Modality and aspect and the thematic role of the subject in Late Archaic and Han period Chinese: obligation and necessity

2.1 The interrelation of aspect and modality

In this representation, t₁ refers to the initial point of the approach/incremental phase E₁; the point t_m refers to the initial point of the second, the resultative phase E₂; t_n refers to a final point of the situation. The point t_m belongs to both phases. This structure accounts, e.g. for the LAC achievement verb 死 sǐ ‘die’ in example 4a which refers to E₂ and to the accomplishment verb 築 zhú in 4b which can refer to E₁ and E₂ depending on the grammatical construction it appears in. One of the distinguishing features between these two categories is the role of the subject. Achievement verbs are usually considered to have a theme subject, and accomplishments have a causer or agent subject.⁶ Note that neither of those constructions has to be marked for aspect by any overt morphological means.

Non-terminative monophasic verbs only consist of a process or state part: E₁ and E₂ are assumed to be identical. The structure of 3b1 accounts for non-terminative (−TELIC) (in)transitive verbs such as live and push, respectively (for push see example 2), 3b2 accounts for state verbs such as 高 gāo ‘high’ in LAC and EMC⁷ in example 4c.

In general, events (accomplishments and achievements) focusing on either t₁ or t_m are compatible with the perfective aspect; states and activities focusing on neither of the final points of a situation are compatible with the imperfective aspect. The interplay of the verb and its arguments, and additionally the employment of adverbs which can e.g. express perfective and imperfective meanings play an important role in the determination of the lexical aspect and in aspectual shifts in Han period Chinese.⁸

For the reading of modal verbs, Abraham proposes a structure similar to that of telic (perfective) and atelic (imperfective) verbs. According to him deontic events are bi- phasic, corresponding to [+TELIC/TERMINATIVE] events, and epistemic events are monophasic corresponding to [−TELIC/TERMINATIVE] events. These structures are summarized in Leiss (2008: 17) in the following way:

In the perfective event, the incremental phase E₁ corresponds to the deontic event; the resultant state E₂ corresponds to the projected (resultative) future event. In epistemic events just as well as in imperfective events, this distinction into two separate phases cannot be observed.

The precise function and the status of this prefix as expressing either aktionsart or perfective aspect are subject to debate (see Besch et al. 2003: 2520; Vogel 1995: 178). The employment of the Germanic aspectual suffix seems to be less obligatory than the systematic and obligatory marking of the perfective and the imperfective aspect in the Slavic languages (Dal and Eroms 2014: §78). Nevertheless, there are some arguments in favour of the hypothesis that Old Germanic had a grammatical aspectual system (Besch et al. 2003), although it cannot be entirely excluded that an analysis of ge- as marker of the lexical aspect (situation type) is more conclusive. Aktionsart morphology is derivational in nature and can be marked or unmarked,¹¹ whereas the viewpoint aspect has to be marked obligatorily in languages with a grammaticalized perfective-imperfective distinction. When the Germanic languages lose the former category of aspect (especially the perfective ge-verbs), they start to develop an elaborate class of deontic and epistemic modal verbs. Modal distinctions had previously been expressed by the interplay of aspectual and temporal marking alone. The diachronic development in the Germanic languages in contrast to other IE languages obviously points to a close and possibly universal relationship between the categories aspect (lexical and/or grammatical aspect) and modality.

The hypotheses on the interplay of aspect and modality presented above are according to Abraham not unchallenged and emerge most obviously from the German data (Abraham and Leiss 2008: xxii). Nevertheless, they will be taken as a point of departure in this paper for the analysis of the Chinese data despite the differences between the aspectual and modal system of Chinese and that of the languages studied in Abraham and Leiss.

2.2 Aspect in Han period Chinese

The first of these processes, the 四聲別意 sì shēng bié yì ‘derivation by tone change’ (e.g. Sagart 1999: 131), is the most prominent and most widely accepted derivational process of Archaic Chinese. It is attested with words of any of the tonal categories A (平 píng), B (上 shǎng), and D (入 rù), which are transformed into category C (去 qù).

The examples demonstrate that according to the reconstructions based on reading variants reported in traditional Chinese lexicography Archaic Chinese might have had several different affixes expressing aspectual variations and related meanings. These are (a) the (more frequent and) generally accepted suffix *-s and (b) a sonorant prefix, reconstructed, e.g. as *ɦ-, deriving the unaccusative, resultative variant from a transitive variant (Baxter 2000; Pulleyblank 1973) (and/or a causative prefix *s-, deriving a causative variant from an unaccusative variant (Mei 2015)). Analyses of the morphology of a large number of representative verbs (for instance in Jin 金理新 2006) provide strong evidence for a morphological differentiation of different verbal aspects, or rather of activity and accomplishment verbs on the one hand and (resultant) states on the other.²³ This distinction based on telicity features provides an argument for a localization of the aspectual morphology in an Inner Aspect Phrase within an articulated VP (vP), as proposed in Meisterernst (2016b) following Travis (2010).²⁴ Additionally, the aspectual morphology of many of the verbs discussed in Jin 金理新 (2006) shows a close relation with the internal argument of the verb. This also supports the hypothesis that it rather concerns the category lexical aspect, usually characterized by derivational morphology, and not the category grammatical aspect. This also accounts for the function of some of the other affixes reconstructed, e.g. in Sagart (1999). The abstract morpheme, which can phonologically be represented by the suffix *-s, or, possibly, by the voicing alternation, reconstructed to be caused by a sonorant prefix *ɦ-, may be labelled with the feature [+RESULT].²⁵Additional to the mentioned affixes a null morpheme referring to telicity features has to be assumed; different items, including the item zero ø, can be inserted due to phonological and other rules of the respective language.²⁶ Functionally, these derivational affixes show a close resemblance to the Germanic prefix ge−/ga-, for which similar functions have been proposed (see Section 2.1). However, the linguistic data attested does not show precisely how long and to which extent these morphological distinctions were productive. During the LAC period, this system had certainly not been productive any longer and started to lose its transparency. At that time, verbs without an overt resultative marker and verbs in the qùshēng reading can identically refer to non-resultative accomplishments. Schuessler (2007: 46) shows that a derived verb in the qùshēng could become an independent, e.g. transitive verb of its own accord.

The loss of the morphology of Archaic Chinese can serve as an explanation for the significant structural changes Chinese underwent from LAC to EMC, e.g. the change from a more synthetic to a more analytic language; this includes the emergence of light verbs, resultative constructions and disyllabification processes. (see e.g. Feng 2014; Hu 2016; Huang 2014; and others). In the aspectual system, distinctions are increasingly expressed by lexical means, i.e. by aspectual adverbs and possibly by sentence final particles, before a new structure for the marking of aspect develops in the EMC period.

From the end of the LAC period, the perfective adverbs²⁷ 既 jì and 已 yǐ appear freely with both atelic and genuinely telic verbs. The atelic verbs are confined to stage-level predicates, i.e. predicates which allow a change of state reading. With state verbs, the adverbial modification induces an inchoative reading and with activity verbs the natural endpoint of the activity is focussed on (Meisterernst 2015a, 2016b). For the verb zhì 治 ‘govern(ed)’ in example 9, the morphological distinctions might still have been transparent at this time. Accordingly, the ratio of instances modified by an aspectual adverb is very low (less than 1% in LAC).²⁸ In 9a, a combination of the perfective adverbs 既 jì and 已 yǐ modifies the unaccusative variant 治 zhì; in 9b and 9c, a perfective adverb modifies transitive predicates with the verbs 治 zhì (chí) and 并 bìng ‘unify(ed)’, respectively. Since causative morphology is also probably generated in the Inner Aspect Phrase, causative examples like 9b and 9c provide an additional argument for the aktionsart hypothesis and the generation of the verbal morphology in an Inner Aspect Phrase (Meisterernst 2016b). Note that all examples are comparatively late. Example 9d represents the non-default employment of a perfective adverb with a state verb²⁹; modified by a perfective adverb a coercion effect is induced, the initial point of the state is focused leading to an inchoative reading (Meisterernst 2016b).

Both the grammatical and the lexical aspect convey information about the temporal structure of a situation and they are closely linked in a compatibility relation. The respective lexical aspect of the verb enhances or prevents a particular grammatical aspectual representation: telic verbs are generally compatible with the perfective aspect, and atelic (state and activity) verbs are generally compatible with the imperfective aspect. These default relations can be shifted by coercion effects from telic to atelic and vice versa when modified accordingly (Meisterernst 2015a, 2016b).³⁰

Another process which can be ascribed to the loss of morphology in LAC and EMC is the disyllabification process in Chinese together with the development of resultative constructions in EMC; some of the latter serve as the source structures of Modern Mandarin aspectual markers.³¹In Modern Mandarin, the aspectual meaning of perfectivity can be expressed by the verbal suffix -le 了. Sybesma (1997, 1999) proposes an analysis of –le as denoting an endpoint or realization. Sybesma (1999: 72) treats LE as a ‘neutral telic marker’, and he analyzes both types of LE as small clause predicates. Sybesma (1994) proposes that the aspectual marker –le in Modern Mandarin and its diachronic development can actually be compared to the Germanic prefix ge-, which expressed completion in Middle Dutch (Sybesma 1994: 41). This claim supports the hypothesis proposed in this paper that a system of derivational affixes can be reconstructed for Archaic Chinese which displays functions comparable to those proposed for the Germanic aspectual prefix ge/ga-. The loss of the aspectual function of this prefix in Old German has been connected to the development of the German modal system; a similar process might also have taken place in EMC, i.e. in a language typologically entirely different from the Germanic languages. In Chinese, the complexity of the modal system only starts to increase in at the end of LAC, the time when the functions of the reconstructed affixes of Chinese started to lose their transparency. As in the early stages of the German language, modal verbs in LAC are to a considerable extent confined to different realizations of the ‘first modal’ (Leiss 2008: 16) ‘can’.³² True deontic verbs only emerge in the EMC period, and the epistemic readings of modal verbs develop even later; this is the typical grammaticalization path for modal verbs. In LAC, epistemic modality is expressed by sentential adverbs; these have to be distinguished semantically and syntactically from modal verbs (Leiss 2008; Meisterernst 2016a). Sentential adverbs modify an entire proposition and are thus less confined in their constraints than modal auxiliary verbs; they are attested with both atelic and telic verbs.³³

2.3 Modality in Han period Chinese

Consequently, the subject plays a crucial role for the analysis of the modal as an epistemic, true deontic (performative), or a circumstantial modal. One of the constraints on a subject of a deontic modal is that it has to be [+HUMAN] (e.g. Meisterernst 2010, 2011). Performativity can only be assumed in cases in which the individual under obligation, i.e. the agent of the verb embedded by the modal, is the addressee or another participant in the conversation (Portner 2009: 189). In this regard, performatives can act like imperatives. This also accounts for the LAC and EMC data. In LAC and EMC, circumstantial and deontic modal values are partly expressed by the same verbs in apparently identical syntactic positions. However, differences can be observed in their scope relations with regard to negation (Meisterernst 2016c); this leads to the assumption that they can also be distinguished syntactically. This has the effect that the modal verb 可(以) kě(yǐ), a circumstantial modal in its affirmative variant, occupies a position different from and lower than its negated deontic variant NEG 可(以) kě(yǐ) NEG. Epistemic modals (i.e. modal adverbs)—and the few deontic modal adverbs—operate on the level of CP (Complemetizer phrase). These facts confirm Hacquard’s hypothesis, but also the hierarchy of functional heads proposed in Cinque (1999).

In the following, only modal auxiliaries, which can express root or deontic modal values in LAC and EMC, will be discussed with particular focus on the lexical aspect of the complement they select. Of the modal verbs expressing possibility, only 可/可以 kě/kěyǐ will be included. The discussion will predominantly be based on the claims proposed in Abraham and Leiss for the relation between the system of the aspectual and the modal markers. It will be proposed that in Chinese, too, a close relation exists between the modal readings and the aspectual (aktionsart) features of the verb. According to the arguments presented in Section 2.2, it will be hypothesised that it is the lexical aspect [+/− TERMINATIVE] instead of the grammatical aspect [+/−PERFECTIVE] which provides a cue for the modal interpretation of the predicate to the effect that the root modal reading is more natural with an event complement than with an atelic complement. The increase in complexity in the modal system of MC observable in the early Medieval Buddhist literature might have been triggered by the loss of the morphological marking of the lexical aspect, similar to the Germanic languages. But this issue still requires more research. Additionally, observations with regard to the Japanese system will be taken into consideration. In Japanese, the aspect-modality link is supposed to be motivated by general cognitive principles (see Abraham and Leiss for Japanese (2008: xix): “The temporality of root modal sentences differs from epistemic modal sentences in that deontic modal sentences require “Speech act time ≠ Event time”, while there are no such restrictions on sentences with epistemic modals… the crucial factor being (not “temporal”, but) “time” referential, rather than aspectual.” According to Abraham and and Leiss in Japanese, the “grammatical aspect only provides a cue to modal interpretation” (Narrog 2008: 279), but does not determine it.

3.1 The modal auxiliary verb 可 kě with double negation: 不可不 bù kě bù, 不可以不 bù kě yǐ bù ‘must’³⁹

In example 14d both, a deontic, future-projecting reading in a strict sense and a generic reading are possible. The deontic reading refers to the particular requirement of individualized situations in the future and the generic reading to general rules and regulations. According to Ziegeler (2008: 55), ‘potentiality’ is the ‘common semantic denominator’ of normative generic expressions and deontic modality.

Thus, 可 kě requires a patient/theme subject and a resultative complement on a regular basis. In both constructions, most of the complements selected refer to events or to states resulting from a previous event either in their transitive or their passivized (or unaccusative) forms. The complement can refer either to E₁ (including t_m) or to E₂ (including t_m) with verbs which have the structure proposed for event (terminative) verbs in Abraham and Leiss (2008: xiii). Temporally, they all have the characteristic: S ≠ E (speech time is not identical with, i.e. it precedes event time),⁴⁵ the structure proposed for deontic modality in Japanese by Narrog (2008), the general structure for deontic modality which typically refers to an obligation performed in the future.

(15)

With a passivized complement, the modal is exclusively speaker oriented and with an agentive complement, it is speaker–addressee oriented; this argues for its analysis as a true deontic modal according to Hacquard (2006).

3.2 The modal verb 當 dāng

The modal function of 當 dāng grammaticalizes from a verb with the basic meaning ‘match, correspond’.⁴⁶ As a modal auxiliary verb, it expresses root necessity: □p ‘it is necessary that p’, roughly corresponding to the modal should in English. In this function, it is regularly attested from the Han period on.⁴⁷ Although it can be employed in true performative deontics, it predominantly appears in indirect suggestions; the agent is frequently unspecified. The verb in its complement is mostly a telic agentive verb in transitive or derived, i.e. passivized/resultative constructions; the obligation is based on laws, rules and norms. Contrary to the strong deontic construction NEG 可(以) kě(yǐ) NEG vP and to the modal auxiliary verb 必 bì, the speaker does not necessarily expect compliance on the side of the frequently only implied agent. As with should in English, the modal force of obligation is weaker than with must. The strength of obligation is induced by the strength of the ordering source for the modal. With strict laws, these ordering sources imply a stronger obligation than with what is, e.g. predetermined by destiny (Meisterernst 2011, 2012). Epistemic values are confined to 當 dāng in the complement of an epistemic, an attitude verb, and do not depend on the modal. After the Han period, the employment of 當 dāng changes, and in the Buddhist literature, 當 dāng tends to express more direct obligations, i.e. performatives. These are frequently characterized by a second or third person subject referring to the addressee and the specified agent of the required action; in these cases, the speaker and the agent of the requested action are not identical.

In the examples in 16, an event verb appears in a passive construction with a theme subject. Although the structure is similar to that of 可 kě with a passive complement, the passive reading is not required syntactically, but depends on the role of the subject. The modal predicate is future-projecting and the complement of 當 dāng refers to a resultant state and to the process leading up to it, i.e. it is [+TELIC]. But at this period, any morphological marking of the resultant state was certainly no longer transparent for the speaker.⁴⁸ Although these examples evidently represent cases of deontic modality, the identification of a particular agent is explicitly avoided. This employment is most typical for 當 dāng in Han Chinese.

The examples in 17 are both transitive and agentive; the agent of the verb is identical to the addressee of the obligation. Example 17a represents one of the less frequent cases in Han period Chinese in which a direct, though polite command, is issued by 當 dāng. In example 17b, the reference time is located in the past and precedes the speech time, and two different times are involved in the modal predicate.⁴⁹ Nevertheless, the modal is still future-projecting.⁵⁰

The example in 18 has [−HUMAN] experiencer subjects; the verbs are intransitive. The verbs in the first clause 衰 shuāi ‘decline’, and 亂 luàn ‘(cause to) be in disorder’ can be both atelic or telic. The verb 治 chí/zhì ‘put in order, govern’, ‘well-governed, in good order’ (see ex. 7) belongs to the verbs for which an aspectual morphology has been reconstructed. The semantic features of the subject are in general assumed to be more typical for epistemic than for deontic readings, but all complements include an event variable and refer to resultant states which are [+TELIC]. Thus, they do not differ significantly from some of the examples discussed above. The modal is future-projecting.

All instances of 當 dāng presented above are future-projecting; the obligation imposed can refer to an agentive event, but also to a future resultant state and the process leading up to it without any agency involved. The latter instances are similar to those with 可 kě with a passivized (unaccusative) resultative complement. In past tense contexts, 當 dāng obtains a counterfactual reading.⁵¹ According to Sparvoli (2015), a counterfactual reading in the past is the typical actuality entailment for deontic modals and one of the possible readings (besides an epistemic reading) in past contexts. The past context can, but does not have to be explicitly marked. In example 19a, the event preceding the modal predicate is marked as completed by the aspectual adverb 已 yǐ; in 19b, the event is located in the past by the adverbial 先 xiān ‘earlier’ in the complement of 當 dāng.

The modal predicate is always future-projecting, although the prospective event is located in the past. The speaker refers to a reference time preceding speech time which serves as the point of reference for the projected future situation time. The counter-factual effect in past tense contexts is produced when both reference time and situation time precede speech time.

The temporal structure of the complement of 當 dāng resembles that of 可/以) kě(yǐ) (E ≠ S). Even if the modal itself is located in the past, i.e. it precedes speech time, the complement is still future-projecting, i.e. it follows reference time and speech time: S,R_E.

(20)

Identical to 可 kě, the modal is exclusively speaker oriented with an unaccusative or passivized complement. With an agentive complement, it is speaker–addressee oriented, this argues for its analysis as a true deontic modal according to Hacquard (2006).

3.3 The modal 必 bì expressing deontic and epistemic modality

Abraham and Leiss (2009) claim that the category source is the distinctive feature of epistemic verbs and epistemic adverbs. Epistemic marking by adverbs does not include a source of information for the epistemic evaluation of the speaker: epistemic adverbs are monodeictic, while epistemic verbs are bi-deictic (Abraham and Leiss 2009: 13) including both the speaker evaluation and the source. The fact that only the speaker evaluation is included in epistemic adverbs can also argue for their availability for expressing future reference, contrary to epistemic verbs.

Although 必 bì seems to be the only direct marker of strict deontic modality in LAC and Han Chinese, the function as an epistemic modal adverb expressing (mostly) future certainty is evidently its predominant function from the earliest instances on. This function of 必 bì can be accounted for by what Coates labels ‘pure logical necessity’, expressing confidence in a logical necessity on the side of the speaker. To obtain a deontic reading of 必 bì, a causative/agentive subject is a necessary condition. This is evidenced by the contrastive examples in 22a and 22b, which contain the verb 立 lì ‘set up, establish’, a default [+TELIC] verb. In example 22a with a non-overt causative/agentive addressee subject, 必 bì has a deontic reading, conveying a direct obligation to the addressee. In 22b with a theme subject, it is epistemic referring to the speaker’s commitment to a future necessity. The verb in 22b has a passive reading. Although a passive reading of the complement is quite natural with the root modal auxiliaries 可以 kě(yǐ) and 當 dāng, this is not the case with deontic 必 bì which requires an agent or a causer subject for a deontic reading.

Although an agentive subject is a necessary condition for the deontic reading, a non-agentive subject is not a necessary condition for an epistemic reading. Both 22c and 22d contain a causative/agentive subject and the [+TELIC] verb 救 jiù ‘rescue’, but 22d has an epistemic reading. The subject of the modal predicate is a third person [+/−HUMAN] subject which renders a deontic interpretation less likely.⁵⁴

3.3.2 Typical instances of the epistemic adverb 必 bì (modal adverb)

The epistemic modal 必 bì is predominantly attested in future-projecting contexts in matrix clauses; it most typically occurs in the apodosis of a conditional or concessive sentence.⁵⁷ The fact that epistemic 必 bì is mostly future-projecting argues against a polysemic modal auxiliary verb 必 bì comparable to the English must expressing both deontic and epistemic values. Future readings are in general not available for the epistemic reading of these verbs (e.g. Coates 1983; Meisterernst 2010; Palmer 2001; Ziegeler 2008). But they are not blocked from epistemic adverbs expressing certainty. According to Nuyts (2001: 77), the appearance of a modal in the apodosis of a conditional sentence argues particularly for an adverbial analysis, expressing “the speaker’s present evaluation (performatively)” of the probability that a particular state of affairs will come about under the conditions given in the protasis (ibid). This definition evidently supports an analysis of 必 bì as a modal adverb ‘certainly’ in most of the cases presented below. In future contexts, the speaker does not relate his commitment to the truth of his deductions from known facts; this would be the default function of an epistemic auxiliary verb. He rather conveys his confidence that under certain conditions (which are not yet true in the real world), his deductions will be true, i.e. “that a certain hypothetical state of affairs under consideration … will occur” (Nuyts 2001: 21). With an epistemic adverb, no source for the commitment is involved (Abraham and Leiss 2009). Epistemic adverbs appear very high in the syntactic structure. They take an entire proposition as their complement; consequently, they are less confined in their selectional restrictions (see Meisterernst 2016a).

The verb 喜 xǐ in example 25 is a genuine state verb; it does not combine with a perfective adverb in LAC and EMC.⁵⁸ Genuine state verbs support an epistemic reading with modal auxiliaries. Additionally, genuine intransitive state verbs do not have an agent or causer subject. The example does not refer to an obligation in the real world, but is assumed to be true by the speaker under the conditions specified in the conditional protasis. All of these argue for an analysis of 必 bì as an epistemic adverb.

1. (25)

   今王事秦, 秦王必喜, 趙不敢妄動。 (Shĭjì: 70; 2298)

   jīn__wáng__shì__qín,__qín__wáng__bì__xĭ,__zhào__bù__găn__wàng__dòng

   Now__king__serve__Qin,__Qin__king__BI__happy,__Zhao__NEG__dare__rash-move

   If you now serve Qin, the King of Qin will certainly be happy, and Zhao will not dare to move rashly.

    

In example 26, future certainty regarding the occurrence of an adverse situation is expressed. The fact that the subject is a [−HUMAN] theme argues against a deontic reading of the modal despite the [+TELIC] verb.

1. (26)

   齊秦合則患必至矣。 (Shĭjì: 70; 2287)

   qí__qín__hé__zé__huàn__bì__zhì__yĭ

   Qi__Qin__join__then__trouble__BI__arrive__SFP

   If Qi and Qin ally, then trouble will certainly arrive.

    

In the examples in 27, nothing argues against a deontic interpretation on a par with example 23b with a first person agentive subject. All verbs are event verbs. Time span adverbials as in example 27a do not provide an argument against a deontic interpretation, since they combine with events identical to deontic modal auxiliary verbs. In this example, the speaker conveys his confidence as a supporting argument for the performative acts, whereas in 27b he conveys his confidence in the occurrence of a future event according to the conditions related in the respective protases. All propositions are future-projecting.

1. (27)

   a. 慎勿與戰, 毋令得東而已。我十五日必誅彭越, 定梁地, 復從將軍。(Shĭjì: 7; 329)

   shèn__wù__yǔ__zhàn,__wú__líng__dé__dōng__ér__yǐ.__wŏ__shí__

   careful__NEG_mod__give__battle,__NEG_mod__order__get__east__CON__finish__I__ten_

   wŭ__rì__bì__zhū__péng__yuè,__dìng__liáng__dì,__fù__cóng__jiàngjūn

   five__day__BI__execute__Peng__Yue,__settle__Liang__territory,__again__follow__general

   Be careful not to join them in fight; just do not order them to get [to] the east. I will certainly execute Peng Yue, pacify the territory of Liang and join you, general, again within fifteen days.

   b. 不勝, 則我引兵鼓行而西, 必舉秦矣。 (Shǐjì 7; 305)

   bù__shèng,__zé__wǒ__yǐn__bīng__gǔ__xíng__ér__xī,__bì__jǔ__qín__yǐ

   NEG__win,__then__I__lead__army__drum__march__CON__west,__BI__conquer__Q in__SFP

   If it (Qin) does not win, then we will lead our troops and, following the beating drums, we will march west, and we will certainly conquer Qin.

    

Examples such as 27, display characteristics typical for deontic modal interpretations, i.e. the modal modifies event predicates with telic verbs with [+HUMAN] agent or causer subjects. Nevertheless, in the examples presented, bì evidently expresses epistemic, and not deontic modality, referring to a certainty on the side of the speaker with regard to the occurrence of a future event, frequently under conditions specified in a conditional protasis. In this regard, they show the same orientation, i.e. a speaker orientation as true deontics do according to Hacquard (2006: 114): the latter display a speaker/addressee–orientation (Sp/A-O). Since modal 必 bì also expresses deontic modality, the speaker orientation of propositions with 必 bì seems to be the semantic link between the deontic and the epistemic functions.

The examples show that no selectional restrictions with regard to the lexical aspect of the verb in the complement of epistemic 必 bì exist. In both the deontic and the epistemic reading of 必 bì, the modal predicate is predominantly future-projecting. Additionally, both readings of 必 bì are speaker oriented; they apparently differ in the fact that deontic 必 bì has a strong addressee additional to the speaker orientation. The agent orientation of 必 bì is much stronger than that of the root and circumstantial modal auxiliary verbs 當 dāng and 可 kě; these are frequently explicitly not directed to a specified addressee. Consequently, in the absence of syntactic devices, it is the syntacto-semantic features of the subject which are relevant for a distinction between the verbal and adverbial function of 必 bì: the deontic reading is confined to an addressee subject that functions as an agent or causer. The epistemic function is not constrained with regard to its subject. Due to the particular semantics of 必 bì as a marker of deontic modality with a strong agent orientation, the temporal structure of the complement of deontic bì 必 differs from that of the root modal auxiliary verbs 可(以) kě(yǐ) and 當 dāng. The verbs in the complement of 必 bì are never passivized and do never refer to a resultant state as they do with 可(以) kě(yǐ) and 當 dāng. They can only refer to E₁ and its final point t_m, but not to the resultant state E₂.

(28)

In its epistemic reading, 必 bì is not confined to this temporal structure, and the temporal part E₂, the (resultant) state part, can also be included in the complement of 必 bì; this is another distinctive feature of the two modal readings. As an epistemic adverb, 必 bì can to a certain extent be compared to the other epistemic markers of Han period Chinese. These—together with some other adverbs expressing factivity and other modal values—appear very high in the hierarchy of adjuncts and they take an entire proposition as their complement (Meisterernst 2016a; Wei 魏培泉 1999). Evidently, epistemic adverbs are not subject to the same constraints with regard to the lexical aspect of their complement as modal auxiliary verbs; they operate on a different syntactic level. Accordingly, they do not provide any counter-evidence to the hypothesis proposed by Abraham and Leiss (2008).