The Role of Prosody and Information Structure in Licensing Ellipsis: Particle Stranding Ellipsis in Japanese

Abstract

Japanese noun phrases typically consist of nouns and particles, but there is an ellipsis phenomenon called Particle Stranding Ellipsis (PSE) where nouns are elided with a particle left. A PF-based deletion analysis of PSE has been proposed, but there are several criticisms against it. Thus, it remains elusive what condition is imposed on the licensing of PSE. In this paper, we will formulate a finer-grained phonological theorization of PSE. Our analysis employs a phonological constraint, StrongStart, and information structural factors like givenness and foci, and characterizes PSE as an edge deletion applying to pragmatically given materials at the left edge of intonation phrases. Under this analysis, information structure plays an important role in ellipsis licensing: givenness feeds and foci bleed PSE. We demonstrate that this analysis can handle data that is problematic for the previous string deletion approach.

1. Introduction

In Japanese, nominal arguments usually consist of an NP and a particle. Particles are too weak to be independent phonologically; therefore, they need a preceding NP as a phonological host. However, Japanese exhibits an ellipsis phenomenon called Particle Stranding Ellipsis (henceforth, PSE), where particles are stranded with their host being phonologically empty. In (1B), the NP Tanaka-kun is elided with the topic particle =wa (accompanied by the interjection particle =ne) stranded.

(1)	A:	Tanaka-kun=wa?
		Tanaka-TITLE=TOP
		“What happened to Tanaka-kun?”
	B:	[Δ]=Wa=ne		kaisya=o	yameta=yo.
			TOP=PRT	company=ACC	quit=PRT
		“He quit the job.”		(Hattori, 1960, p. 452, with a slight modification.)

This construction was first noticed in the study of Japanese linguistics (Hattori, 1960), and recently, generative linguistics has come to pay attention to this phenomenon (Sato, 2012; Goto, 2012; Sato & Maeda, 2019; Sakamoto & Saito, 2018, 2020; Yamashita, 2019; Ono, 2025; among others).

Analyzing the operational mechanisms underlying ellipsis phenomena is an important task in generative linguistics because in ellipsis, there is no surface form pronounced that corresponds to the interpreted meaning. As for PSE, Sato and Maeda (2019) proposed that PF-based string deletion is responsible for this construction. They argue that this deletion mechanism applies under the surface string identity between the antecedent and the elided string. However, there is criticism against their assumption that string identity is sufficient for PSE (Sakamoto & Saito, 2018, 2020). We have no single agreed analysis as to what (identity) conditions are relevant for PSE.

In this paper, we will focus on the issue of what licensing conditions are imposed on PSE, and develop a novel phonology-based mechanism to derive PSE. Building on Weir’s (2022) analysis of left-edge ellipsis in English, we analyze PSE as a left-edge deletion applied to the phonologically weak materials at the left edge of intonation phrases. Our analysis will integrate prosodic and discourse factors into consideration, which will enable us to accommodate the data that is problematic for Sato and Maeda’s analysis.

This paper is organized as follows. In Section 2, we will review the current situation of the study of PSE, with special attention given to the identity conditions imposed on PSE. Section 3 will propose a novel phonology-oriented theory of PSE, under which PSE is formulated as an edge deletion applied to the left edge of intonation phrases. In Section 4, we will demonstrate that our analysis can handle the set of observations that imposes a challenge to string deletion. Section 5 covers the conclusion.

2. Existing Controversies and Licensing Conditions on PSE

In this section, we will examine whether string identity is sufficient for the application of PSE. We will show, based on data from the “dare…mo” construction and COMP stranding, that string identity alone does not correctly deal with these examples. Instead, the applicability of PSE relies not only on string identity but also on prosodic and information structural factors.

The previous analyses of PSE diverge in what environment and conditions are involved in the PSE licensing. Sato and Maeda’s string deletion in (2) seems to presuppose that string identity is a prerequisite for PSE.

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

According to (2), the application of string deletion depends on the identity of the phonetic string. In other words, one may take (2) as saying that string identity is a sufficient condition for PSE. However, Sakamoto and Saito (2018, 2020) demonstrate that string identity is not enough to license PSE; there are cases where strings satisfying string identity cannot be deleted.1 Let us review Sakamoto and Saito’s observations, along with our additional scrutiny.

2.1. “Dare … mo” Construction

The first counterexample against string identity in (2) comes from the “dare … mo (wh … mo)” construction. Japanese indeterminate pronouns like dare (who) are interpreted differently depending on the co-occurring particles. When =mo (also) occurs with dare, it will be interpreted as a universal quantifier, as illustrated in (3).

(3)	[Dare=ga	kaita	hon]	=mo	omoshiroi.
	who=NOM	wrote	book	=MO	interesting
	“The book that everybody wrote is interesting.”
	(Sakamoto & Saito, 2018, p. 354, with a slight modification.)

Sakamoto and Saito point out that the “dare … mo” construction resists the application of PSE, despite the string identity between the antecedent and ellipsis site.

(4)	A:	[Dare=ga	kaita	hon]=mo	omosiroi=no?
		who=NOM	wrote	book=MO	interesting=Q
		“Is the book that everybody wrote interesting?”
	B:	[Dare=ga	kaita	hon]=mo	omosiroi=yo.
		who=NOM	wrote	book=MO	interesting=PRT
		“The book that everybody wrote is interesting.”
	B’: *	[∆]=mo	omosiroi=yo.
		[∆]=MO	interesting=PRT
		Lit. ”[∆]=MO is interesting.”
		(Sakamoto & Saito, 2018, p. 354, with a slight modification.)

If string identity is sufficient for the application of PSE as argued in (2), (4B’) should be grammatical. Sakamoto and Saito argue that the unavailability of (4B’) poses a challenge against the string deletion approach.

2.2. COMP Stranding

Other data that challenges the string identity in (2) comes from COMP stranding. Sakamoto and Saito (2020) cite from Fujii (2016) an observation that COMP =no and =koto cannot be stranded in PSE (though we will slightly modify this empirical generalization with additional observation).

(5)	Speaker A:	[DeNA=ga	katta]=no=wa	mimashita=ka?
		DeNA=NOM	won=COMP=TOP	saw=Q
		“Did you see that DeNA won?”
	Speaker B:	[DeNA=ga	katta]=no=wa	mimasendeshita.
		DeNA=NOM	won=COMP=TOP	saw.not
		“I didn’t see that DeNA won.”
	Speaker B’:	*[∆]=no=wa	mimasendeshita.
		(Fujii, 2016, p. 16, with a slight modification.)

(6)	Speaker A:	[DeNA=ga	katta]=koto=o	shitteimasu=ka?
		DeNA=NOM	won=COMP=ACC	know=Q
		“Do you know that DeNA won?”
	Speaker B:	[DeNA=ga	katta]=koto=wa	shirimasen.
		DeNA=NOM	won=COMP=TOP	know.not
		“I didn’t see that DeNA won.”
	Speaker B’:	*[∆]=koto=wa	shirimasen.
		(Fujii, 2016, p. 16, with a slight modification.)

In (5) and (6), the phonological string DeNA=ga katta (DeNA won) is shared between the antecedent and elided utterances, and the string identity condition is satisfied. Despite this identity, the deletion of the complement clause is not allowed. This observation seems to go against the string deletion approach that adopts string identity. From this data, Sakamoto and Saito (2020) argue against string identity.

Though Sakamoto and Saito’s discussion is important, there is more to be considered. We would like to add a new observation that these COMPs are, in fact, able to be stranded under appropriate prosodic and contextual conditions. The relevant data is given in (7) and (8).

(7)	Speaker A:	[DeNA=ga		katta]=no=wa	mitenai=kedo,
		DeNA=NOM		won=COMP=TOP	not.saw=but
		[Hawks=ga		katta]=no=wa	mitanda=yo=ne?
		Hawks=NO		won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=no=wa		mimashita.
			COMP=TOP	saw
		“Intended: I saw that Hawks won.”

(8)	Speaker A:	[Ano sensei=to		hanashita]=koto=wa	nai=kedo,
		that teacher=with		talked=COMP=TOP	NEG=but
		[(ano sensei=to)		atta]=koto=wa	arunda=yo=ne?
		that teacher=with		met=COMP=TOP	be=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=koto=wa		aru=ne.
			COMP=TOP	be=PRT
		“Intended: I have met that teacher.”

Our observation is that when complement clauses are contrastively focused with a contrastive marker =wa attached, COMP stranding becomes acceptable. This suggests that COMP stranding is controlled by focus-related factors. Given this observation, the correct descriptive generalization should be (9).

(9)	COMPs can be stranded in PSE only when the elided clause satisfies the string identity condition and is contrastively focused.

We added a novel observation to Sakamoto and Saito’s discussion, but the conclusion is the same; string identity alone is not sufficient for the application of PSE. We need to take prosody and information structure into consideration.

It should be noted that contrastive =wa can bear a focal stress on its own instead of its immediately preceding materials (Kawamura, 2010), and that COMP stranding is only observed when the contrastive stress is placed on the COMP in the antecedent clauses, as in (11). If =wa is stressed, particle stranding will be obtained, as shown in (12).

(10)	a:	YAMADA=wa	kita
		Yamada=TOP	came
		“Yamada came (but I don’t know about other people).”
	b:	Yamada=WA	kita
		Yamada=TOP	came
		“Only Yamada came.”

(11)	Speaker A:	[DeNA=ga		katta]=NO=wa	mitenai=kedo,
		DeNA=NOM		won=COMP=TOP	not.saw=but
		[Hawks=ga		katta]=NO=wa	mitanda=yo=ne?
		Hawks=NO		won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=NO=wa		mimashita.
			COMP=TOP	saw
		“Intended: I saw that Hawks won.”

(12)	Speaker A:	[DeNA=ga		katta]=no=WA	mitenai=kedo,
		DeNA=NOM		won=COMP=TOP	not.saw=but
		[Hawks=ga		katta]=no=WA	mitanda=yo=ne?
		Hawks=NO		won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=WA		mimashita.
			=TOP	saw
		“Intended: I only saw that Hawks won.”

Interestingly, (11B) and (12B) have slightly different interpretations, corresponding to the difference between (10a) and (10b). As is observed by Kawamura (2010), the prosody like (10a) has an implicature that the speaker does not have knowledge regarding other people. That is, the prosody in (10a) expresses a kind of incompleteness. On the other hand, (10b) has an exclusive interpretation. Likewise, (11B) expresses an incompleteness reading. Thus, Speaker A may or may not have seen other baseball games. On the other hand, (12B) exhibits an exclusive reading: Speaker A only saw the game where Hawks won. In short, COMP stranding and particle stranding exhibit complementary distribution depending on the different prosody of the antecedent clause.

The other possible answer to (11A) suggested by an anonymous reviewer is to delete the whole of the argument. Again, we find an intonational characteristic in the antecedent clause. Unlike (11B) and (12B), there is no intonational break after the COMP =kedo, and no emphatic stress is placed within the embedded clauses.

(13)	Speaker A:	[DeNA=ga	katta]=no=wa	mitenai=kedo
		DeNA=NOM	won=COMP=TOP	not.saw=but
		[Hawks=ga	katta]=no=wa	mitanda=yo=ne?
		Hawks=NO	won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	mimashita
		saw
		“Intended: I saw that Hawks won.”

These intonational characteristics suggest that in (13), the contrastive interpretation of =wa is suppressed. The observations above suggest that the availability of PSE, including COMP stranding, is controlled by the intonational factors. String identity is not enough to capture this; therefore, an alternative approach that incorporates prosody (and prosody-related semantic/pragmatic factors) should be investigated. To establish such an alternative is the central aim of the rest of this paper.

3. Formulating PSE as Edge Ellipsis

We have seen that string identity is not enough to capture the exact environment where PSE occurs. We need to take prosodic and information-structural factors into consideration. In this section, we will try to construct a novel phonology-oriented analysis of PSE. We base our analysis on Weir’s (2022) phonological analysis of left-edge ellipsis in English, and extend it to an analysis of PSE.2

3.1. Weir (2022) on Left-Edge Ellipsis

Weir (2022) analyzes left-edge ellipsis (LEE) found in colloquial English. LEE is an ellipsis phenomenon where utterance-initial function words are omitted. Examples are given in (14).

(14)	a.	Going to the pub tonight. (= I’m going …)
	b.	Seen the new Star Wars Film yet? (= Have you seen …)
	c.	Postman been yet? (= Has the postman …)	(Weir, 2022, p. 253)

Weir proposes that LEE is triggered by a phonological constraint called StrongStart, which requires that intonation phrases begin with a strong element in a sense to be defined below. Utterance-initial function words are phonologically weak in nature and, thus, inevitably violate StrongStart. To avoid this violation, initial weak materials are elided. This is how Weir analyzes LEE. Below, let us see Weir’s formal implementation of this intuition in detail.

Weir adopts a notion of prosodic hierarchy (see Nespor and Vogel (1986), among others). According to prosodic hierarchy, the phonological component has its own hierarchically structured units; each unit of that hierarchy defines a domain to which different phonological rules, like rhythmic rules, apply. Weir assumes three levels of the phonological units above words: prosodic words (ω), phonological phrases (φ), and intonation phrases (ι).

(15)	Prosodic Hierarchy
	(						)											Intonation Phrase (ι)
	(				)		(	)		(			)					Phonological Phrase (φ)
	(	)		(	)		(	)		(	)		(	)	(	)		Prosodic Words (ω)

Following Selkirk’s (2011) Match Theory, Weir assumes that different syntactic units correspond to different phonological constituents.

(16)	a.	Match clause
		A clause in syntactic constituent structure must be matched by a corresponding prosodic constituent, call it ι, in phonological representation.
	b.	Match phrase
		A phrase in syntactic constituent structure must be matched by a corresponding prosodic constituent, call it φ, in phonological representation.
	c.	Match word
		A word in syntactic constituent structure must be matched by a corresponding prosodic constituent, call it ω, in phonological representation.
		(Selkirk, 2011, p. 439)

These match constraints map the syntactic structure in (17b) onto the phonological representation in (17c). Note that only lexical categories are relevant for match constraints. Function words like determiners are not mapped onto prosodic words, but merely correspond to syllables (σ).

(17)	a.	Has the postman delivered the mail?
	b.	[CP has [TP [DP the postman] [T’ [VP delivered [DP the mail]]]]]
	c.	(ι has (φ(φ the (ω postman)) (φ(ω delivered) (φ the(ω mail)))))

Given the prosodic structure in (17c), Weir further adopts a phonological principle, StrongStart, originally proposed in Selkirk (2011). Selkirk’s original definition is given in (18), which Weir decomposes into sub-constraints in (19) and (20).

(18)	StrongStart
	A prosodic constituent optimally begins with a leftmost daughter constituent which is not lower on the prosodic hierarchy than the constituent that immediately follows:
	* (πn πn+1 …	(Selkirk, 2011, p. 470)

(19)	StrongStart-ι
	An Intonational Phrase should not start with a constituent lower on the Prosodic Hierarchy than a Phonological Phrase.
	(Weir, 2022, p. 261)

(20)	StrongStart-φ
	A Phonological Phrase should not start with a prosodic constituent lower on the Prosodic Hierarchy than a Prosodic Word.
	(Weir, 2022, p. 261)

According to Weir’s (re)formulation of StrongStart-ι, intonation phrases should begin with phonological phrases or intonation phrases. This means that function words at the left edge of intonation phrases will cause a violation of StrongStart-ι. To avoid this violation, the utterance-initial function words are eliminated.

(21)	(ι has (φ(φ the (ω postman)) (φ(ω delivered) (φ the (ω mail)))))

Since LEE applies optionally, Weir supposes another constraint named Max, according to which no ellipsis should occur.

(22)	Max
	Every terminal with a specified contentful realization in the underlying structure (input) must have a realization in the output.
	(Weir, 2022, p. 261)

Max is assumed to float among other constraints; that is, Max competes with StrongStart-ι and either of them wins over the other.3 If StrongStart-ι wins, LEE will be obtained; if Max wins, the non-elliptical form will result. Weir theorizes this competition in optimality theoretic terms (Prince & Smolensky, 1993/2004). The OT-tableau of this analysis is given in (23) and (24). (“*” stands for a violation of that constraint, and “!” indicates that violation is fatal. “**” means that there are two violation of that constraint).

(23)
[Has [[the postman] [delivered [the mail]]]]		StrSt-ι	StrSt-φ	Max
	a. (ι has (φ (φ the postman) (φ delivered (φ the mail))))	* !	**
	b. (ι (φ (φ the postman) (φ delivered (φ the mail))))		** !	*
☞	c. (ι (φ (φ postman) (φ delivered (φ the mail))))		*	**
(Weir, 2022, p. 263, with a slight modification.)

(24)
[Has [[the postman] [delivered [the mail]]]]		Max	StrSt-ι	StrSt-φ
☞	a. (ι has (φ (φ the postman) (φ delivered (φ the mail))))		**	*
	b. (ι (φ (φ the postman) (φ delivered (φ the mail))))	*!	**
	c. (ι (φ (φ postman) (φ delivered (φ the mail))))	**!	*
(Weir, 2022, p. 262, with a slight modification.)

In (23), StrongStart-ι is ranked higher than Max and, therefore, the LEE form is chosen; on the other hand, (24) prioritizes Max, and the non-elliptical form is evaluated as being the most optimal candidate. This is how Weir derives LEE.

LEE and PSE are similar in that both delete utterance-initial strings, ignoring constituency; however, there is also a crucial difference between them. LEE elides inherently weak materials (i.e., function words), while PSE can target lexical categories like NPs. To extend Weir’s analysis to PSE, we need an additional mechanism to turn the elided string of PSE into prosodically weak materials (i.e., a phonological constituent lower than a phonological phrase). To achieve this, we have to resort to information structural factors, which will be introduced in the next subsection.

3.2. Information Structural Factors

As is formulated in Match Theory, prosodic structures are organized with reference to syntactic structures. Syntax, however, is not the sole decisive factor to construct prosodic structures. Information structure also influences prosody. Let us see how information structural factors modify prosodic constituency. The relevant notions are givenness and focus.

First, discourse givenness removes phrasal pitch accents. The lack of pitch accents signals the absence of phonological phrases because phonological phrases are the domain where pitch accents are assigned. This effect is implemented in the following constraint (see also Féry & Samek-Lodovíc, 2006).

(25)	[G]=No-φ (DephraseGiven)
	A [G]-marked constituent in MSO [=morpho-syntactic output] corresponds to prosodic constituent in PI [=phonological input] which is not a φ and contains no φ.
	(Kratzer & Selkirk, 2020, p. 29)

Turning to foci, they have opposite effects. Focused phrases should bear the highest prominence. To contain the prominent accent, focused XP is necessarily a phonological phrase. We adopt (26) to formalize this intuition.

(26)	Stress-Focus (SF):
	A focused phrase has the highest prosodic prominence in its focus domain.
	(Féry & Samek-Lodovíc, 2006, p. 135)

A given XP can be focused at the same time. In such cases, the given focus will also bear prominence. In the answer (27A), the given pronoun functions as an answer focus and receives a phrasal accent.

(27)	Q:	Who did Johni’s mother praise?
	A:	She praised [himi]F.	(Féry & Samek-Lodovíc, 2006, p. 132)

This means that SF overrides DephraseGiven. In addition, an anonymous reviewer points out that in Japanese, givenness does not trigger dephrasing. This means that the Match Constraint is ranked higher than DephraseGiven. Therefore, we postulate the following ranking of these constraints.

(28)	SF >> Match Phrase >> DephraseGiven

3.3. Formulating Predictions: PSE as Edge Ellipsis

Now, we are ready to formulate a new phonology-based analysis of PSE. Let us consider the schematic structure in (29).

(29)	a.	[CP [DP [NP X Y Z][G] Part] ……]
	b.	(ι X Y Z Part …)

In (29a), NP is marked as given.4 Then, NP is not mapped onto a φ. Function words, including particles, are integrated into adjacent prosodic domains (Ito & Mester, 2009). Thus, in the resulting prosodic structure (29b), X Y Z is a prosodic constituent lower than a phonological phrase (i.e., prosodic words or syllables). The string X Y Z violates StrongStart-ι. This violation triggers the deletion of the string X Y Z. At the same time, the Match Phrase, which is a kind of syntax–phonology faithfulness condition, requires that the NP has a corresponding phonological constituent in the resulting phonological representation. To satisfy this requirement, the particle has to host a phonological phrase, which leads to the promotion of particles to a prosodic word, which can be phonologically independent (see Selkirk, 1996, for function words in a strong form when they occur in isolation). Then, the particle is stranded with no phonological host. This is how we derive PSE. [G]-marking facilitates the application of PSE.

On the other hand, if the utterance-initial material is focused, that material has to be mapped onto a phonological phrase to receive phrasal pitch accent. The particle is, again, integrated into the adjacent phonological phrase.

(30)	a.	[CP [DP [NP X Y Z][F] Part] ……]
	b.	(ι (φ X Y Z Part) …)

Since the string X Y Z constitutes a phonological phrase, it satisfies StrongStart-ι. Under this phrasing, there is no violation of StrongStart; therefore, nothing has to be deleted. Rather, to satisfy Max, no ellipsis can occur. In short, [F]-marking has a blocking effect for the application of PSE.

To see how this mechanism works, let us examine the analysis of the baseline example in (1) repeated in (31).

(31)	A.	Tanaka-kun=wa?
		Tanaka-TITLE=TOP
		“What happened to Tanaka-kun?”
	B.	[Δ]=Wa=ne		kaisya=o	yameta=yo.
			TOP=PART	company=ACC	quit.PAST=PART
		“He quit the job.”			(Hattori, 1960, p. 452, with a slight modification.)

When StrongStart-ι is ranked higher than Max, the OT-tableau will be like (32). In this case, a PSE pattern is obtained.5 (Bold face indicates the phrasal stress.)

(32)
[[[Tanaka-kun][G]=wa=ne] [[kaisya=o] yameta=yo]]			Match	[G]= No-φ	StrSt-ι	Max
	a.	[(Tanaka-kun=wa=ne:)φ ((kaisya=o) yameta=yo)φ]ι		* !
	b.	[{Tanaka-kun=wa=ne:}ω ((kaisya=o) yameta=yo)φ]ι	* !		*
	c.	[{Δ=wa=ne:}ω ((kaisya=o) yameta=yo)φ]ι	* !		*	*
☞	d.	[({Δ=wa=ne:}ω)φ ((kaisya=o) yameta=yo)φ]ι				*

On the other hand, in (33), Max is ranked over StrongStart-ι. Under this ranking, a non-elliptical form is chosen.

(33)
[[[Tanaka-kun][G]=wa=ne] [[kaisya=o] yameta=yo]]			Match	Max	[G]= No-φ	StrSt-ι
☞	a.	[(Tanaka-kun=wa=ne:)φ ((kaisya=o) yameta=yo)φ]			*
	b.	[{Tanaka-kun=wa=ne:}ω ((kaisya=o) yameta=yo)φ]ι	* !			*
	c.	[{Δ=wa=ne:}ω ((kaisya=o) yameta=yo)φ]ι	* !	*		*
	d.	[({Δ=wa=ne:}ω)φ ((kaisya=o) yameta=yo)φ]ι		* !

In this way, the floating nature of Max can derive the optionality of PSE.

Importantly, our analysis incorporates prosodic and discourse factors. Thanks to this, we have one important ingredient that Sato and Maeda’s string deletion does not include: givenness and foci have feeding/bleeding effects for PSE. The next section will demonstrate that this component of our analysis plays an essential role in explaining the data from Sakamoto and Saito (2018, 2020) that challenges string identity.

4. Solving Problems

Sato and Maeda’s string identity faces several empirical challenges, which is why Sakamoto and Saito (2018, 2020) argue against the string deletion approach. Contrary to this, our phonology-oriented analysis assumes that givenness and foci play a vital role in licensing PSE. This section aims to demonstrate that our analysis can solve the problems that the string deletion approach faces.

4.1. “Dare … mo” Construction

The first data that counters the string deletion approach comes from the “dare … mo” construction. The relevant observation is repeated below.

(34)	A:	[Dare=ga	kaita	hon]=mo	omosiroi=no?
		who=NOM	wrote	book=MO	interesting=Q
		“Is the book that everybody wrote interesting?”
	B:	[Dare=ga	kaita	hon]=mo	omosiroi=yo.
		who=NOM	wrote	book=MO	interesting=PRT
		“The book that everybody wrote is interesting.”
	B’:	*[∆]=mo omosiroi=yo.
		[∆]=MO interesting=PRT
		Lit. “[∆]=MO is interesting.”
		(Sakamoto & Saito, 2018, p. 354, with a slight modification.)

Even if the string identity is satisfied, the “dare …mo” construction resists the application of PSE. We suggest that what blocks PSE is a focus on wh-phrases; the foci bleed PSE.

Indeterminate pronouns associated with an interrogative COMP are known to be interpreted as wh-phrases (see Kuroda, 1965; and Saito, 2017, among others), and to bear a focus intonation, where wh-phrases are accented and materials that intervene between the wh-phrases and their associated particles are phonologically compressed (see Ishihara, 2007; and Kitagawa, 2017, etc.). For instance, in wh-questions, the wh-phrase bears a focal accent, and “post-focal reduction” follows it.

(35)	Na’oya	ga	NA’ni	o	nomi’ya	de	no’nda	no?
	Naoya	NOM	what	ACC	bar	at	drank	COMPWh
	“What did Naoya drink at the bar?”				(Kitagawa, 2017, p. 499)

We would like to point out that a focus intonation also appears in other usages of indeterminate pronouns. When an indeterminate pronoun appears with =mo and functions as a universal quantifier, it also bears a focus intonation, as follows:

(36)	Dare=ga	kaita	hon=mo	omoshiroi
	anybody=NOM	wrote	book=also	interesting
	“The book that everybody wrote is interesting.”

Given these observations, we assume that indeterminate pronouns are inherently [+F]-marked. If indeterminate pronouns are obligatorily focused, our proposed analysis predicts that the “dare … mo” construction never permits PSE because the focus on the indeterminate pronoun dare bleeds PSE. This is how we account for the observation in (34).

Let us formalize this analysis in optimality theoretical terms. The OT-tableau is given in (37), where the relative clause is tentatively assumed to be phrased into a φ.

(37)
[[[dare=ga[F] kaita hon][G] =mo] omoshiroi]			SF	Match	[G]= No-φ	StrSt-ι	Max
☞	a.	[(((dare=ga)φ (kaita)φ)φ (hon=mo)φ)φ (omoshiroi)φ]ι			*
	b.	[dare=ga kaita hon=mo (omoshiroi)φ]ι	* !	*		*
	c.	[(Δ=mo)φ (omoshiroi)φ]	* !				*
	d.	[Δ=mo (omoshiroi)φ]ι	* !	*		*	*

Assuming that SF overrides other constraints like DephraseGiven and StrongStart, the indeterminate pronoun dare must receive a focal prominence. To realize this prominence, it must be pronounced overtly, and thus ellipsis is blocked. Foci have a bleeding effect in licensing PSE.

4.2. COMP Stranding

Let us move on to the analysis of COMP stranding, which poses another challenge to the string identity approach. The empirical generalization is that COMP stranding is impossible in PSE unless the elided complement clause it takes is focused. We repeat the relevant data: (38) and (39) show cases where embedded clauses are not focused. In these cases, PSE cannot strand the COMPs.

(38)	Speaker A:	[DeNA=ga	katta]=no=wa	mimashita=ka?
		DeNA=NOM	won=COMP=TOP	saw=Q
		“Did you see that DeNA won?”
	Speaker B:	[DeNA=ga	katta]=no=wa	mimasendeshita.
		DeNA=NOM	won=COMP=TOP	saw.not
		“I didn’t see that DeNA won.”
	Speaker B’: *	[∆]=no=wa	mimasendeshita.
		(Fujii, 2016, p. 16, with a slight modification.)

(39)	Speaker A:	[DeNA=ga	katta]=koto=o	shitteimasu=ka?
		DeNA=NOM	won=COMP=ACC	know=Q
		“Do you know that DeNA won?”
	Speaker B:	[DeNA=ga	katta]=koto=wa	shirimasen.
		DeNA=NOM	won=COMP=TOP	know.not
		“I didn’t see that DeNA won.”
	Speaker B’: *	[∆]=koto=wa	shirimasen.
		(Fujii, 2016, p. 16, with a slight modification.)

On the other hand, when embedded clauses are contrastively focused, COMP stranding is permitted, as illustrated in (40) and (41).

(40)	Speaker A:	[DeNA=ga	katta]=NO=wa	mitenai=kedo,
		DeNA=NOM	won=COMP=TOP	not.saw=but
		[Hawks=ga	katta]=NO=wa	mitanda=yo=ne?
		Hawks=NO	won=COMP=TOP	saw.COP=PRT=PRT
		"You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=NO=wa	mimashita.
		COMP=TOP	saw
		“Intended: I saw that Hawks won.”

(41)	Speaker A:	[Ano sensei=to	hanashita]=KOTO=wa	nai=kedo,
		that teacher=with	talked=COMP=TOP	NEG=but
		[(ano sensei=to)	atta]=KOTO=wa	arunda=yo=ne?
		that teacher=with	met=COMP=TOP	be=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=KOTO=wa	aru=ne.
		COMP=TOP	be=PRT
		“Intended: I have met that teacher.”

Notice that in (38) and (39), the polarity is different between the antecedent and the elided clause. This means that (38) and (39) are instances of predicate focus. On the other hand, in (40) and (41), predicates are repeated and thus given. Instead, the complement clauses are contrastively focused. The position of the focus seems to play a decisive role in licensing PSE. Then, the task to be addressed is to clarify the relationship between foci and COMP stranding.

To build a concrete analysis, let us first consider [F]- and [G]-marking of the embedded clauses in (38) and (39). As for [F]-marking, we assume that the whole of the embedded clause (that is, the CP) is [F]-marked. This [F]-marking domain is overtly marked with a contrastive marker =wa: it is attached to the CP.6

We assume that [G]-marking targets TPs, not CPs. This is supported by the fact that TPs can be replaced with an anaphoric expression with the COMP intact.

(42)	Yamada=wa			[[Tanaka=ga		sigoto=o		yameru]	koto]=o	siranakatta.
	Yamada=TOP			Tanaka=NOM		job=ACC		quit	COMP=ACC	not.knew
	[[	sono]	koto]=wa		kare=o		odorokaseta
		that	COMP=TOP		he=ACC		surprised
	“Yamada didn’t know that Tanaka would quit his job. That surprised him.”

The anaphoric expression sono replaces the antecedent TP, which suggests that it is a TP that is [G]-marked. Given this information structure, let us consider the availability of COMP stranding.

In the predicate-focus case, as illustrated in (38) and (39), for example, the embedded TP is [G]-marked, as schematically illustrated in (43).

(43)	[CP TP[G] COMP] Pred[F]

Due to [G]-marking, the embedded TP will not be parsed into a phonological phrase. Moreover, since it is a function word, the COMP is also a phonologically weak material. Thus, the string TP-COMP will cause a violation of StrongStart-ι. To avoid this violation, the deletion must apply to the string TP-COMP. Importantly, COMP stranding is not permitted because the stranded COMP still violates StrongStart-ι. This analysis is implemented in the following OT-tableaux. First, if Max is ranked lower than StrongStart-ι or DephraseGiven, it is the most optimal choice to elide the whole of the CP. No particle is stranded.

(44)
[CP [TP DeNA=ga katta][G]=no]=wa mimasendeshita[F].			SF	Match	[G]= No-φ	StrSt-ι	Max
	a.	[(DeNA=ga)φ (katta=no=wa)φ (mimasendeshita)φ]ι			* !
	b.	[DeNA=ga katta=no=wa (mimasendeshita)φ]ι		* !		*
	c.	[Δ=no=wa (mimasendeshita)φ]ι		* !		*	*
	d.	[Δ=wa (mimasendeshita)φ]ι				* !	*
☞	e.	[ Δ (mimasendeshita)φ]ι					*

(45)
[CP [TP DeNA=ga katta][G]=no]=wa mimasendeshita[F].			SF	Match	[G]= No-φ	Max	StrSt-ι
	a.	[(DeNA=ga)φ (katta=no=wa)φ (mimasendeshita)φ]ι			* !
	b.	[DeNA=ga katta=no=wa (mimasendeshita)φ]ι		* !			*
	c.	[Δ=no=wa (mimasendeshita)φ]ι		* !		*	*
	d.	[Δ=wa (mimasendeshita)φ]ι				*	* !
☞	e.	[Δ (mimasendeshita)φ]ι				*

If Max wins over both StrongStart-ι and DephraseGiven, the non-elliptical form will be chosen.

(46)
[CP [TP DeNA=ga katta][G]=no]=wa mimasendeshita[F].			SF	Match	Max	[G]= No-φ	StrSt-ι
☞	a.	[(DeNA=ga)φ (katta=no=wa)φ (mimasendeshita)φ]ι				*
	b.	[DeNA=ga katta=no=wa (mimasendeshita)φ]ι		* !			*
	c.	[Δ=no=wa (mimasendeshita)φ]ι		* !	*		*
	d.	[Δ=wa (mimasendeshita)φ]ι			* !		*
	e.	[Δ (mimasendeshita)φ]ι			* !

Whichever constraint ranking is assumed, COMP stranding does not occur. Thus, in predicate-focus cases, the COMP is not able to be stranded.

Let us move on to the cases where embedded clauses are foci, as schematically illustrated in (47).

(47)	[[CP TP[G] COMP][F]=wa] Pred

[F]-marking of the embedded CP enforces it to form a phonological phrase to retain its focal prominence. At the same time, [G]-marking of the TP prevents the TP from being mapped to a phonological phrase, which leads to the violation of StrongStart-ι. Thus, the TP has to be elided. These requirements will be satisfied if the COMP is stranded and constitutes a phonological phrase with the TP elided.

OT implementation of the analysis of (47) is given in (48), (49), and (50).

(48)
[CP [TP Hawks=ga katta][G]=no][F]=wa mimashita[F].			SF	Match	[G]= No-φ	StrSt-ι	Max
	a.	[(Hawks=ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !
	b.	[Hawks =ga katta=no=wa (mimashita)φ]ι	* !	*		*
☞	c.	[({Δ=no=wa}ω)φ (mimashita)φ]ι				*	*
	d.	[({Δ=wa}ω)φ (mimashita)φ]ι	* !			*	*
	e.	[Δ (mimashita)φ]ι	* !				*

Even if Max is ranked between DephraseGiven and StrongStart-ι, the same result will be obtained.

(49)
[CP [TP Hawks=ga katta][G]=no][F]=wa mimashita[F].			SF	Match	[G]= No-φ	Max	StrSt-ι
	a.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !
	b.	[Hawks =ga katta=no=wa (mimashita)φ]ι	* !	*			*
☞	c.	[({Δ=no=wa}ω)φ (mimashita)φ]ι				*	*
	d.	[({Δ=wa}ω)φ (mimashita)φ]ι	* !			*	*
	e.	[Δ (mimashita)φ]ι	* !			*

The non-elliptical form can occur when Max is ranked above DephraseGiven.

(50)
[CP [TP Hawks=ga katta][G]=no][F]=wa mimashita[F].			SF	Match	Max	[G]= No-φ	StrSt-ι
☞	a.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι				*
	b.	[Hawks =ga katta=no=wa (mimashita)φ]ι	* !	*			*
	c.	[({Δ=no=wa}ω)φ (mimashita)φ]ι			* !		*
	d.	[({Δ=wa}ω)φ (mimashita)φ]ι	* !		*		*
	e.	[Δ (mimashita)φ]ι	* !		*

Recall that the choice of the answer form depends on the intonation in the antecedent clauses. If =wa, instead of the COMP, is stressed, COMP stranding will not occur. Rather, particle stranding will be preferred.

(51)	Speaker A:	[DeNA=ga		katta]=no=WA	mitenai=kedo,
		DeNA=NOM		won=COMP=TOP	not.saw=but
		[Hawks=ga		katta]=no=WA	mitanda=yo=ne?
		Hawks=NO		won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	[∆]=WA		mimashita.
			=TOP	saw
		“Intended: I only saw that Hawks won.”

Following Kawamura (2010), we assume that the stressed =wa is [F]-marked, with its alternative being =mo (also). Then, we obtain the following representation.

(52)	[[CP TP[G] COMP]=wa[F]] Pred

Given this representation, the OT-analysis of (52) can be given as follows. When Max is ranked lower, particle ellipsis is derivable as in (53) and (54). If Max is ranked higher, as in (55), no ellipsis will occur. Thus, the [F]-marking in (52), armed with other constraints, can account for the choice of particle stranding, rather than COMP stranding, in (51).

(53)
[CP [TP Hawks=ga katta][G]=no]=wa[F] mimashita.			SF	Match	[G]= No-φ	StrSt-ι	Max
	a.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !
	b.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι		* !		*
	c.	[Δ=no ({=wa}ω)φ(mimashita)φ]ι				* !	*
☞	d.	[Δ=no ({=wa}ω)φ(mimashita)φ]ι					*
	e.	[Δ (mimashita)φ]ι	* !				*

(54)
[CP [TP Hawks=ga katta][G]=no]=wa[F] mimashita.			SF	Match	[G]= No-φ	Max	StrSt-ι
	a.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !
	b.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι		* !			*
	c.	[Δ=no ({=wa}ω)φ(mimashita)φ]ι				*	* !
☞	d.	[({Δ=wa}ω)φ (mimashita)φ]ι				*
	e.	[Δ (mimashita)φ]ι	* !			*

(55)
[CP [TP Hawks=ga katta][G]=no]=wa[F] mimashita.			SF	Match	Max	[G]= No-φ	StrSt-ι
☞	a.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι				*
	b.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι		* !			*
	c.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !		*
	d.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι			* !
	e.	[(Hawks =ga)φ (katta=no=wa)φ (mimashita)φ]ι	* !		*

We have also observed that with no contrastive intonation and intonational break after the embedded clause in the antecedent clause, full deletion of the complement clause will be preferred.

(56)	Speaker A:	[DeNA=ga	katta]=no=wa	mitenai=kedo
		DeNA=NOM	won=COMP=TOP	not.saw=but
		[Hawks=ga	katta]=no=wa	mitanda=yo=ne?
		Hawks=NO	won=COMP=TOP	saw.COP=PRT=PRT
		“You didn’t see that DeNA won, but you saw that Hawks won?”
	Speaker B:	mimashita.
		saw
		“Intended: I saw that Hawks won.”

The intonation of the antecedent clause suggests that no focus is put within the complement clauses of mita (saw). Rather, what is relevant is the polarity/predicate focus. Due to this, in the underlying representation of (56B), the complement clause will also lack any contrastive focus. Instead, it should be given, for the complement clause is repeated. At the same time, the predicate of (56B) should be focused because it is the informational center that answers the questions in (56A). Therefore, the [F]- and [G]-marking of (the underlying representation of) (56B) should be (57), which is the same as in (38) and (39).

(57)	[CP TP[G] COMP] Pred[F]

As has been shown in (44) and (45), this focus/givenness pattern will result in no ellipsis (when Max is ranked higher) or full deletion of the complement clause (when Max is ranked lower). Thus, the current analysis can also derive the correlation between the elliptical pattern and the intonation in the antecedent clauses.

This is how our analysis derives the pattern of COMP stranding. [G]- and [F]-marking also play a decisive role in licensing PSE.

5. Conclusions

There have been several approaches to PSE. This paper has examined the string identity condition proposed by Sato and Maeda (2019) and Sakamoto and Saito’s (2018, 2020) criticism against it. We have further provided additional observation as to the availability of COMP stranding, where intonational and information-structural factors play a decisive role in ellipsis licensing.

Then, we have formulated a finer-grained theory of PSE that incorporates phonological and discourse factors. Under our analysis, PSE is a deletion that applies to the given materials at the left edge of intonation phrases. This analysis predicts that givenness and foci have a feeding and bleeding effect for licensing PSE. Especially, [F]-marking patterns explain the data that imposes a challenge against Sato and Maeda’s string deletion approach: the “dare … mo” construction and COMP stranding. As far as the current discussion is correct, it can be said that our proposal is more comprehensive than other previous analysis, for it can predict the non-constituent deletion on the one hand, and capture the fact that PSE is licensed based not only on string identity but also on other phonological and discourse factors. The current proposal tells us the importance of considering phonology and information structure in the study of ellipsis, and of formulating so-called “PF-phenomena” in the way that phonologists would do. A fine-grained phonological theorization will provide us with a clearer picture of the phenomena at issue.