4.2 A Review of the Facts and Analyses Concerning the Blocking Effect

Chapter Four Blocking Effects Revisited 4.1 Introduction

In Chapter Three, I presented a competition-based approach to the resolution of non-local ziji. In section 4.2 I will review the properties of blocking and the problems with previous analyses. The extant accounts in the literature will be shown to be inadequate. In section 4.3 I will propose a competition-based approach which predicts the occurrence of the blocking effect and, most importantly, the different judgments with respect to whether intervening third person NPs induce blocking effects in a sentence with first/second person pronouns as the matrix subject among native speakers of Mandarin Chinese. Section 4.4 summarizes the results.

4.2 A Review of the Facts and Analyses Concerning the Blocking Effect

In the previous studies of the Chinese long-distance reflexive ziji, the following properties about non-agreement blocking have been noted:

(1) a. An intervening first/second person pronoun prevents ziji from referring to a third-person antecedent for all speakers of Chinese. An intervening third person NP prevents ziji from referring to a first/second person antecedent only for some speakers (Pan (2001); Hu and Pan (2002)).

b. A non-subject can also induce the blocking effect (Pollard and Xue (1998)).

c. An intervening plural third person NP does not prevent ziji from referring to a singular third-person antecedent. An intervening singular third person NP prevents ziji from referring to a plural third-person antecedent (Tang (1989)).

d. An intervening first/second person pronoun, whether singular or plural, prevents ziji from referring to another first/second person pronoun, whether singular or plural (Xu (1993); Cole et al. (1993)).

The following examples illustrate (1a-d):

(2) a. Wo

zhidao Lisi

bu xihuan ziji

.(the construal with wo is possible only for some speakers) (Pan (2001))

I know Lisi not like self

I know that Lisi did not like me/himself.

b. Lisi

zhidao wo

bu xihuan ziji

. Lisi know I not like self

Lisi knows that I do not like myself.

(3) Zhangsan

gaosu wo

Lisi

hen ziji

. (Pollard and Xue (1998)) Zhangsan tell me Lisi hate self

Zhangsan told me Lisi hates himself.

(4) a. Ta

zhidao tamen

dui ziji

mei you xinxin. (Tang (1989)) He know they to self not have confidence

He knows that they have no confidence in himself/themselves.

b. Tamen

zhidao ta

dui ziji

mei you xinxin. (ibid.) They know he to self not have confidence

They know that he has no confidence in himself.

(5) a. Wo

zhidao women

dui ziji

mei you xinxin. (Tang (1989)) I know we to self not have confidence

I know that we have no confidence in ourselves.

b. Women

zhidao wo

dui ziji

mei you xinxin.

We know I to self not have confidence We know I have no confidence in myself.

c. Wo

juede ni

dui ziji

mei xinxin. (Pan (2001)) I think you to self no confidence

I think you have no confidence in yourself.

In (2a), the intervening subject is a third-person NP and blocks ziji from referring to

the matrix subject wo for some speakers, although others accept coindexation between

ziji and wo. In (2b), wo blocks the reference of ziji to Lisi. In (3), wo is not a subject,

but it blocks ziji from referring to Zhangsan. In (4a), the intervening intermediate

subject tamen, being a third-person plural NP, does not prevent ziji from referring to

the third-person singular NP ta. By contrast, in (4b), the intervening NP ta blocks ziji

from referring to tamen. In (5), the intervening first/second pronoun, whether singular

or plural, blocks ziji from referring to the antecedent in the matrix clause.

Having seen these cases, let us review the previous accounts for blocking and examine their adequacy more closely here than we did in Chapter Two. Cole and Sung (1994) adopted a purely syntactic account for blocking based on subject-Infl checking.

As we saw in Chapter Two, such an account has difficulty explaining how blocking arises in (3). The object wo is irrelevant to subject-Infl checking, but blocking nevertheless occurs in (3). This account has to be specified as affecting only person features (excluding number features) to account for the absence of blocking in (4a), where both ta and tamen are both third-person NPs. (4b) can be tackled only by a separate requirement, viz. long-distance ziji needs distributive antecedents: tamen is not specified as distributive, so ziji cannot refer to it (see Huang and Liu (2001) and Xu (1993)).

This account runs into difficulty in dealing with (5a-b), where the matrix and intermediate subjects are both first person. Cole et al. (1993) assumed that first person singular and plural pronouns are considered different persons and the first-person plural pronoun is a “fourth person” in Chinese. The evidence they produced is that in Ancash Quechua, there are two first-person plural

pronouns—noqakuna and noqantsik— and only the former, a form without an inclusive suffix -ntsik, is “treated as coreferential by grammatical processes in the language” (p.114). According to them, the latter can be treated as a fourth person.

However, even if this decision is well motivated for Ancash Quechua, it does not seem plausible for the Chinese pronoun women, which makes no distinction between inclusiveness and exclusiveness. Even when women is used in the exclusive sense, i.e.

without including the addressee, the blocking effect exhibited by (5a-b) still occurs.

Therefore, it does not seem reasonable to accommodate (5a-b) into a syntactic theory of blocking based on subject-Infl person feature checking. Lastly, a syntactic account cannot explain the idiolectal/dialectal variation (2) displays.

96 (4b) would be acceptable if accompanied by dou “all, both”, a distributive device.

Let now turn to a pragmatic account for blocking, represented by Huang and Liu (2001). Inspired by Kuno (1972), they employed a direct discourse representation for sentences involving long-distance ziji and attributed blocking to a clash of anchoring points (or pivots) in discourse. Ziji is translated as wo in the direct discourse. For (2b) and (3), the direct discourse representations are as follows:

(6) Lisi zhidao, “wo bu xihuan wo”.

Lisi know, “I not like me”

Lisi knows, “I do not like myself”.

(7) Zhangsan gaosu wo, “Lisi hen wo”.

Zhangsan tell me, “Lisi hate me”

Zhangsan told me, “Lisi hates me”.

Blocking effects arise because of the two instances of wo in the sentence. This is due to perceptual difficulty and the hearer is likely to get confused as to who wo refers to, as they claimed. On the other hand, there is no blocking in (2a) because its direct discourse representation is the following:

(8) Wo zhidao, “Lisi bu xihuan wo”.

I know, “Lisi not like me”

I know, “Lisi does not like me”.

No confusion as to who wo refers to could arise in this case, hence no blocking effect.

The contrast between (4a) and (4b) also follows naturally from their account. As ziji is represented by wo, a singular pronoun, in direct discourse, it cannot refer to tamen, a plural pronoun, in the matrix clause in (4b). The same reason rules out long-distance binding in (5b). However, it runs into problems with (5a). For (5a), the direct

discourse representation is the following:

(9) Wo zhidao, “women dui wo mei you xinxin”.

I know, “we to me not have confidence”

I know, “we have no confidence in me”.

There is no possibility of perceptual here, as both instances of wo refer to the same entity, the external speaker. Huang and Liu would have to attribute the infelicity of (5a) to something else. Xu (1993) claimed that sentences like (5a) are unacceptable

because the person the matrix subject refers to is included in the set of people the embedded subject refers to. This, not any perceptual clash between different pivots, appears to rule out the impossible reading in (5a). However, this account cannot deal with the following:

(10) Wo renwei women dui wo faxian-de zhezhong wuzhi hai xuyao yanjiu.

I think we to I discover DE this-CL substance still need study I think we still have to study the substance that I discovered.

The sentence is perfectly natural with the matrix subject included in the set of people the embedded subject refers to. If so, the unacceptability of (5a) cannot be attributed to set-subset restrictions. We must seek another explanation for (5a).

Furthermore, (5c) also constitutes a problem for the pivot-based approach to blocking. Consider its direct discourse representation:

(11) Wo juede, “ni dui wo mei xinxin”.

I think, “you to me no confidence’.

I think, “you have no confidence in me”.

Again, there is no perceptual clash between pivots, as the two instances of wo in the

sentence refer to one and the same entity, the external speaker. Nevertheless, blocking

occurs in (5c).

There are other problems with the direct discourse representation intended as an attempt to account for the blocking effect. Reconsider (2b) and its direct discourse representation (6). As Kuno (1972) was aware, one immediate problem with (6) is that there are actually infinitely many direct discourse representations for (2b), depending on who the external speaker is. If the speaker is John, then the first instance of wo must refer to John. If the speaker is Bill, then the first instance of wo must refer to Bill.

In other words, wo in indirect speech is a variable. If we want to translate indirect speech into direct discourse, wo must not be kept as it is. Kuno was fully aware of this point, and posited (12b) as the direct discourse representation for (12a):

(12) a. John decided that I should do it.

b. [John decided, “X should do it,”]

and [X is I]

.

Huang and Liu did not translate wo in indirect speech as a variable in the direct discourse representation. But their account for blocking crucially relies on the

confusion caused by the two instances of wo to explain the blocking effect. However, their account would predict blocking to arise in Japanese as well:

(13) Taro

-wa [boku-ga zibun

-o but-ta] koto-o mada urande-i-ru. (Kuno (1978) via Oshima (2004))

Taro-Top I-Nom self-Acc hit-Past fact-Acc still resent-Asp-Pres Taro still resents that I hit him.

Note that the reflexive zibun is also translated as a first person pronoun in the direct discourse representation. Even though there would be two instances of the first person pronoun according to Huang and Liu, (13) displays no blocking effect.

The absence

97 Oshima (2004) mentions an example where the embedded clause contains a beneficiary-oriented verb such as kureru “to lend”, which induces a sort of blocking effect. It is also worth mentioning that in a non-logophoric context, i.e. sentences without a “saying” or “thinking” verb, a first person pronoun

of blocking is expected only under Kuno’s (correct) direct discourse representation.

Next, reconsider (2a). Huang and Liu explained its absence of blocking on the basis of the following direct discourse representation:

(14) Wo zhidao, “Lisi bu xihuan wo”.

I know, “Lisi not like me”

I know, “Lisi does not like me”.

Since both instances of wo refer to the external speaker, there should not be blocking.

Apart from the inability of this account to explain the different judgments among native speakers with respect to (2a), it also predicts no blocking in the following Japanese sentence:

(15) Boku

-wa Mary-ga zibun

-o nikunde iru to omou. (Kuno (1972)) I-Top Mary-Nom self-Acc hating is that think

I think that Mary hates self.

Since the two instances of the first person pronoun would refer only to the external speaker in the direct discourse representation for (15), no blocking should be observed in (15). However, (15) is ungrammatical. On closer inspection, a direct discourse representation analysis is inapplicable to (15) and (2a). It makes sense to assign a direct discourse representation to sentences with third-person matrix subjects because such sentences represent the speech, thought, feeling, etc. of a third person other than the speaker and the hearer. The external speaker is reporting on his/her speech, thought, feeling, etc. However, if the matrix subject is the first person pronoun, as in (2a) and (15), such sentences are already direct speech. In other words, Kuno’s

will induce a blocking effect. This “empathy blocking” I will not discuss because it is not quite relevant to (13).

(correct) direct discourse representation analysis actually rules out long-distance binding of ziji and zibun by the first-person matrix subject in (2a) and (15) respectively.

To sum up, non-agreement blocking cannot be entirely attributed to the pivot clash, and direct discourse representation analysis is problematic as an explanation for the blocking effect. That said, the notion of pivot is still useful for certain blocking phenomena as we will see later. In the following section I will put forward a competition-based account for the blocking effect.

4.3 A Competition-Based Approach to the Blocking Effect

Just as the Antecedent-Seeking Mechanism is a competition-based approach, the best way to tackle the blocking effect, as I will demonstrate, is competition-based, too.

I will build on the insights of Hu and Pan (2002) and Huang and Liu (2001) and propose my approach. First let us re-examine Hu and Pan’s condition on the blocker:

(16) Blocker:

A blocker is an intervening NP that is the most prominent in an obligatory self-ascribing domain containing a reflexive.

An obligatory self-ascribing domain is the minimal complete functional complex that contains an obligatory self-ascriber, i.e. a minimal NP or IP.

One insight of this condition is distinguishing wo or ni from the blocker. Previous studies refer wo or ni as a blocker (see Pollard and Xue (1998) for example). But for Hu and Pan, wo and ni just help to find a blocker; they are not necessarily blockers themselves. The second insight is defining the domain where we can find a blocker.

98 To qualify [+prominent] in the superlative is redundant, as [+prominent] comes only in two values, plus or minus.

99 Suffice it here to say that obligatory self-ascribers are first and second person pronouns.

With these two points in mind, we can propose the following tentative, preliminary version of our blocking theory, which will be modified as the chapter unfolds:

(17) In a domain to be defined later, a blocker is an NP that prevents another NP from anteceding ziji as a result of some sort of competition.

Competition is suggested by the contrast between (2a) and (2b). For some speakers, wo in (2a) seems to outrank the intermediate subject Lisi. For all speakers, Lisi in (2b) seems to be outranked by the intermediate subject wo; there is something about wo that makes it able to block other NPs from anteceding ziji. There is also something about the intermediate subject Lisi that makes it able to block the matrix subject wo from anteceding ziji in (2a) for some speakers. I propose that a differential animacy hierarchy, aided by the prominence hierarchy for the Antecedent-Seeking Mechanism, is responsible for the idiolectal/dialectal variation:

(18) Animacy Hierarchy:

a. first/second pronouns > other human animate NPs > other animate NPs >

inanimate NPs

b. human animate NPs > other animate NPs > inanimate NPs

Given two candidates X and Y, X is marked as [+animate] iff it ranks higher than Y in the animacy hierarchy.

(19) Prominence Hierarchy:

A. a. [+subject] > [-subject]

b. [-dominating, +animate] > [-dominating, -animate]

c. [+dominating, +animate] > [-dominating, +animate]

d. [+dominating, -animate] > [-dominating, -animate]

e. [-dominating, +animate] > [+dominating, -animate]

B. [+dominating, +animate]> [+subject]

Given two candidates α and β
^ only β, not α, is marked as [+prominent] if α

outranks β in terms of [+subject] but is outranked by β in terms of [+dominating,

+animate].

Let us now turn to (2) and examine our tentative theory:

(2) a. Wo

zhidao Lisi

bu xihuan ziji

.(the construal with wo is possible only for some speakers) (Pan (2001))

I know Lisi not like self

I know that Lisi did not like me/himself.

b. Lisi

zhidao wo

bu xihuan ziji

. Lisi know I not like self

Lisi knows that I do not like myself.

Let Lisi compete with wo in (2a). Recall that some speakers reject the coindexation between wo and ziji. For them, (18b) is the relevant animacy hierarchy. According to (18b), Lisi and wo are both [+animate]. Both are also [+subject]. This is a tied

competition, and as we did in Chapter Three, both are marked as [+prominent]. In the context of blockerhood competition, this means that both are blockers. Lisi blocks wo from anteceding ziji. Wo blocks other NPs, if any, from anteceding ziji. Since there is no other NP, wo as a blocker is idle. For speakers who allow ziji to refer to wo, I propose that they adopt (18a) as the relevant animacy hierarchy. According to (18a), only wo, a first-person pronoun, is [+animate] and Lisi is not. Both are [+subject].

Therefore, only wo is marked as [+prominent] and is a blocker. It follows that Lisi does not block wo from anteceding ziji. Now consider (2b). All speakers reject the reference of ziji to Lisi across wo. This follows from our account: either version of the animacy hierarchy assigns [+animate] to wo. Besides, wo is also [+subject] and marked as [+prominent]. Therefore it is a blocker that prevents ziji from referring to Lisi.

To avoid terminological confusion, however, I will say that the winner of

blockerhood competition is marked as [+block] to distinguish it from an NP marked

as [+prominent] and licensed as an antecedent by the Antecedent-Seeking Mechanism.

So far we have been implicitly assuming that only NPs that are allowed as antecedents by the Antecedent-Seeking Mechanism can participate in blockerhood competition: both Lisi and wo in (2) are output of the mechanism. This seems correct, because we will not run a blockerhood competition on the following sentence, in contrast to (2a):

(20) Wo

renwei zheben shu

dui ziji

mei yong.

I think this-CL book to self no use I think that this book is useless to me.

There is no blocking effect. Coincidentally, the intermediate subject zheben shu is marked as [-prominent] by the Antecedent-Seeking Mechanism. It seems that

blockerhood competition can take place only when there are at least two NPs marked as [+prominent]. This and the fact that an NP that does not agree with antecedents marked as [+prominent] in person features triggers the blocking effect help us define the domain where we find a blocker:

(21) Blockerhood Competition Domain:

A minimal clause containing a blocking facilitator and two NPs marked as [+prominent], one of which may be the blocking facilitator itself. Each such domain consists of two and only two territories, which are delimited by the first clausal boundary from the left within the domain.

(22) Blocking Facilitator:

Linearly speaking, any NP not agreeing in person features with candidates which are marked as [+prominent].

With (21) and (22) in place, let us examine (3):

(3) Zhangsan

gaosu wo

Lisi

hen ziji

. (Pollard and Xue (1998)) Zhangsan tell me Lisi hate self

Zhangsan told me Lisi hates himself.

There are two NPs marked as [+prominent] by the Antecedent-Seeking

Mechanism—Lisi and Zhangsan. Wo does not agree in person features with them and is therefore a blocking facilitator. The blockerhood competition domain is the whole sentence, because it is the minimal clause which includes a blocking facilitator and two NPs marked as [+prominent]. There are two territories—the clause with Zhangsan as the subject and the clause with Lisi as the subject. Within the

Blockerhood Competition Domain, the clausal boundary lies at

[Lisi hen ziji]. Each territory contains a blockerhood candidate. Now let Lisi compete with Zhangsan.

Both are [+subject] and [+animate]. So both are marked as [+block]. Lisi is therefore predicted to block Zhangsan from anteceding ziji. Note that the blocking facilitator is not a blockerhood candidate, unlike wo in (2). Reconsider (2b):

(2b) Lisi

zhidao wo

bu xihuan ziji

. Lisi know I not like self

Lisi knows that I do not like myself.

The Blockerhood Competition Domain is the whole sentence, as it is the minimal clause containing a blocking facilitator and two NPs marked as [+prominent]. Here the facilitator is also an NP marked as [+prominent].

So far we have not touched upon the issue of the competition procedure. If

prominence competition follows a procedure, e.g. it takes place upwardly from the

right, it is interesting to ask whether blockerhood competition behaves similarly. In

addition, so far we have only examined cases where each territory contains only one

NP marked as [+prominent]. I will argue that that issue can be resolved only if we

examine sentences where a territory contains two or more NPs marked as [+prominent]. Consider the following:

(23) Wo

zhidao Lisi

renwei Wangwu

taoyan ziji

. I know Lisi think Wangwu hate self

I know that Lisi thinks Wangwu hates self.

The reading concerned is ungrammatical for the speakers who choose (18b) as the relevant animacy hierarchy. The Blockerhood Competition Domain is the whole sentence, as it is the minimal clause containing a blocking facilitator and two NPs marked as [+prominent]. If competition began by having Wangwu compete with Lisi, i.e. in an upward manner, then Wangwu would be predicted to block Lisi from

anteceding ziji because both NPs are [+subject] and [+animate]. Both would be marked as [+block]. This is not a right prediction. On the other hand, if competition begins by having wo compete with Lisi, both will be marked as [+block] because both are [+subject] and [+animate]. Lisi is correctly predicted to prevent wo from

anteceding ziji. We can now revise (17) as follows:

(24) In a Blockerhood Competition Domain, a blocker is an NP that is marked as [+block] and prevents another NP from anteceding ziji as a result of rightward blockerhood competition.

Our approach also predicts the subject-object asymmetry about the blocking effect noted by Liu (1999):

(25) a. Zhangsan

gaosu wo

ziji

mei bei dahui xuanshang. (Liu (1999), via Pan (2001))

Zhangsan tell me self not by conference select

Zhangsan told me that he was not selected by the conference.

b. Zhangsan

gaosu wo

Lisi

hen ziji

. (Pollard and Xue (1998)) Zhangsan tell me Lisi hate self

Zhangsan told me Lisi hates himself.

In (25a), ziji can refer to Zhangsan across wo. In (25b), ziji cannot, or is unlikely to, refer to Zhangsan. This contrast follows from our account. In (25a), there is only one NP marked as [+prominent] by the Antecedent-Seeking Mechanism, i.e. Zhangsan.

Recall that a Blockerhood Competition Domain requires at least two NPs marked as [+prominent]. This means that there is no blockerhood competition in (25a). Hence no blocking effect. On the other hand, (25b) contains two NPs marked as [+prominent]

and therefore its Blcokerhood Competition Domain is the whole sentence. Let Zhangsan and Lisi compete for [+block]. Both are equally prominent in the

(Blockerhood) Prominence Hierarchy and marked as [+block].

Hence the blocking effect.

The rest of the section will demonstrate how our approach accounts for the instances of blocking reported in the literature. We have considered cases where the blocking facilitator occupies a subject position and a matrix object position. But the facilitator occurs elsewhere as well. It may occur in the specifier position of an NP:

(26) Zhangsan

renwei wo

-de jiaoao hai-le ziji

. (Pan (2001)) Zhangsan think my pride harm-Perf self

Zhangsan thinks that my arrogance harmed me.

The Blockerhood Competition Domain is the whole sentence because it is the minimal clause containing the blocking facilitator, and two NPs marked as [+prominent] by the Antecedent-Seeking Mechanism. One such NP in (26) is the

100 Lisi, being [+block], blocks Zhangsan’s referring to ziji. Although Zhangsan is also [+block], it does not block anything, because there is no NP marked as [+prominent] to its left.

facilitator itself. Now let Zhangsan and wo compete for blockerhood. Both are [+subject], and either version of the animacy hierarchy marks wo as [+animate].

Therefore, wo is marked as [+block] and prevents Zhangsan from anteceding ziji.

Hence the blocking effect as observed.

Next, consider the following sentence which differs from (26) in that the containing NP of wode is animate:

(27) Baoyü

yiwei wo

-de xuesheng

bu xihuan ziji

. (Li (1993), via Pan (2001)) Baoyü think my student not like self

Baoyü thinks that my student does not like himself.

There are two NPs marked as [+prominent] by the Antecedent-Seeking Mechanism in (27)—Baoyü and wode xuesheng. The Blockerhood Competition Domain is the whole sentence, as it is the minimal clause containing wo and two NPs marked as

[+prominent]. Let Baoyü compete with wode xuesheng. Both are [+subject] and [+animate]. Therefore both are marked as [+block] and it is correctly predicted that wode xuesheng prevents Baoyü from anteceding ziji.

Now let us consider sentences with ditransitive verbs:

(28) John

shuo Bill

songgei-le wo/ni

yiben ziji

-de shu. (Pan (2001)) John say Bill give-Perf me/you one-CL self ’s book

John said that Bill gave me/you one of his own books.

There are two NPs marked as [+prominent] by the Antecedent-Seeking

Mechanism—John and Bill. The Blockerhood Competition Domain is the whole

sentence, as it is the minimal clause containing the blocking facilitator wo/ni and two

NPs marked as [+prominent]. Now let John compete with Bill for blockerhood. Both

will be marked as [+block] because both are [+subject] and [+animate]. Therefore, it

is correctly predicted that Bill blocks the reference of ziji to John.

Now it is interesting to note that a structurally identical sentence to (28) was given a different grammaticality judgment by Pollard and Xue (1998):

(29) Zhangsan

zhidao Lisi

song gei ni

yizhang ziji

-de xianpian.

Zhangsan know Lisi give to you one-CL self’s picture Zhangsan knew that Lisi gave you a picture of himself/him.

Apparently, our approach cannot tackle (29) because (29) is expected to behave like (28) if blockerhood competition takes place. However, there is reason to believe that (29) does display the blocking effect, albeit a minor form. Consider (29) with (30), which contains only third-person NPs.

(30) Zhangsan

zhidao Lisi

song gei Wangwu

yizhang ziji

-de xianpian.

Zhangsan know Lisi give to Wangwu one-CL self’s picture Zhangsan knew that Lisi gave Wangwu a picture of himself/him.

(30) differs from (29) only in that the former has Wangwu in place of ni. The reference of ziji to the matrix subject Zhangsan is felt to be more acceptable than in (29), indicating that blocking occurs in (29), however subtle it is. In the same vein, consider the following, which Pan (2001) claimed to display no blocking effect:

(31) John

shuo Bill

ba ziji

-de shu songgei-le wo/ni

. John say Bill BA self’s book give-Perf me/you John said that Bill gave his own books to me/you.

Again, our approach predicts Bill to block John from anteceding ziji. However,

consider (32), which has Tom in place of wo/ni:

(32) John

shuo Bill

ba ziji

-de shu songgei-le Tom

. John say Bill BA self’s book give-Perf Tom

John said that Bill gave his own books to Tom.

A comparison between (31) and (32) shows that the latter is more acceptable,

indicating a (minor) blocking effect occurs in (31). Cole et al. (2001) offered a similar example to (31), which they indicated as marginally acceptable:

(33) Zhangsan

zhidao Mali

gen ziji

shuoguo ni xiang qu Taiwan.

Zhangsan know Mary with self said you want go Taiwan

Zhangsan knows that Mary told him/herself that you want to go to Taiwan.

It seems clear that previous studies fail to recognize the relative nature of the blocking effect. But how can we characterize the weakened blocking effect the above sentences display? Our approach provides a way to neatly represent that. Reconsider (29), (31), and (33). One thing they have in common is that the blocking facilitator is preceded by two NPs marked as [+block]. The following and (29) provide an illuminating contrast:

(34) Zhejian shi gaosu Zhangsan

Lisi

songgei ni yizhang ziji

-de xiangpian.

This-CL event tell Zhangsan Lisi give you one-CL self ’s picture This event told Zhangsan that Lisi gave you a picture of himself.

The Blockerhood Competition Domain is the whole sentence, as it is the minimal clause containing the blocking facilitator and two NPs marked as [+prominent]. Let Zhangsan compete with Lisi for blockerhood. The former, being [-subject] and [+animate], is outranked by the latter, being [+subject] and [+animate]. Therefore only Lisi is marked as [+block]. Zhangsan is [-block]. Since the facilitator is not

101 The questionable status of the coindexation between Mali and ziji is due to the fact that one rarely conveys information to oneself. This has nothing to do with blocking.

preceded by two NPs marked as [+block], (34) does not display the weakened blocking effect as we witnessed in the above sentences.

We will shortly discuss other circumstances where the blocking effect can be weakened. But let us now examine the blocking effect in adverbial clauses we alluded to in Chapter Three:

(35) a. Wangwu

bu hui qu, yinwei wo

mei yaoqing ziji

. Wangwu not will go because I not invite self

Wangwu will not go because I did not invite self.

b. Wangwu

bu hui qu, yinwei Lisi

mei yaoqing ziji

. Wangwu not will go because Lisi not invite self Wangwu will not go because Lisi did not invite self.

Also recall that we assign the following structure to (35b):

(36)

I will argue that this structure is compatible with our approach to blocking. The

blocking effect in (35a) can be explained: Wangwu and wo are marked as [+prominent]

by the Antecedent-Seeking Mechanism. The Blockerhood Competition Domain is the whole sentence because it is the minimal clause containing the facilitator and two NPs

Wangwu Spec

Pivot

t NP

bu hui qu I' IP

Spec

yinwei C

Lisi NP

mei yaoqing ziji I'

IP C'

CP IP

Pivot' PivotP

marked as [+prominent]. As Wangwu and wo are [+subject] and either version of the animacy hierarchy marks wo as [+animate], wo is marked as [+block] and prevents Wangwu from anteceding ziji. Also note that if the adverbial clause is adjoined to the left of the matrix IP, blocking is still observed:

(37) Yinwei wo

mei yaoqing ziji

, Wangwu

bu hui qu.

Because I not invite self, Wang not will go

Because I didn’t invite myself, Wangwu will not go.

Our approach still predicts the blocking effect. Since the matrix subject Wangwu moves to [Spec, PivotP], it is a candidate for the Antecedent-Seeking Mechanism. It and wo are marked as [+prominent] and thus can compete for blockerhood. As in (35a), wo is marked as [+block] and prevents Wangwu from anteceding ziji.

Now let us return to the issue of the weakened blocking effect. I propose two other conditions where the blocking effect is generally weakened:

(38) a. Given two NPs marked as [+block], the lower the position assumed by an XP containing the blocking facilitator on the NP accessibility hierarchy with respect to a preceding NP marked as [+block], the weaker the blocking effect displayed by an NP marked as [+block] linearly following the XP.

b. NP Accessibility Hierarchy (revised from Keenan and Comrie (1977)) Subject > Direct Object > Indirect Object > Oblique

(39) The blocking effect is weakened if the blocking facilitator is not a candidate for the Antecedent-Seeking Mechanism.

Again, the theoretical devices employed by our approach facilitate the

characterization of the weakened blocking effect. Consider the following examples:

102 Actually, Keenan and Comrie’s hierarchy also includes genitive NPs. Lehmann (1986) argued that genitives form special subhierarchies. HPSG also employs a version of this hierarchy, without including genitive NPs. Also note that for speakers who perceive little difference in acceptability between (41) and (42), I propose that they collapse indirect objects and obliques into the same category.

(40) Zhangsan

gaosu wo

Lisi

hen ziji

. Zhangsan tell me Lisi hate self

Zhangsan told me Lisi hates himself.

(41) Zhangsan

dui wo

shuo Lisi

hen ziji

. Zhangsan to me say Lisi hate self

Zhangsan said to me that Lisi hates himself.

(42) Bill

cong ni

-nar tingshuo Sue

piping-le ziji

. (Pan (2001)) Bill from you-there hear Sue criticize-Perf self

Bill heard from you that Sue criticized self.

Speakers generally feel there is a difference of acceptability in the long-distance readings of these sentences. The contrast between (40) and (41) is clear. (42) is not indicated with a question mark or asterisk by Pan, although such sentences receive two question marks in Cole et al.’s paper.

However, Cole et al. (2001) did not consider (40) and (41) and hence probably did not compare the relative degrees of acceptability of these three sentences.

Consider the following, which illustrates (39):

(43) a. Baoyü

yiwei wo

-de xuesheng

bu xihuan ziji

. (Li (1993), via Pan (2001)) Baoyü think my student not like self

Baoyü thinks that my student does not like himself.

b. Zhangsan

yiwei nage gen wo shuohua de ren

taoyan ziji

. Zhangsan think that-CL to me talk DE person hate self

Zhangsan thought that the person who talked to me hated self.

The difference between (43a) and (43b) is that wo in the latter is not a candidate for the Antecedent-Seeking Mechanism. This seems to account for the weakened blocking effect.

103 It is interesting to note that Huang and Liu (2001) recognized the marginal status of such sentences, but Pollard and Xue (1998, p.301) are indifferent to the different degrees of acceptability these

sentences display and mark them all with asterisks on the readings concerned.

If a blocking facilitator meet both (38) and (39), we can expect the weakest blocking effect. Consider the following:

(44) Zhangsan

zai wo-de wuzi-li tingshuo Lisi

taoyan ziji

. Zhangsan at my house-Loc hear Lisi hate self

Zhangsan heard in my house that Lisi hated self.

Compare (44) with (42). The long-distance reading is more accessible in (44) than in (40-42). This lends credence to (38) and (39).

So far we have been assuming that NPs marked as [+block] are blockers. Wo and ni are not necessarily blockers themselves. This is an insight from Hu and Pan. Their move also triggers a new question. Is it possible that blockers are not a property of particular NPs, but of something larger? Now let us consider an example that will revolutionize our conception of blocker:

(45) Wo

renwei zhejian shi gaosu Zhangsan

Wangwu

taoyan ziji

. I think this-CL event tell Zhangsan Wangwu hate self

I think this event told Zhangsan that Wangwu hates self.

Speakers who adopt the animacy hierarchy (18b) reject the reference of ziji to wo.

However, our current approach cannot predict this. The Blockerhood Competition Domain is the whole sentence because it is the minimal clause containing two NPs marked as [+prominent] and the blocking facilitator. The boundary of the two territories lies at

[zhejian shi gaosu Zhangsan Wangwu taoyan ziji]. Now let wo compete with Zhangsan. The former, being [+subject] and [+animate], is supposed to outrank the latter, being [-subject] and [+animate]. Only wo would be marked as [+block] and nothing is supposed to prevent ziji from referring to wo, contrary to fact.

Let us explore the following possibility: what if blockers are not particular NPs, but

refer to territories? Competition could be a matter of two warring territories that send their best representative warriors to fight against each other. Recall that each

Blockerhood Competition Domain consists of two territories. The territory on the left has only one warrior, i.e. an NP marked as [+prominent] by the Antecedent-Seeking Mechanism, in our approach. The territory on the right may have more than one warrior. Not every warrior is equally powerful in the right territory, although they are all marked as [+prominent] by the Antecedent-Seeking Mechanism. Some are weaker like Zhangsan in (45). Now if wo compete with Wangwu instead, we will obtain the right result; both are [+subject] and [+animate] and marked as [+block]. Therefore, Wangwu would prevent wo from anteceding ziji. But Zhangsan and Wangwu are equally marked as [+prominent] already. One way to ensure that Wangwu, not Zhangsan, is marked as [+prominent] is to run again the Antecedent-Seeking Mechanism among the NPs already marked as [+prominent] right after the

Blockerhood Competition Domain has been decided, although the purpose is not find the set of legitimate antecedents now. If so, Wangwu, being [+subject] and [+animate], will outrank Zhangsan, being [-subject] and [+animate] and only the former is

re-marked as [+prominent]. Now Wangwu is the representative warrior that is sent to compete with the warrior in the left territory, i.e. wo. Both will be marked as [+block]

and Wangwu blocks wo from referring to ziji. The notion of territory is important at this stage; Although Wangwu blocks the reference of ziji to wo, it will not block the reference of ziji to Zhangsan because Wangwu and Zhangsan belong to the same territory. They are “on the same side”, so to speak. If so, I propose the distinction between [+block] and [+blocker] this way:

(46) [+block] is a property of warriors, whereas [+blocker] is a property of territories.

An NP marked as [+block] will have its territory marked as [+blocker]. An NP

cannot antecede ziji in a blocker, i.e. a territory marked as [+blocker], if the NP is not a member of the blocker.

So far we have tackled a majority of cases of the blocking effect, but we have not accounted for the blocking effect observed in (5a-b). Our current approach cannot deal with such instances because we defined a blocking facilitator as any NP not agreeing in person features with candidates which are marked as [+prominent]. Since wo and women do not differ in person features, it is hard to see how blockerhood competition rules out the unwanted long-distance coindexation in (5a-b). However, I will demonstrate below that these cases are covered by a variant of blockerhood competition. Inspired by Huang and Liu’s approach to blocking, I propose the following conditions:

(47) Pivot-Blockerhood Competition:

The left territory of a Pivot-Blockerhood Competition Domain competes with the right territory according to the Pivothood Prominence Hierarchy. The winner is marked as [+blocker].

(48) Pivot-Blockerhood Competition Domain:

A minimal clause containing a pivot, two subjects associated with verbal

predicates, and two NPs marked as [+prominent], one of which may be the pivot itself.

Each such domain consists of two and only two territories, which are defined by the first clausal boundary from the left within the domain.

(49) Pivothood Prominence Hierarchy:

[+pivot] > [-pivot].

An NP is marked as [+pivot] iff it is an anchoring point, i.e. a pivot, or is linked to an anchoring point.

An NP marked as [+pivot] will have the territory where it belongs marked as [+pivot] too.

104 The two subjects can, of course, be the two NPs marked as [+prominent] themselves. Note also that the subject here does not include genitive NPs.

Let us now consider (5a) and (5b):

(5) a. Wo

zhidao women

dui ziji

mei you xinxin. (Tang (1989)) I know we to self not have confidence

I know that we have no confidence in ourselves.

b. Women

zhidao wo

dui ziji

mei you xinxin.

We know I to self not have confidence We know I have no confidence in myself.

In (5), the Pivot-Blockerhood Competition Domain is the whole sentence because it is the minimal clause containing a pivot and two NPs marked as [+prominent]. As both NPs are [+pivot], their respective territories are also [+pivot], and the competition is tied: both territories are marked as [+blocker]. The right territory, where the

intermediate subject belongs, blocks the matrix subject from anteceding ziji.

There are two features about Pivot-Blockerhood Competition that makes it different from the usual Blockerhood Competition. First, the participants that enter into pivot-blockerhood competition are not particular NPs marked as [+prominent] by the Antecedent-Seeking Mechanism, but territories. Secondly, the prominence

hierarchy is much simplified, compared with the usual prominence hierarchy employed by the Antecedent-Seeking Mechanism and the usual Blockerhood Competition. Consider the following example to illustrate the point:

(50) Wo

renwei zhejian shi gaosu women

ziji

zuo dui le.

I think this-CL event tell us self do right Perf I think this event told us that we did it right.

Both wo and women are [+pivot]. In addition, only wo is [+subject]. If [+subject]

played a role in Pivot-Blockerhood Competition, the object women would not prevent

wo from anteceding ziji. If pivot-blockerhood competition is conceived as an operation between territories, we can predict that wo cannot antecede ziji; the Pivot-Blockerhood Competition Domain is the whole sentence, as it is the minimal clause containing a pivot and two NPs marked as [+prominent]. Each territory is marked as [+pivot] because it is associated with an NP marked as [+pivot]. Both territories are marked as [+blocker] and the right territory prevents wo from anteceding ziji.

Pivot-Blockerhood Competition covers cases more than (5) and (50). Huang and Liu (2001) mentioned the following example illustrative of pivot blocking:

(51) Zhangsan

shuo
ta

qipian-le ziji

. Zhangsan say he cheat-Perf self Zhangsan

said he

treated himself.

If the speaker points at ta as he utters (51), ziji can refer only to ta. Obviously, ta in this case is marked as [+pivot]. If ziji is coindexed with Zhangsan, the latter is also [+pivot] because antecedents of long-distance ziji are pivots (see Huang and Liu (2001) and Cole et al. (2001)). If so, the territories associated with both NPs are also [+pivot] and marked as [+blocker]. Therefore the territory ta is associated with prevents Zhangsan from anteceding ziji.

Another instance of pivot blocking is offered by Carl Lee Baker to Pan (1997, 2001):

(52) John

renwei Bill

zhidao Mark

ba ziji-de shu jie-gei-le ziji-de pengyou.

John think Bill know Mark BA self ’s book loan-to-Perf self’s friend John thinks Bill knows that Mark has loaned his book to his friend.

There are two instances of ziji in the local clause. The factual generalization is that the

two instances of ziji cannot refer to separate long-distance antecedents at the same time. For example, it is ungrammatical if the first instance refers to John and the second instance refers to Bill, or vice versa. In our approach, there are three NPs marked as [+prominent]. If Bill is the long-distance antecedent, it will be marked as [+pivot]. The Pivot-Blockerhood Competition Domain is the intermediate clause, as it is the minimal clause containing two NPs marked as [+prominent] and a pivot. There are two territories in this domain. As Bill is [+pivot], so is the left territory it is associated with. This territory outranks the right territory, which is [-pivot]. The former is therefore marked as [+blocker]. John cannot antecede ziji because it is not in the blocker, as predicted. On the other hand, if John is intended as the long-distance antecedent of one instance of ziji, our approach predicts that unacceptability will arise if Bill antecedes the other instance of ziji at the same time. If both John and Bill are long-distance antecedents, both are marked as [+pivot]. The Pivot-Blockerhood Competition Domain would still be the intermediate clause, as it is the minimal clause (emphasis added) containing a pivot, Bill, and two NPs marked as [+prominent], one of which happens to be Bill here. The territory containing Bill would also [+pivot] and outrank the right territory. Bill’s territory would therefore be marked as [+blocker].

John could not antecede ziji because it would not be in the same territory as Bill. John can be an antecedent only when Bill is not [+pivot].

4.4 Summary

In this chapter I have proposed a competition-based account for the blocking

effect. It has been shown that dialectal/idiolectal variation involving a matrix

first/second-person subject and an intermediate third-person subject can receive an

explanation, whereas this account also correctly predicts that for all speakers an

intervening first/second-person pronoun induces the blocking effect. I have also

demonstrated that different degrees of the blocking effect can be neatly characterized if we employ the theoretical devices of the current approach, aided by the NP

accessibility hierarchy. Finally, I have argued that a different yet related mechanism for the blocking effect, i.e. Pivot-Blockerhood Competition, is necessary, reflecting the insights of Huang and Liu (2001). The latter type of competition is different from the former because unlike the non-agreement blocking competition, the participants that enter into Pivot-Blockerhood Competition are not particular NPs which marked as [+prominent] by the Antecedent-Seeking Mechanism. Also, [+subject] seems largely irrelevant to such competition. What both types of blockerhood competition have in common is that [+blocker] is a property of territories, not particular NPs.

Overall, this chapter has demonstrated that it is feasible to account for the blocking

effect with a competitional model like anaphor resolution of non-local ziji.