Some Notes on Sloppy Identity in Mandarin Sluicing
*Ting-Chi Wei
National Kaohsiung Normal University
Based on the three essential properties of sloppy identity, including c-commanding, lexical identity between a wh-correlate and a wh-remnant, and the na ‘that’ effect, the PF-deletion analysis and the pro sluice analysis will be carefully surveyed to see which one is more tenable in explaining the sloppy identity issues in Mandarin sluicing. It is found that only the former with a full-fledged syntactic representation can capture the essence of sloppy identity, except for the strict-reading-only na-effect. Comparatively, although the pro sluice analysis can easily explain the na-effect and can seemingly account for sloppy identity via the unselective binding and the construal of the E-type pronoun, some theoretical problems still arise in the face of the other two traits. Thus, we propose that sloppy identity is regulated by a stricter version of the syntactic requirement between two conjuncts and can be interpreted in virtue of a modified Dependency Theory (Fiengo and May 1994). That is, sloppy identity is one manifestation of the syntactic parallelism deduced from the fully articulated syntactic structure rather than a result of pro construal.
Keywords: sluicing, sloppy identity, strict identity, PF-deletion, pro sluice analysis
1. Introduction
Linguists who are interested in the issue of sloppy identity almost all pay their attention to the VP-ellipsis (Sag 1976, Williams 1977, Huang 1988a-b, 1991, Fiengo and May 1994, Hoji 1997a-b, 1998, 2003, Otani and Whitman 1991, Kim 1999, etc.). In contrast, very few research focuses on sluicing, except the most extensive studies on sluicing by Merchant (2001).1 Actually, early in Ross (1969), the so-called sloppy identity has already been preliminarily justified as in (1).
(1) I know how to say I’m sorry, and Bill knows how, too.
‘I know how to say I’m sorry, and Bill knows how to say I’m sorry.’ (strict)
‘I know how to say I’m sorry, and Bill knows how to say he (=Bill) is sorry.’ (sloppy)
*
I am very grateful to Wei-Tien Dylan Tsai, Jo-Wang Lin, Chen-Sheng Luther Liu, Audrey Li, Jason Merchant, the audience in TEAL-2 and FOSS-3 and the two anonymous reviewers for the valuable comments and discussions that have greatly improved the content of this paper. Any error is my own responsibility. This study was supported by NSC Grants from Taiwan #95-2411-H-017-003 and #97-2410-H-017-011.
1
Merchant (2001:8) describes that it is very hard to get sloppy reading in sluicing even with some variations on judgment. This view is opposite to that of Ross (1969).
After that, in favor of PF-deletion, Takahashi (1994:271-272) utilizes strict/sloppy identity in Japanese sluicing to argue against the base-generated pro copular analysis (Nishiyama, Whitman, and Yi 1996), which would fail to generate the sloppy reading within its simplex sluice structure, i.e. [pro/sore ‘it’ wh-remnant (copular)]. Later, the same structure is interpreted by Kuwabara (1997) as an underlying cleft structure, whose sloppy identity is blocked by the overt pronoun, which replaces a propositional phrase containing a bound variable. From another viewpoint, Fukaya and Hoji (1999) have observed that both case-marked and non-case-marked sluicing in Japanese can give rise to sloppy reading but involve different derivational processes.2 The former relies on LF-copying analysis (Chung, Ladusaw, and McCloskey 1995, hereafter CLM), while the latter on deep anaphora. The above description reveals that the issue of sloppy identity in Japanese sluicing is still under heated debate. As to Mandarin sluicing, Wang (2002) and Wang and Wu (2006) hold it as evidence to the IP-deletion approach but without providing further elaborations on the issue. On the other hand, under the framework of the pro sluice structure, i.e. [pro (copular) wh-remnant], Adams (2003) rejects the existence of sloppy identity, but Wei (2004) admits that sloppy identity does exist in Mandarin sluicing and further proposes that the derivation of strict and sloppy identity is attributed to the distinction between the nominal pro and the event pro. Only an adjunct wh-remnant anaphoric to event pro can generate the sloppy identity, while an argument wh-remnant referential to nominal
pro cannot. However, more new data show that this argument-adjunct asymmetry will
face an empirical problem.
We find that in addition to the c-commanding requirement (Ross 1967, Takahashi 1994), there is another cross-linguistic restriction on sloppy identity in sluicing; that is, it hinges on a “lexical” identity between the overt wh-correlate and wh-remnant, regardless of the types of wh-remnants involved, except for the case of “contrast sluice” (Merchant 2001:36) and the na ‘that’ effect (Wei 2004). The former allows sloppy identity even in lack of lexical identity between the wh-elements and the latter prohibits it despite its obedience to the lexical consistence. We will show that the “contrast sluice” is not a sluice per se; thus, it is not a counterexample to the identity between wh-elements for sloppy identity. As to the na-effect, it is still analyzed as a sluice due to the fact that na is freely present in any sluicing sentences without changes in meaning, except for its impact on the blockage of sloppy identity reading. The three properties, including c-commanding, identity between wh-words, and
na-effect, will be surveyed by the PF-deletion analysis (Takahashi 1994, Merchant
2001, Chung 2005 and Wang 2002) and the pro sluice analysis (Adams 2003 and Wei
2
The case-marker on the wh-remnant is optional in Japanese. Those with case-markers are called “C(ase)M(arked)-sluicing” and those without “Non-CM-sluicing”. The CM-sluicing requires a linguistic antecedent and respects subjacency, contrary to the non-CM-sluicing.
2004) to see which analysis is more tenable in approaching the essence of sloppy identity in Mandarin sluicing.3
The survey shows that the PF-deletion analysis can capture the characteristics of sloppy identity with respect to c-commanding and lexical identity apart from the
na-effect. In contrast, although the pro sluice analysis can easily explain the na-effect
and can seemingly account for sloppy identity via the unselective binding (Cheng and Huang 1996 and Lin 1996) and the E-type pronoun (Evans 1980), a theoretical problem still arises. Hence, we propose that the sloppy identity can be explained by the deletion analysis via a stricter version of the syntactic identity requirement between two conjuncts, requiring that every item in the sluice clause (including the
wh-remnant) must be identical to an item in the antecedent CP at the level of LF.
Meanwhile, we suggest that the sloppy identity can be interpreted via a modified Dependency Theory (Fiengo and May 1994), which makes use of a syntactic dependency relationship along with the two indexical types, α-occurrence and
β-occurrence, to regulate the reference of a pronoun in strict/sloppy identity. Finally,
we conclude that the sloppy identity is actually a reflection of the syntactic parallelism deduced from the fully articulated syntactic structure rather than a result of pro construal, which, however, may be more successful in explaining the na-effect and the distribution of shi. Therefore, the result of this study implies that sluicing in Mandarin cannot be uniformly dealt with by the same approach as the sloppy identity, since there are still some other issues left unanswered by the PF-deletion analysis, such as
na-effect, the distribution of shi ‘be’ and left-branching modifier. Since the scope of
this paper is limited to the implication of sloppy identity on the derivation of sluicing, we leave the other possibilities open.
The organization of this paper is as follows. Section 2 describes the essential properties of sloppy identity in sluicing. Section 3 surveys the PF-deletion analysis and the pro sluice analysis. In Section 4, we compare these two analyses and conclude this work.
3
A reviewer has raised a conceptual or methodological problem concerning the confusing mixture of the terminology, such as PF-deletion, LF-copying and LF-reconstruction, which had not been seriously distinguished in the previous manuscript. To solve this problem, the priority is to clarify the three terms. According to Takahashi (1994) and Tomioka (2008), in the PF-deletion analysis, before SPELL-OUT, a missing IP starts its life in the derivation with full-fledged structure, which later will be elided at PF but remain intact at LF. In contrast, the LF-copying analysis treats the missing IP to be truly empty at the beginning before SPELL-OUT but to be copied from its antecedent at LF (CLM 1995). That is, the two analyses differ in the structures before SPELL-OUT and the mechanisms such as deletion and copying, respectively. As to the LF-reconstruction, it is a term usually used in the LF-copying analyses to denote the process of recovery as in Williams (1977) and Chao (1987). Hence, we decided to strictly distinguish PF-deletion from LF-copying/LF-reconstruction to avoid terminological confusion. In this paper, the PF-deletion analysis is entertained to interpret the sloppy identity in Sluicing. Hence, there will be a fully articulated syntactic structure at the end of the derivation. As for the LF-copying analysis with regard to sloppy identity, I will leave it open for further research.
2. Sloppy identity in sluicing
The sloppy identity in Mandarin sluicing manifests three general properties: (1) c-commanding, (2) lexical identity between wh-words, and (3) na ‘that’ effect.
Ross (1967) has proposed that for an elided expression to have the sloppy identity, a pronoun relating to the reading must be c-commanded by its antecedent as in (2a); otherwise, the sloppy identity is voided as in (2b). Takahashi (1994:269) claims that sluicing in Japanese also obeys this restriction, indicating that the sluice clause in both languages contains a hierarchical structure just like the antecedent clause. Sluicing in Mandarin also observes this constraint as in (3).
(2) a. Johni knows why hei was scolded, and Mary knows why, too.
‘Johni knows why hei was scolded, and Mary knows why hei was scolded.’ (strict)
‘Johni knows why hei was scolded, and Maryj knows why shej was scolded.’ (sloppy)
b. John’s mother knows why he was scolded, and Mary’s mother knows why, too. ‘Johni’s mother knows why hei was scolded, and Mary’s mother knows why hei was scolded.’ (strict)
(3) a. Zhangsani bu zhidao [ tai weishenme bei ma], dan Lisij zhidao Zhangsan not know he why PASS scold but Lisi know ( shi ) weishenme.4
be why
‘Zhangsani didn’t know why hei was scolded, but Lisij knows why hei/j was scolded.’ (strict/sloppy)
b. [ Zhangsani-de muqin] zhidao [[ tai weishenme bei ma]], Zhangsan-POSS mother know he why PASS scold dan [ Lisij-de muqin] bu zhidao ( shi ) weishenme.
but Lisi-POSS mother not know be why
‘Zhangsani’s mother knows why hei was scolded, but Lisij’s mother does not know why hei/*j was scolded.’ (strict/*sloppy)
Second, we have observed that the derivation of sloppy identity requires “lexical” identity between the overt wh-correlate and wh-remnant, independent of argument-adjunct distinction. It seems that the matching requirement is a
4
Abbreviations: PASS = passive marker; POSS = possessive marker; Q = question marker; NOM = nominative marker; TOP = topic marker; GEN = genitive marker; ACC = accusative marker; PROG = progressive marker; CL = classifier.
cross-linguistic phenomenon. Take English sluicing for example. The wh-adjuncts in (1) and (2a) and the wh-arguments in English translations of (5) and (6) obey this restriction. The same situation also occurs in Japanese sluicing as in (4) and (5) concerning a wh-adjunct and a wh-argument, respectively, (Takahashi’s (1994) (11) and (12), in that order). Mandarin sluicing also manifests the same effect as in (3a) for
wh-adjunct identity and (6) for wh-argument identity.5 Apparently, the example (6) empirically challenges Wei’s (2004) claim that sluicing with a wh-argument cannot derive sloppy identity.
(4) John-wa [ zibun-ga naze sikarareta ka] wakattenai ga, John-TOP self-NOM why was scolded Q not knows but Mary-wa [ naze ka] wakateiru.6
Mary-TOP why Q knows
‘John doesn’t know why he was scolded, but Mary knows why.’ (strict/sloppy)
(5) a. UConn-ga [ soko-no basukettobooru tiinu-ga dare-o sukautosita ka] UConn-NOM it-GEN basketball team-NOM who-ACC scouted Q happyoosita.
announced
‘UConn announced who its basketball team scouted.’ b. Duke-mo [ dare-o ka] happyoosita.
Duke-also who-ACC Q announced ‘Duke announced who, too.’ (strict/sloppy)
(6) Zhangsani zhidao [ shei zai piping tai], dan Lisij bu zhidao shi shei.
Zhangsan know who PROG criticize him but Lisi not know be who ‘Zhangsan knows who is criticizing him, but Lisij doesn’t know who.’
(strict/sloppy)
In line with this lexical identity restriction, we can predict that the derivation of the sloppy reading requires the wh-antecedent to be overtly present; otherwise, only the strict reading is allowed. In (7) and (8), the wh-remnants, why and weishenme ‘why’,
5
Hoji (1990) claims that lexical pronouns in Japanese such as kare ‘he’ and kanojo ‘she’ are not bound pronouns and do not permit sloppy reading in the elliptical structures. Essentially, Takahashi (1994) agrees with this view but he further demonstrates that the anaphor such as zibun ‘self’ and the pronominal or deictic expressions such as sore ‘it’, soko ‘there’, soitu ‘that guy’ permit sloppy identity. There is no such distinction in Mandarin.
6
Takahashi (1994) makes use of the existence of sloppy identity to assimilate Japanese sluicing to English sluicing in terms of IP-deletion account. For other views against this, see Fukaya and Hoji (1999) and Kuwabara (1997).
have merely matrix or embedded strict reading in lack of a sloppy reading when there are no corresponding wh-correlates.
(7) John does not know he was scolded, but Mary know why. (strict)
(8) Zhangsani bu zhidao [ tai bei ma], dan Lisij zhidao ( shi ) weishenme. Zhangsan not know he PASS scold but Lisi know be why
‘Zhangsani didn’t know that hei was scolded, but Lisij knows why Zhangsani was scolded.’ (Embedded strict)
‘Zhangsani didn’t know that hei was scolded, but Lisij knows why Zhangsani does not know that hei was scolded.’ (Matrix strict)
Note that the strict lexical parallelism of sloppy identity will confine the interpretation of wh-remnant within the range of embedded clause. For instance, in (2a) and (3a), the
wh-remnants only target the embedded scope of strict or sloppy reading instead of the
matrix reading. Without wh-correlates, the matrix reading is permissible as in (7) and (8). It indicates that the range of strict/sloppy identity is syntactically determined by the existence of the wh-correlates.
It follows that once the lexical or categorial parallelism of the wh-remnant is lost between two conjuncts, the sloppy reading disappears as well. In (9a) and (10a), the grammatical categories of the two wh-remnants are different; in this case, only the strict identity is generated. In (9b) and (10b, c), though the grammatical categories of the wh-sluices are similar, they actually differ in lexical form and semantic function. This disparity also suppresses the emergence of sloppy identity.
(9) a. John knows what he has done, but Mary wants to know why. (strict) b. John knows why he was scolded, and Mary wants to know when. (strict)
(10) a. Zhangsani zhidao [ shei zai piping tai], dan Lisij bu zhidao ( shi ) Zhangsan know who PROG criticize him but Lisi not know be
weishenme.
why
b. Zhangsani zhidao [tai weishenme bei ma], dan Lisij bu zhidao ( shi ) Zhangsan know he why PASS scold but Lisi not know be
( wei-le-)-shenme liyou.
for-ASP-what reason
‘Zhangsani knows why hei was scolded, but Lisij doesn’t know for what reason hei/*j was scolded.’ (strict)
c. Zhangsani zhidao tai yinggai nian na-yi-ben-shu, dan Lisij bu Zhangsan know him should read which-one-CL-book but Lisi not zhidao ( shi ) sheide shu.
know be whose book
‘Zhangsani knows which book hei should read, but Lisij doesn’t know whose book hei/*j should read.’ (strict)
However, we do find some counterexamples which show sloppy identity even if the lexical identity of the wh-remnant is violated as shown in (11), a construction which is called “contrast sluice” in Merchant (2001:36). The same phenomenon also happens in English as illustrated in the English translation.
(11) Zhangsan zhidao ta yinggai nian na-yi-ben-xiaoshuo, Lisi zhidao ( shi ) Zhangsan know he should read which-one-CL-novel Lisi know be
na-yi-ben-zazhi.
which-one-CL-magazine
‘Zhangsani knows which novel hei should read, and Lisij knows which magazine (hei/j should read).’ (strict/sloppy)
We find that the “contrast sluice” is different from the typical sluicing in several ways. First, the wh-words of the typical sluicing with sloppy identity generally require the same lexico-syntactic forms, whereas those of the “contrast sluice” basically own the different ones. For example in (6), the two shei’s ‘who’ have the identical forms, questioning “the identity of the person who criticizes him”. Similarly, even in (3a), with two identical wh-adjunct weishenme’s ‘why’, the wh-words in question actually refer to the causes of the event, ta bei ma ‘he is scolded’. By contrast, in (11), the two nominal wh-words manifest the different forms; one is xiaoshuo ‘novel’ and the other is zazhi ‘magazine’. Further, from the notion of contrastive focus, the divergence in the construal of the wh-words indicates that there exist two contrasts in (11). One is the contrast between Zhangsan and Lisi, and the other is the contrast between
na-yi-ben-xiaoshuo ‘which novel’ and na-yi-ben-zazhi ‘which magazine’. The middle
‘know’ left. In that sense, the lexical mismatch between the wh-words constitutes a natural consequence of contrast. But there is no such contrast in typical sluicing. Given that the “contrast sluice” is actually not a sluicing at all, it implies that any sloppy identity condition on sluicing will not apply to it; that is, the issue is not our concern here. If that is so, the lexical identity of the wh-elements for sloppy identity is still maintained in our analysis.7
7
A reviewer wonders whether the example such as (i) is a kind of “contrast sluice” with a contrast on
zenmeyang ‘how’ and weishenme ‘why’.
(i) ?? Zhangsan bu zhidao [ ta zenmeyang qu Taipei], dan Lisi zhidao [weishenme ta qu Taipei]. Zhangsan not know he how go Taipei but Lisi know why he go Taipei
‘Zhangsan doesn’t know how he goes to Taipei, but Lisi knows why he goes to Taipei.’
In fact, Merchant (2001:36-7) defines “contrast sluice” as a structure containing the focused
wh-nominals in the argument positions, which are analyzed as variables, making possible the
satisfaction of the entailment condition as shown in (ii). (Readers can refer to Note 13 for the definition of Merchant’s (2001) Focus entailment condition on IP-ellipsis.)
(ii) a. She has [five CATS]F, but I don’t know how many DOGS [IP she has t].
b. IPE’=∃x.she has x, F-clo(IPA)=∃x.she has x.
In other words, Merchant only concerns the issue of entailment, which is jeopardized by the unparallel contrasts, even if he finds a way out of this dilemma via focus. In fact, he is not concerned with the issue whether wh-adjuncts can be focalized. We will try to explain (i) from Merchant’s analysis.
The sentence in (i) is ambiguous with three possible readings as shown in (iii).
(iii) a. Zhangsan doesn’t know how he goes to Taipei, but Lisi knows why Zhangsan does not know how he goes to Taipei.
b. *Zhangsan doesn’t know how he goes to Taipei, but Lisi knows why Zhangsan goes to Taipei how.
c. ??Zhangsan doesn’t know how he goes to Taipei, but Lisi knows why Zhangsan goes to Taipei.
As a typical sluice, the reading (iiia) is more preferable than (iiib), which is ruled out due to the illicit multiple wh-adjuncts within the embedded elided sluice. The “contrast sluice” reading (iiic) is prohibited, mostly because it violates the entailment condition, which is essential to the licensing of sluicing, as illustrated in (iv). When the two wh-adjuncts are focalized in (iva), both clauses fail to entail each other as in (ivb), in contrast to (iib).
(iv) a. Zhangsan bu zhidao [[zenmeyang]F ta qu Taipei], dan Lisi zhidao[[weishenme]F ta qu Taipei].
b. IPE’=∃x.ta qu Taipei by x, F-clo(IPA)=∃x.ta qu Taipei for x.
In fact, Merchant (p.c.) does not relate “contrast sluice” to the issues of sloppy identity, since there is no sloppy identity in English sluicing in his data. In this work, we claim that the entailment condition is only essential to the licensing of sluicing, but not to the derivation of sloppy identity in sluicing, which needs stricter conditions. Hence, the sloppy identity in the non-sluice (11) is not within the scope of this study.
The reviewer has also pointed out that the strict-reading-only (10c) or (21b) will pose a problem to the interpretation of contrast sluice, especially when na-yi-ben shu ‘which book’ and sheide-shu ‘whose book’ may refer to different entities in the respective argument position. Concerning this problem, I agree with the reviewer’s suggestion that the issue of “entity” or “event” should be irrelevant to the discrimination of “contrast sluice” from “typical sluice”. Along this vein, the lack of sloppy identity in (10c) and (21b) is attributed to their non-identical lexico-syntactic forms and their non-contrastive nature of the wh-words (which book/whose book vs. which novel/which magazine in
Third, we further observe that when na ‘that’ or the definite description
na-(NumP/ClP) occurs in front of the wh-remnant, the strict reading is the only option
even if the matching requirement is met. For instance, in (12a), only the embedded strict reading is licit in the case of an argument wh-remnant. In (12b), the scope of the strict reading can be embedded or matrix especially when the wh-remnant is an adjunct. It is worth noting that the copular shi ‘be’ after na ‘that’ is obligatorily required, being different from the optional shi in sluicing as in (10).8 That is, the presence of the demonstrative na will block the sloppy reading. The reason why we still consider (12) as a sluice is due to the fact that na can be added to almost any sluicing sentences without change in meaning, except for the blockage of sloppy identity.
(12) a. Zhangsan zhidao ta yinggai nian na-yi-ben-shu, dan Lisi bu Zhangsan know him should read which-one-CL-book but Lisi not
zhidao na ( yi-ben-shu ) *( shi ) na-yi-ben. know that one-CL-book be which-one-CL
‘Zhangsani knows which book hei should read, but Lisij does not know which book hei/*j should read.’ (strict/*sloppy)
b. Zhangsan bu zhidao ta weishenme bei ma, dan Lisi zhidao Zhangsan not know he why PASS scold but Lisi know na ( ge-yuanyin) *( shi ) weishenme.
that CL-reason be why
‘Zhangsani didn’t know why hei was scolded, but Lisij knows why hei/*j was scolded.’ (Embedded strict/*sloppy)
‘Zhangsani didn’t know why hei was scolded, but Lisij knows why Zhangsani didn’t know why hei is scolded.’ (Matrix strict/*sloppy)
In fact, similar phenomenon also occurs in Japanese sluicing. As noted by Takahashi (1994: 271-2), when the sluice clause contains the pronoun sore “it” and (optionally) the copula, the sloppy reading is not permitted. Thus, (13) has only strict reading: Mary knows why John was scolded.9
(11)). In addition, it is possible that the wh-words of a contrast sluice may undergo focus movement by SPELL-OUT just as a typical sluice does. Since they involve contrastive focuses, lacking in typical sluices, we will not take them into account in this paper.
8
In Mandarin sluicing, the copula shi ‘be’ is optional in front of the wh-remnants other than shei ‘who’ and shenme ‘what’. The latter two require its presence.
9
This paper attempts to consider sloppy identity a common phenomenon across languages such as Mandarin, English, and Japanese. Due to the limit of space, we will only focus on Mandarin and the latter two are left for future research.
(13) John-wa [ zibun-ga naze sikarareta ka] wakattenai ga, John-TOP self-NOM why was-scolded Q not knows but Mary-wa [sore-ga naze ( dearu ) ka] wakateiru.
Mary-TOP it-NOM why is Q knows
‘John doesn’t know why he was scolded, but Mary knows why it is.’10
On the basis of these three properties, we will survey two potential analyses of sluicing, the PF-deletion analysis and the pro sluice analysis, to see which one is more tenable. The former is characterized by its fully articulated syntactic structure before SPELL-OUT and at LF and the latter by its simple sluice clause at all levels.
3. Two analyses on the sloppy identity of Mandarin sluicing
3.1 The PF-deletion analysis
The PF-deletion approach proposed by Takahashi (1994) presupposes that the full-fledged representation of the elided IP structure at LF will naturally explain the strict/sloppy identity in English and Japanese sluicing. Even so, it is a pity that he has not explored this issue further. Likewise, as to Mandarin sluicing, the deletion approach adopted by Wang (2002) and Wang and Wu (2006) does not provide any explanation of the issue in question.11 Below, the three properties of sloppy identity
10
Merchant (2001:120-121) shows that sluicing in English is not equal to “pseudosluicing” for three reasons. First, sluicing can have adjunct remnants and implicit argument remnants, but pseudosluicing cannot as in (i).
(i) a. He fixed the car, but I don’t know how/why/when (*it was). b. They served the guests, but I don’t know what (*it was).
Second, sluicing requires a greatest pitch accent on wh-remnant, whereas pseudosluicing on the copula.
(ii) a. Someone gave me a valentine, but I don’t know WHO.
b. Someone KISSED you, and you don’t remember who it WAS?!?
Third, sluicing does not permit non-D-linked wh-phrases, but pseudosluicing does. (iii) Someone dented my car last night—
a. I wish I knew who (*the hell)! b. I wish I knew who the hell it was!
To argue against this trend, Nishiyama, Whitman, and Yi (1996) claim that when a demonstrative (e.g.
that) is in place of it, the sentences become legitimate. In this paper, I will take the copular analysis
of Mandarin sluicing into account.
(iv) John left, but I don’t know why (that is).
11
in Mandarin sluicing will be checked one by one under the IP-deletion analysis to see if this approach is tenable.12
The c-commanding property, requiring that a pronoun relating to the sloppy identity must be c-commanded by its antecedent, is in conformity with the real essence of the IP-deletion analysis. It implies that only when the elided material after
wh-remnant is fully articulated before SPELL-OUT or at LF can the c-commanding
requirement be satisfied. In other words, the derivation of the sloppy reading hinges on a full-fledged hierarchical structure.
The lexical identity property points to the fact that the sloppy identity must be derived under the circumstance that the overt wh-correlate and wh-remnant is lexically identical. In fact, in terms of parallelism, the PF-deletion analysis requires that every lexical item in the elided IP must be identical to an item in the correlate clause. This idea has been elaborated by Chung (2005) from the Minimalist notion of numeration. Chung argues that the deletion cannot be “merely” constrained by semantic (entailment) conditions alone (Merchant 2001), requiring the nonfocused portions of the antecedent IP and the elided IP to entail each other, mostly because they fail to rule out the prepositional stranding in sluicing as follows.13
12
One of the reviewers has pointed out a serious problem concerning the inconsistent judgment of sloppy identity in English Sluicing between Ross (1969) and Merchant (2001, 2008). Merchant makes use of the comparative ungrammaticality in English Sluicing as in (ib) to argue that VP-ellipsis and Sluicing obey an economy constraint, called the MaxElide, regulating that “if ellipsis targets an XP containing an A’-trace, XP must not be properly contained in any YP that is a possible target for deletion”. Hence, that is the reason why VP-ellipsis in (ia) is licit, while Sluicing in (ib) is unacceptable.
(i) a. Ben knows who he invited, but Charlie doesn’t. (strict/sloppy)
b. ??Ben knows who he invited, but Charlie doesn’t know who. (strict/sloppy)
However, according to Ross’s (1969) judgment, such a difference does not exist. Even though we have found that Mandarin does not have such a judgment contrast between VP-ellipsis and Sluicing (like Ross’s) and does not obey the MaxElide, yet in Mandarin VP-ellipsis, the existence of the
wh-correlate plays no role in the derivation of the sloppy identity, unlike that of Sluicing. Thus, we
conclude that sloppy identity in Sluicing has nothing to do with that in VP-ellipsis. Since the argument of this scenario is based on a set of disputable data and there exist essential differences between the two structures, to avoid any confusion and the weakening of the main argument, we delete it from the text.
13
Merchant’s (2001) Focus entailment condition on IP-ellipsis is based on the definition of e-GIVEN in (i) and is stated in (ii).
(i) e-GIVENness
An expression E counts as e-GIVEN iff E has a salient antecedent A and, module∃-type shifting,
a. A entails F-clo(E), and b. E entails F-clo(A).
(ii) Focus condition on IP-ellipsis
An IP α can be deleted only if α is e-GIVEN.
Take (iii) for example, in which both the correlate clause and sluice clause are mutually entailed as illustrated in (iv). Therefore, the IP-deletion with sluicing is licit in (iii).
(14) a. They’re jealous, but it’s unclear of who. b. *They’re jealous, but it’s unclear who.
Chung (2005) claims that “the choice between pied-piping and preposition stranding is not normally thought to have semantic (true-conditional) consequences” (2005:10). Hence, the sluicing examples in the pairs such as (14a) and (14b) ought to be “semantically equivalent”. That is to say, under Merchant’s semantic conditions, (14b) should be legal as (14a). However, this is not borne out. To compensate for this weakness, in addition to Merchant’s entailment condition, she proposes an extra lexico-syntactic requirement as in (15), demanding that “except for the moved interrogative phrase, the lexical items from which the sluice is constructed must be a subset of the lexical items from which the antecedent CP is constructed” and ensuring that “the ellipsis in sluicing involves no “return to the lexicon”” (2005:11). Let’s see how it works on (14). Given the PF-deletion approach, at the end of the syntactic derivation, (14) will be represented as in (16).
(15) Every lexical item in the numeration of the sluice that ends up (only) in the elided IP must be identical to an item in the numeration of the antecedent CP.
(16) a. They’re jealous, but it’s unclear [of who [they’re jealous --]]. b. *They’re jealous, but it’s unclear [who [they’re jealous of --]].
In (16a), the items in the numeration of the sluice includes of, who, they, be, and
jealous, three of which end up in the elided IP (they, be, jealous) and are each
identical to an item in the numeration of the antecedent CP. Thus, it is licit. In contrast, in (16b), there is an item of, which ends up (only) in the elided IP, but which fails to be identical to an item in the numeration of the antecedent CP. Hence, it is out.
Remember that what Chung’s lexico-syntactic requirement in (15) and Merchant’s (2001) entailment conditions on ellipsis regulate is the legitimacy of sluicing, not the licensing of sloppy identity. For instance, (3a) and (8), as reduplicated together in (17), are predicted to be licit, since they satisfy not only (15) but also the entailment condition. In (17a), at the end of the syntactic derivation after overt focus movement,
(iii) She loves someone, but I don’t know who.
(iv) a. IPA’=∃x [she loves x ], F-clo(IPE)=∃x [she loves x]
b. IPE’ = ∃x [she loves x ], F-clo(IPA)=∃x [she loves x]
In fact, Chung (2005) adopts Romero’s (1998) idea that the antecedent must entail the nonfocused portion of the reduced constituent. That is, she only keeps one half of Merchant’s mutual entailment condition but without giving convincing reasons. In this work, we still use Merchant’s definitions.
whether the wh-correlate weishenme ‘why’ exists or not, every lexical item in the numeration of the sluice that ends up (only) in the elided IP, inclusive of ta ‘he’, bei ‘PASS’, and ma ‘scold’, is identical to an item in the numeration of the antecedent CP. On the other, at the level of LF, as shown in (17b), the antecedent IP entails the elided IP, and vice versa as formulated in (18a, b). It implies that the lexico-syntactic requirement and entailment conditions are only responsible for the licensing of sluicing and that the derivation of sloppy identity in (3a) needs more restrictive constraints to account for the lexical identity property. Otherwise, (3a) and (8) will be undistinguishable.
(17) a. Zhangsan bu zhidao [ta (weishenmei) bei ma], dan Lisi zhidao (shi) weishenmei [ta -- bei ma].14
b. Zhangsan bu zhidao (weishenmei) [ta (ti) bei ma], dan Lisi zhidao (shi) weishenmei [ta ti bei ma].
(18) a. IPA’=∃x[ta x bei ma], F-clo(IPE)=∃x[ta x bei ma] b. IPE’ =∃x[ta x bei ma], F-clo(IPA)=∃x[ta x bei ma]
Rooted in the lexical identity between the wh-correlate and wh-remnant, we propose that sloppy identity relies on a more refined version of identity requirement than (15); that is, every lexical item including the wh-remnant in the numeration of the sluice clause must be identical to an item in the antecedent CP as in (19).
(19) Every item in the numeration of the sluice clause (including the wh-remnant) must be identical to an item in the numeration of the antecedent CP.
The main difference between (15) and (19) lies in the domain of identity. The former is only within the elided IP, not including the wh-remnant, while the latter covers the whole sluice CP clause.
Let’s first check the typical sluicing examples with sloppy identity in (3a) and (6) to see if (19) works. As repeated in (20a, b), every item in the sluice clause including the wh-remnants has a parallel counterpart in the correlate clause. In addition, it seems that (19) can also predict that the examples in (10), as reproduced in (21), lack sloppy identity.
(20) a. Zhangsan bu zhidao [ta weishenme bei ma], dan Lisi zhidao (shi)
weishenmei [ta ti bei ma].
14
b. Zhangsanzhidao [shei zai piping ta], dan Lisi bu zhidao shi sheii [ti zai piping ta].
(21) a. Zhangsani zhidao [ta weishenme bei ma], dan Lisi bu zhidao (shi)
(wei-le)-shenmeliyouj [ta tj bei ma].
‘Zhangsani knows why hei was scolded, but Lisij doesn’t know for what reason Zhangsani was scolded.’ (strict)
b. Zhangsan zhidao [ta yinggai nian na-yi-ben-shui], dan Lisi bu zhidao (shi)
sheide shui [IP ta yinggai nian ti].
‘Zhangsan knows which book he should read, but Lisi doesn’t know whose book.’ (strict)
c. Zhangsanzhidao [sheii zai piping ta], dan Lisi bu zhidao (shi) weishenmej [ti
tj zai piping ta].
‘Zhangsan knows who is criticizing him, but Lisi doesn’t know why.’ (strict)
In (21a), the wh-remnant is distinctive from its corresponding wh-word in lexical form. It is this lexical mismatch that fails to derive the sloppy identity, even though the entailment conditions are satisfied. By the same token, the divergence in form also causes the two wh-words in (21b) to be unable to derive the sloppy identity, despite the satisfaction of entailment conditions. In (21c), the loss of identity between the
wh-remnant and the wh-correlate violates (19). In fact, in this case, the wh-trace left
by shei ‘who’ is actually unbound in the second conjunct. Generally, the unbound trace would lead to ungrammaticality. However, it is still licit. To solve this problem, Merchant (2001:200-208) makes use of Fiengo and May’s (1994) “vehicle change” to explain the similar phenomenon in English by analyzing the wh-trace (t1) in the second conjunct in (22a) as an E-type Pronouns it in (22b). He assumes that the E-type pronoun can be translated as a variable. Given this, both the variable and the E-type pronoun are realized by the same rule as g(2) within entailment conditions as shown in (22c-d). Thus, the Focus condition is satisfied. By the same token, we propose that in (21c) the wh-trace (tj) anaphoric to the wh-correlate shei ‘who’ is also an E-type pronoun and that the sentence satisfies the Focus condition as well.
(22) a. The report details [CP what1 [IP IBM did t1]] and [CP why2 [IP IBM did t1 t2 ]. b. The report details what1 IBM did and why IBM did it1.
c. IPE’ = F-clo(IPE)= IBM did g(2) d. IPA’= F-clo(IPA)= IBM did g(2)
Merchant himself (2001:207) has noted that his Focus condition is defined on entailments, not on structures: “the set quantified over by the wh-phrase (its trace) should be the set picked out by the donkey pronoun in the deleted IP”. Along this line, he further concludes that the Focus condition is insensitive to the distinctions between “regular” pronouns and donkey pronouns. From these statements, the reason why (21c) is licit despite its lack of the sloppy identity reading becomes transparent. It is because it satisfies the Focus condition but does not meet the requirement of sloppy identity. Thus, we can deduce the fact that the sloppy identity requires another stricter identity requirement, which should syntactically discriminate the E-type pronoun from the
wh-correlate shei ‘who’. It can be concluded that vehicle change, just like
lexico-syntactic requirement (15) and Merchant’s entailment condition, may be closely related to the licensing of sluicing but is not essential to the emergence of sloppy identity, which requires a stricter lexical parallelism, as will be specified below.
A reviewer kindly pointed out that (19) is not a sufficient condition that can satisfactorily license the sloppy reading, since the example such as (23) may pose a serious problem to it. For the pronominal ta ‘him’ in the sluice clause is not identical to the counterpart wo ‘I’ in the antecedent CP, which is obviously against (19) and is predicted to be short of the sloppy identity. But, contrary to the fact, the sloppy reading does exist.
(23) Wo bu zhidao wo weishenme bei ma, danshi Lisi zhidao weishenme I not know I why PASS scold but Lisi know why ( ta bei ma).
he PASS scold
‘I don’t know why I was scolded, but Lisi knows why he was scolded.’ (sloppy)
I propose that this dilemma can be resolved by specifying the nature of the 1st person pronouns and further refining (19). Déchaine and Wiltschko (2002) have argued that the 1st and 2nd person pronouns belong to a different syntactic category from the 3rd person pronouns in English. Their main empirical argument lies in the claim that English 1st and 2nd person pronouns cannot be used as bound variables to admit a sloppy identity reading in VP-ellipsis sentences, unlike the 3rd person pronouns as shown in (24). Nevertheless, Rullmann (2004) is skeptical of this claim and gives empirical evidence to support the fact that the variable use of sloppy reading is indeed possible in English as in (25).
(25) a. I got a question I understood, but John didn’t. (strict/sloppy) b. I hope that I will win, but of course you do too. (strict/sloppy) c. I know that John saw me but Mary does too. (strict/sloppy)
Furthermore, the 1st and 2nd person pronouns as bound variables can be strengthened by the example given by Kratzer (1998) in (26), with the proposition that the speaker of the sentence is the only person who has the property λx [x got a question that x understood]. In addition, Rullmann (2004) asserts that the plural pronouns in the bound variable reading in (27a) and (28a) represent a variable ranging over individuals rather than pluralities, which are impossible for a logical reason in (27b) or owing to world knowledge in (28b), respectively.
(26) Only I got a question that I understood.
(27) a. We each/all think we’re the smartest person in the world. b. #We’re the smartest person in the world.
(28) a. Al and I both believed we were going to be elected president. b. #We were going to be elected president.
From the above argument, we propose that the bound variable use of the 1st pronoun can be applied to the wo ‘I’ in (23). That is to say, the wo ‘I’ and the 3rd pronoun ta ‘he’ in the elided clause are both used as variables at the level of LF. These variables can be bound by the nearest NP subjects to derive sloppy identity as shown in (29).
(29) … [weishenmei [x ti bei ma]], … [weishenmei [x ti bei ma]]
It means that the requirement of the purely lexical identity in licensing sloppy identity as depicted in (19) has to be modified to cover the identity of the every item, covert or overt, between the correlate clause and the sluice clause at the level of LF as in (30).
(30) Every item in the sluice clause (including the wh-remnant) must be identical to an item in the antecedent CP at the level of LF.15
15
A reviewer casts doubt on how the lexico-syntactic requirement is satisfied at LF without violating the Interpretability principle on the conditions that after SPELL-OUT, the phonetic features are illegitimate at LF and only the semantic features interpretable at LF remain at LF. Obviously, the reviewer implies that the revised lexico-syntactic requirement in (30) may directly involve the phonetic features and should be dealt with at PF in one way or another. In this paper, we propose that the items with lexical identity are interpreted at the semantic interface, since the LF parallelism
The advantages of this revision are as follows. First, it still retains the ability to explain the deficiency of sloppy identity in (21). In (21a, b), it is the lexical inconsistency between the wh-elements in question that violates (30), even though the other members are the same. As to (21c), it is the differences in the wh-elements and the number of variables between the two conjuncts that cause the violation of (30) as illustrated in (31).
(31) … [sheii [ti zai piping x]], … [weishenmej [E-type proi tj zai piping x]]
Second, (30) can explain why the example in (32) merely contains the strict reading only, which is also raised by one of the reviewers.
(32) Lao Lii bu zhidao [tak weishenme bei ma], danshi Lao Wangj zhidao old Li not know he why PASS scold but old Wang know [ weishenme tak/*i/*j bei ma].
why he PASS scold
‘Old Lii doesn’t know why hek was scolded, but Old Wangj know why hek/*i/*j was scolded.’ (strict)
It follows that (32) confirms to the requirement of (19); therefore, it is predicted to obtain both a strict and a sloppy reading; however, this prediction is not borne out, for (32) denotes only the strict reading. The problem lies in the fact that the pronoun ta ‘he’ here refers to the person other than Lao Li ‘Old Li’ and Lao Wang ‘Old Wang.’ When this kind of discourse “external” reference occurs, it is impossible to have the typical sentence-internal strict or sloppy identity. Here, it reveals that the index of the pronoun seems to play a role in determining the presence or absence of the sloppy identity but (19) fails to predict this distribution in (32). As a matter of fact, we will show that it is the status of pronoun that affects the emergence of sloppy reading rather than the index of pronoun. The latter is a result of the former. We propose that the licensing of sloppy identity in (30) should be adopted instead. Accordingly, in line with Rullmann’s (2004) pronominal analysis, the status of pronoun ta ‘he’ can be either a variable or a non-variable deictic pronoun in the logical form. In the case of the variable use of the pronoun, the condition (30) is met as (29) has illustrated; as a result, the sloppy reading is derived. As to the deictic use of the pronoun, when the pronoun ta ‘he’ internally refers to the same person of the correlate pronoun, Lao Li, of (30) plays crucial roles in deriving the sloppy identity. In other words, the lexical dependency (Fiengo and May 1994), as will be elaborated below, is interpretable at LF. As to the phonetic interface, it is responsible for the deletion of IP (Merchant 2001). Therefore, the lexico-syntactic requirement will not pose a theoretical or conceptual problem to this analysis.
the internal strict reading is derived. On the other hand, both of them may externally refer to the person other than Lao Li and Lao Wang. In this case, the external strict reading is generated. In other words, by discriminating status of the pronouns at the level of LF, the problem posed by (32) can be explained.
So far, it is certain that the satisfaction of (30) ensures the possibility of sloppy identity, but how is it specifically interpreted?16
Suppose that sloppy identity hinges on the LF parallelism of (30), the sluice structure will be identical to its corresponding clause when the bound variable use of pronoun is at work. Below, we propose that a modified Dependency Theory (Fiengo and May 1994) can be used to explain this phenomenon. The basic notions of Dependency Theory are as follows: (i) An occurrence of an index may be independent of other occurrences or dependent on another. The former is called α-occurrence, and the latter β-occurrence. Hence, pronouns anaphoric on elements outside their phrase markers (the second conjuncts) must bear independent α-occurrences; in contrast,
β-occurrences cannot be structurally resolved outside their structures (the second
conjuncts). (ii) An indexical dependency ID is any triple consisting of a sequence of elements, an index, and a structural description: <(cα1, cβ2, …, cβn), I, SD> and (iii) indexical dependencies ID and ID’ are i-copies iff ID and ID’ vary from each other in at most the value of I.17
From the above definitions, it follows that the sloppy identity can be derived only when the dependent β-occurrences are licensed in virtue of the identical syntactic relation holding between the elements bearing β-occurrences and their antecedents. Within this paper, the notion of (i) and (ii) will be adopted, while that of (iii) will be merged with (30) to fit the unique properties of sloppy identity in sluicing. Take the
16
To derive strict/sloppy identity, Sag (1976) and Williams (1977) adopt Partee’s (1973) Derived VP Rule, in which VP is turned into a lambda notation as in (i). The Variable Rewriting Rule in (ic) makes the sloppy identity possible in (id) by variable binding. In contrast, without the step (ic), the Derived VP rule in (ib) will directly derive strict identity via VP rule in (id) by pronominal reference.
(i) a. John visits his children on Sunday and Bill does too.
b. Derived VP Rule: John [λx(x visits his children)] and Bill does too. c. Variable Rewritting Rule: John [λx(x visits x’s children)] and Bill does too.
d. VP Rule: John [λx(x visits x’s children)] and Bill does [λx(x visits x’s children)] too.
Since all these analyses are mainly designed for VP-ellipsis and the literature lacks similar analyses on Sluicing, we will try to approach the sloppy interpretation from another direction.
17
An anonymous reviewer has pointed out that Fiengo and May’s (1994) Dependency Theory adopts the reconstruction analysis at the level of LF. In fact, Fiengo and May analyze the elided VP phrases by using reconstruction just like Williams (1977) and Chao (1987). But their notion of reconstruction does not in itself imply any notion of antecedence in terms of “deletion under identity” or “copying” as those of Williams’ and Chao’s. In their terms (Fiengo and May 1994:192), “reconstruction is an identity condition within the theory of structural representation; it defines what are occurrences of the same phrase marker.” That is to say, it is a notion of syntactic dependency relationship, as will be elaborated below.
sloppy identity in (33) for example. The structural description between the antecedent
Zhangsan and the pronoun ta ‘he’ in the first conjunct is represented as <NP, V, C,
NP> in (33a), since the wh-word weishenme ‘why’ will move to C position at LF. Obviously, (33a) is identical to that of the second conjunct in (33b) under the full-fledged LF structure of the deletion approach. Further, the indexical dependencies ID and ID’ in (33a, b) are considered to be i-copies, since they vary from each other in at most the value of I (1, 2). At this stage, we speculate that to derive sloppy identity in sluicing, the syntactic-identity requirement in (30) has to be triggered to ensure the syntactic identity of each member between the sluice clause and the antecedent clause as listed in (33c). As a consequence, the licensing of β-occurrence for sloppy identity requires a stricter syntactic dependency at LF. In (33), the modified dependency relationship is licensed to derive the sloppy reading via the indexical type
β-occurrences on the bound pronouns ta ‘he’, which requires its indexical value to be
anaphoric on either the antecedents Zhangsan or Lisi in the same conjunct, respectively.18
18
A reviewer has pointed out that the modified Dependency Theory cannot guarantee the correct sloppy reading in (i). In addition, it is also unclear why the pronoun ta ‘him’ in the second conjunct cannot be bound by the embedded subject Lisi. We suggest that even if the pronoun occurs at the position of object in (i), the sloppy reading can still be derived as follows. The structural description between the antecedent Zhangsan and the target pronoun ta ‘he’ in the first conjunct is represented as <NP, V, C, NP, V, NP> in (ia), which is identical to that of (ib). ID and ID’ in (ia, b) are considered as i-copies, since they vary from each other in at most the value of I (1, 2). Finally, to derive sloppy identity in sluicing, the syntactic-identity requirement in (30) has to be triggered to ensure the syntactic identity of each member between the sluice clause and the antecedent clause (ic). As a result, the modified dependency relationship is licensed to derive the sloppy reading via the indexical type β-occurrences on the bound pronouns ta ‘he’, which requires its indexical value to be anaphoric to either the matrix subjects Zhangsan or Wangwu, respectively. However, the pronoun with β-occurrence cannot refer to the embedded subject Lisi, since in Fiengo and May’s (1994) model, it is still a pronoun, not a variable, even though in a sense, the indexical type β-occurrence is just like a bound variable use (cf. Sag (1976) and Williams (1977)). Thus, to avoid violating Binding Principle B, it cannot be bound with the embedded subject Lisi.
(i) Zhangsani bu zhidao Lisij weishenme piping tai/*j, dan Wangwuk zhidao weishenme Lisij
Zhangsan not know Lisi why criticize him but Wangwu know why Lisi piping tak/*j.
criticize him
‘Zhangsani didn’t know why Lisij criticized himi/*j, but Wangwuk knew why Lisij criticized
himk/*j.’
a. ID = <(Zhangsan, ta), 1, <NP, V, C, NP, V, NP>> b. ID’ = <(Wangwu, ta), 2, <NP, V, C, NP, V, NP>> c. The sloppy identity condition (30) is satisfied.
(33) Zhangsan1 bu zhidao [taβ1 weishenme bei ma], dan Lisi2 zhidao [(shi)
weishenmej [taβ2 tj bei ma]]. (sloppy)
a. ID = <(Zhangsan, ta), 1, <NP, V, C, NP>>19 b. ID’ = <(Lisi, ta), 2, <NP, V, C, NP>>
c. The sloppy identity condition (30) is satisfied.
In addition, the sentence can also ambiguously make use of independent
α-occurrence to derive the strict reading as in (34). In the sense of Fiengo and May’s
(1994) interpretation, the sluice clause will be realized as a reconstruction of IP; however, given PF-deletion, it is a full-fledged structure derived by merging. The strict reading is considered a result of pronominal reference via α-occurrence, which is anaphoric on elements outside the second conjunct. In this case, the pronoun ta ‘he’ is anaphoric to its antecedent Zhangsan in the first conjunct.
(34) Zhangsan1 bu zhidao [taα1 weishenme bei ma], dan Lisi2 zhidao [(shi)
weishenmej [taα1 tj bei ma]]. (strict)
When the lexical mismatch occurs as in (21), the lexico-syntactic dependency relationship of β-occurrence fails to be achieved, as shown in (35). In (35a), (30) is violated due to the lexical variation in wh-words, even if the syntactic dependency is
19
In (33), the second conjunct differs from the first one in that the former contains negation bu ‘not’. As a reviewer has pinpointed, are these two SD’s still identical? Fiengo and May (1994:52) define the structural description SD as specifying “the structure that connects the elements that bear occurrences of the index”. In line with the notions originally defined by Chomsky (1955), they let a structural description SD be an ordered string <term1, term2, …, termn> and let a phrase marker P be
a set of n-ary linear factorizations, <category1, category2, …, categoryn>. Then a factorization F of
P satisfies SD if and only if category1 is a term1, category2 is a term2, …, categoryn is a termn.
Further, they let a part of a factorization F be any contiguous substring of factors <categoryi, …,
categoryj> of F. The indexical dependencies are linked to structures by the following definition: A
phrase marker P realizes an indexical dependency ID if and only if: (i) for some f, f part of a factorization F of P, f satisfies SD of ID.
(ii) the factors of f bearing I of ID are (cα1, cβ2, …, cβn), the elements of ID,
(iii) there is no f ’ for which (ii) holds that has fewer factors than f.
When the β-occurrences are structurally resolved, the definition is as follows: cβi is resolved in P if
and only if cβi is an element of an ID realized in P.
From the above definitions, it seems that they have not taken the structures such as negation into account. This can be reconfirmed by the representations of (iv) (p.169). In (ivd), there is also no place for negation.
(iv) a. Max1 thinks heβ1 is strong, Oscar2 thinks heβ2 is strong, but hisα2 father doesn’t think heβ2 is
strong.
b. <(Max, he), 1, <NP, V, NP>> c. <(Oscar, he), 2, <NP, V, NP>> d. <(his, he), 2, <NP, N, V, NP>>
satisfied. Likewise, in (35b), although the wh-words with the same indexical value refer to the same entity and the syntactic dependency is satisfied, their lexical forms differ, defying (30). As to (35c), the difference in the syntactic form of the
wh-elements also violates (30), making β-occurrence impossible. For all these cases,
only the independent α-occurrence is allowed to derive the referential strict reading.20 (35) a. Zhangsan1 zhidao [taα1 weishenme bei ma], dan Lisi2 bu zhidao (shi)
(wei-le)-shenmeliyouj [taα1 tj bei ma]. (strict)
b. Zhangsan1 zhidao [taα1 yinggai nian na-yi-ben-shuα2], dan Lisi3 bu zhidao (shi)
sheide shuα2 [taα1 yinggai nian t2 ]. (strict)
c. Zhangsan1 zhidao [shei2 zai piping taα1], dan Lisi3 bu zhidao (shi) weishenme [t2 zai piping taα1]. (strict)
Finally, as to the property of na ‘that’ effect, the fact that the occurrence of na ‘that’ in front of the wh-remnant hinders sloppy identity poses a serious problem to the PF-deletion analysis, since it is impossible to have the sequence [na (yi-ben-shu)
shi na-yi-ben] ‘which book is that’ reconstructed in the sluice clause and paralleled in
the antecedent clause, violating the syntactic parallelism (Fox and Lasnik 2003) as shown in (36). It implies that the hindrance of sloppy identity of the na-effect cannot be explained by the deletion approach. It may belong to another kind of sluicing structure.
20
Since the lst person and the 2nd person pronoun can be used as bound variables (Rullmann 2004), which can be bound by the nearest NP subjects to derive sloppy identity, one of the reviewers points out that the sentence in (i) may pose a problem to this analysis. More specifically, these two pronouns are β-occurrences and under the lexico-syntactic requirement, (i) should be predicted to have sloppy identity, contrary to the fact.
(i) Zhangsan zhidao woβ weishenme bei ma, dan Lisi bu zhidao weishenme woβ bei ma. Zhangsan know I why PASS scold but Lisi not know why I PASS scold ‘Zhangsan knows why I was scolded, but Lisi doesn’t know why I was scolded.’
Even though (i) satisfies both Chung’s lexico-syntactic requirement and Merchant’s e-GIVENness condition, we still do not think that the wo’s ‘I’ here can be β-occurrences, bound by the nearest subject Zhangsan and Lisi, respectively. From (25)-(28), we can easily observe that the variable use of the lst person and the 2nd person pronoun requires their antecedents to have the same person value. That is, this variable function is strictly antecedent-dependent in the sense of person-to-person consistency or feature-checking. Thus, the 3rd-to-lst person pairing in (i) does not conform to this requirement and contributes to its lack of sloppy identity.
(36) *Zhangsan zhidao [ta yinggai nian na-yi-ben-shu ], dan Lisi bu zhidao Zhangsan know he should read which-one-CL-book but Lisi not know
na ( yi-ben-shu) *( shi ) na-yi-ben [ ta yinggai nian t].
that one-CL-book be which-one-CL he should read
‘Zhangsani knows which book hei should read, but Lisij does not know which book hei/*j should read.’ (strict/*sloppy)
In brief, the deletion approach can successfully account for the c-commanding and lexical identity of sloppy identity in Mandarin sluicing, but it fails to explain the
na-effect. Below, we will turn to the pro sluice analysis (Adams 2003, Wei 2004) to
see how it works.
3.2 The pro sluice analysis
Since Adams’s (2003) pro sluice analysis does not recognize the existence of sloppy identity in Mandarin sluicing, we will only focus on Wei’s (2004) base-generated sluice clause analysis, [pro (shi) wh-remnant], in which the sluice is simply composed of a subject pro and a predicate, (shi)-wh-remnant. The pro can be analyzed as a nominal pro or an event pro, depending on the nature of wh-remnant.
Wh-argument is predicated of nominal pro and wh-adjunct is predicated of event pro
as in (37a, b), respectively. He further asserts that only wh-adjuncts anaphoric to event
pro can generate the sloppy identity, while wh-arguments referential to nominal pro
cannot. Obviously, the second part of this claim has already been falsified by (37a).
(37) a. Zhangsani zhidao [ shei zai piping tai], dan Lisij bu zhidao [pro shi
Zhangsan know who PROG criticize him but Lisi not know be shei].
who
‘Zhangsan knows who is criticizing him, but Lisij doesn’t know who.’
(strict/sloppy)
b. [ Zhangsani bu zhidao [ tai weishenme bei ma]], dan Lisij zhidao
Zhangsan not know he why PASS scold but Lisi know [pro ( shi ) weishenme].
be why
‘Zhangsani didn’t know why hei was scolded, but Lisij knows why hei/j was scolded.’ (strict/sloppy)
Starting from the first c-commanding property, it is obvious that the pro sluice analysis can not faithfully verify this requirement due to its simplex structure within sluice clause. Second, how can this analysis account for the lexical identity between
wh-correlate and wh-remnant? We suggest that the unselective binding analysis (Heim
1982, Nishigauchi 1990, Tsai 1994, Cheng and Huang 1996, and Lin 1996) and E-type pronoun analysis (Evans 1980, Tomioka 1999) may be the candidates. The former ensures the satisfaction of lexical matching requirement, and the latter derives sloppy identity. Given the pro sluice analysis, the structure of sluicing in question can be simply illustrated as below.
(38) … [Whi…], … [pro…Whi]
The sloppy identity relies on the lexical identity between the wh-correlate and
wh-remnant. From the point of view of syntactic parallelism, the identity is hard to
acquire under the simplex sluice analysis. However, it is reminiscent of an unselective binding analysis in dealing with Mandarin bare conditional clause, which requires a necessity operator, denoting the force of a universal quantifier, to bind the two
wh-indefinite variables simultaneously as illustrated in (39a, b) (Cheng and Huang
1996 and Lin 1996). The two wh-variables must be uniform in surface form; otherwise, bare conditionals are disallowed. By analogy, the lexical identity in (3a) and (6) can be rewritten in (39c), showing that unselective binding may be a prerequisite of sloppy identity. The two identical wh-words are bound by an existential closure (Heim 1982 and Diesing 1992), not a universal quantifier due to the uniqueness in sluicing. Thus, we can predict that the example (10) lacking lexical identity between wh-words will show the strict reading only due to its failure to apply the unselective binding.21
(39) a. Shei xian lai, shei xian chi. who first come who first eat ‘If X comes first, X eats first.’ b. NECx [x comes first] [x eats first]
Qx restriction nuclear scope c. [OP(∃x)i[…….[xi……]], [….[pro….xi ]]]
Moreover, the unselective binding can explain why the scope of interpretation is only confined within the embedded clause when sloppy reading is deduced, for the variable in the first conjunct only ranges over the embedded clause; hence, it is
21
impossible to have matrix reading in the case of “donkey sluicing”. Besides, this approach ensures that in donkey sluicing the wh-correlate has to be overt and be lexically identical to the wh-remnant, respecting the Revised Version of the Parallelism Constraint on Operator Binding (PCOB) proposed by Cheng and Huang (1996).
(40) Revised PCOB
In a tripartite structure of quantification Q[A][B], [X1, X2, . . ., Xn] (where n ≥ 1) are variables in A. For every variable in A, there must be an identical variable in B.
Under Revised PCOB, both the wh-correlate and wh-remnant are located in restrictive clause and nucleus scope, respectively, and are required to be overt and identical in form. Once the lexical matching requirement is lost, the structure in question will not be counted as a parallel “donkey sluicing” because of the failure of unselective binding. In this case, the pro, now realized as either event pro or nominal one, seems to dominate the interpretation in sluicing. The pro in (37b), an event one, can refer to the embedded event argument or relate to the whole event. The pro of (37a), a nominal one, is supposed to refer to the previous NP-antecedent. Note that all the nominals in front of pro, including the matrix subject Zhangsan, shei ‘who’, ta ‘he’ and even the matrix subject of the second conjunct Lisi, are not the appropriate NP-correlates, mostly because of the anomaly caused by the coreference. Though unselective binding might ensure the matching requirement in donkey sluicing, it cannot give rise to sloppy identity; to be worse, the construal of pro is also problematic in deriving the sloppy reading.22
Therefore, in addition to the requirement of lexical identity, we can further assume that the pro is actually an E-type pro, being responsible for the derivation of sloppy identity. As we have seen above, the identity of pro as being nominal or eventive fails to explain the phenomenon of sloppy identity. What is the identity of pro in this case? In the pro sluice analysis, pro is the subject of the following wh-predicate, which, along with the wh-correlate in the first conjunct, is bound by an existential operator simultaneously via unselective binding. Through predication or agreement, the pro is anaphoric to the following wh-remnant, which is further bound by the operator coupled with wh-correlate. That is, the complex chain [whi, proi, whi] enables the pro
22 Lin (1996) observes that bare conditionals tolerate wh-adjuncts except weishenme ‘why’. Interestingly, Mandarin sluicing allows almost all adjuncts except zenme(yang) ‘how’, which particularly needs verbal support, differing from the other wh-elements. Additionally, the sluicing data are episodic (uniqueness) and close to the one-case bare conditional (Kadmon 1987) in this respect.
