Chapter 3 Properties and Previous Analyses of the Chinese RNR
3.2 Two previous works on Chinese RNR
3.2.2 Wang (2014a, b)
Wang proposes an approach with True Empty Category (TEC) (Li 2005 and her
subsequent works) plus leftward focus movement (cf. Cheung 2009) to Chinese RNR. TEC is an element that can be generated as null only in subcategorized positions, and the
interpretation of such an null element is obtained via LF-copying or discourse context (Li 2005 and her subsequent works). Following Li, Wang argues that the notion of TEC can be applied to RNR, as subcategorized by a verb in (56a) and by a modificational marker de in (56b).
(56) a. Zhangsan yao mai e=TEC, keshi Lisi bu yao mai e=TEC, wo-de
fangzi
Zhangsan want buy but Lisi not want buy I-DE house ‗Zhangsan wants to buy, but Lisi does not want to buy, my house.‘ (Wang 2014b) b. Lisi kanjian heise de e=TEC gen baise de e=TEC,na
Lisi see black DE and white DE that liang-zhi
guibingou
two-Cl poodle
‗Lisi saw the two poodles, one of which is black and the other is white.‘(Wang 2014b)
88
Let us look at Wang‘s proposal of clausal RNR as in (57).
(57) a. [RNR target [[S V e(=TEC) ] + conj. + [S V e(=TEC) ]]] (Wang 2014b) b. [[[S V e(=TEC)
] + conj. + [S V e
(=TEC)]]
i, # [RNR target, […]i]] (Wang 2014b)
Focus Movement c.
(Wang 2014b)
Her arguments are as follows. First, the RNR target is base-generated as a TEC in each
conjunct as in (57a). The RNR target is assumed to be displaced from the coordinate structure, and the TEC occupies a position to meet the subcategorization requirement of verbs.
Secondly, the rightmost RNR target is actually derived from leftward focus movement of the coordinate structure (cf. Cheung 2009). Seen in (57b), the leftward focus dislocation of the coordinate structure is applied, deriving the seemingly right displaced RNR target. The tree structure is shown in (57c). Third, this movement captures the phonological boundary (#) preceding the RNR target. A clausal RNR is exemplified in (56a). Repeated as in (58).
89
(58) Zhangsan yao mai e=TEC, keshi Lisi bu yao mai e=TEC, wo-de
fangzi
Zhangsan want buy but Lisi not want buy I-DE house‗Zhangsan wants to buy, but Lisi does not want to buy, my house.‘ (Wang 2014b)
Shown in (58), the TEC is originally merged and subcategorized by the verb mai ‗buy‘ in each conjunct. The rightmost RNR target wo-de fangzi ‗my house‘ is the result of the leftward focus movement of the coordinate conjuncts.
Wang argues that TEC accounts for the ungrammaticality of non-arguments as RNR targets (e.g., descriptive complement, resultative complement, duration phrase, and frequency phrase)30 in (59).
(59) Descriptive Complement
*[Zhangsan kan shu kan-de ____], [Lisi ye kan shu kan-de ____], hen kuai Zhangsan read book read-DE Lisi also read book read-DE very fast
(Wang 2014b)
Seen in (59), the RNR target is a descriptive complement hen kuai ‗very fast‘ which obeys
30 Wang‘s claim that TEC occupies the subcategorized position cannot be applied to Dutch, Korean, or Japanese.
Observed in Kluck and Zwart (2009), Dutch RNR can involve conjuncts that are part of a coordinate clausal subject and the RNR target is possible to be a verb as in (i):
(i) Dat Anna een APpel ____, en Roos een baNAAN ___eet is heel vreemd that Anna a apple and Roos a banana eat-3SG is very strange
‗That Anna eats an apple and Roos eats a banana, is very strange‘ (Kluck and Zwart 2009) In addition, An (2007) provides examples of Korean and Japanese RNR in which the RNR target involves the matrix and embedded verbs, as in (ii) and (iii).
(ii) Korean
Tomo-nun Ana-ka ppang-ul, kuliko Nina-num Ana-ka bap-ul, mekess-tako malhayssta T-top A-nom bread-acc and N-top A-nom rice-acc ate-comp said
‗Tomo (said that) Ana (ate) bread and Nina said that Ana ate rice.‘
(iii) Japanese
Tomo-wa Ana-ga pan-o , (sosite) Nina-wa Ana-ga gohan-o, tabeta-to itta T-top A-nom bread-acc and N-top A-nom rice-acc ate-comp said
‗Tomo (said that) Ana (ate) bread and Nina said that Ana ate rice.‘
90
RER but the sentence is illicit. According to Wang, the ungrammaticality of (59) is due to the unsubcategorized RNR target. Since the descriptive complement is not subcategorized by the verb kan-de ‗read-DE‘, the gap cannot be a TEC, leading to ill-formedness. Neither the ATB-movement nor the PF-deletion approach correctly predicts the ungrammaticality of (59) since it obeys RER. However, as we have mentioned in 3.1.1, there are acceptable sentences of RNR containing non-arguments as in (60)-(63). Moreover, if we follow Huang, Li, and Li (2009), the phrase following descriptive and resultative de as in (60) and (61) is also
subcategorized.
(60) Descriptive Complement
[Zhangsan kan shu kan-de ____], [Lisi ye kan shu kan-de ____], jinjin Zhangsan read book read-DE Lisi also read book read-DE interest
you wei
have interest
(61) Resultative Complement
[Zhangsan da-de Lisi ___], [Wangwu ye da-de Lisi______],tou po Zhangsan hit-DE Lisi Wangwu also hit-DE Lisi head break
xie liu
blood flow (62) Duration Phrase
[Zhangsan kan shu kan-le ____], [Lisi ye kan shu kan-le ____], zheng Zhangsan read book read-Asp Lisi also read book read-Asp completely
zheng ba-ge xiaoshi
completely eight-Cl hour (63) Frequency Phrase
[Zhangsan kan shu kan-le ____], [Lisi ye kan shu kan-le ____], chaoguo
91
Zhangsan read book read-Asp Lisi also read book read-Asp over
san-ci
three-time
Turn to the available sloppy reading. According to Wang, sloppy reading can be obtained via LF-copying of the RNR target. This is because a TEC is assumed to be without any specific interpretation. Adopting Panagiotidis‘ (2003) analysis of empty noun,31 Wang claims that a TEC may combine with a null D, deriving the indefinite and definite readings in the clausal RNR. That is, the interpretation of a TEC is flexible. Consider (64) and (65).
(64) Zhangsan xiang mai etec, Lisi ye xiang mai etec, yi ben shu Zhangsan want buy Lisi also want buy one CL book a. ‗Zhangsan wants to buy a book, and Lisi also wants to buy the same one‘
([DP[+def.] [NP etec]]: the book) (the same book) (Wang 2014b) b. ‗Zhangsan wants to buy a book, and Lisi also wants to buy one.‘
([DP[-def.] [NP etec]]: a book) (different books) (Wang 2014b) (65) Zhangsani hen zunjing etec, Lisij ye hen zunjing etec, tai/j/k
—de laoshi
31 Panagiotidis (2003) proposes a unified account of all empty nouns (eN), which are phonologically empty and without any descriptive content. Empty nouns are just as normal lexical entities listed in the lexicon, meaning that no licensing or identification is needed. The interpretation or reference of the empty pronouns is obtained via the pragmatic or LF-copying. See (i).
(i) I Nena puluse [ftina isitiria]i ki o Aris aghoraze e*i/j
the Nena was.selling cheap tickets and the Aris was.buying (Giannakidou and Merchant 1996)
In (i), the empty element e is non-referential and cannot be co-referent with the DP containing its antecedent.
That is, e means tickets or cheap tickets but not the tickets Nena was selling.
Besides, the empty pronouns can combine with D, whose interpretation (definite or indefinite) relies on the semantic of D. In the French example (ii), somebody shows a collection of photos and says
(ii) Les vieilles eN sont drôles.
the.PL old are funny
‗The old ones are funny.‘
The elided DP can refer to any set of old things at LF: a partitive reading (in a proper subset of all the photos shown is funny, a subset of old photos) or non-partitive reading (general comments on the old photos).
92
Zhangsan very respect Lisi also very respect he-DE teacher a. ‗Zhangsan respects Zhangsan‘s teacher, and Lisi also respects Zhangsan‘s teacher.‘
(strict reading) (Wang 2014b) b. ‗Zhangsan respects Zhangsan‘s teacher, and Lisi also respect Lisi‘s teacher‘
(sloppy reading) (Wang 2014b)
In (64), yi ben shu ‗a book‘ can refer to the identical book (strict) or distinct books (sloppy).
Likewise, in (65), the pronoun ta-de laoshi ‗his teacher‘ can refer to either (i) Zhangsan‘s teacher (strict) or (ii) both Zhangsan‘s teacher and Lisi‘s teacher simultaneously (sloppy).32 On the other hand, if the gaps were pro or A‘-variables bound by the RNR target, only the identical reference can be obtained.
Just as the verb subcategorizes a TEC in clausal RNR shown above, the modificational marker de subcategorizes a TEC in a nominal RNR, as in (66).33
(66) a. [RNR target, [ S V [XP DE e(=tec)] + conj. + [XP DE e(=tec)]]] (Wang 2014b)
b. [[ S V [XP DE
e
(=tec)] + conj. + [XP
DEe
(=tec)]]]
i, [RNR target,[ …]i]](Wang 2014b)
32 Although Wang does not explicitly illustrate the strict reading in which the RNR target refers to the subject in the second conjunct (i.e. Lisi), we think her analysis can capture that as well as the strict reading which refers to the subject in the first conjunct (i.e. Zhangsan).
33 Examples of English nominal RNR is given in (i).
(i) I know Sally‘s but not Mary‘s parents. (Johnson 2007)
Focus Movement
93
c.
(Wang 2014b)
Consider (56b) (repeated as (67)) for the example of nominal RNR and its derivation in (68).
(67) Lisi kanjian heise de e=TEC gen baise de e=TEC,
na liang-zhi
Lisi see black DE and white DE that two-Clguibingou
poodle
‗Lisi saw the two poodles, one of which is black and the other is white.‘ (Wang 2014b) (68) [FocP ___ [InnTop na liang-zhi gou [VP
kanjian [
ConjP[heise de e
tec] gen [baise de e
tec]]]]]]
(Wang 2014b)
Shown in (67), the TEC is subcategorized by de in each conjunct, forming the nominal RNR.
After the coordinate structure undergoes the leftward focus movement, the RNR target turns out to be the rightmost of the sentence as in (68).
Now consider (69) and (70).
94
(69) Lisi kanjian [heise gen baise]
de xiongmao
Lisi see black gen white DE panda‗Lisi saw black and white pandas.‘ (mixed reading) (Wang 2014b) (70) Lisi kanjia [[heise] de gen [[baise] de] guibingou
Lisi see black DE and white DE poodle
‗Lisi saw black poodles and white poodles.‘ (split reading) (Wang 2014b)
Wang claims that there are two readings, the mixed and the split ones. If the reading is a mixed one, the coordinate conjuncts are two adjectives as in (69). On the other hand, if the reading is a split one, the coordinate conjuncts are two nominal phrases and should be viewed as RNR as in (70). Note that only one modificational marker de is used in the mixed reading (69) while two de‘s in the split reading (70). However, we observe that there are cases which have only one de but both mixed and split readings are available as shown in (71). Therefore, the number of de in coordinate conjuncts cannot be used to distinguish the adjective
conjuncts from nominal RNR.
(71) heise gen baise
de dongwu hen duo
black and white DE animal very many‗There are many black-and-white animals.‘ (mixed reading)
‗There are many black animals and white animals.‘ (split reading)
Besides, Wang claims that the prosodic boundary preceding the RNR target can be accounted for. Chinese displays a productive tone sandhi: when two underlying third tones are adjacent to each other, the underlying T3-T3 (Low-Low) sequence becomes T2-T3 (Rising-Low) on the surface. However, in (72b), no tone sandhi is observed between the first/second verb and the RNR target which are supposed to be T3-T3 sequences, indicating a
95
prosodic boundary between the first/second verb and the RNR target. Example (72a) is derived from the underlying structure of (72c): the RNR target is separated from the
coordinate structure, and the leftward focus movement is applied to the coordinate structure, giving rise to the RNR target in the rightmost position on the surface. However, the fact that no tone sandhi is not necessarily induced by the leftward focus movement of the RNR target as argued by Wang.34
(72) a. Zhangsan yao mai, Lisi ye yao mai, wu-dong fangzi Zhangsan want buy Lisi also want buy five-Cl house
‗Zhangsan wants to buy, and Lisi also wants to buy, five houses.‘ (Wang 2014b)
b. (Zhangsan yao mai), (Lisi ye yao mai) (wu dong fangzi)
L L L L Underlying tone ---
L# L
L# L Surface tone (Wang 2014b) c. [wu-dong fangzi [[Zhangsan yao mai etec], [Lisi ye yao mai etec]]] (Wang 2014b)Turn the island conditions in Chinese RNR. Example (73) involves an adjunct island while it is still grammatical, indicating that no movement takes place and the TEC is base-generated.
(73) wo hen shengqi yinwei laopo mai-le, Lisi ye hen shengqi yinwei ta-de I very angry because wife buy-Asp. Lisi also very angry because he-DE laopo ye mai-le,
na ge jia baobao
wife also buy-Asp that Cl fake bag
34 I would like to thank Professor Hsiao-hung Iris Wu for pointing it out to me.
96
‗I am angry because my wife bought, and Lisi is also angry because his wife also bought
that fake bag.‘ (Wang 2014b)
However, we do not think (73) is as good as Wang claims. Counterexamples involving adjunct clause is unacceptable as in (74), suggesting that Chinese RNR is subject to island conditions.
(74) *wo bu hui jia gei Zhangsan yinwei wo taoyan, dan Mary hui jia I not will marry to Zhangsan because I hate but Mary will marry gei Zhangsan yinwei ta xihuan, you qian
ren
to Zhangsan because she likes have money people
‗I will not marry Zhangsan because I hate, but Mary will marry Zhangsan because she loves rich people. ‘
Wang also makes a distinction between the inner Topic (RNR targets) and the outer Topic (in the general Topicalization constructions). Following Rizzi‘s (1997) Left Periphery, Wang claims two topic positions (i.e. outer and inner) in the CP periphery, and the RNR targets are base-generated as an inner topic (cf. Shyu 1995 and Paul 2002). Syntactically, (i) an inner topic does not have full lexical content and allows a sloppy reading whereas an outer topic denotes old information and must be specific, (ii) an inner topic can involve an NPI while an outer topic cannot, and (iii) an inner topic must be an argument while an outer topic can be a non-argument. Semantically, the crucial difference between the inner and the outer topic is that the former has identity function and the latter has the inclusion function.
Consider (75).
(75) a. Lisi kaizou-le heise de etec=car gen hongse de etec=car,
97
Lisi drive-Asp black DE and red DE [InnToP
na liang-tai che]
that two-Cl car
(those two cars)=red car + black car (Wang 2014b) b. ?*[OutToP
na liang-tai che], Lisi kaizou-le
heise de etec=car genthat two-Cl car Lisi drive-Asp black DE and hongse de etec=car
red DE
(those two cars) ⊇{red cars, black cars} (Wang 2014b)
According to Wang, in (75a), the inner topic refers to two specific cars (a red one and a black one). If (75b) is acceptable, the outer topic refers to two kinds of cars but not two specific cars.
Regarding (ii), which claims that a NPI can be an inner topic, we do not think that the TEC approach can handle asymmetric NPI licensing as in (76).
(76) Zhangsan mai-le dan Lisi mei mai
renhe shu
Zhangsan buy-Asp but Lisi not buy any book‗Zhangsan bought but Lisi didn‘t buy any book.‘
If the base-generated TEC can get its semantic content via LF-copying or discourse, why can the NPI renhe ‗any‘ be licensed in the first conjunct, where no negative operator is contained?
Table 11 sums up the advantages and disadvantages of Wang‘s analysis. The RER, CSC, and strict/sloppy readings are captured. Furthermore, if the RNR target does not undergo
movement, then we think this approach may possibly handle the constraints on syntactic movements such as non-constituency and RRC violation. TEC with the feature [+definite]
98
induces the internal reading and might account for the relational adjectives (e.g., xiangtong
‗same‘and butong ‗different‘). The disadvantages of her account are as follows: (i) the subcategorization requirement of RNR target does not hold, (ii) TEC does not seem to capture the licensing of relational pronouns (e.g., bici ‗each other‘) in that this type of
pronouns require a plural antecedent composed of the two disjoined subject DPs, and (iii) we do not think that asymmetric sland effects, P-stranding, and NPI licensing can be captured.
This is because Wang seems to assume the same syntactic environment for the licensing of TEC in both conjuncts.
Table 11
Properties claimed to be explained by Wang’s approach with TEC plus leftward focus movement
1. Right Edge Restriction (RER)
2. Non-constituency ?
3. Coordinate Structure Constraint (CSC) 4. Asymmetric island sensitivity ?
5. Asymmetric P-stranding ?
6. Strict/sloppy readings
7. Asymmetric licensing of NPI ? 8. Licensing of relational modifiers ?
3.2.3 Summary
Cheng (2012) argues for an MD approach regarding the strict reading of ACD, the licensing of relational modifiers bici ‗each other‘ and the plural marker -men. Nevertheless, his examples of ACD may not be qualified as ACD sentences because they do not involve VP ellipsis or quantifiers. Besides, the ungrammatical RNR involving relative clause is not due to the island sensitivity as Cheng argues; rather, it is because of the RER violation. On the other hand, Wang (2014a, b) accounts for the illicit RNR target (i.e. non-argument RNR target), the
99
flexible interpretations of the RNR target (i.e. strict/sloppy readings), and she also observes the prosodic boundary between the coordinate conjuncts and the RNR target, and the difference between the inner topic and the outer topic. Still, there are problems of Wang‘s analysis. First, the number of de in coordinate conjuncts cannot determine whether the
sentence is an RNR or not. Second, it is inaccurate to claim that Chinese RNR is not sensitive to island conditions. Third, the leftward movement account is unnatural and complex,
especially because the RNR target is part of the base form while the coordinate structure is derived from the leftward focus movement. Also, it is unclear how this movement is pragmatically motivated, or how a supposed pragmatic feature optionally interacts with the syntax.
3.3 Chapter summary
We first present the general properties of Chinese RNR, most of which are similar to its English counterpart while differences are found as well. Chinese RNR shows asymmetric island effects and asymmetric P-stranding (due to specific constraint in Chinese) whereas English shows asymmetric vehicle change and scope ambiguity, and summative agreement effects. Then we discuss two previous works on Chinese RNR, which are summarized in Table 12. Cheng‘s (2012) MD analysis cannot account for asymmetric island effects, P-stranding, and NPI licensing. Furthermore, he incorrectly attributes the violation of RER (in which part of the relative clause is contained in the target) to the violation of island conditions. On the other hand, Wang‘s (2014a, b) approach with TEC plus leftward focus movement wrongly concludes that (i) the RNR target must occupy a subcategorized position, and (ii) there are no island effects in Chinese RNR. Besides, the motivation of the leftward focus movement in Chinese RNR is unknown.
100
Table 12 Summary of the two works on Chinese RNR
ApproachesProperties Cheng(2012) Wang (2014a, b)
1. Right Edge Restriction (RER)
2. Non-constituency ?
3. Coordinate Structure Constraint (CSC)
4. Asymmetric island sensitivity ? ?
5. Asymmetric P-stranding ? ?
6. Strict/sloppy reading ?
7. Asymmetric licensing of NPI ? ?
8. Licensing of relational modifiers ?
101
Chapter 4
Toward a PF-deletion approach to Mandarin Chinese RNR
Following Ha (2008a, b, c), we propose a PF-deletion analysis to account for the derivation of Chinese RNR. To license a well-formed Chinese RNR, the ellipsis feature of RNR (dubbed as ERNR by Ha) must attach to the contrastive lexical items in non-final conjuncts to get its feature checked with its head of C, and instruct the elements following it to be unpronounced at PF. Besides, the RNR clause and its antecedent must have a mutual entailment relationship. The current analysis is different from Ha in that he assumes that only the RNR target in the first conjunct can be unpronounced at PF, failing to capture the cases of more than two conjuncts (cf. Kluck and Zwart 2009). In addition, Ha‘s approach focuses on English RNR, which is insensitive to island conditions and allows P-stranding; on the other hand, Chinese exhibits asymmetric island effects and asymmetric P-stranding, which can be accounted for under our analysis. The organization of this chapter is as follows. Section 4.1 presents the empirical data against other approaches. Section 4.2 shows the evidence in favor of a PF-deletion account of Chinese RNR. Section 4.3 gives implications for the PF-deletion approach to English RNR, suggesting that the MD approach is also a possible way to derive English RNR. Finally section 4.4 offers a brief conclusion.
4.1 Against other approaches
Based on the asymmetric properties in Chinese RNR, we argue that such phenomena cannot be dealt with under the ATB-movement analysis, Kluck‘s (2007) external remerge, the MD approach, Barros and Vicente‘s (2011) eclectic approach, or Larson (2011) and his subsequent works on the sparse approach.
4.1.1 Against the ATB-movement approach
102
Recall Sabbagh‘s (2007) ATB-movement approach in which a RNR target must obey Rightward Crossing Constraint and Order Preservation. In other words, a RNR target cannot cross any overt phonological elements within the nodes where it originates, and the linear order of RNR can be derived as long as the already built ordering (i.e. fixed ordering after each Spell-Out Domain) is not violated by later Move and Merge. Accordingly, this approach may account for the following properties: Right Edge Restriction, non-constituency,
Coordinate Structure Constraint, and licensing of relational modifiers. The first two properties are captured since a legitimate RNR target is rightmost in each conjunct and no other pronounced elements are crossed, and thus the RNR target could be non-typical constituent of rightward movement. Since RNR is assumed as a type of ATB-movement, it follows Coordinate Structure Constraint and can account for the licensing of relational
modifiers at LF. For instance, the LF structure of (1) can be reconstructed as in (2), where the internal reading is obtained: the identical tune was hummed by Zhangsan and sang by Lisi.
(1)
Zhangsan heng er Lisi changxiangton de qudiao
Zhangsan hum and Lisi sing same DE tune‗Zhangsan hummed and Lisi sang the same tune.‘
(2)
[xiangtonde qudiao] [Zhangsan heng
er Lisi chang]same DE tune Zhangsan hum and Lisi sing
‗The same tune Zhangsan hummed and Lisi sang.‘
However, this approach fails to explain the grammatical contrast regarding the asymmetric phenomena. Consider asymmetric island effects first.
However, this approach fails to explain the grammatical contrast regarding the asymmetric phenomena. Consider asymmetric island effects first.