly, tag questions in imperatives with null subjects can only have 2nd person pronouns, as in (2) also from Zanuttini (2008: 188).
Unique properties of imperative subjects are also seen with overt subjects in English. For instance, overt pronominal subjects in imperatives are restricted to 2nd person, as shown in (3) based on Zanuttini (2008: 189).
(3) You/*I/*he read a book! While 3rd person pronouns are disallowed in imperatives, other 3rd person subjects such as definite DPs, quantifiers and proper names are permitted. As Zanuttini (2008: 190) notes, despite their featural specification, said subjects can not only bind a 3rd person anaphoric object, but also a 2nd person anaphoric object. This is demonstrated with the quantifier subject in the imperative in (4a). Compare with the declarative in (4b) from Zanuttini (2008: 191), where the quantifier subject fails to bind a 2nd person object.
(4) a. Everyone i raise (his i /her i /their i )/your i hand! b. Everyone i should raise his i /her i /their i /*your i hand.
In their seminal work, Zanuttini (2008) and Zanuttini et al. (2012) argue that these special addressee-related interpretive properties of the imperative subject cannot be derived solely from the fact that the core function of imperatives is to direct the addressee to do something. Instead, these facts follow from unique imperative syntax. More specifically, they propose that imperatives consist of a representation of the addressee called the Jussive Phrase (also Jensen 2003;Rupp 2003;Bennis 2006;Isac 2015). The Jussive Phrase, found exclusively in the imperative clause-type, encodes the addressee via an interpretable and valued 2nd person feature. In the presence of a T head that lacks a person feature, the 2nd person feature on the Jussive head agrees with the imperative subject, which occurs with an unvalued person feature to restrict it to the addressee, as is represented in (5). 2 This leads to unique interpretive properties of imperative subjects.
(i) [[run]] = λx : x is the addressee. x runs In this paper, I use the term Jussive Phrase only to indicate the unique 2nd person projection -the requirement for addressee-restriction on the imperative subject does not hinge on the 'minimal semantics' account, and can be analyzed via other accounts of imperative meaning (Kaufmann 2011;Han 1998;2001 etc.).
While (5) is a very promising syntactic analysis of the addressee-relatedness found with imperative subjects, the proposed association between imperative clause-type and the 2nd person projection in syntax, is curious. In recent years, cross-linguistic studies have shown that natural languages exhibit allocutivity, a phenomenon which also involves a syntactic representation of the addressee but which is not restricted to imperatives. To elaborate, allocutivity is a phenomenon, where certain languages have distinct verbal morphology that encodes the addressee of the speech act (Oyharçabal 1993;Miyagawa 2012;Antonov 2015;McFadden 2017;Kaur 2017;Haddican 2018;Alok & Baker 2018;Yamada 2019;Alok 2020 etc.). A classic example is provided by Basque. Consider (6a) and (6b), where a singular male addressee of the speech act is encoded via -k on the verb and the singular female addressee is encoded via -n.
(6) Basque (Oyharçabal 1993: 92-93) a. Pette-k lan egin di-k Peter-erg work do.pfv 3erg-m 'Peter worked.' (said to a male friend) b. Pette-k lan egin di-n Peter-erg work do.pfv 3erg-f 'Peter worked.' (said to a female friend) These allocutive agreement markers, which encode the addressee, his/her gender, number and (non)-honorific properties, are taken to underlie an agreement relation between a representation of the addressee in the left periphery and a functional head. Following Speas & Tenny (2003); Hill (2007); Miyagawa (2012); Zu (2015); (2018); McFadden (2017), among others, I model the representation of the addressee which underlies allocutivity as an argumental Addressee (henceforth Adr). Consider the following proposal provided by Hill (2007) for the Speech Act Projection (SAP) consisting of the Adr. Like the verbal domain consisting of the vP-VP structure, where the specifier of the vP hosts the external argument (agent) and the specifier of the VP, the indirect object (goal), the SAP consists of two layers SAP1 and SAP2. Mirroring the hierarchy of subject and indirect object positions in the verbal domain, the speaker and the addressee are encoded as arguments of the speech act in the specifier of SAP1 and SAP2 respectively. This structure of the SAP and particularly the Adr in the left-periphery lends itself readily to an account of allocutivity -I assume that in allocutive languages, a functional head, say X 0 , is borne with unvalued phi-features, which must be valued by agreement with the argumental Adr to yield allocutivity. For demonstration, see (7).
(based on Hill 2007on Hill : 2099 Given the possibility of agreement across all clause-types with the Adr in certain allocutive languages, this paper addresses the following question: in imperatives with allocutivity, is the Jussive head obligatory to derive addressee-restriction on imperative subjects, with allocutivity acting as a mere add-on, as in (8a)? Or, can the functional projection which hosts allocutivity across clause types replace the Jussive head to derive addresseerestriction on imperative subjects, as in (8b)? (8) the subject, the features valued on Pers 0 by the Adr are shared with the imperative subject, resulting in addressee-restriction on the subject. I situate this difference in the distinct nature of the imperative subject -it is a minimally specified pro which is identified via corresponding agreement (in the sense of Rizzi 1986), in contrast with the declarative subject, which even when covert, does not depend on corresponding agreement for its identification, but instead on discourse factors (Butt & King 1997;Butt 2001).
Furthermore, the ban on generic addressee readings in allocutive imperatives follows from the special properties of Pers 0 . In contrast with the Jussive head which hosts only a 2nd person feature, Pers 0 hosts not only a person feature, but also an extra [status] feature which encodes the speaker-addressee relation (in the sense of Portner et al. 2019). Since the allocutive Pers 0 must agree in not only person but also in the status feature with the Adr, it is blocked from generic addressee contexts.
The proposed analysis of the imperative with allocutivity leads to two important results for addressee-encoding in (imperative) syntax. First, it is indeed the case that imperative subjects require licensing via agreement with a 2nd person feature as proposed by Zanuttini (2008), and contra approaches which derive addressee-restriction solely as a consequence of imperative force (Han 1998;Portner 2004). However, it is incorrect to postulate a dedicated imperative-specific representation of the addressee, i.e. the Jussive head across-the-board. Allocutive languages have functional projections, which undergo agreement with the Adr across clause-types -in its imperative occurrence, the allocutive head can substitute the specialized Jussive head and derive addressee-restriction on the imperative subject. Secondly, there are at least two variants of the functional head, which bears a 2nd person feature and encodes the addressee in syntax: (i) the Jussive head with only a 2nd person feature, and (ii) the allocutive head, with an additional [status] feature that defines the speaker-addressee relation. The presence of allocutive head, which bears both a 2nd person and a status feature, disallows a structure from occurring in contexts with a generic addressee. No such restriction is observed in structures with only the Jussive head, which bears a 2nd person feature, but crucially lacks the status feature.
The paper is structured as follows: section 2 introduces the allocutive imperative with focus on its unique morphological composition vis-à-vis the standard imperative, and its distinct addressee readings. Section 3 analyses allocutivity and provides arguments to rule out the Jussive head in the allocutive imperative. Differences in generic/specific readings of the addressee across the two imperatives are derived in section 4. Employing the findings of sections 3 and 4, section 5 presents the derivation of standard and allocutive imperatives. Section 6 concludes the paper.
Where not cited, data come from the native speaker author's judgments, confirmed by two adult native speakers of Punjabi (from Kanpur, Uttar Pradesh in India).

Introducing the allocutive imperative
This section introduces the allocutive imperative in Punjabi with focus on its two unique addressee-related properties: (i) the allocutive imperative is obligatorily composed of the allocutive ending to the exclusion of the imperative-specific ending, and (ii) the allocutive imperative can only be used in contexts with a specific addressee. 3

Absence of the imperative ending in the allocutive imperative
Following the pattern exhibited by most New Indo-Aryan languages (Masica 1991: 476), Punjabi typically makes imperatives with a bare verb form. I label them standard imperatives. With a 2nd person singular subject, the verb occurs with null morphology as in (11). With a 2nd person plural subject, the verb is marked with -o, see (12).
(11) kitaab paRh-Ø book read-imp.2sg 'Read the book!' The 2nd person endings in the imperative, while historically related to 2nd person forms employed in indicatives (Hewson & Bubenik 1997;Butt & Rizvi 2010 etc.), constitute an imperative-specific paradigm synchronically. As shown in Table 1, the 2nd person singular imperative ending is null, in contrast with the declarative ending eN. 4 In addition to standard imperatives, the language makes another imperative with the allocutive ending, which in Punjabi is available across clause types. Punjabi (Kanpur, Lahore and Gujrat district varieties reported thus far) is an allocutive agreement language -it has distinct verbal forms for the singular and the plural addressee, corresponding to ii/aa and je respectively (Bailey 1912;Akhtar 1997;Butt 2007;Kaur 2017;. These endings comprise a separate paradigm, distinct from the endings in standard imperatives and 2nd person subject agreement in declaratives. Crucially, unlike imperative endings, which are specific to the imperative clause-type, allocutivity occurs across clause-types in Punjabi. I illustrate with declaratives here. In the declarative in (13), the presence of ii encodes a singular utterance addressee. The forms ii and aa are allomorphs, determined by the phonological ending of the preceding verb. When the verb ends in -aa, the form ii is used, as is the case in (13) where the progressive verbal form ends in -aa as a result of agreement with the masculine nominative subject. Contrastingly, when the verb ends in -ii typically due to agreement with a feminine argument, the form of the allocutive marker changes to aa, as shown in (14). In the declarative in (15) The allocutive markers in the above declaratives are optional in that they can be replaced by the minus person (also used as default) form of the 'be'-auxiliary with no change in the truth-conditional semantics of the clause. Consider (16), with the 'be'-auxiliary. This sentence has the same meaning as those in (13) to (15), except the missing overt reference to the addressee.
(17) *kitaab paRh-Ø ii book read-imp.2sg alloc.sg 'Read the book!' (18) *kitaab paRh-o je book read-imp.2pl alloc.pl 'Read the book!' However, the language makes an additional imperative with a distinct verb form, composed of the bare verb stem plus -(e)yaa ending (e.g. kar-eyaa 'do', jaa-yaa 'go' etc.). This verb form is unique to the imperative paradigm, and does not occur elsewhere in the verbal paradigm of the language. 5 Crucially, this imperative composed of the V-(e)yaa form (instead of the bare verb form) must occur obligatorily with the allocutive ending je. 5 At first glance, the V-(e)yaa form in the allocutive imperative seems homophonous with the perfective/ participle verbal paradigm in Punjabi. Consider verbs which end in a consonant, such as parh 'read', vec 'sell' etc. The allocutive imperative form of these verbs is paRheyaa and veceyaa respectively, which is homophonous with the perfective verb form inflected for masculine, singular agreement (also the default agreement form). However, this similarity disappears upon considering verbs which end in vowels, such as khaa 'eat', de 'give' etc. In the allocutive imperative, these verbs occur as khaayaa and deyaa. In contrast, their perfective forms have suppletive morphology; consider khaddaa and dittaa. Moreover, the verb form in the imperative is phi-invariable, as compared to the perfective verb, which shows inflection in number and gender. For instance, the verb 'eat' in the imperative is invariably khaayaa, in contrast with the perfective verb, which is realized as khaddaa for masculine singular (and default) agreement, khaddii for feminine singular agreement, khadde for masculine plural agreement and khaddiyaaN for feminine plural agreement. This shows that the verb form in the allocutive imperative is unique to the imperative paradigm. Thanks to Rajesh Bhatt (p.c) for a discussion on this. In summary, Punjabi has a unique imperative with allocutivity, which disallows the 2nd person ending found in the standard imperative in the language.

Ban on generic addressee use of the allocutive imperative
In addition to its lack of the imperative-specific ending, the second property of the allocutive imperative which differentiates it from the standard imperative is its inability to be used in contexts with a generic addressee. To elaborate, both imperatives pattern alike in exhibiting addressee-restriction on the subject. Crucially however, in contrast with the standard imperative, which can be used for both generic and specific addressees, the allocutive imperative can only be used in contexts with a specific addressee.
To see addressee-restriction on the imperative subject, consider the following facts. Apart from permitting null subjects (as seen in the examples thus far), both imperatives allow pronominal subjects. Crucially, however, these are restricted to 2nd person pronouns. (20) tusii/*asii/*o kitaab paRh-o 2pl/1pl/3pl book read-imp.2pl 'Read the book!' (21) tusii/*asii/*o kitaab paRheyaa je 2pl/1pl/3pl book read alloc.pl 'Read the book!' 6 The allocutive imperative is a real imperative, as evidenced by its performative nature. As is well-noted in the literature, imperatives are performative (Sadock & Zwicky 1985;Han 1998;Portner 2004;Kaufmann 2011;Oikonomou 2016, among others). This is evident from the inability to challenge an imperative. In other words, one cannot intuitively call an imperative "true" or "false". Consider a context where the speaker (teacher) tells the addressee (her students) to read a book. The students can refuse to do so, but they cannot challenge the command with 'That's not true'.
(i) Teacher: Read the book! Student: # That's not true.
This performative character holds of both the standard and the allocutive imperatives in Punjabi.
(ii) a. jaa-o/jaayaa je go-imp.2pl/go alloc.pl Go! b. #ae sac nayii e this truth neg be.prs.3sg 'This is not true.' Further support for addressee-restriction on the subject comes from its binding properties. Even when covert, the subject of both standard and allocutive imperatives can bind a 2nd person anaphoric object, but not a 1st or 3rd person anaphoric object. Punjabi has person-variable possessive forms, which must match the person feature of their antecedent to be bound. This is demonstrated for the 1st person possessive pronominal object in (22), which can only be subject-bound by a 1st person subject. Apart from the restriction on the overt pronominal subject and binding properties, the addressee-restriction on imperative subjects is also evidenced by agreement facts. Both imperative types in Punjabi can occur with quantifier subjects such as 'everybody', as shown in (25) and (26). The reference of 'everybody' must overlap with the set of addressees, failing which the imperative form cannot be used. Crucially, even in the presence of a 3rd person quantifier subject, the form of the imperative verbal complex remains invariable, i.e. it inflects for 2nd person agreement and not 3rd person.
(25) sabb kitaab paRh-o everybody book read-imp.2pl 'Everybody read a book!' (26) sabb kitaab paRheyaa je everybody book read alloc.pl 'Everybody read a book!' Contrast the above pattern with declaratives shown below, where a 3rd person quantifier subject must occur with corresponding 3rd person agreement, and not 2nd person agreement, see (27). The 3rd person agreement can optionally be replaced by the allocutive marker je; however, there is no overlap between the subject and the addressee in (27).
(27) sabb kitaab paRh raye ne/je everybody book read prog.m.pl be.prs.3pl/alloc.pl 'Everybody is reading a book.' The above observations for both standard and allocutive imperatives clearly show that subjects of both imperatives are restricted to the addressee, which underlies the presence of an agreeing 2nd person feature in the functional spine of the imperative clause. Importantly however, only the allocutive imperative is restricted to contexts with a specific addressee, as I discuss below.
We typically understand the addressee as a speech act participant who is being spoken to directly by the speaker of the speech act. However, as Pak (2015); Portner et al. (2019), among others, propose, this pre-theoretic notion of the 'addressee' needs further nuance. There are contexts where the speaker interacts directly with the addressee, yielding a specific use of the addressee. However, we also find utterance contexts, which involve a non-specific/generic addressee. To see this difference, consider the following imperatives from English. The utterance in (28) identifies the addressee as a specific individual who is interacting with the speaker and is being requested to have tea. This instance of the addressee is called the interlocutor-addressee. On the other hand, (29), in its natural usage, does not pick a specific addressee who is being spoken to directly by the speaker at a given point of time. Here, the addressee "is merely the individual(s) towards whom the prohibition is targeted or directed" (Portner et al. 2019: 4), which could be anyone visiting the zoo. This divide is also attested in Punjabi imperatives. The allocutive imperative must always refer to an interlocutor-addressee, while the standard imperative can be used to refer to a generic or non-specific addressee. Recall that je is the plural allocutive marker. As a plural form, it must identify the relation of the addressees with the speaker, which can be either honorific or non-honorific. The marker je can also be used as a singular form, but only to refer to an honorific addressee.
(30) buaa kholeyaa je door open alloc.pl 'Open the door!' (to (non)-honorific addressees OR a singular honorific addressee) The standard imperative can also be used in an interaction like the allocutive imperative. Consider (31), where the presence of -o must identify the speaker-addressee relation -it can be used for (a) honorific plural addressees, (b) non-honorific plural addressees, or (c) a singular honorific addressee. Crucially, however, unlike the allocutive imperative, the standard imperative is also permitted in contexts where the addressee cannot be identified, and the speaker-addressee relation cannot be determined. There are two contexts which illustrate this. First, only the standard imperative is felicitous with the indefinite pronoun 'someone' as the subject, as shown in (32). Note that in its occurrence in the standard imperative, it is obligatory for the subject 'someone' to overlap with the addressee or a subset if there is a set of potential addressees, failing which this structure is ungrammatical. This is illustrated in (33), where the addressee and the 3rd person subject of the event of opening the door are distinct. Only a phi-feature matching subjunctive verb form is permitted in such sentences. Secondly, standard imperatives can be used in mottos and public orders -contexts which lack an interlocutor-addressee that the speaker is directly talking to (Pak 2015;Portner et al. 2019). Consider the standard imperative used as a motto in (34) and as a general public order in (36). The allocutive imperative is disallowed in such contexts, as shown in (35)  In summary, we have seen two imperatives in Punjabi -the standard imperative and the allocutive imperative. While both imperatives show addressee-restriction on the subject, there are two critical differences between them. First, while the standard imperative employs the 2nd person imperative ending and bans allocutivity, the allocutive imperative lacks the 2nd person imperative ending and appears instead with obligatory allocutivity. Secondly, while the standard imperative is felicitous in contexts with either a generic or a specific addressee, the allocutive imperative can be felicitously used only with a specific addressee, whose (non)-honorific relation with the speaker can be determined. Given these morphosyntactic and interpretive differences, we are faced with the following questions: a. Assuming that the imperative-specific 2nd person ending is the overt realization of the Jussive head, does the allocutive imperative also consist of a Jussive head, albeit covert? Or, does the allocutive imperative lack the Jussive head altogether such that the allocutive head licenses the imperative subject? b. How is the ban on generic addressee readings obtained in the allocutive imperative in contrast with the standard imperative, which allows these readings?

Ruling out a covert Jussive head in the allocutive imperative
This section determines the presence/absence of the (covert) Jussive head in the allocutive imperative. I will first present issues with (8a), which assumes a uniform syntax consisting of the Jussive head across both standard and allocutive imperatives, and then provide independent evidence for (8b) which rules out the Jussive head from the allocutive imperative.
Assuming that the allocutive imperative in Punjabi consists of the Jussive head like other (standard) imperatives would underlie the following structure in (38), repeated from (8a). (38) underlies a two-probe system with both the allocutive probe and the Jussive head (indicated in bold). The Jussive head agrees with the pro subject to restrict it to the addressee. The allocutive head, labeled as X-alloc, agrees with the Adr in the left-periphery and exhibits allocutivity. However, there are two main issues with extending this analysis to Punjabi: the first pertains to stipulated optionality at PF, and the second to deriving distinct addressee-related interpretive effects. We begin with the issue of stipulated optionality at PF. Assuming (38) for Punjabi requires us to assume that while the structure consists of both the Jussive and the allocutive head, the former is left unrealized in the allocutive imperative. However, this is a stipulation in view of existing literature on allocutivity across (un)related languages. It is well-noted across allocutive systems with phi-agreement which include languages such as Basque, Tamil, Magahi and Punjabi, among others, that in the presence of two instances of 2nd person agreement, i.e. 2nd person subject agreement and allocutivity, it is the latter which remains unrealized and not the former. Consider the following declaratives from Magahi in (39)-(40). There is no ban on alloctivity with 1st person subject agreement. However, in the presence of 2nd person subject agreement, allocutivity is disallowed. Similarly, in imperatives, it is allocutivity, which is ruled out in the presence of 2nd person imperative agreement, as in (41).

(40)
Magahi (Alok 2020) ham dauR-l-i-au I run-pfv-1-alloc.nhon 'I ran.' (said to a friend) Alok & Baker (2018) and Alok (2020) argue that these facts follow from a two-probe approach (as in 38) in conjunction with a morphophonological economy constraint. Essentially, it is claimed that the C-T domain of allocutive languages hosts an allocutive probe, which is distinct from the subject agreement probe (T in declaratives, Jussive-T complex in imperatives). The two probes operate independently in syntax, allowing licensing of all subjects with allocutivity. The ban on allocutivity with 2nd person agreement then follows from a morphophonological economy requirement such as Kinyalolo's Constraint, according to which, "in a word (phonologically defined), AGR on one head is silent if and only if its features are predictable from AGR on another head" (see Kinyalolo 1991;Carstens 2005). Following the Person Licensing Condition in Baker (2008), the authors propose that the 2nd person pronoun is a variable bound by the Adr, which acquires its features precisely by being bound by the Adr. By virtue of being in a relationship of variable binding, T-agreement triggered by the 2nd person subject and allocutive agreement triggered by the Adr are essentially the 'same'. This allows only one instance of 2nd person agreement per C-T. Cross-linguistically, this instance corresponds to 2nd person subject agreement, and allocutivity is dropped.
The ban on allocutivity in standard imperatives in Punjabi can be argued to follow directly from the above account. One could argue that in the presence of both imperative agreement and allocutivity in standard imperatives, allocutivity remains unrealized. However, employing the same account also to explain the ban on imperative-ending in the allocutive imperative amounts to saying that there is optionality at the level of PF -in the standard imperative, it leaves the allocutive ending unrealized, which is also the pattern attested in declaratives where allocutivity remains unpronounced with 2nd person subject agreement (as will be discussed in detail in the next subsection). In the allocutive imperative, in contrast, the 2nd person imperative ending is not spelled-out. While this mechanically works, this is a just so story which restates the problem in terms of PF-optionality.
The second problem that this approach faces pertains to the distinct interpretive effects. If the underlying structure of both the allocutive and standard imperatives is the same (as in 38), with variation situated solely at PF, it is unclear as to what explains the ban on generic addressee readings with the allocutive imperative but not the standard imperative. To elaborate, if specific addressee readings ensue from allocutivity in syntax, we should obtain them not only in the allocutive imperative which realizes allocutivity overtly, but also in the standard imperative, where it is left unrealized.
In view of the above arguments, I contend that the account in (38/8a) cannot be assumed for the allocutive imperative in Punjabi. Moreover, as I show in the following sections, Punjabi provides independent evidence in favor of the option in (8b), where the allocutive head replaces the Jussive head to license the imperative subject.
Specifically, I will show that unlike in Magahi, where allocutivity is banned only with 2nd person, Punjabi bans allocutivity with both 1st and 2nd person subject agreement in indicatives. This distribution of allocutivity vis-à-vis 1st/2nd person agreement cannot be explained by the two-probe system in (38). Instead, it is argued to follow from the presence of a unique person probe, which agrees with the Adr, upon failing to agree in person with the subject.
Extending this one-probe analysis to the allocutive imperative means that the C-T consists of a unique person probe. Given obligatory allocutivity, this probe must correspond to the allocutive head, which agrees with the Adr, and not the imperative-specific Jussive head. On the flipside, the sole person probe in the standard imperative corresponds to the Jussive head. Such an analysis not only resolves the issue of stipulated PF-optionality, but also explains the availability of standard imperative in generic contexts easily as ensuing from the complete absence of the allocutive head in the structure.

Distribution of allocutivity with person agreement
This section presents the distribution of allocutivity, focusing on its occurrence vis-à-vis subject agreement in order to better determine the composition of person probes in the C-T domain -does it host two distinct person probes, one for the subject and one for the Adr, or does it host a unique person probe?
As seen previously in section 2, allocutivity in Punjabi can occur across different clause types. It is realized by free morphemes, which occur in place of the tensed auxiliary in the clause-final position after aspect, voice, modal elements and negation. This is shown in (42) The only item that can follow the allocutive marker (or the tensed auxiliary) is the polar question item kii. The PolQ item can occur both in the sentence-initial and final positions, as in (43). Allocutivity precedes the clause-final occurrence of the polar question particle, as shown in (44).

(43)
(kii) karan-ne miraa-nuu kitaab dittii e (kii) polq Karan-erg Mira-dat book give.pfv.f.sg be.prs.3sg polq 'Has Karan given the book to Mira?' karan-ne miraa-nuu kitaab dittii je/e kii Karan-erg Mira-dat book give.pfv.f.sg alloc.pl/be.prs.3sg polq 'Has Karan given the book to Mira?' Allocutivity in Punjabi has a curious distribution vis-à-vis the tensed auxiliary, as I now illustrate. The (person-variable) tensed auxiliary in Punjabi is a stand-alone form, which occurs to the right of the verb. The verb hosts aspectual information, and shows agreement in number and gender but not person. 7 The auxiliary encodes temporal information (present/past), and phi-agreement in person and number. The availability of person-inflected auxiliaries is determined by a conjunction of two factors: (a) case-alignment in the language, and (b) phi-completeness of subject agreement (in contrast with object agreement, which is always minus person).
Punjabi is an aspect based split-ergative language (Bhatia 1993;Deo & Sharma 2006;Bhatt 2007;Kaur 2016;Chandra & Kaur 2017 etc.). Subjects are nominative case valued in the imperfective but ergative/oblique in the perfective (transitive) domain. Only nominative subjects control phi-agreement on the verbal complex. Objects across aspectual specifications receive either an unmarked accusative or an overt differential object marker/DOM as determined by its properties -only the unmarked object triggers agreement. Crucially, only subject agreement triggers agreement in full phi -the verb shows agreement in number and gender, and the auxiliary shows agreement in person and number. Object agreement, on the other hand, is invariable for person. The verb agrees in number and gender, and the auxiliary inflects for number alone. Given this, it is only with 1st/2nd person nominative subjects that occur in the imperfective aspect that 1st/2nd person-inflected auxiliaries obtain. This is shown in (45) In the remaining scenarios, i.e. (a) subject agreement with 3rd person nominative subjects, (b) object agreement in the presence of ergative subjects, which do not control agreement, and (c) default agreement in the presence of ergative subjects and DOM objects, personinflected auxiliaries fail to obtain. Instead, the auxiliary occurs in its 3rd person form (e/ne). Following standard assumption (cf. Harley & Ritter 2002 and others), I treat 3rd person as being minus person. Consider (46) to (47)  The same auxiliary-forms obtain with object agreement in the presence of an (un)marked ergative subject in the perfective domain. This subject fails to control agreement on the verb, which instead agrees with the unmarked object in number and gender. Consider (48), where the unmarked feminine, singular object 'cat' controls agreement on the verbal complex. 9 Crucially, object agreement is not person-variable. 1st/2nd person pronominal objects bear obligatory differential object marking/DOM, -nuu. As a consequence of being marked, there is no possibility for (person) agreement with objects (Kaur 2016).

(48)
asii/ona-ne billii paalii e 1pl.obl/3pl.obl-erg cat.f.sg raise.pfv.f.sg be.prs.3sg 'We/they have raised a cat.' The minus person form e also obtains when both the subject and the object fail to control agreement, resulting in default agreement. Consider (49) with the oblique subject and the DOM object. In such a scenario, neither of the arguments controls verbal agreement, yielding a default (3sg) form of the auxiliary, e.
(48) asii oss billii-nuu paaleyaa e 1pl.obl that cat.f.sg-dom raise.pfv.m.sg be.prs.3sg 'We raised that cat.' The distribution of the present tensed auxiliary forms, in view of (a) case-alignment, and (b) phi-completeness of only subject agreement, is summarized in Table 2. Given the distribution of the auxiliary forms, we find that allocutivity is disallowed with the 1st and 2nd person nominative subjects which must occur with corresponding personinflected auxiliaries, as shown in (50).
1st/2nd person nominative subject *asii/tusii billii paalde je 1pl.nom/2pl.nom cat.f.sg raise.hab.m.pl alloc.pl 'We/you raise cats.' Allocutivity can only occur in configurations which license a minus person auxiliary. This means that allocutivity is grammatical in configurations with 3rd person nominative subjects as in (51), with object agreement configurations as in (52), and in default agreement configurations, as in (53) The distribution of allocutivity vis-à-vis the past tense auxiliary is the same -it cannot co-occur with 1st/2nd person nominative subjects, as in (54). (54) *asii/tusii billii paalde saaN/so je 1pl.nom/2pl.nom cat.f.sg raise.hab.m.pl be.pst.1pl/2pl alloc.pl 'We/you used to raise cats.' Allocutivity is only permitted in configurations which result in a minus person auxiliary. Crucially, unlike in the present tense, where the minus person auxiliary cannot be overtly realized in the presence of allocutivity as in (55), in the past tense the minus person auxiliary and allocutivity can co-occur, as shown in (56) In summary, allocutivity in Punjabi is mutually exclusive with 1st/2nd person agreement. 10 I take this fact to suggest that across tense specifications, the C-T in Punjabi hosts a unique person probe, which can either agree with the 1st/2nd person subject yielding person-inflected auxiliaries, or with the Adr to yield allocutivity but not both. The next section determines the structural location of allocutivity and provides a derivational account. 10 There is one outlier case, where 1st person agreement (but not 2nd person agreement) can co-occur with allocutivity. This pertains to the future indicative, as shown in (i) and (ii).
(i) maiN bajaar jaa-v-aaN-g-aa je 1sg.nom market go-subj-1sg-fut-m.sg alloc.pl 'I will go to the market.' (ii) *tusii bajaar jaa-v-o-g-e je 2pl.nom market go-subj-2pl-fut-m.pl alloc.pl 'You will go to the market.' This seems to refute the one-probe analysis for Punjabi. However, I argue that there is strong motivation to treat the future indicative as a unique case, and not on par with standard present/past indicatives. Consider the rather atypical composition of the future verb in the language. Unlike the present/past indicatives, the future verb does not host a person bearing auxiliary. Instead, the person feature appears attached to the subjunctive morpheme, which is located before the future morpheme -g-. Following the observation by Butt & Lahiri (2013); also Butt & Poudel (2012), I argue that this is due to the development of the modern future structure from a bi-clausal configuration -present-day gaa-marked constructions descend from a periphrastic future construction, in which 'go' took a complement clause. Examples of such periphrastic future constructions are relatively common (see Bybee et al. 1994), and can be observed, for example, in present-day English (iii).
(iii) I'm going to eat.
The modern future structure in HU/Punjabi has developed from a similar structure like in (iii), except the matrix predicate selects a subjunctive clause. Given this bi-clausal structure, it is possible to posit two agreement loci -one in the matrix clause, and another in the person-inflecting embedded subjunctive clause.

(iv) [IP [Pers[Pers:_] [gaP[NG:_] [IP [PersP[Pers:_] [subjMoodP [AspP [vP…]]]]]]]
I claim that both of these person agreement loci exist in the reanalyzed form. However, the higher agreement locus remains unrealized. Since the person forms/auxiliaries are fused with tense markers and tense is not semantically compatible with the future (Kush 2011), the higher PersP remains unvalued (and unrealized), as shown in (v) for Punjabi-same facts hold for HU.
(v) maiN baazaar jaa-v-aaN-g-aa (*aaN) 1sg.nom market go-subj-1sg-fut-m.sg be.prs.1sg 'I will go to the market.' However, the allocutive auxiliary in Punjabi is non-tensed, and only bears phi-features. It can therefore occupy the matrix Pers 0 slot, resulting in the co-occurrence of allocutivity with 1st person subject agreement in the future. Therefore, I maintain that the C-T in Punjabi has one person probe. The future should be treated as deviant -it has two probes, as motivated by its historical development.

Deriving allocutivity-one person-probe analysis
To analyse the above facts, I propose that the Punjabi clause structure has a unique person probe situated at Pers 0 above T 0 . Allocutivity obtains upon agreement with this probe at Pers 0 but only when it fails to be valued by the subject in the first agreement cycle (see Kim 2019 for a similar proposal for the verbal suffix -si in Korean which alternates between subject honorification and allocutive agreement).
There are two crucial components to the analysis. First, I follow Miyagawa (2012); (2017) in treating allocutivity as agreement in the 2nd person feature between a probe and the Adr. Like 2nd person pronouns, the Adr occurs with 2nd person, number, gender and honorificity features, which can also be realized on the probe. 11 Secondly, the specific implementation of allocutivity I employ here involves two agreement mechanisms that are well-motivated in syntax: (i) (phase-bound) bidirectional probing, and (ii) sequential agreement with multiple goals, i.e. Cyclic Agree. Bidirectional probing, as is wellmotivated for phenomena such as switch reference, adjectival agreement, polyvalent case (Merchant 2006;Baker 2008;Arregi & Hanink 2018, among others), allows a probe to seek a goal in either direction and agree with it. For instance, the embedded C in switch references systems like Washo probes both downward to agree with the embedded subject and also upward to agree with the matrix subject (Arregi & Hanink 2018). See the schema in (57), where C 0 undergoes agreement in two directions simultaneously.
Note that there are two logical possibilities for bidirectional probing -the probe can target both goals in the same derivational step and hence manifest Multiple Agree (Hiraiwa 2001), as is the case for the probe at C 0 in Washo, or it can target the two goals sequentially and exhibit Cyclic Agree (e.g. Béjar & Rezac 2009;Despić et al. 2019, among others). I propose Cyclic Agree for the probe at Pers 0 in Punjabi. This probe can target two 11 A reviewer notes that cross-linguistically, facts show that allocutive agreement may reflect number/gender/ (non)-honorificity feature of the addressee but never the person feature (following McFadden 2017). This makes it unclear as to how does the unvalued person probe play a role in allocutive agreement. I note that the lack of person feature in allocutive morphology across languages is not an established fact. It is indeed the case that in Tamil, the allocutive marker is missing the person agreement morpheme. It is realized as -ŋgae, and not the 2nd plural agreement form -iiŋgae. However, this does not seem to be the case cross-linguistically. The most telling case is that of Basque, where allocutive agreement, and 2nd person singular and familiar ergative agreement are homophonous (Hualde et al. 2003;Haddican 2018 etc.). The 2sg familiar, masculine ergative agreement as well as allocutive agreement is realized as -a-/-k, while the corresponding feminine forms for both paradigms are -na-/-n. The facts in Basque raise a more general issue for those who assume the absence of 2nd person specification on the allocutive marker: if not via a 2nd person feature, how does the allocutive marker encode the addressee specification? In other words, what featural specification on the allocutive probe dictates that it must target the Adr (in contrast with other arguments)? In view of these issues, I adopt the standard assumption and treat 2nd person feature as encoding the addressee specification (Harley & Ritter 2002;Zanuttini 2008 etc.) Moreover, while cross-linguistic extensions remain to be worked out, my account of allocutivity for Punjabi provides an interesting solution to the above problem. Unlike Basque, the allocutive marker and 2nd person subject agreement in Punjabi are distinct. Also, since the forms fuse various features, one cannot tell if the allocutive marker does or does not contain the 2nd person feature. For those who advocate the lack of 2nd person feature on the allocutive marker, the issues discussed earlier prevail. However, assuming a 2nd person feature, in conjunction with Cyclic agree solves the problem. In my system, allocutivity is the result of second cycle agreement, which means that the distinct form of the allocutive marker (in contrast with 2nd person subject agreement) does not necessarily underlie a distinct featural make-up. Instead, it could be described as an instance of second-cycle effects, which essentially correspond to contextual allomorphy that results from agreement between an argument (say X) and a probe on a different cycle, as compared to agreement between another argument Y and the same probe (cf. Béjar & Rezac 2009). arguments. Crucially, it searches its domain one argument at a time -it looks downward first and attempts an agreement relation. If the first instance of probing satisfies the probe, it bleeds probing of all additional goals. Only upon failure of the first agreement cycle, the second cycle of agreement is initiated -this instance of agreement is upward. Putting this technical machinery together, allocutivity in Punjabi will be analysed as second-cycle agreement of the Adr with the unique person probe in the C-T which fails to agree with the subject in the first agreement cycle. But before we proceed to do this, I establish the structural location of the allocutive marker in the language.
I propose that je occupies the Pers 0 head, separate from T 0 which hosts temporal information. Support for the presence of two heads, where one hosts temporal information but not person and the reverse holds for the other head, comes from two facts: (a) the decomposition of the forms of the 1st and 2nd person agreement auxiliaries in the language, and (b) the co-occurrence of the past tense auxiliary and the allocutive marker.
As shown in Table 3, 1st and 2nd person agreement forms can clearly be decomposed into the phi-agreement morpheme corresponding to -aaN (1sg,1pl), -aiN (2sg) and -o (2pl), and the tense morpheme realized via ∅ and s-for the present and the past tense respectively. This decomposition suggests that tense is located on a distinct syntactic head than the person and number feature bundle. More evidence is provided by the co-occurrence of the past tense auxiliary and the allocutive marker. Recall that in the past tense, which is realized overtly via s-, je can co-occur with the 3rd person past tense auxiliary form (as shown in ex. 56 previously). The co-occurrence of two auxiliary forms (sii/sann and je) clearly attests to the presence of two syntactic heads. The head spelled out as sii/ sann encodes temporal information realized via the affix s-. The Pers 0 head in contrast does not encode temporal information -it bears the person feature and is realized as je. 12 In view of the above facts, I propose that the C-T domain in Punjabi consists of two adjacent heads T 0 and Pers 0 , where T 0 hosts temporal information while the person probe is situated on Pers 0 . 1st and 2nd person tensed auxiliaries spell out the two adjacent heads together. The 3rd person auxiliary only spells out the minus person T head, leaving je to spell out Pers 0 .
With these assumptions in place, I propose the following analysis of allocutivity in Punjabi. T 0 and Pers 0 probe independently. The probe at T 0 agrees with the subject -this agreement relation values nominative case on the subject. As for the person probe at Pers 0 , it has two possible sources for its valuation. Assuming Cyclic Agree, the probe first looks 12 While we focus on only the tense and person distribution in this paper, it appears that the number probe is present on both heads in the T-Pers complex. With a 1st/2nd person subject, both the probes at T and Pers are valued by the subject. In the presence of a 3rd person subject, there are two possibilities -(i) the probe at T agrees with the 3rd person subject in number, and the probe at Pers remains unvalued since the subject is minus person, or (ii) the probe at T agrees with the subject, and the probe at Pers agrees with the Adr -both instances of agreement are realized overtly in the past tense. It is crucial that the entire phi-probe on Pers is valued by the same goal. Thus, if the person probe agrees with the 1st/2nd subject, the number probe must also copy the features of the subject. In the presence of a 3rd person subject, which cannot value the person probe, both person and number probes on Pers 0 are valued by the Adr. For ease of exposition in this paper, however, I only illustrate the person feature on Pers 0 (and not number).
in its c-command domain. In the presence of 1st and 2nd person subjects, the person feature on the probe is checked, and is labeled as Pers subj . This obligatory person agreement is motivated by the Person Licensing Condition (Béjar & Rezac 2003), as given in (58).
(58) Person Licensing Condition (PLC): An interpretable 1/2 feature must be licensed by entering into an Agree relation with a functional category. (Béjar & Rezac 2003: 53) Obligatory person agreement with the 1st and 2nd person subject bleeds the next cycle of agreement with the Adr, ruling out allocutivity. Consider the schema in (59). With 3rd person nominative subjects, the (number) probe at T 0 is valued by the subject, which is valued as nominative by T 0 in return. As for the person probe, it probes downward to agree with the 3rd person subject, but remains unchecked since 3rd person is minus person. This triggers a second cycle of agreement where Pers 0 looks upward to target the Adr and yields allocutivity, as in (61) (13)-(16), allocutivity is optional in that the auxiliary can occur in its default form, which means that the person probe can remain unvalued for person. This is not an issue if one assumes with Preminger (2011) that phi-features on functional heads can remain unvalued without causing a derivational crash, as long as agreement is attempted. With 3rd person subjects, the person probe attempts agreement with the subject. However, since the subject is 3rd person, and hence person-less, the person feature cannot be valued. This results in the insertion of a default form. The second question, pertaining to the choice between the default auxiliary and allocutivity is unresolved at the moment. The presence of allocutivity in declaratives is pragmatically conditioned -while preferred in honorific contexts, it is formally optional. The reviewer is right in noting that this optionality is strange within an agreement based analysis since agreement is generally obligatory when the factors that permit it are in place. Thus, for instance, subject agreement in Punjabi is obligatory and cannot be omitted in the same way that allocutivity can be. There are two plausible ways to encode this optionality. First, optionality is the result of allocutivity being second-cycle agreement while subject agreement obtains from the first cycle of agreement. Alternatively, optionality of allocutive agreement could plausibly be related to properties of agreement in the C-domain. For instance, in her excellent overview of complementizer agreement/CA, Van Koppen (2017) notes that CA too is optional for speakers of Bavarian (cf. Mayr 2010), and a dialect of Flemish (Vanacker 1949), among others.
In such configurations, T 0 is valued by the subject while the person probe at Pers 0 is valued by the Adr. Features on the two heads are spelled out separately, as is evident from the past tense. In the past, features on T 0 are realized overtly, while in the present tense, they remain unrealized. The person probe on Pers 0 valued by the Adr is realized as je.
(62) a. T pres ⟷ ∅ b. T past ⟷ sii/sann (63) Pers alloc ⟷ je It must be noted that since allocutivity results from the second-cycle of agreement on the person probe, the distinct 2nd person morphology of allocutivity (je) as compared to 2nd person subject agreement (-o) in Punjabi is not surprising and can be treated as an instance of second-cycle effects (in the sense of Béjar & Rezac 2009).
In summary, allocutivity does not underlie a distinct allocutive probe in Punjabi. The C-T in Punjabi hosts a unique instance of the person feature at Pers 0 , which has two possible sources of valuation -the 1st/2nd person subject or the Adr but not both. In a given derivation, only one of the potential goals triggers agreement. If the valuing goal is the subject, we get subject agreement. If Pers 0 targets the Adr upon failure to be valued by the subject, it bears allocutivity.

Independent evidence for the one person probe system
In this section, I provide independent evidence for the one person-probe system from a subject-object interaction in Punjabi. Punjabi has a 3rd person animate clitic suu, which in its object occurrence is disallowed with 1st/2nd person subjects (Akhtar 1997;Butt 2007;Kaur 2017;Kaur & Raynaud 2019). Following Kaur (2017), I show that this ban results from the presence of a unique person probe in C-T, which can either license the 1st/2nd person subject, or the object clitic, which ends up in the same agreement domain as the subject.
The 3rd person pronominal clitic suu occurs post-verbally to replace a 3rd person singular argument. Consider the following example in (64), where suu, occupying the position of the auxiliary, replaces a 3rd person subject. Similarly, in (65) In its subject occurrence, suu is unaffected by the person feature on the object, which can be 1st/2nd or 3rd. Objects do not control person agreement in the language. However, in its object occurrence, suu manifests person effects, i.e. it can replace a non-agreeing 3rd person object but only when the subject is also 3rd person, as seen in (65). The presence of a 1st/2nd person subject with an object suu results in ungrammaticality, as shown in the following example in (66). The sentence is grammatical only with corresponding 1st/2nd person agreement auxiliaries. 14 (66) maiN/tuu vekhdaa aaN/eN/*suu 1sg.nom/2sg.nom see.hab.m.sg be.prs.1sg/be.prs.2sg/*3sg.cl 'I/you see him/her.' An important thing to be noted here is that only the 3rd person clitic pronoun suu is subject to the person effects. There is no incompatibility between a 3rd person full object pronoun and an agreeing 1st/2nd person subject. This pattern with 3rd person clitic pronominal object suu, which occupies the auxiliary position, is akin to the ban on allocutivity with agreeing 1st/2nd arguments. Examining the properties of suu, these facts follow from a one-probe system. The analysis in a nutshell is as follows: like 3rd person dative indirect object clitics in PCC languages, suu in Punjabi can only reference animate entities and can be argued to bear a person feature, albeit negatively specified (cf. Adger & Harbour 2007, among others). Punjabi is a DOM language (Bhatia 1993;Kaur 2016), which requires 3rd person animate objects (among 1st/2nd person pronouns and reflexives) to undergo object shift. This triggers the movement of the object clitic to the edge of vP. However, since suu unlike full pronouns is morphologically deficient (in the spirit of Cardinaletti & Starke 1994;Déchaine & Wiltschko 2002;Roberts 2010 etc.), it cannot be licensed via the PP based DOM within vP and targets the agreement based licensor, i.e. T-Pers complex, which is also the locus of subject licensing. This results in a ban on suu in the presence of 1st/2nd person subjects.
Despite its occurrence in the position occupied by the auxiliary, suu is a pronominal clitic. First, unlike agreement markers which do not restrict the co-occurrence of the argument that they agree with, object-indexing suu can never co-occur with an overt object.
(68) karan-ne o-nuu i vekhyaa e/*suu i Karan-erg 3sg-dom see.pfv.m.sg be.prs.3sg/3sg.cl 'Karan saw him/her.' Secondly, it has been observed that agreement typically is not restricted in terms of the semantic properties of the goal (Corbett 2006). In contrast, clitics exhibit semantic restrictions with regard to the argument that they can co-reference (Jaeggli 1982;Suñer 1988;Kramer 2014, among others). Patterning with clitics (albeit without any clitic-doubling), suu exhibits semantic restrictions in that it can only reference animate and specific DPs. Using it to refer to an inanimate entity is infelicitous.
14 Like je, suu is grammatical with person-specified agreement in the future marked sentences.

(69)
Context: Karan has broken a statue. The speaker points at the broken statue and says … #karan-ne toR dittaa suu Karan-erg break give.pfv.m.sg 3sg.cl 'Karan has broken it.'

(70)
Context: Karan has hit a cat. The speaker points at the cat and says … karan-ne maareyaa suu Karan-erg hit.pfv.m.sg 3sg.cl 'Karan has hit her.' No such (in)animacy-sensitivity is observed for the agreement marker, which agrees with the unmarked object, regardless of its (in)animacy.
(71) karan-ne muurtiyaaN/billiyaaN vekhiyaaN ne Karan-erg statue.f.pl/cat.f.pl see.pfv.f.pl be.prs.3pl 'Karan has seen the statues/cats.' The above diagnostics show that suu patterns with a pronoun and not an agreement marker. Moreover, it is also animate, in contrast with inanimate pronouns like 'it' in English. However, suu is distinct from the full 3rd person pronoun in that it behaves like an X 0 as is evidenced by (a) its inability to be focused, and (b) fixed position in the verbal complex. Akhtar (1997) notes that the arguments suu stands for are unstressed and not the focal part of the sentences. This can be seen directly from the inability of suu to stand for wh-words.
(72) a. karan-ne ki-nuu maareyaa Karan-erg who-dom hit.pfv.m.sg 'Who did Karan hit?' b. karan-ne maareyaa suu Karan-erg hit.pfv.m.sg 3sg.cl 'Karan hit him/her.' but not 'Who did Karan hit?' Furthermore, unlike the full pronoun, which can be scrambled since Punjabi, like Hindi-Urdu allows scrambling with information-structural effects (cf. Kidwai 2000 etc.), suu must occur in the clause-final position replacing the auxiliary. (74) (*suu) karan-ne (*suu) maareyaa suu 3sg.cl Karan-erg 3sg.cl hit.pfv.m.sg 3sg.cl 'Karan hit him/her.' I take the above mentioned behavior of suu to indicate that it is an X 0 /XP item (in the sense of Chomsky 2001). Put together, these properties show that suu is a pronominal clitic, which needs licensing akin to 3rd person full object pronouns. However, unlike a full pronoun, it is structurally deficient. Question: what is the nature of this licensing?
Punjabi is a DOM language, which requires 1st/2nd person object pronouns, proper names and reflexives to obligatorily occur with DOM, realized as -nuu, as in (75)  With 3rd person pronouns, the presence/absence of the marking correlates with interpretive effects. The animate 3rd person full pronoun in the object position requires obligatory differential object marking/DOM, as shown in (77) Following recent work by Kalin (2017); (2018), I treat DOM as the manifestation of a licensing requirement on certain object nominals in a given language. Restricting the discussion to pronouns, I propose that the [person] feature on object pronouns must be licensed via DOM. Given the interpretive effects associated with DOM on 3rd person pronouns, not only 1st/2nd person objects but also 3rd person full animate pronominal objects enter the derivation with a person feature, albeit negatively specified (Adger & Harbour 2007). The presence of the person feature on the object triggers object shift to the edge of vP. Clear evidence for this movement is provided by the position of -nuu marked direct object/DO vis-à-vis the indirect object/IO in ditransitives (Bhatt & Anagnostopoulou 1996;Kidwai 2000;Kaur 2016 for more evidence). In ditransitives, the IO typically precedes the DO in Punjabi, (80). However, in the presence of a DOM bearing DO, this order must reverse, (81).  Akin to full object pronouns, suu originates in the complement of V. Due to the presence of a positively specified 3rd person feature, it raises to the inner specifier of vP and 'tucksin' below the subject (in the sense of Richards 1999 etc.). However, since it is defective, it cannot host the adpositional DOM realized as -nuu. Consequently, it targets the only functional head with a person probe, i.e. Pers 0 . As schematized in (83), Pers 0 bearing the person feature probes its c-command domain and agrees with the object suu for person, which then adjoins to the T-Pers complex.

(84) *[T-Pers [Pers:1/2] [vP 1st/2nd subj [Pers:1/2] [vP obj-suu [Pers:3] [VP… V]]]]
If the C-T domain in Punjabi had two person probes, both a 1st/2nd person subject and 3rd person animate object clitic would have been permitted, contra facts. This subjectobject interaction clearly demonstrates that Punjabi has a unique person probe in the C-T, which can either agree with a 1st/2nd person subject, a 3rd animate object clitic or the Adr. 17 This is summarized in Table 4.

Interim summary
This section has argued that the distribution of allocutivity vis-à-vis both 1st and 2nd person subject agreement in Punjabi cannot be explained by the presence of two probes, which predicts 1st person agreement to be allowed with allocutivity, contra facts. Instead, this distribution is better explained from the presence of a unique person probe. Allocutivity obtains upon agreement with this probe, but only when it fails to be valued for person by the subject. 16 I depict minus person 3rd person items (including 3rd person subjects) as lacking the person feature altogether. In contrast, both full and clitic 3rd person animate object pronouns are shown as carrying a [Pers:3] feature. The presence of this feature requires licensing, akin to the licensing of 1st/2nd person feature. 17 Unlike je, suu is not permitted in the past tense. This could potentially be due to the similar phonological forms of the minus person past auxiliary (sii/sann) and suu. In the present tense domain, I analyze it identically to je in that T 0 has null exponence, and suu like je, occupies Pers 0 .
Extending this syntax of allocutivity (and the C-T) to imperatives, the facts in the allocutive imperative follow. Allocutive imperatives obligatorily host the allocutive marker. Given our analysis of allocutivity, this means that the imperative structure must host a unique Pers alloc , which forms a complex with the minus person T 0 . Moreover, the oneperson probe approach also requires that in the presence of Pers alloc , no other person probe is available in the C-T domain to agree in person with the subject and license it. This rules out the co-occurrence of Jussive head which bears a 2nd person feature and agrees with the imperative subject to license it.
Given the independent support for the one-probe analysis and the problems pertaining to (a) stipulated optionality at PF, and (b) inability to explain interpretive effects that (38/8a) raises for allocutive imperatives in Punjabi, this section has convincingly shown that the allocutive imperative in Punjabi is best analysed via (8b), which lacks a Jussive head. Instead, the functional projection which hosts allocutivity across clause types, i.e. Pers alloc , replaces the Jussive head to derive addressee-restriction on imperative subjects.

Explaining the ban on generic addressee readings
Having determined that the locus of 2nd person feature in the allocutive imperative is not the Jussive head, but instead the allocutive head, the next question pertains to the difference in addressee readings across standard and allocutive imperatives. This section follows Portner et al. (2019) to claim that the ban on generic addressee readings in allocutive imperatives follows from an extra encoding of the [status] feature on the allocutive head, in contrast with the Jussive head which hosts only a 2nd person feature, but not a [status] feature. Crucially, however, it differs from Portner et al. (2019) to argue that the allocutive marker in Punjabi does not realize c, the head associated with performative meaning in their system.
The division in (non)-specific addressee uses seen in Punjabi imperatives is not unique and has been discussed for Korean imperatives (Pak 2015;Portner et al. 2019). Korean is an allocutive language which encodes the speaker's relation with the addressee via speech style particles/SSPs (Martin 1992;Sohn 1999;Kim-Renaud & Pak 2006;Pak 2015). The language has six SSPs: formal, polite, semiformal, familiar, intimate, and plain. Except for the plain SSP, all particles invoke a specific addressee. Consider the imperative in (85) with a polite SPP, which can only be uttered to an honorific addressee, signaling the identification of a specific addressee in turn.

(85)
Korean (Pak 2015: 139) Kamki keli-si-ess-unikka swui-si-eyo cold catch-hon-pst-since rest-hon-imp.pol 'Since you have a cold, rest!' The plain SSP, however, is unique in that it may occur both in contexts with and without a specific addressee. It has two forms in its imperative occurrence -(u)la or -ela/-ala, which are not in free variation. -(u)la can occur only when the imperative is used for a generic addressee, as in mottos, rallies etc. For illustration, see (86a). Its usage in contexts with a specific addressee, as in (86b), where a teacher is talking to a student, is infelicitous. Only -ela/-ala can be used in such contexts.

(86)
Korean (Portner et al. 2019: 8-9) a. Na-lul ttala-o-la! 1sg-acc follow-come-imp 'Follow me!' (Rally cry) b. Na-lul ttala-o-#la/ala! 1sg-acc follow-come-imp 'Follow me!' (Specific student-teacher interaction) Portner et al. (2019) analyze this difference as resulting from the extra encoding of "politeness". Building on existing studies which encode speech act participants in syntax, the authors postulate a dedicated cP in the clausal periphery, which corresponds to the layer of structure that interfaces with the context of utterance. The meaning of c consists of both the propositional content, and the politeness meaning. Specifically, the c head bears a meaning that encodes the relation between the speaker and the addressee. This meaning is modeled as a [status] feature. Building on Kim-Renaud & Pak (2006), the [status] feature has values that represent an ordering (or hierarchical) relation (<, =, >) between speaker (Sp) and addressee (Adr). Linguistic forms that encode (im)politeness are in turn treated as realizations of c. Given this account, the difference in specific/generic readings in Korean imperatives follows directly. Assuming the clause structure in (87), where c hosts the [status] feature, Portner et al. (2019) propose that speech style particles in Korean are a manifestation of c. The [status] feature on c in combination with the 2nd person feature on the Jussive head in imperatives is spelled out as -ela/-ala. On the other hand, -(u)la is a simplex form, which only realizes the Jussive head. Since -ela/-ala encodes the [status] feature, it is restricted to contexts with a specific addressee, in contrast with -(u)la, which lacks it and can thereby appear only in contexts with an addressee whose relation with the speaker cannot be defined. To explain the ban on generic addressee readings in Punjabi allocutive imperatives, I follow Portner et al. (2019) to claim that these effects obtain due to the obligatory extra encoding of the [status] feature in the allocutive marker since it must always define a speaker-addressee relation, regardless of the clause type. In contrast, the standard imperative is composed of the Jussive head alone, which only bears a 2nd person feature but not the [status] feature. However, I differ from Portner et al. (2019) and show that unlike the SSPs in Korean, which are spelled out at c, the allocutive marking je is not realized at c; instead, it is realized lower in the structure at Pers 0 -in addition to its person specification, Pers 0 also hosts an unvalued [status] feature. Since je obligatorily spells out the [status] feature it is restricted to contexts with a specific addressee. Based on the interaction between subject agreement and allocutivity in Punjabi, the previous section has already established that je is realized at Pers alloc located in the T-region, and not at c. More evidence for this claim comes from embedding facts. Portner et al. (2019) provide an interesting correlation between the locus of syntactic realization of an utterance-oriented particle (SSPs, allocutive markers etc.) and its (un)embeddability. Specifically, the authors argue that utterance-oriented particles that are syntactically realized at c cannot be embedded. Consider the following example from Korean. The language allows embedding of imperatives, but only when they occur with the plain SSP -(u)la, which does not encode the [status] feature, and is thereby not associated with c. In contrast, the complex SSP -ela/-ala, by virtue of realizing c, cannot occur in embedded domains.

(88)
Korean (Portner et al. 2019: 8) Yumi-ka Inho-hanthey [choysen-ul ta ha-la/*ela-ko] malha-ess-ta Yumi-nom Inho-to best-acc all do-imp-comp say-pst-dec.plain 'Yumi told Inho to do his best.' The authors take the above facts to follow from the type of meanings that are encoded at c. Essentially, cP is taken to host non-propositional, performative meaning that cannot be represented as an individual. However, following Rosenbaum (1967) and Kiparsky (1970), argument clauses are, in some respect, more 'nominal' than root or adjunct clauses. Since the cP cannot have the nominal character required for a phrase to serve as one of the arguments of a lexical functor such as a verb, it cannot be embedded.
Examining the correlation between the realization of c and its (un)embeddability for Punjabi, we find that the allocutive marker in Punjabi can be embedded. As Kaur & Yamada (2019) show, Punjabi allows embedded allocutivity in finite domains, restricted to the complement clauses of speech predicates such as 'tell', 'say', 'speak', 'ask'. Consider the following example for illustration. 18 (89) karan-ne keyaa ki mira kal aayegii je Karan-erg say.pfv.m.sg that Mira tomorrow come.fut.f.sg alloc.pl 'Karan said that Mira will come tomorrow.' Evidence for the embedded structure being a true instance of indirect speech comes from the possibility for a question word associated with the embedded clause to take scope over the matrix clause to form a direct question (following diagnostics used in Anand & Nevins 2004;Crnič & Trinh 2009, among others).
(90) kidde-naal karan-ne keyaa ki [mira aayegii je] who.gen-with Karan-erg say.pfv.m.sg that Mira come.fut.f.sg alloc.pl 'With whom did Karan say that Mira will come?' Thus, the availability of embedded allocutivity in Punjabi further confirms that the allocutive marker is not realized at c. Instead, je spells out the Pers alloc head. 19 We can now model the ban on generic readings of the addressee in allocutive imperatives. I claim that Pers alloc enters the derivation with not only an unvalued person but also an unvalued status feature. The person feature is valued by the Adr. I also situate the status feature on the Adr. 20 Upon agreement with the Adr, Pers alloc ends up with not only 18 In the Punjabi variant under consideration, embedding of imperatives with allocutivity seems marginally allowed too. The same observation holds for standard imperatives. Regardless of clause-type, there is variation in the availability of embedded allocutivity. Three Punjabi speakers of Lahore variety (Pakistan) who were consulted did not permit allocutivity on the embedded verb. 19 It must be noted here that the possibility to embed allocutivity is not unique to Punjabi. While some languages like Basque, Jingpo etc. restrict allocutivity to root contexts, recent studies on Tamil (McFadden 2017), Magahi (Alok 2020) and Japanese (Kaur & Yamada 2019;Yamada 2019) have shown that allocutivity can occur in embedded domains in these linguistic systems. 20 The possibility for the status feature to be situated at Adr is also entertained by Portner et al. (2019), at least in contexts with an overt vocative. However, I suggest that regardless of the overt vocative, there is evidence in favor of the [status] feature being located on the Adr. Embedded allocutivity in languages like Punjabi, Magahi, Tamil etc. can show indexical shift, which means that it can refer to the reported and not the utterance addressee. In such instances, the allocutive marking on the embedded verb does not match the allocutive marking on the matrix verb. Consider (i) from Magahi, where the embedded allocutive marker encodes the non-honorific reported addressee, while matrix allocutivity encodes the high-honorific utterance addressee.
the person specification of the addressee, but also its (non)-honorific status vis-à-vis the speaker. Since je spells out this entire feature bundle consisting of 2nd person and (non)honorific status features, its presence in the allocutive imperative restricts the imperative to contexts with a specific addressee. Recall from section 2.2 that je is a plural form, used for both honorific as well as nonhonorific addressees. In addition, je can also be used as a singular form -it only refers to an honorific addressee in this case.
To explain the above-mentioned usages of je, I follow the analysis of the French 2nd person plural pronoun vous as proposed in Ackema & Neeleman (2018). Like je, vous can be employed for a plural addressee who may be honorific or non-honorific. Additionally, vous can be employed for a singular honorific addressee. Ackema & Neeleman (2018) propose that in syntax, vous consists of the following featural specification: [2 PL HON]. 21 To explain the availability of the singular, honorific addressee reading, the plural number feature on vous is argued to get deleted at LF, in the presence of honorific feature value, as in (92). This deletion in (92) can be understood as the LF counterpart of the notion of impoverishment, which takes place at PF (Nevins 2008). It yields a 2nd person pronoun, which is underspecified for number, and is therefore compatible with both a singular and a plural honorific addressee reading. Support for the claim that vous has a plural feature in syntax, which is deleted only post-syntactically comes from plural agreement on the verb with vous even when it refers to a singular honorific addressee.
(93) French (Wechsler 2004: 255) Vous êtes/*es loyal 2pl be.2pl/be.2sg loyal.m.sg 'You (one male addressee, polite) are loyal.' (i) Magahi (Alok & Baker 2018) Santeeaa Banteeaa-ke kahk-ain ki Ram-ke Sita-se baat karek-e chah-au Santee Bantee-dat told-alloc.hhon that Ram-dat Sita-ins talk do-inf should-alloc.nhon 'Santee told Bantee that Ram should talk to Sita (said to a teacher).' The example in (i) poses a challenge for a c-based account. Given that c cannot be embedded given its special semantics, one is forced to posit a unique instance of c in the matrix clause. This root c could be argued to value the [status] feature on all instances of allocutivity across matrix and embedded domains. However, this predicts obligatory matching between the features of embedded and matrix allocutivity, which is not borne out. In contrast, recent studies (Bhadra 2018;Alok 2020) have shown that there are null DP coordinates corresponding to 'speaker' and 'addressee' in every finite clause. If the above-mentioned representations of the addressee can be shown to host their own [status] feature, the facts in (i) can be explained. Since each of the addressee representations hosts its individual [status] feature, the embedded Adr feature can shift independently to encode the features of the reported addressee realized as the indirect object of the matrix verb. The matrix Adr, in contrast, encodes the properties of the utterance addressee. This allows them to have distinct specifications of allocutivity. While this analysis remains to be evaluated further, the facts in (i) raise doubts about situating the [status] feature on the unembeddable c. I therefore posit it on the Adr itself. 21 Ackema & Neeleman (2018) follow a distinct featural system for person composed of [PROX] and [DIST] features, but nothing is this paper hinges on that. I therefore employ the more conventional system, consisting of 1st and 2nd person as values for the feature attribute [person]. I follow the same proposal for je -when the allocutive form consists of the featural specification [2 PL HON] in syntax, the plural number feature on je can get deleted at LF, yielding a feature bundle, which is underspecified for number but specified for honorificity, i.e. [2 HON]. This feature bundle is spelled out as je, accounting in turn for its usage as an honorific form, both for singular and plural addressees. In contrast, when the featural specification of je in syntax is [2 PL NHON], no such deletion of the number feature happens. This prevents je from being used as a singular non-honorific form. Instead, Punjabi employs the dedicated singular allocutive forms ii/aa to refer to a singular non-honorific addressee.
In summary, je spells out the following feature bundles given in (94). Since its featural composition always consists of not only phi-featural specification, but also status specification, it can only refer to a specific addressee.

Deriving the allocutive and standard imperatives
With Zanuttini (2008) and Zanuttini et al. (2012), this paper has shown that all imperatives host a syntactic representation of the addressee, which agrees with the subject. However, employing the allocutive imperative in Punjabi, this paper has also demonstrated that the addressee representation does not have to correspond to the imperativespecific Jussive head. The allocutive imperative in Punjabi, characterized by (i) the lack of imperative-specific ending and (ii) restriction to contexts with a specific addressee, has been argued to lack the Jussive head. Instead, Pers 0 , which undergoes agreement with the Adr to yield allocutivity (Pers alloc ) replaces the Jussive head to derive addressee-restriction in the allocutive imperative. Moreover, its featural specification consisting of not only an unvalued person but also an unvalued status feature derives specific addressee readings.
In view of these findings, this section models the derivation of imperatives in Punjabi.
To encode imperative meaning, I follow Han (1998); and Medeiros (2015), among others in assuming that imperative syntax hosts an illocutionary force operator in C 0 . This operator associates an imperative sentence with an illocutionary force, which in turn allows a speaker to use the sentence to perform a speech act. Specifically, the operator is a function that takes a proposition (p) and turns it into a "directive action".
By performing a directive action, the speaker instructs the addressee to update the Plan Set, a separate discourse component which is a set of propositions that specifies the addressee's intentions to bring about p. Thus, an imperative is a clause-type, which is an instruction to the addressee to add p to his/her plan set. 22

Standard imperative
Consider an instance of the standard imperative composed of the bare verb stem with 2nd person endings, as shown in (97) (repeated from ex. 12).
(97) kitaab paRh-o book read-imp.2pl 'Read the book!' The syntax of the standard imperative hosts a C head with an imperative operator -I designate this via C imp . The C imp occurs with an irrealis T complex. Recall that the T complex in Punjabi consists of two heads T 0 and Pers 0 . The imperative T head has an irrealis feature. 23 Following Zanuttini (2008) and Zanuttini et al. (2012), it also has a case feature to license the subject but crucially no person. The Pers 0 in imperatives is a special imperative head, on par with the Jussive head (Zanuttini 2008), which enters the derivation with a 2nd person feature. 24 Crucially, it hosts only a 2nd person feature but not a [status] feature, and is labeled as Pers imp . In line with the oneprobe system advocated in this paper, Pers imp is the only person probe in the imperative C-T domain. There are two empirical upshots to this claim. First, in the absence of an additional person probe with an unvalued person feature which can target the Adr to yield allocutivity, we correctly rule out the co-occurrence of allocutivity with imperative-specific agreement in standard imperatives. Secondly, in the complete absence of the allocutive head, the availability of standard imperatives in contexts with a generic addressee also follows. Pers imp hosts only a 2nd person feature, and there is no allocutive head with an unvalued [status] feature to be valued by the Adr, which can restrict the structure to specific addressee contexts.
Compare this with the two-probe approach in (38/8a). According to this approach, the structure of the standard imperative in Punjabi would host both the Pers imp head, which would agree with the subject, and the Pers alloc head to agree with the Adr. In the presence of both imperative and allocutive agreement, the latter is ruled out by morphophonological economy constraints. Assuming the two-probe approach for the standard imperative would compel us to posit a non-uniform system -standard imperatives use a two-probe system with an active morphophonological constraint while allocutive imperatives and declaratives underlie a one-probe approach. Furthermore, the presence of an allocutive head with a [status] feature would value the [status] feature (via agreement) in syntax, like in the allocutive imperative. It is unclear as to why the covertness of allocutivity at PF in the standard imperative should have a distinct semantic-pragmatic effect as compared 22 While I assume Han's account for clearer exposition of clause-typing and the featural composition of the C-T, the proposed account of addressee-restriction can be adapted to other existing approaches to imperative meaning. 23 I treat this instance of T as lacking any temporal specification -the presence of [irrealis] feature on T captures the intuition that the proposition describes an unrealized situation. It is in principle possible to replace the irrealis T with a Mood head. 24 It is possible that the Pers imp /Jussive head enters the derivation with an unvalued person feature but inherits this feature from the Adr during the derivation, as suggested in Zanuttini (2008). However, the mechanism for this inheritance is left unclear -is the Jussive head, bearing an unvalued person feature, relativized to agree upward such that it targets the Adr as soon as it enters the derivation. Or does it pattern akin to the allocutive head, which first probes downward to target the subject, and upon failing to be valued probes upward? Since there is no clear diagnostic to test this, I treat Pers imp /Jussive head as being a special person head which enters the derivation with a pre-valued 2nd person feature but no status feature.
to its overt realization in the allocutive imperative. I thereby maintain a one-probe system for the standard imperative in Punjabi.
With the above-mentioned composition of T-Pers imp , we proceed to the nature of the imperative subject, which while typically covert, can be realized overtly and is hence best analyzed as pro, and not a PRO (cf. Chomsky 1981;Rizzi 1986, among many others). Following the basic tenets of the Government and Binding (GB) and Minimalist framework, DPs can be partitioned according to their needs with respect to abstract Case. Overt DPs and pro require normal abstract Case, whereas PRO either doesn't require Case or requires a special "null" Case (Martin 2001). In finite domains, the T head is a nominative case valuing head, which can license overt DPs and pro. In non-finite domains on the other hand, only a PRO can be licensed as the subject due to the lack of a case-licensing T 0 . Given the availability of both overt 2nd person pronouns and also some quantificational 3rd person items as imperative subjects, the imperative subject cannot be treated as PRO which must always be covert. I therefore treat the imperative subject position as a pro, in the sense of Rizzi (1986). The pro is inherently unspecified for phi-feature values, which must be valued in relation to person features located on functional heads via agreement. Clear motivation for the subject not hosting the 2nd person feature, but rather inheriting it from the agreeing functional head in imperative comes from the observation that the verbal complex obligatorily appears with 2nd person morphology, which cannot vary for person even with minus person (3rd person) quantificational subjects. I repeat the relevant example for both the standard and allocutive imperative here in (98) -the observation holds for either manifestation of the 2nd person feature -o or je.
(98) saare jaa-o/jaayaa je all go-imp.2pl/go alloc.pl 'Everyone go!' In summary, the standard imperative hosts an irrealis T 0 with Pers imp . The T-Pers complex has a pre-specified 2nd person feature and a nominative case feature, which agree with the pro subject to value it as 2nd person and thereby restrict it to the addressee. Consider the tree in (99).

(99) Standard imperative in Punjabi
The structure in (99) consists of C imp , which occurs with an irrealis T-Pers imp complex, where Pers imp bears an interpretable and valued instance of the 2nd person feature, and T 0 bears a valued case feature. The subject, i.e. pro enters the derivation with an unvalued person feature as well as an unvalued case feature. Assuming downward agree (Chomsky 2000;Chomsky 2001), the T-Pers imp complex searches its c-command domain for the closest active goal; it finds the subject, which hosts an unchecked case feature. Upon agreement, the subject is valued for case and receives a 2nd person feature, resulting in addressee-restriction, as required. Crucially, since Pers imp lacks the [status] feature, generic addressee readings can obtain.

Allocutive imperative
The allocutive imperative is analysed on par with the standard imperative except for two differences: (a) the clause-type free Pers 0 (specifically, Pers alloc ) and not the imperativespecific Pers imp constitutes the licensor of the 2nd person feature on the subject in allocutive imperatives, and (b) the allocutive head agrees not only in person but also in status feature with the Adr, which restricts it to contexts with a specific addressee. The structure of the allocutive imperative also consists of a C head with an imperative operator, which occurs with an irrealis T-Pers complex. However, unlike the irrealis T complex in the standard imperative which has a pre-valued 2nd person bearing Pers imp , the irrealis T complex in the allocutive imperative hosts Pers 0 , which enters the derivation with an unvalued person and status feature, and must be valued by agreement with the Adr. Recall our analysis of allocutivity: the Pers 0 head with the unvalued person feature probes downward to agree with the subject; however, since both Pers 0 and the pro subject are unvalued for features, this agreement cycle fails. This gives way to the second cycle of agreement -Pers 0 probes upward and agrees with the Adr. This values the features on Pers 0 . To ensure that these features are transmitted to pro, I invoke the Feature Sharing view (Pesetsky & Torrego 2007) in (100), albeit with a modification in directionality of agreement probing.
(100) Feature sharing view of Agree: an unvalued feature F (a probe) on a head H at syntactic location α (Fα) scans its c-command domain for another instance of F (a goal) at location β (Fβ) with which to agree. Replace Fα with Fβ, so that the same feature is present in both locations.
The feature sharing view of agreement allows the possibility to value a feature at more than one location as a consequence of an application of Agree that involves only one of these locations. Consider a derivation with two unvalued occurrences of F. Agree between these two occurrences of F yields a structure that contains only one occurrence of F with two instances; this is shown with indices in (101). When either of these instances of F undergoes agreement with a valued occurrence of F in a third syntactic location, the result will be a valued feature F present at three locations, as schematized in (102) Assuming the feature sharing approach to agreement coupled with bidirectional Agree, we can resolve the agreement relation between the pro subject, Pers 0 and Adr. Recall that Pers 0 first agrees with the subject, but is not valued since the imperative pro lacks pre-specified feature values. This application of agreement between the subject and Pers 0 yields a structure that contains only one occurrence of the person feature with two instances. When Pers 0 probes upward in the next cycle and receives a value from the Adr, the person feature on both the subject and Pers 0 get valued. Consider (103). 25

(103) Allocutive imperative in Punjabi
In summary, in the absence of Pers imp , the pro subject agrees with the Pers alloc head, which receives its features via agreement with the Adr. This agreement relation between the subject and the Adr, mediated via the allocutive head, restricts the subject to the addressee, as required by imperative syntax. Moreover, the presence of [status] feature on the allocutive head bans the allocutive imperative from occurring in contexts with a generic addressee.
A question that remains is as follows: given the availability of allocutivity across clause types, what prevents agreement between the Adr, Pers alloc and the subject in declaratives? In other words, why is the declarative subject not restricted to the addressee in the presence of Pers alloc , which also hosts a 2nd person feature? There are two possible answers to this question. The first possibility is to argue that feature transfer from the Pers alloc head to the subject is context-sensitive -in the presence of C-imp, the Pers alloc head transmits 25 As an alternative to the above mentioned cyclic agree based account of allocutivity assumed for allocutive imperatives, it is possible to posit that the language has two types of Pers heads -Pers1 with a [person] feature alone, and Pers2 with [person] plus [status] features. Pers1 corresponding to Pers imp is a special category that occurs only in imperatives and enters the derivation with a 2nd person feature. In contrast, Pers2 can occur across clause-types. In a declarative clause, Pers2 is unvalued for person and status features, which must be valued by agreement with the Adr via the second agreement cycle (assuming Cyclic agree). In the presence of C-imp, these features on Pers2, however, are pre-valued. This system arguably makes imperative syntax simpler -a pre-valued 2nd person projection (with or without the [status] feature) agrees with the imperative subject via downward agree to restrict it to the addressee. However, the main issue with this account is that it necessitates two distinct lexical entries for the Pers2 head as conditioned by clause-typeone with pre-valued features and another with unvalued features. This approach also underlies that while allocutivity in declaratives obtains via cyclic agreement with the Adr, in imperatives, no such derivation is required. In view of these shortcomings, I maintain a uniform account of allocutivity across clause types.
features to the subject. If the C head does not have an imperative operator, there is no feature transfer to the subject. While plausible, such an approach arguably suffers from the same premise that the paper attempts to question -akin to associating a person projection with a clause-type, this approach associates a certain derivational step with a clause-type.
A better solution lies in the distinct nature of subjects across imperative and declarative domains. We have seen that the imperative subject is a pro which enters the derivation without any pre-specified values for the person feature and must receive them from the agreeing head in order to be licensed. The declarative domain in Punjabi, however, does not seem to have an underspecified pro, even when the subject is covert. Unlike the imperative domain, in the declarative domain, licensing of the covert subject is not contingent on agreement (Butt & King 1997;Butt 2001). Consider the imperfective sentence with a null subject in (104) -similar to Italian, the absence of null subject in this structure seems to correlate with corresponding agreement on the verb. However, like in Japaneselike languages, which lack phi-agreement, a null subject in the declarative domain can also be licensed in the absence of corresponding subject agreement. Consider the example in (105), which is a perfective structure where the verb agrees with the object and not the oblique subject. The subject can be null, notwithstanding. If the null subject in (105) were a minimally specified pronoun requiring licensing via agreement with the functional head, it is unclear how it would be identified as a 1st person singular in the absence of corresponding agreement on the verbal complex. Butt & King (1997) and Butt (2001) link pro-drop in HU/Punjabi declaratives to the discourse structure of the language. Employing a range of HU film dialogs, Butt & King (1997) propose that only old information such as continuing topics or background information can be dropped. Consider (106) from HU -the Punjabi sentence is exactly the same.
(106) Hindi-Urdu (Butt & King 1997) [ye] topic bhi meri tarah haiN. jahaaN pro topic dana 3pl.nom also 1sg.gen.f.sg like be.prs.3pl where pro seed.m.sg.acc dekha, udhar pro topic gaye or pet bhar kar pro topic see.pfv.m.sg there pro go.pfv.m.pl and stomach fill having pro uR gaye rise go.pfv.m.pl 'They (topic) are also like me. Where (they cont.topic) see a seed, there (they cont.topic) go and having filled (their) stomach, (they cont.topic) fly away.' As shown in (106), the 3rd person plural pronoun, which is the continuing topic, is dropped throughout the dialogue at four instances. The possibility to drop the continuing topic may or may not correlate with agreement/case. In the second instance, there is no corresponding agreement on the verb dekha, which shows agreement with the object. Moreover, the case of the dropped argument in this second instance is not even nominative. Instead, given the transitive perfective configuration, it should be ergative. In the third and fourth instances, however, the verb shows corresponding 3rd person plural agreement. While a larger examination of the issue is awaited and the question remains as to what governs the distribution of covert arguments in the declarative domain, it is clear that the determining factor is not agreement. I take this preliminary observation to suggest that even when phonologically unrealized, the covert subject pronoun in Punjabi declaratives patterns with an overt pronoun which enters the derivation with pre-valued phi-features. 26 In the absence of a true minimal pronoun subject in the declarative domain, there is no reason to transfer the allocutive features from Pers alloc to the subject. Subject licensing, regardless of its (c)overt phonological realization takes place as discussed in section 3. 1st/2nd person subjects agree with the person probe at Pers 0 to be licensed -no further agreement between Pers 0 and the Adr takes place. This renders the issue of transferring allocutive features to the subject moot. 27 As for 3rd person subjects, they enter into agreement with the T head, where they receive a nominative case. Due to the lack of person on the 3rd person subject, the probe at Pers 0 remains unvalued. It agrees upward and receives the 2nd person feature from Adr. However, since the 3rd person subject does not have a [person] attribute, valued or unvalued to begin with, it is clear that no feature transfer would take place between Pers 0 and the subject. 26 The system here posits two distinct types of pro-one which is minimal and another which is born with all its features in place like an overt pronoun. A solution to this proliferation of pro-types would be to posit PRO as the imperative subject. However, as discussed previously, given the ability of the imperative subject to be overt (and also 3rd person, albeit only when quantificational), this is not tenable. 27 While there is no agreement between 2nd person pronouns and the allocutive marker, they must match in features, as shown in (i) from Punjabi.
(i) karan twaaDe-naal/*tere-naal bajaar gayaa sii je Karan.nom 2pl.gen-with/2sg.gen-with market go.pfv.m.sg be.pst.3sg alloc.pl 'Karan went to the market with you.' This can be analysed via Alok & Baker (2018). According to Baker's (2008) version of the Person Licensing Condition, a 2nd person pronoun acquires its features precisely by being bound by the Adr in the periphery. Alok & Baker (2018) argue that since the 2nd person pronoun is variable-bound by the Adr, and the allocutive marker obtains by agreement with the same Adr, it follows that the features on the two items, i.e. the allocutive marker and the 2nd person pronoun match. There is an outstanding problem for this account, however. Some allocutive languages may not always exhibit the matching requirement. Consider (ii) from Korean, where the 2nd person pronoun is a plain/nonhonorific form, but the allocutive marking is honorific.
(ii) Korean (Pak 2015: 136) Ne-na cal ha-si-eyo you-na well do-hon-dec.pol 'Mind your own business.' (Lit: 'YOU do well.') As a reviewer suggests, this could ensue from Korean speech particles being a realization of a head 'c', without being in agreeing relation (Agree). I agree with this line of thought, also because such mismatches are found in languages without phi-agreement (and not in Magahi, Punjabi etc, which show phi-agreement. However, while Portner et al. (2019) argue that the Korean speech style particles/SSPs realize c and are not the result of agreement, they do not predict a mismatch between the SSPs and 2nd person pronouns. In fact, in section 5.2 of their paper, they predict obligatory matching: 'Interlocutor is an argument of c and, when c binds a variable in the clause, its features are passed on to that variable. We therefore predict that, in Korean, speech style particles and second person pronouns always show the same value for the status feature.' This claim however is incorrect, as shown by (ii). An investigation into the issue of mismatches in features on the allocutive marker and 2nd person pronouns is in progress.
To sum up, allocutive imperatives pattern with standard imperatives in that they host a functional projection, which agrees with the subject to restrict it to the addressee. However, this projection varies across the two imperatives -it corresponds to the imperative-specific Pers imp in the standard imperative and the clause-type free Pers alloc in the allocutive imperative. In its declarative occurrence, the clause-type free Pers alloc does not transfer 2nd person features from the Adr to the subject, which is distinct from the underspecified pro subject of the imperative. These findings are summarized in Table 5.

Conclusion
The paper set out with the primary objective of investigating the syntax of addresseeencoding in imperatives in light of allocutivity, which can occur across clause-types in some languages. In their seminal work, Zanuttini (2008) and Zanuttini et al. (2012) argue that addressee-restriction on the imperative subject is obtained by agreement with a unique and obligatory 2nd person projection which is specific to directives. This paper argued that linking the 2nd person projection, which agrees with the imperative subject to restrict it to the addressee, to a specific clause-type is incorrect. The empirical motivation came from the allocutive imperative in Punjabi, which is composed of the allocutive ending to the exclusion of the imperative-specific ending. Furthermore, this paper also corroborated the distinct syntactic encoding of a specific interlocutor-addressee vis-à-vis a generic addressee, as proposed in Portner et al. (2019). Like Korean, Punjabi imperatives clearly manifest a distinction in specific/generic readings of the addressee -only the allocutive imperative is restricted to contexts with a specific addressee. These interpretive facts were argued to follow from special properties of the allocutive head. Unlike the Jussive/Pers imp head which only hosts a 2nd person feature and is therefore allowed in generic addressee contexts, the allocutive head hosts both a person as well as a status feature, which encodes the speaker-addressee relation. Since the allocutive head must agree in not only person but also in the status feature with the Adr, it is blocked from generic addressee contexts. Finally, a word about variation in the role of allocutivity in imperatives needs to be said. As discussed in section 1, in Korean imperatives, speech style particles treated on par with allocutive markers, occur in addition to the Jussive head, which agrees with the subject. The speech style particles only provide extra information about the relation between the speaker and the addressee. In contrast, as we saw for allocutive imperatives in Punjabi, the allocutive head is not ornamental; instead it replaces the Jussive head to agree with the subject. A plausible explanation for this difference lies in the syntax of allocutive markers across the two languages. With Portner et al. (2019), Korean speech style particles are a realization of the head c, and crucially do not underlie an agreement relation in syntax. The imperative subject, however, needs to be restricted to the addressee via an agreement relation with a functional head bearing a 2nd person feature. This role can only be performed by the Jussive head in Korean imperatives, resulting in its obligatoriness. In contrast, allocutivity in Punjabi is realized at Pers 0 (in the T region) as a result of an agreement relation with the Adr. Moreover, Pers 0 does not constitute an allocutive probe distinct from the subject agreement probe. I suggest that syntax of allocutivity in Punjabi is therefore well-suited to mediate an agreement relation between the Adr and the subject, akin to the Jussive head. Further validation of the proposed locus of variation will depend in part on what can be discovered about the composition of imperatives in other allocutive languages.