To appear in 2009 Language and Linguistics 10.3.
Unifying the Long Passive and the Short Passive:
On the Bei Construction in Taiwan Mandarin
*One-Soon Her
Graduate Institute of Linguistics & Research Center of Mind, Brain, and Learning
National Chengchi University
This paper has three goals: to show that the Mandarin bei is a unified construction, to explore a universal characterization of the passive, and then to demonstrate that the bei construction is a genuine passive. The bei construction comes in two forms, one with an overt agent, known as the long passive, and the other without, known as the short passive. Previous accounts assign two very different structures to the two forms and posit two lexically different
bei’s. This paper, however, based largely on newly discovered natural occurring data in
Taiwan Mandarin, demonstrates that bei behaves rather similarly with or without an overt agent and thus dismisses the short passive and argues for a unified bei. Specifically, the bei construction is analyzed, within the Lexical-Functional Grammar, as the passive counterpart of the active ba construction. The paper then explores a universal characterization of the passive and a typology of the syntactic assignment of the demoted agent, and it demonstrates that bei’s agent phrase, demoted to object, is in line with the universal characterization of passive. Key words: long passive, short passive, universal, bei, ba, demotion, topicalization
1. Introduction
The precise nature of the Mandarin bei construction, along with the ba construction, has been one of the most contentious issues in Chinese linguistics. Even the part-of-speech of the two elements, examples of which are given in (1), has long been a topic of debate. While many, e.g., Wang (1970), Li and Thompson (1981), Her (1985-6), J. Huang (1982), L. Huang (1990), Y. Li, (1990), McCawley (1992), and Tsao (1996), have viewed them as prepositions, others, e.g., Hsueh (1989), Her (1989), Chiu (1993), Ting (1998), J. Huang (1999), Bender (2000), and D. Li (2003) have argued that they are in fact verbs.1
(1) a. Amei bei (gemi) weizhu le. Amei BEI fans encircle PERF2 ‘Amei was encircled (by the fans).’ b. Gemi ba Amei weizhu le.
fans BA Amei encircle PERF
‘The fans encircled Amei.’
* I gratefully acknowledge the financial support by the NSC grant 97-2410-H-004-112-MY2 and the constructive comments by the two anonymous Language and Linguistics reviewers. I also thank Simon Smith and my research assistants Meng-Ying Chen, Hsiao-Chien Feng, Yu-Ying Ho, Wan-Ju Huang, Li-Hsin Ning, Guang-Zhong Wu, and Yi-Ting Xie, for their assistance with data-collection and proofreading. I thank Jen Ting for the discussions we had a few years back when the idea of a unified
bei first came to me. I am however solely responsible for the content presented here.
1 Another lexical item in the language that involves a similar debate is the dative gei; see Her (2006) and Ting and Chang (2004).
2
The following abbreviations are used: CL (classifier), PASS (passive marker), PERF/PFV (perfective maker), PRT (particle).
Notice that in (1a) the agent phrase after bei need not appear. In the literature, the bei construction with the overt agent has come to be known as the long passive, and the one without, the short passive. All researchers who made this formal distinction have assigned two distinct structures to the two passives, based on claims that the behavior of bei varies depending on the presence or absence of the agent phrase (e.g., Her 1989 and 1991, Ting 1998; J. Huang 1999; J. Huang et al 2008). Most notably this is based on the observation that the short passive does not seem to allow long-distance gaps, as shown in the example in (2), taken from J. Huang (1999).
(2) a. Zhangsan bei Lisi pai jingcha zhua-zou le. (J. Huang 1999: 450 (51a)) John BEI Lee send police arrest PERF
‘John was “sent-police-to-arrest” by Lee.’
b.*Zhangsan bei pai jingcha zhua-zou le. (J. Huang 1999: 450 (52a)) John BEI send police arrest PERF
This paper disputes such claims and aims to accomplish three things: first to demonstrate that the short form essentially does not exist, as it is simply the long passive with a covert agent NP; to then propose a formal account of bei as a unified verb; and finally to show that bei fits in a universal characterization of passive and is precisely the passive counterpart of ba.
A notable feature of this paper is the source of its data. Unlike most of the previous studies on the syntax of Mandarin bei, this study focuses on Taiwan Mandarin (cf., Cheng 1985, Hsu 2006, Her 2009). Given the wide distribution of Mandarin in China and in the world, the language now comes in a great many varieties. To foster a greater accountability and reliability of the data employed and thus to minimize controversy over grammaticality issues, a single dialect is selected. Also, unlike most other papers on bei, the majority of examples used in this study are natural occurring data found on the Internet with Google search within the domain of Taiwan, further aided by grammaticality judgment and other solicited data from native Taiwan Mandarin speakers. Smith (1999: 15), a renowned generativist, has this to say regarding the kinds of data useful in the generativist research:
While idealization is necessary, it must be emphasized that idealization away from speech errors, for instance, still allows one to use performance mistakes such as slips of the tongue as evidence… All our understanding of linguistic knowledge…has to be supported by evidence, and where that evidence comes from is limited only by our imagination and ingenuity. (Smith 1999:15, emphasis in original)
Chomsky himself, in spite of the widely circulated misconception, in fact does not object to the use of corpus data per se. Here is his precise position first hand:
You don’t take a corpus, you ask questions. You do exactly what they do in the natural sciences. ... You have to ask probing questions of nature. That’s what is called experimentation, and then you may get some answers that mean something. ... You can take as many texts as you like, you can take tape recordings, but you’ll never get the answer. (Chomsky, quoted in Aarts (2000: 6))
Thus, as long as one starts out with a probing question of nature, there is simply no good reason for the overwhelmingly one-sided reliance on introspected data in the mainstream generative theorizing. However, the situation is changing as more and
more linguists are now using corpus data. Brian Joseph (2004: 382), then editor of Language, confirms this observation, with reference to the articles of the journal, “…we seem to be witnessing…a shift in the way some linguists find and utilize data – many papers now use corpora as their primary data, and many use internet data”. In this paper, I thus also aim to show that Fillmore (1992:35) is right, that, in addition to introspected data, corpora allow the establishment of new facts, some of which one ‘couldn't imagine finding out about in any other way’.
The paper thus aims to answer this probing question: does the short passive really exist in Mandarin Chinese? It is organized into 5 sections. After this introduction, section 2, based on data from Taiwan Mandarin, shows that bei behaves rather similarly with or without the agent phrase and thus that positing two distinct bei’s and two different structures for the long passive and the short passive cannot be justified. Section 3 then proposes a unified formal account of the bei construction within the Lexical-Functional Grammar (LFG) (cf., Bresnan 1982, 2001). In section 4, a universal characterization of the notion of passive and the typology of the demoted external argument are explored, and bei will be shown to be precisely the passive counterpart of the active ba construction. Section 5 concludes the paper.
2. Unifying the long passive and the short passive
This section will demonstrate the empirical facts of the long passive and the short passive respectively and the A’-dependency analysis for the long passive and the A-movement analysis for the short passive, which both J. Huang (1999) and Ting (1998) advocate. New evidence introduced here will establish the inadequate empirical coverage by the accounts offered in Ting (1998) and J. Huang (1999) and also the fact that the short passive in essence does not exist as it allows the same range of syntactic behavior as the long passive. Thus, it will be shown that bei is in fact one single unified lexical item. Note also that the account of bei in the book The Syntax of Chinese (J. Huang, A. Li, and Y. Li 2008) is essentially adopted from J. Huang (1999).
2.1 Empirical facts on the long passive
The Chinese long passive is called ‘long’ for two reasons: it has an overt agent phrase immediately after bei, and, perhaps more importantly, it allows the affected theme in the passive to be identified with a long-distance gap. The example in (3) is taken from J. Huang (1999) and Ting (1998) cites similar examples.
(3) Zhangsan bei Lisi pai jingcha zhua-zou le. (J. Huang 1999 : 440 (24)) John BEI Lee send police arrest PERF
‘John was “sent-police-to-arrest” by Lee.’ 3
(4) John was arrested by the police sent by Lee.
As seen in (3), the long passive bei construction is very different from the English be-passive in (4), where the passivized theme, John, must be identified with the logical object of the passivized verb, arrested. However, in (3), while the
3 As English passive allows no unbounded gaps, there are no good translations available in the passive form. This paper thus follows J. Huang (1999) and indicates the long distance passive this way.
expressed agent, Lisi, is the agent who sent the police, the matrix subject, Zhangsan, is identified with the object gap of zhuazou ‘arrest’, which is further embedded in the clause headed by pai ‘send’.
In terms of argumentation, in order to prove that the long passive allows long-distance gaps, one only needs a few well-formed examples. It is not necessary, and in fact not possible, to demonstrate that all long-distance gaps are well-formed in the long passive. The unbounded nature of the filler-gap relationship in the long passive is indeed easily verified by many similar examples found on the Internet. Two such examples are listed below in (5) and (6).
(5) Duoshao kunnan dou bei ta shefa kefu le. many difficulty all BEI he manage overcome PERF
‘All of the many difficulties were “managed-to-overcome” by him.’ (6) Bei ta qitu nuyi de ziyou renmin.
BEI he attempt enslave DE free people
‘The free people who were “attempted-to-enslave” by him.’
However, Ting (1998) and J. Huang (1999) further observe that the object position of the transitive verb that heads the final embedded clause need not be empty and can be filled by an overt resumptive pronoun, as in (7).
(7) Zhangsani bei Lisi piping-le tai yidun.
John BEI Lee criticize-PERF he once ‘Zhangsan was criticized once by Lisi.’
Ting (1998) and J. Huang (1999) also note that the long passive allows the particle suo, as in (8), while the only other construction where suo appears is a relative clause with a relativized object, as in (9).
(8) Ta-de zhengci bu bei fating suo caixin. he-POSS testimony not BEI court SUO believe ‘His testimony was not believed by the court.’ (9) Ta jinnian suo chuban de shu dou hen hao.
he this-year SUO publish DE book all very good
‘Books which he has published this year are all very good.’
Given the consensus that suo, a remnant from Classical Chinese, is an objective pronoun of sort and must be accompanied by an empty object position (Chiu 1993, 1995; Ting 1995, 2003), both J. Huang (1999) and Ting (1998) take it as further evidence that the long passive must involve a gap or a resumptive pronoun. They thus presume that in the long passive there must be a lower element co-indexed with the matrix subject and that this element must be either a gap or an overt pronoun. In other words, if there is no gap or the gap is lexically filled by a non-pronominal NP, the long passive is ungrammatical. Ting (1998: 342 (56)) cites the following examples as evidence.
(10) a.*Zhangsan bei liumang da-le Lisi. John BEI hoodlum hit-PERF Lee
‘John was affected by the hoodlum’s hitting Lee.’ b.*Zhangsan bei wawa ku-le.
John BEI baby cry-PERF
‘John was affected by the baby’s crying.’ c.*Zhangsani bei Lisi piping-le zijii.
John BEI Lee criticize-PERF self ‘Zhangsan was criticized by Lee.’
Again, in terms of argumentation, for the position of Ting (1998) and J. Huang (1999) to hold, there cannot be any well-formed case of the long passive that does not contain a gap or resumptive pronoun. However, such well-formed long passives abound and again can be easily verified by many examples found on the Internet. Two are given in the discourse cited in (11).
(11) a. Zongci ... gan-le jingcha zheme jiu, Zongci be-PERF police so long ‘Zongci … has been a cop for so long,’ b. diyi ci yudao sha-bu-si de ren,
first time encounter kill-not-dead DE person
‘for the first time he encountered a person that couldn’t be killed;’ c. bujin bei ta sha-le tongshi,
not-only BEI he kill-PERF partner
‘he (Zongci) not only was “killed-partner” by him,’ d. hai chadian bei ta pao-le.4
also almost BEI he escape-PERF
‘but also almost had him escaped on him.’
In (11c), the verb sha ‘kill’ has its object lexically filled by tongshi ‘partner’, not a resumptive pronoun. In (11d), the verb pao ‘escape’ is intransitive and does not allow an object; thus, a gap or resumptive pronoun is not even possible. In (11c), the only link available between the matrix subject, Zongci, and tongshi ‘partner’, the object of sha ‘kill’, is that the former may be interpreted as the possessor of the latter. Like (11c), in (12) and (13), there are no gaps in the embedded clause and there are no links whatsoever between the implicit matrix subject, which presumably is the speaker, and any element in the embedded clause.
(12) Yinwei wode dayi, bei ta chenggong le.5 because-of my carelessness BEI he succeed
‘Because of my carelessness, he has succeeded (on me).’
4 This is a genuine example found in a novel posted on a website in Taiwan, and its URL is: www.nch.com.tw/novel.php?nid=10868&cat=40917&sid=15.
(13) Meiyou tianli, hai zhende bei ta qudao laopo le.6 no justice even true BEI he marry wife PERF
‘There is no justice, it’s even true that he found a wife and got married.’ Clearly then, there are indeed some well-formed long passive sentences where no gaps or resumptive pronouns can be identified, (11d) and (13) being the clearest and non-controversial examples.
2.2 A’-dependency for the long passive
Both J. Huang (1999) and Ting (1998) contend that the Mandarin long bei passive should have the structure and derivation shown in (15), similar to the English tough-construction as analyzed by Chomsky (1980), shown in (14).
(14) This problemi is tough [CP OPi (for you) to solve ti]. |___ predication ____||____ movement ____| (15) IP NP V' V IP NP V' V NP
Zhangsan bei OP Lisi da-le t Zhangsan was hit by Lisi.
...
... IP NOP
↑
There are several important structural features in the analysis of (15). First of all, the long passive bei selects an NP as its subject and an IP clause as its complement. Secondly, it is assumed that the final embedded clause is always headed by a transitive verb; thus, in (15) the verb da ‘hit’ must be transitive. Thirdly, the object of this verb may be occupied by a null operator (NOP). In terms of derivation, the object of the final embedded clause, if it is an NOP, moves to adjoin the embedded IP. Thus, the fourth structural feature is that the long passive bei must lexically specify that its subject controls the NOP in its complement IP. Since the landing site of the movement involved here is a non-argument position, it is A’-movement in nature.
However, as mentioned earlier, the object of the transitive verb that heads the final embedded clause may also be filled by an overt resumptive pronoun, as in (16). According to Ting (1998), the A’-movement analysis in (15), arguably, does not apply here. Instead, the NOP, now based-generated as adjunction to the IP, enters into an A’-binding relation with the resumptive pronoun. Note that the control relation between bei’s subject and the NOP in its complement IP still holds.
(16) Zhangsani bei [IP OPi [IP Lisi piping-le tai yidun]]
John BEI Lee criticize-PERF he once ‘Zhangsan was criticized once by Lisi.’
Aside from the non-movement A’-binding account for long passive resumptive pronouns, which is preferred by Ting (1998), it is conceivable for the relation between the filler Zhangsan and the resumptive pronoun gap ta ‘he’ to be accounted for by movement. J. Huang (1999: 443) also cites such sentences as further evidence for the A’-movement, or wh-movement, which accounts for the long passive but offers no further details. Zhang (2004), for example, proposes to extend Kayne’s (2002) cluster-split hypothesis and view a resumptive pronoun and its antecedent as a cluster integrated into the relevant structure. The antecedent then moves, thus leaving the resumptive pronoun in-situ at the trace position of its antecedent.
(17) Zhangsani bei [IP OPi [IP Lisi piping-le ti-tai yidun]]
John BEI Lee criticize-PERF he once ‘Zhangsan was criticized once by Lisi.’
As shown in (17), the NOP and the resumptive pronoun ta ‘he’ are first generated in the same position as a cluster [OPi-tai]; the NOP then moves, leaving ta
‘he’ standing alone at the NOP’s trace position. Assuming that the particle suo is an object resumptive pronoun, one can extend the same A’-binding or A’-movement account outlined for (16) and (17). The A’-dependency account offered by Ting (1998) and J. Huang (1999) is thus adequate for long passives with a gap or a resumptive pronoun. However, this account’s strict requirement of a syntactically-licensed dependency relation, by either movement or binding, forces the prediction that long passives where there are no gaps and no resumptive pronouns are ill-formed. As seen in (11), (12), and (13), this is incorrect.
Furthermore, this A’-dependency account also predicts that the final embedded verb in the long passive is always transitive. This again is not always the case. Some examples found on the Internet are given in (18).
(18) a. Wo bei gugu fa-zhan.7 I BEI aunty punish-stand
‘I was “punished-to-stand-in-a-corner” by my aunty.
b. Pengyou bu shou niao xing, friend not familiar bird trait
weishi deshihou you bei ta taotuo.8 feed when again BEI it escape
‘(My) friend wasn’t familiar with the traits of the bird. At the time of feeding, (he) again had it escape on him.’
c. Zui ou-de shi juran hai bei ta chenggong-le.9 Most upsetting be surprisingly even BEI he succeed-PERF ‘What’s most upsetting is that, surprisingly, he even succeeded.’
d. Wo ..la-bu-zhu fengzheng-de xi xian, bei ta feizou-le.10 I hold-not-on kite-POSS thin line BEI it fly-away-PERF
‘I couldn’t hold on to the kite’s thin line and had it fly away on me.’ In these four examples, fa-zhan ‘punish-to-stand’, taotuo ‘escape’, chenggong ‘succeed’, and feizou ‘fly away’ are all intransitive verbs and do not allow an object. In these long passives, a dependency relation between a non-existent object and the matrix subject is simply impossible. The A’-dependency account, which by definition requires such a dependency relation, is too restricted and fails to cover the full range of bei long passives.
2.3 Empirical facts on the short passive
As for the short passive, according to J. Huang (1999) and Ting (1998), the need to distinguish it from the long passive is not so much the fact that it has no overt agent phrase but the fact it allows no long-distance gaps. J. Huang (1999) lists the sentence in (19a) as evidence and Ting (1998) also gives a similar example, as in (19b).
(19) a.*Zhangsan bei pai jingcha zhua-zou le. (J. Huang 1999: 450 (52)) John BEI send police arrest PERF
‘John was “sent-police-to-arrest”.’
b.*Lisi bei pai wo zhua-zou le. (Ting 1998: 324 (13b))
Lee BEI send I arrest PERF ‘Lee was “sent-me-to-arrest”.’
Without a proper context, the examples in (19) are admittedly odd; however, a far more important point to be noted is that, in order to justify the distinction of the short passive, it is not enough to show that the short passive does not allow one particular long-distance gap or even some long-distance gaps. Instead, one must demonstrate that all long-distance gaps are barred when bei appears without an overt agent phrase. Consequently, one only needs to demonstrate that the short passive in fact allows some long-distance gaps to unify the long passive and the short passive in this regard. And well-formed examples abound. Sentences in (20) have exactly the same object-control
8 diary.yam.com/pets/article/595223
9 www.xxxs.cn/html/kumingdaizongcai/1003.html 10 www.wzwx.com/lqyy/yy/424e.htm
structure as that of those in (19). Sentences in (21) have the subject-control structure. All are authentic examples from the Internet.11
(20) a. Wo gang zoujin gongyuan, I just enter park
jiu bei pai jingcha qing jin-le paichusuo.12 at-once BEI sent police invite enter-PERF precinct-station ‘Just as I was entering the park, I was “sent-police-to-invite-into-
the-precinct-office”.’
b. Tongdao dou yijing bei pai bing bashou.13 passage all already BEI send troop guard ‘All passages have been “sent-troops-to-guard”.’ (21) a. Gongsi-de wanglu bei qitu ruqin.14
company-POSS network BEI attempt hack
‘The company network has been “attempted- to-hack”.’ b. Ziliao bei shefa kaobei le.15
document BEI manage copy PERF
‘The documents have been “managed-to-copy”.’
Thus, it is pointless to dispute the grammaticality of the examples in (19) per se, as the grammaticality and authenticity of examples in (20) and (21) conclusively demonstrate that the long passive and the short passive are rather similar in allowing some long-distance gaps. In other words, there is no exclusive correlation between an agent phrase and a long-distance gap.
Ting (1998) and J. Huang (1999) also point out that while the long passive allows a resumptive pronoun at the gap, as shown in (22a) and (23a), the short passive requires the gap to be lexically unfilled, as shown in (22b) and (23b).
(22) a.?Zhangsani bei wo piping-le tai yidun. (Ting 1998: 322 (6))
John BEI I criticize-PERF he once ‘John was criticized once by me.’
b. Zhangsani bei piping-le (*tai) yidun. (Ting 1998: 324 (13))
John BEI criticize-PERF he once ‘John was criticized once.’
11 An anonymous reviewer suggested that frequency counts be provided for both short passives and long passives to show that the former is not marked. I contend that this is not necessary, for two good reasons besides the fact that it is rather unpractical. First, frequency counts are never an issue in syntactic studies that rely on introspection, which form the greatest major of generative works. More importantly, even if it turns out to be true that well-formed short passives are less frequent than their long passive counterparts, thus more marked, it would not change the logical conclusion that the two forms share an identical structure. In short, it is the grammaticality of the data that matters.
12 www2.tianya.cn/new/Publicforum/Content.asp?idWriter=0&Key=0&strItem=no16&idArticle= 115318&flag=1 13 kuangge.net/writings/read.aspx?chapterid=6636&writingid=104 14 download.microsoft.com/download/6/b/3/6b3eb4a5-1ba1-4e37-a501-73b977f9a5c8/032906_ WindowsServer_am.ppt 15 www.mobile01.com/topicdetail.php?f=226&t=51836&p=8
(23) a. Zhangsani bei Lisi da-le tai yixia. (J. Huang 1999: 443 (34))
John BEILee hit-PERF he once.
‘John was hit once by Lee.’
b. Zhangsani bei da-le (*tai) yixia. (J. Huang 1999: 452 (55))
John BEI hit-PERF he once ‘John was hit once.’
First of all, as J. Huang (1999: 443) has noticed, in (22a) and (23a), where the long passive allows a resumptive pronoun, the additional adjunct phrase yidun ‘once’ or yixia ‘once’ is necessary to make the pronoun more acceptable, as shown in (24). Note also that even (22a) is marked as marginal by Ting (1998).
(24) *Zhangsani bei Lisi da-le tai. (J. Huang 1999: 427 (5a))
John BEILee hit-PERF he.
‘John was hit by Lee.’
Clearly then, even in long passives a resumptive pronoun in the gap position is subject to intricate restrictions, which require further research to clarify but are not
immediately relevant to our discussion.16 What is important here is whether
resumptive pronouns are exclusive to long passives or they are likewise well-formed in some (thus not all) short passives. Again, plenty instances are found; four examples from the Internet are given in (25).
(25) a. Ta bai pa bei renwei tai wufa guanjiao haizi.17
he father afraid BEI consider he fail discipline children
‘His father was afraid to be considered that he failed to discipline his children.’
b. Shuijiao de reni dou hui bei shuocheng tai shangxian-le.18
sleep DE person all will BEI describe-as he on-line-PERF ‘Anyone who is sleeping would be described as he is “on-line”.’ c. Tai bei duozou-le tai-de zhongsheng quanyi.19
he BEI deprive-PERF he-POSS lifelong rights ‘He was deprived of his lifelong rights.’
None of the bei short passives in (25) has an overt agent but each has a resumptive pronoun at the gap position. In (25a), the matrix subject ta ‘he’ is co-indexed with the object of the verb renwei ‘consider’, ta ‘he’, a resumptive pronoun. In (25b), the matrix subject is shuijiao de ren ‘a person who is sleeping’, which is co-indexed with the lower ta ‘s/he’. In (25c), the resumptive pronoun ta ‘he’ serves as the possessor of the retained object and is co-indexed with the matrix subject.
16
Ning (2008) in fact discovers a general preference for gaps over resumption in Chinese relative clauses.
17 www.pczone.com.tw/vbb3/archive/t-86499.html 18 www.student.tw/db/archive/index.php?t-108308.html 19 http://www.grass-land.com/qqc/03/qqc0311b.html
As further evidence to the resumptive pronouns in the short passives in (25), examples in (26) and (27) show that the position occupied by the resumptive pronouns can also be a reflexive ziji ‘self’ or a gap, respectively. These are also genuine examples from the Internet.
(26) a. Tongxinglianzhei bu xiang bei renwei zijii shi guaiwu.20, 21
homosexuals not want BEI consider self be freak ‘Homosexuals do not want to be considered freaks.’
b. Nii...jiu zheyang bei qiangzou-le zijii–de qingmeizhuma!22, 23
you exactly this-way BEI rob-PERF self-POSS childhood-sweetheart ‘This is exactly how you were robbed of your childhood sweetheart!’ (27) a. Nii bei renwei ei shi shengming de ruozhe.24
you BEI consider be life DE weakling
‘You are considered to be a weakling in life.’
b. Woi ..bei qiangzou-le ei qian hai bei dashang-le ei jiao.25
I BEI rob-PERF money also BEI injure-PERF foot ‘I was robbed of (my) money and also had (my) foot injured.’
Thus, even though (24) is ill-formed and, like long passives, not all short passives allow resumptive pronouns, it is without question that some short passives do. Again, there is no exclusive correlation between an agent phrase and a resumptive pronoun. The long passive and the short passive are again alike in this regard. In all the short passive examples in (25) to (27) an overt agent phrase can be inserted after bei and turn the short passives into long passives that are grammatical. Short passives in (26) are used to illustrate this point, shown in (28).
(28) a. Tongxinglianzhei bu xiang bei renmen renwei zijii shi guaiwu.
Homosexuals not want BEI people consider self be freak
‘Homosexuals do not want to be considered freaks by people.’ b. Nii...jiu zheyang bei ta qiangzou-le zijii–de qingmeizhuma!26
you exactly this-way BEI he rob-PERF self-POSS childhsweetheart ‘This is exactly how you were robbed of your childhood sweetheart by him!’
20 www.ffoundation.com.tw/modules/newbb/viewtopic.php?topic_id=2445&forum=9
21 An anonymous reviewer questions whether ziji here is a pronoun or an adverb. According to Tsai (2002), ziji as an adverb can mean willingly, by oneself, by nature, or automatically. None of these meanings exists in this example and the copular verb shi indicates ziji here serves as the nominal subject.
22 blog.yam.com/ausky/article/11500862
23 An anonymous reviewer commented that (26b) here and (25c) above can be accommodated under Thompson’s (1973) retained object analysis. That indeed is the case. However, note that these examples are only meant to show that a resumptive pronoun is allowed in the short passive as well. 24 buddhism.lib.ntu.edu.tw/BDLM/cbs/class1/dstext/3_2B92703048.doc
25 www.bwsk.net/yq/s/suwaner/lhzd/002.htm 26 blog.yam.com/ausky/article/11500862
Furthermore, to date, in the relevant literature it has been widely assumed to be a fact that the particle suo is allowed in the long passive but not in the short passive. J. Huang (1999: 451), for example, gives the examples in (29) and (30) as evidence.
(29) *zhexie shiqing bu neng bei ___ suo liaojie. (J. Huang 1999: 451 (53)) these thing not can BEI SUO understand
‘These things cannot be understood.’
(30) *ni zuijin dui ta-de xingwei kongpa hui bei ___ you recent to he-POSS behavior afraid will BEI suo chixiao. (J. Huang 1999: 451 (54))
SUO laugh-at
‘I’m afraid your recent behavior toward him will be laughed at.’
J. Huang points out this contrast between the long passive and the short passive because if the missing agent is lexically filled in (29) and (30), for example with tamen ‘people’ or wairen ‘others’, the sentences will be good. However, the wei-passive, which does allow suo in its short passive, shown in (31), offers a first clue that the above assumption may be too hasty.
(31) Ta bu wei suo dong. he not WEI SUO affect ‘He was not affected.’
While noticing the fact presented in (31), Ting (1998: 347, fn. 9) contends that this ‘does not pose a problem for the proposed analysis that the bei-passive without a lexical logical subject does not license the presence of suo’ and is determined that ‘bei-passives, in contrast to wei-passives, do not co-occur with suo in the absence of the lexical subject in modern Chinese.’ Ting further disregards the second, and much stronger, clue that in Classical Chinese suo does occur in the bei short passive. Ting cites an example from San Guo Yan Yi (Romance of the Three Kingdoms), shown in (32).27
(32) Bi bei suo qin. definitely BEI SUO capture ‘Will definitely be captured.’
As Chiu (1995), J. Huang (1999), and Ting (1998, 2003, 2005) have duly noted, as a remnant of Classical Chinese, the particle suo is limited only to a literary style in modern Chinese. Thus, it is more often found in writing than in speech. Given that suo is a historical remnant and also that in Classical Chinese it does occur in bei short passive, it should be no surprise at all, contrary to the conventional assumption, that the same usage is found in modern Mandarin, but one can also expect its use to be highly restricted. Aside from the numerous instances of bu bei suo dong ‘not be affected’ found in actual use, which is parallel to bu wei suo dong in (31), there are also many other examples in the conventionalized [bu bei suo V] pattern found on the Internet. Two are given in (33a-b). Bu is a stative negator (Lin 2006) and as such can in principle freely occur with the bei..suo construction. Therefore, not surprisingly,
27 It is generally agreed that this novel was written towards the end of the Yuan Dynasty (1279-1368) and the beginning of the Ming Dynasty (1368-1644), thus roughly in the 14th century.
examples of [bei suo V], without the negative item, are also found; (33c-e) are three such examples.
(33) a. Bu bei suo ai shi tongku de zhemo.28 not BEI SUO love be painful DE torture ‘Not to be loved is a painful torture.’ b. Ni nanmian bu bei suo pian.29
you unavoidably not BEI SUO trick ‘Unavoidably you would be tricked.’
c. Bu zhidao laogong hui-bu-hui bei suo youhuo.30 not know husband will-not-will BEI SUO seduce
‘(I) don’t know whether (my) husband will be seduced or not.’ d. Ni neng ai ren, ye yiding hui bei suo ai.31
you can love person also surely will BEI SUO love
‘(If) you can love people, (then) surely you’ll be loved also.’ e. Ren ruo tai zhizhuo, bi jianghui bei suo kun.32
Person if too stubborn surely will BEI SUO trap ‘If one is too stubborn, one will surely be trapped.’
Furthermore, while the grammaticality judgment of J. Huang’s examples in (29) and (30) is corroborated by myself and my half dozen native Taiwan Mandarin informants, an interesting ‘adverb effect’ has been noted in the experiments conducted. English is well-known for its that-trace effect, as any standard textbook on generative syntax attests. However, Culicover (1993a, 1993b) notes that certain fronted adverbials eliminate that-trace effect, as shown in (34); this is known as the ‘adverb effect’.33
(34) a.*Robin met the man whoi Leslie said that ti was the mayor of the city.
b. Robin met the man whoi Leslie said that for all intents and purposes ti
was the mayor of the city. (Culicover, 1993a: 98)
Similarly, it seems that when in the short passive bei and suo are separated by an adverbial PP phrase, acceptability likewise increases. The sentence in (35a) is made up by the author and accepted by more than a dozen native speakers of Taiwan Mandarin, while (35b) is a genuine example found on the net.
28 www.i-part.com.tw/diary/diary_viewpage.php?o=1171640&d=47 29 www.book188.com/files/article/html/18/18961/330666.html 30 forum.frontier.org.tw/women/viewtopic.php?topic=26658&forum=8&8 31 tw.myblog.yahoo.com/juju10022000/article?mid=926&prev=1203&next=570&l=f&fid=29&sc=1 32 www.ck101.com/forums/viewthread.php?tid=1061899
33 Kandybowicz (2006) notes that this effect was discovered by Bresnan (1977: 194) and popularized by Culicover (1993a).
(35) a. Henduo ren jiushi bei *(yi zhe zhong huangyan) suo pian. many people exactly BEI with this kind lie SUO trick ‘Many people were exactly tricked with such lies.’
b. Zhe ge wenti de gengjinyibu taolun keneng bei zai yixie this CLS issue POSS further discussion perhaps BEI at some
jieduan suo bei xuyao.34
stage SUO BEI need
‘Further discussions on this issue are perhaps needed at some stages.’ Once again, in order to prove that suo does appear in the short passive, a few well-formed examples are sufficient. Thus, the previous conventional assumption held by J. Huang (1999) and Ting (1998), among others, is incorrect, as some of the short passives, like some of the long passives, do indeed allow the particle suo.
An anonymous reviewer challenges the legitimacy of the short passive bei..suo data in modern Chinese, arguing that, as claimed by Ting (2003), “suo is licensed in Classical Chinese but not in modern Chinese due to two different grammatical systems in modern and Classical Chinese”. Therefore, “how can the author know that examples in (33) are not produced by imitation of Classical Chinese grammar?” However, note that the same point must apply to long passive [bei NP suo] as well. Ting and J. Huang claim that suo distinguishes between the long passive and the short passive in Modern Chinese, based solely on intuition; the data in (33) and (35) merely show that this is not true. Thus, as an imitation of Classical Chinese grammar or not, suo does not distinguish the long passive and the short passive in modern Chinese.
Finally, J. Huang (1999: 449) observes that place adverbials may occur with the long form, as in (36a), but not the short form, as in (36b). Ting (1998: 350) makes the same observation, shown in (37).
(36) a. Zhangsan bei Lisi zai xuexiao pian-zou le. John BEI Lee at school abduct PERF ‘John was abducted at school by Lee.’
b.*Zhangsan bei zai xuexiao pian-zou le. John BEI at school abduct PERF ‘John was abducted at school.’
(37) *Zhangsan bei zai gongsi li piping-le. John BEI at company inside criticized-PERF ‘John was criticized in the company.’
Nonetheless, once again such a purely introspected observation is contradicted by data from naturally occurring texts. Two counterexamples are given here in (38).
(38) a. Qizi bei zai gonggong changsuo baoli ouda yi ci.35 wife BEI at public place violence batter one time
‘(My) wife has been battered once in a public place.’
34 http://www.tipo.gov.tw/cooperation/cooperation_4_3_2.asp 35 http://hk.epochtimes.com/7/9/10/51266.htm
b. Yi zhi laoshu bei zai jiujing zhong jinpao-le yi nian.36 one CL mouse BEI at alcohol inside immerse-PERF one year
‘A mouse has been immersed in alcohol for a year.’
Our discussion thus far has thus established one simple fact, that is, the short passive does not exist, as bei’s behavior is essentially the same with or without an agent phrase. Bei allows an optional agent phrase and an optional gap, which is unbounded in nature. Contra Her (1991), Ting (1998), and J. Huang (1999), among others, the short passive is quite simply an agentless version of the long passive.
2.4 A-movement for the short passive
Based on the conventional misconception that the short passive strictly requires a local gap, Ting (1998) and J. Huang (1999) propose that, the bei short passive involves A-movement, not A’-movement as in the long passive. J. Huang (1999), adopting Hoshi’s (1991, 1994a, 1994b) analysis of the English get-passive and Japanese ni-passive, assigns the short passive the structure and derivation in (39).
(39) IP NP V' V VP NP V' V NP Lisi bei PRO da-le t
Lisi got hit ....
i i i
Ting (1998: 340) argues for the same analysis for the short passive, as shown in (40b); (40a), as a contrast, shows the analysis proposed for the long passive.
(40) a. [Zhangsani bei [IP OPi [ IP Lisi piping-le ti ]]. (long passive)
John BEI Lee criticize-PERF ‘John was criticized by Lee.’
b. [Zhangsani bei [VP PROi piping-le ti ]]. (short passive)
John BEI criticize-PERF ‘John was criticized.’
There are several important features in the analysis of (39) and (40b). First of all, the short passive bei selects an NP as its subject and a VP as its complement. Secondly, it is assumed that this complement VP is always headed by a passivized transitive verb, e.g., da ‘hit’ in (39) and piping ‘criticize’ in (40b) and its object must
be PRO. Given the standard assumption that a passivized transitive verb has its external theta role assignment and the accusative case assignment absorbed, the PRO object thus must move to VP’s spec position to avoid violating the case filter and to satisfy the EPP. This analysis thus crucially involves NP-movement, an A-movement. However, like the long passive, the short passive also involves a control relation, except that here the matrix subject controls the subject of the VP complement.
It is immediately clear that this A-movement analysis is inadequate for the range of short passive data covered in section 2.3 and leaves unaccounted for the short passives with long-distance gaps, as in (20) and (21), resumptive pronouns, as in (25), and the particle suo, as in (33) and (35). Furthermore, like the long passive, a short passive may also be without a gap, resumptive pronoun, or suo. In other words, the embedded verb, if transitive, may have its object position lexically filled by a full NP, as in (41), and the embedded verb may also be intransitive and forbids an object entirely, as in (42). Again, these are all genuine examples from the Internet.
(41) a. Bei gai-ming Xu Zhituo, Xu Zhian bu nu fan xiao.37 BEI change-name Xu Zhituo Xu Zhian not angry instead laugh ‘Having been “changed-name-to- Xu-Zhituo”, Xu Zhian didn’t
get angry, but laughed instead.’
b. Hai you qita ren bei qiangzou-le qian.38 still have other person BEI rob-PERF money
‘There are other people who have been “robbed-money”.’
c. Bei jiechuan mianju de wo hen wuzhu.39 BEI lift mask DE I very helpless
‘I, who have been “mask-lifted”, am very helpless.’40 d. Minzhong paidui deng-zhe bei da pigu.41
folks line-up wait-PROG BEI spank bottom
‘Folks lined up and were waiting to have their bottoms spanked.’ (42) a. Jintian zhuadao xiao gong juzi danshi bei paozou-le.42
today catch young male orange but BEI escape-PERF ‘Today (I) caught the young male orange (cat), but had him escape
on me.’
b. Po-hou chuangru youzhiyuan, xiaofangrenyuan weibu,
nasty-monkey break-in kindergarten firemen encircle-capture
reng bei taotuo.43 still BEI escape
‘A naughty monkey broke in a kindergarten. The firemen tried to encircle and capture it, but still had it escape on them.’
37 210.244.31.140/news/entertainment/ent8/3889532.shtml 38 www.people.com.cn/BIG5/paper447/802/103759.html 39 www.gaotang.cc/bbs/xiangxi.asp?lbid=8&id=1413
40 Jiechuan mianju ‘lift mask’ also has the idiomatic meaning of exposing one’s lies. 41 www.libertytimes.com.tw/2004/r-s/r-s040123-1.htm
42 www.purlc.idv.tw/book_read.asp?book=b_animal&ssn=2065 43 www.ettoday.com/2004/01/03/124-1566973.htm
c. Women you bei tiao-piao-le.44
we again BEI bounce-check-PERF
‘Again, we were “bounced-check”.’45
d. Mei nian you liu bai ge xuesheng bei tuixue.46
every year have six hundred CL student BEI withdraw-school
‘Every year 600 students are “withdrawn-from-school”.’
The problem presented by examples in (41), precisely, is that the object position of the passivized transitive verb is not occupied by an empty element but a non-pronominal full NP. Thus, no movement is possible, as this full NP remains in situ. Examples in (42) present a different but no less serious problem, that is, there is no object position at all, as all the embedded verbs here are intransitive. Thus, the A-movement account also fails to cover the full range of Chinese long passives.
Finally, given the fact that the short passive and the long passive share essentially the same range of syntactic behavior, the A’-dependency account and the A-movement account can both be rejected for the Mandarin bei construction.
2.5 Passive compound verbs
In the discussion above, all bei passives are syntactically derived, as bei and the embedded verb are both free morphemes. There are also bei-V sequences where the V is no longer a free morpheme in modern Mandarin. Some examples are listed in (43).
(43) a. bei-bu ‘to be arrested’ b. bei-qin ‘to be arrested’
c. bei-lu ‘to be captured’
d. bei-fu ‘to be captured’
e. bei-hai ‘to be killed’
f. bei-po ‘to be forced to…’
g. bei-jian ‘to be raped’ h. bei-fen ‘to be burned’
i. bei-ji ‘to be laughed at’
j. bei-feng ‘to be mocked’
k. bei-ji ‘to be hit’
l. bei-qie ‘to be stolen’
m. bei-duo ‘to be robbed/stolen’
n. bei-dong ‘to be passive’
Reynolds (1995), Ting (1998), J. Huang (1999), among others, have treated such bei-V sequences as lexical items derived by a lexical process. Such a lexical analysis is well-founded, for two reasons. First, as a bound morpheme, the V here no longer stands on its own in modern Mandarin. And, more importantly, these bei-V sequences observe lexical integrity, as the standard tests of separability can attest. However, there is clear evidence that the lexical approach cannot be extended to short passives where
44 www.ettoday.com/2004/01/03/124-1566973.htm
45 The literal meaning of tiao-piao is having a bounced check and its idiomatic reading is having a broken promise.
the V is a free morpheme, as adverbials can freely intervene and lexical integrity cannot be justified. The reader is referred to the references given above for further details. The rest of the paper will focus on the syntactically derived bei passives.
3. Formal analysis of Bei
This section builds on what has been established so far, i.e., bei is a single verbal predicate, and demonstrates that bei is a three-place predicate, not a two-place predicate entailed under the analysis by Ting and J. Huang. Bei will be shown to be precisely the passive counterpart of the verb ba. A formal analysis of bei’s lexical structure and the syntactic structure of the bei passive will be rendered within the Lexical Functional Grammar (LFG).
3.1 Bei as a three-place predicate
It is clear that bei, as a unified verb, requires a clausal complement. What is not clear is what syntactic category this complement is. The core of this issue essentially rests on the grammatical status of the optional agent NP. Under the A’-dependency analysis proposed by Ting (1998) and J. Huang (1999), bei subcategorizes for two syntactic arguments: a subject and an IP complement. The optional agent NP is thus the subject of the embedded IP, as shown in (44). However, Her (1989, 1991) argues instead that bei subcategorizes for three syntactic arguments: subject, object, and a VP complement, as in (45), and more specifically that it is an object-control verb.
(44) Zhangsani bei [IP OPi [ IP (Lisi) piping-le ti]
John BEI Lee criticize-PERF ‘Zhangsan got criticized (by Lisi).’
(45) Zhangsan [VP bei [NP (Lisii)] [VP PROi piping-le e]]47
John BEI Lee criticize-PERF ‘Zhangsan got criticized (by Lisi).’
Ting and J. Huang further argue that bei in (44) is an ECM (exceptional case marking) verb, which assigns accusative case, exceptionally, to the subject of an adjacent IP. Bei is thus like believe in (46). Evidence that Ting and J. Huang have cited in support of the ECM analysis comes from binding, as in (47), and adjacency facts, as in (48).
(46) Mary believed (*sometimes) *heri/herselfi to be a good linguist.
(47) Zhangsani bei *tai/zijii pian-le.
John BEI him/self trick-PERF ‘John got tricked by himself.’
47 An anonymous reviewer questions why the coindexation does not go in a different way, e.g.,
Zhangsan coindexed with PRO and the object of bei with object of the embedded verb. First of all, the
data indicate such relationships are incorrect. Secondly, as a general principle, object control supersedes subject control; thus, the promise-type of verbs is exceptional.
(48) Zhangsani bei (*zuotian) (*xianran) zijii pian-le.
John BEI yesterday obviously self trick-PERF ‘John obviously got tricked by himself (yesterday).’
However, clearly, (47) and (48) are no evidence for the ECM analysis. The ECM is merely a stipulation within the Chomskyan transformational theory to avoid a raising-to-object analysis and has been discredited by empirical arguments (e.g., Culicover and Jackendoff 2005). Rather, data in (47) and (48) present the simple fact that syntactically the agent NP behaves exactly like an object of bei, no more and no less. While believe in (46) does not assign any theta role to its object, it will be argued momentarily that bei does theta-mark the agent NP. Incidentally, bei, under the analysis by Ting and J. Huang, is also not exactly like believe, because of the NOP adjoined to the IP, which intervenes between bei and the agent NP. Thus, J. Huang (1999) needs to stipulate that an adjoined null category does not block ECM, while Ting (1995: 138) proposes to extend Baker’s (1988) theory that abstract case relationship must be expressed at PF to account for it.48
Another observation casts serious doubt on the IP analysis. According to Fu (1994) and Hsieh (2001), sentence adverbials occur as IP adjuncts only, never as VP adjuncts. Thus, if bei indeed requires an IP complement, sentence adverbials should occur, but they do not.
(49) Lisi xianran/dagai/huoxu pian-le Zhangsan. Lee obviously/probably/possibly trick-PERF John ‘Lee has obviously/probably/possibly tricked John.’ (50) *Zhangsan bei Lisi xianran/dagai/huoxu pian-le.
John BEI Lee obviously/probably/possibly trick-PERF ‘John obviously/probably/possibly got tricked (by Lee).’
The contrast between (49) and (50) demonstrates that the post-bei sequence is not an IP. Again according to Fu (1994) and Hsieh (2001), VP adjuncts include temporal adverbs, manner adverbs, and various PPs. Thus, as expected within the VP analysis of (45), the post-agent sequence in the bei construction allows all these.
(51) Wo zuotian zai taibei yi jiachao chenggongde pian-le ta. I yesterday at Taipei with fake-money successfully trick-PERF he ‘Yesterday I successfully tricked him with fake money in Taipei.’ (52) Ta bei wo zuotian zai taibei yi jiachao chenggongde pian-le.
he BEI I yesterday at Taipei with fake-money successfully trick-PERF ‘Yesterday in Taipei he was tricked by me with fake money successfully.’ Given the fact that the post-bei sequence is not an IP but the post-agent sequence is a VP, the only conclusion is that bei requires three syntactic arguments, a subject, an object, and a VP complement, confirming the analysis of (45). The next issue that needs to be addressed is whether the NP object is theta-marked by bei or not. In both
(53a) and (54a), him behaves like the object of the matrix verb; however, it is theta-marked by persuade in (54a) but not theta-marked by believe in (53a).
(53) a. Mary believed him to be the best linguist. b. Mary believed him to be likely the best linguist. (54) a. Mary persuaded him to be the best linguist.
b.*Mary persuaded him to be likely the best linguist.
Note that in the raising, or ECM, construction of (53b), the nonthematic object of believe can be the controller of the nonthematic subject of likely, but the thematic object of the control verb persuade in (54b) cannot. This provides us with a test with the raising verb kanqilai ‘seem’ to determine whether bei is a raising or control verb.
(55) a.*Xianji bei zhongguoren kanqilai qiang-le.49 initial-advantage BEI Chinese seem obtain-PERF b.*Taiwan bei zhongguoren kanqilai qiang-le xianji.
Taiwan BEI Chinese seem obtain-PERF initial-advantage Bei behaves like a control verb in (55). Another clue to the object’s thematic nature is the fact that it can be syntactically empty. A nonthematic argument, e.g., the English expletive it, which by definition bears no semantic role, is only required syntactically. A nonthematic argument is therefore always syntactically overt.
(56) Zhangsan bei e pian-le. John BEI trick-PERF ‘John got tricked.’
The empty object in (56) thus must be thematic. Note that the optionally empty position after bei is not an assumption, but a fact, once it has been established that the long passive and the short passive belong to the same construction. The semantics of bei also confirms its requirement of three event participants. In (56), in spite of the empty object, someone is responsible for the event that happened to Zhangsan, the event being that this someone tricked Zhangsan. Under this analysis, bei is a three-place predicate requiring three theta roles with a meaning that approximates (57):
(57) bei <x y z>, x is responsible for (adversely) affectingy as z describes
We will discuss the syntactic assignment of bei’s thematic roles in section 4. We will now examine bei’s semantic property under the A’-dependency analysis by Ting and J. Huang, repeated in (58) below.
49 Note that the active counterpart is well-formed: (i) Zhongguoren kanqilai qiang-le xianji.
Chinese seem obtain-PERF initial-advantage ‘The Chinese seem to have obtained the initial advantage.’
(58) Zhangsani bei [IP OPi [ IP (Lisi) piping-le ti] John BEI Lee criticize-PERF ‘Zhangsan got criticized (by Lisi).’
According to J. Huang (1999), even though bei c-selects an NP subject and an IP complement, bei is in fact a one-place predicate, requiring only one argument semantically, which is linked to the subject. However, the bei construction involves a complex predicate: bei being the primary predicate and the IP structure the secondary predicate. 50 The IP structure is interpreted as the secondary predicate of the matrix subject due to the coindexing of the NOP and the matrix subject. According to this logic, (58) gets this interpretation: Zhangsan ended up with the property of being an x such that Lisi criticized x.
However, in (58), Lisi is not only the one that criticized Zhangsan, it is also the entity responsible for Zhangsan’s ending up being criticized. Thus, the first drawback with this semantic analysis is that it completely leaves out the entity responsible for the subject ending up with a certain property. A much more serious problem is the fact that this analysis incorrectly predicts the obligatoriness of a null object in the bei construction. This analysis thus fails to account for bei passives that do not host a gap. Many examples are given in section 3; two more examples here should suffice.
(59) a. Gannaidi you bei Yangji jichu liang zhi anda.51 Kennedy then BEI Yankee hit two CL hit
‘Kennedy then suffered two more hits by the Yankees.’ b. Yinggelan xian bei duishou dazhen.52
England first BEI opponent try
‘England suffered a try by its opponent first.’
In (59a), the verb jichu ‘hit’ is transitive and has a lexically filled object liang zhi anda ‘two hits’. The sentence thus has the meaning approximating ‘Kennedy was adversely affected by the Yankees for making two hits’. The verb dazhen in (59b), a VO compound meaning literally ‘to reach goal’, is an intransitive verb meaning to score a touchdown in American football or a try in English football. The sentence thus means approximately ‘England was adversely affected by its opponent in that the latter scored a try’. In both sentences, there is nothing in the embedded clause to be coindexed with the matrix subject. The A’-dependency analysis thus fails to account for such bei passives.
To summarize, here are the steps we have taken in our argumentation. First, we established the verbal status of the unified bei. We then ruled out the analysis where bei subcategorizes for a subject and an IP complement. We confirmed that bei instead requires a subject, an object, and a VP complement. Whether bei is a two-place (raising) predicate or three-place (control) predicate hinges on the object’s thematic status. Given that only thematic arguments can be syntactically unexpressed, the decision in turn can be based on whether bei allows its object to be syntactically unexpressed. The fact that
50 J. Huang (1999) argues that bei’s IP complement turns into a secondary predicate by way of ‘lambda-abstraction’ and is thus no longer an argument.
51 blog.pixnet.net/mlb688/post/3981072
the object of bei is optional thus indicates it is a role-bearing argument. Bei is therefore a three-place predicate.
3.2 Between ba and bei
As mentioned in the beginning of the paper, the part-of-speech of Mandarin ba has also been a contentious issue in Chinese linguistics. However, Bender (2000) and D. Li (2003) both argue, convincingly, that it is a verb, not a preposition. They demonstrate that ba is a three-place predicate that subcategorizes for a subject, an object, and a VP complement. The similarity between her analysis of ba and our analysis of bei is obvious. Thus, between ba and bei, the subject and the object are inverted, as shown in (60). Unlike ba, bei allows the agent NP to be optional.
(60) a. Ta ba juzi bo-le pi. he BA orange peel-PERF peel(n) ‘He peeled the orange.’
b. Juzi bei (ta) bo-le pi. orange BEI he peel-PERF peel(n) ‘The orange was peeled (by him).’
Bender (2000: 127) formally renders the lexical entry of ba in (61) within the grammatical framework of Lexical Functional Grammar (LFG), and the ba’s semantics is informally given in (80). Under this analysis, (60a) has the c-structure and f-structure given in (63a-b).
(61) ba V
(↑ PRED) = ‘ba <(↑ SUBJ)(↑ OBJ)(↑ XCOMP)>’53 (↑ OBJ) = (↑ XCOMP TOPIC)
(62) (↑ SUBJ) is responsible for the fact that (↑ OBJ) turns out as (↑ XCOMP) describes.
(63) Ta ba juzi bo-le pi. he BA orange peel-PERF peel(n) a. c-structure54 IP NP VP Ta V NP VP ba juzi V NP bo-le pi
53 Bender (2000) uses the function COMP. I modify it to be XCOMP, which is more in line with the LFG conventions.
54 Note that LFG c-structures are annotated with functional equations, which link the c-structure and the f-structure. For the sake of simplicity, such annotations are not included.
b. f-structure55
PRED ‘ba <(↑SUBJ) (↑OBJ) (↑XCOMP)>’ OBJ [‘orange’]
SUBJ [‘3sg’] TOPIC
XCOMP PRED ‘peel <(↑SUBJ) (↑OBJ)>’ SUBJ [….]
OBJ [‘peel’]
An important feature of Bender’s account is that ba, unlike other pivotal verbs, does not have its object control the XCOMP’s SUBJ; instead, it controls its TOPIC, indicated by the curved line in the f-structure in (63b). In LFG, TOPIC is grammaticalized discourse function, subject to the Extended Coherence Condition (ECC) (Bresnan and Mchombo 1987: 8):
The extended coherence condition requires that all functions in f-structure be bound. An argument function . . . is bound if it is the argument of a predicator (PRED). An adjunct is bound if it occurs in an f-structure containing a PRED. Finally, a topic or focus is bound whenever it is functionally identified with or anaphorically binds a bound function.
In the f-structure of (63b), the TOPIC in the complement clause, juzi ‘orange’, enters a possessor-possessed relation with OBJ in the local f-structure and thus anaphorically binds the latter, as indicated by the dotted line. This topic analysis of ba’s object also accounts for the unbounded nature of the gap, if any, in the embedded VP complement. Compare (64) and (65).
(64) Ta ba zhe jian shi jiao ren gaosu-le Lisi e. he BA this CL matter ask someone tell-PERF Lee ‘He has asked someone to tell Lee about this matter.’ (65) Zhe jian shi bei ta jiao ren gaosu-le Lisi e.
this CL matter BEI he ask someone tell-PERF Lee
‘This matter has been “asked-someone-to-tell-Lee-about”.’
In (64), the TOPIC of the embedded clause, which is controlled by ba’s object, zhe jian shi ‘this matter’, enters into a long-distance control relation with the (empty) object of the final embedded clause. The ba sentence in (64) finds its bei counterpart in (65), where the matrix subject likewise enters a long-distance control relation with the same object. According to Tsao (1986), a non-controversial topic in Chinese has the following properties.
55 Note that the f-structure information is also drastically simplified. For example, the matrix subject function should have a full set of attribute-value pairs indicating person, number, and PRED. The current ‘3sg’ and ‘orange’ (for the object function) is a ‘short-hand’ convention in the LFG literature.
(66) Topic properties (Tsao 1986: 4)
a. Topic invariably occupies the S-initial position of the first sentence in a topic chain.
b. Topic can optionally be separated from the rest of the sentence by one of the four pause particles: a (ya), na, me, and ba. c. Topic is always definite or generic.
d. Topic is a discourse notion; it may, and often does, extend its semantic domain to more than one sentence.
e. Topic is in control of the pronominalization or deletion of all the coreferential NPs in a topic chain.
f. Topic, except in cases where it is also subject, plays no role in such processes as reflexivization, passivization, and Equi-NP deletion.
Thus, while LFG’s ECC can be taken to be a universal constraint, the possible relations listed in (66) can be seen as Mandarin-specific restrictions on the kind of adjuncts that are allowed to serve for the incorporation of the topic (Bender 2000: 128). According to Tsao (1986) and Bender (2000), the only properties that ba’s object NP does not have are the initial position and pause particles. Ba’s object corresponds to bei’s subject in our analysis, and note that the bei subject shares all six properties listed by Tsao, including the initial position and all pause particles. Additionally, Tsao (1986) finds a close parallelism between pre-subject non-argument topics and the ba-NPs in the relations they enter with the comment clause, i.e., they allow virtually the same range of relations, including location, “relational” adverbials, possessor-possessed relation, and whole-part relation, precluding only time adverbials and class-member relationships. The following examples are Bender’s (2000: 125) adaptation from Tsao (1986: 17-19).
(67) Locative adverbial
Ta ba bilu sheng-le huo. he BA fireplace build PERF fire ‘He built a fire in the fireplace.’
(68) “Relational” adverbial
Ta ba nei jian shi xie-le yi feng baogao. he BA that CL matter write-PERF one CL report. ‘What he did with that matter was write a report about it.’ (69) Possessor-possessed
Ni neng bu neng ba nei ben shu jiang yidian jia? you can not can BA that CL book lower a-little price ‘Can you reduce the price of that book a little?’
(70) Whole-part
Zhangsan ba wu ge pingguo chi-le san ge. John BA five CL apple eat-PERF three CL ‘John ate three of the five apples.’
Note that exactly the same patterns are found in the bei construction, with the subject and object inverted, as shown in (71)-(74) below. To further emphasize the point that there is no significant distinction between the long passive and the short passive, again note that the bei object in all these examples can be left empty.
(71) Locative adverbial
Bilu bei (ta) sheng-le huo. fireplace BEI he build PERF fire
‘What he did to the fireplace was build a fire in it.’ (72) “Relational” adverbial
Nei jian shi bei (ta) xie-le yi feng baogao. that CL matter BEI he write-PERF one CL report. ‘What he did with that matter was write a report about it.’ (73) Possessor-possessed
Nei ben shu bei (ta) jiang-le yidian jia. that CL book BEI he lower-PERF a-little price
‘What he did with the book was reduce its price a bit.’ (74) Whole-part
Wu ge pingguo bei (ta) chi-le san ge. five CL apple BEI he eat-PERF three CL
‘What he did with the five apples was eat three of them.’
I will now give one final piece of evidence for the topic status of ba’s object and its bei counterpart. One interesting feature associated with the Mandarin word formation of V-lai-V-qu ‘V-come-V-go’, meaning to do something repeatedly, is that the compound is detransitivized even if the V is transitive; however, this suppressed internal argument, though barred from surfacing as the object, may appear as the topic as well as in the ba and bei constructions (Chang 2007).
(75) a.*Ta zhengtian jiang-lai-jiang-qu nei jian shi. he whole-day talk-come-talk-go that CL matter b. Nei jian shi, ta zhengtian jiang-lai-jiang-qu. that CL matter he whole-day talk-come-talk-go ‘That matter, he talks about all day long.’
c. Ta zhengtian jiang-lai-jiang-qu de nei jian shi he whole-day talk-come-talk-go DE that CL matter ‘The matter which he talks about all day long.’ (76) a. Ta ba nei jian shi zhengtian jiang-lai-jiang-qu.
he BA that CL matter whole-day talk-come-talk-go
‘What he does with that matter is talk about it all day long.’ b. Nei jian shi bei ta zhengtian jiang-lai-jiang-qu.
that CL matter BEI he whole-day talk-come-talk-go ‘That matter is talked about by him all day long.’
In (75b), the absorbed argument appears as the sentential topic, and it is relativized in (75c). Within LFG, a relativized element also takes on a topic function (e.g., Bresnan 2001: 183; Falk 2001: chp. 6). This topicalized NP is ba’s object, as in (76a), and bei’s subject, as in (76b). Bender’s analysis of ba-NP is thus taken to be correct and the same analysis is extended to bei’s subject.
However, Bender’s analysis of ba is incomplete in that it leaves the subject of the embedded clause unaccounted for. The f-structure in (63b) is repeated below in (77). An XCOMP’s SUBJ, by definition, must be functionally or anaphorically controlled. In (77), it is unclear how SUBJ in the embedded clause is fulfilled. This issue is left for future research in Bender (2000: 129).
(77) Ta ba juzi bo-le pi. he BA orange peel-PERF peel(n)
PRED ‘ba <(↑SUBJ) (↑OBJ) (↑XCOMP)>’ OBJ [‘orange’]
SUBJ [‘3sg’] TOPIC
XCOMP PRED ‘peel <(↑SUBJ) (↑OBJ)>’
SUBJ [….] ? OBJ [‘peel’]
The primary reason for Bender’s hesitation to have ba’s subject, the obvious choice, as the controller in (77) is because she argues, rather convincingly, that the simple transitive verbs in the embedded simplex clause are in fact unmarked passives. These are the core cases of ba. Thus, in (78), the embedded verb chai ‘demolished’ is an unmarked passive, with its agent role suppressed from syntactic assignment and its subject fulfilled by an anaphoric relation with the TOPIC, fangzi ‘house’. The matrix subject thus cannot control the lower subject.
(78) Lısì ba na jian fangzi chai-le.
Lee BA that CL house demolish-PERF ‘Lee demolished that house.’
PRED ‘ba <(↑SUBJ) (↑OBJ) (↑XCOMP)>’ OBJ [‘house’]
SUBJ [‘Lee’] TOPIC
XCOMP PRED ‘demolished <(↑SUBJ)>’ SUBJ [….]
However, even when ba’s subject does not always control the subject of the first embedded predicate, it is linked to the agent role of the embedded predicate, as
Bender (2000: 128) clearly recognizes. Thus, in (78), ba’s subject, Lee, is identified with the passive chai’s ‘demolished’ agent, which is nonetheless suppressed by passivization. This control relation is thus not realized syntactically. Accordingly, in (77), the matrix subject does indeed control the lower subject, which is linked to the agent role, of the active verb bo ‘peel’, as shown in (79).
(79) Ta ba juzi bo-le pi. he BA orange peel-PERF peel(n)
PRED ‘ba <(↑SUBJ) (↑OBJ) (↑XCOMP)>’ OBJ [‘orange’]
SUBJ [‘3sg’] TOPIC
XCOMP PRED ‘peel <(↑SUBJ) (↑OBJ)>’ SUBJ [….]
OBJ [‘peel’]
In fact, a more accurate generalization is that ba’s subject is always linked to the
logical subject, or Ô in LFG’s terminology, of the first embedded predicate. LFG’s
current formalism does provide a way to specify this. The function that links theta roles with grammatical functions (GF) is γ; thus γ(Ô) refers to the GF linked to the
logical subject.56 Thus, ba’s lexical entry in (79) can now be made complete with an additional functional equation, as the last equation in (80).
(80) ba V
(↑ PRED) = ‘ba <(↑ SUBJ)(↑ OBJ)(↑ XCOMP)>’ (↑ OBJ) = (↑ XCOMP TOPIC)
(↑ SUBJ) = (↑ XCOMP γ(Ô))
Given that the object of bei is the passive counterpart of the subject of ba, it also controls the GF linked to the logical subject, or Ô, of the first embedded predicate.
3.3 Bei’s lexical form
The discussion above demonstrate that bei requires three argument roles. Syntactically it subcategorizes for a subject, an object, and an open complement. Also, the matrix subject controls the topic of the embedded complement. Bei’s lexical form is formalized within LFG and given in (81) and its semantics is stated informally in (82). Under this analysis, (60b) has the c-structure and f-structure given in (83).
(81) bei V
(↑ PRED) = ‘bei <(↑SUBJ) (↑OBJ) (↑XCOMP)>’ (↑ SUBJ) = (↑ XCOMP TOPIC)
(↑ OBJ) = (↑ XCOMP γ(Ô))
﹁(↑ OBJ) => (↑ OBJ PRED) = PRO
