The thematic hierarchy and causativity

YAFEI LI

T H E T H E M A T I C H I E R A R C H Y A N D C A U S A T I V I T Y *

Chinese resultative compounds provide evidence that the thematic hierarchy is not the only factor determining the link between theta roles and syntactic arguments. Rather, it interacts with the causative hierarchy and when in conflict, can be overridden by the latter. This "asymmetrical" interaction between the two factors is shown to be responsible for the complex and, in some cases, totally unexpected patterns of ambiguity that resnltative compounds demonstrate in various syntactic contexts. To the extent that the presence/absence of the causative hierarchy can be accurately predicted and its interaction with the thematic hierarchy explicitly specified, this paper lends support to the multi-dimensional nature of human language.

O. I N T R O D U C T I O N

In the government-binding framework, it is often taken for granted that the linking between theta roles of a lexical item and syntactic argument positions is sufficiently constrained by the thematic hierarchy in which Agent, for instance, is more prominent than Theme. In this paper, I consider a violation of the thematic hierarchy, characteristically exhibited by resultative compounds in Chinese. I show that the complicated semantic and syntactic behaviors of these compounds are best explained if the thematic hierarchy can be violated, and that the condition for such violations can be precisely formulated. The basic data is given in section 1, followed by arguments against possible analyses in which syntactic movement is used to avoid a thematic hierarchy violation. My own analysis is presented and defended in section 2. In the !ast section, my analysis is briefly compared with Grimshaw's (1990) theory of aspectuality.

1. B A S I C D A T A AND M O V E M E N T - B A S E D A N A L Y S E S

Chinese has a productive resultative V-V compound in which the first verbal morpheme refers to the causing event (hereafter referred to as

* The data as shown in (4) was brought to my attention by Bao Zhiming. I thank Jane Grimshaw for discussing the basic analysis with me at its beginning stage. Grimshaw (1990) has important influence on the ways in which my analysis is presented. This revised version benefits greatly from the detailed comments of three N L L T reviewers and Carol Georgo- poulos. I am also grateful to Emily Huang, Teng Qiuyun, Wang Yuyan, and Xie Qing for providing judgments on the Chinese data.

Natural Language and Linguistic Theory 13: 255-282, 1995. © 1995 Kluwer Academic Publishers. Printed in the Netherlands.

256 YAFEI LI

Voaus) and the second refers to the resulting event (hence Vre~). An example is given in (1): ~

(1) Taotao zhui-lei-le Youyou le

Taotao chase-tired-asp Youyou LE

a. Taotao chased Youyou and as a result Youyou got tired. b. Taotao chased Youyou and as a result Taotao got tired.

Li (1990a) argues that the ambiguity of (1) results from optionally identify- ing the single theta role of Vr~ lei 'tired' with either one of the two theta roles of Voa~s zhui 'chase' and assigning the identified pair of theta roles to a single argument in the sentence. The two alterantives are represented in (2): z

(2)a. V(l(2-a)) (yielding (la)) b. V(1-a(2))(yielding (lb))

/ N / N Vcaus Vres Veaus Vres

I I I I chase tired chase tired

11 (2)) (a) (1 (2)) (a) In (2a), the single theta role a of Vr~ is identified with the theta role 2 of Voaus and the identified pair 2-a is assigned to the object of the sentence, Youyou, as indicated by the two pairs of angled brackets surrounding them

1 In this paper, asp stands for an aspect marker in Mandarin, and LE for the sentence-final particle that often occurs with declarative sentences about past events. The examples remain well formed without LE, though less natural as stand-alone sentences. For some native speakers, the reading in ( lb) is harder to obtain than the one in (ia). However, the type of reading represented by ( lb) can be made more salient with examples like (i):

(i) Taotao ting-fan-le neishou get le

Taotao listen-to-bored-asp that song LE

Taotao listened to that song (for so many times that he) got bored.

In this article, I will stick to examples like (1) both because many native speakers do get the ambiguity in (1a-b) and because it provides a minimal pair that helps simplify subsequent discussions. An early but extensive lexical analysis of the resultative compounds can be found in Thompson (1973). 2 I follow Li (1990a) in using " ( . . . )" to indicate an argument structure. The theta roles of Vcaus are represented as whole numbers 1, 2 , . . . , and the theta roles of Vre~ are represented as English letters a, b . . . . . The natural sequence of these symbols carry no significance. " - " is used to indicate theta-identification, a thematic operation first proposed in Higginbotham (1985). Adapting Grimshaw's (1990) notations to mine, I will surround the thematically most prominent theta role (or pair of theta roles) with one pair of angled brackets, the next prominent theta role with two pairs of angled brackets, etc.

T H E T H E M A T I C H I E R A R C H Y A N D C A U S A T I V I T Y 257

in the argument structure of the compound. Thus, Youyou is interpreted as both the one who was chased and the one who got tired as the result of chasing. On the other hand, a of Vre~ is identified with the theta role 1 of V . . . . in (2b). With 1-a assigned to the subject of the sentence, Taotao, the latter is both the one who did the chasing and the one who ended up becoming tired. This analysis is adopted in the rest of the current paper.

It is also shown in Li (1990a) that (1) cannot have the reading in (3a). (3b) represents the argument structure that would be required by such a reading:



a. *Youyou chased Taotao and as a result Taotao got tired. b. (1 (2-a))

Taotao Youyou

Note that Youyou cannot be interpreted as the chaser (i.e., the receiver of 1 from V~aus) and Taotao cannot be the chasee (by receiving 2 from Voaus). Under the widely accepted assumption that the thematic hierarchy of the theta roles must 'align' with the structural prominence of the syntactic argument positions (Jackendoff 1972, Higginbotham 1985, Grim- shaw 1990, Li 1990a, among many others), the impossibility of (3a-b) is expected. With the most prominent theta role of V .. . . (--i) assigned to the object of the sentence and the less prominent theta role of VCau~ (= 2) to the subject, the thematic hierarchy is not respected, making the reading unavailable. See Li (1990a) for details.

Given this analysis of (3), it comes as a surprise that (1) also allows a third reading:



a. Youyou chased Taotao and as a result Youyou got tired. b. (1-a (2))

Taotao Youyou

In (4), as in (3), the two theta roles of V .. . . zhui 'chase' are 'inverted' in their assignment, with the thematically less prominent 2 assigned to the subject position and the most prominent i to the object position. Examples

258 YAFEI LI

like (4) raise two closely related questions: What accounts for this violation of the thematic hierarchy? And why is the same violation not possible in (3)? Presented with such anomalous data, one is naturally tempted to appeal to syntactic movement, thus avoiding the claim that the example results from theta role assignment in violation of the well-supported thematic hierarchy. In the rest of this section, I show that a movement approach is not available to (4).

1.1. Deriving (4) with Movement in a Mono-Clausal Structure

Suppose that (4) has a mono-clausal structure. Then it is theoretically possible that the sentence-initial NP Taotao is the D-structure object raised to a higher position, accompanied by raising the compound verb over the D-structure subject Youyou. (5) illustrates the derivation:

(5) [CP Taotaoi chase-tiredj [XP Youyou tjti]]

If XP -- VP, there may be several positions for Taotao to land; if XP = IP, Taotao can only occupy the SpecC position. But independently of where Taotao and the compound verb land outside VP, (5) allows assigning theta roles in compliance with the thematic hierarchy. Such an analysis is untenable, however, because the same derivation will also incorrectly allow the ungrammatical (3), which differs from (4) only in assigning the single theta role of lei 'tired' to Taotao.

That (4) cannot be analysed with a structure like (5) is further supported by the fact that the postverbal NP in (4) can be introduced in a preverbal position by ba:

(6) Taotao ba Youyou zhui-lei-le

Taotao ba Youyou chase-tired-asp

Youyou chased Taotao and as a result Youyou got tired.

The precise nature of the ba construction is still unclear, but it is not controversial that when the verb is transitive, ba can only (conditionally) introduce the object and never the subject (cf. Li and Thompson 1981, Goodall 1986, Cheng and Ritter 1987, A. Li 1990). If Youyou in (4) and (6) were the D-structure subject, one would have to explain why ba can introduce the subject only in this construction. On the other hand, if Youyou is the object in these sentences, as I will argue throughout this paper, its introduction by ba will follow from the general properties of ba without ad hoc stipulations.


1.2. Deriving (4) with Movement in a Bi-Ctausal Structure

It is well known that the sentence-initial NP Taotao in (4) (as well as in (6)) is also understood as the cause of Youyou's becoming tired even though it is Youyou who chased Taotao (cf. Wang 1958, Huang 1988). A possible scenario in which (4a) is appropriate for the sentence is that Taotao was so good at running long distance that Youyou became tired from chasing him. Drawing on this fact, (4) may be given another plausible analysis, represented in (7): 3

(7) VP

NP V'

Taotao V XP

CAUSE NP

t L Youyou V

zhui-tei 'chase-tired'

At S-structure, the compound V in the embedded XP moves to the matrix V, a phonetically empty causative verb, following Baker's (1988) analysis of verb incorporation. This analysis not only reflects the semantics, that Taotao made Youyou chase him and eventually become tired, but also correctly yields the linear order [Taotao chase-tired Youyoul without anomalous theta-role assignment.

Attractive as it may be, such an analysis faces empirical problems. First, since XP in (7) contains a ~¢np predicate (cf. note 3), it is expected to allow adverbials in it. This is confirmed by (8), which differs from (7) only in having a lexical causative verb shi 'make':

(8)a. (?)Taotao shi Youyou henkuai zhui-lei-le

Taotao make Youyou quickly chase-tired-asp

Taotao made Youyou quickly become tired from chasing (him).

3 It is irrelevant to the present analysis whether the embedded XP is CP, IP, or VP. See Li (1990b) for arguments that it should be VP.

260 YAFEI LI

(8)b. Zheishou ger shi Youyou henkuai chang-fan-le

this song make Youyou quickly sing-bored-asp

This song made Youyou quickly become bored of singing (it).

The position of the adverb is normally before the verbal compound chang-fan 'sing-bored', as is typical of Mandarin Chinese. If the embedded verbs move to the matrix V when the latter is phonetically empty, the adverbs should be stranded and occur at the end of the sentence:

(9)a. *Taotao zhui-lei-lei Youyou henkuai t i le

Taotao chase-lei-asp Youyou quickly LE

Intended reading: =(8a)

b. *Zheishou ger chang-fan-lei Youyou henkuai t i le

this song song sing-bored-asp Youyou quickly LE

Intended reading: = (8b)

The ungrammaticality of these clause-final adverbs is unexpected if the sentences in (9) (and (4)) are derived from (7) through verb incorporation. In contrast, the sentences are acceptable without the adverbs - (9a) forms a minimal pair ~ith (4) while (9b) contrasts with (10):

(10) Zheishou ger chang-fen-lei Youyou le

this song sing-bored-asp Youyou LE

Youyou sang this song and as a result became bored.

Secondly, there are clear cases in which the lexical causative verb can be used but the stipulated empty causative verb cannot:

( l l )a. Youyou chang-fan-le

Youyou sing-bored-asp

Youyou sang and as a result became bored.

b. Wuxiuzhide pailian shi Youyou chang-fan-le

endless rehearsal make Youyou sing-bored-asp

Endless rehearsals made Youyou become bored from singing.

C. *Wuxiuzhide pailian chang-fan-le Youyou le

endless rehearsal sing-bored-asp Youyou LE

Intended reading: Same as (l ib).


(12)a. Taotao bing-dao-le

Taotao sick-fall-asp

Taotao was down with sickness.

b. Jiankude gongzuo shi Taotao bing-dao-le

tough work make Taotao sick-fall-asp

Tough work (conditions) brought Taotao down with sickness.

c. *Jiankude gongzuo bing-dao-le Taotao le

tough work sick-fall-asp Taotao LE

Intended reading: Same as (12b).

(13)a. Youyou pao-lei-le

Youyou run-tired-asp

Youyou ran and as a result became tired.

b. ?Jiaolian shi Youyou pao-lei-le

coach make Youyou run-tired-asp

The coach made Youyou run and get tired.

c. *Jiaolian pao-lei-le Youyou le

coach run-tired-asp Youyou LE

Intended reading: Same as (13b).

Some (b) sentences may sound slightly odd (13b) or need a specific context ( l lb) , but they are far better than the (c) sentences. If the latter had a D-structure like (5), which the (b) sentences also share, the contrast would be unexpected.

Note that the ungrammaticality of the (c) sentences cannot be attributed to some peculiar selectional property of the empty causative verb, because of the contrast between the ungrammatical ( l lc) and the grammatical (10): Both sentences use the same compound chang-fan 'sing-bored', and in both sentences the clause-initial NP is inanimate. On the other hand, the contrast can be explained by the theta criterion, which requires that each argument receive a theta role (Chomsky 1981). (14) below indicates that only ger 'song' can receive a theta role from chang 'sing':

(1.2) Youyou chang ger/*pailian

Youyou sing song~rehearsal

Youyou sings songs/*rehearsals.

262 YAFEI LI

As a result, the clause-initial NP wuxiuzhidepailian 'endless rehearsal' in (11c) cannot receive a theta role from the compound chang-fan 'sing- bored'. If there is no empty causative verb in (11c) to assign a theta role to it, the ungrammaticality of the example follows. In contrast, the same compound can assign a theta role to the clause-initial NP zheishou ger 'this song' in (10b), making it grammatical. The same account applies to the contrast between the (b) and (c) sentences in (11) through (13): Each sentence-initial NP in a (b) sentence receives a theta role from the lexical causative verb shi 'make', thus satisfying the theta criterion. But this account entails that theta roles are assigned inversely in (10b) and (4). 4

There are apparent exceptions to the conclusion that the sentence-initial NP must receive a theta role:

(15)a. Taotao de gushi xiao-si wo le

Taotao 's story laugh-die me LE

Taotao's story made me laugh to death.

b. Nachang jihuang e-si-le henduo ren

that famine hungry-die-asp many people

Many people starved to death in that famine.

Since both V .. . . and Vr~ in these compounds look like intransitives and each assigns its single theta role to the object, the subject in each sentence apparently has no theta role from the compound. There are reasons to

4 Given that the sentence-initial NP must be thematically related to the compound, two other analyses may be devised to avoid inverse theta role assignment. One is to assume that (4) and (10) involve empty operator movement comparable with the English tough-movement:

(i) NPli CAUSE[Opi[NP2 V-V ti]

Subsequent verb-raising will yield the right word order. But (i) is ad hoc because no known lexical causative verb/affix requires an empty Op. Furthermore, it fails to explain why inverse theta role assignment only happens with resultative compounds. The other alternative analysis is to make the empty causative verb assign no theta role to the matrix subject position (cf. Pesetsky 1995):

(ii) 0 CAUSE [NP2 V-V NP1]

The surface word order derives from moving the compound to matrix V and NP1 to the empty subject position tl. This analysis has the same two problems as (i). In addition, one wonders why NP1 must raise, and where the embedded subject NP2 gets Case. If NP2 gets Case from the matrix verb CAUSE, then CAUSE behaves inconsistently with Burzio's generalization by assigning accusative Case but no subject theta role; if NP2 gets Case from the embedded INFL, then the complement clause is a full CP and thus moving NP1 to 0 should be blocked as super-raising, in addition to the problem that cross-linguistically verb incorporation never seems to cross a full clause boundary (Li (1990b).

THE T H E M A T I C H I E R A R C H Y AND C A U S A T I V I T Y 263

believe, however, that (16) are not counterexamples. It is true that xiao 'laugh' in (15a) is usually used intransitively, but (16) is also acceptable, in which xiao in a compound is clearly used causatively with the interpretation ' m a k e . . . laugh':

(16) Taotao de gushi zui xiao-ren

Taotao 's story most laugh-people

Taotao's story made people laugh most. (=was the funniest)

Thus, a plausible assumption is that it is this causative xiao that serves as V .. . . in (15a) and assigns a theta role to the subject Taotao's story. 5 Turning to (15b), notice that the subject is actually an oblique argument, indicating the time of the event described by the compound. So it is reasonable to assume that the subject receives a temporal theta role from the compound or has an intrinsic theta role like other temporal adverbials (Carol Georgopoulos, personal communication).

Further evidence against positing an empty causative verb as in (8) is provided by the V-de construction, the syntactic counterpart of the resultative compound in which V~au~ carries a suffix -de and is followed by a clause headed by an independent Vre~ (see Li to appear for an analysis of the V-de construction):

(17) Taotao zhui-de Youyou feichang lei (cf. (1) and (4))

Taotao chase-de Youyou very tired

a. Taotao chased Youyou and as a result Youyou became very tired.

b. Youyou chased Taotao and as a result Youyou became very tired.

s A reviewer wonders if fan 'bored' as used in (10b) is also transitive, given (i):

(i) Neishou get zui fan-ren

that song most bore-people

That song bores people most.

If fan were used transitively in (10b), the example would be exactly the same in theta rote assignment as the ungrammatical (ii) below with transitive Vr~:

(ii) *wushu xue-shang-le Taotao le

martial-art learn-injure-asp Taotao LE

Intended reading: Taotao learned martial art and got injured by it.

The reader is referred to Li (1990a) for an account of why examples with a transitive Vre~ like (ii) are ungrammatical. Since (10b) is good, fan must be intransitive.

264 YAFEI LI

Now compare the following examples:

(18)a. Neige gushi shi [Taotao yilian-santian xiao-de duzi-teng]

that story make Taotao for-three-days laugh-de belly-hurt

For three days, that story made Taotao laugh so much that his belly hurt.

b. Neige gushi shi [Taotao xiao-de yilian-sanfian duzi-teng]

that story make Taotao laugh-de for-three-days belly-hurt

That story made Taotao laugh so much that his belly hurt for three days.

(19) Neige gushi xiao-de Taotao yilian-santian duzi-teng

that story laugh-de Taotao for-three-days belly-hurt

= (18b) :b (lSa)

When the adverbial yilian-santian 'for-three-days' immediately precedes Voau~ (in (18a)), its scope is over the whole event of laughing-made-belly- hurt; but when it immediately precedes V ~ (in (18b)), what lasted three days is only the belly-hurting event. Since xiao in (19) is used causatively, it may seem plausible to derive (19) from (20) through verb-raising:

(20) that story CAUSE [Taotao (for three days) laugh-de (for three days) belly hurt]

But this analysis wrongly predicts that (19) is ambiguous with respect to which event lasted for three days, because the adverbial may immediately precede either Vres or the trace of V ... . . The fact that (19) only has the reading of (18b) follows naturally if xiao 'laugh' enters syntax as a causativized verb.

In addition, (18) and (19) show that a causative reading does not necessarily result from an empty causative verb plus syntactic verb incorporation even when such a derivation is otherwise plausible. Consequently, it is not automatically preferable to take (8) as the structure for (4).

Given all the problems (8) has, I conclude that it provides no viable explanation for the inverse theta role assignment in (4). And in the light of section 1, I will assume that (4) as well (1) and (3) have monoclausal structures in which the preverbal NP is in subject position and the postverbal NP in object position, consistent with the general SVO word order of Chinese.


2. C A U S A T I V E R O L E S AND THE C A U S A T I V E H I E R A R C H Y

In this section, I show that conditioned violations of the thematic hierarchy can explain both the contrast between (4) and (3), and several other related properties of resultative compounds. Specifically, I propose that in addition to the set of theta roles assigned by lexical words and regulated by the thematic hierarchy, there are two causative roles (c-roles), Cause and Affectee, that arguments acquire when they are associated in a particular way with two causally related predicates. The causative hierarchy, in which Cause is more prominent than Affectee, interacts with the thematic hierarchy to determine the linking between theta roles and syntactic argument positions. It is this interaction that accounts for inverse theta role assignment.

2.1. Motivation for C-Ro&s and the Causative Hierarchy

Notice that the four readings of (1) discussed so far exhaust atl the logically possible variations of theta role assignment by the compound zhui-Iei 'chase-tired' if the single theta role of Vres is to be identified with one of the theta roles of V ..... These four possibilities are summarized in (21):

(21) Taotao zhui-lei-le Youyou (= (1))

Taotao chase-tired-asp Youyou

a. T. chased Y. and Y got tired. (ta) b .T . chased Y. and T. got tired. (lb)

(1 (2-a)) (1-c~ (2))

I I I I Taotao Youyou Taotao Youyou

c. Y. chased T. and T. got tired. (3) d .Y . chased T. and Y. got tired. (4)

(1 (2-a)) ( t -a (2))

">4 ">4 Taotao Youyou Taotao Youyou

Though the basic meanings of these interpretations are always A chased B and as a result A or B became tired, there is a remarkable difference between (21a) and (21d) on the one hand, and (21b) on the other. In- terpreted as (21a), the sentence strongly suggests that by chasing Youyou, Taotao was responsible for Youyou's becoming tired. Similarly, if interpreted as (21d), the sentence means that by behaving in certain ways (e.g.,

266 YAFEI LI

not letting Youyou catch him), Taotao was responsible for Youyou's becoming tired. Remarkably, what (21a) and (21d) have in common is the extra meaning that Taotao has somehow put Youyou into the state of being tired. In contrast, no such additional meaning is associated with (21b), which simply means that Taotao chased Youyou and got tired as a result. It is this semantic difference that made Wang (1958) and Huang (1988) call examples like (21a, d) causatives and those like (21b) noncausa- tives.

Based on this semantic difference, I propose that in (21a, d), Taotao receives a Cause role and Youyou an Affectee role. 6 Cause and Affectee are different in kind from theta roles such as Agent and Patient. First, they are assigned to arguments that already bear theta roles. Taotao, for instance, receives Agent in (21a) and Patient in (21d) from Voaus (cf. (14) and the discussion thereon) in addition to Cause. Second, they are not part of the argument structure of either V~,us or V~s. The meaning of the verb zhui 'chase' does not indicate that its Patient argument is a cause for anything, though this argument in (21d) (i.e., Taotao) bears the Cause reading. Third, Cause and Affectee are not consistently associated with any particular arguments of a given verb morpheme. In (21a), the Agent argument Taotao of zhui 'chase' is Cause and the Patient argument You- you is Affectee, whereas in (21d), the Patient is Cause and the Agent is Affectee. All these properties of Cause and Affectee point to two con- clusions: (A) Cause and Affectee are different linguistic entities from theta roles. Given their evident relation to causativity, I will call them causative roles, c-roles for short. (B) C-roles do not come from the individual verb morphemes in a resultative compound but rather are a property of the compound as a whole. Because resultative compounds obviously differ from monomorphemic verbs in overtly representing the internal structure of a causal event with V .. . . and V~s, I will adopt the strongest hypothesis in (22): 7

6 Huang (1988) uses Causer and Causee to describe the same fact. I choose Cause and Affectee only to avoid possible confusion. For many people, Causer is linked to agentivity and/or volition. This is clearly not the case here. In (21d), Taotao is the cause for Youyou's being tired, but it is not necessarily true that Taotao agentively initiated the chase or intentionally made his chaser tired. Also consider (10b), in which that song is understood to have caused Youyou to become bored, though it is not agentivc nor even animate. For the separation of Cause from agentivity and/or volition, see Comrie (1976), Rosenberg (1975), Pesetsky (1987), Grimshaw (1990), Jackendoff (1990), among others. 7 While (22) applies to the current data directly, it is possible that it actually applies at more than one level of linguistic representation. Suppose that at the lexico-conceptual level, kill can be decomposed into °x DO-SOMETHING-TO y and as a result y DIE'. Then x will acquire the Cause reading and y the Affectee reading at that level. Further exploration of this possibility is beyond both the immediate concern and the limited length of the present


(22) C-roles are assigned by a predicate P only if the internal structure of a causal event is overtly represented by P.

Though a resultative structure is necessary to create c-roles, it is not sufficient, since the arguments in (21b) don't bear them. An examination of (21) shows a correlation between c-roles and the way Vres assigns its theta role a. In (21a, d), a is assigned to the object position, and the NP arguments show the Cause and Affectee readings; in (21b), a goes to the subject position, and no c-roles are assigned. (23) below further shows that whether Vcaus assigns a theta role to the object position has no effect on the assignment of the Affectee role:

(23) Taotao tiao-fan-le Youyou le

Taotao jump-bored-asp Youyou LE

a. Taotao made Youyou bored by jumping.

b. <i <a>>

L Taotao Youyou

I I Cause Affectee

In (23), Youyou only receives a from Vr~fan 'bored' but is still understood as Affectee (i.e., one whose boredom was caused by Taotao). Below is the first approximation of the descriptive conditions for c-role assignment (a condition similar to (24a) is proposed in Grimshaw 1990):

(24)a. The argument in the subject position receives the c-role Cause from a resultative compound if it receives a theta role only from Voau~;

b. The argument in the object position receives the c-rote Affectee fi'om a resnltative compound if it receives a theta role at least f r o m Vre s.

Recall that the NP in subject position must be thematically associated with the compound according to the theta criterion (section 1.2). So (24a) can be restated below:

(24)a'. The argument in the subject position receives the c-role Cause

paper. See Li (to appear) for an example of applying (22) 'cyclicly' at different levels of representation.

268 YAFE, I L I

from a resultative compound only if it does not receive a theta role from Vre~.

For all the examples discussed in this section, (24a') plus the theta criterion is equivalent to (24a) in effect, because V~aus now becomes the only available theta role assigner for the Cause argument in subject position. But (24a') also provides a natural account of (15b), repeated below:

(25) Nachang jihuang e-side henduo ren

that famine hungry-die-asp many people

Many people starved to death in that famine.

As noted earlier, nachang ]ihuang 'that famine' may well carry an intrinsic temporal theta role and therefore does not directly receive a theta role from V~aus. By (24a), this NP cannot be assigned the Cause role. But according to (24a'), it can, provided that it is not a thematic argument of Vres. s Since native speakers' judgment indicates that in (25) that famine is understood as the cause for many people's starvation to death, (24a') will be adopted in place of (24a) in the rest of this paper.

The intuitions behind (24) are clear. The argument of Vres is understood as Affectee because it refers to the participant of the resulting event and therefore the one that bears the consequence, whether or not it also participates in the causing event. And an argument that is not associated with the resulting event acquires the Cause role because its referent, as a necessary part of the whole causal event, directly or indirectly participates in bringing about the resulting event. In other words, Cause and Affectee are the roles arguments play in a causal structure when they are associated with the components of this structure in certain ways, and such associations are directly reflected through theta role assignment specified in (24). More- over, (24) encodes an asymmetric relation between the two c-roles, with Cause more prominent than Affectee, as reflected by the fact that the Cause argument must be in the subject position and the Affectee argument in the object position. This relation will be referred to as the causative hierarchy.

8 Note that in this analysis of (25), it does not matter whether the temporal NP is regarded as modifying V ~ or the whole compound.


2.2. Interactions between the Causative Hierarchy and the Thematic Hierarchy

That the causative hierarchy has the ability to 'override' the thematic hierarchy is indicated by the well-formed (21d), repeated below:



a. Youyou chased Taotao and as a result Youyou became tired.

b. (1-a (2})

">4 Taotao Youyou

i I Cause Affectee

Even when theta role assignment violates the thematic hierarchy, (26) is acceptable as long as the subject and object arguments are assigned Cause and Affectee in accordance with the causative hierarchy. Since the thematic hierarchy is respected both cross-linguistically and in Chinese other~ wise (cf. Li 1990a, also see below), I propose (27):

(27) Theta roles can be assigned contrary to the thematic hierarchy if the arguments receiving them are assigned c-roles in ways compatible with the causative hierarchy.

(26) sheds light on how the thematic hierarchy restricts theta role assignment. Though the theory only requires that theta roles be assigned according to the thematic hierarchy, the tacit assumption seems to be that a theta role is not available for assignment if a less prominent one is not yet assigned. (26) shows that this assumption cannot be correct. If the order in which theta roles are assigned strictly followed the thematic hierarchy, (26) would be impossible because in the presence of the thematically less prominent theta role 2, the more prominent 1 could never be assigned to the object position, neither could 2 be assigned 'later than' 1 to the subject position. The only way to guarantee that the inverse reading can be obtained is to assign theta roles of a verb in random order. This in turn means that the thematic hierarchy as a constraint on theta role assignment is checked only 'after' theta roles are randomly assigned. Namely, the thematic hierarchy does not constrain the process of theta role assignment; rather, it is a constraint on the result. In fact, it must

270 YAFEI LI

apply after c-roles are assigned and the causative hierarchy is satisfied. Otherwise, the inverse theta role assignment in (26) would still be ruled out before the causative hierarchy applies to save it. (28) summarizes how theta role assignment, c-role assignment, and the two hierarchies interact in this theory:

(28) Step 1: Randomly assign theta roles to syntactic argument positions;

Step 2: Where possible, assign c-roles to these positions according to (24);

Step 3: Check the result of theta role assignment according to (27).

(28) immediately accounts for the paradigm in (21), repeated below. Recall that Taotao is in the subject position and Youyou in the object position. C-roles are also shown, when present:

(29)a. (1 (2-a}}

Taotao Youyou

I I Cause Affectee

c. *(1 (2-a}) ">C Taotao Youyou

b. (1-a (2))

L 1 Taotao Youyou

d. (1-a (2)}

Taotao Youyou

I I Cause Affectee

In (29a), as the first step, 1 of Vcatt s is randomly assigned to Taotao and the identified theta roles 2-a to Youyou. Then the compound assigns Cause to Taotao and Affectee to Youyou, according to (24). Finally, theta role assignment is checked according to (28). It satisfies the thematic hierarchy, but this has no effect on the well-formedness of the sentence, because each argument has received a c-role in accordance with the causative hierarchy. Hence, the acceptable interpretation that Taotao chased You- you and made Youyou tired. In (20b), the way theta roles are assigned prevents Taotao and Youyou from receiving c-roles under (24), making it 'non-causative', as Wang (1958) terms it (cf. 1.2). But since there is no violation of the thematic hierarchy, the reading is permitted. Random

THE THEMATIC HIERARCHY AND CAUSATIVITY 271

theta role assignment also generates (20c-d). In (20d), Taotao only receives 2 from V . . . . . thus qualifying as Cause under (24a); and Youyou acquires the Affectee reading under (24b) by receiving a from V~e~ in addition to 1 of V . . . . ). With c-roles assigned in accordance with the causative hierarchy, the thematic hierarchy is overridden as specified in (27), thus allowing the inverse theta role assignment. In contrast, theta role assignment in (2%) prevents the arguments from receiving c-roles

from the compound. In the absence of c-roles, (27) requires that theta role assignment obey the thematic hierarchy, making the intended reading of (29c) ungrammatical.

2.3. Inverse Theta Role Assignment in the Ba-Construction

I first consider two properties of the morpheme ba, and then show that their interaction with resultative compounds is correctly accounted for by (28).

When used with a resultative compound, ba must introduce the Affectee argument. 9 Recall that a sentence like (21) has three readings. When the object is introduced by ba, only two of them remain:

(30) Taotao ba Youyou zhui-lei-le

Taotao BA Youyou chase-tired-asp

a. Taotao chased Youyou and as a result Youyou got tired. b. Yonyou chased Taotao and as a result Youyou got tired. c. *Taotao chased Youyou and as a result Taotao got tired.

(30a-b) correspond to (21a) and (21d) respectively, while (30c) corresponds to (21b). Notice that the two possible readings in (30) are precisely those in which the subject and object receive the Cause and Affectee

9 Following Huang (1982), Travis (t984), Wang (1985), Cheng (1988), and A. Li (1990), I assume that the argument introduced by ba receives its theta role from a verb instead of from ba directly, For the purposes of this paper, it makes no difference whether the argument is base-generated in the preverbal position or moves there, though I share Goodall's (1989) view that, at least in some cases, base-generation is unlikely, For the observation that the argument introduced by ba is affected/disposed, see Wang (1947), Li and Thompson (1981), and Cheng (1988). In this paper, I am concerned with some specific properties of ba, not a full account of the morpheme.

272 YAFEI LI

roles. This follows if ba must introduce the Affectee argument when used with a resultative compound.i°

Another property of ba is that it can assign a Case to the NP it introduces, in addition to the Case assigned by the verb. In (31-32), I assume that the argument structure of Venus chang 'sing' is (1) and that of Vr~s wang 'forget' is (a (b)):

(31) Taotao ba Youyou chang-wang-le [N~" yao shuo de hua] Taotao ba Youyou sing-forget-asp want say de words

a. Taotao sang (so beautifully that) Youyou forgot what she (Youyou) wanted to say.

b. (1 (a

I I Taotao Youyou

Cause Affectee

(h)))

1 what Youyou wanted to say

lo A reviewer notes that zhui 'chase' when used alone cannot have its object introduced by ba, and suggests that this may help clarify why (30c) is impossible:

(i) *Taotao ba Youyou zhui-te

Taotao ba Youyou chase-asp

Intended reading: Taotao chased Youyou.

The idea is that if ba cannot introduce Youyou in (i), it should not introduce Youyou in (30c) because in both cases, Youyou only receives 2 from zhui. This is tmlikely to be the reason, given the examples below:

(ii) Taotao ba Youyou da-le

Taotao ba Youyou beat-asp

Taotao beat Youyou.

(iii) Taotao ba Youyou da-lei-le

Taotao ba Youyou beat-tired-asp

4: Taotao beat Youyou and as a result Taotao became tired.

The object of da 'beat' can be introduced by ba, and the compound da-lei 'beat-tired'As well-formed, but still (iii) is unacceptable with Youyou receiving only 2 from da.

THE THEMATIC H I E R A R C H Y AND CAUSATIVITY 273

(32) *Taotao chang-wang-le Youyou [NP yao shuo de hua] ~1

Taotao sing-forget-asp Youyou want say de words

Intended reading: Same as (30).

With no theta role identification taking place, each of the three theta roles must be assigned to an NP argument. The ungrammaticality of (32) is expected if a resultative compound resembles most other Chinese verbs in assigning only one Case, leaving one of the two postverbat NP arguments Caseless. Given the well-formed (31), in which ba introduces Youyou, it is reasonable to assume that yao shuo de hua 'what she wanted to say' receives Case from the compound verb while Youyou has its Case from baJ 2 Note that Youyou is the Affectee argument since it receive a from Vres,

Given these two properties of ba (i.e., introducing an Affectee argument and assigning an extra Case), we predict that the other internal argument in (31-32), yao shuo de hua 'what she wanted to say', cannot be introduced by ba:

(33)a. *Taotao ba [NP yao shuo de hua] chang-wang-le Youyou le

Taotao ba want say de words sing-forget-asp Youyou LE

Intended reading: Same as (31).

b. (I <a <b>)>

Taotao what shewanted to say Youyou

b I Cause Affectee

If ba must introduce the Affectee argument, the NP 'what she wanted to say' must be interpreted as Affectee in (33). By the causative hierarchy, the most prominent argument (i.e., Taotao) has the causatively most prominent Cause role and the second most prominent argument gets the causatively less prominent Affectee role. In (33), this would require that

xl This sentence may be interpreted to mean 'Taotao sang s o . . . that he (i.e., Taotao) forgot what Youyou wanted to say', that is, with the postverbal NP Youyou analyzed as the subject of the relative clause modifying the object hua 'words'. This interpretation is odd pragmatically, and in any case it is irrelevant to the discussion in the text. lz This analysis is adopted from Li (1990a), which argues that theta-identification is imposed on a compound when there are more theta roles from its components than the number of Cases available to license NP arguments.

274 YAFEI LI

the NP 'what she wanted to say' be the second most prominent argument in the sentence. Note that it receives the theta role b from V .... leaving the least prominent internal argument, Youyou, with a from V~s. Since Youyou does not bear any c-role, it must receive a theta role in a way consistent with the thematic hierarchy, as required by (27). In (33), however, the thematic hierarchy is violated because Youyou is structurally less prominent than the NP 'what she wanted to say' but receives a thematically more prominent theta role, a. Hence (33) is ruled out. In contrast, (31) is well-formed with the least prominent internal argument 'what she wanted to say' receiving the least prominent theta role b. Note that under this analysis, there is no need to specify which theta role of Vre~ the Affectee argument must receive when Vr~s has more than one of them (cf. (24b)). The Affectee argument must receive a but not b in order for a less prominent argument to satisfy the thematic hierarchy,

For the same reason, we predict that when both Voau~ and V~e, are transitive, the compound can take three NP arguments and permit inverted theta role assignment as long as ba introduces the argument that receives the most prominent theta role from Vr~. In the context that Taotao used a house as the gambling price and lost it, (34) is acceptable:

(34) Neipan qi ba Taotao xia-shu-le yi-suo fangzi

that-round chess ba Taotao play-lose-asp a house

a. Taotao played that chess game and as a result lost a house. b. (1-a (2 (b)})

X I that chess game Taotao a house

t l Cause Affectee

Since that chess game only receives 2 from V ..... it qualifies as Cause; and Taotao with a from Vr~s qualifies as Affectee. Hence inverse theta role assignment is possible. A house is neither Cause nor Affectee, but it receives the least prominent b from V .... so the thematic hierarchy is also satisfied.

Further evidence that (28) interacts with the properties of ba in determining the behaviors of resultative compounds come from examples like (35), in which the compound is composed of a transitive Voaus and an intransitive Vr,s, with the intended interpretation that A cursed B with the result that C left (ma °curse' has (1 (2)) and zou 'leave' has (a)):

(35)a.

T H E T H E M A T I C H I E R A R C H Y A N D C A U S A T I V I T Y

*Taotao ma-zou-le Youyou keren (le)

Taotao curse-leave-asp Youyou guest LE

275

b. *Taotao ba Youyou ma-zou-le keren (le)

Taotao ba Youyou curse-leave-asp guest LE

No matter which of the three arguments did the cursing, which one was cursed, and which one left, the sentences are totally unacceptable. This result is expected of (35a), which may be argued to violate the Case filter with two postverbal NPs. But why is (35b) just as bad, given that ba may provide an extra Case? All the six logically possible ways to randomty assign the three theta roles to arguments in (35b) are listed in (36), and all of them must be ruled out:

c .

e .

(36) a. *(1 (2 (a))) b. *(1 (2 (a)))

1 1 X Taotao Youyou guest Taotao Youyou guest

*(1 (2 (a>)) d. *(1 (2 <a>>>

Taotao Youyou guest Taotao Youyou guest

*(1 (2 (a))) f. *(1 (2 (a))}

Taotao ¥ouyou guest Taotao Youyou guest I

i I 1 l Cause Affectee Cause Affectee

(36a) corresponds, for instance, to the reading that Taotao cursed Youyou and as a result the guest left.

Since ba must introduce the Affectee argument of the compound, (36a- d) are directly ruled out because Youyou does not receive any theta role from Vres, though there may be additional reasons why some of them are ungrammatical. In the last two cases, the syntactic subject Taotao qualifies as Cause and Youyou qualifies as Affectee. Consider (36e) first, which presumably gives the reading that Taotao cursed the guest and therefore made Youyou leave:

276 YAFEI LI

(37) (1 {2 (a)))

I X Taotao Youyou guest

Cause Mfectee

In this-case, the argument with 2 ( = 'guest') has no c-role, so by (27), whether 2 is assigned to the argument with the correct prominence should be determined by the thematic hierarchy. As shown in Li (1990a, 1993), Vcaus is the morphological head of the compound and the thematic prominence of each theta role of the head must be strictly maintained, in the sense that the most prominent theta role of V . . . . must be assigned to the most prominent argument, the second most prominent theta role must be assigned to the second most prominent argument, etc. As such, 2 must be assigned to Youyou, but it actually is assigned to the least prominent argument keren 'guest'. So (37) is ruled out as a violation of the thematic hierarchy. Next consider (36f), which is intended to mean that the guest cursed Taotao and thus made Youyou leave:

(38) (1 12 (a)))

Taotao Youyou guest

I I Cause Mfectee

With the argument receiving I from Vcaus (i.e., keren 'guest') being neither Cause nor Affectee, it must be the most prominent syntactic argument in the sentence to satisfy the thematic hierarchy. Since keren is the least prominent argument, (38) is excluded. Thus none of the six cases in (36) is grammatical.

2.4. Inverse Theta Role Assignment and Passivization

Though resultative compounds generally allow passivization with bei, inverse theta role assignment is excluded. If the active sentence has multiple readings as in (21), the one with inverse theta role assignment becomes impossible in the passive counterpart; if the active sentence only has the inverted reading (for pragmatic reasons, as in (10b)), passivization leads to ungrammaticality:


(39) Taotao bei Youyou zhui-lei-le

Taotao bei Youyou chase-tired-asp

a. Taotao was chased by Youyou and became tired as a result. (cf. (21a))

b. *Taotao chased Youyou and became tired as a result. (cf. (21d))

(40) *Youyou bei neishou ger chang-fan-le (cf. (10b))

Youyou bei that song sing-bored-asp

Intended reading: Youyou sang that song and as a result became bored.

This incompatibility between inverted theta role assignment and passivization follows from my analysis and a peculiar property of the passive morpheme bei.

In (39a), the complement of bei, i.e. Youyou, is understood as Cause and the surface subject Taotao as Affectee. This suggests that c-roles are assigned at D-structure. Otherwise (39a) would be a violation of the causative hierarchy, with the Affectee argument structurally more prominent than the Cause argument. Furthermore, it is well known that bei only operates on verbs with an affected internal argument. Since a song is not affected by singing it, (41a) cannot be passivized. But killing a kangaroo does affect its physical state, so (42b) is well-formed. For rel- evant discussion and references, see Li and Thompson (1981).

(41)a. Taotao chang-le neizhi get

Taotao sing-asp that song

Taotao sang that song.

b. *Neizhi ger bei Taotao chang-le

that song bei Taotao sing-asp

(42) a. Taotao sha-le neizhi daishu

Taotao kill-asp that kangaroo

Taotao killed that kangaroo.

b. Neizhi daishu bei Taotao sha-le

that kangaroo bei Taotao kill-asp

That kangaroo was killed by Taotao.

278 YAFEI LI

This property of bei sheds light on why (39) also fails to mean that Taotao was chased by Youyou and as a result Youyou became tired (cf. (21b)). Recall that in such an interpretation, the arguments do not exhibit the Cause-Affectee readings. If the affected internal argument corresponds to the Affectee argument in the case of a resultative compound, it is expected that the structure with this intended reading cannot be passivized.

Now consider the ways theta roles are assigned at D-structure in (39a-

(1 (2-a)) (for (39a)) (43)b. (1-a (2)) (for (39b))

I I X Youyou Taotao Youyou Taotao

I I I I Cause Affectee Cause Affectee

(43)a.

b):

In (43a), passivization with bei is allowed since the internal argument of the compound is the Affectee. To see why passivization is excluded in (43b), consider what is meant by an internal argument. In all the cases I know of, an internal argument has two crucial properties. Structurally, it is the complement of a verb, and thematically, it receives an internal theta role from the verb. In (43b), the object argument qualifies as an internal argument structurally, but not thematically because it does not receive the internal theta role from the compound. Taking advantage of this observation, I propose that just as the notion subject must be distinguished from the notion external argument (Grimshaw 1990), object and internal argument are also two different notions. Subject and object are purely structural notions, while external and internal arguments reflect their thematic relations with the verb. In particular, an internal argument of a verb is an argument in the object position receiving an internal theta role from the verb. Returning to (43b), we have seen that bei, instead of simply passivizing a verb with an object, is sensitive to the specific thematic relation between the verb and its object. This indicates that bei operates on an internal argument rather than a structural object. It follows that (43b) is not allowed in a passive sentence because the Affectee argument is not the internal argument of the compound verb. Hence the absence of the reading in (39b) and (40).13

13 It is worth pointing out that this account of (39-40) holds independently of the specific analysis of bei. All the proposals I know of can incorporate it easily. For some such proposals, see A. Hashimoto (1971), Chu (1973), Huang (1982), Feng (1990), A. Li (1990), and Goodall (1991).

THE THEMATIC H I E R A R C H Y AND CAUSATIVITY 279

3. C A U S A T I V E H I E R A R C H Y VS. A S P E C T U A L H I E R A R C H Y

In order to explain why English psych-verbs like fear have an Experiencer subject and a Theme object whereas those like scare haste a Theme subject and an Experiencer object, Grimshaw (1990) proposes that the arguments of a verb are structured along two dimensions, aspectual and thematic, and that the subject of a sentence must be the aspectually most prominent argument (section 2.3). Furthermore, she argues that the aspectually most prominent argument is the one associated only with the activity part of (44), which provides a template for the event structure of every verb (also see Pustejovsky 1991):

(44) event

activity state

Now the event of The movie scared the children may be decomposed into the initial activity of the movie doing something (to the children) and the final state of the children being in fear. ~4 Since the movie is a participant of the initial activity but not one of the final state, it is the aspectually most prominent argument for the verb scare and thus is the subject, though, as Theme, it is thematically less prominent than the children, which receives the Experiencer theta role. The reader is referred to Chap- ter 2 of Grimshaw (1990) for details.

Grimshaw's analysis of English psych-verbs are similar to my analysis of Chinese resultative compounds in at least two ways. First, she also assumes the existence of a hierarchical system independent of the thematic hierarchy which, when in conflict, overrides the latter in determining the distribution of theta roles. Secondly, the resultative compounds in Chinese are practically an overt representation of the event structure in (44), with Voa~ denoting the state part. 12 1 am inclined to believe that these similarit- ies are not accidental and that the two analyses are likely to reflect the properties of the same component of the language faculty. It is beyond the goal of the present paper to adequately explore the possibility of

14 See also Jackendoff (1983) and Hale and Keyser (1986, 1987) on decomposing causative verbs in similar ways. 15 Most resultative compounds are composed of an activity verb as Vo~u~ and a state verb such as Vr~, but there are exceptions as discussed in Li (1993):

(i) Taotao e-ku-le

Taotao hungry-cry-asp

Taotao was so hungry that he cried.

2 8 0 Y A F E I L1

merging these two analyses. Below, I consider a case in which the two theories diverge.

By definition, the effect of my causative hierarchy is restricted to a proper subset of compounds (and possibly verbs, see note 8) because not every theta role assignment of every verb or verbal compound gives rise to c-role assignment - only when c-roles are assigned can the causative hierarchy exist. In contrast, Grimshaw's more general aspectual dimension is meant to determine the assignment of every individual theta role of every verb (Grimshaw, section 2.5). Now consider (lb) again, repeated in (45):



a. Taotao chased Youyou and as a result Taotao got tired.

b. (1-a {2)>

P I Youyou Taotao

Recall that with the intended reading, the sentence does not indicate which argument is Cause and which one is Affectee. If the aspectual dimension associated with a verb always determines the distribution of theta roles to arguments, then it must have determined that Taotao with 1-a is aspectually more prominent than Youyou with 2 because Taotao is the subject while Youyou is the object. But now the aspectual prominences reflected in this theta role assignment are exactly the opposite of what would be consistent with the causative hierarchy, according to which the argument with the more prominent c-role (i.e., Cause) should be exclus- ively associated with Voau~ (i.e., have the theta role 2) while the one with the less prominent c-role (Affectee) should be associated with Vre~ (i.e., be assigned l-a) (cf. (24)). Since the causative hierarchy and the aspectual dimension seem to have conflicting standards in determining the prominences of the two arguments in (45), it is not clear to me at this stage how the two can be merged. 16

16 A possibility is that in the same dimension where Cause and Affectee exist, there are other roles R1 . . . . , Rn all of which (including Cause and Affectee) are ranked by a single hierarchy H. Then the two arguments in (44) are still assigned R~ and Rj according to H even when they do not bear Cause and Affectee. A question to be answered before adopting this approach is why the effects of R1, • • •, Rn in determining theta role assignment are apparently redundant: Suppose that in (44), Taotao has Ri and Youyou has R# Then it must be that Ri is more prominent than Rj according to H, and presumably it is this hierarchical relation that determines the particular theta role assignment in (44). But this is unnecessary,

THE THEMATIC HIERARCHY AND CAUSATIVITY 281

To summarize, the rather complicated and unexpected behaviors of the Chinese resultative compounds can be explained if we assume (A) that in a structurally represented resultative construction, an argument may receive both some theta role(s) and a c-role, (B) that each set of roles is linked to syntactic arguments according to the respective hierarchy, and (C) that the effects of the thematic hierarchy may be overriden by the causative hierarchy. The fact that a similar theory can be independently motivated by the analysis of English psych-verbs indicates the common structural core of the phenomena at issue.

REFERENCES

Baker, Mark: 1988, Incorporation: A Theory of Grammatical Function Changing, University of Chicago Press, Chicago.

Cheng, Lisa: 1988, 'Aspects of the Ba-Construction', in Carol Tenny (ed.), MIT Lexicon Project Working Papers 24, Cambridge, pp. 73-84.

Cheng, Lisa and Elizabeth Ritter: 1987, 'A Small Clause Analysis of Inalienable Possession in Mandarin and French', in James Blevins and Juli Carter (eds.), NELS 18, pp. 65-78.

Chomsky, Noam: 1981, Lectures on Government and Binding, Foris, Dordrecht. Chu, Chauncey: t973, 'The Passive Construction: Chinese and English', Journal of Chinese

Linguistics 1, 437-470. Comrie, Bernard: 1975, 'Review of Jackendoff 1972', Linguistics 160, 71-86. Feng, Shengli: 1990, 'The Passive Construction in Chinese', unpublished manuscript, Univer-

sity of Pennsylvania. Goodall, Grant: 1986, 'On Argument Structure and L-marking with Mandarin Chinese ba',

talk presented at NELS 17. Goodall, Grant: 1989, 'Evidence for an Asymmetry in Argument Structure', Linguistic

Inquiry 20, 669-674. Goodall, Grant: 1991, 'Mandarin Chinese Passives and the Nature of Case Absorption', talk

presented at the 3rd North American Conference on Chinese Linguistics, Ithaca. Grimshaw, Jane: 1990, Argument Structure, MIT Press, Cambridge. Hale, Kenneth and Jay Keyser: 1986, 'Some Transitivity Alternations in English', MIT

Lexicon Project Working Papers 7, Cambridge. Hale, Kenneth and Jay Keyser: 1987, 'A View from the Middle', MIT Lexicon Project

Working Papers 10, Cambridge. Hashimoto, Anne: 1971, 'Mandarin Syntactic Structures', Unicorn 8, 1-146. Higginbotham, James: 1985, 'On Semantics', Linguistic Inquiry 16, 547-593. Huang, James: 1982, Logical Relations in Chinese and the Theory of Grammar, Ph.D.

dissertation, MIT. Huang, James: 1988, 'Wo Pao De Kuai and Chinese Phrase Structure', Language 64, 274-

31t. Jackendoff, Ray: 1972, Semantic Interpretation in Generative Grammar, The MIT Press,

Cambridge. Jackendoff, Ray: 1983, Semantics and Cognition, MIT Press, Cambridge. Jackendoff, Ray: 1990, Semantic Structure, MIT Press, Cambridge. Li, Audrey: 1990, Order and Constituency in Mandarin Chinese, Kluwer, Dordrecht°

because the thematic hierarchy is already sufficient to guarantee that t-a is assigned to Taotao and 2 is assigned to Youyou (cf. 2.2).

282 YAFEI LI

Li, Charles and Sandra Thompson: 1981, Mandarin Chinese: AFunctional Reference Gram- mar, University of California Press, Berkeley.

Li, Yafei: I990a, 'On V-V Compounds in Chinese', Natural Language & Linguistic Theory 8, 177-207.

Li, Yafei: I990b, 'X°-binding and Verb-incorporation', Linguistic lnquiry 21, 299-426. Li, Yafei: 1993, 'Structural Head and Aspectuality', Language 69, 480-504. Li, Yafei: to appear, 'Chinese Resultative Constructions and the UTAH', in J. Packard

(ed.), Word Formation in Chinese, Mouton de Gruyter, Berlin. Pesetsky, David: 1987, 'Binding Problems with Experiencer Verbs', Linguistic lnquiry 18,

126-140. Pesetsky, David: 1995, 'Zero Syntax', MIT Press, Cambridge. Pustejovsky, James: 1991, 'The Syntax of Event Structure', Cognition 41, 47-81. Rosenberg, M: 1975, 'Generative vs. Interpretive Semantics', Foundations of Language 12,

561-582. Thompson, Sandra: 1973, 'Resultative Verb Compounds in Mandarin Chinese: A Case for

Lexical Rules', Language 49, 361-379. Travis, Lisa: 1984, Parameters and Effects of Word Order Variation, Ph.D. dissertation,

MIT. Wang, Li: 1947, Zhongguo Xiandai Yufa, China Press, Shanghai. Wang, Li: 1958, Hanyu Shirao, Science Press, Beijing. Wang, Peter: 1986, 'Additional Support for the Transformational Approach to Mandarin

Ba and Bei', in S. Teng (ed.), Readings in Chinese Transformational Syntax, Crane Publishing Co., Taipei, pp. 67-75.

Received: 8 January 1992 Revised: 24 November 1993

Department of Linguistics University of Wisconsin at Madison Madison WI 53706

The thematic hierarchy and causativity

Documents

Transcript of The thematic hierarchy and causativity