approved for publication in Noncanonical Passives, ed. Artemis Alexiadou & Florian Schäfer. John Benjamins. 2013. 1 Variations in Non-Canonical Passives* C.-T. James Huang Harvard University Abstract. This paper shows that non-canonical passives (like English get- and Chinese bei-passives) exhibit a chameleonic character that makes them amenable to a control and/or a raising analysis—depending on context and lexical choice. Such passives are formed by superimposing on the main predicate a higher semi-lexical verb whose meaning may include one or more points in the causative-unaccusative spectrum. Such semi-lexical verbs may differ in their ‘bandwidths’ along the spectrum, exhibiting lexical, contextual and idiolectal micro-variation—accounting for the controversies that have often arisen in the literature. The paper demonstrates the need for fine-grained decomposition in argument structure theory by providing several concrete cases, and ends with a case study of so-called ‘give-passives’ in Mandarin, arguing that they are raising constructions involving an existential use of gei ‘give’ and an implicit affectee which gives rise to their passive-like meanings. Keywords: non-canonical passives, passivization cartography, control-raising controversy, get-passives, bei-passives, gei-passives, Mandarin 1. Introduction 2 Cross-linguistically, there are two main strategies to form a passive sentence: first, by intransitivizing (rather, unaccusativizing) the main verb, with ensuing syntactic consequences, as illustrated by the familiar be-passive in English: (1) Johni was kicked ti by Bill. Second, by superimposing a de-causativized experiential verb on the main predicate which itself remains in active form. The bei-passive in (Mandarin) Chinese involves such a strategy: (2) Zhangsani bei [Lisi ti-le ei liang-xia] Zhangsan bei [Lisi kick-Perf twice] ‘Zhangsan got twice kicked by Lisi.’ A third strategy makes a passive by a combination of the first two strategies. This is illustrated by the English get-passive. (3) Johni got [ei arrested ti by the police]. We shall be concerned with the ‘non-canonical’ passives as illustrated in (2)-(3). The analysis of such passives has been the subject of long controversies. Based on 3 investigation of extensive earlier accounts, Huang (1999) and Huang, Li and Li (2009, henceforth HLL) concluded that two types of bei passive should be distinguished—the long and the short passive—depending on whether they contain the agent phrase or not. The long passive is analyzed as involving a semi-lexical verb bei complemented by a null operator construction predicated on the subject, as depicted in (4): (4) Zhangsani bei [ OPi [ Lisi kicked ti twice ]] predication movement In the short passive, the semi-lexical verb bei takes a VP complement with an A-moved PRO controlled by the matrix subject, as in (5): (5) Zhangsani bei [vP PRO kicked ti twice ] control movement According to the analysis, the long passive thus has a syntax akin to that of a tough construction. With bei analyzed as an experiential verb that selects an ‘affectee’ subject, (4) has the meaning ‘Zhangsan ended up as an affectee x, such that Lisi kicked x twice.’ The short passive has a syntax somewhat similar to that of ‘John hopes PRO to be forgiven’, and (5) has the meaning ‘Zhangsan ended up as an affectee x, such that x was kicked twice’. A central property of the analysis represented by (4)-(5) is that each case exhibits two dependencies and, crucially, that the upper dependency is one of predication or control, but not one of movement. This property distinguishes itself from most previous analyses.1 4 The proper analysis of the English get-passive has also been a matter of controversy in the recent past (involving authors such as Alexiadou 2005, Butler and Tsoulas 2006, Fox and Grodzinsky 1998, Haegeman 1985, Hoshi 1994a, etc.). The debates concern whether (3) should be analyzed in terms of raising as in (6) or control as in (7): (6) Johni got [ti raising arrested ti by the police]. (Raising analysis) (7) Johni got [PROi arrested ti by the police]. control (Control analysis) In view of the recent discussions we would like to reconsider the issue of whether the non-canonical passives should be analyzed in terms of control/predication or in terms of raising. After examining the various recent discussions, it will be our claim that both analyses are in fact possible, for certain types of sentences, and that this ‘chameleon character’ of the bei and get passives follows from a quasi-cartographic approach to argument structure. In Section 2 we proceed to this point and explain the proposal in some detail, showing how it resolves some aspects of the current controversy. In Section 3 we illustrate how different verbs with similar meanings are employed in different languages and dialects in the expression of non-canonical passives, and how they may differ in the ‘bandwidths’ of their meanings along the causative-unaccusative continuum. In Section 4 we focus on a case study of gei ‘give’, which in Northern Mandarin has an additional passive-like meaning. We take up the discussion in Shen and Sybesma (2010) and propose that the passive-like meaning comes from the unaccusative meaning 5 ‘happen’ of gei (near the lower end of the causative-unaccusative continuum). Section 5 provides a summary and concludes the paper. 2. On the chameleonic character of the get and bei passives Alexiadou (2005) revived the raising analysis of English get passives drawing in part on arguments by Haegeman (1985), and Fox & Grodzinsky (1998). One sort of support comes from the alternation of the passive with the causative: (8) a. Bill got John arrested by the police. b. John got arrested by the police. [=3] This clearly parallels the familiar causative-unaccusative alternation of John broke the window and The window broke. This justifies a raising analysis as depicted in (6) in parallel to the received analysis of unaccusative break. Another type of support for raising comes from the putative absence of an implicit agent in the get passive. The relevant observation (after Fox & Grodzinsky 1998) is that, whereas an implicit agent licenses a purpose clause in a be-passive as in (9a), a getpassive does not seem to have an implicit agent to license the clause in (9b): (9) a. The ship was sunk [PRO to collect insurance money]. b. *The ship got sunk [PRO to collect insurance money]. 6 A similar contrast obtains below, on the interpretation that the adverbial refers to the purpose of the implicit agent. (10) a. The book was torn on purpose. b. *The book got torn on purpose. The putative absence of an implicit agent is then taken as suggesting that a get-passive is actually an adjectival passive, analogous to get excited and get angry, which involves no implicit agent and is a clear case of raising. A third major source of support points to the possibility of idiom-chunk subjects which, according to standard assumptions, must have been raised from below get: (11) a. (?) In the end, advantage always gets taken of John. b. (?) Tabs got kept on Jane Fonda by the FBI. The raising analysis, however, is at odds with Hoshi’s (1994a, 1994b) control analysis of English get-passives and Japanese ni-passives, the control/predication analysis of Chinese bei-passives of Huang (1999) and HLL (2009), and some recent arguments from Butler and Tsoulas (2006). Interestingly, one argument in favor of control (produced by Hoshi) capitalizes on the (relative) ill-formedness of idiom-chunk subjects when compared with canonical be-passives (and a parallel contrast between ni-passive and niyotte-passive in Japanese):2 7 (12) a. Considerable advantage was taken of John. b. Tabs were kept on Jane Fonda by the FBI. Whereas the raising proponents judge (11a-b) to be good, the control proponents consider them bad. While this may reflect dialectal or idiolectal differences, it seems also true that for those who accept (11), (12a-b) still sound better. Another argument for control/predication also concerns sentences with purpose clauses and intentional adverbs, like (9)-(10), but the focus is on a different aspect of their meanings. As first observed by Fiengo and Lasnik (1974), such adverbials may refer to the intention or purpose of the surface subject of get, but not the subject of be, suggesting that the subject of get is base-generated with its independent theta-role:3 (13) a. *John was burned deliberately. b. John got burned deliberately. (14) a. *The pedestrian was hit [PRO to collect insurance money]. b. The pedestrian got hit [PRO to collect insurance money]. Additional arguments for the non-raising analysis come from long-distance passivization and cases involving the relative pronoun suo: (15) zhe-feng xini bei wo jiao ta pai Lisi qing ta meimei ji-zou le. this letteri bei I ask him send Lisi request his sister mail Perf This letteri bei [I asked him to dispatch Lisi to request his sister to mail ti] 8 (16) zhe-ge jieshi bu neng bei ta suo liaojie. This explanation not can bei he SUO understand ‘This explanation could not be understood by him.’ As argued in Huang (1999) these strongly support A-bar movement followed by predication, as depicted in (4), ruling out raising as a case of ‘improper movement’. There are other arguments put forth in the literature for one or another’s favorite analysis, but the above review suffices to illustrate the problem that facts exist that seem to jointly support two conflicting analyses, depending on what facts are considered and on one’s judgments about the grammaticality and meaning of certain sentences. In the next section, we shall advance the view that these facts do not lead to contradictory conclusions; they simply reflect the chameleonic character of the non-canonical passives. Both the control and the raising analysis are possible, depending on the context, the scenario, and lexical choice. 3. The fine structure of the non-canonical passive By the chameleonic character we mean that verbs like English get, Chinese bei, Japanese rare, etc., may behave sometimes as raising verbs and at other times as control verbs, and in some cases a given sentence may be ambiguous.4 9 For example, when get occurs with idiom-chunk subjects (for the speakers who accept them), it is being used as a raising verb. The subject is in a non-thematic position in such cases. But when it occurs with a purpose clause or an intentional adverb predicating on the subject, get is being used as a control verb, taking a thematic (experiencer or affectee) subject. There is no conflict in these situations, because a sentence with an idiom-chunk subject, being non-referential, cannot take a subjectoriented adverb anyway. The same can be said of Chinese bei sentences—both the agentive long passive and agentless short passive. For example, chui niu lit. ‘blow a cow’ is a verb-object idiomatic phrase meaning ‘to bluff’ and the object may appear as the subject of a bei passive: (17) niu dou bei (ta yi-ge ren) chui guang le cow all bei he one-CL person blow finished Perf Literal: All the cows got blown away clean (by him single-handedly). Idiomatic: All the bluffing was done (by him single-handedly). The idiomatic meaning would require a raising analysis under standard considerations, but inclusion of any intentional adverb would force a literal meaning, as expected.5 In such cases, as in the event of long-distance passivization or a passive with suo, the control/predication analysis is required. Note crucially that idiom chunks like niu do not occur with long-distance or suo-passives, either, again as expected. The situation with Japanese rare is similar. As reported in Hoshi (1994a, 1994b), Japanese passives may involve idiom-chunks as subjects (and in such cases the agent 10 phrase must be realized by a PP headed by niyotte). In the context of an intentional adverb, a control analysis is required, and the agent (if present) will take a dative case.6 The view being advanced here implies that many passive sentences would be analytically ambiguous. In particular, consider a simple passive that contains no idiomchunk subject or any subject-oriented adverbial, or one in Chinese that involves no longdistance dependency or a dependency with suo, or one in Japanese whose agent is not headed by niyotte. What would be the proper analysis of such non-canonical passives? Simple logic says that they may be analyzed in either way—and this is indeed the claim we shall maintain. The source of the ambiguity is that the passive verb may or may not select a (thematic) subject. This would be the idealized ‘neutral case’ without consideration of the context or scenario being described by a given sentence. In each actual case, however, the selection of a given verb with or without a thematic subject may be made depending on the given scenario, e.g., whether a volitional subject is involved even if no overt intentional adverb is provided. This idealized ‘neutral case’ is also subject to speaker variations and lexical choice. For example, speakers vary with respect to the degree of acceptability about sentences with idiom-chunk subjects. We take this to be an example of idiolectal micro-variation—in the treatment of a given passive verb as being preferably a dyadic or a monadic predicate. Finally, not all lexical choices are equal. As we shall see below, in each language several verbs may be used for the expression of a non-canonical passive, but while a given verb may preferably be used as a dyadic predicate, another may allow a dyadic and a monadic use equally, and still another may only have a monadic use.7 11 Now a new question arises: Why should the non-canonical passive exhibit such chameleonic characters? (And why doesn’t the canonical be-passive do so?) We submit that the answer lies in what might be called the ‘passivization cartography’ in the spirit of Rizzi (1997), Cinque (1999), and much other work: (18) THE PASSIVIZATION CARTOGRAPHY: Non-canonical passives are formed by superimposing on the main predicate a higher semi-lexical verb whose meaning may include one or more points in the causative-unaccusative spectrum. (19) THE CAUSATIVE-UNACCUSATIVE CONTINUUM: cause > let > witness > undergo > be affected by > become > exist > be (20) Verbs may differ in having different ‘bandwidths’ along the spectrum. The reason why English get, Chinese bei and Japanese rare exhibit both control and raising behaviors is that they all occupy more than one point in the continuum in (19), points paraphrased as undergo (with an experiencer subject), and be affected by (with an affectee subject), and become (unaccusative, with no thematic subject). In addition, English get also has the causative reading, whereas bei and rare do not. Speakers may differ with respect to the finer details on this description, or on their preferences for any of the points identified. Note that the canonical be-passive can now be seen as part of the continuum, as a construction formed by superimposing the verb be on the main passive predicate. The 12 reason why the be-passive does not exhibit chameleonic character is simple: be means be at the rightmost end of the continuum. The chameleonic character of the non-canonical passive means that in addition to the passive voice, it may also express (possibly a number of) what Chvany (1993) calls ‘bystander voice’ readings.8 As Chvany shows, the ‘bystander voice’ (neither purely active nor entirely passive) is itself a continuum. Depending on the superimposing verb, the meaning of a sentence in bystander voice ranges from ‘mediated causation’ to permission, to ‘more or less voluntary observation’. Some examples follow: (21) Bystander voice continuum (Chvany 1993): a. HAVE: The general had the deserter shot/be shot/get shot. b. GET: Ivan purposely got the spy killed/to be killed. c. GET: Ivan’s carelessness got the spy killed. (*to be killed) d. LET: Mary let the child be scolded. e. SEE: The soldier saw his buddy (be(ing)) shot. The sense of causation decreases from (21a) to (21e). In case the object is missing or coindexed with the subject, a sense of passivity arises, as in a get-passive like John got killed or John had himself arrested. Although the last case with see is not commonly used as a passive in English, a ‘see-passive’ was quite common in Archaic Chinese, as is in Modern French. Famous examples from the Classical texts include: jian xiao ‘see laugh’ (=get laughed at); jian xin ‘see trust’ (=get trusted), jian yong ‘see employ’ (=get 13 employed); jian yi ‘see suspect’ (=get suspected), etc. An example of the see-passive in French is given below (Alain Peyraube, p.c.): (22) Il s’est vu embarqué à la fin de la manifestation He see-pronominal taken-away at the end of demonstration ‘He was taken away at the end of the demonstration.’ In addition to English have and get, the famous French faire also participates in expressions that range from varying degrees of causative to passive readings (see Labelle, this volume, inter alia). And it has also been demonstrated that, in German, the various non-canonical passive forms exhibit a continuum of the ‘strengths’ of passivity, with overlapping ‘bandwidths’: the kriegen passive (also possible with the auxiliaries bekommen or erhalten; see Bader & Häussler, this volume and Lenz, this volume), the sich lassen passive (Pitteroff 2011), and the haben passive (see Businger, this volume and Gese, this volume). And in Chinese, in addition to bei in Mandarin and see in Archaic, other passive and passive-like verbs may include jiao ‘cause, have’, rang ‘let’, gei ‘give’ in Mandarin, and other items in other dialects. Outside of passives, it is also well known that certain verbs fluctuate between raising and control-like readings. In English, familiar examples include begin, continue, threaten, promise. Both sentences in (23) are ambiguous between a raising and control construal: (23) a. John promises to be annoying. b. John threatens to be annoying. 14 In Chinese, the following sentence with kending also exhibits a control-raising ambiguity:9 (24) Zhangsan kending [e mingtian hui lai] Zhangsan certain tomorrow will come a. Zhangsan was certain that [e] will come tomorrow. [e = pro, control-like] b. Zhangsan is certain to come tomorrow. [e = trace, raising] But the same is not true of the near synonym queding ‘be certain’. Another fluctuating verb is kongpa ‘be afraid’, but not the near synonyms haipa ‘be afraid’ and pa ‘be afraid’ (with the last one exhibiting some idiolectal variations): (25) a. wo kongpa jintian hui xia yu. I fear today will fall rain ‘I am afraid that it will rain today.’ b. jintian kongpa hui xia yu. today afraid will fall rain It will probably rain today. 15 Summarizing, we have seen that evidence exists for both the raising and the control analysis for non-canonical passives, and that both strategies should indeed be available given the chameleonic nature of the superimposing semi-lexical predicate, which may occupy one or more points on the causative-unaccusative continuum. We see that the semi-lexical predicates may vary with respect to their bandwidths on this continuum, and that speakers vary in the exact details of the lexical variations. This state of affairs shows that a fine-grained lexical decomposition of the superimposing predicate, perhaps in terms of a cartographical approach to argument structure, allows us to see the differences. Incidentally, this state of affairs should not be surprising from what we otherwise know from current views of synchronic variations and diachronic changes. The microvariations are to a great extent the results of micro-parametric changes over time— changes which may proceed at different rates for different lexical items and for different speakers. In the remainder of this paper we would like to address an aspect of lexical choice and turn to a case study of gei ‘give’ as a passive verb (the ‘give-passive’) in Mandarin, again demonstrating the degree of lexical variation on the causative-unaccusative continuum. 4. On lexical choice and the syntax of the so-called ‘give-passive’ in Mandarin Chinese dialects differ with respect to the set of de-causativized semi-lexical predicates they use for the expression of passives. In Southern Min, the verb hoo retains a full spectrum of the meanings on the continuum: give, cause, let, undergo, and a range of 16 direct and indirect passive meanings. A similar situation is true of Cantonese and several other southern dialects.10 In Mandarin, the primary passive verb is bei, which is historically related to the causative verb bei ‘to cover’ through stages of grammaticalization (Zhang 1994) and which, in contrast to hoo ‘give’ in Min, has a pretty narrow bandwidth, used only to express direct or indirect passive voice. For the higher regions of the bystander voice, Mandarin uses rang (or jiao), with weak causative ‘have’ or permissive ‘let’ meanings. The following sentences with rang do not convey a sense of passivity:11 (26) Zhangsan rang nüer xue xiaotiqin. Zhangsan let daughter study violin ‘Zhangsan had his daughter study violin.’ (27) Zhangsan rang Lisi ti-le liang xia. Zhangsan let Lisi kick-Perf two time ‘Zhangsan let Lisi kick him twice.’ (28) Zhangsan rang Lisi pao-zou-le. Zhangsan let Lisi run-away-Perf ‘Zhangsan had/let Lisi run-away. A strong sense of passivity arises with (27-28) if the lower predicate is further embedded under gei ‘give’: 17 (27’) Zhangsan rang Lisi gei ti-le liang xia. Zhangsan let Lisi give kick-Perf two time ‘Zhangsan got kicked by Lisi twice.’ (28’) Zhangsan rang Lisi gei pao-zou-le. Zhangsan let Lisi give run-away-Perf ‘Zhangsan had Lisi run away on him. As indicated, (27’) is translatable as a direct passive, and (28’) as an indirect (adversative) passive. The addition of gei to (26) does not produce the same effects, however; in fact, it renders (26) ungrammatical: (26’) *Zhangsan rang nüer gei xue-le xiaotiqin. Zhangsan let daughter give study-perf violin This state of affairs is puzzling. In the remainder of this paper we shall be concerned with the syntax of the ‘give-passive’. It has long been observed that the standard short passive with bei can often be paraphrased with gei: (29) Zhangsan bei/gei daibu le. Zhangsan bei/gei arrest Perf ‘Zhangsan got arrested.’ 18 For this reason gei is sometimes taken to be a passive verb analogous to bei. But this idea cannot go too far, since their similarities cease as soon as other facts are considered. In an important recent study of the construction, Shen and Sybesma (2010) pointed out further major properties of the gei-VP. (Some observations were made earlier in Tang 2002.) First, gei-VP may occur with intransitives but bei-VP is impossible: (30) xiaogou gei/*bei bing le. dog gei/*bei sick Asp ‘The little dog got sick.’ (31) xiao-niao gei/*bei fei-le. little-bird gei/*bei fly-Perf ‘The little bird flew away.’ (32) xiao-shu gei/*bei zhang-wai le little-tree gei/*bei grow-crooked ‘The little tree grew crooked.’ Secondly, gei-VP may occur in a ba-construction, under the pre-verbal ba-object, but beiVP may not: (33) ta ba yu dou gei/*bei chi-le. he BA fish all gei/*bei eat-perf ‘He ate all the fish.’ 19 This contrast is in fact more general: gei-VP may occur under jiao and rang (‘let’ as in (26-28)), or under bei, but bei-VP may not: (34) chuangzi rang Lisi gei/*bei dapo-le. window let Lisi give/*bei break ‘The window got broken by Lisi.’ (35) chuangzi jiao Lisi gei/*bei dapo-le. window let Lisi give/*bei break ‘The window got broken by Lisi.’ (36) chuangzi bei Lisi gei/*bei dapo-le. window bei Lisi give/*bei break ‘The window got broken by Lisi.’ And although gei-VP may be embedded under a bei-VP (as in (36)), the latter may not be embedded under the former: (37) *chuangzi gei Lisi bei dapo-le. window give Lisi bei break-Perf ‘The window got broken by Lisi.’ 20 A third important property of the gei-VP is that it cannot take an unergative VP as its complement. An unaccusative is fine as in (30-32), but an unergative is bad under gei:12 (38) a. *ta gei ku le. he got cried perf b. *ta gei xiuxi le. he got rested perf We should add that a transitive full VP is also bad under gei, unless its object has been fronted under passivization as in (34-36): (39) a. *ta gei xue xiaotiqin le. he got study violin perf b. *ta gei dapo-le piqiu le. he got break-perf ball perf Finally, Shen and Sybesma point out that while the addition of gei to a sentence does not affect its independent grammaticality, it signals the existence of an ‘external force’ whose identity is somewhat slippery and hard to grasp. 21 (40) xiaoniao feizou le. little-bird flew-away Perf ‘The little bird flew away.’ (41) xiaoniao gei feizou le. little-bird gei flew-away Perf (Some ‘external force’ is involved.) In addition to the primary meaning that the little bird flew away, (41) says that this event is surprising or undesirable to someone, but this affectee is only implicit but not overtly expressed. Summarizing, then, the gei-VP construction exhibits the following four properties: (a) it may occur with an intransitive predicate, (b) it may not occur with an unergative predicate, (c) it may occur below other semi-lexical verbs (jiao, rang, ba, bei), but not above them, (d) its occurrence licenses an implicit ‘external force’. Two questions arise: What is the proper analysis of the gei-VP, and what explains these properties? We propose that gei is a semi-lexical unaccusative verb with the meaning of existential ‘happen’.13 Thus we claim that the so-called ‘give-passive’ is not a passive, but an existential, raising construction that happens to convey a passive-like reading. In particular, the verb gei ‘give’ has two senses, each occurring at one extreme of the causative-unaccusative continuum: the former being the 3-place predicate of giving (x causing y to have z), and the latter being an unaccusative, existential predicate meaning 22 ‘happen’. For some speakers, gei also has the additional bystander voice reading of ‘let’ or ‘sustain/undergo’. We take the one-place unaccusative verb to be derived from subject-suppression from a two-place bystander verb, as in (42a)-(42b). When subject raising takes place, we have a gei-VP sentence as in (42c): (42) a. kanshou gei [fanren pao-le]. guard sustain prisoner run-away ‘The guard had the prisoner running away.’ (2-place bystander) b. [e] gei [fanren pao-le] happen prisoner run-away ‘It happened that the prisoner ran away. (subject suppressed) c. fanreni gei [ ti pao-le] prisoner happen run-away ‘It happened that the prisoner ran away.’ (raising) The situation is not unlike the alternation between two-place and raising certain: (43) a. John is certain that Bill will run away. b. It is certain that Bill will run away. c. Bill is certain to run away. 23 That a ‘give’ verb can have an existential ‘happen’ reading is quite natural in view of some cross-linguistic evidence. One example comes from the informal expression ‘What gives?’ in English, which basically means ‘What’s happening?’ or, e.g., ‘What’s the matter so you are so late?’ (See Joseph 2000 for some remarks on its origin.) And in German, of course, geben is also used as an existential verb: (44) Es gibt eine Fliege in der Suppe. There exists a fly in the soup. (45) Es gibt einen Fehler in Ihrer Argumentation. There exists a flaw in your argument. Now let’s see how the observed properties of the gei-VP construction may fall out under the assumption that gei is an existential raising verb similar to ‘happen’. First, gei may occur with an intransitive VP (while bei cannot), as shown in (30)-(32). This is so because, as a raising verb, gei takes a clausal complement whose subject raises (regardless of whether there is an object), whereas bei, as a passive verb, takes a transitive clause whose object is subject to movement (as null operator or as PRO). (Cf. John happened to arrive on time vs. *John got arrived on time.) Secondly, the complement VP of gei cannot be unergative—either an intransitive as in (38) with ku ‘cry’, xiuxi ‘rest’ or a full transitive VP as in (39) with ‘study violin’ or ‘break the ball’. It can be an unaccusative as in (30)-(32) with ‘fly away’, etc., or a transitive VP with an object gap as in (29) and (33)-(36). Note that in these latter cases, 24 the transitive VP is understood as being predicated on the missing or fronted object, thus has the semantics of a passivized—viz. unaccusativized—VP. So this reduces to the generalization that gei may take an unaccusative but not a pure unergative VP complement. We take this generalization to be a reflection of semantic selection of the unaccusative gei. There is a well known selective relation between semi-lexical (or light) verbs and their complements: e.g., DO selects an action, CAUSE an action or event, BECOME a state or property, etc., so it is most natural that an unaccusative semi-lexical verb gei, with the light verb meaning OCCUR, takes an unaccusative, existence-denoting VP as its complement.14 This explanation also accounts for the third property observed: that gei can be embedded under jiao, rang, ba or bei, but not the other way around, as partially shown in (27’)-(28’), (33)-(36) as opposed to (37).15 This is because, among the light verbs that occur in the causative-unaccusative or bystander continuum, CAUSE, LET, AFFECT, UNDERGO (the basic meanings of jiao, rang, ba, bei respectively) occupy higher positions than HAPPEN and EXIST (the meanings of gei). Finally, what about the fact that a gei-VP construction gives rise to the existence of an ‘external force’ whose identity is somewhat slippery? We claim that this external force is simply an implicit Affectee, the argument that loses its argument position when subject suppression takes place but that still exists as a haunting phantom. Thus, consider the paradigm in (42) again. Given a context like (42a), we know that the subject kanshou ‘guard’ is the Affectee. If the subject is suppressed (so affective gei ‘sustain’ becomes unaccusative gei ‘happen’), with consequent raising we get (42c), which can be understood implicitly as meaning that it happened to the guard that the prisoner ran away. 25 Without any context, the implicit affectee could be the speaker or someone else whose identity is ‘somewhat slippery’. Again the situation is not unlike the paradigm in (43). If a context like (43a) is known, then (43b-c) convey a mental state implicitly true of John; otherwise the speaker is the default implicit experiencer.16 In this section, we have seen examples that demonstrate the claim that languages differ in the lexical choice of a superimposing light predicate for the making of a noncanonical passive, and that the same verbs may have different bandwidths as used in different languages or dialects. Thus while TSM and other southern Chinese dialects use hoo or its cognates,17 the typical Mandarin passive verb is bei. And while hoo possesses a full spectrum of lexical meanings in the causative-unaccusative continuum, its Mandarin cognate gei has primarily two uses, the ditransitive ‘give’ and the existential ‘happen’. A similar situation obtains for German geben. We show that the so-called ‘gei-passive’ is not a passive per se, but an existential raising construction which, by virtue of the existence of an implicit affectee, conveys passive or passive-like meanings. 5. Summary and conclusion This paper has shown that all non-canonical passives are not equal. Although their chameleon-like character has led researchers to opposing views concerning their proper analysis in the past, we have seen that both the raising and control analyses may be appropriate depending on specific circumstances. This state of affairs arises because there is a continuum between causativity and unaccusativity, and that languages and 26 dialects differ in the choices of the semi-lexical predicates for the expression of passive and passive-like meanings. As we have seen, a fine-grained decompositional (or cartographic) approach to the meanings of the superimposing light predicates allows us to see the differences among various forms and their meanings. 27 References Alexiadou, Artemis. 2005. A note on non-canonical passives: the case of the get-passive. In Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz, & Jan Koster (eds), 13–21. Berlin: Mouton de Gruyter. Alexiadou, Artemis. 2012. Non-canonical passives revisited: parameters of non-active voice, to appear in Linguistics. Butler, Jonny & Tsoulas, George. 2006. Get-passives, raising, and control. Cheng, Lisa L.-S., Huang, C.-T. James, Li, Y.-H. Audrey & Tang, C.-C. Jane. 1999. Hoo, hoo, hoo: syntax of the causative, dative and passive constructions in Taiwanese. In Contemporary studies on the Min dialects [Journal of Chinese Linguistics monograph series 14], Pang-Hsin Ting (ed), 146–203. Berkeley: Project on Linguistic Analysis, University of California, Berkeley. Cheng, Lisa, Huang, C.-T. James, Li, Y.-H. Audrey & Tang, C.-C. Jane. 1993. Three ways to get passive. Ms., UC Irvine, USC and Academia Sinica. Chvany, Catherine. 1993. Bystander voice in English: a generalization masked in some versions of theta theory. In Causatives and Transitivity, Bernard Comrie and Maria Polinsky (eds), 241-252. Amsterdam: John Benjamins. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. 28 Collins, Chris. 2005. A smuggling approach to the passive in English. Syntax 8: 81-120. Collins, Peter. 1996. Get passives in English. Word Englishes 15: 43-56. Feng, Shengli. 1995. The passive construction in Chinese. Studies in Chinese Linguistics 1: 1-28. Fox, Danny & Yosef Grodzinsky. 1998. Children’s passive: a view from the by-phrase. Linguistic Inquiry 29: 311–332. Haegeman, Liliane. 1985. The get-passive and Burzio’s generalization. Lingua 66: 53– 77. Hoshi, Hiroto. 1994a. Passive, Causative, and Light Verbs: a Study on Theta Role Assignment. PhD dissertation, University of Connecticut. Hoshi, Hiroto. 1994b. Theta-role assignment, passivization, and excorporation. Journal of East Asian Linguistics 3: 147-173. Huang, C.-T. James. 1982. Logical Relations in Chinese and the Theory of Grammar. PhD dissertation, M.I.T. Huang, C.-T. James. 1999. Chinese passives in comparative perspective. Tsing Hua Journal of Chinese Studies 29: 423–509. Huang, C.-T. James. 2011. On bei xiaokang and the putative birth of a new passive construction. Invited speech at the 19th annual conference of the International Association of Chinese Linguistics, at Nankai University, Tianjin, June 2011. Huang, C.-T. James, Li, Y.-H. Audrey & Li, Yafei. 2009. The Syntax of Chinese. Cambridge: Cambridge University Press. Joseph, Brian. 2000. What gives with what gives? Contribution to Jorge Hankamer WebFest. 29 Lasnik, Howard & Fiengo, Robert. 1974. Complement object deletion. Linguistic Inquiry 5: 535-571. Lien, Chinfa. 2008. Special types of passive and causative constructions in TSM. In Chinese Linguistics in Leipzig, Redouane Djamouri, Barbara Meisterenst and Rint Sybesma (eds), 223-237. Paris: CRLAO, EHESS. Liu, Na Kristy. 2011. Where is the affectee in an adversative passive? Ms., Beijing Language and Culture University. Liu, Na Kristy. 2012. On the Syntax, Semantics and Typology of Non-Canonical Passives. PhD dissertation, Beijing Language and Culture University. Matthews, Stephen and Virginia Yip. 2008. Passive, unaccusative and pretransitive constructions in Chaozhou. Chinese Linguistics in Leipzig, Redouane Djamouri, Barbara Meisterenst & Rint Sybesma (eds), 163-174. Paris: CRLAO, EHESS. Matthews, Stephen, Huiling Xu and Virginia Yip. 2005. Passive and unaccusative in the Jieyang dialect of Chaozhou. Journal of East Asian Linguistics 14: 267-298. Pitteroff, Marcel. 2011. On the properties of German sich-lassen middles. Talk presented at the AG 9 - Non-Canonical Passives, 33. Jahrestagung der DGfS, Georg-AugustUniversität, Göttingen. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar, Liliane Haegeman (ed), 281-337. Dordrecht: Kluwer. Reed, Lisa. 2011. Get-passives. The Linguistic Review 28: 41–78. Shen, Yang and Sybesma, Rint. 2010. Jufa jiegou biaoji ‘gei’ yu dongci jiegou de yansheng guanxi [On the marker gei and its relation to the derivation of VP structure]. Zhongguo Yuwen 2010(3): 222-237. 30 Tang, Sze-Wing. 2002. Hanyu de gei shi bu shi beidong biaoji? [Is Chinese gei a passive marker?]. Ms. in Chinese, Chinese University of Hong Kong. Tang, Sze-Wing. 2003. Hanyu Fangyan Yufa de Canshu Lilun. [A parametric theory of Chinese dialectal grammar]. Beijing: Peking University Press. Tang, Sze-Wing. 2006. A minimalist view on the syntax of BECOME. In Minimalist views on language design, Changguk Yim (ed), 301-311. Seoul: Hankook Publishing Company and the Korean Generative Grammar Circle. Ting, Jen. 1998. Deriving the bei-construction in Mandarin Chinese. Journal of East Asian Linguistics 7: 319–354. Toyota, Junichi. 2008. Diachronic Change in the English Passive. London: Palgrave MacMillan. Zhang, Hongming. 1994. The grammaticalization of bei in Chinese. Chinese Languages and Linguistics 2: 321-360. 31 Notes *Acknowledgments. In addition to the Göttingen workshop in February 2011, parts of this paper have been presented on other occasions, including colloquia at National Sun Yat-Sen University (Taiwan), the Chinese University of Hong Kong, MIT (Ling Lunch), and presentations at the Harvard-Australia Workshop on Language, Learning and Logic hosted by Macquarie University, Sydney, and the 7th Conference of the European Association of Chinese Linguistics (EACL-6) hosted by Universita ca Forscari, Venice. For very helpful comments on earlier versions I am indebted to Artemis Alexiadou, Gluglielmo Cinque, Stephen Crain, Yang Gu, Jeremy Hartman, Chuansheng He, Toru Ishii, Ed Keenan, Cornelius Kubler, Thomas Lee, Henk van Riemsdijk, Luigi Rizzi, Florian Schäfer, Yang Shen, Shu-Ing Shyu, Dylan Tsai, Coppe van Urk, Ting-Chi Wei. The final revision has benefited particularly from extensive discussions with Na Kristy Liu and from generous suggestions by an anonymous reviewer and the editors, which have helped strengthen the paper in both content and presentation. 1 The analysis of the long passive extended Feng (1995) and addressed certain concerns arising therefrom. The distinction between the long and short passives is also made in Ting (1998). The PRO-movement analysis of the short passive adopts the proposal of Cheng, et al (1993), which in turn is an adaptation of Hoshi’s (1994a, b) analysis of English get-passives. An early argument that the short passive is not derived from the long passive by syntactic agent-deletion was made in Huang (1982). 2 Jeremy Hartman (p.c.) also pointed out the contrast below: (i) a. ?There were arrested many people yesterday. b. *There got arrested many people yesterday. 32 3 Whereas (9)-(10) show that control by an implicit agent is possible with a be- but not a get-passive, the similar sentences in (13)-(14) show an opposite pattern, with control by the surface subject of get but not of be. Although the facts in (13)-(14) argue for a control-type analysis for (13b) and (14b), the question does arise as to why the sentences (9b) and (10b) do not allow construal with the implicit agent. A natural explanation suggests itself: the impossibility of (9a)-(10b) is due to the possibility of (13b)-(14b). In particular, the availability of a closer potential controller (i.e., the thematic subject of get) blocks the control by the implicit agent (which is part of a covert PP). So (9b) and (10b) are out, not because of the absence of an implicit agent, but because the purpose clause and adverb are forced to be construed with the inanimate subject of get. (As pointed out by Florian Schäfer, p.c., ‘potential’ should be understood in a technical sense.) This view allows for special circumstances that may facilitate control by implicit agents, thus overriding the blocking. As shown by Butler and Tsoulas (2006: 6) and Reed (2011: 55), certain sentences similar to (9b) and (10b) do indeed allow control by the implicit agent, with the choice of an appropriate main verb or when a more vivid context is provided. As a reviewer also points out, the Romance FP (faire par) construction seems to readily allow control by an implicit agent when a matrix subject is pragmatically excluded as a controller. However, the same does not happen with (9b) and (10b) even with an 33 inanimate subject. Clearly, the question of what counts as a ‘potential controller’ deserves further investigation. Regardless of the status of an implicit agent in English get-passives, it is clear that, an implicit agent is active in Chinese short bei-passives. The following example contains two volitional adverbs, one predicated on the subject of bei and the other on the implicit agent. The subject is deliberate and the implicit agent is malicious: (i) Zhangsan guyi bei buhuaihaoyidi piping-le. Zhangsan deliberately bei maliciously criticize-Per ‘Zhangsan deliberately got maliciously criticized.’ Finally, the idea of analyzing a Chinese bei-passive as an adjectival or resultative passive (where an implicit agent would be absent) would be entirely unmotivated, as bei is never used with adjectival or resultative complements. (A new type of passive with apparent adjectival complements has emerged in recent years in Mainland China, but it should be analyzed with a covert main verb akin to CONSIDER; see Huang 2011 for details.) 4 This view as it relates to the English get passive has been independently advanced by Reed (2011), as it turns out, who takes the construction to be three-way ambiguous, admitting versions of a raising structure (with verbal or adjectival complementation, as in Haegeman 1986 and Alexiadou 2005) and a control structure. Alexiadou (2012) offers a detailed new analysis showing the English get passive to be ambiguous between a passive and an anti-causative interpretation, and analyzing it as a middle-voice construction 34 whose head is underspecified between these two interpretations. The Chinese bei passive, on the other hand, does not involve adjectival complementation, but may take a verbal or clausal complementation. Furthermore, Liu (2012) shows that it has no anti-causative interpretation. Taken together, this shows that the get passive is more versatile, or more chameleon-like, than the bei passive. 5 Insertion of guyi ‘deliberately’ after dou ‘all’ in (17) gives rise to the literal meaning that all the cows intentionally got blown away by him. This shows that the presence of a volitional adverb forces a control analysis. 6 Toru Ishii (p.c.) has pointed out that while certain ni-passives in Japanese strongly disallow idiom-chunk subjects, others allow them in varying degrees. Thus the ni passive (ii) is only slightly less natural than the niyotte-passive. (i) John-ga [paatii-no masseki]-o kegasi-ta John-Nom [party-Gen bottom]-Acc make-dirty-Past ‘John had the honor of being present at the party.’ (ii) ?[Paatii-no masseki]-ga John-ni kegas-are-ta party-Gen bottom-Acc John-by make-dirty-Pass-Past (iii) [Paatii-no masseki]-ga John-niyotte kegas-are-ta party-Gen bottom-Acc John-by make-dirty-Pass-Past This may be seen as another reflection of the chameleonic character of the rare-passive. 35 7 Since the agentive long passive can also involve a raising structure as in (17), a nontrivial question arises as to how raising is possible given the original arguments in Huang (1999) and HLL (2009) that the long passive involves clausal complementation with the agent phrase occupying the subject position of the complement clause. This structure poses no problem for the original analysis involving null operator movement, but it does create a problem for the A-movement of the embedded object across the agent subject to the matrix subject position. Here I adopt the new analysis proposed in Liu (2012), according to which the ‘clausal’ complement is in fact a vP with the agent argument in its Spec, and A-movement of the embedded object may proceed by clause-internal scrambling without violating minimality or invoking any smuggling operation of the sort proposed in Collins (2005). Given what’s indicated in the preceding note, the situation of certain Japanese nipassives maybe similarly treated. 8 We thank Yang Gu (p.c.) for pointing out the relevance of this work to us. For related work on ‘voice continuum’ and the development of English passives, see Toyota (2008). Also of relevance is Collins (1996), who notes that descriptive approaches to the English get passive have resorted to a continuum on which to determine the relative passivity of a get-participle construction. 9 The English counterpart of the Chinese example, certain, fits in a control-like structure as well as a raising structure: 36 (i) a. John is certain that [he] will win. [control-like] b. John is certain to win. [raising] In fact, some speakers accept a control reading for (ib) as well, thus reflecting idiolectal micro-variations. 10 For examples and detailed discussions, see Cheng et al (1999), Tang (2003) and relevant references therein. The lexical counterpart of get in Chinese is de in Mandarin, but de is used in resultative constructions mainly, though in the Hakka dialect of Youxian it is also used to form the passive (Chuansheng He, p.c.). 11 The approach proposed here to the chameleonic character of passives does not explain why certain verbs can realize certain points on the causative-unaccusative continuum, while others cannot. It should be clear from elsewhere in the paper that the bandwidth of interpretations of a given item does not necessarily form a single continuum, either. We have nothing new to offer in answer to such questions, other than pointing to rates of grammaticalization and lexical replacement as possible factors, in equally descriptive terms. The situation is not unlike the question why threaten in English exhibits a controlraising ambiguity while warn does not, or why raising is possible with Chinese guji ‘estimate’ but not with qidai ‘expect’ or yuliao ‘predict’. 12 Sentence (30) is to be understood in a context where a dog’s getting sick means a reduction in the number of healthy members of a certain group. Similarly, in (31), fei 37 ‘fly’ must be understood as feizou ‘fly away’ and does not denote a simple activity. In these senses, the verbs are being used as unaccusatives. 13 We depart here from Shen and Sybesma’s (2010) proposal treating the gei-VP as a middle construction. Their hypothesis is at odds with certain known properties of the middle construction, and it is not clear how it derives all the properties of the gei-VP as observed. The proposal made here is closer to Tang (2006) in spirit, though not in detail. 14 The selectional dependency between a light verb and the main predicate (similar to famous have-be auxiliary selections in languages that exhibit them) is not unlike the selection between a classifier (a ‘light noun’) and the main noun. The deontic-epistemic alternation of certain modal auxiliaries also illustrates a similar effect: John must come (deontic) vs. John must have come (epistemic), the former with a complement that denotes an action, the latter with one that denotes an event in existence. 15 Any example with gei above rang, ba, jiao in a way similar to (37) with gei above bei would be entirely ungrammatical. 16 Just as a suppressed agent or experiencer may be overtly expressed (in the form of a by- or to-PP), so one may expect the implicit affectee to be optionally explicit. While this is not borne out in Mandarin for lack of an appropriate affectee marker, it is instantiated in Taiwanese Southern Min (TSM) by a ka-phrase. See Liu (2011) for discussion of the location of the affectee phrase. (i) chiao-a long hoo ka goa puei liaoliao khi-a. bird-DM all give KA I fly-away finish INCH-SFL ‘All the birds flew away clean on me.’ 38 17 See Cheng, et al (1999) and Lien (2008) for additional discussion of hoo in TSM. See Mathews and Yip (2008) and Matthews, Xu and Yip (2005) for the passive and unaccusative uses of ‘give’ in Chaozhou.