Antecedent-contained argument ellipsis in Japanese

Abstract

This article aims to provide new and solid evidence for the observation made in the previous literature that antecedent containment needs to be resolved in overt syntax in cases of antecedent contained deletion in Japanese. The evidence is new in the sense that it is based on argument ellipsis, which the recent literature has convincingly shown to be available in Japanese, rather than verb-stranding VP-ellipsis, the existence of which has been debated. It is shown how to construct relevant data using elliptic clausal arguments in the hope that similar research will be conducted with data from other languages where argument ellipsis is assumed to be available. It is argued that cases of antecedent-contained argument ellipsis in Japanese can be accounted for by the PF deletion analysis or the derivational copying analysis, but not by the LF copying analysis that posits that copying applies at the very end of the LF derivation.

1 Introduction

Does human language allow the kind of ellipsis, schematically shown in (1), where an elided part (XP₂) is contained in its antecedent (XP₁), or what has been called antecedent-contained deletion (ACD) in the literature?

(1)	… [XP1 … [XP2 e] …] …

This configuration is exemplified by cases like (2) cited from May (1985) and Lasnik (1999).¹

(2)	Dulles suspected everyone Angleton did.

Here the verb phrase (VP) inside the relative clause is elided and arguably takes the matrix VP as its antecedent. Then, (2) seems to have the following structure (throughout this article, elided elements are indicated with gray shading):

(3)	Dulles [VP1 suspected everyone Angleton did ]

Many researchers have studied cases like this (Baltin 1987; Bouton 1970; Fox 2002; Hornstein 1994; Kennedy 1997; Lasnik 1993; May 1985; Sag 1976, among many others). A consensus has been that the elided constituent needs to be dislocated somehow so that it can escape from its antecedent.

One influential analysis by May (1985) claims that quantifier raising (QR) is responsible for resolving antecedent containment, as illustrated below.

(4)	[TP [NP everyone Angleton did ] [TP Dulles [VP1 suspected t]]]

As the object of the main clause is a quantified phrase, it is subject to QR and is adjoined to TP at LF. In the resulting structure in (4), the elided VP is outside the antecedent VP and importantly, the two VPs have parallel structures, licensing ellipsis of VP₂.

One of the consequences of this analysis is that antecedent containment can be resolved covertly as QR is a covert operation. We can have cases of overt resolution like (5).

(5)	How many of the languages that Ron can does Harry speak?

The object of the entire sentence is a wh-phrase (how many of the languages that Ron can) here and hence undergoes wh-movement overtly. The QR analysis claims that QR can serve the same purpose as wh-movement though it takes place covertly. An issue arising from this line of analysis is whether resolution of antecedent containment can really be covert.

Baltin (1987) takes issue against the QR analysis, claiming that resolution must be overt as in (5). Baltin (1987) proposes an alternative analysis of (2) making use of extraposition, according to which the example is analyzed as below:

(6)	Dulles [VP1 suspected everyone] [CP Angleton did ]

In (6), the relative clause undergoes string-vacuous extraposition and is dislocated from the matrix VP. The two VPs, VP₁ and VP₂, are separated, so that the former can license the ellipsis of the latter.²

The purpose of the present article is to consider what Japanese can say about the issue of whether resolution of antecedent containment can be covert or must be overt. In order to do this, we need to have cases of ACD in Japanese. The previous literature on the topic argues that resolution must take place overtly in Japanese. Takahashi (1996a, b) and Abe (2019) construct cases of what they call ACD in Japanese assuming that some sort of VP-ellipsis is available in the language. Their data, however, must be subject to reconsideration, as explained in detail in section 2. In section 3 of this paper, we present a new kind of ACD in Japanese making use of ellipsis of clausal arguments. As the availability of argument ellipsis in Japanese is firmly grounded, we can provide clearer data for the research on ACD in general. In section 4 we examine whether resolution of antecedent containment can be covert in Japanese, ultimately showing that it must be overt, as argued by Takahashi (1996b) and Abe (2019) on the basis of less clear data. We observe that resolution is independent of quantifier scope, which is unexpected under the QR analysis. In section 5 we consider how cases of ACD in Japanese can be accounted for by the existent approaches to ellipsis, showing that, whereas the PF deletion analysis and the derivational copying analysis can handle them, the LF copying analysis, assuming that copying applies to very final LF representations, cannot. Section 6 concludes the whole discussion.

2 Background

In order to have a case of antecedent-contained deletion like (2) in a language, that language needs to allow VP-ellipsis. Concerning Japanese, Otani and Whitman (1991) argue that some cases of the null object construction in the language are analyzed as involving VP-ellipsis.³ Takahashi (1996a, b) points out that Otani and Whitman’s (1991) idea makes it possible to construct cases of ACD in Japanese.

Otani and Whitman (1991) first observe that (7b), if preceded by (7a), can have the so-called sloppy reading that Mary threw out her own letter, too.

(7)	a.	John-wa	zibun-no	tegami-o	sute-ta.
		John-top	self-gen	letter-acc	throw.out-past
		‘John threw out his letter.’
	b.	Mary-mo	[e]	sute-ta.
		Mary-also		throw.out-past
		‘lit. Mary threw out, too.’

The verb sute ‘throw out’ is a transitive verb in Japanese and hence (7b) contains a null object. Otani and Whitman argue that (7b) in fact involves VP-ellipsis, as shown below.⁴

(8)	a.	[TP John-wa [T' [VP zibun-no tegami-o tV] [T suteV [T ta]]]]
	b.	[TP Mary-mo [T' [T suteV [T ta]]]]

The crucial assumption here is that verbs move to T in Japanese, so that the main verb survives VP-ellipsis in (8b). The VPs in (8a–b) contain the traces of the verbs, which undergo movement to the position of Tense. Assuming also that (8b) has the object identical to the object in (8a), namely zibun-no tegami-o ‘self’s letter,’ we have identical VPs in (8a–b), allowing the VP in (8b) to be elided (ellipsis is indicated with gray shading).⁵ Note that the availability of the sloppy interpretation in (7b) is properly accounted for by the analysis in (8b).

Given this line of analysis, the example in (9b), which superficially appears to have a null indirect object and a null direct object, is also taken to involve VP-ellipsis.⁶

(9)	a.	John-wa	zibun-no	kodomo-ni	hon-o	yonde-age-ta.
		John-top	self-gen	child-dat	book-acc	read-give-past
		‘John read his child a book.’
	b.	Mary-mo	yonde-age-ta.
		Mary-also	read-give-past
		‘lit. Mary read, too.’

When (9b) has the sloppy reading that Mary read her own child a book, too, it is analyzed as involving VP-ellipsis, as shown below.

(10)	a.	[TP John-wa [T' [VP zibun-no kodomo-ni hon-o tV] [T yonde-ageV [T ta]]]]
	b.	[TP Mary-mo [T' [T yonde-ageV [T ta]]]]

The VP in (10b) is identical to the VP in (10a) and hence is eligible for ellipsis.

Then, Takahashi (1996a, b) points out that cases like (11) can be taken to involve ACD.⁷

(11)	John-ga/mo	zibun-no	kodomo-ni	[NP [RC	Mary-ga	[e]	yonde-age-ta]
	John-nom/also	self-gen	child-dat		Mary-nom		read-give-past
	subete-no	hon]-o	yonde-age-ta.
	all-gen	book-acc	read-give-past
	‘lit. John (also) read his child all the books that Mary read.’

This example is supposed to have the following structure, where the Japanese words are indicated with their glosses for the purpose of illustration:

(12)

The VP inside the relative clause is intended to take the matrix VP, which is indicated with a box, as its antecedent and be elided as indicated with gray shading. Because the elided VP is contained in the matrix VP, the sentence is a case of ACD. Assuming sloppy interpretation to be a crucial probe for VP-ellipsis, Takahashi (1996a, b) observes that (11) does not have the sloppy reading that John read John’s child all the books that Mary read to Mary’s child, drawing the conclusion that VP-ellipsis of the sort shown in (11) is not possible. This should be contrasted with the English counterpart below.

(13)	John read his child all the books Mary did.

This is a case of ACD in English and fully acceptable and ambiguous between the strict and sloppy interpretation.

Takahashi (1996a, b) goes on to point out that if the direct object containing the relative clause is preposed by scrambling, the sloppy reading becomes available.

(14)	[NP [RC	Mary-ga	[e]	yonde-age-ta]	subete-no	hon]-o	John-ga/mo
		Mary-nom		read-give-past	all-gen	book-acc	John-nom/also
	zibun-no	kodomo-ni		yonde-age-ta.
	self-gen	child-dat		read-give-past
	‘lit. All the books that Mary read, John (also) read his child.’

The availability of the sloppy reading in (14) means that VP-ellipsis is possible there. The example is assumed to have the following structure:

(15)

The NP containing the relative clause is adjoined to TP by scrambling (Saito 1985). As a result, the containment relationship between the two VPs is resolved. Comparing (11) and (14), Takahashi (1996b) concludes that ACD resolution must take place in overt syntax in Japanese.

Takahashi’s (1996a, b) argument presented above is based on Otani and Whitman’s (1991) hypothesis that VP-ellipsis is available in Japanese and thus is undermined by the reanalysis of the null object construction due to Oku (1998) and Kim (1998), which is now called the argument ellipsis analysis. These authors argue that in languages like Japanese, arguments, rather than verb phrases, are subject to ellipsis, as illustrated below.

(16)	a.	[TP	John-wa	[VP [NP	zibun-no	tegami]-o	sute-ta]]
			John-top		self-gen	letter-acc	throw.out-past
	b.	[TP	Mary-mo	[VP			sute-ta]]
			Mary-also		self-gen	letter-acc	throw.out-past

This is the analysis of (7a–b) in terms of argument ellipsis. The object in (16b) is identical to the object in (16a) and hence can be elided. Note that what is elliptic here is just the object NP, a consequence being that the phenomenon in question does not necessitate movement of verbs.

The argument ellipsis analysis is in part motivated by Oku’s (1998) observation that the null object construction in question does not allow a reading where an adjunct is understood. Consider the following examples cited from Oku (1998).

(17)	a.	Bill-wa	kuruma-o	teineini	arat-ta.
		Bill-top	car-acc	carefully	wash-past
		‘Bill washed a car carefully.’
	b.	John-wa	[e]	araw-anakat-ta.
		John-top		wash-neg-past
		‘lit. John did not wash.’

Anteceded by (17a), (17b) is a null object construction. Note that (17a) contains the adverb teineini ‘carefully.’ Oku (1998) notes that (17b) does not mean that Bill did not wash a car carefully, but just that Bill did not wash a car. The absence of the adjunct-including reading argues against the VP-ellipsis analysis, which would predict that the reading should be possible just as it is available in the following example in English:

(18)	Bill washed a car carefully, but John didn’t.

The second clause in (18) has the VP elided and can mean that John did not wash a car carefully. Under the argument ellipsis analysis, which assumes that arguments but not adjuncts are eligible for ellipsis, (17a–b) are analyzed as follows:

(19)	a.	[TP Bill-wa [VP kuruma-o teineini arat] ta]
	b.	[TP John-wa [VP araw] anakat-ta]

The object is elided in (19b), where the adverb is simply missing.

With the emergence of the argument ellipsis analysis, the cases that Takahashi (1996a, b) assumes to involve ACD need to be reconsidered. The relevant cases in (11) and (14) are now analyzed as in (20a–b), respectively.

(20)	a.	John-ga/mo	zibun-no	kodomo-ni	[NP [RC	Mary-ga
		John-nom/also	self-gen	child-dat		Mary-nom
					yonde-age-ta]	subete-no
		self-gen		child-dat	read-give-past	all-gen
		hon]-o	yonde-age-ta.
		book-acc	read-give-past
		‘lit. John (also) read self’s child all the books that Mary read .’
	b.	[NP [RC	Mary-ga
			Mary-nom	self-gen	child-dat
		yonde-age-ta]	subete-no	hon]-o	John-ga/mo	zibun-no
		read-give-past	all-gen	book-acc	John-nom/also	self-gen
		kodomo-ni	yonde-age-ta.
		child-dat	read-give-past
		‘lit. All the books that Mary read , John (also) read self’s child.’

The indirect objects in the relative clauses are subject to argument ellipsis. Notice that there is no containment relationship between the indirect objects in the main clauses and the ones in the relative clauses and hence that these should have nothing to do with ACD.

In fact, Sakamoto (2016) points out that (20a–b) can be subsumed under the generalization pointed out by Abe (2009) that elided arguments cannot be c-commanded by their antecedents.⁸ This is exemplified by the following sentence:

(21)	Ken-ga	zibun-no	hahaoya-ni	[Hana-ga	[e]	hon-o	okut-ta
	Ken-nom	self-gen	mother-dat	Hana-nom		book-acc	send-past
	to]	itta.
	that	tell-past
	‘lit. Ken told self’s mother that Hana sent a book.’

Here the indirect object is missing from the embedded clause. It might arise from argument ellipsis as shown below, where the indirect object in the embedded clause is intended to be elided under the identity with the matrix indirect object.

(22)	Ken-ga	zibun-no	hahaoya-ni	[Hana-ga
	Ken-nom	self-gen	mother-dat	Hana-nom	self-gen	mother-dat
	hon-o	okut-ta	to]	itta
	book-acc	send-past	that	tell-past
	‘lit. Ken told self’s mother that Hana sent a book.’

If this were possible, the embedded clause should have the sloppy reading that Hana sent her own mother a book. This reading is not available, showing that argument ellipsis cannot apply here. This conforms to the generalization above as the indirect object in the matrix clause c-commands the indirect object in the lower clause.

Returning to (20a–b), we notice that while the matrix indirect object c-commands the indirect object in the relative clause in (20a), the former does not c-command the latter in (20b). Hence, the contrast between (20a–b), which Takahashi (1996a, b) attributes to the presence or absence of antecedent containment, is subsumed under Abe’s (2009) generalization and can be accounted for by the theory of cyclic derivation as proposed by Sakamoto (2016).

Sakamoto’s (2016) reanalysis of what seem to be cases of ACD in Japanese is based on the views of Kim (1998) and Oku (1998) that the null object construction in question involves not V-stranding VP-ellipsis, but argument ellipsis. Since Oku (1998) argued against the availability of V-stranding VP-ellipsis on the basis of (17) and other similar data, it has been an issue in Japanese syntax whether it is really absent. Funakoshi (2016) argues that it is available, observing that there are speakers that can have the reading where the adjunct is understood in (17b). The crucial examples in (17) are analyzed as below.

(23)	a.	[TP	Bill-wa	[VP	kuruma-o	teineini	tV] [T	aratV-ta]]
			Bill-top		car-acc	carefully		wash-past
		‘Bill washed a car carefully’
	b.	[TP	John-wa	[NegPtNeg [T
			John-top		car-acc	carefully
			arawV-anakatNeg-ta]]
			wash-neg-past
		‘John did not wash ’

Funakoshi (2016) supports his analysis noting that the relevant adjunct-including reading becomes unavailable if the object in (23b) shows up overtly.

(24)	a.	Bill-wa	kuruma-o	teineini	arat-ta.
		Bill-top	car-acc	carefully	wash-past
		‘Bill washed a car carefully.’
	b.	John-wa	kuruma-o	araw-anakat-ta.
		John-top	car-acc	wash-neg-past
		‘lit. John did not wash a car.’

The object surfaces in (24b), where it is difficult to have the adjunct-including reading. The presence of the object indicates that VP-ellipsis does not apply to (24b) and hence that the sentence cannot be analyzed as involving the adjunct on the premise that adjuncts alone are not eligible for ellipsis (namely, that they are elidable only if they are contained in other constituents eligible for ellipsis).

Following Funakoshi (2016), Abe (2019) points out examples of ACD in Japanese that he assumes to involve V-stranding VP-ellipsis, using adjunct-including readings as a way to ensure its involvement. The following is a case in point:

(25)	Taroo-wa	subayaku	zibun-no	inu-ni	[NP [RC
	Taroo-top	quickly	self-gen	dog-dat
	Hanako-ga	tuketeage-nakat-ta]		kubiwa]-o	tuketeage-ta.
	Hanako-nom	put-neg-past		collar-acc	put-past
	‘Taroo put the collar Hanako didn’t on his dog quickly.’

Notice that the main clause contains the adverb subayaku ‘quickly.’ Abe (2019) considers whether it is understood in the relative clause as well. (25) is assumed to have the following structure, where the words are indicated with their English glosses for convenience:

(26)

The VP (shaded in gray) in the relative clause is intended to take the main VP (indicated with a box). As the former is contained in the latter, this is a case of ACD. And importantly, Abe (2019) observes that the adjunct cannot be understood in the relative clause, concluding that VP-ellipsis is not allowed in (25).

Then Abe (2019) points out that if antecedent containment is resolved by dislocating the NP containing the relative clause by cleft formation, the relevant interpretation becomes possible.

(27)	[CP	Taroo-ga	subayaku	zibun-no	inu-ni	tuketeage-ta
		Taroo-nom	quickly	self-gen	dog-dat	put-past
	no]-wa
	that-top
	[NP [RC	Hanako-ga	tuketeage-nakat-ta]	kubiwa-o]	da.
		Hanako-nom	put-neg-past	collar-acc	be
	‘It was the collar Hanako didn’t that Taroo put on his dog quickly.’

The cleft construction in Japanese consists of the presuppositional clause headed by the complementizer no ‘that,’ which is followed by the topic marker, and the focused element, which is followed by the copula (Hoji 1990; Hiraiwa and Ishihara 2012, among others). Abe (2019) observes that (27) can be understood to mean that Taroo quickly put on his dog the collar that Hanako did not quickly put on her dog. Comparing (25) and (27), he concludes that antecedent containment needs to be resolved overtly in Japanese.⁹

(25) and (27) are somewhat unfortunate examples, however, because they are a little too complex in that they need to be judged in terms of the adjunct-including interpretation as well as the sloppy interpretation. In addition, Abe (2019) notes that the antecedent containment in (25) cannot be resolved by scrambling the NP containing the relative clause, as shown below.

(28)	[NP [RC	Hanako-ga	tuketeage-nakat-ta]	kubiwa]-o	Taroo-wa
		Hanako-nom	put-neg-past	collar-acc	Taroo-top
	subayaku	zibun-no	inu-ni	tuketeage-ta.
	quickly	self-gen	dog-dat	put-past
	‘lit. The collar Hanako didn’t, Taroo put on his dog quickly.’

Abe observes that this does not have the adjunct-including reading or the sloppy reading, attributing it to his assumption that the relevant ellipsis process, namely V-stranding VP-ellipsis in Japanese, cannot take place backward.

This assumption can be contested as the literature has shown that ellipsis in general can take place forward or backward. Cases of backward ellipsis are found easily.

(29)	a.	Anyone who can [e] should speak English.
	b.	Ken-ga	[e]	hihansi-nakat-ta	node,	Hana-ga
		Ken-nom		criticize-past	because	Hana-nom
		zibunzisin-o	hihansi-ta.
		self-acc	criticize-past
		‘lit. Because Ken didn’t criticize, Hana criticized herself.’
	c.	Boku-ni-wa	[naze	[e]	ka]	rikai-deki-nai
		me-to-top	why		q	understand-can-neg
		ga,	Ken-ga	taigakusi-ta.
		though	Ken-nom	quit.school-past
		‘Though I cannot understand why, Ken quit school.’
	d.	Ken-no	[e]-wa	yoku-nakat-ta	ga,	Hana-no
		Ken-gen	-top	good-neg-past	though	Hana-gen
		taido-wa	rippa	dat-ta.
		attitude-top	good	be-past
		Though Ken’s was not good, Hana’s attitude was good.’

The example in (29a) involves backward VP-ellipsis in English. The first clause in (29b) can have the sloppy reading that Ken did not criticize himself and hence can involve backward argument ellipsis.¹⁰ (29c–d) are cases of sluicing (Takahashi 1994) and NP-ellipsis (Saito and Murasugi 1990) in Japanese, respectively, which take place backward. It is not very clear, therefore, whether the absence of the adjunct-including reading in (28) should be ascribed to backward ellipsis.

We may construct and examine simpler examples instead of (25), (27), and (28). The following examples may be relevant:¹¹

(30)	a.	Taroo-ga	subayaku	[NP [RC
		Taroo-nom	quickly
		Hanako-ga	tikaduk-e-nakat-ta]
		Hanako-nom	approach-can-neg-past
		hito]-ni	tikaduk-e-ta.
		person-dat	approach-can-past
		‘Taroo could quickly approach the person Hanako couldn’t.’
	b.	[CP	Taroo-ga	subayaku
			Taroo-nom	quickly
		tikaduk-e-ta]	no-wa	[NP [RC
		approach-can-past	that-top
		Hanako-ga	tikaduk-e-nakat-ta]	hito]-ni	da.
		Hanako-nom	approach-can-neg-past	person-dat	be
		‘It was the person Hanako couldn’t that Taroo could quickly approach.’
	c.	[NP [RC	Hanako-ga	tikaduk-e-nakat-ta]
			Hanako-nom	approach-can-neg-past
		hito]-ni	Taroo-ga	subayaku
		person-dat	Taroo-nom	quickly
		tikaduk-e-ta.
		approach-can-past
		‘lit. The person Hanako couldn’t, Taroo could quickly approach.’

Here the point is whether the adjunct subayaku ‘quickly’ can be understood in the relative clauses so that (30a), for instance, can mean that Taroo could quickly approach the person Hanako could not quickly approach. My own judgment is that while it is quite difficult to have the adjunct-including interpretation in (30a) and (30c), it is at best obscure whether the reading is really available in (30b).¹² It would be safe, therefore, to say that the use of adjunct-including construal may not be a very reliable way to construct cases of ACD in Japanese or to examine how ACD is resolved in the language.¹³

Of relevance to our discussion in the next section is Takita’s (2018) argument that null complement clauses can arise through argument ellipsis in Japanese. Takita considers cases of what he calls antecedent-contained clausal argument ellipsis such as the following:

(31)	a.	John-wa	[Mary-ga	[CP e]	it-ta	atode]	zibun-o
		John-top	Mary-nom		say-past	after	self-acc
		hihansi-ta.
		criticize-past
		‘lit. John criticized himself after Mary said.’
	b.	John-wa	[Mary-ga	[CP	kare-ga	zibun-o	hihansi-ta
		John-top	Mary-nom		he-nom	self-acc	criticize-past
		to]	it-ta	atode]	zibun-o	hihansi-ta.
		that	say-past	after	self-acc	criticize-past
		‘John criticized himself after Mary said that he criticized himself.’

In (31a), the bracketed part is an adjunct clause modifying the main clause. The verb itta ‘said’ in the adjunct clause has a null complement clause, whose content corresponds to the main clause. Thus, if spelled out, (31a) is represented as in (31b), where the complement clause means that he (John) criticized himself.

Takita (2018) analyzes (31a) as follows (the words are indicated with their English glosses just for convenience):

(32)

The adjunct clause as well as the matrix subject is adjoined to the main clause, so that there are three segments (CP₁, CP₂, and CP₃) of the matrix CP. The complement clause in the adjunct, indicated with gray shading, takes CP₃ as its antecedent and is elided by argument ellipsis.

Takita’s (2018) main concern is to argue that null complement clauses like the one in (31a) do not arise through V-stranding VP-ellipsis. Funakoshi (2014) argues that elliptic complement clauses must result from V-stranding VP-ellipsis, but Takita (2018) points out that it could not handle cases like (31a). According to the V-stranding VP-ellipsis analysis, (31a) should be analyzed as below rather than as in (32).

(33)

Inside the adjunct clause in (33), the verb say moves to T, and the VP containing the complement CP and the trace of the verb is assumed to be elided. Crucially, Takita (2018) notes, there is no appropriate VP in the main clause that could antecede the elided VP: the matrix VP does not contain a verb (or its trace) selecting a clausal complement.

Takita (2018) does not touch on the issue of antecedent containment directly: he just assumes that it needs to be resolved for relevant sentences to be grammatical, as shown in (32). Nonetheless, his argument that null complement clauses can arise through argument ellipsis will be important in our discussion in the next section.

To sum up, it is at best unclear whether there are clear and convincing cases of ACD in Japanese. The data constructed based on the assumption that V-stranding VP-ellipsis is available in Japanese may not serve the intended purpose. The assumption depends on the possibility of sloppy interpretation and adjunct-including construal, but it has turned out that sloppy interpretation can be handled without recourse to V-stranding VP-ellipsis and that the availability of adjunct-including construal varies among speakers. The lack of reliable empirical support in turn obscures the plausibility of the generalization made by Takahashi (1996b) and Abe (2019) that antecedent containment must be resolved in overt syntax. It is necessary, therefore, to find more reliable data involving ACD in Japanese that enables us to examine whether the generalization in question is really attested. The following two sections are devoted to addressing these issues.

3 Toward antecedent-contained argument ellipsis

Let us return to the typical case of ACD like (33a) with the relevant structure in (33b).

(33)	a.	Harry read every book Ron did.
	b.	Harry [VP1 read [NP every booki [RC Ron did ]]]

The constituents that are in a containment relationship here are the two VPs: VP₁ and VP₂. When one wishes to construct cases of ACD in Japanese, it is necessary to come up with a category other than VP because it is not clear whether VP is elidable in the language. Then we need to have sentences that have the structure indicated below, where XP ≠ VP.

(34)	… [XP1 … [NP [RC … [XP2 … ei …] …] Ni] …] …

The lower occurrence of XP, namely XP₂, contains a gap associate with the head noun of the relative clause and is contained in XP₁. XP₁ and XP₂ need to meet the following conditions:

(35)	a.	XP2 needs to be a constituent out of which relativization is possible.
	b.	XP2 needs to be of the category that can be elided.
	c.	XP2 needs to be contained in XP1, which is of the same category as XP2.

I show below that CP meets those conditions in Japanese.

Let us start by confirming that relativization can involve movement in Japanese. Ishii (1991) provides an argument based on reconstruction effects. While Ishii considers examples containing the reflexive karezisin ‘himself,’ we can observe the relevant effect with the reciprocal pronoun otagai ‘each other.’

(36)	[RC	Ken	to	Hana-ga	ei	hihansi-ta]	otagai-no	ronbuni
		Ken	and	Hana-nom		criticize-past	each.other-gen	paper
	‘each other’s papers Ken and Hana criticized’

The head of the relative clause contains the reciprocal, which can be bound by the subject of the relative clause Ken to Hana ‘Ken and Hana.’ Another indication of movement is the possibility of relativization of idiom chunks (see Morita 2013).

(37)	a.	Ken-ga	sono	mondai-ni	keri-o	tuke-ta.
		Ken-nom	that	issue-to	end-acc	attach-past
		‘Ken put an end to that issue.’
	b.	[RC	Ken-ga	sono	mondai-ni	ei	tuke-ta]	kerii
			Ken-nom	That	issue-to		attach-past	end
		‘the end Ken put to that issue’

The expression keri-o tuke ‘attach an end’ in (37a) is an idiomatic verb phrase in Japanese meaning ‘put an end.’¹⁴ The object part of the idiom can be relativized as in (37b).

Let us note that CPs or complement clauses allow relativization out of themselves, meeting condition (35a) (Ishii 1991, among others).

(38)	a.	[RC	Ken-ga	[CP	Hana-ga	ei	kai-ta	to]
			Ken-nom		Hana-nom		write-past	that
		omottei-ru/syoogensi-ta]	ronbuni
		think-npast/testify-past	paper
		‘the paper Ken thinks/testified that Hana wrote’
	b.	[RC	Ken-ga	[CP	Hana-ga	ei	migak-u	no]-o
			Ken-nom		Hana-nom		polish-npast	that-acc
		mokugekisi-ta]	kabini
		witness-past	vase
		‘the vase Ken witnessed Hana polish’

In (38a), relativization takes place out of the complement CP headed by the complementizer to ‘that.’ In (38b), the complement CP is headed by the complementizer no, which appears with perception verbs and factive verbs, and it contains a gap associated with the head noun.

The involvement of movement in cases like (38a–b) can be confirmed with the examples below.

(39)	a.	[RC	Rei-ga	[CP	Ken	to	Hana-ga	ei
			Rei-nom		Ken	and	Hana-nom
		hihansi-ta	to]	it-ta]	otagai-no	ronbuni
		criticize-past	that	say-past	each.other-gen	paper
		‘each other’s papers Rei said that Ken and Hana criticized’
	b.	[RC	Rei-ga	[CP	Ken-ga	sono	mondai-ni	ei
			Rei-nom		Ken-nom	that	issue-to
		tuke-ta	to]	omottei-ru]	kerii
		attach-past	that	think-npast	end
		‘the end Rei thinks that Ken put to that issue’

(39a) exhibits a reconstruction effect and (39b) is a case of relativization of an idiom chunk.

Complement clause CPs can be elided in Japanese, meeting the condition in (35b) (Takita 2018; Takahashi 2020). Consider (40) first.

(40)	a.	Ken-wa	[CP	Hana-ga	sono	supai-ni	at-tei-ta	to]
		Ken-top		Hana-nom	that	spy-dat	meet-asp-past	that
		syoogensi-ta.
		testify-past
		‘Ken testified that Hana met that spy.’
	b.	Rei-mo	[CP	Hana-ga	sono	supai-ni	at-tei-ta	to]
		Rei-also		Hana-nom	that	spy-dat	meet-asp-past	that
		syoogensi-ta.
		testify-past
		‘Rei testified that Hana met that spy, too.’
	c.	Rei-mo	[e]	syoogensi-ta.
		Rei-also		testify-past
		‘lit. Rei testified, too.’

(40a) is intended to antecede (40b–c). While the complement clause is repeated in (40b), it is omitted in (40c), which nonetheless has the same interpretation as (40b): the complement clause that Hana met that spy is understood in (40c).

A similar observation can be made with a different sort of complement clause. In (41), the complement clauses are headed by the complementizer no, which appears with perception verbs.

(41)	a.	Ken-wa	[CP	Hana-ga	sono	biru	kara
		Ken-top		Hana-nom	that	building	from
		deteku-ru	no]-o	mokugekisi-ta.
		come.out-npast	that-acc	witness-past
		‘Ken witnessed Hana come out of that building.’
	b.	Rei-mo	[CP	Hana-ga	sono	biru	kara
		Rei-also		Hana-nom	that	building	from
		deteku-ru	no]-o	mokugekisi-ta.
		come.out-npast	that-acc	witness-past
		‘Rei witnessed Hana come out of that building, too.’
	c.	Rei-mo	[e]	mokugekisi-ta.
		Rei-also		witness-past
		‘lit. Rei witnessed, too.’

(41a) serves to antecede (41b–c). The complement clause is repeated in (41b) but is omitted in (41c).

That the null complements in (40c) and (41c) can involve ellipsis is demonstrated by the fact that extraction is possible out of them. In the examples below, (42a) serves to antecede (42b–c), where the complement CP is repeated in (42b) but omitted in (42c).

(42)	a.	Ken-ga	[CP	Hana-ga	at-tei-ta	to]	syoogensi-ta
		Ken-nom		Hana-nom	meet-asp-past	that	testify-past
		no-wa
		that-top
		sono	supai-ni	da.
		that	spy-dat	be
		‘It was that spy that Ken testified that Hana met.’
	b.	Rei-ga	[CP	Hana-ga	at-tei-ta	to]	syoogensi-ta
		Ken-nom		Hana-nom	meet-asp-past	that	testify-past
		no-wa
		that-top
		kono	supai-ni	da.
		this	spy-dat	be
		‘It was this spy that Rei testified that Hana met.’
	c.	Rei-ga	[e]	syoogensi-ta	no-wa	kono	supai-ni	da.
		Ken-nom		testify-past	that-top	this	spy-dat	be
		‘lit. It was this spy that Rei testified.’

All the sentences here are cases of the cleft construction in Japanese, which is assumed to involve movement (Hoji 1990; Hiraiwa and Ishihara 2012). The focused elements sono supai-ni ‘that spy-dat’ and kono supai-ni ‘this spy-dat’ are associated with the gaps in the complement clauses in (42a–b), respectively. What is noteworthy is (42c), which minimally differs from (42b) in that the complement clause is omitted but has the same interpretation as (42b): namely, (42c) means that it was this spy that Rei testified that Hana met. This indicates the presence of hidden structure in (42c), as illustrated below.

(43)	Rei-ga						syoogensita	no-wa
	Ken-nom	Hana-nom		meet-asp-past		that	testify-past	that-top
	kono	supai-nii	da
	this	spy-dat	be
	‘lit. it was this spyi that Rei testified ’

The complement clause indicated with gray shading contains a trace associated with the focused element and is elided under the identity with the complement clause in the antecedent. This directly explains the fact that the focused element is interpreted as the object of atteita ‘met’ in (42c).

An argument of the same sort can be provided for the empty complement in (41c).

(44)	a.	Ken-ga	[CP	Hana-ga	deteku-ru	no]-o	mokugekisi-ta
		Ken-nom		Hana-nom	come.out-npast	that-acc	witness-past
		no-wa	sono	biru	kara	da.
		that-top	that	building	from	be
		‘It was from that building that Ken witnessed Hana come out.’
	b.	Rei-ga	[CP	Hana-ga	deteku-ru	no]-o	mokugekisi-ta
		Rei-nom		Hana-nom	come.out-npast	that-acc	witness-past
		no-wa	kono	biru	kara	da.
		that-top	this	building	from	be
		‘It was from this building that Rei witnessed Hana come out.’
	c.	Rei-ga	[e]	mokugekisi-ta	no-wa	kono	biru
		Rei-nom		witness-past	that-top	this	building
		kara	da.
		from	be
		‘lit. It was from this building that Rei witnessed.’

(44a) is obtained by applying cleft formation to (41a) so that the PP sono biru kara ‘from that building’ is focused and detached from the complement clause. (44a) can antecede (44b–c), both of which involve cleft formation as well. While the complement clause is realized overtly in (44b), it is omitted in (44c). Of importance is the fact that (44c) has the same meaning as (44b). This is directly accounted for if (44c) is analyzed as follows:

(45)	Rei-ga						mokugekisi-ta
	Rei-nom	Hana-nom		come.out-npast	that-acc		witness-past
	no-wa	kono	biru	karai	da
	that-top	this	building	from	be
	‘lit. it was from this buildingi that Rei witnessed ’

This shows that the empty complement clause in (44c) arises through ellipsis and has hidden syntactic structure, whereby the interpretation where the focused PP is associated with the predicate detekuru ‘come out’ is obtained.

Finally, let us observe that complement clauses (or CPs) can meet condition (35c): namely, that they can be embedded under complement clauses.

(46)	a.	Ken-wa/mo	[CP1	Hanai-ga	[NP	Rei-ga
		Ken-top/also		Hana-nom		Rei-nom
		[CP2
		kanozyoi-ga	at-tei-ta	to]	syoogensi-ta
		she-nom	meet-asp-past	that	testify-past
		supai]-ni	at-tei-ta	to]	syoogensi-ta.
		spy-dat	meet-asp-past	that	testify-past
		‘Ken (also) testified [CP1 that Hana met the spy Rei testified [CP2 that she met]].’
	b.	Ken-wa/mo	[CP1	Hanai-ga	[NP	Rei-ga
		Ken-top/also		Hana-nom		Rei-nom
		[CP2
		kanozyoi-ga	migak-u	no]-o	mokugekisi-ta
		she-nom	polish-npast	that-acc	witness-past
		kabin]-o	migak-u	no]-o	mokugekisi-ta.
		vase-acc	polish-npast	that-acc	witness-past
		‘Ken (also) witnessed [CP1 Hana polish the vase Rei witnessed [CP2 her polish]].’

(46a–b) may be difficult to process due to their complex structure: each of them has four clauses and involves center embedding. Nonetheless, they are grammatical and become understandable if they are read slowly, for example. Note that in each of (46a–b), the higher CP (CP₁) contains the lower CP (CP₂).

Having confirmed that complement CPs satisfy all the three conditions in (35), we are now ready to use them to construct cases of antecedent-contained argument ellipsis. The examples in (47a–b) are obtained by eliding the most embedded clauses (CP₂) in (46a–b), respectively.

(47)	a.	*	Ken-wa/mo	[CP1	Hana-ga	[NP	Rei-ga
			Ken-top/also		Hana-nom		Rei-nom
			[CP2 e]
			syoogensi-ta	supai]-ni	at-tei-ta	to]	syoogensi-ta.
			testify-past	spy-dat	meet-asp-past	that	testify-past
			‘lit. Ken (also) testified that Hana met the spy Rei testified.’
	b.	*	Ken-wa/mo	[CP1	Hana-ga	[NP	Rei-ga
			Ken-top/also		Hana-nom		Rei-nom
			[CP2 e]
			mokugekisi-ta	kabin]-o	migak-u	no]-o	mokugekisi-ta.
			witness-past	vase-acc	polish-npast	that-acc	witness-past
			‘lit. Ken (also) witnessed Hana polish the vase Rei witnessed.’

Significantly, (47a–b) are unacceptable when intended to have the same meanings as (46a–b), respectively.¹⁵ Namely, it is impossible or extremely difficult to understand (47a) as meaning that Ken (also) testified that Hana met the spy that Rei testified that she (Hana) met and to understand (47b) as meaning that Ken (also) witnessed Hana polish the vase Rei witnessed her (Hana) polish.

Of equal significance is the fact that (47a–b) are turned into acceptable sentences once the NPs containing the elided CPs are dislocated and placed in the initial positions of the sentences by scrambling.

(48)	a.	[NP	Rei-ga	[CP2 e]	syoogensi-ta	supai]i-ni	Ken-wa/mo
			Rei-nom		testify-past	spy-dat	Ken-top/also
		[CP1	Hana-ga	ti	at-tei-ta	to]	syoogensi-ta.
			Hana-nom		meet-asp-past	that	testify-past
		‘lit. The spy Rei testified, Ken (also) testified that Hana met.’
	b.	[NP	Rei-ga	[CP2 e]	mokugekisi-ta	kabin]i-o	Ken-wa/mo
			Rei-nom		witness-past	vase-acc	Ken-top/also
		[CP1	Hana-ga	ti	migak-u	no]-o	mokugekisi-ta.
			Hana-nom		polish-npast	that-acc	witness-past
		‘lit. The vase Rei witnessed, Ken (also) witnessed Hana polish.’

The preposed NPs in (48a–b) can be understood to mean the spy that Rei testified that Hana met and the vase that Rei witnessed Hana polish, respectively, showing that ellipsis is involved in the complement clauses indicated as CP₂.

For the sake of completeness, we may note that there is another possibility of applying scrambling to (47a–b). While the NPs containing the elided CPs are scrambled in (48), the higher CPs (namely, CP₁) may be able to be subject to scrambling. In that case, the following examples are obtained:

(49)	a.	*	[CP1	Hana-ga	[NP	Rei-ga	[CP2 e]	syoogensi-ta
				Hana-nom		Rei-nom		testify-past
			supai]-ni	at-tei-ta	to]i	Ken-wa/mo	ti	syoogensi-ta.
			spy-dat	meet-asp-past	that	Ken-top/also		testify-past
			‘lit. That Hana met the spy Rei testified, Ken (also) testified.’
	b.	*	[CP1	Hana-ga	[NP	Rei-ga	[CP2 e]	mokugekisi-ta
				Hana-nom		Rei-nom		witness-past
			kabin]-o	migak-u	no]i-o	Ken-wa/mo	ti	mokugekisi-ta.
			vase-acc	polish-npast	that-acc	Ken-top/also		witness-past
			‘lit. Hana polish the vase Rei witnessed, Ken (also) witnessed.’

The results are that (49a–b) do not show improvement in acceptability. This is natural because scrambling does not resolve antecedent containment in (49a–b), where the elliptic complement clauses (CP₂) are contained in their antecedents (CP₁).

We have observed in (48) that scrambling can help resolve antecedent containment. Let us note that cleft formation can do so, too. The following examples are obtained from (47a–b) by applying cleft movement to the NPs containing the elliptic gaps:

(50)	a.	Ken-ga	[CP1	Hana-ga	ti	at-tei-ta	to]
		Ken-nom		Hana-nom		meet-asp-past	that
		syoogensi-ta	no-wa	[NP	Rei-ga	[CP2 e]	syoogensi-ta
		testify-past	that-top		Rei-nom		testify-past
		supai]i-ni	da.
		spy-dat	be
		‘It was the spy Rei testified that Ken testified that Hana met.’
	b. ?	Ken-ga	[CP1	Hana-ga	ti	migak-u	no]-o
		Ken-nom		Hana-nom		polish-npast	that-acc
		mokugekisi-ta	no-wa	[NP	Rei-ga	[CP2 e]	mokugekisi-ta
		witness-past	that-top		Rei-nom		witness-past
		kabin]i-o	da.
		vase-acc	be
		‘It was the vase Rei witnessed that Ken witnessed Hana polish.’

In clear contrast to (47a–b), (50a–b) are acceptable with the relevant readings.¹⁶

Comparison of (47a–b) with (48a–b) and (50a–b) shows that elliptic complement clauses need to escape from their antecedents overtly: if they are contained in the antecedents in overt syntax, it results in unacceptable sentences.

4 Resolution of Antecedent Containment and Quantifier Scope

We have observed that antecedent containment needs to be resolved overtly in Japanese. The relevant examples considered in the last section have NPs that are interpreted as definite (see the English translations of (47a–b) and (48a–b), where the NPs containing the elliptic CPs are shown as definite expressions). To examine the relationship between resolution of antecedent containment and quantifier scope, let us change those NPs into quantificational expressions. The examples below are obtained by turning supai ‘spy’ in (47a) and kabin ‘vase’ in (47b) into universal expressions (every spy and all the vases, respectively).¹⁷

(51)	a.	*	Ken-wa/mo	[CP1	Hana-ga	[NP	Rei-ga
			Ken-top/also		Hana-nom		Rei-nom
			[CP2 e]
			syoogensi-ta	dono	supai]-ni-mo	at-tei-ta	to]
			testify-past	every	spy-dat-also	meet-asp-past	that
			syoogensi-ta.
			testify-past
			‘lit. Ken (also) testified that Hana met every spy Rei testified.’
	b.	*	Ken-wa/mo	[CP1	Hana-ga	[NP	Rei-ga
			Ken-top/also		Hana-nom		Rei-nom
			[CP2 e]
			mokugekisi-ta	subete-no	kabin]-o	migak-u	no]-o
			witness-past	all-gen	vase-acc	polish-npast	that-acc
			mokugekisi-ta.
			witness-past
			‘Ken (also) witnessed Hana polish all the vases Rei witnessed.’

The modification does not have any consequence on the acceptability of the examples: (51a–b) are still degraded.

Proponents of the QR analysis, according to which antecedent containment can be resolved by the covert operation of quantifier raising, would say that the unacceptability of (51a–b) is expected given that QR is usually clause-bounded. In order to place the elliptic CPs (CP₂) outside of the antecedent CPs (CP₁), the universal NPs would need to move into the matrix clauses. That this sort of long-distance QR is impossible is shown by the following examples:

(52)	a.	Darega-ga	[CP	Hana-ga	dono	supai-ni-mo
		someone-nom		Hana-nom	every	spy-dat-also
		at-tei-ta	to]	syoogensi-ta.
		meet-asp-past	that	testify-past
		‘Someone testified that Hana met every spy.’
	b.	Dareka-ga	[CP	Hana-ga	subete-no	kabin-o
		someone-nom		Hana-nom	all-gen	vase-acc
		migak-u	no]-o	mokugekisi-ta.
		polish-npast	that-acc	witness-past
		‘Someone witnessed Hana polish all the vases.’

In (52a–b), the universal quantifiers in the embedded object positions cannot take scope over the existential quantifiers in the matrix clauses: the only interpretation available is the one where the latter takes scope over the former. If long-distance QR were possible, the impossible readings would be available.

It turns out that (51a–b) can be improved by moving the quantified NPs into the matrix clauses overtly by scrambling, as shown below.

(53)	a.	[NP	Rei-ga	[CP2 e]	syoogensi-ta	dono
			Rei-nom		testify-past	every
		supai]i-ni-mo	Ken-wa/mo	[CP1	Hana-ga	ti
		spy-dat-also	Ken-top/also		Hana-nom
		at-tei-ta	to]	syoogensi-ta.
		meet-asp-past	that	testify-past
		‘lit. Every spy Rei testified, Ken (also) testified that Hana met.’
	b.	[NP	Rei-ga	[CP2	e]	mokugekisi-ta
			Rei-nom			witness-past
		subete-no	kabin]i-o	Ken-wa/mo	[CP1	Hana-ga
		all-gen	vase-acc	Ken-top/also		Hana-nom
		ti	migak-u	no]-o	mokugekisi-ta.
			polish-npast	that-acc	witness-past
		‘lit. All the vases Rei witnessed, Ken (also) witnessed Hana polish.’

These examples are as good as (48a–b), serving as another indication that antecedent containment needs to be resolved overtly in Japanese. But they have additional important implications for the relationship between quantifier scope and resolution of antecedent containment. It has been noted in the literature on scrambling in Japanese (Oka 1989; Tada 1993; Saito 2004; etc.) that while clause-internal scrambling can affect the scope of scrambled quantified phrases, long-distance scrambling cannot. Let us consider the examples below for illustration.

(54)	a.	Dareka-ga	dono	sensei-mo	sonkeisi-tei-ru.
		someone-nom	every	teacher-also	respect-asp-npast
		‘Someone respects every teacher.’ (∃ > ∀, *∀ > ∃)
	b.	Dono	sensei-mo	dareka-ga	t
		every	teacher-also	someone-nom
		sonkeisi-tei-ru.
		respect-asp-npast
		‘lit. Every teacher, someone respects.’ (∃ > ∀, ∀ > ∃)

When two quantifiers occur in the underlying order, the one in a higher position asymmetrically takes wide scope over the one in a lower position, a phenomenon known as scope rigidity. Thus, the existential quantifier in the subject position asymmetrically takes wide scope over the universal quantifier in the object position in (54a). If the universal quantifier is preposed over the existential quantifier by clause-internal scrambling as in (54b), the sentence becomes ambiguous with either quantifier being able to take wide scope over the other (Kuroda 1971; Kuno 1973).

Let us turn to (55), where two quantifiers belong to different clauses.

(55)	a.	Dareka-ga	[CP	Ken-ga	dono	sensei-ni-mo
		Someone-nom		Ken-nom	every	teacher-dat-also
		at-ta	to]	omot-tei-ru.
		meet-past	that	think-asp-npast
		‘Someone thinks that Ken met every teacher.’ (∃ > ∀, *∀ > ∃)
	b.	Dono	sensei-ni-mo	dareka-ga	[CP	Ken-ga         t
		Every	teacher-dat-also	someone-nom		Ken-nom
		at-ta	to]	omot-tei-ru.
		meet-past	that	think-asp-npast
		‘lit. Every teacher, someone thinks that Ken met.’ (∃ > ∀, *∀ > ∃)

In (55a), the existential quantifier in the matrix subject position asymmetrically takes wide scope over the universal quantifier in the embedded object position. (55b) is derived by moving the universal quantifier into the matrix clause by long-distance scrambling. The scopal interpretation does not change, however: the universal quantifier cannot have wide scope over the existential quantifier.

We have so far made two observations about long-distance scrambling: it can resolve antecedent containment but cannot affect quantifier scope. We can now examine whether the two — resolution of antecedent containment and quantifier scope — are related or not. The following examples are obtained from (51a–b) by replacing the matrix subjects with existential quantifiers:

(56)	a.	*	Dareka-ga	[CP1	Hana-ga	[NP	Rei-ga
			someone-nom		Hana-nom		Rei-nom
			[CP2 e]	syoogensi-ta	dono	supai]-ni-mo	at-tei-ta
				testify-past	every	spy-dat-also	meet-asp-past
			to]	syoogensi-ta.
			that	testify-past
			‘lit. Someone testified that Hana met every spy Rei testified.’
	b.	*	Dareka-ga	[CP1	Hana-ga	[NP	Rei-ga
			someone-nom		Hana-nom		Rei-nom
			[CP2 e]	mokugekisi-ta	subete-no	kabin]-o	migak-u
				witness-past	all-gen	vase-acc	polish-npast
			no]-o	mokugekisi-ta.
			that-acc	witness-past
			‘lit. Someone witnessed Hana polish all the vases Rei witnessed.’

As indicated, the modification does not affect the acceptability of the examples. They are as degraded as (51a–b).

Let us then apply long-distance scrambling to the universal quantifiers in (56a–b).

(57)	a.	[NP	Rei-ga	[CP2 e]	syoogensi-ta	dono
			Rei-nom		testify-past	every
		supai]i-ni-mo	dareka-ga
		spy-dat-also	someone-nom
		[CP1	Hana-ga	ti	at-tei-ta	to]
			Hana-nom		meet-asp-past	that
		syoogensi-ta.
		testify-past
		‘lit. Every spy Rei testified, someone testified that Hana met.’ (∃ > ∀, *∀ > ∃)
	b.	[NP	Rei-ga	[CP2 e]	mokugekisi-ta	subete-no
			Rei-nom		witness-past	all-gen
		kabin]i-o
		vase-acc
		dareka-ga
		someone-nom
		[CP1	Hana-ga	ti	migak-u	no]-o
			Hana-nom		polish-npast	that-acc
		mokugekisi-ta.
		witness-past
		‘lit. All the vases Rei witnessed, someone witnessed Hana polish.’ (∃ > ∀, *∀ > ∃)

The examples in (57a–b) are better than (56a–b) with each of them having one scopal reading. In (57a–b), the readings where the existential quantifiers have wide scope over the universal quantifiers are possible, but the reversed-scope interpretations are impossible. The last observation indicates that resolution of antecedent containment can be dissociated from quantifier scope. In (57a–b), the universal quantifiers containing the elided CPs need to move into the matrix clauses to have antecedent containment resolved but they take embedded scope.

There may be native speakers of Japanese who are not very certain about the interpretation of (57a–b). To have a clear judgment of those examples, which only have the readings where the existential quantifiers have wide scope over the universal quantifiers, it may be useful to compare them with data that permit the reversed scope interpretations. Let us first consider the following examples in English:

(58)	a.	Some girl said Harry admired every teacher.
	b.	Some girli said shei admired every teacher.

As is known in the literature, QR is clause-bounded. In (58a), the scope of every teacher is confined to the embedded clause, and the example only has the reading in which the existential quantifier has wider scope than the universal quantifier. Note that (58b) is minimally different from (58a) in the choice of the embedded subject: the embedded subject in (58b) is the pronoun she bound by the matrix subject. This difference has a significant influence on the scopal interpretation of the sentence, allowing the universal quantifier to have wide scope over the existential quantifier. Grano and Lasnik (2018) argue that when embedded clauses have pronominal subjects bound by the matrix subjects, they cease to function as phases, allowing otherwise local operations to occur across the clausal boundary. Thus, in (58b), the universal quantifier can undergo QR to the matrix clause and have scope over the existential quantifier.

Bearing this in mind, let us consider (59a–b).

(59)	a.	Darega-ga	[CP	Hana-ga	dono	supai-ni-mo	at-tei-ta	to]
		someone-nom		Hana-nom	every	spy-dat-also	meet-asp-past	that
		syoogensi-ta.
		testify-past
		‘Someone testified that Hana met every spy.’ (∃ > ∀, * ∀ > ∃)
	b.	Daregai-ga	[CP	zibuni-ga	dono	supai-ni-mo	at-tei-ta	to]
		someone-nom		self-nom	every	spy-dat-also	meet-asp-past	that
		syoogensi-ta.
		testify-past
		‘lit. Someone testified that self met every spy.’ (∃ > ∀, *∀ > ∃)

As observed with (55a), (59a) is scopally unambiguous, limited to the reading in which the existential quantifier has wider scope than the universal quantifier. (59b) is obtained by replacing the embedded subject Hana in (59a) with the reflexive pronoun zibun ‘self’ bound by the matrix subject. Unlike the English example in (58b), (59b) does not become ambiguous, which is expected because Japanese is a language with scope rigidity (see the discussions about (54) and (55)). Let us now apply scrambling to the embedded object and move it into the matrix clause:

(60)	Dono	supai-ni-mo	daregai -ga	[CP	zibuni -ga	t	at-tei-ta
	every	spy-dat-also	someone-nom		Hana-nom		meet-asp-past
	to]	syoogensi-ta.
	that	testify-past
	‘lit. Every spy, someone testified that self met.’ (∃ > ∀, ∀ > ∃)

This example minimally differs from (55b) in that the reflexive pronoun, instead of the name, occurs as the embedded subject. But it now enables the moved universal quantifier to have wide scope.

Bearing this in mind, let us modify (57a) and change the embedded subject into a reflexive pronoun bound by the matrix subject, as follows:

(61)	[NP	Rei-ga	[CP2 e]	syoogensi-ta	dono	supai]i-ni-mo	darekaj-ga
		Rei-nom		testify-past	every	spy-dat-also	someone-nom
	[CP1	zibunj-ga	ti	at-tei-ta	to]	syoogensi-ta.
		self-nom		meet-asp-past	that	testify-past
	‘lit. Every spy Rei testified, someone testified that self met.’ (∃ > ∀, ∀ > ∃)

This example allows the reading in which the universal quantifier takes wide scope over the existential quantifier as well as the reading in which the existential quantifier takes wide scope over the universal quantifier.¹⁸ If we compare (57a) with (61), it is clearly difficult to obtain the construal in which the universal quantifier takes wide scope over the existential quantifier in (57a). Again, (57a) indicates that resolution of antecedent containment can be independent of quantifier scope in Japanese.

5 Explaining the data

How can we account for the examples in (48), (53), and (57), where long-distance scrambling resolves antecedent containment? We assume with Saito (1989, 2003, 2004, and 2005) and Bošković and Takahashi (1998) that long-distance scrambling is subject to total reconstruction, which explains the fact noted in (55b) that long-distance scrambled quantifiers usually cannot take scope in the positions where they are moved.¹⁹ Let us consider (57a), repeated below as (62), as a representative case.

(62)	[NP	Rei-ga	[CP2 e]	syoogensi-ta	dono	supai]i-ni-mo	dareka-ga
		Rei-nom		testify-past	every	spy-dat-also	someone-nom
	[CP1	Hana-ga	ti	at-tei-ta	to]	syoogensi-ta.
		Hana-nom		meet-asp-past	that	testify-past
	‘lit. Every spy Rei testified, someone testified that Hana met.’ (∃ > ∀, *∀ > ∃)

In the view that ellipsis involves PF deletion, this has the following derivation, which is shown using the English glosses and word order just for convenience:

(63)	a.	(the pre-scrambling representation)
		someone testified [CP1 that Hana met [NP every spyi Rei testified [CP2 that Hana met ti]]]
	b.	(the post-scrambling representation)
		[NP every spyi Rei testified ]j, someone testified [CP1 that Hana met tj]
	c.	(the PF representation)
		[NP every spyi Rei testified [CP2 e]]j, someone testified [CP1 that Hana met tj]
	d.	(the LF representation)
		[TP someonej [TP tj testified [CP1 that Hana [VP [NP every spyi Rei testified [CP2 that Hana met ti]]j [VP met tj]]]]]

Long-distance scrambling of the embedded object yields (63b), which feeds into the PF and the LF component. The PF representation in (63c) is derived by deleting CP₂, indicated with gray shading in (63b). PF deletion can successfully locate the antecedent of CP₂, namely CP₁, as they are separated.²⁰ The LF representation in (63d) is obtained from (63b) by reconstructing the scrambled NP into the embedded clause and additionally applying QR to the quantified phrases (the matrix subject quantified phrase and the embedded object quantified phrase are adjoined to TP and VP, respectively). Note that (63d) only yields the reading in which the existential quantifier takes wide scope over the universal quantifier. There is nothing wrong with the derivation in (63), which means that the data in question can be explained by PF deletion.

Let us note that the counterpart of (62) without scrambling, namely (56a), can be accounted for easily by the PF deletion analysis. (56a) is repeated as (64), with its schematic derivation given in (65).

(64)	*	Dareka-ga	[CP1	Hana-ga	[NP	Rei-ga	[CP2 e]
		someone-nom		Hana-nom		Rei-nom
		syoogensi-ta	dono	supai]-ni-mo	at-tei-ta	to]	syoogensi-ta.
		testify-past	every	spy-dat-also	meet-asp-past	that	testify-past
		‘lit. Someone testified that Hana met every spy Rei testified.’

(65)	a.	(the overt syntactic representation)
		someone testified [CP1 that Hana met [NP every spyi Rei testified ]]
	b.	(the PF representation)
		someone testified [CP1 that Hana met [NP every spyi Rei testified [CP2 e]]]
	c.	(the LF representation)
		[TP someonej [TP tj testified [CP1 that Hana [VP [NP every spyi Rei testified [CP2 that Hana met ti]]j [VP met tj]]]]]

The representation in (65a) is obtained in overt syntax and it feeds into PF and LF. Notice that CP₂ is contained in its supposed antecedent (CP₁) in (65a), so that ellipsis of CP₂ cannot be allowed on the premise that antecedent-contained deletion is impossible without containment resolution. The LF representation in (65c) is obtained via QR and ends up as identical to (63d). But the sentence is ruled out because the deletion operation applying between (65a–b) is illicit.

Let us next consider how (62) can be handled by the LF copying analysis. Under the assumption that copying applies to the final output in the LF component (Williams 1977), it is analyzed in the following way:

(66)	a.	(the pre-scrambling representation)
		someone testified [CP1 that Hana met [NP every spyi Rei testified [CP2 e]]]
	b.	(the post-scrambling representation)
		[NP every spyi Rei testified [CP2 e]]j, someone testified [CP1 that Hana met tj]
	c.	(the PF representation)
		[NP every spyi Rei testified [CP2 e]]j, someone testified [CP1 that Hana met tj]
	d.	(the representation after reconstruction and QR)
		[TP someonej [TP tj testified [CP1 that Hana [VP [NP every spyi Rei testified [CP2 e]]j [VP met tj]]]]]
	e.	(the representation after copying)
		[TP someonej [TP tj testified [CP1 that Hana [VP [NP every spyi Rei testified [CP2 that Hana [VP [NP every spyi Rei testified]j [VP met tj]]]] [VP met tj]]]]]

The copying analysis assumes that the elliptic part is generated as empty in the syntactic component. Thus, CP₂ is just empty in (66a). It remains so after scrambling applies to the embedded object, as shown in (66b), which feeds into the PF and the LF component. In the PF representation in (66c), CP₂ remains empty. On the LF side, reconstruction and QR apply to (66b), deriving (66d). Now, to recover the empty CP, its antecedent (CP₁) is copied onto CP₂, yielding (66e). Notice that (66e) contains another occurrence of the empty CP, indicated with gray shading, which needs to be recovered by copying, resulting in infinite regress. This shows that the LF copying analysis with the assumption that copying applies to the final LF output cannot account for the cases where antecedent containment is resolved by long-distance scrambling.

Sakamoto (2016) puts forth the idea that copying applies derivationally in a phase-by-phase fashion. Specifically, he assumes that elements included in the domain that undergoes Transfer in the sense of Chomsky (2000) can be reused to construct another structure, and that reused elements do not have phonetic content because they have been transferred. Let us consider how this version of the copying analysis can deal with the relevant example in (62), repeated below as (67) with the English glosses and in the English word order.

(67)	[NP every spyi Rei testified [CP2 e]]j, someone testified [CP1 that Hana met tj]

As Sakamoto (2016) assumes that different clauses (or more precisely, phases) can be constructed independently and that relative clauses can be late-merged (Lebeaux 1988), we consider the derivations of the main clause and the relative clause separately.

The main clause is constructed as shown in (68).

(68)	(construction of the main clause)
	a.	[CP1 that Hana met every spy]
	b.	[CP1 every spyi that [TP Hana met ti]]
	c.	[vP someone v [VP testified [CP1 every spyi that [TP Hana met ti]]]]
	d.	[vP every spyi [vP someone v [VP testified [CP1 ti' that Hana met ti]]]]
	e.	[TP someonek T [vP every spyi [vP tk v [VP testified [CP1 that Hana met ti]]]]]
	f.	[TP every spyi [TP someonek T [vP ti“ [vP tk v [VP testified [CP1 that Hana met ti]]]]]]

First, the embedded clause (CP₁) is constructed in (68a), which contains just every spy in the object position as the relative clause is not introduced at this point. Since the object ultimately undergoes long-distance scrambling, it moves to the edge of the embedded CP in (68b).²¹ Then, the matrix vP is built in (68c), which is followed by (68d), where the embedded object moves to the edge of vP. Since vP is a phase, the matrix VP including CP₁ is transferred at this point. Subsequently, the matrix TP is built with raising of the subject in (68e) and scrambling (or TP adjunction) of the embedded object in (68f).

The relative clause is built as shown below.

(69)	(construction of the relative clause)
	a.	[VP testified ]
	b.	[vP Rei v [VP testified ]]
	c.	[TP Reim T [vP tm v [VP testified ]]]
	d.	[CP OP [TP Reim T [vP tm v [VP testified ]]]]

The verb testified needs to be combined with a CP. The occurrence of CP₁ included in the transferred VP in (68d) is reused, as indicated with gray shading in (69a). Note that this instance of copying avoids the problem of antecedent containment precisely because it copies an element in one structure onto a separate structure. Subsequently, vP and TP are formed in (69b) and (69c), respectively. Let us assume with Ishii (1991) that relativization involves the empty operator OP in Japanese. Then, it is directly base-generated at the edge of the relative clause, as indicated in (69d).²² The empty operator is ultimately coindexed with the head NP every spy in the main clause and its reused traces.

Then, the relative clause in (69d) is late-merged into the main clause in (68f), yielding (70), where the late-merged relative clause is indicated in boldface.

(70)	(late merger of the relative clause)
	[TP [NP every spyi [CP OP [TP Rei testified]]]
	[TP someone testified [CP1 that Hana met ti]]]

This feeds into PF and LF. With regard to the PF side, as the gray-shaded CP does not contain phonetic content, it is not pronounced, correctly giving rise to the actual pronunciation of the sentence. On the LF side, the scrambled NP ought to undergo reconstruction and the quantifiers are subject to QR. Therefore, the derivational copying analysis seems to be capable of accounting for the data where antecedent containment is resolved by long-distance scrambling.

One may wonder whether the derivational copying analysis can handle cases of ACD in Japanese where antecedent containment is not resolved overtly, such as (64), repeated below as (71) with the English glosses and word order.

(71)

*

someone testified [CP1 that Hana met [NP every spyi Rei testified [CP2 e]]]

The main clause and the relative clause can be built independently, as shown below.

(72)	(construction of the main clause)
	a.	[CP1 that Hana met every spy]
	b.	[vP someone v [VP testified [CP1 that [TP Hana met every spy]]]]
	c.	[TP someonek T [vP tk v [VP testified [CP1 that Hana met every spy]]]]

(73)	(construction of the relative clause)
	a.	[VP testified ]
	b.	[vP Rei v [VP testified ]]
	c.	[TP Reim T [vP tm v [VP testified ]]]
	d.	[CP OP [TP Reim T [vP tm v [VP testified ]]]]

As scrambling does not apply to the quantified NP every spy, it remains in situ in (72a–c). At the point when the matrix vP is formed in (72b), its complement VP is transferred. Because CP₁ is part of the transferred VP, it becomes available for reuse, and it is reused to form the relative clause, as shown in (73a), where it is merged with the verb testified. Then, the rest of the relative clause is built as shown in (73b–d). Note that as in (69), the empty operator is directly base-generated at the edge of the relative clause in (73d) and needs to form a chain with every spy contained in the reused CP. We can assume this to be permissible because the empty operator is ultimately associated with the head NP every spy in the main clause.²³

Now the relative clause in (73d) needs to be late-merged with the NP every spy in the main clause. There arises a problem here, according to Sakamoto (2016). At the point of the derivation, when the relative clause is late-merged with every spy in the main clause, the latter has already been transferred and hence cannot be targeted. This explains the ungrammaticality of (71) and the other similar cases of ACD in Japanese where antecedent containment is not resolved overtly.

To summarize, among the three approaches to ellipsis, the PF deletion analysis and the derivational copying analysis can explain the cases of ACD in Japanese. Importantly, the LF copying analysis that assumes that copying applies at the very final LF representation cannot account for them.

6 Conclusion

We have observed that it is at best unclear whether the cases of ACD in Japanese studied in the previous literature really count as involving ACD. As an alternative, we have proposed to take advantage of argument ellipsis, which the recent literature has shown convincingly to be available in the language. We have constructed cases of antecedent-contained argument ellipsis by means of elliptic complement clauses and arrived at the conclusion that antecedent containment must be resolved overtly in Japanese. While this has already been pointed out by Takahashi (1996b) and Abe (2019), we have demonstrated it with more reliable data. We have further argued that cases of antecedent-contained argument ellipsis in Japanese can be accounted for by the PF deletion analysis or the derivational copying analysis, but not by the LF copying analysis that posits that copying applies at the very end of the LF derivation.

Among the empirical contributions of the present article is the demonstration that cases of ACD can be constructed in Japanese by means of argument ellipsis without recourse to any sort of VP-ellipsis. While VP-ellipsis has a limited cross-linguistic distribution, argument ellipsis has been shown to be available more widely (Gribanova 2020; Kim 1999; Landau 2018; Sato 2014, 2015; Sato and Karimi 2016; Şener and Takahashi 2010; Simpson et al. 2013; Takahashi 2007, 2013, 2014, 2020; etc.). Hopefully we will be able to examine the findings of this article in a variety of languages with argument ellipsis to advance comparative research on ACD and shed new light on its issues. This is left for future research.