Toward a PF-deletion approach

4.2 Toward a PF-deletion approach

We have shown that Chinese RNR cannot be explained by the movement, the MD, the external remerge, the eclectic, or the sparse approaches. Specifically, given that (i) no island effects in the second conjunct and (ii) prepositions in Chinese can be stranded in the second conjunct, the RNR target is closely connected to the second conjunct. Rather than being raised, it stays within the adjacent conjunct. For this reason, we propose a PF-deletion approach to derive Chinese RNR. We will justify the validity of our proposal by presenting empirical evidence. Before getting into the details, let us state the basic assumption about the Chinese RNR construction. As observed in Wang (2014b) that Chinese RNR contains a coordinate structure, we assume that coordination is a necessary (or sufficient) condition for Chinese RNR.³⁵

In addition, we assume that in Chinese RNR, there are two entities of the RNR target which have fully represented structures in both conjuncts. Along the line of Merchant‘s ellipsis feature of sluicing (ES), we adopt Ha‘s (2008a, b, c) ellipsis feature of RNR (ERNR) and propose the licensing requirements of Chinese RNR as follows. Syntactically, E_RNR is attached to pre-RNR elements (which are contrastively focused) in the non-final conjuncts and has to be checked from the respective head of C. Phonologically, the checked E_RNR unpronounces the RNR target. Semantically, ERNR must satisfy e-GIVEN, i.e. the coordinate

35 As observed in Huybregts and Van Riemsdijk (1985), among others, de Vreis (2013) also notes there are RNR-like constructions which involve syntactic subordination as the Dutch example below:

(i) Het kan moeilijk zijn om syntactische __ van semantische factoren te onderscheiden it can difficult be to syntactic from semantic factors to distinguish

‗It can be hard to distinguish syntactic ____ from semantic factors.‘ (de Vreis 2013)

113

conjuncts must mutually entail each other. Consider (13) for illustration.

(13)

[&P[CP F[Zhangsan xihuan[ERNR]

<jufa>]], dan [

CP Lisi taoyan jufa.]

Zhangsan like syntax but Lisi hate syntax ‗Zhangsan likes, but Lisi hates syntax.‘

In (13), Zhangsan xihuan in the first conjunct is contrastively focused with Lisi taoyan in the second conjunct, so the ERNR feature can be attached to it and get checked to agree with the head of C. The E_RNR feature then guides PF-deletion of the RNR target, yielding one RNR target on the surface. Semantically, the antecedent Lisi taoyan jufa ‗Lisi hates syntax‘ turns to variables at LF and they are existentially closed by F-closure, yielding ∃x∃R[x R syntax]; the same applies to the RNR clause Zhangsan xihusn jufa ‗Zhangsan likes syntax‘ , yielding

∃x∃R[x R syntax]. Taken together, the antecedent and the RNR clause create sets of

alternative that are mutually entailed. Since the ERNR feature satisfies all the three requirements, a well-formed RNR is obtained.

Recall that the disadvantage of Ha‘s analysis is the inability to explain the RNR involving more than two conjuncts. Precisely, how can all the RNR targets in the non-final conjuncts be deleted if there is only one ERNR feature in the first conjunct which can be checked with the closet head of C? In order to accommodate those cases of RNR, we depart from Ha in the following aspect. Rather than entering only the first conjunct, ERNR can be attached to lexical items in the non-final conjuncts as long as they are contrastively focused.

And all the focused elements have to check the feature value with its own head of C in order to be interpretable. Moreover, to account for the specific properties of RNR in Chinese, we propose that (i) asymmetric island effects can be captured by a general constraint—deletion within an island in coordinate structure is prohibited, and (ii) asymmetric P-stranding can be seen as an extension from Merchant‘s (2001) generalization—P-stranding is consistent in

114

ellipsis sites.

Let us illustrate how the current proposal captures the properties of Chinese RNR. First, look at non-constituency in (14), where the RNR target contains two objects of the verb and cannot undergo leftward movement or be clefted.

(14)

Zhangsan zengsong er Lisi ye zengsong -le, Wangwu yi-ben

shu

Zhangsan give and Lisi also give-Asp Wangwu one-Cl book

‗Zhangsan gave and Lisi also gave Wangwu a book.‘

If we follow Huang, Li and Li‘s (2009) analysis of VP structure, (14) is in fact a type of constituent. Along the line of Larson (1988) and Chomsky (1995), VP internal arguments are assigned theta-roles by the verb, and outside the VP, there is a verbal head v which introduces Agent. Structurally speaking, VP is the complement of v as illustrated in (15).

(15) vP

NP v‟

v VP

Consequently, the double-object construction (16a) has the structure in (16b), where the NP is the subject, and XP1 and XP2 are the two objects. Note that the lexical verb moves to v to obtain the linear order of subject-verb-object1-object2 in Chinese as in structure (16b).

XP1 V‘

V XP2

115

(16)

a. ta di-gei gege yi-hu jiu he pass-give brother one-Cl wine

‗He passed his brother a jug of wine.‘

b. vP

NP1 v‟

v VP

he pass-give brother t a jug of wine

Shown in (16b), after the V-to-v movement, NP2 and NP3 are the constituents which we hear.

Adopting Huang, Li and Li‘s analysis of double-object constructions, the RNR example (14) has the structure as in (17).

NP2 V

V NP3

116

Seen in (17), the RNR target Wangwu yi-ben shu ‗Wangwu a book‘ in the first and the second conjunct is a constituent under VP.

The RNR examples involving the Coordinate Structure Constraint are exemplified in (18).

117

‗There‘s a grocery store near Zhangsan‘s home, a convenience store next to Lisi‘s office and one next to Wangwu‘s school.

In (18), the ERNR feature enters the contrastively focused lexical elements Zhangsan jia fujin

you in the first conjunct and Lisi gongsi pangbian you in the second conjunct, each of which

checks with its C head to be interpretable. However, the phonological and semantic licensing requirements are not satisfied. The E_RNR should instruct the elements following it to be unpronounced, namely zahuo dian ‗grocery store‘ in the first conjunct and bianlishangdian

‗convenience store‘ in the second conjunct. But pronouncing the former violates the phonological condition. What‘s worse, semantically, the first conjunct does not create the same set of alternatives as the second and the third conjuncts. To be specific, the F-closure of the first conjunct is ∃x∃R[x R grocery store] whereas that of the second and the third

conjuncts are ∃ x∃ R[x R convenience store]. Without meeting the phonological and the semantic licensing requirements, (18) is ill-formed.

Turn to asymmetric island effects in Chinese RNR. Recall that island insensitivity in English RNR favors the PF-deletion analysis since no movement is involved (cf. chapter 2).

On the other hand, island effects are observed asymmetrically in the first conjunct of Chinese RNR. If there is no movement involved in the Chinese RNR, can the PF-deletion approach account for the asymmetric island effects in Chinese RNR? The answer is positive. In fact, such effects can be reduced to a more general prohibition against deletion within an island in coordinate structures. To show this general ban, we would like to consider VP ellipsis in English. The reason why we consider VP ellipsis in English but not in Chinese is because we assume with Li (2005 and her subsequent works) that in Null Object Construction, the empty position must be empty at overt syntax as a last resort.³⁶ Consider (19) and (20) for the

36 Li (2005) and her subsequent works argue for a True Empty Category (TEC) as a last resort of empty object based on the availability of strict/sloppy readings as in (i) and indefinite reading as in (ii). The interpretations of

118

illustration of VP ellipsis in coordinate structures. The only difference between (19) and (20) is that the latter involves VP ellipsis within an adjunct island. The ungrammatical (20)

indicates that deletion cannot be applied to elements inside an island in coordinate structures.

(19)

John likes novels, and Bill thinks John does [e].

(20)

*John likes novels, and Bill buys a lot of novels because John does [e].

Given this general ban, under the PF-deletion approach, a Chinese RNR sentence involving deletion within an island should be ruled out even if the E_RNR meets the licensing requirements. This prediction is borne out as demonstrated in (21) and (22).

(21)

*Lisi mashang jujue-le Wangwu de tiyi yinwei ta taoyan

‗Lisi rejected Wangwu‘s suggestion right away because Mary who he hates also joined, but Zhangsan still considers whether to join the debate club or not.‘

(22)

Lisi mashang jujue-le Wangwu de tiyi yinwei ta taoyan de Lisi right.away reject-Asp Wangwu DE suggestion because he hate DE

the TEC is claimed to be obtained via LF-copying of its antecedent as in (i) or via discourse context as in (ii).

(i) Zhangsan kanjian-le ta-de zhaopian, Mali ye kanjian-le e.

Zhangsan see-Asp he-DE picture Mali also see-Asp

a. ‗Zhangsan saw his picture, and Mali also saw Mali‘s picture.‘ (sloppy reading) b. ‗Zhangsan saw his picture, and Mali also saw Zhangsan‘s picture.‘ (strict reading) (ii) Zhangsan dagai hai mei zhaodao e

Zhangsan probably still not find

‗Zhangsan probably still has not found something.‘

119

Mali ye jiaru-le bianlun she dan Zhangsan haizai kaolu yao-bu-yao Mary also join-Asp debate club but Zhangsan still consider yes-not-yes jiaru

bianlun she

join debate club

‗Lisi rejected Wangwu‘s suggestion right away because Mary who he hates also joined the debate club, but Zhangsan still considers whether to join the debate club or not.‘

Example (21) has a non-elliptical counterpart in (22), where the RNR target being part of the adjunct is base-generated in the right edge of each conjunct. The contrast between (21) and (22) is accounted for since the former violates the general prohibition on the coordinate structures, i.e. deletion within an island is not allowed. To support the general ban on deletion within an island, consider the examples of English VP ellipsis, where island effects are observable (cf. Chomsky 1995, Chomsky and Lasnik 1993, Merchant 2001, Lasnik 2007, among others).

(23)

They said they heard about a Balkan language, but I don‘t know which Balkan language they said they heard about. (Lasnik 2007)

(24)

*They said they heard about a Balkan language, but I don‘t know which Balkan

language they did. (Lasnik 2007)

Example (23) has the counterpart of VP ellipsis in (24), where the wh-extraction is unacceptable. To account for the fact, Lasnik argues for the structure as in (25).

(25)

They said they heard about a Balkan language, but I don‘t know which Balkan language [_TP they T [_AspP did <_VP say that they head about a Balkan langauge>]]. (Lasnik 2007)

120

According to Lasnik, the deletion targets VP, leaving the two possible maximal projections AspP and TP. With the assumption that one of them is an island, the presence of it leads to the ill-formedness. Following the logic, Lasnik further predicts a well-formed VP ellipsis can obtain if a movement of the antecedent clause is involved as in (26).

(26)

? I know which book John said that Mary read, but YOU don‘t know which one he did.

(cf. *I know that John said that Mary read a certain book, but I don‘t know which one he

did.) (Lasnik 2007)

Seen in (26), the object of know in the antecedent clause is extracted out of the TP, and that in the ellipsis clause is as well extracted. After the extraction, no element is left in AspP or TP. It follows that no island effects are observeable. That is, syntactic violation is fine as long as it can be fixed/deleted at PF (i.e. island repair).The unacceptability of (26) shows that there is a parallel strcutre in some cases of ellipsis, supporting the argument of internal syntactic structure in the ellipsis site.

Next we turn to asymmetric P-stranding in Chinese RNR. We observe that Chinese disallows P-stranding in RNR just as in non-RNR sentences.

(27)

*women cai taolun wan guanyu, dan tamen yijing biaojue wan guanyu we just discuss finish about but they already vote finish about

biye luxing de xingcheng

graduation trip DE itinerary

‗We just discussed about, but they already voted about, the graduation trip itinerary.‘

(28)

women cai taolun wan guanyu biye

luxing de xingcheng

we just discuss finish about graduation trip DE itinerary dan tamen yijing biaojue wan guanyu

biye luxing de xingcheng

121

but they already vote finish about graduation trip DE itinerary

‗We just discussed about the graduation trip itinerary, but they already voted about the graduation trip itinerary.‘

Illustrated in (27), the elided NP, which is the object of the preposition, leads to

ungrammaticality. On the other hand, the non-elliptical counterpart in (28) is grammatical.

In Chinese, the similar prohibition on P-stranding in RNR and non-RNR sentences can be seen as an extension to Merchant‘s (2001) generalization on P-stranding in (29).

(29) Preposition Stranding Generalization

(Merchant 2001) A language L will allow preposition stranding under sluicing iff L allows preposition stranding under regular wh-movement.

Merchant observes that languages (e.g., English, Swedish, Danish, Icelandic) which have P-stranding independently allow this option in sluicing³⁷ as in (30); languages which do not have P-stranding (e.g., Greek, German, Dutch, Russian) do not allow the option in sluicing as in (31).

(30)

a. Peter was talking with someone, but I don‘t know who. (Merchant 2001) b. Sluicing

Peter‘s mom will get angry if he talks with someone from his class, but I don‘t

remember who. (Merchant 2001)

37 Sluicing is a construction in which the ellipsis takes place but leaves the wh-phrase as in (i) and is reconstructed as in (ii). We assume an analysis of sluicing where it is generally viewed as the result of wh-movement out of IP followed by IP-deletion at PF (Ross 1969, Merchant 2001, among others).

(i) Bill drank something, but he doesn‘t know what.

(ii) Bill drank something, but he doesn‘t know what_i he drank t_i.

122

(31)

Greek

a. I Anna milise me kapjon, alla dhe ksero *(me) pjon the Anna spoke with someone but not I.know with who

(Merchant 2001) b. Sluicing

I mitera tou Giannis tha thimosi an milisi me kapjon apo the mom of Giannis FUT get.angry if he.talks with someone from tin taksi tou, alla dhe thimame *(me) pjon

the class his but not I.remember with who

‗Giannis‘s mom will get angry if he talks with someone from his class, but I don‘t

remember who.‘ (Merchant 2001)

Likewise, Wang and Wu (2006) claim that sluicing in wh-in-situ languages involves overt

wh-movement followed by IP deletion just as wh-movement languages. The only difference

is that the overt wh-movement in the former type of languages is to meet a [+Focus] feature requirement while the latter is to meet a [+wh] feature. Wang and Wu observe the

unavailability of P-stranding in Chinese sluicing as in (32). As indicated in (33), the

derivation is like: the preposition gen ‗with‘ is pied-piped with wh-word shei ‗who‘, followed by the deletion of IP.

(32)

Lisi gen ren qu wan, keshi wo bu zhidao (shi) *(gen) shei [_IP Lisi t_i qu wan ] Lisi with person go play but I not know be with who Lisi go play

‗Lisi went on a trip with someone, but I don‘t know with who.‘ (Wang and Wu 2006)

(33)

…keshi wo bu zhidao [CP shi [FocP gen sheii [IP Lisi ti qu wan ]

Seen from the above, the unavailability of P-stranding in Chinese sluicing (which is also a

123

type of ellipsis) and RNR suggests that ellipsis may be involved in the RNR constructions.

Therefore, a PF-deletion approach to RNR is desirable. As for McCloskey‘s (1986)

generalization that some languages allow P-stranding in RNR but not in non-RNR sentences, we think that RNR in these languages may not be derived under the PF-deletion approach but under the MD approach.

Turn to sloppy reading as in (34), where the RNRed pronoun can be a bound variable and co-indexed with the NP in the same conjunct. The pronoun in the first conjunct refers to

Zhangsan and that in the second conjunct refers to Lisi.

(34)

Zhangsani xihuan[ERNR] <tai

de laoshi>, dan Lisi

j taoyan, ta j

de laoshi

Zhangsan like he DE teacher but Lisi hate he DE teacher

Following Ha (2008a, b, c), the E_RNR feature enters the derivation and attaches to the contrastive focused elements Zhangsan xihuan ‗Zhangsan likes‘ in the first conjunct. After checking with its head of C to get valued, the E_RNR feature instructs the PF component not to pronounce the elements following it in the same conjunct. Note that the deletion operation occurs at PF, so the semantic properties of the pronouns are not affected by the deletion operation. At LF, the antecedent turns to bound variables which are existentially closed as

Lisi λx [x hates x‟s teacher]; the RNR clause has the F-closure as Zhangsan λy [ y likes y‟s teacher]. The outcome of F-closure of the antecedent and the RNR clause is still the same,

∃x∃R [x R x‟s teacher].Since the syntactic, phonological, and semantic requirements have

been met, sloppy reading is captured under the current account.

Now we look at asymmetric NPI licensing. First recall that we think there is a vehicle change operation in Chinese RNR but the effects are invisible because of an independent

constraint in Chinese: R-expressions have to precede pronouns in sentences (cf. section 3.1.2).

Although the vehicle change effects in the nominal equivalences are invisible, another type of

124

vehicle change in the suppletive paradigms (e.g., some/any) can be observed in asymmetric licensing of NPI. Under ellipsis, polarity items and nonpolarity items are interchangeable (cf.

Sag 1976, Fiengo and May 1994, Johnson 2001, among others). Consider (35).

(35)

a. Max talked to someone, but Oscar didn‘t. (Fiengo and May 1994) b. *Max talked to someone, but Oscar didn‘t talk to someone.

c. Max talked to someone, but Oscar didn‘t talk to anyone.

According to Fiengo and May (1994), (35a) has the reconstructed counterpart in (35b), where

the nonpolarity someone leads to ungrammaticality. Since the elements in the paradigm are not distinctive in the reconstruction, someone can be treated as anyone as in (35c). Following Ha‘s (2008a, b, c) analysis of English RNR, we argue that this type of vehicle change is applicable in Chinese RNR as well. Consider (36) and (37).

(36)

a. Zhangsan mai-le dan Lisi mei mai renhe

jinianpin

Zhangsan buy-Asp but Lisi not buy any souvenir

‗Zhangsan bought, but Lisi didn‘t buy any souvenirs.‘

b. *Zhangsan mai-le

renhe jinianpin dan

Lisi mei mai renhe jinianpin Zhangsan buy-Asp any souvenir but Lisi not buy any souvenir

‗Zhangsan bought any souvenirs, but Lisi didn‘t buy any souvenirs.‘

c. Zhangsan mai-le

yixie jinianpin dan Lisi

mei mai renhe jinianpin Zhangsan buy-Asp some souvenir but Lisi not buy any souvenir

‗Zhangsan bought some souvenirs, but Lisi didn‘t buy any souvenirs.‘

(37)

*Lisi mei mai dan Zhangsan mai-le

renhe jinianpin

Lisi not buy but Zhangsan buy-Asp any souvenir ‗Lisi didn‘t buy, but Zhangsan bought any souvenirs.‘

125

Sentence (36a) has the unelided counterpart as in (36b), containing two entities of the RNRed NPI renhe ‗any‘. At this point, renhe ‗any‘ is not acceptable in the first conjunct due to the lack of a local negative operator. By applying vehicle change and changing renhe ‗any‘ with

yixie ‗some‘ at PF as in (36c), the result is grammatical. In (37), however, the operation of

vehicle change cannot be applied to the second conjunct since the RNRed NPI renhe ‗any‘ in the second conjunct is not in an ellipsis site. Taken together, the mismatch between the polarity and nonpolarity should be treated the same in ellipsis sites. And the vehicle change effects are captured under the PF-deletion analysis.

Regarding the licensing of relational modifiers, we assume with Ha (2008b) that the single copy of the relational modifier makes the covert quantifier raising available, inducing an internal reading. Note that we would need an additional assumption which allows the PF component to be accessible to the LF component, rather than treating these two components inaccessible to each other.

(38)

Zhangsan yong shou er Lisi yong qiubang da-le bici

Zhangsan with hand while Lisi with bat hit-Asp each.other

‗Zhangsan hit Lisi with hand, while Lisi hit Zhangsan with a bat.‘ (Cheng 2012)

(39)

[bici] Zhangsan yong shou, er Lisi yong qiubang da-le

each.other Zhangsan with hand while Lisi with bat hit-Asp

‗Each other, Zhangsan hit with hand while Lisi hit with a bat.‘

Example (38) is obtained by the deletion of the RNR target in the non-final conjunct (i.e. the E_RNR feature meets the licensing requirements). Consequently, the only one RNR target in the second conjunct undergoes the covert movement to scope over the coordination as in (39).

As summed up in Table 13, we justify our analysis by accounting for all the properties under the PF-deletion operation: Right Edge Restriction, non-constituency, Coordinate

126

Structure Constraint, asymmetric island sensitivity, asymmetric P-stranding, sloppy reading, asymmetric licensing of NPI, and the licensing of relational modifiers.

Table 13 Summary of the approaches to Chinese RNR

Approaches

Properties Cheng(2012) Wang (2014a, b) Current

proposal

So far, we have illustrated the PF-deletion approach to Chinese RNR and have solved the puzzles of asymmetric islands effects and P-stranding. In this section, we would like to address a more general question of whether such an approach is applicable to the English counterpart.

The properties of English RNR have already been discussed in Chapter 2. Among the ATB-movement approaches, Sabbagh‘s (2007) version solves the RNR sentences that violate the movement conditions (i.e.Right Roof Constraint violation and island insensitivity) and captures the non-constituency, licensing of relational modifiers and scope ambiguity.

Nevertheless, these properties can also be accounted for under the PF-deletion or the MD approach as well. There are insufficient arguments for the ATB-movement approach to English RNR and failures to capture other properties (i.e. sloppy reading, asymmetric vehicle change, asymmetric licensing of NPI, and summative agreement). In the following, we show

Nevertheless, these properties can also be accounted for under the PF-deletion or the MD approach as well. There are insufficient arguments for the ATB-movement approach to English RNR and failures to capture other properties (i.e. sloppy reading, asymmetric vehicle change, asymmetric licensing of NPI, and summative agreement). In the following, we show

在文檔中 漢語右部節點提升結構 (頁 119-133)