(2) α is the governing category for β iff α is the minimal category containing β, a governor of β, and a SUBJECT accessible to β.

Chapter Two Literature Review 2.1 Introduction

It is well known that Chomsky’s (1981) Binding Theory (henceforth BT) has received much attention in theoretical linguistics, in part due to the consequences that BT brings to grammar as a whole, e.g. the notion of reconstruction and the postulation of PRO, and in part due to the cross-linguistic diversity of the referential possibilities of nominal expressions that have come to our attention. Among the various

phenomena associated with BT, one of the most noticeable is known as long-distance reflexives (henceforth LDRs), which are reflexives with antecedents falling outside the governing category and apparently contradict the standard formulation of Principle A of BT. Hence revisions and attempts have been proposed to tackle LDR within the generative and other traditions, albeit to varying degrees of success. The point of departure common to them is Chomsky’s (1981) formulation of Principle A:

(1) An anaphor is bound in its governing category.

(2) α is the governing category for β iff α is the minimal category containing β, a governor of β, and a SUBJECT accessible to β.

Based on this version of Principle A, many scholars either analyze LDRs in a way consistent with the current formulation by proposing LF-movement, thereby rendering long-distance binding into a sequence of local bindings (Huang and Tang (1991), Huang and Liu (2001), Cole and Sung (1994), Cole and Wang (1996), and Cole et al.

(2001), among others) or reject the crucial ingredient characteristic of that approach,

i.e. a strict c-command relation between an antecedent and a reflexive, and instead

adopt alternative analyses couched in non-Chomskyan frameworks (Pollard and Xue

(1998, 2001), Xue and Popowich (2001), Pan (2001, 2002)). Theoretical and

notational differences aside, these non-Chomskyan approaches also make different empirical predictions which their Chomskyan counterparts are said to fail to capture.

Below I offer an assessment of the two types of current analysis of LDRs in Mandarin Chinese.

2.2 Movement analyses of Chinese ziji

In this section, I shall review proposals put forward by Huang and Tang (1991), Huang and Liu (2001), Cole and Sung (1994), and Cole et al. (2001). The hallmark of these analyses is that ziji is treated as involving LF-movement, thereby maintaining the locality restriction imposed on anaphors by the classical BT. What differs is the type of movement ziji undergoes. For Haung et al., the crucial syntactic operation is the adjunction of ziji to each IP, whereas for Cole et al., it is the movement of ziji as a head via intermediate landing sites to the INFL of each clause. It is at the IP-adjoined position and the INFL that ziji is subject to Binding Condition A (henceforth BCA) for Huang et al. and Cole et al. respectively.

2.2.1 Huang and Tang (1991)

Huang and Tang’s (henceforth HT) approach to Chinese LDR involves

IP-adjunction in conjunction with some auxiliary assumptions (to be stated shortly).

One is to recognize that BT is responsible for two sets of nominal features—

phi-features (including person, number, and in many languages, gender) and

reference-index (R-index for short) at two levels. Their proposal draws its inspiration from the following contrast between (3), (4), and (5):

(3) a. John

knows that Bill

likes pictures of himself

.

b. John

knows that pictures of himself

, Bill

likes.

(4) a. Zhangsan shuo Lisi zui xihuan taziji de shu.

Zhangsan say Lisi most like himself’s book Zhangsan

said Lisi

likes himself’s

book most b. Zhangsan shuo taziji de shu, Lisi zui xihuan.

Zhangsan say himself’s book, Lisi most like

Zhangsan

said that, himself’s

books, Lisi

likes most (5) Zhangsan shuo Lisi chang piping ziji.

Zhangsan say Lisi often criticize self Zhangsan

said Lisi

often criticizedl self

.

In (3b) and (4b), both himself and taziji can refer to the matrix subject and the

embedded one, whereas in (3a) and (4a), they can only refer to the embedded one. In (5), ziji can refer either to the embedded one or the matrix one, in stark contrast to (3a). On the whole, himself and taziji pattern together, unlike ziji.

HT surmised that this difference in binding behavior is tied to their content of phi-features. Whereas ziji is unspecified for person and number, taziji and himself carry third person, singular features.

They also capitalized on the fact that himself can refer to John in (3b) as a corollary of topicalization, a syntactic movement that moves the reflexive to a position above the embedded IP, and therefore theorized that ziji in (5) moves to a similar position, IP, covertly at LF, in a fashion parallel to (3b). In this IP-adjoined position, ziji can form a new governing category in which it can find its antecedent.

What allows such a movement is postulated to have a bearing on the

(non-)content of phi-features. The ziji-type reflexives have to satisfy BT twice to

receive their phi-features and R(eference)-index at S-S and LF respectively, and the

taziji-type reflexives, with inherent phi-features, need to satisfy BT only at S-S, in

order to receive the R-index. In other words, there is no need for the compound

reflexive taziji to undergo such an LF movement because at S-S, its R-index can be

assigned.

In this approach, HT also attempted to account for the well-known blocking effects associated with ziji, as exemplified below:

(6) Ni shuo Zhangsan chang piping ziji.

You say Zhangsan often criticize self

You said that Zhangsan often criticized himself.

Since ziji gets its phi-features from Zhangsan at S-S, it has to undergo IP-adjunction to get its R-index at LF. At LF, however, its phi-features differ from those of the matrix subject, ni, and R-indexing is therefore impossible. The blocking effect in this case is thus explained. But consider the following slightly more complicated sentence:

(7) Zhangsan

shuo wo

juede Lisi

zongshi piping ziji

. Zhangsan say I feel Lisi always criticize self

Zhansgan

say I

feel Lisi

always criticized self

.

If we apply the approach to (7), ziji will receive its phi-features from Lisi at S-S and then at LF adjoins to the lowest IP, where it should look for an R-index-bearing NP.

Since it is incompatible with the intermediate subject wo, it cannot receive an R-index (because compatibility of phi-features is a prerequisite for a reflexive to be bound by an R-index-bearing antecedent). But that would not prevent it from adjoining further up to the intermediate IP, where it could receive an R-index from Zhangsan. But in fact such a prediction is not borne out: ziji cannot refer to Zhangsan. In view of this problem, HT proposed the following stipulation:

(8) An anaphoric

-index (i.e. that received by inheritance under binding) can be retained only if it is directly bound.

3 Ziji can refer to Lisi if it stays where it is at LF.

In (7), once ziji adjoins to the lowest IP, it clashes with the intermediate subject wo with respect to phi-features and therefore cannot be directly bound by it. According to (8), its phi-features cannot be retained and it therefore cannot have a long-distance antecedent.

HT also incorporated Tang’s (1989) notion of sub-command into their BT:

(9) β sub-commands α iff β is contained in an NP that c-commands α or that sub-commands α, and any argument containing β is in subject position.

This definition is needed to rule in the following binding relation between ziji and its antecedent:

(10) [Wo

de jiaoao]

hai-le ziji

. My pride hurtPerf self My pride hurt self

They argued that since the c-commanding NP is inanimate and therefore ineligible as the antecedent but only wo can antecede ziji, this notion of sub-command is necessary.

In short, HT proposed a bi-level application of BT for Chinese reflexives to account for the observation that Chinese taziji is local and ziji can be long-distance bound. For the reflexives specified for ph-features, BT applies at S-S to determine their reference; for those unspecified for ph-features, BT applies at S-S and LF successive-cyclically to determine their phi-features and reference.

2.2.2 Problems with Huang and Tang (1991)

HT’s approach is meant to capture blocking effects and to maintain the locality

thesis of classical BT. However, as Pollard and Xue (1998) noted, objects, in addition

to subjects, can induce blocking as well:

(11) Zhangsan

gaosu wo

Lisi

hen ziji

. Zhangsan tell me Lisi hate self

Zhangsan told me that Lisi hates self.

This is a crucial example against HT’s treatment for blocking. Assuming that the embedded clause is on the same layer as gaosu and wo, the reflexive ziji, once adjoined to the lowest IP, need only receive an R-index from Zhangsan. Since it has received the phi-features, which are compatible with those of Zhangsan, from Lisi at S-S, nothing should prevent ziji from being bound to Zhangsan, and the resulting coindexation should be legitimate, contrary to fact. In other words, HT’s approach to blocking is too weak. Another weakness lies in the stipulation (8), repeated below:

(8) An anaphoric

-index (i.e. that received by inheritance under binding) can be retained only if it is directly bound.

HT offered the following pair of sentences as independent evidence:

(12) a. Zhangsan

shuo ziji

de shu, Lisi

zui xihuan.

Zhangsan say self’s book, Lisi most like

Zhangsan

said that self’s

book, Lisi

likes most.

b. Zhangsan

shuo taziji

de shu, Lisi

zui xihuan.

Zhangsan say himself’s book, Lisi most like

Zhangsan

said that himself’s

book, Lisi

likes most.

(12a) shows that ziji, with its phi-features received at S-S from Zhangsan, can only be directly bound by it but is not directly bound by Lisi whereas (12b) indicates that taziji, with its inherent phi-features, can be bound either by Zhangsan or, via a chain or reconstruction, by Lisi. However, there is every reason to believe (11) is

empirically incorrect. Consider the following:

(13) Zhejian shi zhengming-le duiyu ziji

-de cuo, Zhangsan

sihao meiyou huiyi This thing prove-Perf for self’s mistake, Zhangsan little not-have remorse.

This incident proved that for self’s mistake, Zhansgan didn’t have the slightest remorse.

In (13), ziji receives its phi-features from Zhangsan, probably via a chain but the former is not directly bound by the latter. By (11), its phi-features could not be retained, and the sentence is predicted to be ungrammatical, contrary to fact.

Whatever the cause of the ungrammaticality (12a) on the intended reading, it cannot be due to (11).

Let us now examine another hallmark of HT’s proposal, the distinction between phi-features and reference features (also known as R-index). It seems that this very distinction motivates the bi-level application of BT in Chinese; ziji needs to receive phi-features and R-index at S-S and LF respectively, so Binding Condition A applies twice, whereas taziji and the English reflexive himself only need to be subject to BCA at S-S to receive their R-index. This account predicts that reflexives with inherent phi-features cannot be bound cross-clausally. However, cross-linguistic research has falsified this prediction. For example, Cole et al. (1994) provided a counterexample from Italian:

(14) Credo che [Mario

sostenga che [tu abbia parlato di sé

e della sua famiglia in I believe that Mario claims that you spoke about self and about his family on

TV]].

TV.

The Italian reflexive sé is specified for phi-features, but long-distance is still possible

across a clause whose subject contains incompatible phi-features. This is unexpected

given HT’s analysis. O’Grady (1987) offered another counterexample from Korean:

(15) Bob-i

Harry-ka

John-uy

caki

-eytayhan chayk-ul po-ass-ta-ko Bob-Nom Harry-Nom John’s self-about book saw

malha-yess-ta.

said

Bob

said that Harry

saw John’s

book about self

.

Here, the Korean reflexive caki is inherently third person. Given HT’s analysis, BT should apply to caki at S-S to fix its R-index and it should refer to John only.

However, it is feasible to have the higher subjects as its antecedents. There are also reflexives which don’t have inherent phi-features and which however cannot undergo LF movement to refer to a cross-clausal antecedent in certain contexts (cf. Progovac (1992) for the Russian reflexive svoj).

An obvious counterexample from Chinese against HT’s bi-level analysis is given below:

(16) Wo

renwei zhexie shi zhengming-le ziji

dangshi zuo cuo le.

I think these thing prove-Perf self at-that- time do wrong PER.

I thought that these incidents proved that I did wrong at that time.

Their analysis requires ziji to be bound at S-S for phi-features before adjoining to the embedded IP to receive an R-index from wo. This cannot work, because at S-S the c-commanding NP in the local governing category would supply third person, plural features to ziji, which therefore cannot be bound by wo, due to a feature clash. In other words, HT’s proposal predicts (16) to be ungrammatical, contrary to fact.

Finally, HT’s inclusion of subcommand into grammar constitutes another

4 The Korean reflexive caki is considered a pronominal in Cole et al. (1990). But it is treated as an anaphor in O’Grady (1987) and Sohng (2004). There is supposed to be considerable idioletcal/dialectal variation with this form.

problem. Sub-command is treated as an alternative to c-command embodied in the classical BT. It applies when the c-commanding subject is inanimate. That the subject is inanimate is the conditioning factor for sub-command. Please see section 2.4 for detailed discussion. As we shall see, the same comments apply to Cole et al.’s Feature Percolation.

2.2.3 Huang and Liu (2001)

In view of the aforementioned observation that non-subjects can induce blocking, Huang and Liu (henceforth HL) offered a revised version of IP-adjunction analysis, exempting syntax from accounting for blocking effects and shifting the burden to pragmatics by employing Kuno’s (1972) insights. This effectively means that a syntactic approach to blocking has been rejected. Also absent now is the bi-level application of BT sensitive to the content of phi-features and R-index. HL analyzed long-distance ziji as a reflex of the first person pronoun wo in the direct discourse representation. That is, an LDR is treated as involving an ‘underlying’ representation as follows:

(17) Zhangsan

manyuan Lisi chang piping ziji

. Zhangsan

complain Lisi often criticize self

.

Zhangsan complained that Lisi often criticized self.

(18) Zhangsan manyuan, “Lisi chang piping wo.”

Zhangsan complained, “Lisi often criticized me.”

(18) is the direct discourse representation of (17). As a welcome consequence, this

analysis captures the fact that the matrix subject Zhangsan is said to be aware, or

disposed to say, that the person Lisi criticized was Zhangsan himself in that (18)

contains wo. In contrast, Zhangsan would not necessarily be aware of this if we used

ta ‘he’ instead of ziji:

(19) Zhangsan

manyuan Lisi chang piping ta

.

Zhangsan

complained that Lisi often criticized him

.

The referent of the matrix subject Zhangsan was not necessarily aware that he himself was the person who was often criticized by Lisi. A possible scenario is that Zhangsan lost his memory in a car accident and his family played a video tape recording what he and Lisi used to do. Zhangsan, having watched it, said that Lisi criticized a man a lot without realizing he was the man. However, we, as the external speaker of (19), know that Zhangsan was the man. HL, following Chierchia (1989), called (19) a de re reading, in contrast to (17), a de se reading.

HL argued that this pragmatic account can explain blocking effects in a way that avoids the problems that would arise with any syntactic account. Blocking is teated as a consequence of a perspective clash. Take (20) for example:

(20) a. Zhangsan juede wo zai piping ziji.

Zhangsan feel I at criticize self

Zhangsan feel that I am criticizing self.

b. Zhangsan juede, “wo zai piping wo.”

Zhangsan feel I at criticize me Zhangsan feel, “I am criticizing me.”

In (20a), ziji can only refer to wo as a typical case of the blocking effect. In (20b), the direct discourse representation, wo in object position corresponds to ziji. As we can see, there are two instances of wo now, one referring to Zhangsan and the other to the external speaker. The hearer is therefore likely to get confused in this situation as to who wo refers to. The observed blocking effect then is the result of a perspective clash.

Similarly, if wo were replaced by ni “you” in (20a), blocking would still occur

because ni is anchored to the external speaker and ziji, once translated into wo, is

anchored to the internal speaker—still a perspective clash.

This pragmatic account also explains some cases of blocking that resist a syntactic analysis. For example, non-subject blocking as well as subject blocking is explained now.

(21) Zhangsan

gaosu wo

Lisi

hen ziji

.

In other words, the pragmatic account covers more cases of blocking than a syntactic account and thus should be preferred. Number asymmetry also seems to be explained.

Consdier (22):

(22) a. Lisi

shuo tamen

chang piping ziji

. Lisi say they often criticize self

Lisi

said that they

often criticized self

. b. Tamen

shuo Lisi

chang piping ziji

. They

said that Lisi

often criticized self

.

(22a) is acceptable with ziji referring to the matrix subject whereas (22b) is not. This follows naturally if we assume the following direct discourse representations:

(23) a. Lisi

shuo, “tamen

chang piping wo.”

b. Tamen

shuo, “Lisi

chang piping wo.”

It is obvious that long-distance binding in (22a) is grammatical because there is no perspective clash; wo can naturally refer to Lisi. This is not the case in (22b) because wo is inherently singular and incompatible with tamen, a plural pronoun, although there is no perspective clash involved here.

HL treated Chinese LDR as a logophor in that it, as they supposed, displays

logophoric properties— it describes the consciousness, the source of speech, or the

physical location of the internal protagonist.

They also recognized that not every instance of ziji is a logophor, and the dividing line between a non-logophoric reflexive and a logophor is the traditional notion of governing category. The reflexives that fall within the governing category do not display logophoricity effects. Consider the following:

(24) Zhangsan piping-le ziji-de pengyou.

Zhangsan criticize-Perf self’s friend Zhangsan criticized his own friend.

Zhangsan may or may not know that the person he criticized was his own friend. If (24) is embedded as a complement of a saying verb and long-distance binding is possible, then the matrix subject must be aware of who was criticized, i.e. he self-ascribes to himself the proposition expressed by the complement clause:

(25) Lisi

shuo Zhangsan piping-le ziji

-de pengyou.

Lisi say Zhangsan criticize-Perf self’s friend Lisi

said Zhangsan criticized his

own friend.

To bring this account in line with the formal semantic theory developed by Chierchia (1989), HL proposed adjoining a long-distance, or logophoric, ziji to IP, or as an alternative, moving ziji to the Spec of a CP-type functional category which takes an IP as its complement. Once ziji gets there, it can be bound by the long-distance

antecedent. Such a landing site is termed [Spec, SourceP] to reflect that ziji refers to the source, which entails the consciousness of the internal protagonist (Sells 1987).

5 See Sells (1987) for detailed discussion of the three components of logophoricity. SOURCE represents the source of communication; SELF refers to the mental state or consciousness of the protagonist; and PIVOT is the spatio-physical location of the protagonist.

2.2.4 Problems with Huang and Liu (2001)

HL’s proposal represents a considerable improvement over HT (1991) in that it has abandoned the empirically inadequate stipulations needed to account for blocking effects and the bi-level BT sensitive to the content of phi-features. It is now the task of pragmatics to explain blocking effects, with the consequence that successive-cyclic LF movement is unnecessary (see their endnote 27).

However, the pragmatic account is far from problem-free. Recall that the analysis translates a long-distance ziji into wo in the direct discourse representation. Thus the following is grammatical because there is no perspective clash in the direct discourse representation (26b):

(26) a. Lisi

shuo tamen

chang piping ziji

. (=22a) Lisi

said that they

often criticized self

. b. Lisi shuo, “tamen chang piping wo.” (=23a)

However, to explain why wo would induce blocking in (27a), HL gives (27b) as its direct discourse representation:

(27) a. Lisi

shuo wo

chang piping ziji

. Lisi said I often criticized self.

b. Lisi shuo, “wo chang piping wo.”

It is claimed that once (27a) is translated into (27b), there are two occurrences of wo, one anchored to the internal speaker and the other to the external speaker. This would cause perceptual confusion. Hence the blocking effect. However, one cannot but wonder why the embedded subject in (27a) remains unchanged in (27b). In (27a), wo refers to the external speaker. There are two possible situations in which (27a) may be

6 It remains unknown why Huang and Liu explicitly stated that their analysis is not successive-cyclic on the ground that blocking is tackled by pragmatics in their approach (see their endnote 27 and the following discussion). Successive-cyclicity is supposed to be a general property of movement.

true: either Lisi talked to me and said that I often criticized myself, or he told someone else that I often criticized myself. In the first case, wo ‘I’ in (27a) should be translated into ni ‘you’ in the direct discourse representation. In the second case, it should be translated into ta ‘s/he’ or any other third-person NP. In neither case should wo remain unchanged. If so, this pragmatic account for blocking is based on erroneous

translation of the blocking-inducing first and second person pronouns. Secondly, it fails to explain important dialectal/idiolectal variation; many speakers do not recognize a coreferential reading between wo and ziji in (28):

(28) Wo

juede ta

dui ziji

mei xinxin. (Battistella and Xu (1990)) I think he to self no confidence

I think he has no confidence in himself

This is unexpected given the pragmatic account, since (29), the direct discourse representation of (28) is fully acceptable with no perspective clash involved:

(29) Wo juede, “ta dui wo mei xinxin.”

The following non-corerefential reading between wo and ziji is also unpredicted:

(30) a. wo

renwei wo-men

dui ziji-de

zhengzhuang hai xuyao liaojie.

I think we to self’s symptoms still need understand I think we still need to understand our symptoms.

b. wo renwei, “women dui wo-de zhengzhuang hai xuyao liaojie”.

As in Lisi

shuo tamen

chang piping ziji

, there is no perspective clash involved. The perspective is that of the internal speaker in (30). Nothing in this account should bar ziji from referring to wo; the non-coreferential reading is therefore left unexplained.

The other hallmark of HL’s analysis is the governing category as the dividing

line between anaphoric and logophoric ziji. The following example is argued to illustrate the point:

(31) Zhangsan

kuajiang-le changchang piping ziji

de naxie ren.

Zhangsan praise-Perf often criticize self DE those people Zhangsan praised those people who often criticize self.

HL argued that the sentence indicates that antecedent Zhangsan was aware that the person who was criticized was him. This appears to gain support from the following:

(32) ?? Zhangsan

kuajiang-le houlai sha si ziji

de naxie ren.

Zhangsan praise-Perf later kill die self-DE those people.

Zhangsan praised those people who later killed self.

They claimed that (32) is not so acceptable because normally Zhangsan could not be aware of the event of those people killing him at the time he praised them. This, as they argued, is because an antecedent outside the governing category of ziji must self-ascribe the event. However, there is reason to believe that (31) does not

necessarily have a de se reading. We may embed (31) in a context in which Zhangsan is unaware that the people who he helped often criticized him:

(33) Zhangsan

zai bu zhiqing de qingkuang xia kuajiang-le changchang piping ziji

Zhangsan at no know DE situation down praise-Perf often criticize self

de naxie ren. Ruguo ta zhidao tamen cengjing piping ziji, ta yiding buhui bang DE those person. If he know they ever criticize self, he certainly will not help tamen.

them

Zhangsan unknowingly helped those who often criticized self. If he had known that they criticized him before, he certainly would not have helped them.

As for (32), the unacceptability does not necessarily have to be attributed to the fact

that the referent of Zhangsan could not self-ascribe the event. Consider (34):

(34) Dangshi Zhangsan

jiandao-le yizhang rihou hui wei ziji

dailai caifu de fapiao Then Zhangsan pick-Perf one day-after will for self bring fortune-DE receipt.

At that time Zhangsan picked up a receipt that would later bring him a fortune.

In (34), the referent of Zhangsan was not necessarily aware that the receipt he picked up would bring him a fortune in the future. In other words, he could not self-ascribe the event. Nevertheless, (34) is acceptable. Whatever the cause of the ill-formedness of (32) must be something else.

Finally, like Huang and Tang (1991), Huang and Liu’s (2001) approach says nothing about subject orientation.

(35) Zhangsan

gaosu Lisi

Wangwu

kanbuqi ziji

. (Battistella (1989)) Zhangsan tell Lisi Wangwu look-down-on self

Zhangsan told Lisi that Wangwu looked down on self.

If ziji adjoins the lowest IP, or moves to some Spec position, it is c-commanded by Lisi and Zhangsan. Both should be able to antecede ziji, contrary to the judgment.

2.2.5 Cole and Sung (1994) and Cole et al. (2001)

In this subsection, I shall examine the proposals presented by Cole and Sung (1994) and Cole et al. (2001).

As with HT (1991) and HL (2001), their approach involves LF movement to account for the properties of Chinese ziji; however, what distinguishes their analysis from HT and HL is that the final landing site of ziji is

7 I review these proposals together within this subsection because their syntactic mechanisms for ziji are essentially the same. What differs between the earlier version, i.e. Cole and Sung (1994) and the more recent Cole et al. (2001) is the inclusion of a pragmatic component in the analysis of Chinese blocking facts in the latter. Related works include Cole et al. (1993), Cole and Wang (1996) and Li (1993).

INFL.

The possibility of ziji, or other reflexives, to participate in long-distance binding is attributed to their morphological form: only morphologically simple reflexives such as ziji can move as a head successive-cyclically to INFL, whereas morphologically complex reflexives such as ta-ziji cannot. Hence the local nature of binding with taziji and the availability of ziji in long-distance binding. Long-distance binding is licensed by checking the Agr features on I with the Spec. If a clash of features occurs, then binding fails. Hence the typical blocking effect. Consider below how blocking is derived:

(36) Zhangsan

rewei wo

zhidao Lisi

taoyan ziji

. Zhangsan think I know Lisi hates self.

On this account, long-distance binding is blocked because ziji, once moved to the intermediate INFL, fails in the checking process; since it carries third person singular features, it is incompatible with the intermediate first person subject.

Hence the blocking effect.

Cole and Sung noted that blocking occurs only in languages with no AGR features. For example, Chinese and Korean display blocking and their AGR does not contain inherent phi-features.

On the other hand, Italian has rich AGR and does not display blocking. This observation and animate sub-commanding antecedents

motivate their Feature Percolation Principles (FPP), presented below as an auxiliary device for the head-movement analysis:

(37) Feature Percolation Principles:

8 To be precise, the landing site is Agr.

9 Note that the phi-features on ziji are freely generated here.

10 Note that only the Korean reflexive casin exhibits blocking; the Korean third person reflexive caki doesn’t display phi-feature blocking.

a. The features of the mother node and the features of the daughter nodes will be identical.

b. If the features of the daughter nodes conflict, the mother node will have the features of the head node.

Consider the following Italian example in light of FPP:

(38) La signora

dice che io giaccia presso di sé

. The woman says that I lie near self.

The woman orders that I lie near her.

If sé adjoins to the lowest INFL, the former will not match with the latter in features.

However, according to (37b), if the features of the daughter nodes, i.e. INFL and sé, clash, the Infl will take the features of the head, i.e. first person singular, and check with the intermediate subject. This step violates no proposed principles; then sé moves up to the matrix INFL; again, it is the features of the head, not of sé, that check with the matrix subject. There is no violation whatsoever here. And the binding by La signora is thus predicted, with no blocking effects.

FPP is also responsible for binding by animate sub-commanding antecedents.

According to Cole et al. (1993), an inanimate NP lacks the feature [antecede α], where α takes as its value the referential index of an animate NP. The [antecede α] of a sub-commanding animate NP will percolate to the higher inanimate NP which c-commands ziji. If [Spec, NP] and [N], both being the daughter nodes of an NP, are both animate, as in Zhangsan

de baba

, the [antecede i] conflicts with the [antecede j].

However, according to (37b), baba, as a head, will transfer its feature [antecede j] to the whole NP, the mother node. Thus it is predicted that only Zhangsan

de baba , not Zhangsan

, can bind ziji, as in (39):

(39) Zhangsan

de baba

taoyan ziji

.

Zhangsan’s father hates himself.

Now let us examine the extended proposal made by Cole et al. (2001). Like HL (2001), they noted that subjects and non-subjects alike induce blocking—an

observation that suggests that their head movement analysis is inadequate, as blocking is treated as a clash between the INFL, where ziji is adjoined, and the subject. Since the checking mechanism concerned does not take place between ziji and a verb, the analysis predicts that no blocking would arise when an intervening object carries different phi-features than the subject, as in (21), repeated below as (40):

(40) Zhangsan

gaosu wo

Lisi

hen ziji

. Zhangsan told me Lisi hated him.

Therefore, they included a pragmatic account based on HL’s into their analysis. On the assumption that antecedents of LDR ziji can only be internal pivots (in the sense of Sells (1987)), they argued that (pragmatic) blocking occurs because a first or second person pronoun introduces the external speaker as the pivot, thus eliminating the possibility of any internal pivot and potential LD antecedent. This proposal also explains why ziji cannot, or is unlikely to, refer to Zhangsan in the following:

(41) Zhangsan

renwei wode pengyou

hai-le ziji

. Zhangsan think my friend hurt-Perf self Zhangsan thought my friend hurt himself.

The blocking effect witnessed in (41) is unexpected given the head-movement

analysis, in that the intermediate subject, wode pengyou, is third person singular and

agrees with the matrix subject in phi features. However, the pragmatic account of

blocking correctly predicts that ziji cannot take Zhangsan as its antecedent due to

conflicting centers of deixis—that of Zhangsan and that of the external speaker.

Apart from the pragmatic approach to blocking, Cole et al. retained the grammatical approach to blocking in view of the typological observation that blocking occurs only in languages with no overt verb agreement, e.g. Chinese and Malayalam, as noted in Cole et al. (1993). The pragmatic approach obviously cannot explain this cross-linguistic pattern.

Secondly, they notice that the following sentences display different degrees of strength with respect to the blocking effect:

(42) a. wo

zhidao Wangwu

hen ziji

. I know Wangwu hates self.

b. Wangwu

zhidao wo

hen ziji

. Wangwu knows I hate self.

c. Wangwu

zhidao Lisi

hen ziji

. Wangwu knows Lisi hates self. .

Whereas reference of ziji to the first person matrix subject is somewhat bad in (42a), reference to the third person matrix subject is totally unacceptable in (42b). (42a) and (42c) would be wrongly predicted to be of equal acceptability under the pragmatic account, because neither involves a pivot clash. It therefore seems that a grammatical account is inevitable, as Cole et al. argued. (42a) is somewhat unacceptable because it violates grammatical conditions, although it doesn’t violate any pragmatic ones. (43b) is completely unacceptable because it violates both pragmatic and grammatical

conditions.

2.2.6 Problems with Cole and Sung (1994) and Cole et al. (2001)

It has been long noted since HT (1991) that the head movement analysis violates the Empty Category Principle (ECP), since ziji must move out of an island such as an adverbial clause or a relative clause to allow long-distance binding by a matrix subject

11 However, this cross-linguistic pattern is questionable. See Chapter Four for discussion on the Japanese reflexive zibun.

under the analysis.

In response to this criticism, Cole and Sung (1994) claimed that the CP of the relative clause or the adverbial clause is L-marked only when the bare reflexive adjoins to a head. The resulting CP is therefore not a barrier. Hence no ECP effect.

However, as Progovac (1993) noted, this move wrongly predicts that Wh-extraction should be possible, given that the CP, once ziji is adjoined to a head governing the CP (e.g. V) is no longer an island.

(43) *Xiaoming

bu xihuan nage weisheme piping ziji

de laoshi?

Xiaoming not like that why criticize self ’s teacher

Xiaoming does not like the teacher who criticizes self why?

(43) still exhibits the ECP effect, despite the supposed L-marking of the CP by the adjoining of ziji to the complementizer.

12 Li (1993) accounted for binding into islands. His approach is also a version of head movement analysis. However, he dissociated blocking from head movement so that he didn’t address blocking effects. Furthermore, unlike Cole and Sung, he did not propose to eliminate the barrierhood of the CP by any means; instead, he proposed to delete the offending traces resulting from head movement of ziji so that the ECP applies vacuously to these instances of LDRs. Since the ECP is a representational constraint, once the intermediate traces have disappeared, no ECP effect results. This avoids the wrong prediction noted by Progovac. However, this approach has its own share of problems. For instance, as with other versions of head movement analysis, it predicts that only bare reflexives, not compound reflexives, can participate in long-distance binding (See(44) in the text). Secondly, in order to rule in and out some cases of binding into sentential adverbials (which the antecedent fails to c-command at S-S), he made use of reconstruction; this predicts that the adverbial that cannot reconstruct to a position c-commandable by an antecedent does not exhibit LDRs. A case in point is the sentential conditional clause headed by ruguo “if”. However, this analysis would also predict that the adverbial clause headed by jinguan “despite the fact that” does not allow LDRs because it cannot reconstruct to a sufficiently low position in the matrix clause. This contradicts the fact:

(i) Jinguan Zhangsan zema zijii, Lisii haishi dai-zhe xiaorong.

Although Zhangsan scold self, Lisi still carry-Dur smile Although Zhangsan scolded him, Lisi was still wearing a smile.

(ii) *Lisi [jinguan Zhangsan zema ziji] haishi dai-zhe xiaorong.

Secondly, the analysis is built on the belief that only monomorphemic reflexives can be long-distance bound. However, this is empirically incorrect, as witnessed by the following:

(44) Zhangsan renwei zhexie qian hai-le taziji.

Zhangsan think these money harm-PERF himself.

Zhangsan thought this money harmed him.

In (44), taziji is a compound reflexive; it cannot be analyzed as an emphatic pronoun similar to English he himself. As Audrey Li noted

, a pronominal with the emphatic marker ziji cannot occur in object position, whereas there is no such restriction on names:

(45) Lisi

hai-le ta

/Zhangsan-ziji

/taziji

.

Lisi

harmed him/Zhangsan himself/*him himself

.

In other words, the sequence taziji in object position must be a reflexive.

Thirdly, since head movement analysis is in part to maintain the locality restriction on binding, it cannot accommodate the following instance of binding:

(46) Wo

renwei zhexie shu hai-le ziji

. I think these book harm self I thought these books harmed me.

As long-distance binding is treated as a sequence of local bindings under the head movement analysis we have seen (recall that the grammatical component dealing with blocking is preserved in Cole et al (2001)), the full acceptability of (46) is

13 Her observation appeared in Cole and Sung (1994) as personal communication, although Liejiong Xu made the same point as early as 1986, according to Battistella and Xu (1990)’s article.

unexplained; suppose, as Cole and Sung (1994) did, that ziji has the first person singular features freely generated on it. It then moves to the local Infl, where it enters into Spec-head feature checking. At this step there should be a featural conflict, as the intermediate subject zhexie shu is third person plural as well as inanimate. However, the sentence is perfectly acceptable, contrary to the prediction.

Fourthly, Cole and Sung (1994) treated the object of psych-verbs

as an LF subject, thereby accounting for the possibility of binding relation between the reflexive ziji and a superficial object.

However, as the following sentence demonstrates, such coreferential relations are not limited to psych-verbs:

(47) Zhangsan

dui ziji

mei xinxin (de shi) dui Wangwu

zaocheng-le hen da de daji.

Zhangsan to self no confidence DE event to Wangwu cause-Perf very big DE blow

(The fact) that Zhangsan had no confidence in self caused a serious blow to Wangwu.

In Cole et al. (1994), the surface subject of psych-verbs are “reconstructed” to a VP-internal position sufficiently low to be c-commanded by the experiencer object, an LF subject in their terms. Even if we could adopt such a movement in (47), Wangwu, being inside a PP, still could not c-command ziji. Therefore this casts doubt on the motivation for treating the experiencer object of psych verbs as an LF subject.

Fifthly, feature percolation is retained throughout various versions of the head movement analysis. It has the effect of rendering binding by a sub-commanding animate NP into a case of local binding, as (10) illustrates, repeated below as (48):

(48) Wo

de jiaoao

hai-le ziji

. My pride hurt-Perf self

14 Psych verbs are exemplified by shi…nanguo “make…sad”.

15 I will discuss psych-sentences in more detail in section 3.4, Chapter Three.

My pride hurt me.

Before the problem is introduced, let it be noted that the antecedent of a locally bound reflexive does not have to be aware that the action that the utterance describes

happened to him. In other words, he does not have to self-ascribe the event, as (49) shows:

(49) Zhangsan hai-le ziji.

Zhangsan harm-Perf self Zhangsan harmed himself.

Now let us examine an example from Teochew from Cole et al. (2001):

(50) Ah Meng

gai chia

hai-liao kaki

Ah Meng’s car harmed self Ah Meng’s car harmed itsef/him.

If the FPP applies to (50), we would have a case of local binding—the subject bearing the feature [antecede i] should be able to bind kaki as an eligible locally

c-commanding antecedent. This prediction is not borne out. According to Cole et al.

themselves, kaki requires additional requirements, i.e. a de se requirement.

This is unexpected, as there is no de se or other discourse-pragmatic requirement on a locally bound reflexive. This suggests that FPP incorrectly assimilates (50) into local binding.

Now let us examine the way that blocking is treated. Recall that Cole et al.

recognized grammatical as well as discourse blocking. Consider the following:

16 Cole et al. cited the example to show that the bare reflexive kaki in Teochew is subject to a de se requirement in a non-local environment. However, they fail to notice what consequence this example has on the FPP.

17 It is also probable that their claim that the infelicity of (50) is due to the absence of a de se requirement on the intended antecedent is incorrect. Consider the equivalent of John’s pride harmed self in all Chinese dialects. Presumably, it would be perfectly grammatical, but the referent of John does not have to self-ascribe the event any more than does Ah Meng in (50).

(51) a. Wo

renwei ni

taoyan ziji

. I think you hate self

I thought you hated self.

b. Wo renwei, “ni taoyan wo”.

I thought, “you hate me”.

According to the grammatical account, ziji, with its first person singular features freely generated, would violate Spec-head agreement when it adjoins to the

intermediate Infl. But in the direct discourse representation of (51a), there is no pivot clash, as the only pivot is the external speaker. We should therefore expect (51a) to only show reduced acceptability with a question mark. But this sentence is entirely unacceptable.

Finally, the head movement analysis incorrectly rules out the following sentence.

Under the analysis, ziji must land at Infl to receive its reference. However, once it arrives at the matrix Infl, there is no appropriate c-commanding binder because Zhangsan is inside a PP. It is hard to see how the head movement analysis accounts for the acceptability. Note incidentally that the IP adjunction account cannot cope with (52), either because Zhangsan, residing in a PP, fails to c-command ziji.

(52) Zhe xiang Zhangsan

zhengming-le zhexie yao dui ziji

yidian yong ye meiyou.

This to Zhangsan prove-Perf these medicine to self at-all use not-have This indicated to Zhangsan that this medicine was totally useless for him.

2.2.7 Interim Summary

Both the IP-adjunction analysis and the head-movement analysis started out as

attempts to maintain the (local) classical Binding Theory and to account for certain

phenomena previously thought to involve core grammar, e.g. blocking effects. They

differ in that only the latter attempts to deal with subject orientation. What their more

recent versions have in common is the transferal of part of the workload to pragmatics.

A purely pragmatic approach to blocking adopted by Huang and Liu (2001) seems insufficient, as it fails to consider significant cross-dialectal/idiolectal variation, for instance; a mixed approach along the lines of Cole et al. (2001) inherits some problems from the grammatical component. Equipped with the Spec-head checking mechanism as part of the BT, the head-movement approach manages to cling to what I call the strong locality thesis, with long-distance binding treated as a sequence of local bindings, whereas the IP-adunction analysis manifests a weaker version of the locality thesis by maintaining a local relation between an antecedent and a reflexive with functional projections such as SourceP, SelfP, and PivotP where ziji lands to enter into a c-command relation with the antecedent; the earlier HT (1991)’s

successive-cyclic movement analysis has been explicitly abandoned and long-distance binding is therefore not a sequence of local bindings anymore in this approach.

However, the success at sticking to the locality thesis, whether the strong or weak version, is achieved only at the cost of empirical coverage; for example, (52) remains recalcitrant to both of the refined proposals of the movement analyses we have examined.

2.3 Non-Movement Analyses of Chinese ziji

In this section, I will review proposals by Pollard and Xue (1998, 2001) and Pan

(2001) and Hu and Pan (2002). What they have in common is the rejection of LF

movement to account for the phenomena related to Chinese ziji. Pollard and Xue

(henceforth PX) couched their analysis within the framework of Head-Driven Phrase

Structure Grammar (HPSG) by treating c-command as irrelevant to binding and

instead employing o-command (to be introduced below shortly). On the other hand,

Hu and Pan employed an optimality-theoretic approach incorporating a variety of

primitives from different linguistic components, such as tree configuration and

whether the verb is inherently non-reflexive, i.e. the verb whose two arguments are necessarily disjoint in reference. Both lines of inquiry represent a drastic departure from the Chomskyan BT.

2.3.1 Pollard and Xue (1998, 2001)

The subsection gives a sketch of the mechanisms that PX employed to

characterize binding and coreference relations in Chinese and English.

What most noticeably distinguishes it from the Chomskyan approaches is the inclusion of o-binding:

(53) O-Binding:

X (locally) o-binds Y iff X and Y are co-indexed and X (locally) o-commands Y;

Y is (locally) o-free if it is not (locally) bound.

As we can see, o-binding crucially involves o-command, which is defined below:

(54) O-Command:

X o-commands Y just in case X is a less oblique co-argument of some Z that dominates Y. In case Z=Y, X is said to locally o-commands Y.

To understand o-command, it is necessary to list the obliqueness hierarchy:

(55) The Obliqueness Hierarchy:

SUBJECT< PRIMARY OBJECT< SECONDARY OBJECT< OTHER COMPLEMENTS

Their version of Principle A for American English is stated as follows:

18 Note that binding refers to a coreference relation that stands in some syntactic configuration between an antecedent and a reflexive. Not all coreference relations involve binding.

(56) Principle A for American English:

An anaphor must be locally o-bound if it has a referential (=nonexpletive) local o-commander.

If an anaphor does not have a local o-commander, it is exempt from (56) and may have an antecedent subject to semantic, pragmatic and discourse constraints. This proposal is claimed to have an advantage over the classical BCA, in view of the following example:

(57) John’s

campaign requires that pictures of himself

be placed all over town.

Because himself has no local o-commander in the local domain of the nominal predicate pictures, its antecedent need only obey semantic, pragmatic and discourse constraints. Under the classical BCA analysis, himself, without an accessible subject in the lower clause, must find its antecedent in the higher domain. However, it could not find a c-commanding antecedent there and is predicted to be ungrammatical, contrary to fact.

This formulation, however, cannot be applied to British/Literary English, where examples of the following kind abound:

(58) a. (…) his

wife was equally incredulous of her innocence and suspected himself

, the pastor, to be the cause of her distress, (…).

b. But Rupert

was not unduly worried about Peter’s opinion of himself

.

Nor can it apply to Chinese:

(59) Zhangsan

shuo Lisi taoyan ziji

.

Zhangsan said Lisi disliked him.

In (58) and (59), the reflexive has a local o-commander, but is not locally o-bound—a fact that contradicts (56). Therefore PX puts forward the following principle, called Principle R, for British/Literary English:

(60) Principle R:

A reflexive must either be locally o-bound or interpreted in accordance with certain (English-specific) pragmatic/discourse constraints.

(60) differs from (56) in that the former does not require a reflexive to be obligatorily bound when it has a local o-commander, thereby capturing the facts in British/Literary English.

For Chinese, PX proposed the following principle, called Principle Z:

(61) Ziji must either be o-bound by a subject or interpreted in accordance with certain (Chinese-specific) pragmatic/discourse constraints.

(61) differs from (60) in that o-binding is restricted to the subject in Chinese, whereas there is no such restriction in English. Consider (62) as evidence:

(62) Zhangsan

gei Lisi

yi-zhang ziji-de

xiangpian.

Zhangsan give Lisi one-CL self’s photo.

Zhangsan

gave Lisi

a photo of himself

.

In (62), ziji can only refer to the subject as the antecedent, in contrast to (63) below in English. This conforms to the traditional observation about subject-orientation in Chinese.

(63) John

gave Bill

a picture of himself

.

In (63), himself can refer to John or to Bill. Both o-command himself and can be

legitimate antecedents.

There are at least four cases where ziji can have an antecedent that does not obey the syntactic constraint of being o-bound by a subject. Firstly, for sentences involving psych-verbs, a non-subject NP can qualify as the antecedent only by obeying

non-syntactic constraints under their analysis:

(64) Zhangsan

xiangxin ziji-de

xiaohai mei de jiang de xiaoxi shi Lisi

hen nanguo.

Zhangsan believe self’s child not get prize DE news make Lisi very sad.

Zhangsan

believed that the news of his

child not getting a prize made Lisi

very sad.

In (64), ziji can have either Zhangsan or Lisi as the antecedent. Under the current analysis, Lisi is a possible antecedent because it is a SELF, in Sells’s (1987) taxonomy of logophoricity, although it is not an o-binder of ziji.

Secondly, the syntactically non-obligatory nature of PX’s approach also gains support from ziji with an antecedent in the universe of discourse:

(65) Zheyang yilai nage xuesheng

jiu bu shuohua-le, yanjing cong ziji

-de shen-shang

Thus then that student then not speak-Perf, eye from self’s body-on yikai-le.

move-away-Perf

Then, that student stopped talking, and his eyes moved away from him.

Since the antecedent need not be in the same sentence as ziji, it is hard to see how an obligatory syntactic approach à la Chomsky (1986) can accommodate such cases. The reading of (65) where the antecedent does not refer to nage xuesheng is possible only when an entity is prominent in the discourse.

19 SELF refers to the entity whose consciousness or state of mind is being reported.

Thirdly, sub-commanding antecedent is also treated as licensed by pragmatic, discourse conditions. Consider the following:

(66) a. Zhangsan

-de baba

de qian bei ziji

-de pengyou touzou-le.

Zhangsan’s father’s money BEI self’s friend steal-Perf Zhangsan’s

father’s

money was stolen by his

friend.

b. Zhangsan

-de baba

de qian bei ziji

-de pengyou touzou-le. Mama-de shu ye Zhangsan’s father’s money BEI self’s friend steal-Perf. Mother’s book also

bei ziji

-de pengyou touzou-le. Ta

ji-de ku qilai.

BEI self’s friend steal-Perf. He worry cry start.

Zhangsan’s

father’s

money was stolen by his

friend. His mother’s book was also stolen by his friend. He was so worried that he started crying.

(66a) is provided by Tang (1989) as a case of subcommand. However, if it is

embedded in a context like (66b), Zhangsan can be the antecedent. This suggests that a sub-commanding antecedent should not be analyzed as the result of a syntactic process along the lines of Cole et al. (1994) and Huang and Tang (1991). Rather, its antecedency should be obtained by pragmatic factors, such as topicality, as PX aruged.

Moreover, PX’s approach would treat as pragmatically licensed an instance of ziji contained in an adjunct that is not part of a complement. Consider the following:

(67) Zhangsan

shuo Wangwu

bu hui qu, yinwei Lisi

mei yaoqing ziji

. Zhangsan say Wangwu not will go because Lisi not invite self.

Zhangsan

said that Wangwu

woud not go because Lisi

did not invite him

/himself

.

In (67), Lisi can antecede ziji because it is a local o-binding subject. So can Zhangsan

because it is a subject of the clause taking a complement clause and ipso facto a less

oblique coargument of the complement clause containing ziji. Wangwu cannot

antecede ziji because the former is not a less oblique coargument and o-commanding subject of the latter; ziji is contained in the adverbial clause. Wangwu can hardly rely on pragmatic factors here to obtain antecedency either, presumably because it is not prominent in discourse, according to PX. It is not a topic like Zhangsan. By contrast, the following sentence, being chopped off from (67), is acceptable if ziji refers to Wangwu, although o-binding does not hold here. This is possible when the speaker assumes Wangwu’s viewpoint:

(68) Wangwu

bu hui qu, yinwei Lisi

mei yaoqing ziji

. Wangwu not will go, because Lisi not invite self

Wangwu

will not go, because Lisi

didn’t invite him

/himself

Such examples illustrate the cases which PX treated as non-syntactic uses of ziji. If an NP as an antecedent satisfies the syntactic condition of being an o-commanding subject, it is free from the nonsyntactic conditions, and vice versa. An NP can still satisfy both types of conditions at the same time, as they are not exclusive of each other.

In short, the nonsyntactic conditions in PX’s approach include logophoricity, and discourse prominence, which may manifest in a variety of ways such as topicality (as in (66b)), and contrastiveness.

2.3.2 Problems with Pollard and Xue (1998, 2001)

As PX’s approach does not require a reflexive to stand in a c-command relation to the antecedent, it allows an antecedent not to c-command (or rather, o-command) a reflexive and is therefore free from some of the problems we have seen with

movement analyses. However, as it employs Huang and Liu (2001)’s analysis to

tackle blocking, it suffers from the same problems.

Besides, because syntactic binding is not obligatory in their approach, this raises the question whether the syntactic condition of their account, e.g. binding to an o-commanding subject, is falsifiable. No sentence in which the antecedent is a

non-subject could be considered a counterexample, since any antecedent could always fulfill some non-syntactic condition. The question is how to falsify the syntactic requirement, when it is optional. If we want to falsify it, we have to look for correct sentences in which the antecedent is not an o-commanding subject and does not satisfy the non-syntactic requirement either. Only if such sentences exist can we falsify the proposed syntactic requirement. But this could be done only if we could independently verify what the non-syntactic factors for licensing long-distance reflexives are. But if we want to do this, we have to look for correct sentences with long-distance reflexives which do not satisfy the syntactic requirement to determine whether the non-syntactic requirements really hold. This entails that we know what the syntactic requirement is. This is a circular quest and unlikely to be successful.

Furthermore, Restricting syntactic binders to o-commanding subjects fails to explain the contrast between (67) and the following:

(69) Zhangsan

shuo Wangwu

yinwei Lisi

mei yaoqing ziji

er bu hui qu.

Zhangsan say Wangwu because Lisi not invite self therefore not will go Zhangsan says that Wangwu, because Lisi didn’t invite self, will not go.

Speakers recognize a difference in acceptability between (67) and (69) if ziji refers to Wangwu. (69) is considerably better than (67) on the intended reading.

But in neither sentences is Wangwu an o-binder.

20 That said, to deal with (65) and (66), all syntactic accounts available have to assume that syntactic binding is optional.

21 Pollard and Xue (1998) claimed that in (69) it is more difficult to construe Wangwu than Zhangsan or Lisi as an antecedent. But the important contrast is that between (69) and (67) with respect to Wangwu as an antecedent.

Moreover, since they claim that syntactic binding is not obligatory, to treat sub-commanding antecedents in (66) as pragmatically licensed induces a conceptual conflict in their approach. If syntactic binding is optional, then nothing prevents us from treating binding to sub-commanding NPs as syntactically licensed. In fact, if one thinks of syntactic binding as reflecting preferable judgments, Tang’s (1989) judgment on (66a) indicates that syntactic binding is at work. (66b) may well just illustrate that syntactic binding is optional and pragmatics influences our decision as to what can antecede ziji. It therefore seems that Pollard and Xue wanted to treat (66) as pragmatically licensed just because (66) cannot be coped with by o-binding, their syntactic mechanism.

2.3.3 Pan (2001) and Hu and Pan (2002)

In this subsection, I will review the proposals by Pan (2001) and Hu and Pan (2002; henceforth HP). The latter represents a revised version of the former and eliminates the separate treatments for local and long-distance binding.

Unlike various analyses in the Chomskyan framework, both versions are characterized by the disuse of c-command as a strict requirement on the antecedent. The former version differs from latter by employing self-ascription in its condition on the antecedenthood of ziji. Self-ascription refers to the believer’s ascription of a property to himself (in this case, it is equal to the de se belief), or the speaker’s ascription of a property to a person who is not necessarily self-conscious.

The following is the theory for

22 For Haihua Pan’s treatment of local binding, please refer to Pan (1995, 1997, 1998).

23 For instance, in the case that the proposition “John thinks he is bleeding” is true, John does not necessarily know that he himself is the person who he thinks is bleeding; it could well be that he was kidnapped by some evil scientist and locally anesthetized and is looking at a monitor showing the part of his body on which the scientist is doing an experiment in a laboratory. In other words, there is nothing that makes it obligatory to think that John is self-conscious in the world where the proposition is true. Of course, John may be so, but this interpretation of the proposition is the result of the speaker’s ascription of the property to John.