The blocking effect in Vietnamese

Abstract

This article explores a restriction on non-local binding in Vietnamese—the blocking effect—including a systematic comparison with its Mandarin Chinese counterpart. Our finding is that the blocking effect in Vietnamese appeared to be rather different from that in Mandarin but, in fact, employs essentially the same syntactic mechanism. While binding of Mandarin ziji is governed by a [+participant] feature, binding of the Vietnamese anaphor mình is governed by a [+author] feature. Together with the assumption of the presence of a silent performative frame, this derives that binding of Vietnamese mình yields what one may call an Author effect.

1 Introduction

Many languages have anaphors that allow an antecedent beyond the local domain, but only some exhibit a Blocking effect as illustrated by the Mandarin Chinese example in (1) from Huang and Tang (1991).

(1)	a.	Zhangsani	renwei	Lisij	hai	le	zijii/j.
		Zhangsan	think	Lisi	hurt	ASP	self
		‘Zhangsan thought that Lisi had hurt him/himself.’
	b.	Zhangsani	renwei	wo/nij	hai	le	ziji*i/j.
		Zhangsan	think	I/you	hurt	ASP	self
		‘Zhangsan thought that I/you had hurt myself/yourself/*him.’

The Mandarin reflexive ziji may refer to a remote antecedent (1a), but intervention of a 1st or 2nd person pronoun as in (1b) blocks this long-distance binding of ziji, which is generally argued to be due to a mismatch in person features between the intervening subjects wo/ni and the matrix one Zhangsan.¹ According to Cole et al. (1993) and Cole and Sung (1994), only languages without overt subject-verb agreement are likely to induce the blocking effect and Mandarin Chinese is one of that kind.

Similarly to Mandarin Chinese, Vietnamese does not show subject-verb agreement, and Vietnamese has an anaphoric element mình ‘body’, which, like Chinese ziji, allows non-local binding (cf. 2).

(2)	Nami	nghĩ	Hùngj	biết	Mai	thích	mìnhi/j.
	Nam	think	Hung	know	Mai	like	body
	‘Nam thinks Hung knows Mai likes him (=Nam, Hung).’

This makes it interesting to consider Vietnamese in this respect, and indeed, if Hùng in (2) is replaced by the first person pronoun tôi, Nam is not available as an antecedent of mình, showing that Vietnamese exhibits a blocking phenomenon as well (see Doan 2020 for relevant observations and earlier discussion; see also Doan 2022; Chou and Vu 2022).

The aim of the present contribution is to provide an account of why this is so, taking into account the differences and similarities between Vietnamese and Mandarin Chinese.

Preparing the ground for our discussion, it is useful to point out one difference between mình and its Mandarin counterpart ziji. While Mandarin ziji freely allows coargument binding, coargument binding of mình is subject to restrictions. More specifically, in the variety of Vietnamese we base our analysis on, mình can only be used to express reflexivity in combination with the verbal reflexive particle tự ‘self’, where English would use himself as illustrated in (3a) (cf Doan 2022; Tran 2009: 66), although there is some regional variation.² Under these conditions the pronominal nó can also have a local antecedent.³ When tự is present, it always enforces a local interpretation.⁴

(3)	a.	Lani	tự	trừng phạt	mìnhi/nói.
		Lan	Self	punish	body/3sg
		‘Lan punished herself.’
	b.	Lani	trừng phạt	mình*i/nó*i.
		Lan	punish	body/3sg
		‘Lan punished *herself.’

As shown in Reuland et al. (2020) (contra most of the preceding literature), ziji is a complex anaphor consisting of a reflexivizing element zi and a pronominal element ji. By contrast, mình is monomorphemic, qualifying as simplex (Doan 2022), and like other simplex anaphors, such as Dutch zich and its counterparts in Scandinavian, it cannot be co-argument bound, unless this is licensed by some other factor (see Reuland et al. 2020 for a succinct overview of the issue). This will be relevant for the choice of examples since in the variety of Vietnamese we based the exposition of our analysis on, the local subject will not qualify as a possible antecedent for this reason. For our eventual analysis of non-local binding, the regional variation reported with respect to local binding of mình plays no role, however.⁵

What is important to observe is that mình virtually always also allows a speaker value from discourse, indicated by the subscript ‘sp’ (cf. Sect. 7 for further discussion), but only in the absence of tự:

(4)	Lan	(*tự)	trừng phạt	mìnhsp.
	Lan	self	punish	body
	‘Lan punished me.’

We will now consider the Vietnamese blocking phenomenon in more detail. We illustrate it in the following examples:⁶

(5)	a.	Nami	nghĩ	bạnj		biết	Mai	thích	mìnhi/j/sp.
		Nam	think	friend.add		know	Mai	like	body
		‘Nam thinks you know(s) Mai likes him/you/me.’
	b.	Nami	nghĩ	tôij	biết	Mai	thích	mình*i/j.
		Nam	think	1sg	know	Mai	like	body
		‘Nam thinks I know Mai likes me.’

(6)	Tôii	nghĩ	Namj	biết	Mai	thích	mìnhi/j.
	1sg	think	Nam	know	Mai	like	body
	‘I think Nam knows Mai likes him/me.’

In (5a), the antecedent of mình can be the intermediate subject, bạn as the addressee, or the matrix subject Nam, or a speaker value, as discussed above. In contrast, mình in (5b) can only be coreferential with the intermediate subject, namely the first person pronoun tôi ‘I’, and binding of mình by the matrix subject Nam is blocked. Hence, a blocking pattern arises. When the first person doesn’t intervene, as in (6), no blocking is observed.

The following questions then arise: What blocks mình from having the long-distance antecedent in (5b)? How can we account for the blocking phenomenon in Vietnamese? Can we subsume Vietnamese blocking effects under the same type of blocking effects in other languages and, specifically, Mandarin?

We will pursue an analysis in which the first person pronoun tôi in (5b) plays a role as a blocker, and in which, due to its intervention, mình cannot receive a value from the matrix subject Nam. Our analysis of binding of the anaphoric element mình is based on Multiple Agree (Chomsky 2000, 2001; Hiraiwa 2001, 2002, and others) following Giblin (2016)’s approach to Mandarin Chinese. Our derivation of the blocking effect will be inspired by the main idea of Giblin's approach, but significantly modified in the details.

The structure of the article is as follows. In Sect. 2, we will revisit a general description and discussion of the Blocking effect in Mandarin Chinese. In Sect. 3, we introduce Multiple Agree. Section 4 presents Giblin’s account of the Blocking effect in Mandarin Chinese. Sections 5 and 6 explore the properties of the blocking phenomenon in Vietnamese and show how its manifestation contrasts with its Chinese counterpart. Section 7 introduces Ross (1970)’s Performative Hypothesis. Section 8 will be dedicated to developing an account for the blocking effect in Vietnamese in which we introduce the person-feature geometry proposed by Béjar and Rezac (2009) and investigate whether Giblin’s account would extend to Vietnamese. We will then argue that Giblin’s approach does not fully carry over to Vietnamese, as the blocking pattern for mình is distinct from that of Chinese ziji. Hence, an alternative account is required. Section 9 summarizes our findings and concludes with a discussion of the blocking effect as it occurs with other nominal expressions such as kinship and status terms.⁷

2 Revisit the blocking effect in Mandarin Chinese

In order to understand the specific properties of the blocking effect in Vietnamese, a comparison with blocking in Mandarin is useful. Its main features are illustrated below:

(7)	a.	Zhangsani	renwei	Lisij	zhidao	Wangwuk	xihuan	zijii/j/k.
		Zhangsan	think	Lisi	know	Wangwu	like	self
		‘Zhangsan thinks Lisi knows Wangwu likes himself/herself.’
	b.	Zhangsani	renwei	wo/nij	zhidao	Wangwuk	xihuan	ziji*i/*j/k.
		Zhangsan	think	I/you	know	Wangwu	like	self
		‘Zhangsan thinks I/you know Wangwu likes him/*me/*you/himself.’
								(Cole et al. 2006)

In (7a), Zhangsan, Lisi and Wangwu are 3rd person and are all available as antecedents of ziji. In (7b), the matrix subject Zhangsan and the intermediate subjects wo (‘I’) or ni (‘you’) differ in person, Zhangsan being 3rd person and wo and ni being 1st and 2nd person, respectively. This configuration prohibits the reflexive ziji from being bound by the matrix subject Zhangsan; the intermediate subject (wo or ni) itself is not admissible as an antecedent either.⁸

Intervention of a 2nd or 1st person pronoun blocks the long-distance binding of ziji even when these pronouns are not subjects themselves (see Huang and Tang 1991, and also Giblin 2016). This is illustrated in (8); note that, in fact, the non-subject Lisi in (8a) is able to bind ziji, as we will discuss below and in Sect. 5:

(8)	a.	Zhangsani	renwei	Lisij	de	jiao’ao		hai-le		zijii/j.
		Zhangsan	think	Lisi	of	arrogance		harm-Perf		self
		‘Zhangsan felt that Lisi’s arrogance harmed him.’
	b.	Zhangsani	renwei	woj	de	jiao’ao	hai-le		ziji*i/j.
		Zhangsan	think	I	of	arrogance	harm-Perf		self
		‘Zhangsan felt that my arrogance harmed *him/me.’
								(Huang and Tang 1991)

One striking difference between the blocking effect in Mandarin and Vietnamese is that in the latter language only the 1st person is a blocker (cf 5).⁹ As we will see there are some other differences as well. These differences will be among the main issues to be discussed later in Sects. 5 and 6.

As noted, there is substantial literature on non-local binding and the blocking effect in Mandarin, with different types of implementations. They vary from approaches in which non-local binding of ziji is effected by (covert) syntactic movement (as in Battistella 1989; Cole et al. 1990; Cole and Sung 1994) to approaches where non-local binding of ziji is effected by a discourse-based operation involving logophoricity (as in Huang and Liu 2001). We believe that these previous approaches to non-local binding in Mandarin may face a range of challenges (which we are unable to discuss at length here, for reasons of space). It will be sufficient to note that, for example, the head-movement approach cannot account for the blocking effect by non-subjects; see Huang and Tang (1991) and Wong (2021), for instance. As discussed in Wong (2021), the approach in Huang and Liu (2001) sets out to reduce the blocking effect to a conflict in perspective between clause mates. This does not cover cases where the blocking is caused by a more distant intervener, as is also the case in Vietnamese. Moreover, these approaches do not seem to easily carry over to Vietnamese. For instance, the fact that only 1st person is a blocker in Vietnamese is hard to reconcile with an approach along the lines of Huang and Liu (2001). In addition, it is not clear that logophoricity plays a comparable role in Vietnamese as it does in Mandarin. We refer to Charnavel et al. (2017) and Wong (2021) for overviews and further discussion.

Huang and Tang (1991) develop an account that provides an answer to the role of non-subjects in binding and blocking as shown in (8). To account for this pattern, they propose that the c-command condition in binding is too strong for Mandarin Chinese. Rather, it is governed by sub-command, as defined in (9) (Tang 1989: 101):¹⁰

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

According to (9), the sub-commanding NP Lisi in (8a) can bind ziji, inherently animate, since it is the most prominent potential subject contained in the c-commanding NP Lisi de jiao’ao ‘Lisi’s arrogance’. As is standard, long-distance binding of ziji by the matrix subject Zhangsan is not affected. However, the intervention of the first person pronoun wo as the sub-commanding subject in (8b) blocks ziji from being bound remotely by the matrix subject Zhangsan.

In elaborating their proposal, they develop a system that provides a detailed account of the data considered in their paper, but like the other earlier proposals in the literature it relies on the use of indices, thus violating the inclusiveness condition (Chomsky 1995).¹¹ We will refrain from further discussing its details, but as we will see in Sect. 5, the sub-command configuration itself is also relevant for Vietnamese.

Unlike other extant approaches, Giblin (2016) develops a proposal that is compatible with the inclusiveness condition. Building on the works of Progovac (1992, 1993) and Reuland (2005, 2011), Giblin proposes a syntactic account for non-local binding and the blocking effect in Mandarin Chinese. Binding of phi-feature deficient anaphors such as ziji is established by forming an Agree-based dependency (Chomsky 2000, 2001, 2008; Reuland (2005, 2011). More specifically, Giblin’s approach uses the operation of Multiple Agree, as proposed by Hiraiwa (2001, 2005). We therefore opt to pursue an account here that builds on Giblin’s approach.

A more detailed discussion of Giblin’s approach will be presented in Sects. 3 and 4.

3 Multiple agree

As discussed in Chomsky (2000, 2001), Agree is a syntactic operation. The appeal to its existence is justified by the fact that, quite commonly, in natural languages, different constituents share features. Technically, in Chomsky’s implementation, Agree takes place between an element that is unvalued for some relevant feature (a probe) and an element that can supply such a value (a goal), and is subject to the requirement that the probe c-commands the goal. The domain in which a probe can look for a value constitutes its search domain. Thus, the c-command domain of a probe contains its search domain. Such a probe and a goal are in a feature-checking relation. However, the theory of Agree proposed by Chomsky cannot deal with cases where a multiple feature-checking operation occurs in Japanese, such as in Raising to Object and Clefting (see Hiraiwa 2001, 2005; Ura 1996). Extending the theory of Agree, Hiraiwa proposes a theory of Multiple Agree in which a single probe can simultaneously agree with multiple goals in its search domain. Multiple Agree is characterized as in (10a) from Hiraiwa 2001, 2002, revised and elaborated in Hiraiwa (2005), as in (10b):

(10)	a. Multiple Agree: (Multiple feature checking) with a single probe is a single simultaneous syntactic operation; Agree applies to all the matched Goals at the same derivational point derivationally simultaneously.
	b.

Since anaphor binding may include interpretive dependencies between one antecedent and multiple anaphors, Multiple Agree is potentially better suited for modeling it than the original Agree operation. Multiple Agree as defined in (10) possesses two characteristic features, namely simultaneity and multiplicity. Together they mean that the probe searches down its domain to match the nearest goal, then postpones valuation until it finds all other possible goals, and Agree applies to all the matched goals in one fell swoop.

As noted, Mandarin Chinese ziji can be non-locally bound, as illustrated in (11):

(11)	Zhangsani	renwei	Lisij	hen	zijii/j.
	Zhangsan	think	Lisi	hate	self
	‘Zhangsan thinks that Lisi hates self.’

In accounting for long-distance binding of ziji in Mandarin Chinese, Giblin (2016) proposes that there is a matrix C⁰, which starts out unvalued for some relevant phi-feature, specifically a [+participant] feature (see Sect. 4), and looks for a value in its search domain. C⁰ finds a value on the matrix subject and gets valued. Subsequently, the matrix T⁰ and all embedded T⁰s receive this value from C⁰, hence indirectly from the matrix subject.

The element ziji is phi-feature deficient, and assuming that a subordinate TP contains an occurrence of ziji, this element will be visible for probing by the embedded T⁰ and share the relevant value. What results is a phi-feature dependency with the matrix subject, which is interpreted as binding. Giblin assumes that the dependency between C⁰ and elements lower in the structure are not blocked by intervening complementizers and other phase boundaries (see also Bošković 2007). Hence, long-distance binding is accounted for.

Note that, like Mandarin, Vietnamese lacks subject-verb agreement. Hence an embedded T⁰ need not match its feature with the local subject. The derivation is represented in (12–13). The tree in (12) presents the starting point: The elements C⁰, T⁰ and the reflexive ziji all start out unvalued for phi-features; the arrow represents the search operation of C⁰. It probes for a valued feature in its domain and finds a value on the matrix subject DP_i, which then causes C⁰ to be valued.

(12)

Due to the effect of Multiple Agree, the operation valuing C⁰ simultaneously values the relevant phi-feature on the matrix T⁰ and the embedded T⁰ 's, as illustrated in (13).

(13)

The embedded T⁰ bears the same value as the matrix subject, and is able to share these values with ziji; hence, ziji ends up being bound by the latter, as in (14):

(14)
	(Giblin 2016: 141)

Next, consider how Giblin’s account accommodates the binding relation between the embedded antecedent DP_j and ziji. Informally, what we see is that the binding relation between anaphor and antecedent is indirect: It is mediated by a head c-commanding both the antecedent and the anaphor. As assumed earlier, being in a Multiple Agree operation, the probe C⁰ can check multiple goals in its search space; hence it can probe beyond DP_i, and continue looking for a match with another goal; In (15) it finds one, namely DP_j. In Giblin's system, one of these DPs will be used to value the features of the probe, from which the binding relation follows (but see the next section for a restriction). If the matrix DP_i values the features of the probe, ziji gets bound by the matrix DP_i. Alternatively, if the intermediate DP_j values the probe, the latter will bind ziji. This option is represented in (15) and (16):

(15)

(16)
	(Giblin 2016: 142)

Summarizing (15) and (16):

(15): The elements C⁰, two instances of T⁰, and the reflexive ziji all start out unvalued; the arrow represents the search operation of C⁰ in its domain. It may find DP_j, which values C⁰.

(16): After C⁰ gets valued by DP_j, both the matrix T⁰ and the embedded T⁰ inherit the matched features from C⁰. The features inherited from C⁰ then pass down to ziji, which ends up being bound by DP_j.

4 Giblin (2016)’s account of the blocking effect in Mandarin Chinese

As discussed, non-local binding in Mandarin is subject to the Blocking effect. Giblin proposes an analysis of this effect, which we will summarize below, using (17) as an example:

(17)	Zhangsani	renwei	wo/nij	zhidao	Wangwuk	xihuan	ziji*i/*j/k.
	Zhangsan	think	I/you	know	Wangwu	like	self
	‘Zhangsan thinks I/you know Wangwu likes him/*me/*you/himself.’
							(Cole et al. 2006: 23)

As (17) shows, ziji cannot be bound by a third-person matrix subject NP like Zhangsan if there is a 1st/2nd-person pronoun such as wo/ni in the search space of the matrix C⁰. Note that this intervening element need not be in a position that would make it a potential antecedent (see Giblin 2016: 110–113 for details). Here, we will restrict the discussion to the main features of the approach presented in Giblin (2016). It is based on the following conditions:

(18)	A Condition on Multiple Agree: Multiple Agree can take place only under non-conflicting feature specifications of the agreeing elements.

In light of (18), two arguments cannot have contrasting specifications for person when entering Multiple Agree; otherwise, the sentence is ungrammatical. But it is possible for one argument to be fully specified while the other lacks specification.

Giblin’s account of the blocking effect is based on the feature system for personal pronouns proposed in Béjar and Rezac (2003, 2009). They make use of the semantic categories [+/−participant] and [+/−speaker], with the encoding of person feature specifications formulated as in (19):

(19)

As shown in (19), on the left-hand side of the table, while π, representing person, is shared by all pronouns, the participant feature is only shared by 1st/2nd person. The latter form a contrastive relation in that the 1st person is not only assigned a marked [+participant] value shared with the 2nd person, but also contains the marked [+speaker] value, which is absent in the 2nd-person entry. Furthermore, the table expresses that the [+speaker] value will always go together with a [+participant] value.¹² The right-hand side of the table presents a short-hand version of the left-hand side in which the number 3 stands for person, 2 refers to the participants, and 1 to the speaker.

Giblin’s account for the Blocking effect now works in the following manner. He proposes that the probe C⁰ merged in the matrix clause is unvalued for a [+participant] feature and searches for a goal as a source for a value. The valuation operation of C⁰ will be prohibited if it violates the requirement of Contiguous Agree (Nevins 2007):

(20)	Contiguous Agree (informally): There can be no interveners between P and x that are not in the domain of relativization that includes x.

In the condition, P represents the probe, and x is the goal. The domain of relativization represents a set of feature values that are in some relevant sense related. Thus, no interveners in the path from the probe to the goals are allowed to be ‘too different’ from the feature the probe searches for.¹³ For the case of Mandarin Chinese, this amounts to the requirement in (21):

(21)	i. The C0 probe searches for a particular feature, namely [+participant].
	ii. A convergent derivation will occur when there are no unmarked values of
	[participant] that intervene between the probe and the featural specification that it is looking for. That is, there can be no [−participant] DPs that occur between the probe and a [+participant] DP.
							(Giblin 2016:147)

In a nutshell, the feature [−participant] is 'too different' from a [+participant] feature in the sense of Nevin's contiguity requirement. Let us see how this works in sample derivations. (21i) is illustrated in (22) (Giblin 2016: 54), where the intervention of the [−participant] DP between the probe C⁰ and the [+participant] DP as the goal, which is a better match, causes a violation of Contiguous Agree. Hence, the derivation crashes, which is indicated by the star. Crucially, the [+participant] feature is shared by 1st/2nd-person pronominals. Hence, its role in the derivation entails that both 1st/2nd-person pronominals cause a blocking effect.

(22)

The Contiguous Agree constraint works effectively in dealing with the fact that blocking can be caused by interveners that are not themselves possible binders, an issue that is problematic for other approaches.

Although Giblin’s approach works well for Mandarin, we will see that it does not fully carry over to Vietnamese, as the blocking pattern for mình is distinct from that of Chinese ziji. Hence, an alternative account must be provided. We will argue that, unlike Chinese, the blocking effect in Vietnamese is not caused by a violation of Contiguous Agree with respect to the [+participant] feature but by a violation with respect to the [+author] feature. In the end, then, the source for the blocking effect in Vietnamese will be closer to that in Mandarin than the differences would lead one to initially expect.

5 A blocking effect in Vietnamese

As already mentioned in the introduction, non-local binding of Vietnamese mình also shows a blocking effect. The examples are repeated here as (23), (24), and (25):

(23)	Nami	nghĩ	Hùng/bạnj		biết	Mai	thích	mìnhi/j/sp.
	Nam	think	Hung/friend.add		know	Mai	like	body
	‘Nam thinks you/Hung know(s) Mai likes him/you/me.’
(24)	Nami	nghĩ	tôij	biết	Mai	thích	mình*i/j.
	Nam	think	1sg	know	Mai	like	body
	‘Nam thinks I know Mai likes me.’
(25)	Tôii	nghĩ	Namj	biết	Mai	thích	mìnhi/j.
	1sg	think	Nam	know	Mai	like	body
	‘I think Nam knows Mai likes him/me.’

The blocking configuration in Vietnamese differs from that in Mandarin Chinese in that the second person pronoun does not serve as a blocker, as illustrated in (26):¹⁴

(26)	Nami	nghĩ	màyj		biết	Mai	thích	mìnhi/j.
	Nam	think	2sg		know	Mai	like	body
	‘Nam thinks you know Mai likes him/you.’

Furthermore, in Mandarin Chinese the intervening first person pronoun is itself ruled out as an antecedent, whereas it is licit in Vietnamese, as in (24). Similarly to Chinese, an intervening third-person NP does not serve as a blocker in Vietnamese, as shown in (25). Here mình refers to the speaker realized as the first-person pronoun tôi in the matrix clause, while binding by the intermediate antecedent Nam is also acceptable. Mình can optionally refer to the speaker when there is no first-person antecedent as in (23).

In general, there are two cases of the blocking effect in Vietnamese, which are given as follows:

• (i) There is a blocking effect when a first-person pronoun subject structurally intervenes between mình and a more remote potential antecedent.

• (ii) The blocking effect is also induced when a sub-commanding, first-person pronoun, inherently animate, intervenes between mình and a more remote potential antecedent.

Case (i) is illustrated in (24) in which the intervening subject of the first person tôi triggers a blocking effect. The same pattern holds in cases where the predicates are various kinds of thinking and saying verbs such as ngờ ‘doubt’, tiết lộ ‘reveal’, tin ‘believe’, hiểu ‘understand, tưởng ‘mistakenly guess’, quên ‘forget’. See (27), (28), and (29):

(27)	a.	Nami	ngờ	là	Hùngj	đã	tiết lộ	với
		Nam	doubt	that	Hung	PST	reveal	with
		mọi người	rằng	Thu	ghét	mìnhi/j/sp.
		everybody	that	Thu	hate	body
		‘Nam doubted that Hung revealed to everybody that Thu hated him/me.’
	b.	Nami	ngờ	là	bạnj	đã	tiết lộ	với
		Nam	doubt	that	friend.add	PST	reveal	to
		mọi người	rằng	Thu	ghét	mìnhi/j/sp.
		everybody	that	Thu	hate	body
		‘Nam doubted that you revealed to everybody that Thu hated him/you/me.’
	c.	Nami	ngờ	là	tôij	đã	tiết lộ	với
		Nam	doubt	that	1sg	PST	reveal	With
		mọi người	rằng	Thu	ghét	mình*i/j.
		everybody	that	Thu	hate	body
		‘Nam doubted that I revealed to everybody that Thu hated me.’

(28)	a.	Nami	tin	Hùngj	sẽ	hiểu	rằng	Thu	luôn
		Nam	believe	Hung	will	understand	that	Thu	always
		muốn	bảo	vệ	mìnhi/j/sp.
		want	protect	body
		‘Nam believed Hung would understand that Thu alwayswanted to protect him/me.’
	b.	Nami	tin	bạnj	sẽ	hiểu	rằng	Thu	luôn
		Nam	believe	friend.add	will	understand	that	Thu	always
		muốn	bảo	vệ	mìnhi/j/sp.
		want	protect	body
		‘Nam believed you would understand that Thu always wanted to protect him/you/me.’
	c.	Nami	tin	tôij	sẽ	hiểu	rằng	Thu	luôn
		Nam	believe	1sg	will	understand	that	Thu	always
		muốn	bảo	vệ	mình*i/j.
		want	protect	body
		‘Nam believed I would understand that Thu always wantedto protect me.’

(29)	a.	Nami	tưởng	Hùngj	đã	quên	rằng	Thu	lừa
		Nam	suppose	Hung	already	forget	that	Thu	deceive
		mìnhi/j/sp.
		body
		‘Nam supposed Hung forgot that Thu deceived him/me.’
	b.	Nami	tưởng	bạnj	đã	quên	rằng	Thu	lừa
		Nam	suppose	friend.add	PST	forget	that	Thu	deceive
		mìnhi/j/sp.
		body
		‘Nam supposed you forgot that Thu deceived him/you/me.’
	c.	Nami	tưởng	tôij	đã	quên	rằng	Thu	lừa
		Nam	suppose	1sg	already	forget	that	Thu	deceive
		mình*i/j.
		body
		‘Nam supposed I already forgot that Thu deceived me.’

The examples (27c), (28c), and (29c) show that the presence of an intervening first-person pronoun tôi simply excludes long-distance binding of mình by the matrix subject Nam. By contrast, the intervening (low) status term bạn as the addressee in (27b–29b), like the 2nd person pronoun mày in (26), and the third person NPs in (27a–29a), does not trigger a blocking effect. Similar patterns hold for sentences in (30) and (31) where the matrix predicates are verbs of perception such as nghe ‘hear’ and thấy ‘see.

(30)	a.	Nami	nghe	Hùngj	tiết	lộ	với	mọi	người	rằng
		Nam	hear	Hung	reveal		with	everybody		that
		Thu	ghét	mìnhi/j/sp.
		Thu	hate	body
		‘Nam heard Hung reveal with everybody that Thu hated self.’
	b.	Nami	nghe	bạnj	tiết	lộ	với	mọi	người	rằng
		Nam	hear	friend.add	reveal		with	everybody		that
		Thu	ghét	mìnhi/j/sp.
		Thu	hate	body
		‘Nam heard you reveal with everybody that Thu hated self.’
	c.	Nami	nghe	tôij	tiết	lộ	với	mọi	người	rằng
		Nam	hear	1sg	reveal		with	everybody		that
		Thu	ghét	mình*i/j.
		Thu	hate	body
		‘Nam heard me reveal with everybody that Thu hated me.’

(31)	a.	Nami	thấy	Hùngj	tiết	lộ	với	mọi	người	rằng
		Nam	see	Hung	reveal		with	everybody		that
		Thu	ghét	mìnhi/j/sp.
		Thu	hate	body
		‘Nam saw Hung reveal with everybody that Thu hated self.’
	b.	Nami	thấy	bạnj	tiết	lộ	với	mọi	người	rằng
		Nam	see	friend.add	reveal		with	everybody		that
		Thu	ghét	mìnhi/j/sp.
		Thu	hate	body
		‘Nam saw you reveal with everybody that Thu hated self.’
	c.	Nami	thấy	tôij	tiết	lộ	với	mọi	người	rằng
		Nam	see	1sg	reveal		with	everybody		that
		Thu	ghét	mình*i/j.
		Thu	hate	body
		‘Nam saw me reveal with everybody that Thu hated me.’

These sets of sentences indicate that the predicates do not play a role in causing the blocking effect, but the first-person pronoun does.

As indicated in (ii), a sub-commanding NP of the first person also yields a blocking pattern. See (32). However, unlike in Mandarin (Giblin 2016: 45), in Vietnamese a 1st-person pronoun in object position does not act as a blocker, as illustrated in (33):

(32)

a.

Hùngi

nghĩ

tính

kiêu

ngạo

của

Namj

đã

làm

hại

mìnhi/j/sp.

Hung

think

CL arrogance

of

Nam

PST

cause

harm

body

‘Hung thought that Nam’s arrogance harmed him/me.’15

b.

Nami

nghĩ

tính

kiêu

ngạo

của

tôij

đã

làm

hại

mình*i/j.

Nam

think

CL

arrogance of

1sg

PST

cause

harm

body

‘Nam thought that my arrogance harmed me.’

(33)	a.	Johni	nói	với	tôij	là	Nam	ghét	mìnhi/j.
		John	say	with	1sg	COMP	Nam	hate	body
		‘John said to me that Nam hates him/me.’
	b.	Johni	luôn	nhắc	tôij	là	Nam	ghét	mìnhi/j.
		John	always	remind	1sg	COMP	Nam	hate	body
		‘John always reminds me that Nam hates him/me.’

In (32a), mình may take the matrix subject Hùng or the sub-commanding Nam as its antecedent or it may receive a speaker value from discourse. On the other hand, the presence of the first-person pronoun tôi as the sub-commanding element in (32b) results in a blocking effect. Here, mình can only be bound by the first-person pronoun tôi, while its remote binding by the matrix subject Nam is ruled out. In (33), however, the intervening tôi does not keep John from acting as an antecedent for mình, contrary to what happens in Mandarin.¹⁶

Note that, as a subcase, blocking does also occur when mình functions as a nominative anaphor. Consider (34):

(34)

a.

Hùngi

nghĩ

tôij

sẽ

thừa

nhận

mình*i/j

đã

chỉ

trích

Mai.

Hung

think

1sg

will

admit

body

PST

criticize

Mai

‘Hung thought I would admit that I criticized Mai.’

b.

Tôii

nghĩ

Hùngj

biết

mìnhi/j

đã

chỉ

trích

Mai.

1sg

think

Hung

know

body

PST

criticize

Mai

‘I thought Hung knew he/I criticized Mai.’

As shown in (34a), with the intervention by the first person pronoun tôi, binding of mình by the remote antecedent Hùng is blocked. By contrast, in (34b), when the intervening subject is a third person expression, namely Hùng, binding by Hùng as well as binding by the first-person pronoun tôi are fine. Note that this differs from what Giblin (2016: 169–170) observed for Mandarin Chinese, where ziji in subject position is exempt from binding requirements and can have a non-local antecedent even in the presence of [+participant] interveners.

As in the cases discussed above, long-distance binding of mình as a possessor is blocked as well when a first-person pronoun intervenes. Consider (35):

(35)	Nami	nghĩ	tôij	biết	Thuk	thích	khu	vườn	của	mình*i/j/k
	Nam	think	1sg	know	Thu	like	garden		POSS	body
	‘Nam thought that I knew Thu likes self's garden'.

In (35), mình may take the local subject Thu or the first-person pronoun tôi as its antecedents. However, binding of mình by the matrix subject Nam is ruled out as a result of the blocking effect.

The question of how to accommodate both the differences and the similarities between the blocking effects in Vietnamese and Mandarin will be addressed and resolved in the next sections.

6 The difference in the blocking effect between Vietnamese and Mandarin Chinese

The evidence presented so far shows that whatever causes the blocking effect shows up differently in Vietnamese than it does in Mandarin Chinese. As we will show, one factor is that the blocking effect in Vietnamese is based on the [+author] feature, rather than on the [+participant] feature. In the previous section, we discussed possible combinations of third-person subject NPs with lower subjects as potential interveners. Here, we will further examine the combinations of the first-person pronoun tôi/tao and the second-person pronoun mày as the higher subjects with the different options for lower subjects. See (36) and (37):

(36)	a.	Tôii	biết	nój	tin		Hà	không	ghét	mìnhi/j.
		1sg	know	3sg	believe		Ha	not	hate	body
		‘I knew he believed that Ha did not hate self.’
		=>1>3
	b.	Taoi	biết	màyj	nghĩ		Hà	không	ghét	mìnhi/j.
		1sg	know	2sg	think		Ha	not	hate	body
		‘I knew you thought that Ha did not hate self.’
		=>1>2
	c.	Tôii	nói	tôii		nghĩ		Hà	không	ghét	mìnhi
		1sg	say	1sg		think		Ha	not	hate	body
		‘I said I thought that Ha did not hate self.’
		=>1>1

(37)	a.	Màyi	biết	nój	nói	Hà	không	tin		vào	mìnhi/j/*sp.
		2sg	know	3sg	say	Ha	not	believe		in	body
		‘You knew he said that Ha had no confidence in self.’
		=>2>3
	b.	Màyi	nói	màyi	tin		Hà	không	ghét	mìnhi/*sp.
		2sg	know	2sg	believe		Ha	not	hate	body
		‘You said you believed that Ha did not hate self.’
		=>2>2
	c.	Màyi	nói	taoj	tin		Hà	không	ghét	mình*i/j.
		2sg	know	1sg	believe		Ha	not	hate	body
		‘You said I believed that Ha did not hate self.’
		=>*2>1

The sentences in (36) and (37) show that there is a blocking effect only in the case of (37c) where the first-person pronoun tao serves as the intervener and triggers a blocking effect. Note that (37a,b) deserve attention in one other respect, namely that the speaker interpretation of mình is not available here; see Sect. 8 for discussion.

The difference between the blocking effect in Vietnamese and that in Mandarin Chinese is summarized in Table 1 (based on the Vietnamese facts that we have established so far) and Table 2 (representing the facts from Mandarin Chinese provided by Li 1990).

Table 1 The blocking versus non-blocking patterns in Vietnamese

Further properties:

• i. In Vietnamese, the intervener that causes the blocking can itself act as an antecedent of mình.

• ii. A [+author] element in object position does not cause blocking.

• iii. Mình in subject position is subject to blocking.

Table 2 The blocking versus non-blocking patterns in Mandarin Chinese (summarized by Li (1990); slightly modified)

Further properties:

• i. In Mandarin, the intervener that causes the blocking cannot itself act as an antecedent of ziji.

• ii. A [+participant] element in object position does cause blocking.¹⁷

• iii. Ziji in subject position is not subject to blocking.

However, as we will see, there are two other factors as well. One is the fact that mình can virtually always be valued as the speaker from the discourse. In the traditional literature, this has been taken to indicate that, alongside the anaphor mình, there is also a (1st-person) pronominal mình. We argue, instead, for a uniform analysis of mình. This is achieved by deriving the speaker interpretation by the optional insertion of a performative frame, which will be discussed in more detail in Sect. 7, below. The other major factor, to be discussed in more detail in Sect. 8, we hypothesize to reside in the optional merger of a complementizer in complement clauses.

7 The performative hypothesis (Ross 1970) and its application to Vietnamese

Consider the following sentences from Ross (1970):

(38)	a. Prices slumped.
	b. Even Rodney’s best friends won’t tell him.

(39)	a. I promise you that I won’t squeal.
	b. I sentence you to two weeks in the Bronx.

Elaborating Austin (1962), who distinguishes constative sentences as in (38) from performative sentences in (39), Ross (1970) proposes that every sentence is embedded under a performative frame: a covert syntactic structure containing a representation of the speaker, the hearer, and a performative verb. The proposal is formulated as in (40):

(40)	The performative analysis: All declarative sentences occurring in contexts where first-person pronouns can appear, derive from deep structures containing one and only one superordinate performative clause whose main verb is a verb of saying.

Ross uses the term deep structure, which is currently no longer in use, but refers to a structural representation of a sentence before movement and deletion operations. He proposes that a sentence like (38a) will have a deep structure as in (41).

(41)

(41) contains a performative frame “I—performative V—you” as the highest clause. In Ross’s analysis, this frame is subsequently deleted. In more current terms, one would say that the elements of the frame are syntactically represented but are not realized at PF (Phonological Form). This is the form in which we will adopt it.

To demonstrate the syntactic visibility of the performative frame, Ross provides thirteen arguments in English, seven of which are dedicated to postulating a higher subject NP I; three are to prove that verbs in the silent clause must be verbs of saying with the feature composition [+communication, +linguistic, +declarative]; and the other three arguments are to provide the evidence for a 2nd-person indirect object.

Let us revisit some arguments that are of relevance to our later discussion. The first argument we would like to relate to is the existence of the 1st-person subject in the hidden clause when the visible clause contains picture-NPs such as picture of oneself, story or portrayal of oneself, etc. The argument is based on a similarity between the sentences in (42) and (43):

(42)	a. Tad knew that it would be a story about himself
	b. Mike will not believe that this is a photograph of himself.
	c. I promised Omar that it would be a poem about himself.
							(Ross 1970)

(43)	a. This is a picture of myself.
	b. (I Vtold you) this is a picture of myself.

The sentences in (42) feature a construction in which the reflexive pronoun himself embedded in a picture-NP can refer to the NP in the higher clause and where himself is anaphoric. If the performative analysis is adopted, the fact that myself in (43) is licit without an overt antecedent can be accommodated, given that there is in fact an antecedent, namely the 1st-person subject of the silent higher clause.

An argument for a silent 2nd-person antecedent is provided by the contrast in (44):

(44)	a.	Kick yourself.
	b.	*Kick themselves.

In order to capture this contrast, the structure must contain an element that may serve as antecedent for yourself, but not for themselves.

Furthermore, according to Ross, the silent verb must be a verb of saying which bears the features [+communication, +linguistic, +declarative]; otherwise, the sentence is ungrammatical. See (45):

(45)	Tomi said/declared/asserted/*laughed/*groaned/*snorted that Ann could swim,
	but nobody believed himi.

Thus, Ross’s performative analysis expresses that there is a silent performative clause in the highest position in every declarative sentence.

We are not the first to apply the performative hypothesis to Vietnamese. For earlier proposals, we refer to Trinh and Truckenbrodt (2018) and Trinh (2022). Trinh and Truckenbrodt employ the performative frame in order to derive the fact that, in Vietnamese, names cannot be bound. Trinh uses it in accounting for the fact that names can be used to refer to the speaker or the hearer.

As indicated above, we use the performative frame to account for the existence of a speaker interpretation for mình. Consider the structures in (46).

(46)	a.	Nami		khen	mình*i/sp.
		Nam		praise	body
		‘Nam praises *himself/me.’
	b.	( Tôisp	kể)	Nami	khen	mình*i/sp.
		(1sg	tell)	Nam	praise	body
		‘(I’m telling that) Nam praised me.’

As already noted, unlike Mandarin ziji, mình cannot be co-argument bound (in the variant we are discussing). Hence, in (46a) the local subject Nam is not available as an antecedent. Even though no 1st-person pronoun is realized, mình in (46a) gets a 1st-person interpretation. The question is how this interpretation is assigned. In fact, the availability of this interpretation follows straightforwardly if we adopt the performative hypothesis. As noted, we assume that every sentence in Vietnamese optionally contains a syntactically-expressed, but silent, first-person pronoun as the subject of a silent verb of saying or thinking (and a silent second-person indirect object, which we will not extensively discuss), as illustrated in (46b):

The interpretation of mình as a speaker now follows on the same footing as in (45). With the silent performative frame tôi kể ‘I tell’, mình receives the speaker value from the silent 1st-person subject. This, then, accounts for the availability of the speaker interpretation in the absence of an overt 1st-person antecedent.

As noted by an anonymous reviewer, mình can also be interpreted as the addressee, like Chinese ziji. However, as discussed in Pham (2002) this use is rather rare, limited to a very special context such as conversations between husband and wife or when used to address someone intimately. Moreover, like Chinese ziji, it seems to be limited to a structural configuration when there is no antecedent available in the sentence, as in (47):

(47)	Mình	ăn cắp	rồi	còn	đổ lỗi	cho người khác.
	body.add	steal	already	also	blame	for person another
	'You stole it but still blamed the others.'

In the next section, we will show in detail how Vietnamese non-local binding patterns can be accounted for.

8 Our analysis: the interaction of mình with the [+author] feature

In this section, we will systematically account for the following facts, including the differences between Vietnamese and Mandarin.

• i. In Vietnamese, a speaker interpretation of mình is in principle always available (with an exception to be discussed).

• ii. In Vietnamese, only a 1st-person intervener causes blocking, unlike in Mandarin where both 1st- and 2nd-person interveners cause blocking.

• iii. In Vietnamese, the intervener that causes the blocking can itself act as an antecedent of mình in its domain, whereas in Mandarin such an intervener cannot bind ziji.

• iv. In Vietnamese, mình in subject position is subject to blocking, unlike ziji in Mandarin.

• v. In Vietnamese, a [+author] element in object position does not cause blocking, unlike that in Mandarin where it does.

As we will see, property i. follows from the optional insertion of a performative frame and the assumption that, in Vietnamese, C⁰ probes for the [+author] feature rather than for the broader [+participant] feature that it probes for in Mandarin. This seems like a minimal stipulation needed to derive the differences. Thus, henceforth we will refer to the blocking effect in Vietnamese as the author effect.¹⁸

We will now derive property i. in more detail, and then proceed to illustrate the binding procedure more generally. We start with the proposal that, in Vietnamese, sentences are optionally embedded under a silent performative frame (Ross 1970), which is visible to the syntax, including a silent 1st-person subject. This entails that a sentence such as (48a), with indices omitted, may have the structure in (48b) with possible binding dependencies as indicated, or the structure of (48c). It is the latter structure that gives rise to the availability of the speaker value for mình (note that here and elsewhere in this section indexings given are relative to a derivation, so the range of interpretations available for (48a) is provided by (48b) together with (48c), in line with a comment by an anonymous reviewer).

(48)	a.	Nam	nghĩ	(rằng)	Hùng	biết	(rằng)
		Nam	think	(that)	Hung	know	(that)
		Thu	thích	mình.
		Thu	like	body
	b.	Nami	nghĩ	(rằng)	Hùngj	biết	(rằng)
		Nam	think	(that)	Hung	know	(that)
		Thu	thích	mìnhi/j.
		Thu	like	body
		‘Nam thought that Hung knew Thu likes him (= Nam, Hùng)’
	c.	[Tôisp kể [Nami	nghĩ	(rằng)	Hùngj	biết	(rằng)
		[1sg tell Nam	think	(that)	Hung	know	(that)
		Thu	thích	mìnhsp]]
		Thu	like	body
		(I told that) ‘Nam thought that Hung knew Thu likes me.’

As discussed in Sect. 3, the general mechanism of syntactic binding is based on Multiple Agree. Thus, as in Hirawai (2005) and Giblin (2016)’s approaches, one probe can agree with multiple goals simultaneously. The general structure of (48c) is as in (49). C⁰ initially has an unvalued [+author] feature, and so does mình. Hence C⁰ probes to value this feature. By assumption, the NP_valφ of the performative frame is 1st-person and will contain a valued [+author] feature as part of its feature specification. If so, it will value C⁰. C⁰ shares its value with mình by Multiple Agree, yielding (50).

(49)	[C0uφ [NPvalφ [T0uφ ..V.. [T10uφ…. mìnhuφ .... ]]]]→

(50)	[ C0valφ [NPvalφ [T0valφ..V.. [T10valφ… mìnhvalφ.... ]]]]

Chain formation is possible given that Vietnamese lacks an obligatory C⁰ introducing subordinate clauses.¹⁹ In line with Giblin's assumption for Mandarin, we assume that in the absence of intermediate complementizers, the dependency between C⁰ and elements lower in the structure is not blocked by phase boundaries (see also Bošković 2007). Thus, the silent first person qualifies as the NP_valφ in (48), and mình is valued as the speaker, as in (49). This, then, derives property i, according to which a sentence like (48) allows an interpretation of mình as the speaker.²⁰

However, there is another interpretation in which mình is valued as the 3rd-person expression Nam. This is achieved by an alternative derivation under which the silent performative frame is absent (recall we assumed that this frame is optional). So, consider again the structure of (49/50), but now the italic part is not silent and just corresponds to the initial part Nam nghĩ 'Nam think' of (48b). Here, Nam is valued for φ-features, but does not contain the [+author] feature. However, as discussed by Giblin, Preminger (2014) developed a theory of valuation in which the impossibility of achieving full valuation does not lead to a crash. Informally, it is possible to settle for a 'next best'. So, even if C⁰ probes for an [+author] feature, if it cannot find this feature on the NP_valφ, it settles for what it can find there, for instance, whatever represents a 3rd-person feature.²¹ Giblin develops this formally in terms of the feature structure for pronouns proposed by Béjar and Rezac (2003, 2009) as in (19), repeated here:

(19)

In (19) [π] is shared by all pronouns and other 3rd-person expressions. Therefore, and more precisely, what C⁰ in (49) settles for is [π]. This feature will be shared with mình and binding obtains. This accounts for the option in (48b) where Nam binds mình.²²

The next question is how to account for the option where mình is bound by the intermediate subject Hùng in (48b). A straightforward way to capture this, is to assume that the matrix verb nghĩ ‘think’ may optionally select for a CP. If so, the structure that allows the intermediate subject to act as a binder of mình is one in which the complement of the verb nghĩ ‘think’ is a CP with a C⁰ bearing an unvalued [+author] feature and the same procedure applies to that CP.

Summarizing, the reading of (48b) where Nam is the binder is illustrated in (51). The matrix C⁰ is merged. Upon merger, C⁰ probes its search domain for the [+author] feature, does not find it, but sees the 3rd person as next-best on the matrix subject NP Nam and gets valued. The T⁰ projections inherit the [π] feature from C⁰ and then share it with mình, which leads to its being bound.

(51)	[CP C0uφ→valφ	[Namvalφ	nghĩ	Hùngj	biết	Thu	thích	mìnhuφ→valφ=Nam]]
		Nam	think	Hung	know	Thu	like	body
		‘Nam thinks Hung knows Thu likes him.’

To see how the alternative is derived where the intermediate subject is the binder, consider (52). Here the verb nghĩ ‘think’ is taken to select a CP with a C head. Let’s refer to this intermediate C as C₁⁰. Thus C₁⁰ will have been merged in the intermediate clause. C₁⁰ gets valued by the embedded subject Hùng and transfers its [π] feature to mình, which results in the interpretation of mình as bound by Hùng.

(52)	Nam	nghĩ	[CP1 C10uφ→valφ	[Hùngvalφ	biết	Thu	thích	mìnhuφ→valφ=Hùng]]
	Nam	think		Hung	know	Thu	like	body
	‘Nam thinks Hung knows Thu likes him.’

Note that there will also be a matrix CP, but its C⁰ cannot probe beyond the embedded C₁⁰ due to minimality.²³ This, then, accounts for binding by an intermediate subject.

The next task is to account for property ii: In Vietnamese a 1st-person intervener causes blocking. To see how this works, consider a case like (53), where the first-person pronoun tôi intervenes. If we assign (53) the structure of (54), we have a configuration that violates Contiguous Agree (see Sect. 4) since Nam, not valued for the [+author] feature, intervenes between C⁰ probing for this feature and tôi that is valued for this feature (see the structure in (22) with part replaced by author). Since Contiguous Agree is a general formal constraint, it would not be plausible to assume that it would not apply in Vietnamese. Thus, under this structure, tôi is a blocker. Since Contiguous Agree is violated, no Agree chain is formed, which is indicated by _*→valφ in (54). Consequently, the first-person interpretation of mình is not derived either.

(53)	Nami	nghĩ	(rằng)	tôij	biết	(rằng)	Thu	thích	mình*i/j.
	Nam	think	(that)	1sg	know	(that)	Thu	like	body
	‘Nam thought that I knew Thu likes *him/me.’

(54)	[C0uφ*→valφ	[Namvalφ	nghĩ	tôi	biết	Thu	thích	mìnhuφ*→valφ=*Nam/*me]]
		Nam	think	1sg	know	Thu	like	body
		‘Nam thinks I know Thu likes *him/*me.’

Thus, property ii. has been derived.

Consider next, property iii. Unlike in Mandarin Chinese, the blocker can itself serve as an antecedent of mình. Note now that, given our assumptions, there is indeed another derivation, which does yield the intermediate 1st person as an antecedent. As we saw in the discussion of (52), Vietnamese has the option of merging an intermediate C. Consider again the sentence in (53) but now under the option of merging a C₁⁰ as the complement of the verb nghĩ ‘think’ as in (55):

(55)	[CP C0uφ→valφ	[Namvalφ	nghĩ	[CP C10uφ→valφ	[tôi+author	biết
		Nam	think		1sg	know
		Thu	thích	mìnhuφ→valφ=+author]]]
		Thu	like	body
		‘Nam thinks I know Thu likes *him/me.’

Under this option, mình is effectively valued by the intervening 1st person without a contiguity violation. Upon merger, C₁⁰ probes its search domain for an [+author] feature and finds the first-person pronoun tôi, which intrinsically bears the [+author] feature. C₁⁰ gets valued, thus the complement T₁ and embedded T⁰s inherit this feature from C₁⁰ and finally value mình. This configuration satisfies Contiguous Agree as there is no unmarked [author feature] intervening between the probe C₁⁰ and the [+author] tôi. As a result, mình ends up having the author interpretation. Note that this derivation says nothing about the binding possibilities of lower subjects. We know that in (53) and (55) Thu is not available as an antecedent of mình, but this is due to the fact that mình cannot be coargument bound. If the possibility of tôi as an antecedent in (55) is due to the optional presence of the complementizer C₁⁰, this would make us wonder what happens if the mình is further embedded, for instance as a possessive. If so, locality would not prevent Thu from binding mình. Let’s therefore consider the sentence in (56a). Assuming that complementizers can always be optionally inserted, one possibility is the structure in (56b).

(56)	a.	Nami	nghĩ	(rằng)	tôij	biết	(rằng)	Thuk
		Nam	think	(that)	1sg	know	(that)	Thu
		thích	khu	vườn	của	mình*i/j/k.
		like	CL	garden	of	body
		‘Nam thought that I knew Thu likes self's garden'.
	b.	Nami	nghĩ	[C10	[tôij	biết	[C20uφ→valφ	[Thuvalφ
		Nam	think		1sg	know		Thu
		thích	[khu	vườn	của	mìnhuφ→valφ]]]]]
		like	[CL	garden	of	body]
		‘Nam thought that I knew Thu likes self's garden'.

In this structure, C₂⁰ is a minimality barrier for C₁⁰. Given our reasoning so far, mediated by C₂⁰, the local subject Thu should be able to bind mình and, in fact, it does. This indicates that the analysis proposed is indeed on the right track. Note that the derivation of a case like *2> 1… in (36c) with mày as the matrix subject is no different from the cases of *3>1 … discussed here. Like a 3rd person matrix subject, a 2nd person matrix subject causes a contiguity violation. But merging an intermediate C licenses the 1st-person interpretation of mình.

For sake of completeness, consider next (57a) with tôi as the matrix subject. We may assume that the performative frame is absent as the first-person pronoun tôi bears the [+author] feature and the derivation proceeds as in the case of (57b). C⁰ is merged and gets valued by the first person pronoun tôi. The [+author] feature from C⁰ is inherited by the matrix T⁰ and the embedded T⁰ that, in turn, values mình yielding the first-person interpretation.

(57)	a.	Tôii	nghĩ		(rằng)	Namj	biết	(rằng)
		1sg	think		(that)	Nam	know	(that)
		Thu	thích		mìnhi/j.
		Thu	like		body
		‘I thought that Nam knew Thu likes me/him.’
	b.	[C0φ→valφ	[Tôivalφ=author	nghĩ	(rằng)	Nam	biết	(rằng)
			1sg	think	(that)	Nam	know	(that)
			Thu	thích	mìnhvalφ=author]]
			Thu	like	body
			‘I thought that Nam knew Thu likes me/him.’

As assumed, a C₁⁰ with an unvalued feature can also optionally be merged in the intermediate clause as in (58). As in the other cases discussed, C₁⁰ constitutes a minimality barrier for the matrix C⁰. It probes in its search domain and gets valued by the intermediate subject Nam. The T⁰ projections inherit the value from C₁⁰ and transfer to mình. As a result, mình can also take Nam as its potential antecedent.

(58)	[C0=unval	[Tôi	nghĩ	(rằng)	[C10uφ→valφ	[Namvalφ	biết	(rằng)
		1sg	think	(that)		Nam	Know	(that)
		Thu	thích	mìnhvalφ]]]]
		Thu	Like	body
		‘I thought that Nam knew Thu likes me/him.’

The approach we are exploring also allows us to account for the occurrence of the author effect when the first person pronoun tôi is a sub-commander serving as a possessor, as in (59a). The relevant options are shown in (59b) and (59c).

(59)	a.	Nami	nghĩ	[[tính	kiêu ngạo	của		tôij=author ]]
		Nam	think	CL	arrogance	of		I
		đã	hại	mình*i/j=author].
		PST	harm	body
		'Nam thought that my arrogance harmed me.’
	b.	[C0=unval	[Nami	nghĩ	[[tính	kiêu ngạo của		tôij=author]
			Nam	think	CL	arrogance of		I
			đã	hại	mình*i/j]]]
			PST	harm	body
	c.	Nami	nghĩ	[C10=unval	[[tính	kiêu ngạo của		tôij=author ]
		Nam	think		CL	arrogance of		I
		đã	hại	mình*i/j=author]]
		PST	harm	body

We assume that a sub-commanding tôi is available as a target for probing. As a consequence, the configuration in (59b) violates contiguity, and under that derivation neither Nam nor tôi will be able to bind mình. An alternative derivation is available if the complement of the verb nghĩ ‘think’ is headed by a C₁⁰ as in (59c). By assumption, C₁⁰ is underspecified for phi-features, in particular it has an unvalued author feature that it seeks to value. If the nearest NP in its c-command domain (its nearest goal) is inanimate, such as tính kiêu ngạo của tôi ‘my arrogance’ is, for example, it is considered an impossible source, and is therefore skipped for being invalid. The next available element is then its specifier. If the specifier would be tôi, that settles it. Thus, C₁⁰ finds tôi as a target, gets valued and now binds mình without a contiguity violation.²⁴

Consider now property iv. with mình a nominative anaphor. In this respect, Vietnamese differs from Mandarin. In Mandarin subject ziji is exempt and not sensitive to blocking. In Vietnamese subject mình is sensitive to blocking. Consider therefore (60):

(60)	a.	Hùngi	nghĩ	tôij=author	sẽ	thừa nhận	mình*i/j
		Hung	think	1sg	FUT	admit	body
		đã	chỉ trích	Mai.
		PST	criticize	Mai
		'Hung thought I would admit that I criticized Mai.’
	b.	[C0=unval	[Hùngi	nghĩ	[tôij=author	sẽ	thừa nhận
			Hung	think	1sg	FUT	admit
		[mình*i	đã	chỉ trích	Mai]]]
		body	PST	criticize	Mai
	c.	[Hùngi	nghĩ	[C10=unval	[tôij=author	sẽ	thừa nhận
		Hung	think		1sg	FUT	admit
		[mìnhj	đã	chỉ trích	Mai]]]]
		body	PST	criticize	Mai

The derivation will proceed as in the other cases we discussed. In the case of (60b), there will be a contiguity violation, but in the case of (60c), with an intermediate C₁⁰, mình will be bound by tôi. The simplest assumption to account for the difference between Vietnamese and Mandarin is that, unlike what Giblin assumes for Mandarin, T⁰ in Vietnamese has a residual phi-feature that enters in an agree-relation with mình in subject position, thus making it visible for probing and chain formation along the lines of the given derivation.

Finally, consider property v: In Vietnamese tôi in object position does not act as a blocker, while in Mandarin a [+participant] element does. In order to act as a blocker, an element must be visible for probing. Consider then the configurations in (33), repeated here as (61a, b):

(61)	a. Johni	nói	với	tôij	là	Nam	ghét	mìnhi/j.
	John	say	with	me	that	Nam	hate	body
	‘John said to me that Nam hates him/me.’
	b. Johni	luôn	nhắc	tôij	là	Nam	ghét	mìnhi/j.
	John	always	remind	me	that	Nam	hate	body
	‘John always reminds me that Nam hates him/me.’

To account for the pattern in (61a) it suffices to assume that the preposition với 'with' creates a domain that is opaque for probing, whereas the corresponding structure in Mandarin is not. The case of (61b) is perhaps less straightforward, but it suffices to assume that in Vietnamese oblique marked arguments carry a functional layer that protects them from probing.²⁵

This, then, derives the main patterns of non-local binding in Vietnamese listed at the beginning of this section with some minimal stipulations.

Let’s now come back to the issue left open in Sect. 6. That is, why does the presence of a second-person pronoun mày block a speaker interpretation of mình? See (62) for illustration:

(62)	Nami	nghĩ	màyj	biết	Mai	tấn công	mìnhi/j/*sp.
	Nam	think	2sg	know	Mai	attack	body
	‘Nam thinks you know Mai attacks him/you/*(me).’

Interestingly, kinship terms like em ‘younger brother/sister’, status terms such as bạn ‘friend’, and proper names with, prima facie, the same interpretation, namely that of the addressee, do not block the speaker interpretation. See (63a) and (63b):

(63)	a.	Nami	nghĩ	bạn/emj		biết	Mai
		Nam	think	friend.add/kin.younger.add		know	Mai
		tấn	công	mìnhi/j/sp.
		attack		body
		‘Nam thinks you know Mai attacks him/you/me.’
	b.	[Tôi kể [Nami	nghĩ	bạn/emj		biết	Mai
		tấn	công	mìnhi/j/sp.]]
		[I tell [Nam think friend.add/kin.younger.add know Mai attack body]]
		‘[I told that] Nam thinks you know Mai attacks him/you/me.’

This restriction has a rather different type of explanation. It is due to the fact that honorificity is an important factor in Vietnamese, and sentences must respect a certain degree of harmony in honorificity. The form mày reflects a high degree of informality, almost rudeness, whereas bạn ‘friend’ is rather neutral, and em 'younger brother' is more intimate than 'familiar'. The high degree of informality expressed by mày does not match with the degree of esteem a speaker is supposed to have for herself. Therefore, the presence of mày is incompatible with the presence of the performative frame. Consequently, the speaker interpretation, which depends on the presence of the performative frame, is absent in (62). From the opposite end, a high degree of formality, as one may find in status terms, has the same effect; see (64).

(64)	Nami	nghĩ	thầyj	biết	Mai	tấn công	mìnhi/j/*sp.
	Nam	think	stat.male teacher.add	know	Mai	attack	body
		‘Nam thinks you know Mai attacks him/you/*(me).’

Here, the status of a teacher appears to be too high for compatibility with the performative frame. As observed by an anonymous reviewer, this analysis predicts that embedding (62) under an overt performative frame has the same effect. As shown by the impossibility of a speaker interpretation in (65), this predication is borne out:

(65)	[Tôi kể	[Nami		nghĩ	màyj	biết	Mai	tấn công	mìnhi/j/*sp]]
	I	tell	Nam	think	2sg	know	Mai	attack	body
	‘[I told that] Nam thought you knew Mai attacked him/you/*me.’

We will conclude this contribution with a discussion of alternative forms of reference to the speaker, their role in blocking, and what this tells us.

9 Forms of reference to the speaker and blocking and conclusions

Vietnamese has a rich inventory of forms that can be used to refer to the speaker and the addressee. These include proper names like Hùng or Mai, kinship terms such as anh ‘elder brother’, em ‘younger brother/sister’, and status terms like thầy ‘male teacher’, etc²⁶. Our discussion here will be limited to expressions with a speaker role. There is a clear difference between such terms and pronominals. Pronominals are dedicated to a certain role. A form like tôi is always used for the speaker, never for the addressee or a third party. It is an important issue to what extent the use of non-pronominals in what one intuitively might understand as pronominal roles, is just a free discourse-based use, or whether it is somehow syntactically encoded. Interestingly, the blocking effect in Vietnamese may shed light on this issue. Under the account given, blocking is an effect that is intrinsically related to properties of feature chains, in the form of the contiguity requirement. If the use of non-pronominals in ‘pronominal’ roles were just a free discourse-based process, one would expect that non-pronominals would not give rise to intervention effects. However, they do. As illustrated in (66), not only the first person pronoun tôi can serve as an intervener in a blocking configuration, but also kinship terms, status terms, and proper names that are used to self-address can produce the blocking effect.

(66)	a.	Nami	nghĩ	anhj	đã	biết	Mai	không
		Nam	think	kin.elder brother.sp	PST	know	Mai	NEG
		tin	mình*i/j.
		trust	body
		‘Nam thought I knew Mai did not trust me/*him.’
	b.	Nami	nghĩ	thầyj	biết	cái	Mai	tố cáo
		Nam	think	stat.male teacher.sp	know	CL	Mai	denounce
		mình*i/j.
		body
		‘Nam thought I knew Mai denounced me/*him.’
	c.	Nami	nghĩ	Hùngj	biết	Mai	ghét	mình*i/j.
		Nam	think	sp	know	Mai	dislike	body
		‘Nam thought I knew Mai disliked me/*him.’

The contrast in (66a,b,c) shows that the kinship term anh, the status term thầy, and the proper name Hùng, referring to the speaker, all induce the blocking effect, prohibiting long-distance binding of mình by the matrix subject Nam, and leaving only the speaker/author value for mình. Since, clearly, these non-pronominals cannot have the value [+author] feature intrinsically, they must receive it from the frame in which they appear. This presupposes a left periphery that is at least as rich as assumed in works such as Delfitto and Fiorin (2011)—see the discussion in Reuland (2015) —and which allows such elements to obtain a valued [+author] feature by being linked to the relevant position in the left periphery. Further pursuing this issue would lead us beyond the scope of this article. It will therefore be left for future research.

In this article, we provided an account of the blocking effect in Vietnamese. We took as our starting point the approach developed in Giblin (2016) for Mandarin. The differences between the binding patterns of Vietnamese mình and Mandarin ziji follow from the factors outlined below:

• i. Mình is simplex, whereas ziji is complex.

• ii. In Mandarin, C⁰ searches for a valued [+participant] feature, whereas the feature searched for in Vietnamese is [+author].

• iii. In Mandarin, only the root clause has a C⁰, whereas in Vietnamese a C⁰ can optionally be merged to each complement clause.

• iv. Vietnamese allows the optional merger at the root of a performative frame containing a silent 1st-person subject pronoun.

Vietnamese has a rich system of non-pronominal forms, including proper names, kinship terms, and status terms that may receive a speaker value in interpretation. Prima facie, the blocking effect in Vietnamese appeared to be rather different from that in Mandarin, but once considered in detail, it turned out that the basis mechanism is quite similar to that in Mandarin, the difference being largely reducible to the factors in (ii), (iii) and (iv), above.

Our analysis of the blocking effect in Vietnamese and its similarities and differences with the blocking effect in Mandarin is testimony to the fruitfulness of a modular approach to the complexity of linguistic phenomena.