3.2.1 Degree modification contexts

Two kinds of degree modification contexts distinguish between adjectives and COS verbs. The classic kind of degree modification with intensifiers such as hěn ‘very’ (see (14) above), fēicháng ‘extremely’ (see (29a)), and post-predicate jí-le ‘to the utmost’ illustrated in (29b), is possible for gradable adjectives, as noted earlier.

1. (29)

   1. (a) shù-yè fēicháng hóng

      tree-leaf extremely red

      ‘The leaves are extremely red.’

   2. (b) shù-yè hóng jí-le

      tree-leaf red utmost-prt

      ‘The leaves are extremely red.’

They cannot, however, modify a predicate such as zuì ‘intoxicated’, and bìng' ‘sick', see (30a, b).

1. (30)

   1. (a) *Sānmáo hěn/fēicháng zuì/bìng

      Sanmao very/extremely intoxicated/sick

      Intended: ‘Sanmao is very/extremely drunk/sick.’

   2. (b) *Sānmáo zuì/bìng jí-le

      Sanmao intoxicated/sick utmost-prt

      Intended: ‘Sanmao is extremely drunk/sick.’

This is not because zuì ‘intoxicated’ or bìng ‘sick’ are non-gradable. Rather, degree modification for these predicates must be expressed in another way. To encode the attainment of excessive intoxication/illness, a -de complement structure is employed, with the modifying phrase hěn lìhai ‘seriously’, as shown in (31).Footnote ¹⁵

1. (31) Sānmáo zuì-de/bìng-de hěn lìhai

   Sanmao intoxicated-vprt/sick-vprt very serious

   ‘Sanmao is intoxicated/sick to a serious extent.’

Now putative adjectives such as gāo ‘tall/high’, hóng ‘red’, bái ‘white’, etc., may also combine with -de hěn lìhai ‘to a serious extent’, as is shown in (32a). Importantly, though, in this case they can only take on a COS interpretation. When modified by hěn ‘very’ (32b), the predicate bái ‘white’ can describe a congenitally high degree of whiteness in hair. When bái ‘white’ occurs with -de hěn lìhai, it must encode a change to a high degree of whiteness, and (32c) is infelicitous because the context indicates the improbable situation of a baby's hair somehow either turning white or having turned white at birth.

1. (32)

   1. (a) Sānmáo de tóufa bái-de hěn lìhai

      Sanmao assoc hair white-vprt very serious

      ‘Sanmao's hair turned drastically white.’

   2. (b) Sānmáo yì shēng-xià-lai tóufa jiù hěn bái

      Sanmao one born-down-come hair JIU very white

      ‘Sanmao's hair was (very) white since birth.’

   3. (c) #Sánmáo yì shēng-xià-lai tóufa jiù bái-de hěn lìhai

      Sanmao one born-come-down hair JIU white-vprt very serious

      ‘Sanmao's hair turned drastically white the moment he was born.’

What does this pattern of interpretation indicate? Only on the stative interpretation can putative adjectives such as hóng ‘red’, bái ‘white’, gāo ‘tall/high’ combine with degree modifiers such as hěn ‘very’. Predicates such as zuì ‘intoxicated’ and bìng ‘sick’ cannot occur with these modifiers, but rather, must be modified with de hěn lìhai ‘to a serious extent’. The same holds of the COS interpretation of hóng ‘red’, bái ‘white’, etc. This indicates that zuì ‘intoxicated’ and the COS interpretation of bái ‘white’ must share some property. Furthermore, this property cannot be shared between zuì ‘intoxicated’ and the stative interpretation of bái ‘white’, since the former cannot occur with hěn ‘very’. I propose that this property is simply that of being a COS verb. Predicates such as bái ‘white’ may correspond either to an adjective or a COS verb. In contrast, zuì ‘intoxicated’ can only be a COS verb.

3.2.2 Negation

The preceding conclusion receives support from how adjectives and COS verbs interact with negation marking. Researchers have noted that the two negation morphemes in Mandarin, bù and méi, combine with predicates of different event structure categories. Different explanations have been proposed for these distinctions (Li & Thompson Reference Li and Thompson1981: 421–440; Huang Reference Huang1988; Ernst Reference Ernst1995; Lin Reference Lin2003), but one clear contrast is that bù is associated with negating the presence of some state, whereas méi is associated with negating the occurrence of an event (Lin Reference Lin2003). This point is demonstrated by Lin with the examples in (33), with lǎo ‘old’. With bù, lǎo ‘old’ has a stative meaning, see (33a), and with méi, it indicates COS, see (33b) (see also Liu Reference Liu2010: 1036).

1. (33)

   1. (a) tā kàn-shàngqu yì diǎn dōu bù lǎo

      3sg look-appear one little all neg old

      ‘He is not old at all in appearance.’ (Lin Reference Lin2003: 437, ex. (23))

   2. (b) tā kàn-shàngqu yì diǎn dōu méi lǎo

      3sg look-appear one little all neg old

      ‘He hasn't become old at all in appearance.’ (Lin Reference Lin2003: 437, ex. (24))

Turning to basically COS predicates, we find that they cannot be negated using bù in a declarative sentence, as shown in (34a).Footnote ¹⁶ Rather, to claim that someone is not drunk, méi must be used (34b).

1. (34)

   1. (a) *Sānmao bú zuì

      Sanmao not intoxicated

      Intended: ‘Sanmao isn't/didn't get drunk.’

   2. (b) Sānmáo méi zuì

      Sanmao not intoxicated

      ‘Sanmao isn't/didn't get drunk.’

Converging with this contrast, Lin (Reference Lin2003) notes that certain adjectives, such as cōngming ‘intelligent’ – essentially those that describe human propensities (Dixon Reference Dixon1982: 21) – cannot be negated by méi, as shown in (35b). This would mean that such adjectives have no COS counterparts.

1. (35)

   1. (a) tā bù cōngming

      3sg neg intelligent

      ‘He is not clever.’

   2. (b) *tā méi cōngming

      3sg not intelligent

      ‘He has not turned clever.’ (Lin Reference Lin2003: 437, ex. (25))

The negation patterns support the hypothesis that verbs such as zuì ‘intoxicated’ are basically COS, and thus also eventive, since they allow negation by méi but not by bù. They also support the related hypothesis that predicates such as lǎo ‘old’ (also gāo ‘tall/high’ and hóng ‘red’, discussed above) have both state and COS meanings. Finally, human propensity adjectives such as cōngming ‘intelligent’ seem to be stative only, as also suggested by their incompatibility with -le (Liu Reference Liu2010: 1035):Footnote ¹⁷

1. (36) *Zhāngsān cōngming/bèn le

   Zhangsan smart/stupid perf

   ‘Zhangsan got smart/stupid.’ (Liu Reference Liu2010: 1035, ex. (83a))

3.2.3 Interim summary

To sum up, predicates such as lǎo ‘old’, hóng ‘red’, guì ‘expensive’, zāng ‘dirty’, and pàng ‘plump, fat’ occur with degree modifiers such as hěn ‘very’ and fēichang ‘extremely’, and are negated with bù, on a stative interpretation. In such cases, the predicate may be assumed to be an adjective. These predicates may also occur with méi ‘not’ and be modified with -de hěn lìhai ‘to a serious extent’, but only on a COS interpretation. In such cases, we may consider them COS verbs. In contrast, verbs such as zuì ‘intoxicated’ and bìng ‘sick’ are basically COS. They cannot be modified with hen ‘very’ and fēichang ‘extremely’, but only with -de hěn lìhai ‘to a serious extent’. They can only be directly negated by méi, and not by bù. These patterns are set out in (37):

1. (37) Modification patterns of adjective/COS verb alternation pairs and basic COS verbs

   

Strictly speaking, these patterns allow us to conclude only that hóng ‘red’, gāo ‘tall/high’, etc. alternate between state and COS, whereas predicates such as zuì ‘drunk’, bìng ‘sick’, etc. are simply COS verbs. They do not immediately lead to the conclusion that the stative meanings of the alternating predicates are rendered by adjectives. That is, these diagnostics per se cannot yet rule out the possibility that hóng ‘red’, gāo ‘tall/high, etc. are simply verbs with both state and COS interpretations. Below, I show that these predicates can and should be distinguished from stative verbs.

4.1.1 COS verbs in predicative position

The degree modification and negation contexts used to distinguish adjectives from intransitive COS verbs in the last section reflect the eventive nature of these verbs. Using these contexts as diagnostics, predicates such as liè ‘crack’ in (11) above and suì ‘shatter’, which may be modified by -de hěn lìhai ‘to a serious extent’ but not by hěn ‘very’, and are negated by méi but not by bù, may be classified as COS verbs.Footnote ²² At least prima facie, there is no reason to believe that such verbs have a (zero-)derived stative form.

Now, RVCs such as qiāo-suì ‘knock-shatter’ in (46a) below may not be modified by hěn ‘very’ (presumably because they are eventive), nor by -de hěn lìhai ‘to a serious extent’ (probably because they are not gradable). Still, like intransitive COS verbs such as liè ‘crack’ and suì ‘shatter’, they are also negated with méi rather than bù, as seen in (46b), indicating that they are non-stative.

1. (46)

   1. (a) Sānmáo qiāo-suì-le bēizi

      Sanmao knock-shatter-perf cup

      ‘Sanmao knocked on the cup, causing it to shatter.’

   2. (b) Sānmáo méi/*bù qiāo-suì bēizi

      Sanmao neg knock-shatter cup

      ‘Sanmao didn't cause the cup to shatter by knocking on it.’

That is, causative RVCs differ from COS verbs in having a causer argument, but otherwise are equally consistently eventive with respect to the relevant diagnostics. I now show that detransitivized forms of causative RVCs are also eventive and not stative.

Causative RVCs in Mandarin show a transitivity alternation demonstrated in (47):

1. (47)

   1. (a) Sānmáo qiāo-suì-le bēizi

      Sanmao knock-shatter-perf cup

      ‘Sanmao knocked on the cup, causing it to shatter.’

   2. (b) bēizi qiāo-suì-le

      cup knock-shatter-perf

      ‘The cup got knocked on and shattered.’

Tan (Reference Tan1991) argues that these are zero-derived passive verbs, with no agent phrase realized. A similar assumption is made in Ting (Reference Ting2006), where these verbs are called middles. For current purposes, the actual label is unimportant. The crucial point is that Mandarin RVCs show detransitive uses in which the patient nominal is realized as subject (see Tan Reference Tan1991 and Ting Reference Ting2006 for arguments to this effect).

These detransitive uses retain an agentive meaning, however, as illustrated by their compatibility with agent and instrument phrases (Keyser & Roeper Reference Keyser and Roeper1984), in (48a). In contrast, (48b) shows that the COS verb suì ‘shatter’ itself is non-agentive.

1. (48)

   1. (a) bēizi yóu tāmen/yòng chuí-zi qiāo-suì-le

      cup by 3pl/use hammer knock-shatter-perf

      ‘The cup was shattered by them/with a hammer’.

   2. (b) *bēizi yóu tāmen/yòng chuí-zi suì-le

      cup by 3pl/use hammer shatter-perf

      ‘The cup shattered *by them/*with a hammer.’

Unsurprisingly, the agentive detransitive RVC is also eventive. Like suì ‘shatter’, discussed above, and transitive qiāo-suì ‘knock-shatter’ (see (46b)), the detransitived RVC qiāo-suì ‘knock-shatter’ in (49) is also negated by méi.

1. (49) bēizi méi/*bù qiāo-suì

   cup neg knock-shatter

   ‘The cup didn't get shattered from being knocked on./The cup was not knocked on.’

Such examples suggest that detransitive RVCs in predicate position, like their transitive counterparts and like intransitive COS verbs, do not have stative readings. I assume that qiāo-suì ‘knock-shatter’ in (49) has a lexical representation corresponding to a passive verb, along the lines of (50), with its agent argument existentially bound.Footnote ²³

1. (50) λy λs λe′ λe ∃x [become(s)(e) ∧ shattered′(y)(s) ∧ cause(e′, e) ∧ knock′(x, y)(e′)]

I show further below that such predicates in attributive position concomitantly do not show adjectival behaviour in general.

4.1.2 No RVC-derived adjectives

In this subsection, I turn to RVCs in attributive position, and show that these verbs do not lend themselves to deverbal adjective formation. Two points demonstrate this. First, RVCs can only modify nominals in the presence of de, and second, they may always retain their original agentive and eventive interpretations. The examples in (51) show that an RVC such as dǎ-pò ‘break’ (lit. ‘hit-break’) cannot modify a noun in the absence of de.Footnote ²⁴

1. (51)

   1. (a) yì zhī dǎ-pò de bēizi

      one cl hit-break assoc cup

      ‘a broken cup’

   2. (b) *yì zhī dǎ-pò bēizi

      one cl hit-break cup

      ‘a broken cup’

The following examples show that an RVC in prenominal position continues to be agentive (52a), and eventive (52b).

1. (52)

   1. (a) yì tái yóu xīn yuángōng nòng-huài de jīqì

      one cl by new employee make-bad assoc machine

      ‘a machine that was broken by (a) new employee(s)’

   2. (b) yì tái méi/*bú nòng-huài de jīqì

      one cl NEG make-bad assoc machine

      ‘a machine that was not broken’

These converge with the obligatory presence of de. While any phrase may combine with de to modify a noun (20c), only adjectives may directly modify a noun without de. That RVCs cannot do so suggests they are verbs here. That they allow agentive and eventive interpretations indicates the same. That is, examples such as (51a) contain a relative clause modifier. Below, I show that COS verbs are somewhat more flexible than RVCs, but they also do not productively form the base for a derived adjective.

4.1.3 No COS-derived adjectives

I now show that basic intransitive COS verbs do not regularly give rise to adjectives. I argue that although some basic COS verbs appear to engage in adjectival behaviour, this pattern is not productive, and in some cases, must receive a V–N compounding analysis.

I demonstrate this generalization first with the verb miè ‘extinguish’ in (26b), repeated here.

1. (26)

   1. (b) huǒ miè-le

      fire extinguish-perf

      ‘The fire went out.’

Unlike zuì ‘intoxicated’ and bìng ‘sick’ discussed in Section 3 above, the verb miè ‘extinguish’, is not gradable, as shown in (53a). It is, however, negated using méi rather than bù, as (53b) shows, consistent with its COS interpretation in (26b).Footnote ²⁵

1. (53)

   1. (a) *huǒ miè-de hěn lìhai

      fire extinguish-vprt very serious

   2. (b) huǒ méi/*bú miè

      fire neg extinguish

      The fire isn't extinguished.

(53b) indicates that miè is a COS verb. The contrast in (54) below shows that miè ‘extinguish’ cannot directly modify a noun, see (54a). The verb miè ‘extinguish’ can only occur prenominally within a relative clause, as in (54b), which the preferred presence of -le further suggests.

1. (54)

   1. (a) *yì gēn miè làzhú

      one cl extinguish candle

      Intended: ‘an extinguished candle’

   2. (b) yì gēn miè(-le) de làzhú

      one cl extinguish-perf assoc candle

      ‘an extinguished candle’

Some COS verbs such as pò ‘break’ and huài ‘break down’ in (55) sometimes seem able to directly modify a noun, but I would argue that, to the extent that pò ‘break’ and huài ‘break down’ are adjectival, it is probably via some kind of pragmatic coercion, or due to analogy with homophonous adjectives, which I describe immediately below.

1. (55)

   1. (a) ?yì zhī pò bēizi

      one cl break assoc cup

      ‘a broken cup/a shabby cup’

   2. (b) ?yì tái huài jīqì

      one cl break.down machine

      ‘a broken down machine’

   3. (c) …tuī-zhe huài chē lái zhǎo nǐ

      push-dur break.down car come seek 2sg

      ‘pushing a broken down car to come looking for you.’ (PKU)

Several intransitive predicates, including pò ‘shabby/break’ and huài ‘bad, wicked/break down’, actually each have two senses, one corresponding to an adjective and one to a COS verb. The form pò may be adjectival, meaning ‘shabby, of poor quality’, or it may be a COS verb, meaning ‘break’. Consistent with the current assumptions, the adjectival sense occurs felicitously in attributive position without the associative marker de when modifying a nominal, but the verbal sense is far less felicitous in this position.

A search on the PKU corpus for the two senses of pò provides support for this interpretation. A search for pò fángzi ‘shabby house’, which shows the adjectival sense of pò, yielded 84 examples in the relevant sense. A search for pò bēizi ‘broken cup’, which taps the COS sense, yielded only two relevant examples, and only one of these clearly involved a broken cup, while the other could be interpreted as describing a cup in poor condition. Modification contexts suggest the same division. The sixteen instances of hěn pò in the PKU corpus could all be interpreted as ‘very shabby’. The objects described included machines, vehicles, furniture, books, shoes and cloth items, but not brittle objects such as cups and vases that typically break or shatter. In cases involving items made from cloth or paper, pò could also have meant ‘torn’, but each example was equally compatible with a ‘shabby’ interpretation. To supplement the PKU data, I also conducted informal Google searches for certain strings. Tellingly, a Google search for huāpíng hěn pò ‘vase very shabby’ turned up only one example, and that single instance clearly meant that the vase in question was of poor quality. A Google search for bēizi hěn pò returned 10 usable results.Footnote ²⁶ Of these, there were six unique examples, all with a ‘shabby cup’ interpretation. A PKU search for pò de hěn lìhai ‘badly broken’ returned no results, but a Google search returned numerous instances, many describing human skin being broken, a situation hardly conceivable as shabbiness. This result is compatible with the understanding that -de hěn lìhai ‘to a serious extent’ modifies a COS verb (see Section 3.2.1 above), which in the case of pò would correspond to the ‘break’ sense rather than the ‘shabby’ sense.

Similarly, the form huài may correspond to either a COS verb meaning ‘to break down’ (as of machines), or an adjective meaning ‘bad’ or ‘wicked’ (of people).Footnote ²⁷ This distinction underlies the contrast in acceptability of the examples in (56)–(57).

1. (56)

   1. (a) *zhè gè nàozhōng hěn huài

      this cl alarm.clock very bad

      Intended: ‘This alarm clock is very faulty.’

   2. (b) zhè gè nàozhōng huài-de hěn lìhai

      this cl alarm.clock bad-vprt very serious

      ‘This alarm clock is badly broken.’

1. (57)

   1. (a) zhè gè rén hěn huài

      this cl person very bad

      ‘This person is wicked.’

   2. (b) *zhè gè rén huài-de hěn lìhai

      this cl person bad-vprt very serious

      Intended: ‘This person became seriously wicked.’

The sense of huài describing mechanical malfunction cannot be modified by hěn ‘very’, as is shown in (56a), but allows degree modification with -de hěn lìhai ‘to a serious extent’, as in (56b), both hallmarks of a COS verb. The sense of huài describing a character flaw behaves in exactly the opposite way, as seen in (57).

A PKU search for huài N also yielded results consistent with the assumption that its COS, ‘break down’ sense would be less available in prenominal position than the adjectival sense of ‘bad/wicked’. A search for the string huài chē ‘broken down car’ yielded six relevant examples, and huài jīqì ‘broken machine’ yielded no results. In contrast, the search string huài háizi ‘naughty child’ yielded 90 examples.Footnote ²⁸

Some other COS verbs, such as bìng ‘sick’, also seem able to directly modify nouns, e.g. bìng-rén ‘patient, lit. sick person’, bìng-fáng ‘sick room’. But these sequences are best considered V–N compounds as their meanings are non-transparent. A bìng-rén ‘patient’ is not necessarily ill, while bìng de rén ‘sick person’, where de here can be treated as marking a relative clause, does describe someone who is ill. And a bìng-fáng ‘sick room’ clearly does not describe a room that is ill, but a room for sick people, and *bìng de fáng ‘sick assoc room’ is correspondingly unacceptable.

On the basis of these generalizations, I conclude that intransitive COS verbs also do not constitute a regular basis for deverbal adjective formation, although in Section 4.4.2 below I will discuss why they seem to allow limited adjective formation.

4.3.1 Verbs of cooking

The examples in (68)–(69) provide names of dishes which incorporate the verbs describing how the dishes are cooked, a common naming strategy in Chinese cuisine. While the verb of cooking itself is sufficient in some cases, such as in (68), in other cases, such as in (69), it includes adverbial modification. As (70) shows, the verbs of cooking involved are transitive verbs.

1. (68)

   1. (a) chǎo (de) ròu-piàn

      sauté assoc meat-slice

      ‘sautéed pork slices’

   2. (b) zhá (de) dòufu

      deep.fry assoc tofu

      ‘deep fried tofu’

1. (69)

   1. (a) (qīng-)zhēng (de) yú

      clear-steam assoc fish

      ‘steamed fish’

   2. (b) (bái-)zhŭ (de) dàn

      white-cook assoc egg

      ‘boiled egg’

1. (70) chúshī chǎo-le bō-cài/zhēng-le yú

   chef sauté-perf spinach/steam-perf fish

   ‘The chef sautéed spinach/steamed fish.’

Clearly distinct from the COS verbs and RVCs discussed above is the ability of these verbal elements to modify nouns even in the absence of the associative particle de. Many of the names of dishes formed in this fashion are highly conventionalized, and some may be compounds, but some are productive, with a compositional meaning, and show characteristics of phrasal syntax. For instance, as is shown in (71) they license ellipsis of the head noun in a parallel structure (Paul Reference Paul2005, as discussed in Section 2.3 above).

1. (71)

   1. (a) wǒ bù chī zhá dòufu, (qīng-)zhēng de hái kěyi

      1sg neg eat deep.fry tofu clear-steam assoc still acceptable

      ‘I don't eat deep fried tofu, steamed (tofu) is OK.’

   2. (b) jīntiān méi yǒu bái-zhŭ  dàn, chǎo de nì yào ma?

      today neg have white-cook egg sauté assoc 2sg want qprt

      ‘There are no boiled eggs today. Do you want fried ones?’

Some of these modifiers allow an eventive interpretation, as indicated by negation with méi in (72a), and can occur with an agent phrase in the presence of de, as indicated in (72b).Footnote ³² Others, however, do not allow an eventive or agentive interpretation even with de. This is the case for bái-zhŭ ‘white-cook’ in (73).Footnote ³³

1. (72)

   1. (a) yì pán méi/#bu chǎo *(de) bōcài

      one plate neg sauté assoc spinach

      ‘a plate of spinach that has not been sautéed’

   2. (b) yì pán yóu mŭqīn chǎo *(de) bōcài

      one plate by mother sauté assoc spinach

      ‘a plate of spinach sautéed by mother’

1. (73)

   1. (a) *yí gè méi bái-zhŭ de dàn

      one cl neg white-cook assoc egg

      Intended: ‘an egg that has not been boiled’

   2. (b) *yí gè yóu mŭqīn bái-zhŭ de dàn

      one cl by mother white-cook assoc egg

      Intended: ‘an egg boiled by mother’

How should we interpret the contrast between (72) and (73)? Recall from Sections 4.1.2 and 4.1.3 above that RVCs such as dǎ-pò ‘hit-break’ in (51) and COS verbs such as mìe ‘extinguish’ in (54) may only modify a nominal with de present, suggesting that in modifier position, they occur in a relative clause. This assumption is compatible with the eventive (and agentive, in the case of the RVC) properties they retain. Their behaviour is also consistent with the assumption that a deverbal adjective cannot be modified by eventive and agentive modifiers, reasonable enough given that adjectives are stative. These assumptions now suggest a reason for the contrast between (72) and (73). First of all, note that in (71b), the behaviour of bái-zhŭ ‘white-cook’ indicates that it modifies dàn ‘egg’ at a phrasal level, suggesting bái-zhŭ dàn ‘white-cooked egg, i.e. boiled egg’ is not a lexical compound. We thus cannot account for the ungrammaticality of (73) by, say, assuming bái-zhŭ ‘white-cooked’ to be part of a compound.

We can, however, account for the contrast between (72) and (73) by assuming that in (72) chǎo ‘sauté’ occurs in a relative clause, just like dǎ-pò ‘hit-break’ or miè ‘extinguish’ in (51) and (54), while in (73) bái-zhŭ ‘white-cooked’ is an adjective. Thus, the former but not the latter allows eventive and agentive modifiers. This step allows us to capture a three-way distinction. First, RVCs and COS verbs such as dǎ-pò ‘hit-break’ in (51) and mìe ‘extinguish’ in (54) have eventive/agentive interpretations, but never modify a noun without de. This suggests that, as prenominal modifiers, they correspond only to relative clauses. Second, verbs of cooking such as chǎo ‘sauté’ in (72) show eventive/agentive properties in the presence of de, but they may also modify a noun without de. This suggests that as prenominal modifiers, they may constitute relative clauses (allowing eventive/agentive modifiers, but only with de), or they may be adjectives (allowing de to be absent). Third, prenominal modifiers such as bái-zhŭ ‘white-cooked’ in (73b) do not show eventive/agentive properties, but may modify a noun without de. This suggests that such modifiers are not relative clauses, but are only adjectival. Overall, these distinctions point to the prenominal modifiers in (68)–(69) as having an adjectival representation available.

4.3.2 Agentive and instrumental modifiers

Another set of potential deverbal adjectives consists of an eventive verb, with a preceding noun describing an instrument or agent of the event. Some examples are listed in (74)–(75). Like the modifiers based on verbs of cooking, these phrases may directly modify nouns without de, although de is always possible.

1. (74)

   1. (a) zì-zhì (de) gāodiǎn

      self-make assoc pastry

      ‘homemade pastries’

   2. (b) shǒu-huì (de) tú-àn

      hand-draw assoc pattern

      ‘handpainted patterns’

   3. (c) kǒu-shù (de) lìshǐ

      oral-recount assoc history

      ‘history in the oral tradition’

1. (75)

   1. (a) mín-bàn (de) tuántǐ

      civilian-run assoc group

      ‘civilian-run group’

   2. (b) xiàn-zhuāng (de) shū

      thread-assemble assoc book

      stitch-bound book

As with verbs of cooking, these modifiers fall into two groups. The modifiers in (74) are similar to chǎo ‘sauté(ed)’, but distinct from both dǎ-pò ‘hit-break’ and bái-zhŭ ‘white-cooked’, in that they allow the presence of an agent-denoting phrase, as in (76).

1. (76)

   1. (a) yì xiē yóu běn diàn zì-zhì *(de) gāodiǎn

      one some by this store self-make assoc pastry

      ‘some pastries homemade by our store’

   2. (b) yì xiē yóu yìrén shǒu-huì *(de) tú-àn

      one some by craftsperson hand-draw assoc pattern

      ‘some patterns handpainted by artisans’

   3. (c) yí duàn yóu zhǎngzhě kǒu-shù *(de) lìshǐ

      one section by elder oral-recount  assoc history

      ‘a stretch of history orally recounted by the elders’

Going by the discussion of (73a, b) above, these modifiers would be of the chǎo ‘sauté’ category: They may occur in a relative clause, as in (76), retaining the agentivity of the verb, or they may constitute adjectives, as in (74), allowed to modify the nominal with de absent.

However, as (77) shows, the modifiers in (75) do not allow agent phrases.

1. (77)

   1. (a) *yí gè yóu jū-mín mín-bàn (de) tuántǐ

      one cl by reside-people civilian-establish assoc group

      Intended: ‘a civilian group established by residents’

   2. (b) *yì běn yóu dà chūbǎnshè xiàn-zhuāng (de) shū

      one cl by big publisher thread-assemble assoc book

      Intended: ‘a book stitch-bound by a big publisher’

This behaviour corresponds to that of bái-zhŭ ‘white-cooked’, which also does not allow an agent phrase, seen in (73b) above. These modifiers – bái-zhŭ ‘white-cooked’, mín-bàn ‘civilian-established’, xiàn-zhuāng ‘thread-assembled’ – however, do modify nominals directly while also allowing an optional de.

These patterns suggest that the agent/instrument–verb sequences here can have adjective status. The examples in (76) suggest that those sequences in (74) may be ambiguous with a relative clause structure, while (77) suggests that those in (75) are only adjectival. The table in (78) sets out the different possibilities for interpreting the pre-nominal modifiers discussed above.

1. (78) Deverbal adjectives and relative clause pre-nominal modifiers in Mandarin

   

Most noteworthy in (78) is the irregular relationship it reflects between verbs encoding some COS and their related adjectives. First, the relationship between verb classes and deverbal adjectives is non-uniform. In particular, as noted in Section 4.1.3 above, some intransitive COS verbs may form deverbal adjectives, but only in a limited fashion. Verbs of cooking and verbs of creation have deverbal adjectival counterparts, among which some, e.g. bái-zhŭ ‘white-cooked’ seem to be only adjectival. I have labelled kǒu-shù ‘orally-recounted’ as a verb of information transmission. In modifier position, kǒu-shù ‘orally-recounted’ may be an adjective or a relative clause. Unlike for verbs of cooking and verbs of creation, I have not been able to find a verb of information transmission that is only adjectival in modifier position. Second, note also the three-way distinction discussed above between verbs that have no deverbal adjective counterparts (RVCs, some intransitive COS verbs), those that do (most others in the table), and those that, in prenominal position, only correspond to deverbal adjectives (those in the last two rows of the table in (78) above). The varied situation presented by eventive verbs and their deverbal adjectival counterparts contrasts with the relatively regular COS verb alternations shown by adjectives (see (37) above). This suggests that a rule-based derivation may not be the right way to approach deverbal adjective formation in Mandarin. If so, how do these deverbal adjectives come about? I turn to this question directly below.

4.4.1 Deverbal adjectives are not productive

Unlike deadjectival COS verbs, which are quite regular, the deverbal adjectives discussed above are relatively conventionalized and domain-specific. For instance, although any food item described with a particular verb of cooking has to have undergone the process named, simply undergoing that process does not necessarily qualify the food to be named with the relevant modifier. Many dishes require intermediate steps of processing some ingredient in some way before that ingredient is combined with the rest of the ingredients to form the dish. Thus, for a certain dish, pork slices may be first sautéed briefly, then taken off the heat while something else is being prepared. This intermediate step does not necessarily qualify the pre-sautéed pork slices as chǎo ròu-piàn ‘sautéed pork slices’.

The flexibility of verbs of cooking is also domain-specific: There seems to be no correspondingly varied choice of modifiers for nouns naming drinks. For instance, there is no obvious way to express James Bond's famous ‘shaken not stirred’ distinction via prenominal modification without de.

Similarly, adjectives such as mín-bàn ‘civilian-established’ are restricted to the contrast between government and civilian as agents of an event. Analogous adjectives include guān-bàn ‘official-established’ and mín-yíng ‘civilian-administered’. There are, however, no corresponding adjectives for other social groups in agentive capacity, for instance, there is no adjectival *(xué)shēng-bàn ‘student-established’ or *(jiā)zhǎng-bàn ‘parent-established’. That is, in nominal modifier position, these phrases must be followed by de.

These observations demonstrate the non-productive nature of the deverbal adjectives discussed, suggesting that they are not systematically available.

4.4.2 The conceptual-pragmatic basis of deverbal adjectives

If the deverbal adjectives discussed above are not derived by rule or systematic polysemies, how do they come about? I suggest that these deverbal adjectives have a conceptual-pragmatic basis, specifically, that the modified NP containing these adjectives describes a culturally- or contextually-salient class of entities. Several points support this interpretation.

First, as seen in (79), these deverbal adjectives are attributive and do not occur in predicative position, except in the shì … de construction, as in (80).Footnote ³⁴

1. (79)

   1. (a) *zhè pán bōcài chǎo

      this plate spinach sauté

      Intended: ‘This plate of spinach is sautéed.’

   2. (b) *zhè duàn lǐshǐ kǒu shù

      this cl history oral recount

      Intended: ‘This history is in the oral tradition.’

1. (80)

   1. (a) zhè pán bōcài shì chǎo de

      this pl spinach be sauté DE

      ‘This spinach is sautéed.’

   2. (b) zhè duàn lǐshǐ shì kǒu shù de

      this cl history be oral recount DE

      ‘This history is in the oral tradition.’

As we saw in Section 4.2 above, these properties are manifested by other adjectives, such as nán ‘male’ and zhēn ‘real’, which may be understood to have a characterizing function. It seems plausible that words that are used to characterize at least certain kinds of objects, when combined with the relevant nominal, would describe a culturally-salient class of entities.

Second, the ability to occur prenominally in the absence of de – the reason these words were identified as adjectives – is itself indicative of a characterizing function. As noted in Section 2.1 above, even basic property concept adjectives such as bái ‘white’ and gāo ‘tall/high’ do not modify all nominals with equal felicity in the absence of de (Zhu 1956/Reference Zhu1980). The examples in (15) above, repeated here, illustrate one such case.

1. (15)

   1. (a) bái (de) zhǐ

      white assoc paper

      ‘white paper’  (Zhu Reference Zhu1980: 7)

   2. (b) bái #(de) shǒu

      white assoc hand

      ‘white hand’ (Zhu Reference Zhu1980: 10)

Paul (Reference Paul, Anderl and Eifring2006: 306) suggests that a pre-nominal modifier in the absence of the associative particle de ‘is possible provided the resulting NP provides a natural, plausible classification, with the modifier presented as a defining property’ (see also Li & Thompson Reference Li and Thompson1981: 119). Paul draws an analogy between nominal modification without de in Mandarin with prenominal adjectival modification in English, which shows ‘similar semantico-pragmatically motivated gaps’, discussed at length in Bolinger (Reference Bolinger1967). Bolinger (Reference Bolinger1967) argues that in English, adjectives in attributive position tend to have a ‘characterizing’ function. For instance, the same adjective in attributive and predicative position may have different effects on meaning, as in the contrast between the English expressions the visible stars (stars that are inherently visible, being of high magnitude) and the stars visible (on a cloudy night). The characterizing function similarly distinguishes between the availability of present participle compounds: a home-loving girl as opposed to the lack of a mistake-erasing secretary or a husband-waking wife, which ‘must await the day when we have some interest in characterizing secretaries as mistake-erasing or wives as husband-waking’ (Bolinger Reference Bolinger1967: 7). The availability of prenominal adjectives without de in Mandarin appears to be subject to a similar condition.

The deverbal adjectives discussed in Section 4.3 above, based on verbs of cooking and verbs with agent and instrument modifiers, are easily compatible with the cultural salience rubric, given the importance of cuisine in Chinese culture. Many of the deverbal adjectives discussed above also describe methods of production or creation, one way artifacts may be classified (Hilpinen Reference Hilpinen and Zalta2011).

Cultural salience is not the only factor licensing deverbal adjectives. Contextual factors may allow a class of objects to at least temporarily take on salience. This is presumably what is at work in the felicity conditions of huài ‘break down’ in the shì … de environment in (65) discussed in Section 4.2 above. COS verbs often describe changes resulting in decay, disrepair, and malfunction, perhaps not usually the kinds of categories objects are classified by. This may be why it is more difficult to find them in descriptions with a characterizing function such as the shì … de construction and in prenominal modifier position without de. But they are found in contexts where these categories are salient, e.g. for an auto mechanic, broken cars are of occupational significance. Indeed, huài ‘break down’ would be felicitous in prenominal position without de, e.g. huài chē ‘broken down car’ in the same context as it would be felicitous with shì … de, namely when the category of objects being described is salient in some way, e.g. broken down cars being pointed out to or by an auto mechanic.Footnote ³⁵

A further example in this vein is provided by the intransitive COS verb duàn ‘to break/snap apart’ in (81a).

1. (81)

   1. (a) diàn-xiàn duàn-le

      electric-cable snap-perf

      ‘The electric cables snapped.’

   2. (b) tā gěi yì tuán duàn (de) diàn-xiàn chán-zhù-le

      3sg give one mass snapped assoc electric-cable tangle-stop-perf

      ‘She was tangled in a mess of severed electric cables.’ (PKU)

   3. (c) duàn (de) xiàn shì yīn de fúhào

      broken assoc line be yin assoc symbol

      ‘A broken line is the symbol for yīn.’ (PKU)

   4. (d) huàn-zhě kěnéng shì xuè-guǎn duàn-le

      sickened-one possible be blood-vessel snap-perf

      ‘It could be that the patient's blood vessel had snapped.’ (PKU)

   5. (e) *yì gēn duàn xuè-guǎn

      one cl snap blood-vessel

      Intended: ‘a snapped blood vessel’

This word may occur prenominally with optional de, suggesting that it has an adjectival representation, as in (81b, c), but these uses are limited when compared with its verbal use.Footnote ³⁶ For instance, (81b) is in the context of a wartime novel, describing what happens during a family's flight from a city in which an air raid had recently occurred. In such a context, snapped cables are likely to be a salient category. Another putative adjectival use is in a specialized context such as (81c), which describes the use of a broken line to represent the concept of yīn in Taoist cosmology.Footnote ³⁷ As (81d, e) suggest, however, duàn ‘snap/break apart’ is less restricted as a COS verb than as a result state adjective. I could find no instances of duàn xuèguǎn ‘snapped blood vessel’, which would indicate adjectival duàn ‘snapped’, either in the PKU corpus or through a Google search. In contrast, in the PKU corpus, I found five instances of xuèguǎn duàn-le ‘blood vessel snapped’, where duàn ‘snap’ would be verbal, and numerous further such instances through a Google search.

The conventionalized and context-dependent qualities of deverbal adjectives form a clear contrast with the regularity of deadjectival COS verbs, suggesting that some of them, e.g. those based on verbs of cooking, are simply listed in the lexicon, while some are probably formed ‘on the fly’, e.g. duàn ‘to break/snap apart’, hence the apparently random discrimination between, say, blood vessels and electric wiring.

4.5.1 Why no RVC-derived adjectives?

I have argued above that the deverbal adjectives discussed in this paper are not systematically available, and those that are available are either separately listed in the lexicon, or derived ‘on the fly’. This provides a plausible explanation for the rarity of deverbal adjectives based on COS verbs, and the domain-specificity of those based on certain transitive verbs, which are somewhat more numerous. One puzzle remains, however. This is the seemingly categorical inability of RVCs to show adjectival behaviour, i.e. to occur in attributive position of a noun phrase with de. I have no account of this asymmetry, but in this subsection, I speculate on a possible reason for it. I believe the crux of the matter lies in the lack of overt morphology in Mandarin specialized to signaling the relevant grammatical operations that verbs (and adjectives) participate in.

Let us first consider the simpler case of intransitive COS verbs. Drawing on Kratzer (Reference Kratzer2000), I assume a stativizing function (82) – possibly associated with a zero morpheme – that may apply to intransitive COS verbs, as shown in (83).Footnote ³⁸ I assume that the operation, being non-overt, is costly, and that it is either conventionalized for particular verbs or applied when the context allows the kind of interpretation discussed above.

1. (82) Stativizer: λR λx λs ∃e R(x)(s)(e)

2. (83)

   1. (a) pò ‘to break (intransitive)’: λx λs λe [become(s)(e) ∧ not-whole′(x)(s)]

   2. (b) Stativized pò ‘broken’:

      λR λx λs ∃e R(x)(s)(e)(λx λs λe [become(s)(e) ∧ not-whole′(x)(s)])

      =λx λs ∃e [become(s)(e) ∧ not-whole′(x)(s)]

One way RVCs may be prevented from giving rise to a derived stative is simply to assume that they are of the wrong type. The stativizing function selects for a function of type <e,<s,<v,t>>>, but an RVC such as cā-gān ‘wipe dry’, when detransitivized, as in (84), is of the type <e,<s<v,<v,t>>>>.

1. (84) Passive/detransitivized cā-gān ‘be wiped-dry’:

   λy λs λe′ λe ∃x [wipe′(x, y)(e′) ∧ dry′(y)(s) ∧ become(s)(e) ∧ cause(e′, e)]

The distinction may seem trivial, but a more intuitive way of understanding it is that a phonologically non-overt grammatical operation that involves existential binding over events (as opposed to existential closure at the clause level) may not affect more than one event argument.

What of deriving statives from verbs of cooking and the other verbs discussed above? There are different ways to approach these, but one important point is that these verbs do not specify change of state in the same way as RVCs do. Although verbs of cooking such as chǎo ‘sauté’ allow an inference that the theme of the event is cooked to some extent, I believe this inference is not specified in the event structure of the verb.Footnote ³⁹ This distinction shows up in negation contexts with méi, such as in (85).

1. (85)

   1. (a) wǎn méi cā-gān

      bowl neg wipe-dry

      ‘The bowl wasn't wiped dry.’

      (i.e. ‘The bowl was wiped, but didn't get dry.’ Or: ‘The bowl wasn't wiped.’)

   2. (b) cài méi chǎo

      vegetable neg  sauté

      ‘The vegetables were not sautéed.’

      (i.e. ‘No sautéing of the vegetables took place.’ Not: ‘The vegetables were sautéed but not cooked.’)

RVCs under méi have two interpretations: (85a) may mean that there was wiping but the bowl did not become dry (the favoured interpretation), or that no wiping occurred. The example in (85b), with chǎo ‘sauté, can only mean that no sautéing event took place, not that there was sautéing but the vegetables did not get cooked.

A proper representation that captures these generalizations must await further work, but for current purposes, we may distinguish chǎo ‘sauté’ (and other verbs of cooking) from ‘true’ COS verbs by representing them as activity verbs, as in (86a).

1. (86)

   1. (a) chǎo ‘sauté’: λy λx λe sauté′(x, y)(e)

   2. (b) stative chǎo ‘sautéed’: λy λt ∃e ∃x [sauté′(x, y)(e) ∧ τ(e) <t]

To accommodate the associated COS inference, we could further assume a meaning postulate stating that, for any event of sautéing, there is a corresponding change of the theme to some degree of being cooked. A stativized meaning for such a verb, then, could perhaps be better captured via some version of Kratzer's (Reference Kratzer2000) ‘resultant state’ meaning, as a function from an individual to the set of times that are preceded by (the runtime τ(e) of) an event e in which this individual undergoes sautéing, as in (86b).Footnote ⁴⁰ Again, I would assume such an operation to be costly, given its non-overt nature, and to be conventionalized for particular verbs, or to apply only in particular contexts.

4.5.2 Stativizing morphology?

On a related but more general note: Why is there an asymmetry between deadjectival verbs and deverbal adjectives? I speculate again that the lack of deverbal result state adjectives is due to the lack of overt verbal morphology such as a participial morpheme that is specialized for forming result state predicates from verbs encoding change, since this could mean that there is no consistent source of stative predicates for adjective formation. Such a hypothesis would suggest that deadjectival verb formation, which is also not overtly indicated, is somehow less ‘costly’ than a stativizing operation, perhaps because it does not involve existential binding over an event.

The durative particle -zhe presents a further potential problem for the notion that deverbal adjective formation in Mandarin is somehow limited by the lack of overt morphology. Jaxontov (Reference Jaxontov and Nedjalkov1988) proposes, for instance, that -zhe shows stativizing properties. This observation is based on the ability of -zhe to combine with verbs of putting such as fàng ‘put’, bǎi ‘place’ and guà ‘hang’, illustrated in (87). Since V-zhe may describe situations such as that in (87b) in which no actual causing action has taken place to bring about a change of location of the theme, this suggests V-zhe is some kind of a derived stative verb.

1. (87)

   1. (a) qiáng-shàng guà-zhe yì fú huà

      wall-upon hang-dur one cl painting

      ‘On the wall hung a painting.’

   2. (b) tiān-shang guà-zhe jǐ kē xīng

      sky-upon hang-dur  few cl stars

      ‘In the sky hung a few stars.’

While this proposal may be correct for verbs of putting, the morpheme -zhe cannot be considered a particle for deriving result states in general. It may attach to a variety of verbs, including activity verbs, as in (88a), and stative verbs, as in (88b–d), where the interpretation it yields is clearly that the event or state described is ongoing at the reference time of the clause. With the latter it is often associated with locative inversion, as shown in (88c) (Pan Reference Pan1996), though this is not necessarily the case, see (88d). For further discussion of -zhe, see Yeh (Reference Yeh1993), Wu & Kuo (Reference Wu and Kuo2003), Wu (Reference Wu2007).

1. (88)

   1. (a) tā kàn-zhe wǒ …

      3sg  look-dur 1sg

      ‘(S)he looked at me …’

   2. (b) Qīngzàng gāoyuán-shang yǒu-zhe bǎi- xuě-ǎiǎi de xuě fēng

      Qinghai-Tibet plateau have-dur white-snow-snowy assoc snow peak

      ‘On the Qinghai-Tibetan plateau there are white snow-capped mountains.’ (PKU)

   3. (c) zhè-lǐ zhù-zhe 2.8 wàn Zhèjiāng rén

      here live-dur 28,000 Zhejiang people

      ‘Here live 28,000 people from Zhejiang province.’ (PKU)

   4. (d) tā  … zài gōng-lǐ zhù-zhe

      3sg  be.at palace-within live-dur

      ‘She lived in the palace.’ (PKU)

That -zhe can combine with originally stative predicates suggests that it should not simply be considered stativizing morphology. That it can combine with activity predicates that do not encode a change of state indicates also that -zhe has functions other than creating a result state predicate. In addition, while -zhe may combine with verbs of change of location such as guà ‘hang’ in (87) above, it may not combine with other COS verbs, regardless of their valence, as (89a) shows.

1. (89)

   1. (a) *bēizi (dǎ-)pò-zhe

      cup hit-break-dur

      Intended: ‘The cup is broken.’

   2. (b) guà-zhe *(de) yīfu

      hang-dur assoc clothes

      ‘hanging clothes’

This shows that whatever its stativizing properties, it does not apply to all COS predicates. Finally, V-zhe never shows adjective-like behaviour (89b). It cannot modify nominals without de, further suggesting that even if -zhe does derive result state predicates from verbs of caused change of location, they remain verbs and are not converted into adjectives. Therefore, at the very least, we can conclude that V-zhe is not a source for a regular derivation of deverbal adjectives. It might eventually turn out upon investigation that the -zhe with verbs of putting is distinct from the durative -zhe in (88), but this is a question for another paper, and I do not discuss it further here.

Summing up this section briefly, adjectives cannot be derived systematically from COS verbs. Deverbal adjectives based on COS verbs do exist, but they are better considered as listed separately or formed ‘on the fly’ rather than being systematically available. These deverbal adjectives are allowed if the modified NP they occur in describes a culturally- or contextually-salient category of objects.