Revisiting Particle-Stranding Ellipsis: A Critical Comparison of Two Analyses

Abstract

This paper presents novel evidence that particle-stranding ellipsis in Japanese is best accounted for by PF-deletion rather than by its theoretical competitor, LF-copying. I begin by examining a central prediction of the LF-copying analysis, which states that overt extraction is categorically ruled out, and show that this prediction is not supported by the empirical data. Additional evidence comes from covert across-the-board movement, as I demonstrate that particle-stranding ellipsis can occur in environments that are argued to involve this type of movement. This finding presents a serious derivational challenge to the LF-copying theory, given the widely accepted view that covert across-the-board movement is not permitted in the grammar. In addition to these syntactic observations, I present previously unreported prosodic evidence showing that particle-stranding ellipsis involving the negative polarity item -sika can exhibit focus intonation. As the LF-copying analysis cannot account for this prosodic pattern, the data provide strong support for the PF-deletion account. Finally, I show that these findings are well explained by the phonology-based deletion model that was originally proposed in the literature.

1. Introduction

Particle-stranding ellipsis (PSE) constitutes a relatively recent addition to the inventory of Japanese ellipsis phenomena (Sato & Ginsburg, 2007; Nasu, 2012; Sato, 2012; Goto, 2014; Shibata, 2014; Sato & Maeda, 2019; Sakamoto & Saito, 2018; Takita, 2020; among others). As illustrated by the response to (1a) in (1b), this type of ellipsis typically occurs utterance-initially, leaving behind a particle such as a case marker or a focus particle:1

(1)	a.	John-ga	doo	sita-no?
		John-nom	how	did-q
		‘What did John do?’
	b.	-Ga	kaisya-o	yameta-rasii	yo
		-nom	company-acc	quit-seem	prt
		‘John quit his company.’		(adapted from Goto, 2014, p. 103)

The derivational nature of PSE has been the subject of considerable debate, partly because of this particular positional property. Two main theoretical approaches have emerged in the literature: LF-copying (Sakamoto & Saito, 2018) and PF-deletion (Shibata, 2014; Sato & Maeda, 2019). In this paper, I begin by testing a prediction derived from the LF-copying analysis, namely that overt extraction is disallowed. I then undertake a theoretical investigation of the mechanisms underlying PSE, with particular attention to prosody-oriented empirical observations. The novel data revealed here, such as the presence of focus intonation in certain instances of PSE, have significant implications for our understanding of how PSE operates, and strongly favor a PF-deletion analysis over its LF-copying counterpart. Finally, I show that these findings are well accounted for by the prosody-based approach originally proposed by Shibata (2014) and later refined by Sato and Maeda (2019).

The structure of this article is as follows: Section 2 provides an overview of PSE, with a particular focus on Sakamoto and Saito’s (2018) LF-copying analysis. Section 3 presents empirical counterarguments to this analysis, drawing on evidence from overt extraction and covert across-the-board (ATB) movement. Section 4 presents the novel data indicating that PSE sentences with a stranded -sika particle may display focus intonation, as discussed by Ishihara (2003, 2007), thereby providing new support for a PF-deletion analysis. I further demonstrate that the PF-deletion mechanism is best explained by the interaction between string deletion (Sato & Maeda, 2019) and the prosody-based licensing condition on PSE first proposed by Shibata (2014). Section 5 comprises the conclusion.

2. PSE and LF-Copying

Sakamoto and Saito (2018) argue that PSE involves ellipsis rather than the use of covert pro-forms, such as null pronouns. Their observation is based on the well-established claim that surface anaphora, such as VP-ellipsis, require an overt linguistic antecedent (Hankamer & Sag, 1976). Sakamoto and Saito note that PSE also requires an overt linguistic antecedent, as illustrated in (2):

(2)	[Context: Mary is a very cute girl, and every boy in her class has a crush on her. When Maryi enters the classroom, …]
	a.	[DP	Kanozyo]i-ga	kita!	b.	[DP pro]i	kita!	c.	*[DP ∆]i-ga	kita!
		she-nom		came			came		-nom	came
		‘Shei came!’				Lit. ‘[DP pro]i came!’			Lit.‘[DP∆]i-NOM came!’
(Sakamoto & Saito, 2018, p. 3)

In the given context, (2a) and (2b), which involve overt and null pronouns, respectively, are both acceptable. By contrast, the PSE variant in (2c) is ungrammatical. This grammatical asymmetry implies that PSE is not the result of a structure that involves phonetically null pronouns.

Sakamoto and Saito (2018) also claim that extraction is only expected to be possible from a PSE domain if it involves ellipsis, not a null pronoun. This is based on the uncontroversial observation that surface anaphora, despite lacking a phonological exponent, retains an internal syntactic structure. The contrast shown in (3) illustrates that overt extraction is only permitted from surface anaphora, which follows naturally from the presence of such an internal structure:

(3)	a.	I know which book1 Mary [VP read t1], and which book2 Mary didn’t [VP read t2].
	b.	* I know which book1 Mary [VP read t1], and which book2 Mary didn’t [VP do it].
		(Fiengo & May, 1994, p. 247)

If Sakamoto and Saito are correct in claiming that PSE involves surface anaphora, it follows that extraction from a PSE site should be possible, just as it is with VP-ellipsis, as in (3a). Inalienable possessor constructions, such as (4), serve as a useful testing ground for this prediction, for which Kikuchi (1994) claims that the genitive-marked possessor undergoes covert raising out of the object nominal, as shown in (4b):

(4)	a.	John-ga	kinoo	[DP Mary-no		tume]-o	kitta
		John-nom	yesterday		Mary-gen	nail-acc	cut
		‘(Lit.) John clipped [Mary’s nail] yesterday.’
	b.
		(Sakamoto & Saito, 2018, p. 352)

Kikuchi’s motivation for postulating covert raising comes from the well-known interpretative contrast regarding floating numeral quantifiers (FNQs), as shown in (5):

(5)	a.	Taro-wa		kuruma-o		2-dai	kowasita
		Taro-top		car-acc		2-cl	broke
		‘Taro broke 2 cars.’
	b.	*	Gakusei-tati-ga	[PP [DP	kuruma]-de]			2-dai	kita
			student-pl-nom		car-in			2-cl	came
		‘Students came in 2 cars.’			(Sakamoto & Saito, 2018, p. 352)

Given that an FNQ must be c-commanded by its associated NP (Miyagawa, 1989), the contrast in (5) suggests that only the bare DP, but not the PP-internal DP, can c-command the FNQ. Kikuchi then points out that the FNQ-nominal association is impossible when the object nominal consists of a simple nominal, as in (6a); however, it becomes possible when an inalienable possessum nominal occupies the object position, as shown in (6b):

(6)	a.	*	John-ga	[DP	tomodati-no	kuruma]-o	3-nin	norimawasita
			John-nom	friend-gen		car-acc	3-cl	drove.around
			‘John drove his friend’s 3 cars around.’
	b.	Hanako-wa		[DP	kodomo-tati-no	tume]-o	3-nin	kitta
		Hanako-top		child-pl-gen		nail-acc	3-cl	cut
		‘Hanako clipped 3 children’s nails.’				(adapted from Kikuchi, 1994, p. 82)

The ungrammaticality observed in (6a) is expected, as the c-command requirement is not satisfied at any point in the derivation. On the surface, (6b) is predicted to pattern similarly, as the relevant requirement also appears to be unmet. However, Kikuchi claims that genitive possessors within inalienable possessum nominals undergo covert raising, thereby allowing possessors like kodomo-tati-no, ‘children-PL-GEN’, to c-command their associated FNQ, ultimately meeting the c-command requirement.

Sakamoto and Saito (2018) show that genitive possessors undergo covert raising, even in cases in which PSE applies to inalienable possessum nominals, thus allowing for the FNQ-nominal association:

(7)	a.	Taro-wa		[DP	nezumi-no me]-o	50-piki	sirabeta-n-da	yo	ne?
		Taro-top			mouse-gen eye-acc	50-cl	examined-nml-cop	prt	prt
		‘Did Taro examine 50 mice’s eyes?’
	b.	[DP ∆]-o		100-piki	sirabeta-n-da	yo
			-acc	100-cl	examined-nml-cop	prt
		‘He examined 100 mice’s eyes.’				(adapted from Sakamoto & Saito, 2018, p. 353)

With (7a) serving as the antecedent and involving covert possessor raising, the inalienable possessum nominal in (7b) undergoes PSE, leaving only the accusative case marker. Of note, the FNQ-nominal association is still available in (7b), suggesting that the possessor covertly moves out of the PSE domain and c-commands the FNQ from that position.

Sakamoto and Saito (2018) then argue for an LF-copying analysis, in which a phonetically unrealized part of PSE is covertly retrieved from an antecedent discourse. On this basis, (7b) undergoes the following derivation:

(8)	a.	[DP mouse-gen eye]-acc 100-cl examined-nml-cop	prt
	b.		prt

In (8a), the missing domain is covertly copied via LF-copying. The possessor nezumi, ‘mouse’, is then raised out of the phonetically missing DP, thereby satisfying the c-command requirement for the FNQ, as indicated by the dotted arrow.

Based on LF-copying, the availability of covert extraction, as illustrated in (8b), is naturally expected. But what about overt extraction? This theory makes a strong prediction concerning the overt/covert distinction, namely, that overt extraction is predicted to be categorically ruled out (see also Sakamoto, 2020). In this regard, it may be beneficial to consider Sakamoto and Saito’s (2018, p. 352, fn. 2) comment that ‘[w]hether overt extraction is possible out of a PSE domain is hard to check due to the fact that PSE is possible only in the “string-initial” position’. Notwithstanding this remark, the authors suggest that the universal quantificational interpretation associated with the focus particle -mo offers indirect support for their prediction that overt extraction is impossible. To examine this, consider (9). The focus particle -mo yields a universal quantificational reading when it is associated with a wh-phrase (Kuroda, 1965; Nishigauchi, 1990; Shimoyama, 2001; among others), as shown in (9a). This reading remains available even when intervening material separates the two elements, as in (9b):

(9)	a.	Taro-wa	dono	kuruma-mo	aratta
		Taro-top	which	car-mo	washed
		‘Taro washed every car.’
	b.	[[Dare-ga	syookaisita]	gakusei]-mo	kasikokatta
		who-nom	introduced	teacher-mo	was.smart
		‘The student(s) that everybody introduced was smart.’

Based on this, Sakamoto and Saito observe that a stranded -mo resists the intended universal quantificational interpretation, even when the preceding discourse contains a wh-phrase:2

(10)	a.	[DP [Dare-ga		kaita]	hon]-mo	omosiroi	no?
		who-nom		wrote	book-mo	interesting	q
		‘Is [DP the book [that everybody wrote]] interesting?’
	b.	[DP [Dare-ga		kaita]	hon]-mo	omosiroi	yo
		who-nom		wrote	book-mo	interesting	prt
		‘[DP The book [that everybody wrote]] is interesting.’
	b′.	*	[DP ∆]-mo	omosiroi		yo
			[DP ∆]-mo	interesting		prt
		Lit. ‘[DP ∆]-mo is interesting.’			(adapted from Sakamoto & Saito, 2018, p. 354)

Sakamoto and Saito argue that their LF-copying theory, when combined with D. Takahashi’s (2002) particle-movement analysis, provides a neat explanation of this interpretative fact. In Takahashi’s analysis, -mo is base-generated, together with its associated wh-phrase, and undergoes overt movement to its surface position. According to Takahashi’s analysis, (10a) undergoes the following derivation:

(11)	a.	[[who-mo-nom wrote] book] interesting
	b.
		(Adapted from Sakamoto & Saito, 2018, p. 355)

Given this line of analysis, the ungrammaticality of (10b) follows naturally from the LF-copying theory, since no overt extraction from a covertly structured ellipsis domain is allowed.

Although the LF-copying-based theory appears to be appealing, I argue that a closer examination reveals that PSE does, in fact, allow overt extraction. If this is indeed the case, it directly contradicts a central prediction of the LF-copying theory, indicating the need for an alternative formulation.

3. Challenges to LF-Copying

Against the backdrop of Sakamoto and Saito’s (2018) LF-copying theory of PSE, this section presents a set of arguments that pose significant empirical challenges to the analysis in question. Section 3.1 illustrates that overt extraction does involve PSE, contrary to the prediction according to the LF-copying theory. In Section 3.2, I argue that the LF-copying account encounters a derivational problem when covert ATB movement is considered in depth.

3.1. Overt Extraction

Let us begin with overt extraction. Contrary to the prediction of the LF-copying analysis, in this subsection, I demonstrate that overt extraction from PSE sites is possible. The first case to consider is raising-to-object constructions (Hiraiwa, 2001; Tanaka, 2002; among others), as illustrated in (12):

(12)	[TPTaro-ga	[vPAi-oi	kokorokara	[CP ti	baka-da	to]	omotta]]
	Taro-nom	Ai-acc	really		foolish-cop	c	thought
	‘Taro really considered Ai to be a fool.’

The CP-internal accusative argument Ai is moved to the matrix domain in (12), appearing beyond the matrix adjunct kokorokara, ‘really’. When this sentence is reformulated as a yes-no question, as in (13a), the answer in (13b) now undergoes PSE:

(13)	a.	[TPTaro-ga	[vP Ai-oi	kokorokara [CP ti	baka-da	to]	omotta]]	no?
		Taro-nom	Ai-acc	really	foolish-cop	c	thought	q
		‘Did Taro really consider Ai to be a fool?’
	b.	[[CPTo]j [TP	Ken-ga [vP	Akemi-oi	kokorokara	tj	omotta]]]	yo
		c	Ken-nom	Akemi-acc	really		thought	prt
		‘Ken really considered Akemi to be a fool.’

In (13b), the embedded CP moves to the sentence-initial position via scrambling, ultimately leaving the overt C head to behind. The sentence undergoes the following derivational steps, which are informally represented below:

(14)	i.	[TP really [CP Akemi-acc foolish-cop c] thought ] prt
	ii.
	iii.
	iv.	[[CP ___ foolish-cop c] [TP Ken-nom Akemi-acc really ___ thought ]] prt

Beginning with the base structure in (14i), (14ii) demonstrates that the accusative argument Akemi moves from the embedded CP into the matrix clause. In (14iii), the CP is scrambled across the matrix subject Ken, and finally, in (14iv), it is elided via PSE. These derivational steps suggest that the accusative-marked Akemi is extracted from the clausal argument that ultimately undergoes PSE, suggesting that the clause in question displays an overt internal structure.3

Raising-to-subject constructions (Hiraiwa, 2001, 2010, among others) in (15), together with the raising-to-object constructions, reinforce the current conclusion. The steps in their derivation are shown below:

(15)	a.	[TP Taro-gai	minna-ni		[CP ti	baka-da		to]	omow-arete-iru]	no?
		Taro-nom	everyone-dat			foolish-pres		c	think-pass-pres	q
		‘Is Taro thought to be stupid by everyone?’
	b.	[[CP To]j	[TP	Mary-gai	minna-ni		tj	omow-arete-iru]]		yo
		c		Mary-nom	everyone-dat			think-pass-pres		prt
		‘Mary is thought to be stupid by everyone.’

(16)	i.	[TP everyone-dat [CP Mary-nom foolish-pres c] think-pass-pres] prt
	ii.
	iii.
	iv.	[[CP ___ foolish-pres c] [TP Mary-nom everyone-dat ___ think-pass-pres]] prt

As shown in (16i), the matrix subject Mary is first base-generated within the embedded CP. Later in the derivation, it moves to the matrix Spec, TP position in (16ii), after which the CP scrambles over it in (16iii). Once displaced, the CP undergoes PSE in (16iv), producing the surface structure in (15b). These observations clearly point to overt extraction from the PSE sites, casting serious doubt on the LF-copying analysis, which does not predict surface representations such as (13b) and (15b).

Yet another argument regarding overt extraction is based on PP-internal N′-deletion, which is first developed by Takita and Goto (2013). Example (17a) illustrates a canonical case of N′-deletion (Saito & Murasugi, 1990); (17b) shows that N′-deletion is still possible when a quantifier functions as its remnant, an observation that is crucial to Takita and Goto’s proposal:

(17)	a.	Taro-no	taido-ga	yoi	ippou	Hanako-no	Δ-ga	yokunai
		Taro-gen	attitude-nom	good	while	Hanako-gen	-nom	not.good
		Lit. ‘Taro’s attitude is good while Hanako’s Δ is not good.’
	b.	Sannin-no	sensei-ga	kita	ippou	gonin	Δ-ga	kaetta
		three-gen	teachers-nom	come	while	fine	-nom	left
		Lit. ‘three teachers came while fine Δ left.’
		(Adapted from Takita & Goto, 2013, p. 221)

Consider (18), in which PP-internal N′-deletion takes place:

(18)	a.	Kinseizin-wa	[ kinoo-no	kaseizin-no	koogeki	kara]	seikansita
		Venusians-top	yesterday-gen	Martians-gen	attack	from	survived
		Lit. ‘Venusians survived [from yesterday’s attack by Martians]’
	b.	Suiseizin-wa	[kyoo-no	Δ	kara]	toosoosita
		Mercurians-top	today-gen		from	run.away
		Lit. ‘Mercurians run away [from today’s Δ]’
	c.	Suiseizin-wa	[kyoo-no	Δ	e]	taiousita
		Mercurians-top	today-gen		to	responded
		Lit. ‘Mercurians responded [to today’s Δ]’
	d.	Suiseizin-wa	[kyoo-no	Δ	de]	hiheisita
		Mercurians-top	today-gen		with	got.exhausted
		Lit. ‘Mercurians got exhausted [with today’s Δ]’
		(Adapted from Takita & Goto, 2013, pp. 217–218)

With (18a) as the antecedent for (18b–d), while the postposition stranded in (18b) is identical to that in (18a), this is not the case in (18c–d). This suggests that PP-internal N′-deletion tolerates mismatches in postposition. Turning to (19), we observe that quantifiers can surface as remnants of PP-internal N′-deletion:

(19)	a.	Kinseizin-wa		[subete-no	kaseizin-no	koogeki kara]		seikansita
		Venusians-top		all-gen	Martians-gen	attack	from	survived
		Lit. ‘Venusians survived [from all attacks by Martians]’
	b.	Suiseizin-wa		[hotondo	Δ	kara]	toosoosita
		Mercurians-top		most		from	run.away
		Lit. ‘Mercurians run away [from most Δ]’
	c.	*	Suiseizin-wa	[hotondo	Δ	e]	taiousita
			Mercurians-top	most		to	responded
			Lit. ‘Mercurians responded [to most Δ]’
	d.	*	Suiseizin-wa	[hotondo	Δ	de]	hiheisita
			Mercurians-top	most		with	got.exhausted
			Lit. ‘Mercurians got exhausted [with most Δ]’
		(adapted from Takita & Goto, 2013, p. 218)

With (19a) as the antecedent, (19b), which employs the same postposition as (19a), is grammatical. In contrast, (19c–d), where different postpositions are stranded, are ungrammatical. Based on the contrast between (18) and (19), Takita and Goto conclude that PP-internal N′-deletion is blocked when (i) a quantifier serves as a remnant and (ii) the stranded postposition differs from that of the antecedent.

Takita and Goto’s (2013) analytical basis follows M. Takahashi’s (2011) formulation of N′-deletion, according to which the NP structure differs slightly depending on whether an ellipsis remnant includes a quantifier. Accordingly, the relevant structure for (17a) is (20a), while the quantifier-stranding version in (17b) corresponds to (20b), in which the K head overtly moves to Q, and the Spec-head licensing condition on the ellipsis is satisfied as a result.4

(20)	a.		b.
				(adapted from Takita & Goto, 2013, p. 221)

Let us next consider how M. Takahashi’s (2011) analysis of N′-deletion accounts for the puzzling pattern of PP-internal N′-deletion above. Assuming that Japanese postpositions, such as case markers, belong to the K category, Takita and Goto (2013) analyze the elided structure in (19) as in (21), in which the K head undergoes overt movement to Q, thereby licensing the ellipsis of its KP complement:

(21)	a.		(=(19a))
	b.		(=(19b))
	c.		(=(19c/d))
		(Takita & Goto, 2013, p. 222)

The derivation in (21) suggests that the head movement of K only occurs when a quantifier remains after N′-deletion. Given that this process is akin to V-stranding VP-ellipsis, according to Takita and Goto, it is naturally expected that such N′-deletion is constrained by the well-known rigid identity condition on ellipsis (Goldberg, 2005), which requires the remnant (i.e., the head) to be identical to the antecedent. The ungrammaticality of (19c–d) thus follows, since the heads that undergo movement differ between the antecedent (19a) and the targets (19c/d). The same explanation applies to (18). The structures involved are as follows:

(22)	a.	[KP kinoo-no [NP kaseizin-no koogeki] kara]	(=(18a))
	b.	[KP kinoo-no [NP kaseizin-no koogeki] kara]	(=(18b))
	c.	[KP kinoo-no [NP kaseizin-no koogeki] e/de]	(=(18c/d))
		(Takita & Goto, 2013, p. 223)

In the absence of a QP projection, head movement does not take place. Consequently, the rigid identity condition on ellipsis does not come into play, making postpositional mismatch possible. On this basis, Takita and Goto maintain that the structures and derivations outlined in (21) and (22) are empirically well justified.

With these elements in place, let us examine whether overtly raised postpositions are compatible with PSE. If they are, this would suggest that there is an internal structure within the PSE site, thereby arguing against the LF-copying analysis. The answer is in the affirmative:

(23)	a.	[Subete-no	kaseizin-no		koogeki	kara]	kinseizin-wa	seikansita	no?
		all-gen	Martians-gen		attach	from	Venusians-top	survived	q
		‘Did Venusians survive from all attacks by Martians?’
	b.	-Kara	pro (zannennagara)			seikansinakatta		yo
		-from		unfortunately		not.survived		prt
		‘(Unfortunately) they did not survive from all attacks by Martians.’

With (23a) as the antecedent, (23b) shows that the postposition -kara, ‘from’, which survives N′-deletion, can undergo PSE. Moreover, various other postpositions and case markers also occur in post-ellipsis positions via PSE, which further confirms the current argument:

(24)	a.	[ Subete-no	kaseizin-no	koogeki-de]	doseizin-wa	hiheisita	no?
		all-gen	Martians-gen	attack-with	Saturnians-top	got.exhausted	q
		‘Did Saturnians get exhausted with all attacks by Martians?’
	b.	-De	pro	hiheisita	wakedewanai	yo
		-with		got.exhausted	it.is.not.the.case	prt
		‘It is not the case that they got exhausted with all attacks by Martians.’

(25)	a.	[Subete-no	kaseizin-no	koogeki-e]	doseizin-wa	taioosita	no?
		all-gen	Martians-gen	attack-to	Saturnians-top	responded	q
		‘Did Saturnians respond to all attacks by Martians?
	b.	-E	pro	taioosita	yo
		-to		responded	prt
		‘They responded to all attacks by Martians.’

(26)	a.	[Subete-no	kaseizin-no	koogeki-ga]	doseizin-o	nayamaseta	no?
		all-gen	Martians-gen	attack-nom	Saturnians-acc	annoyed	q
		‘Did all attacks by Martians annoy Saturnians?’
	b.	-Ga	suiseizin-o	nayamaseta-nda	yo
		-nom	Mercurians-acc	annoyed-cop	prt
		‘They annoyed Mercurians.’

(27)	a.	[Subete-no	kaseizin-no	koogeki-o]	doseizin-ga	yarisugosita	no?
		all-gen	Martians-gen	attack-acc	Saturnians-nom	withstood	q
		‘Did Saturnians withstand all attacks by Martians?’
	b.	-O	suiseizin-ga	yarisugosita-nda	yo
		-acc	Mercurians-nom	withstood-cop	prt
		‘Mercurians withstood them.’

Examples (23) and (24–27) above indicate that ellipsis sites are equipped with full-fledged syntactic structures, leading to the ellipsis sites not being reconstructed via LF-copying.

Incidentally, data from N′-deletion provides further grounds for questioning the LF-copying theory. In their discussion of instances of N′-deletion such as (17a), which I repeat below, Saito and Murasugi (1990) argue that the -no in the pre-ellipsis position is not pronominal:

(28)	Taro-no	taido-ga	yoi	ippou	Hanako-no	Δ	-ga	yokunai
	Taro-gen	attitude-nom	good	while	Hanako-gen		-nom	not.good
	Lit. ‘Taro’s attitude is good while Hanako’s Δ is not good.’
	(adapted from Takita & Goto, 2013, p. 221)

Building on Kamio’s (1983) observation that the pronominal -no can only replace concrete and not abstract nouns, as shown in (29), Saito and Murasugi conclude that the -no in (28) functions as a genitive case marker rather than as a pronoun:

(29)	a.	[katai sinnen-o		motta]	hito
		firmconviction-acc		have	person
		‘a person with a firm conviction’
	b.	*	[katai no-o motta] hito	(adapted from Saito & Murasugi, 1990, p. 287)

Taking this a step further, Saito et al. (2008) argue that the genitive case marker -no is inserted in accordance with the rule formulated by Kitagawa and Ross (1982).5

(30)	Mod-Insertion:6
	[NP … XP Nα] → [NP … XP Mod Nα], where Mod = no.
	(adapted from Saito et al., 2008, p. 249; cf. Kitagawa & Ross, 1982, p. 23)

As suggested by Kitagawa and Ross (1982), Saito et al. (2008), and others, the insertion rule operates at PF, which suggests that linear order is essential for example (30). However, according to the LF-copying analysis, no-stranding PSE examples such as the one below appear to be problematic, as they lack any preceding phonetic material in the PF component:

(31)	a.	Taro-wa	Hati-no	nintaizuyosa-ni	kandousita	no?
		Taro-nom	Hati-gen	patience-gen	impressed	q
		‘Was Taro impressed by Hati’s patience?’
	b.	-No	tyuuseisin-ni	kandousita-n-dayo
		-gen	loyalty-gen	be.impressed-nml-cop.prt
		‘He was impressed by Hati’s loyalty.’

In (31b), which follows (31a), the sentence-initial genitive case marker does not occur in the environment designated in (30) in the PF component and thus, never satisfies the condition for no-insertion. The grammaticality of (31b) is therefore unexpected according to the LF-copying theory, as a covertly recovered structure, by definition, lacks phonological content.7

3.2. Covert Across-the-Board Movement

An additional challenge to the LF-copying-based analysis of PSE is due to covert ATB movement. Drawing on examples such as (32), Bošković and Franks (2000) persuasively argue that such covert movement does not exist in the grammar.

(32)	a.	Some delegate represented every candidate and nominated every candidate.
	b.	* Who said [that John bought what] and [that Peter sold what]?
		(Bošković & Franks, 2000, pp. 113, 110, respectively)

In (32a), every is not allowed to take wide scope over some, suggesting that covert movement from each conjunct is prohibited. The ungrammaticality of (32b) further demonstrates that what cannot undergo covert ATB extraction from each conjunct, even though English generally permits partial wh-in-situ in multiple wh-questions. Considered together, these examples indicate that covert ATB movement is not a legitimate grammatical option.

Against this background, we now turn to left-node raising constructions in Japanese, as illustrated in (33a), which represent the mirror image of right-node raising in English (33b). Two sentences are conjoined in such constructions, with their shared object DP (e.g., keeki, ‘cake’, in (33a)) being fronted:

(33)	a.	Keeki-o	John-ga	tukuri,	(sosite)	Mary-ga	tabeta
		cake-acc	John-nom	make	and	Mary-nom	ate
		‘The cake, John made, and Mary ate.’
	b.	John made, and Mary ate the cake.				(Nakao, 2010, p. 156)

As Nakao (2010) points out, it is logically possible that the ‘shared’ object is not structurally shared; instead, each gap may have a distinct grammatical source: The object in the first conjunct could undergo scrambling, while the second could involve a phonologically null pronoun, options that are freely available in Japanese, as shown in (34):

(34)
	(adapted from Nakao, 2010, p. 156)

Although Japanese left-node raising constructions can be reduced to the interaction of scrambling and a phonetically null pronoun in this scenario, Nakao argues that this does not tell the entire story, and proposes that constructions such as (33a) can be better explained by an ATB movement analysis. Let us briefly review her arguments.

Left-node raising constructions are reported to resist case mismatch between the two gap positions, as illustrated in (35).

(35)	a.	??Mary-ni	John-ga	hana-o	okuri,	Tom-ga	nagusameta
		Mary-dat	John-nom	flower-acc	send,	Tom-nom	comforted
		‘(To) Mary, John sent a flower, and Tom comforted.’
	b.	??Mary-o	John-ga	dansu-ni	sasoi,	Tom-ga	raburetaa-o	kaita
		Mary-acc	John-nom	dance-to	invite,	Tom-nom	loveletter-acc	wrote
		‘(To) Mary, John invited to a dance, and Tom wrote a love letter.’
		(Nakao, 2010, p. 157)

In (35a), okuru, ‘send’, takes a dative object, while nagusameru, ‘comfort’, takes an accusative one. Similarly, in (35b), sasou, ‘invite,’ is an accusative-assigning predicate, and kaku, ‘write’, assigns a dative case. Both examples are degraded. However, as (36) illustrates, null object constructions with the same predicates do not show the same degradation, which points to an important difference from left-node raising constructions:

(36)	a.	Mary-ni	John-ga	hana-o	okutta.	Tom-wa	pro	nagusameta
		Mary-dat	John-nom	flower-acc	sent	Tom-top		comforted
		‘John gave a flower to Mary. Tom comforted (her).’
	b.	Mary-o	John-ga	dansu-ni	sasotta.	Tom-wa	pro	rabu	retaa-o	kaita
		Mary-acc	John-nom	dance-to	invited	Tom-top		love	letter-acc	wrote
		‘John invited Mary to a dance. Tom wrote a love letter (to her).’
		(Nakao, 2010, p. 157)

Another of Nakao’s (2010) observations concerns distributive scope. As Abels (2004) notes, the shared objects in English right-node raising constructions allow for a distributive interpretation. In particular, (37a), but not (37b), can be interpreted to mean that the song that John sang and the song that Mary recorded were two different songs:

(37)	a.	John sang, and Mary recorded, two quite different songs.
	b.	John sang two quite different songs, and Mary recorded two quite different songs.
		(Abels, 2004, p. 51)

It is intriguing that left-node raising in Japanese also permits a distributive interpretation. In (38), the shared object hutatsu-no betsubetsu-no kyoku, ‘two separate songs’, can be interpreted as John sang one song, Mary recorded another, and the two songs were different:

(38)	Hutatu-no	betubetu-no	kyoku-o	John-ga	utai,	Mary-ga	rokuonsita
	two-gen	separate-gen	song-acc	John-nom	sing,	Mary-nom	recorded
	‘Two separate songs, John sang, and Mary recorded.’				(Nakao, 2010, p. 159)

Of note, the pro-occurring counterpart in (39) does not allow for the distributive reading. Consequently, the sentence is interpreted as referring to the same two songs:8

(39)	Hutatu-no	betubetu-no	kyoku-o	John-ga	utatta.		Mary-ga		pro	rokuonsita
	two-gen	separate-gen	song-acc	John-nom	sang		Mary-nom		recorded
	‘John sang two separate songs. Mary recorded (them).’							(Nakao, 2010, p. 159)

With these observations in mind, Nakao argues that left-node raising constructions are best explained by ATB movement. According to her analysis, the derivation in (40) is assigned to (33a):

(40)

Nakao’s ATB movement analysis finds crosslinguistic support from Polish, in which the ATB movement requires that the case of both gap positions be uniform:

(41)	a.	Coacc		Jan	lubi	tacc	i	Maria	uwielbia tacc?
		what		Jan	likes		and	Maria	adores
		‘What does Jan like and Maria adore?’
	b.	*	Coacc	Jan	lubi	tacc	i	Maria	nienawidzi tgen?
			what	Jan	likes		and	Maria	hates
			‘What does Jan like and Maria hate?’						(Citko, 2003, p. 89)

Bearing Nakao’s (2010) ATB analysis in mind, we now examine (42), in which (42a) is a yes-no question modeled after Nakao’s original example in (33a), and the response is given via PSE in (42b):

(42)	a.	Keeki-o	John-ga	tukuri,	(sosite)	Mary-ga	tabeta	no?
		cake-acc	John-nom	make	and	Mary-nom	ate	q
		‘Did John make the cake and Mary eat it?’
	b.	-O	John-ga	tukuri,	(sosite)	Hanako-ga	tabeta(-n-desu)
		-acc	John-nom	make	and	Hanako-nom	ate-nml-cop
		‘John cooked the cake and Hanako ate it.’

Given the grammaticality of (42b), the LF-copying analysis must derive it using covert ATB movement, as shown in (43):

(43)	i.	John-nom cake-acc make, (and) Hanako-nom cake-acc ate(-nml-cop)
	ii.

After the objects cake are covertly copied into each relevant position in (43i), they are moved to the sentence-initial position via covert ATB movement in (43ii). However, the derivation goes against the conclusion reached by Bošković and Franks (2000), who argue that ATB movement cannot take place covertly.9

In this section, I have shown that the core prediction of the LF-copying theory of PSE (i.e., the ban on overt extraction from a PSE site) is not supported. Furthermore, I have demonstrated that PSE is fully compatible with the ATB environment, contrary to Bošković and Franks’ (2000) widely accepted conclusion that ATB movement cannot occur covertly. If the arguments in this section are correct, they significantly undermine the explanatory power of the LF-copying analysis.

4. PSE Meets Focus Intonation

In this section, I present a previously unobserved prosody-oriented fact: PSE can exhibit focus intonation when accompanied by the focus particle -sika. In addition to the empirical arguments discussed thus far, this prosodic fact provides compelling support for the PF-deletion analysis of PSE (Shibata, 2014; Sato & Maeda, 2019) instead of the LF-copying alternative (Sakamoto & Saito, 2018). I further demonstrate that Sato and Maeda’s (2019) string deletion, when combined with Shibata’s (2014) phonology-based licensing condition on PSE, provides a coherent explanation of the precise mechanism underlying PF-deletion.

4.1. Focus Intonation

Considerable attention has been devoted to focus intonation in Japanese (Nagahara, 1994; Deguchi & Kitagawa, 2002; Ishihara, 2003, 2007; Sugahara, 2003, among others). Of note, Ishihara (2003) observes that a key trigger for focus intonation is found in wh-interrogative sentences. Focus intonation in Japanese consists of the following two prosodic phenomena:

(44)	Focus Intonation Pattern (FIP) in Japanese:
	a.	Prosodic-focalization:
		The pitch register of an item bearing focus is raised.
	b.	Post-focal reduction:
		The pitch register of the domain after the item bearing focus is lowered.
		(adapted from Ishihara, 2003, p. 33)

Now, consider the minimally different pair of declarative and interrogative sentences in (45), together with their corresponding pitch tracks from Ishihara (2003):

(45)	a.	Naoya-ga	nanika-o	nomiya-de	nonda
		Naoya-nom	something-acc	bar-loc	drank
		‘Naoya drank something at the bar.’
	b.	Naoya-ga	nani-o	nomiya-de	nonda	no?
		Naoya-nom	what-acc	bar-loc	drank	q
		‘What1 did Naoya drink t1 at the bar?’			(adapted from Ishihara, 2003, p. 52)

The interrogative sentence in (45b) differs from the declarative sentence in (45a) in that it contains nani, ‘what’, instead of nanika, ‘something’, as the object, and ends with the Q-particle no, which marks the interrogative scope. Unlike the example in Figure 1 with its steady declination, Figure 2 shows that the object has a focus-induced pitch prominence, followed by a pitch reduction (highlighted in gray). These prosodic characteristics, which Ishihara (2003) terms the FIP, are particularly important when considered in relation to the FIP-wh-scope correspondence stated below:

(46)	FIP-wh-scope correspondence:
	The domain of FIP corresponds to the scope of wh-question.	(Ishihara, 2003, p. 59)

Wh-question sentences in Japanese canonically mark their interrogative scope with the assistance of Q-particles such as no and ka. As Ishihara (2003) observes, the scope domain in question falls under the domain of focus intonation, consisting of the F₀-rise on a focused item (i.e., a wh-phrase) and the post-focus reduction that follows it. In other words, the scope domain that is syntactically marked by a Q-particle is also marked prosodically through FIP. Concrete examples of this are shown in (47) and (48), in which the former is the matrix wh-question sentence, and the latter is the embedded type:

(47)	Naoya-wa	[Mari-ga	nani-o	nomiya-de	nonda	to]	imademo	omotteru	no?
	Naoya-top	Mari-nom	what-acc	bar-loc	drank	c	even.now	think	q
	‘What1 did Naoya still think that Mari drank t1 at the bar?’

(48)	Naoya-wa	[Mari-ga	nani-o	nomiya-de	nonda	ka]	imademo	oboeteiru
	Naoya-top	Mari-nom	what-acc	bar-loc	drank	q	even.now	remember
	‘Naoya still remembers what1 Mari drank t1 at the bar.’						(Ishihara, 2003, pp. 53–55)

The pitch tracks associated with these examples are provided below as Figure 3 and Figure 4:

In (47), the wh-phrase takes the matrix scope, which is marked by the Q-particle no. In contrast, the wh-phrase in (48) takes the embedded scope, which is also marked by the Q-particle, in this case, ka. The pitch tracks demonstrate that the scope domains of these examples correspond to the domains of FIP, that is, the raised focus F₀ peaks on the wh-arguments and the post-focus reductions marked by the shading. Notice that the post-focus reduction is canceled after the scope domain ends, as shown in Figure 4.10

4.2. Focus Intonation of -sika

Based on the general observations about focus intonation discussed thus far, we now turn our attention to the negative polarity item (NPI) -sika, which typically attaches to an NP and produces the ‘only NP’ reading in combination with negation (Muraki, 1978; Aoyagi & Ishii, 1994; Tanaka, 1997), as demonstrated in (49):

(49)	Taro-sika	ringo-o	tabe-na-katta
	Taro-sika	apple-acc	eat-neg-pst
	‘Only John ate apples.’

Ishihara (2007) observes that sentences with the NPI -sika display a type of focus intonation comparable to that seen in wh-questions. The main points of his observations are presented below:

(50)	a.	A focal F0-rise of the phrase to which an NPI attaches.
	b.	A post-focal downtrend on all the material following the NPI until the negation that bind the NPI.
	c.	A pitch reset after the negation.	(Ishihara, 2007, p. 69)

According to example (50), an XP with -sika exhibits a pitch prominence that is associated with focus, followed by a lowering of the post-focus pitch that continues until the corresponding negation. The region following the negation marks an upward pitch reset. Let us explore this pattern further using the examples in (51):

(51)	a.	Naoya-wa	[Mari-sika	ramu-o	nomiya-de	noma-nakat-ta	to]	Yumi-ni	itta
		Naoya-top	Mari-sika	rum-acc	bar-in	drink-neg-pst	c	Yumi-dat	said
		‘Naoya said to Yumi that only Mari drank rum at the bar.’
	b.	Naoya-sika	[Mari-ga	ramu-o	nomiya-de	nonda	to]	Yumi-ni	iwa-nakat-ta
		Naoya-sika	Mari-nom	rum-acc	bar-in	drank	c	Yumi-dat	say-neg-pst
		‘Only Naoya said to Yumi that Mari drank rum at the bar.’
		(adapted from Ishihara, 2007, p. 70)

While both the -sika phrase and its licensing negation are located within the embedded clause in (51a), they are located in the matrix domain in (51b). The expected intonation patterns of these examples are provided below, in which the boxes and the shaded areas denote a focal F₀-rise and a post-focus downtrend, respectively:11

(52)	a.	Naoya-top [Mari-sika rum-acc bar-in drink-neg-pst c] Yumi-dat said
	b.	Naoya-sika [Mari-nom rum-acc bar-in drank c] Yumi-dat say-neg-pst

As demonstrated in (53), the NPI -sika may appear as a stranded particle in PSE. I argue that sentences like (53b) exhibit focus intonation, as outlined in (52), even under ellipsis. The relevant test examples are presented in (54) and (55), which are modeled after Ishihara’s (2007) original examples:

(53)	a.	Taro-wa	LGB-sika	yoma-nakat-ta-no?
		Taro-top	LGB-sika	read-neg-pst-q
		‘Did Taro read only LGB?’
	b.	-Sika	pro	yoma-nakat-ta	yo
		-sika		read-neg-pst	prt
		‘pro (=Taro) read only LGB.’

(54)	a.	Mari-sika	[Taro-ga	itumo	mizu-o	nomu	to]	minna-ni		iwa-nakat-ta-no?
		Mari-sika	Taro-nom	always	water-acc	drink	c	everyone-dat		say-neg-pst-q
		‘Did only Mari tell everyone that Taro always drink water?’
	b.	-Sika	[Akemi-ga	itumo	mizu-o	nomu	to]	minna-ni	iwa-nakat-tayo
		-sika	Akemi-nom	always	water-acc	drink	c	everyone-dat	say-neg-pstprt
		‘Only Mary told everyone that Akemi always drinks water.’

(55)	a.	Mari-wa	[Taro-ga	itumo	mizu-sika	noma-nai		to]	minna-ni		itta-no?
		Mari-top	Taro-nom	always	water-sika	drink-neg		c	everyone-dat		said-q
		‘Did Mari tell everyone that Taro always drinks only water?’
	b.	-Sika Mari-wa	[Akemi-ga	itumo	noma-nai	to]	minna-ni			itta	yo
		-sika Mari-top	Akemi-nom	always	drink-neg	c	everyone-dat			said	prt
		‘Mari told everyone that Akemi always drinks only water.’

In (54b), the sentence-initial -sika is associated with the matrix negation, whereas it is associated with the negation within the embedded CP in (55b). The expected intonation patterns are illustrated below:

(56)	a.	-sika [Akemi-nom alwayswater-acc drink c] everyone-dat say-neg-pst prt
	b.	-sika Mari-top [Akemi-nom always drink-neg c] everyone-dat said prt

I argue that the predicted focus intonation patterns in (56) are indeed attested. The pitch tracks obtained for (54b) and (55b) are shown in Figure 5 and Figure 6 below:12

Figure 5, which presents the pitch contour obtained from (54b), demonstrates that -sika begins with a higher pitch, followed by a post-focus downtrend. Figure 6, which corresponds to (55b) and is more relevant to our discussion, reveals a raised pitch at the beginning of the sentence, with the following domain showing a pitch reduction extending toward the negation (see note 11) and a notable pitch reset thereafter.

It is interesting that the intonation patterns of sentences without -sika, such as (57a) and (57b), contrast with their -sika counterparts, as illustrated in Figure 7 and Figure 8, respectively:

(57)	a.	Mari-ga	[Akemi-ga		itumo	mizu-o	nomu	to]	minna-ni
		Mari-nom	Akemi-nom		always	water-acc	drink	c	everyone-dat
		iwa-nakat-ta	yo
		say-neg-pst	prt
		‘Only Mari told everyone that Taro always drinks water.’
	b.	Mizu-o	Mari-wa	[Akemi-ga		itumo	noma-nai	to]	minna-ni	itta	yo
		water-acc	Mari-top	Akemi-nom		always	drink-neg	c	everyone-dat	said	prt
		‘Mary told everyone that Taro always drinks only water.’

Figure 7 and Figure 8, both of which lack the sentence-initial -sika, do not exhibit the post-focus reduction that is observed in Figure 5 and Figure 6; as a result, the significant pitch reset effect triggered by the post-focus reduction is also missing in Figure 8.

The defending PF-deletion analysis captures the observed prosodic facts in a straightforward manner. Let us consider how (54b) is derived under the PF-deletion theory:

(58)	i.	Mari-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prt
		(at syntax)
	ii.	Mari-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prt
		(at the PF component)
	iii.	Mari-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prt
		(at the PF component)

In (58i), the syntax, including the sentence-initial PSE position, contains a fully specified structure that permits the computation of the focus intonation between Mari-sika and its associated negation.13 Once this structure is passed to the PF component, the focus intonation associated with the syntactically established relationship between the two items surfaces in (58ii), prior to the phonological deletion of the sentence-initial PSE element Mari, as shown in (58iii), as desired.

Conversely, the current prosodic observations provide strong evidence against the LF-copying analysis of PSE, as this analysis fails to account for the attested focus intonation patterns. More specifically, a syntactic structure that is recovered covertly cannot trigger the computation of focus intonation, as, by definition, it lacks phonetic content. To explore this more concretely, let us consider how the same example (54b) is derived under the LF-copying theory:

(59)	i.	-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prt
		(at syntax)
	ii.	Mari-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prT
		(at syntax and LF)
	iii.	-sika [Akemi-nom always water-acc drink c] everyone-dat say-neg-pst prt
		(at the PF component)

Starting from the base structure in (59i), Mari is covertly recovered from the preceding discourse via LF-copying, as shown in (59ii). However, upon spell-out to the PF component, the phonological representation, shown in (59iii), no longer contains the copied element. As a result, the observed focus intonation pattern is not predicted to surface under the LF-copying theory.14

In fact, several arguments in support of the PF-deletion theory of PSE have been proposed in the literature. Among them, Shibata’s (2014) phonology-based licensing condition on PSE in (60), which requires the left edge of the utterance (i.e., u) to be aligned with the left edge of an intermediate phrase (i.e., i) and for the stranded particle to have prosodic focus, is particularly worth considering in this context:

(60)	PSE is licensed in: [[ X …]i ]u, where X is a stranded particle and is focused.
	(Sato & Maeda, 2019, p. 363; originally from Shibata, 2014)

Assuming that Shibata’s licensing condition in (60) is correct, Sato and Maeda (2019) formulate a generalized version of string deletion in (61), originally proposed by Mukai (2003), as the theoretical backbone for (60). In (61), U_E stands for an elliptical utterance, whereas U_A refers to an antecedent non-elliptical utterance.

(61)	String Deletion in the Phonological Component
	String deletion may apply to a contiguous phonetic string in UE at PF, regardless of its syntactic constituency, if UA exhibits the identical phonetic string.
	(Sato & Maeda, 2019, p. 367)

As Sato and Maeda (2019) discuss, string deletion applies to a contiguous sequence of phonological strings at the PF component, allowing the deletion process to ignore syntactic constituency. The authors provide several examples, including, but not limited to, (62), in which PSE targets the verbal complex, together with the preceding subject nominal:

(62)	a.	Chomsky-ga	sangatu-ni	rainiti-suru-rasii	yo
		Chomsky-nom	March-in	visit.Japan-do-seem	prt
		‘It seems that Chomsky is visiting Japan in March.’
	b.	-Mitai-desu	ne
		-seem-cop.pol	prt
		‘It seems that he is visiting Japan in March.’
		(Control:	Kare-ga	sangatu-ni	rainiti-suru-mitai-desu	ne)
			he-nom	March-in	visit.Japan-do-seem-cop.pol	prt
			‘It seems that he is visiting Japan in March.’
		(adapted from Sato & Maeda, 2019, p. 368)

Based on the uncontroversial assumption that the verbal complex rainiti-suru-mitai-desu, ‘visit.Japan-do-seem-cop.pol’, is formed via morphological merger (Halle & Marantz, 1993; Shibata, 2015), Sato and Maeda argue that the sentence-initial PSE position in (62b) is derived via string deletion since the set of items that undergoes PSE clearly does not form a syntactic constituent.

I argue that the empirical observations of focus intonation presented above are neatly accounted for by the interaction between (60) and (61), thereby reinforcing Sato and Maeda’s (2019) proposal that Shibata’s licensing condition on PSE is grounded in the mechanism of string deletion. More precisely, sentences containing a stranded NPI -sika satisfy Shibata’s licensing condition due to the particle’s focus-marking function and subsequently undergo phonological deletion via string deletion. Thus, the current prosodic evidence finds a natural theoretical grounding within the theory of PF-deletion.

In summary, I have first presented empirical observations that PSE can exhibit focus intonation when accompanied by the NPI -sika (Ishihara, 2007). This prosody-based fact constitutes significant evidence that the mechanism underlying PSE involves PF-deletion rather than LF-copying, as the latter cannot account for the observed focus intonation patterns. I have further argued that this prosodic observation aligns well with the PF-deletion-based approach to PSE originally proposed by Shibata (2014) and developed further by Sato and Maeda (2019), thereby fitting naturally within the theory of PF-deletion.

5. Conclusions

In this paper, I have conducted a theoretical investigation that aims to deepen our understanding of the nature of PSE. Drawing on empirical data from overt extraction from a PSE site, covert ATB movement, and the focus intonation of the NPI -sika, I have demonstrated that the LF-copying analysis of PSE, first proposed by Sakamoto and Saito (2018), suffers from empirical limitations. As an alternative, I have argued for a PF-deletion account, showing that Shibata’s (2014) phonology-based licensing condition, in conjunction with Sato and Maeda’s (2019) string deletion, accounts for the precise mechanism underlying PF-deletion. Unless future empirical evidence suggests otherwise, the present analysis not only advances our understanding of PSE, but also provides valuable insights into ellipsis–prosody interactions, a topic that remains underexplored in the current literature.