2.1 The speaker-addressee relation: two types of directive force

To recap the discussion in Section 1, the two grammatical categories of qing receive different interpretations (cf. (3a–b)): qing¹ encodes obligation imposed by the speaker on the addressee to get him/her to do something to make a future state of affairs or action come true, whilst qing² encodes permission by the speaker for the addressee to make true a future state of affairs or to perform a desired action, according to a set of norms. Context plays a crucial role in distinguishing between the two categories, which are in turn represented by two types of the speaker-addressee relation.

The context in (12) shows that the manager (speaker) has authority to impose an obligation on the employees (addressees) by asking them to wear the company uniforms. In this context, only the use of qing¹ is felicitous rather than qing².

Consider another context in (13), where the principal does not force the students to sit down but instead, permits them to sit down without imposing obligation on them. The use of qing² is hence felicitous.

The above contrasts are taken to show that qing¹ is closely tied to the obligation imposed by the speaker upon the addressee in a given context, whereas qing² is devoid of this property. If this line of thinking is on the right track, the contrast between (14a) and (14b) can be understood if an angry girl asks her boyfriend to get out of the room at her request in (14a) instead of giving her boyfriend permission to decide whether to leave the room in (14b).

In this subsection, the distinction between qing¹ and qing² is captured by the two types of directive force, which, in turn, instantiate two kinds of the speaker-addressee relation: qing¹ represents a command where the speaker has administrative or social authority over the addressee, whereas qing² is used to communicate permission in a context where the speaker intends to get the addressee to perform an action which does not necessarily have to be true in a future time. It follows that the two types of the speaker-addressee relations are identified and encoded by command and permission, respectively.

2.2 The optional presence of vocatives

Note that the subject of the qing imperative situated either after qing¹ or preceding qing² cannot be analyzed as a vocative. First, the referent of the subject in the two types of qing imperatives can be syntactically identified by a vocative in the leftmost position or the rightmost position, as shown in (15a–b) and (16a–b), where the vocative agrees with the subject in number and person.

Moreover, vocatives are outside of the thematic grid of the verb but are related to propositional information and occupy a clause-external position (Moro 2003).⁷ (15a–b) and (16a–b) prove that the vocatives are optional and their presence identifies a group of addressees or a single addressee present in the immediate discourse. The vocatives in (15a–b) and (16a–b) are, however, infelicitous, if referring to the set of addressees not in the discourse. Last, vocatives are not restricted to imperatives (Portner 2005; Alcazar & Saltarelli 2014).

2.3 The designated structural position of qing¹ and qing²

Though the two categories of qing under investigation can be distinguished, less discussed is their structural positions. The two categories behave differently when co-occurring with the imperative mood adverb qianwan ‘by all means’ (Cinque 1999; Li 2017; contra Hsiao 2012). As evident in (17a–b), qing¹ must precede the imperative adverb.⁸

(18a–b), in contrast, indicate that qing² is blocked before or after the adverb qianwan. The grammaticality of (18a) is greatly improved, however, if there is a pause between the subject nimen ‘you’ and qing², as shown in (19). It is tempting to analyze nimen ‘you (pl)’ under discussion as a vocative. Yet, this line of thinking is immediately challenged by (20), where the genuine vocative is able to co-occur in the sentence-initial position with the subject.

Two conclusions can be drawn here. First, if qing² is considered to be an imperative marker, it is not clear why qing² resists co-occurring with the imperative mood adverb qianwan ‘by all means’ in (18a–b). Second, qing¹ and qing² are not situated in the same structural position in terms of their co-occurrence with the mood adverb qianwan.

2.4 Outer affective constructions in Mandarin

In Mandarin, outer affective constructions are interpreted as encoding a speech act of exclamation (or exclamative) associated with a speaker/addressee-oriented construal and a presupposition that goes beyond the speaker's expectation (Tsai 2015a, 2015b). (21) says that it is beyond the speaker's expectation that Zhangsan drank three bottles of wine upon the speaker. Crucial to our discussion here is that (21) can be interpreted as performing a speech act, such as ‘I command that Zhangsan should drink three bottles of wine upon me’. In other words, (21) can encode a degree of obligation in the sense that in the context set up in (21), Zhangsan should not perform the wine-drinking action upon the speaker.

Along this line of thinking, if (21) encodes directive force (obligation), it is predicted that qing¹ is compatible with the outer affective construction, because as shown in Section 2.1, qing¹ bears the directive act of command. The prediction is borne out in (22a), and the infelicity of qing² in (22b) can be explained accordingly. As the outer affective construction encodes the directive act of command resulting from the evaluative adverb juran ‘unexpectedly’, which can be recast as obligation in the immediate context, and qing², denoting permission, is not compatible with this obligation context.⁹

2.5 Two types of deontic modality

Two categories of qing encode different types of deontic modality in Mandarin. As shown in (23a), the deontic modal auxiliary yao and its corresponding deontic adverb bixu are permitted in the qing¹ imperative clause (Tsai 2015c). The occurrence of the deontic modal auxiliary and the deontic adverb, in contrast, is not felicitous with the qing² imperative in (23b). The infelicity can be ascribed to the fact that qing² does not encode obligation imposed on the addressee, which is crucial in the context set up in (23).

It should be further noted that the deontic modal auxiliary yao and the adverb bixu are permitted to co-occur in (23a), only one of them occurs in (24a–b), and both of them are covert in the qing¹ imperative in (24c) with the interpretation of obligation.

The contrast between (23a) and (23b) is taken to illustrate that qing¹ encodes the addressee-oriented deontic modality (obligation), and the encoding arises from the implicit deontic modal not present in the qing¹ imperative in (24c), which can be overtly realized as in (23a) and (24a–b).

In contrast, qing² encodes a type of deontic modality in which the subject of the imperative is required to perform an action, according to a set of norms in the absence of the speaker's obligation imposed on the addressee. Analyzed along the lines of Tsai & Portner's (2008) system of Mandarin modals, the deontic modal auxiliary yao and its licensed adverb bixu encode the subject-oriented deontic modality in which the subject is permitted to perform a desired action according to a set of norms.

2.6 Summary

Table 1 summarizes the structural properties and pragmatic effects of qing¹ and qing² surveyed in Section 2. The properties will be accounted for by a syntax-pragmatics analysis to be developed in Section 4.

It must be mentioned that the two types of deontic modality discussed in Section 2.5 fare well with Isac's (2015) two types of deontic modality in the imperative context, as shown in (26a–b). In the true deontic modality in (26a), the obligation is put on the addressee, and the subject-oriented deontic modality in (26b) requires that permission is given to the subject. Interpreted along the lines of Tsai & Portner's (2008) system, the true deontic modality and the subject-oriented deontic modality are recast respectively as the addressee-oriented and the subject-oriented deontic modality, both of which can be realized as the deontic modal head yao and its corresponding adverb bixu (see (5)). Based on the observation in Section 2.5, qing¹ encodes the addressee-oriented deontic modality, while qing² encodes the subject-oriented deontic modality, as they have different modal interpretations in their imperative contexts.

3.1 Illocutionary act phrase (Woods 2015)

Woods (2015) argues for two types of please in English, according to their distributional and interpretative differences. One is an overt head of the IA⁰ that projects the illocutionary act phrase (IAP) in the CP periphery and marks an utterance as a request, while the other is an adverbial whose use pertains to politeness determined by contextual factors. As shown in (27), (28) and (29), the clause-initial please is strongly restricted to imperatives and interrogatives, and expresses direct requests, and the clause-final please has a wider distribution and is compatible with various clause types in the b examples of (27), (28) and (29).

The sentence-initial please has two properties. First, as a request maker, the clause-initial please can be used in a non-polite context and seems to attenuate non-polite utterances, as is evident in (30). Second, the sentence-initial please in (31) introduces a conversational implicature, which can be cancelled.

Following Hill's (2007) system of the speech act layer, Woods argues that the sentence-initial please is an IA⁰ (illocutionary act) dominating CP, which is the locus of marking clause types.

(32)

Woods' (2005) analysis (the amended version of Hill (2007))

(Woods 2015, 371)

By contrast, the clause-medial please can occur in declaratives with deontic modal force expressing necessity. As shown in (33a–b), the deontic modal auxiliaries will and must are allowed to occur with the clause-medial please.

Similarly to the clause-initial please, the clause-medial please is used in situations in which an expression of politeness may have been forced, and is even sarcastic. Admittedly, the clause-medial please can be interpreted as a request in (35).

Woods argues that the clause-medial please is merged to a position above vP, and that what makes it distinct from the clause-initial please is the lack of ForceP, suggesting the existence of an independent functional projection above vP (in the Mittelfeld/the IP-internal domain) dedicated to the encoding of discourse, for example IAP. Moreover, it modifies a predicate by marking it as a request and occurs only in declaratives that have the same modal force as a request.

While tempting at first glance, Wood's analysis cannot account for the properties in Table 1. For instance, though the IAP analysis of please can explain the encoding of directive force if we assume an extension of the CP layer that is responsible for syntactic substantiation of the illocutionary force, it cannot account for the distinctions between the two grammatical categories of qing (see Table 1). Woods's analysis is insightful in two ways, however. First, there exists an extended functional projection at the top of the CP periphery responsible for the syntax-pragmatics interface computation of directive force. Second, deontic modality plays a role in deriving the interpretation of imperatives; that is, the clause-medial please is base-generated in the vP periphery and is tied to modals expressing necessity.

3.2 Light verb phrase (Alcázar & Saltarelli 2008)

In English and Spanish, imperatives can be formed by means of merging ‘imperative’ verbs with bare verbs: go/come in English (36a-c) and the suffix –ve in Spanish (37a-b). Alcázar & Saltarelli (2008) argue for the light verb analysis of imperatives, in which a light verb projects vP^demand and the speaker-addressee relation is determined by two vP layers.

(38) schematizes the derivation of the verb oyeve in (37a), demonstrating that the speaker-addressee relation is structurally represented by vP^demand taking vP as its c-commanded complement such that e₁^speaker is able to perform a hearing action imposed on the e₂^ADDRESSEE.

(38)

Alcázar & Saltarelli's (2008) light verb analysis of (37a)

Nerveless, there are compelling reasons to believe that this analysis cannot be adopted for the two categories of qing in Mandarin. First, head movement as an operation to activate the speaker-addressee relation in the vP domain suffers the burden of proof in Mandarin, as verbs are devoid of imperative morphology in Mandarin. Second, as argued by Huang (2015), there is no strong feature (outside of the vP layer) triggering verb movement in modern Mandarin. It follows that verb movement as a strategy to form imperatives in Mandarin is not supported.

3.3 Jussive phrase (Zanuttini 2008)

Zanuttini (2008) argues for an imperative-specific projection, JussiveP, in the CP layer conveying the directive force and licensing the subject of the imperative. As shown in (39), Jussive⁰ acts as a probe establishing a proper probe-goal Agree relation with the subject externally merged at [Spec, vP], accounting for the second person restriction imposed on the imperative subject.

(39)

Zanuttini's (2008) JussiveP analysis

One merit of the JussiveP analysis is that it nicely captures the second person restriction imposed on the subject of the imperative, but it fails to account for the designated position of the subject of the imperative ni ‘you’ residing in the CP domain and the imperative mood adverb qianwang ‘by all means’ that is compatible with qing¹ but not qing².

3.4 Summary

Table 2 is a summary of the analyses reviewed above and illustrates how they explain the observed properties in Section 2. It is noted that the analyses are not intended for the licensing of vocatives and the imperative mood adverb in imperative clauses. What is more, one shared view is that the speaker-addressee relation is visible to syntactic computation, encoded by functional categories projecting in the extended zones of CP (IAP in the IAP analysis and vP^demand in the light verb analysis). These analyses suggest that imperative construals are licensed in the CP periphery. None of the analyses, however, can disambiguate the two types of directive force encoded by two categories of qing.

4.1 Theoretical assumptions

First, I follow Cheng (1997) in assuming that C⁰ is the locus of clause typing information, which is recast as Force⁰ in Rizzi's (1997) split-CP system. Second, I adopt Frascarelli & Jiménez-Fernández's (2016) view that the imperative is a mood rather than a type of illocutionary force, and is dependent on both a hidden illocutionary force activated in the matrix clause and discourse information about the speaker-addressee relation.¹¹ To represent the speaker-addressee relation in the syntax, I follow Haegeman & Hill's (2013) adapted version of Speas & Tenny's (2003) syntax-pragmatics system of speech act, as schematized in (40), in which a supra-subordinate speech layer merges to CP and thematic p(ragmatic)-arguments, including speaker, addressee, and utterance content, are built into the syntax and licensed within their own domains. This interface analysis is more advantageous to the JussiveP analysis (i.e., Zanuttini 2008; Frascarelli & Jiménez-Fernández 2021, among others) in two regards.¹² First, in addition to representing the speaker-addressee relation in the syntax, the sa*P analysis captures the encoding of directive act in a way that the speaker (=agent) in the highest position c-commands the addressee (goal) and is in the qualified position to perform an action (such as ask, order, suggest, etc.) by means of which the speaker imposes the utterance content on the addressee. Thus, one type of directive force encoded by qing¹ (see Section 2.1) can be structurally encoded as I hereby order that you…, though not overtly spelt out. This analysis fares well with the Performative Hypothesis (Ross 1970), along the lines of which root clauses are selected by higher covert predicates (i.e., I order that… for imperatives).¹³ Second, it has been proposed that the core function of JussiveP is to both encode the imperative clause type and the addressee by valuing the person feature as the second person feature (see Hill 2007, Zanuttini 2008 and Zanuttini, Pak & Paul 2012). However, as discussed above, clause typing is determined by mood in Mandarin (Li 2006). Hence, granted the sa*P analysis, SA⁰ licenses the presence of vocatives (i.e., [Spec, SAP]) within the addressee/hearer domain. When bearing the [iperson^2nd] feature, SA⁰ serves as an active probe searching for a proper goal within its c-command domain to value the [uperson] feature of the subject in imperatives. Based on the above reasons, the JussiveP analysis is not motivated in the current work.

(40)

To support the above two assumptions, it is necessary to discuss how clause typing on Force⁰ connects to the speech act layer, as ForceP serves as the gateway toward the syntax-pragmatics interface. Therefore, to ensure the success in activating the syntax-pragmatics interface, I implement Kellert & Lauschus's (2016) version of the sa*P analysis in which sa*⁰ creates a dependency on Force⁰ by means of probe-goal relations. Here for concreteness, SA⁰ is endowed with an interpretable speech act feature [iSa] and an uninterpretable force feature [uForce] that can be checked by a matching clause typing feature in the interrogative clause encoding directive speech act. (41b) illustrates the derivation of (41a) under Kellert & Lauschus's (2016) analysis.¹⁴ Assuming a feature mechanism of the probe-goal relation (Chomsky 2000), Sa⁰, as an active probe, establishes a proper Agree relation with C⁰, the goal, while valuing [uSa] on C⁰ as encoding a directive speech act, and its unvalued [uForce] is in turn valued by [iForce] on C⁰ to encode interrogative clause.

b.

Directive Force/Interrogative clause: I hereby ask that you…

One complication added to the current discussion is that I follow Baker's (2008) bidirectional approach to Agree, according to which the target (the probe or the goal) of valuation may either c-command or be c-commanded by the controller, allowing both upward or downward probing, as formulated in (42).

Moreover, to capture the deontic interpretations of the two categories of qing, I follow Tsai & Portner's (2008) hierarchy of modals in Mandarin, as schematized in (5) and repeated in (43) for the reader's convenience, where there are two categories of deontic modality which can be realized as the deontic modal auxiliary yao: one is the addressee-oriented deontic modal auxiliary in the CP periphery and the other is the subject-oriented deontic auxiliary in the vP periphery.^15,16

(43)

The hierarchy of modals in Mandarin (adapted from Tsai & Portner 2008, 4)

Third, as argued in Li (2006), mood determines clause typing in Mandarin, with the idea being that a directive force can be compatible with different clause types (i.e., interrogatives, Yes/No questions, imperatives, etc.). I adopt Kempchinsky's (2009) idea that Force⁰ contains an unvalued world feature [uW] which has to be checked and deleted by a mood head via identification in order to be identified as irrealis (subjunctive) or realis (indicative). In this light, the [uW] feature on Force⁰ needs to be checked by either the addressee-oriented deontic modal (the command reading) or the subject-oriented deontic modal (the permission reading). This line of thinking is not without theoretical support. Han (2000) argues that imperatives contain an illocutionary operator along with directive force, encoded in the syntax by two features ([directive] and [irrealis]) on C⁰.

With the assumptions ready in place, we will develop a syntax-pragmatics interface analysis of the two types of qing imperatives in Section 4.2 and Section 4.3, and demonstrate that the analysis can capture the syntax-pragmatics interface computation of imperative mood, clause typing and the speaker-hearer relation.

4.2 Structural configuration of qing¹

I propose that qing¹ is a head of imperative mood (represented by Mood_imp⁰ in the CP periphery). (45) illustrates a step-by-step derivation of the qing¹ imperative in (44).^17,18,19

(45)

The derivation of (44)

a. Step 1: The subject of the qing1 imperative undergoes topicalization to [Spec, TopP] to check the peripheral feature (Tsai 2015a).

b.

Step 2: Qing1 undergoes head movement to Force0 to check the [uW] feature and types the clause as imperative (irrealis).

c.

Step 3: As the speech act layer merges to ForceP, sa*0 with the [iSa] and the [uForce] establishes an Agree relation with Force0 which bears the [uSa] and the [iForce] to undergo the Bidirectional Agree to have the unvalued features valued. The subject with the [uperson] is valued by SA0 as [iperson2nd].

There is ample evidence for the step-by-step derivation presented in (45a–c). First, it is motivated to claim that the encoding of obligation by qing¹ pertains to Mod⁰_deo in (45b), whose interpretative effect is observed by its overt head yao and its corresponding adverb bixu licensed within this domain. Kruger (2012) maintains that deontic modality is available in the interpretation of an imperative clause that obliges the addressee to conform to the proposition presented or a request imposed by the speaker in the immediate context. Apparently, deontic modality has a direct bearing on imperative mood. This line of reasoning is supported by the imperative interpretation of qing¹ in which the speaker imposes obligation on the addressee by committing him/her to perform an action or mandating that state of affairs be true. Take (44), for example. The speaker hereby obliges the addressee Xiaodi to conform to the request that he must keep the matter in mind. Likewise, it has been reported in previous works on imperatives in other languages (Han & Lee 2007 for imperatives in Korean) that imperatives associated with request or command are encoded by deontic modality. Second, the subject of the imperative clause undergoes obligatory topicalization to check the peripheral feature on Top⁰ (Tsai 2015a). The second person restriction is ascribed to the subject whose [uperson] feature is valued by SA⁰ through an Agree relation, which bears the pragmatic role of the addressee. SA⁰, which acts as an active probe and values the [uperson] feature on the subject as [iperson^2nd], enters a proper Agree relation with the subject ni ‘you’ at [Spec, TopP], consistent with Zanuttini's (2008) view that a syntactic relation is formed via Agree resulting in the same person feature in both positions (e.g., the Jussive⁰ and the subject). Third, qing¹ is a Mood⁰ that undergoes head movement to Force⁰ to specify the imperative mood via the feature checking of the [uW] feature, following Koopman's (2007) idea that ‘force’ has to be typed as imperative or declarative.²⁰ In other words, the process of clause typing is achieved by Mood⁰-to-Force⁰ movement. This line of thinking adds weight to Cormany's (2014) view that a clause contains a syntactic feature which specifies its type and participates in syntactic interaction with other functional elements. Third, to ensure the success in activating the syntax-pragmatics interface, I implement Kellert & Lauschus's (2016) version of the sa*P in that sa*⁰ is endowed with an interpretable speech act feature [iSa] and an uninterpretable force feature [uForce] that can be checked by the [uSa] and the [iForce] feature on Force⁰ via Bidirectional Agree, as schematized in (45c).

Note that the analysis in (45a–c) accounts for the observed properties of qing¹ in a more principled manner. First, the root clause properties of qing¹ imperatives can be ascribed to the role of qing¹ in activating the probe-goal relation between the speech act layer and the CP layer. The temporal and the conditional adverbial clause, which are arguably reduced structures (Haegeman 2012) and lack a full-fledged CP structure, do not hinge upon the speech act layer for their licensing configurations. As expected, qing¹ is not compatible with these two types of adverbial clauses.²¹ Second, the speaker-addressee relation in the qing¹ imperative is syntactically represented by the speech act layer above ForceP. Moreover, the second person identification of the subject of the qing imperative is constrained by SA⁰ through the person feature checking.

4.3 Structural configuration of qing²

In contrast to the head status of qing¹, qing² is an imperative mood adverb. Recall from Section 2.3 that qing² is not permitted to co-occur with the imperative mood adverb qianwan ‘by all means’, due to the co-occurrence restriction on two adverbs of the same type. Moreover, qing² lacks the obligation that the speaker imposes on the addressee; it encodes the subject-oriented deontic modality that contributes to the function of permission. As shown in (46), the qing² imperative does not involve the subject-oriented obligation of the speaker that intends to get the addressees to complete filling out the application forms. Rather, (46) is felicitous in a context in which the guests must fill out the application forms to apply for subsidies, according to the regulations. Thus, the permission is imposed on the subject according to the regulations, rather than from the speaker, which amounts to the subject-oriented deontic modality (see Section 2.5).

Given the division of labor between the speech act and clause typing as in the case of qing¹, it remains to be accounted for how clause typing is achieved, if qing² is an imperative mood adverb merging to [Spec, MoodP_imp] and Mood⁰-to-Force⁰ movement is not active for clause typing (i.e., the feature checking of the [uW] on Force⁰).

It has been proposed in the literature that the syntax-pragmatics mapping of the imperative can be achieved via movement operations: (i) V⁰ undergoes head movement to C⁰ or Jussive⁰ to check the [+imp] feature and bears imperative morphology (see Frascarelli & Jiménez-Fernández (2021) for a more detailed discussion), or (ii) a phrase with relevant properties of the imperative is pied-piped to [Spec, ForceP]. As argued in Koopman (2007), Force⁰ can be typed as declarative or imperative by moving a constituent that has the relevant properties to Force⁰. Thus, Force⁰ is typed as declarative when FinP is attracted to [Spec, ForceP] but is typed as imperative when ImpP is attracted to [Spec, ForceP]. Koopman (2007) analyzes Dutch and German imperative constructions and argues for a model of clause typing that hinges upon phrasal movement targeting a maximal projection which contains the imperative properties. In this model, there are two ways of forming imperatives in Dutch and German, as schematized in (47a–b). In Dutch, ImpP moves to [Spec, ForceP] in the spirit of Kayne (1998), if TopP is absent in (47a), while in German, TopP, which contains ImpP, undergoes pied-piping to [Spec, ForceP].²² Notice that in (47b), when [Spec, TopP] is not occupied by a phrase, the verb at Imp⁰ needs to undergo upward head movement to Top⁰ such that it is not deeply embedded and can establish a licensing configuration with Force⁰.

(47)

Two movement operations for clause typing

a. Dutch: V-to-Imp0 followed by ImpP to [Spec, ForceP]

b.

German: V0-to-Imp0-to-Top0 followed by pied piping of TopP to [Spec, ForceP]

To account for the properties of qing², I motivate the analysis in (47b) by assuming that the qing² clause is typed as imperative by phrasal movement of TopP to [Spec, TopP]. (48) illustrates a step-by-step derivation of (46). As shown in (48b), TopP undergoes pied-piping to [Spec, ForceP] and the [uW] on Force⁰ is valued by the [iW] on Mood_imp⁰. As the derivation unfolds up to the merger of the speech act layer with ForceP, the [iForce] feature on Force⁰ serves as an active probe undergoing upward probing and values the [uForce] feature on SA⁰, and the [iSA] on SA⁰ in turn values the [uSA] feature on Force⁰ via downward probing. Thus, the speech act layer is valued as saying that I hereby permit that you…. Similarly, SA⁰ establishes an Agree relation with the subject nimen ‘you (pl)’ for the [iperson^2nd] feature and identifies the referents in the context.²³

a.

Step 1: The subject with the [uperson] feature undergoes topicalization to [Spec, TopP]

b.

Step 2: TopP undergoes pied-piping to [Spec, ForceP] to create a licensing configuration in which the [uW] on Force0 is checked by the [iW] on Moodimp0.

c.

Step 3: As the speech act layer merges to ForceP, sa*0 with the [iSa] and the [uForce] feature establishes an Agree relation with Force0 which bears the [uSa] and the [iForce] feature to undergo the Bidirectional Agree to have the unvalued features valued. The subject with the [uperson] is valued by SA0 as [iperson2nd].

4.4 Theoretical implications for the syntax of imperatives in Mandarin

To recap, the proposed analysis presents a transparent syntax-pragmatics mapping relation between the speaker-addressee relation and the imperative mood, and has shown that the two categories of qing in Mandarin imperatives are derived by means of two movement operations. This work has argued for the speech act layer at the periphery of CP that is responsible for the syntax-pragmatics interface computation of the speaker-addressee relation to be acutely sensitive to the context, lending weight to Cormany's (2014) view that a speech act field at the periphery of CP is activated to enter the syntactic derivation and participates in the syntactic-pragmatic interface computation of imperatives. Under the proposed analysis, the directive act of qing¹ (obligation) arises from the addressee-oriented deontic modality which imposes obligation on the addressee to get the desired action to come true, while the directive force of qing² (permission) contains the subject-oriented deontic modality in which the subject (the addressee) is permitted to perform an action or make true the state of affairs according to a set of norms. The postulation of deontic modality in imperative clauses has been supported by previous works (Han & Lee 2007; Zanuttini 2008; Kruger 2012; Cormany 2013, among others). Kruger (2012) has convincingly claimed that deontic modality ranges over concepts like obligation and necessity closely associated with the meaning of imperatives in general. What is more, as mentioned in Section 2, qing has its verbal origin and retains its argument structure in the vP domain, while the two categories of qing in the imperative context are functional categories, lose argument structure and are employed to encode the imperative mood in the CP domain. The co-existence of different categories of qing displays a trait of retaining their verbal origins on the grammaticalization path and leaving their cognates through syntactic projections throughout the vP, TP and CP layer (see Tsai 2015c, 275 for the cognates of modal auxiliaries in Mandarin), consistent with Roberts & Rousseau's (1999) minimalist view on syntactic change that grammaticalization can be regarded as shift ‘up the tree’ to a functional category. As illustrated in (7) and (8), qing still retains its verbal usage with the full argument structure while being developed as the mood categories that are realized as the imperative mood head (qing¹) or the imperative adverb (qing²) in the CP periphery. It is worth noting that the two types of deontic modality in the qing imperative also reflect the spectrum of the speaker's mood in cartographic terms through the subject-oriented deontic modality in the vP periphery to the addressee-oriented deontic modality in the CP periphery, consistent with Tsai's (2015c) view that Mandarin modals showcase various usages of a single modal from the lexical layer to the complementizer layer.

From a crosslinguistic perspective, imperatives are associated with deontic modality (Kruger 2012 for Old English, Han & Lee 2007 for Korean, among others). As discussed in Kruger (2012), the interpretation of deontic modality ranges over concepts like obligation, necessity and ability, which are the basics of imperative meanings. Of great interest to the current discussion is that the two types of deontic modality in the qing imperatives display their interpretative effects and have their overt realizations.

Moreover, the syntax-pragmatics interface analysis developed in this paper is advantageous to Yang's (2010) structural analysis of Mandarin imperatives in several regards. As shown in (49), Yang argues that the matrix imperative has an imperative operator (Op^imp) that encodes force, and assumes that Mood⁰ contains the syntactic and semantic information of imperatives. Particularly, force and mood are not overtly marked.

(49)

Yang's (2010) proposed analysis of Mandarin imperatives

However, the structure in (49) fails to provide a structural position for qing¹ as well as the vocative, as shown in (50a–b), where qing¹ is compatible with the mood categories yao and buyao. Furthermore, the division of force, mood and deontic modality is not clearly defined in Yang's analysis. For example, the speaker's directive act imposed on the addressee is not structurally represented, and the speaker-addressee relation, which is acutely sensitive to the immediate context, is not accounted for.

Particularly, the current proposal is able to account for (50a–b) in a more principled manner by adding modifications to the structure in (49). (51) instantiates the structure of (50a) under the syntax-pragmatics interface analysis defended in this paper. First, according to the analyses developed in Section 4.2 and 4.3, buyao and yao are deontic modal auxiliaries projecting ModP_deo and qing¹ is Mood⁰. Second, clause typing is achieved by head movement of qing¹ from Mood_imp⁰. Third, the fact that the subject must have the second person feature results from the thematic identification between SA⁰ and the subject at [Spec, TopP], achieved via the checking of the [uperson] feature on the subject by SA⁰. Last, the speaker-addressee relation is structurally represented by the speech act layer that externally merges to ForceP and enters the syntax-pragmatics interface via the Bidirectional Agree between SA⁰ and Force⁰.

(51)

The proposed structure of (50a)