The right node raising analysis of coordinated wh-questions in Japanese

Abstract

In this paper, I argue for the right node raising (RNR) analysis of coordinated wh-questions in Japanese, according to which verbs or their larger projections are moved rightward across-the-board in the coordinated structures, with the conjoined wh-phrases staying in their original VP domains. I demonstrate that this analysis can properly capture the following properties of this construction: (i) the conjoined wh-phrases retain the in-situ property of wh-phrases in this language; (ii) they behave as if they make a constituent; (iii) they are sensitive to the clause-mate condition. The most crucial theoretical implication of my arguments for the RNR analysis is that the backward ellipsis analysis is inaccessible to coordination in this language. This is further confirmed by the behaviors of what I call backward gapping, which is also amenable to the RNR analysis. I also examine whether this implication holds cross-linguistically, and reach only the tentative conclusion that it might not accord with what has been found out by the bi-clausal analysis of coordinated wh-questions in other languages.

1 Introduction

In this paper, I argue that Japanese coordinated wh-questions are most properly analyzed in terms of right node raising (RNR), that is, rightward across-the-board (ATB) movement of verbs or their larger projections a la Johnson’s (2009) analysis of English gapping. It is shown that this new analysis is made possible due to the well-known properties of this language: head-finality and in-situ wh-phrases. As far as I know, it has been standard to analyze Japanese coordinated wh-questions as an instance of backward ellipsis (cf. Ishii 2014; Kasai 2016). Thus, according to this analysis, an instance of this construction such as (1a) is analyzed roughly as in (1b):¹

In (1b), the material except for who-NOM in the first conjunct gets deleted since it is recoverable from the corresponding material in the second conjunct. I argue against this analysis, pointing out some serious empirical problems. I propose an alternative analysis of this construction that involves rightward ATB movement of verbs or their larger projections. According to this proposal, (1a) will be analyzed roughly as follows:

I argue that this analysis solves the problems pointed out for the ellipsis analysis in a natural way.

The paper is organized as follows: In Sect. 2, I argue against the backward ellipsis analysis of Japanese coordinated wh-questions, pointing out several empirical problems with it. In Sect. 3, I argue for the RNR analysis of this construction, demonstrating that these problems are accommodated in natural ways and providing further consequences of this analysis. In Sect. 4, I examine what implications my RNR analysis of Japanese coordinated wh-questions has for analyzing coordinated wh-questions in other languages. In Sect. 5, I consider another Japanese construction that has given rise to the debate between the backward ellipsis analysis and the RNR analysis and argue for the latter.

2 Arguments against the backward ellipsis analysis

As Ishii (2014) observes, Japanese coordinated wh-questions do not show any restriction with respect to the possible combination of the coordinated wh-phrases, as shown below:

(3)	a.	(Ittai) dare-ga	sosite nani-o	Mary-ni	ageta no? (two argument wh’s)
		(the hell) who-NOM	and what-ACC	Mary-DAT	gave Q
		‘Lit. Who1 and what2 (the hell) t1 gave Mary t2?’
	b.	(Ittai) nani-o	sosite doko-de	kimi-wa	tabeta no? (argument-adjunct wh’s)
		(the hell) what-ACC	and where	you-TOP	ate Q
		‘Lit. What1 and where (the hell) did you eat t1?’
	c.	(Ittai) itu	sosite doko-de	kimi-wa John-ni	atta no? (adjunct-adjunct wh’s)
		(the hell) when	and where	you-TOP John-DAT	saw Q
		‘When and where (the hell) did you see John?’ (Ishii 2014: 89)

This pattern of facts is similar to that found in coordinated wh-questions in multiple wh-fronting languages such as Bulgarian, Polish and Russian, and is dissimilar to that found in non-multiple wh-fronting languages such as English and German, which do not allow two wh-arguments to be coordinated, as shown below with English examples:

(4)	a.	*Who and what bought?
	b.	*Who and what gave to Mary?
(5)	a.	What and why did you eat?
	b.	When and where did you see John? (Ishii 2014: 90)

Ishii (2014) follows Citko and Gračanin-Yuksek (2013) in assuming the mono-clausal analysis for coordinated wh-questions in multiple wh-fronting languages, according to which the multiple wh-phrases are coordinated during the derivation and the resulting amalgam is moved to the Spec of an interrogative clause. Given that Japanese is not a multiple wh-fronting language, Ishii assumes the alternative bi-clausal analysis for Japanese coordinated wh-questions, according to which the multiple wh-phrases of this construction move to the Spec’s of separate interrogative clauses. Although Ishii mentions two types of bi-clausal analysis for the construction in question, one in terms of multi-dominant structure, proposed by Gračanin-Yuksek (2007) and Citko and Gračanin-Yuksek (2013), and the other in terms of backward ellipsis, proposed by Kazenin (2002) and Lipták (2003, 2011), he simply adopts the latter analysis without any argument and tries to account for the difference between Japanese and English coordinated wh-questions by the availability of scrambling.²

According to Ishii’s (2014) analysis, (1a), for instance, is analyzed roughly as follows:

(6a) shows the underlying structures of the two conjuncts that are built up till TP. Ishii claims that from these underlying structures, who-NOM in the second conjunct undergoes sideward movement to Spec-TP in the first conjunct, and then undergoes further movement to Spec-CP. Assuming the mechanism of clausal typing proposed by Cheng (1991), according to which interrogative clauses need to be marked as such either by overt interrogative Cs or overt wh-movement into their Spec’s, Ishii claims that the overt movement of who-NOM is necessary since the interrogative C of the first conjunct is null. The second wh-phrase what-ACC in the second conjunct, on the other hand, does not have to undergo wh-movement to Spec-CP since the second conjunct is overtly marked by the Q-marker no, so Ishii assumes that it instead undergoes scrambling, adjoining to TP in the second conjunct. These series of operations give rise to the representation given in (6b), and the output form of (1a) is derived by deleting the TP of the first conjunct under identity with the corresponding part of the second conjunct.

One may raise a couple of questions about the technical implementation of this analysis. It seems odd to assume that the first conjunct lacks an underlying subject, unlike the second conjunct, as shown in (6a), and obtains one by way of sideward movement from the latter. Why not assume instead that who-NOM in (6a) is simply base-generated in Spec-TP of the first conjunct and pro is posited for the subject of the second conjunct?³ Further, there seems to be no reason to assume that what-ACC in the second conjunct undergoes scrambling, other than deriving the correct word order, given that Japanese is a wh-in-situ language. Note that in this case, the scrambling in question is obligatory, given the unacceptability of the following sentence:

(7)	*Dare-ga	sosite	Mary-ni	nani-o	katte-ageta	no?
	who-NOM	and	Mary-DAT	what-ACC	buy-gave	Q
	‘Lit. Who1 and for Mary what2 t1 bought t2?’

I will not dwell on these technical questions any further here but rather point out more general and serious problems with the ellipsis analysis. First of all, there is a factual question about the possible positions the conjoined wh-phrases may occupy in coordinated wh-questions in Japanese. As expected from his ellipsis analysis, Ishii (2014) holds that the conjoined wh-phrases cannot appear in situ. This does not sound right, however, as witnessed by the fact that (3b, c) have their variants in which the conjoined wh-phrases appear in situ:⁴

(8)	a.	Kimi-wa (ittai)	nani-o	sosite	doko-de	tabeta	no?
		you-TOP (the hell)	what-ACC	and	where	ate	Q
		‘Lit. You ate what and where (the hell)?’
	b.	Kimi-wa (ittai)	itu	sosite doko-de	John-ni	atta	no?
		you-TOP (the hell)	when	and where	John-DAT	saw	Q
		‘Lit. You saw John when and where (the hell)?’

One might object that in these cases, the topic phrase kimi-wa ‘you-TOP’ occupies a position higher than the interrogative CPs, so that they may be compatible with Ishii’s analysis in which the first conjoined wh-phrase moves to Spec-CP. However, (8a, b) can be embedded without changing their grammaticality:

(9)	a.	Watasi-wa	[kimi-ga	(ittai)	nani-o	sosite	doko-de tabeta ka] wakara-nai.
		I-TOP	you-NOM	(the hell	what-ACC	and	where     ate   Q   know-not
		‘Lit. I don’t know [Q you ate what and where (the hell)].’
	b.	Watasi-wa	[kimi-ga	(ittai)	itu	sosite	doko-de John-ni atta ka]
		I-TOP	you-NOM	(the hell)	when	and	where John-DAT saw Q
		wakara-nai.
		know-not
		‘Lit. I don’t know [Q you saw John when and where (the hell)].’

Ishii (2014) provides the following example to make his point (the judgment of this sentence is his):

(10)	??John-wa	[Bill-ga	(ittai)	dare-ni	sosite	nani-o      ageta koto]-o
	John-TOP	Bill-NOM	(the hell)	who-DAT	and	what-ACC gave   fact-ACC
	sitteiru	no?
	know	Q
	‘Lit. John knows the fact that Bill gave who and what (the hell)?’ (Ishii 2014: 92)

Personally, I find no degradation with this sentence. It is predicted under Ishii’s analysis that if the conjoined wh-phrases are preposed sentence-initially, the resulting sentence should be acceptable. (10) is not a good sentence to test this prediction since its embedded clause constitutes a complex NP island. Let us instead compare the following examples:

(11)	a.	John-wa	[Bill-ga (ittai)	dare-ni	sosite	nani-o ageta to]
		John-TOP	Bill-NOM (the hell)	who-DAT	and	what-ACC gave COMP
		omotteiru	no?
		think	Q
		‘Lit. John thinks that Bill gave who and what (the hell)?’
	b.	(Ittai) dare-ni	sosite nani-o	John-wa	[Bill-ga	ageta to]
		(the hell) who-DAT	and what-ACC	John-TOP	Bill-NOM	gave COMP
		omotteiru	no?
		think	Q
		‘Lit. Who and what (the hell) does John think that Bill gave?’

There is no significant difference in acceptability between these sentences (to me, (11a) is more natural than (11b)). Furthermore, it is possible to put conjoined wh-phrases even within islands:⁵

(12)	a.	John-wa [(ittai)	dare-ni	sosite	nani-o	ageta] hito-o
		John-TOP (the hell)	who-DAT	and	what-ACC	gave person-ACC
		sagasiteiru	no?
		is.looking.for	Q
		‘Lit. John is looking for the person who gave who and what (the hell)?’
	b.	John-wa [Mary-ga	(ittai) dare-ni	sosite	nani-o	ageta node]
		John-TOP Mary-NOM	(the hell) who-DAT	and	what-ACC	gave because
		okotteiru	no?
		is.angry	Q
		‘Lit. John is angry because Mary gave who and what (the hell)?’

Thus it is reasonable to conclude, contrary to what Ishii (2014) claims, that the in-situ property of wh-phrases in Japanese carries over to the conjoined wh-phrases in coordinated wh-questions. It is not at all clear how such sentences as (11a) and (12a, b) are derived under the backward ellipsis analysis proposed by Ishii (2014).

It is now interesting to compare the above examples of coordinated wh-questions with those that Ishii (2014) calls “forward sluicing”. The latter counterpart of (1a) is something like the following:

(13)	Dare-ga	Mary-ni	katte-ageta	no,	sosite	nani-o?
	who-NOM	Mary-DAT	buy-gave	Q	and	what-ACC
	‘Lit. Who bought e for Mary and what?’

Ishii claims that this sentence is derived under his ellipsis analysis in such a way that the first and second conjuncts are swapped: while dare-ga ‘who-NOM’ in the first conjunct undergoes scrambling since this clause is headed by the Q-marker no, nani-o ‘what-ACC’ in the second conjunct undergoes wh-movement to Spec-CP since it lacks an overt Q-marker, and deletion applies to the material in the second conjunct except for the fronted wh-phrase. Aside from the technical details of this analysis, it is reasonable to claim that the second conjunct of this construction involves ellipsis with a remnant wh-phrase. As far as I can determine, there is no reason to assume that the wh-phrase in the first conjunct undergoes any movement, as witnessed by the fact that the first conjunct can be embedded within another clause without changing grammaticality:

(14)	Anata-wa	[dare-ga	Mary-ni	katte-ageta to]	omotteiru no,	sosite nani-o?
	you-TOP	who-NOM	Mary-DAT	buy-gave COMP	think       Q	and what-ACC
	‘Lit. You think that who bought e for Mary and what?’

Suppose, on the other hand, that the remnant wh-phrase in the second conjunct undergoes overt wh-movement in accord with Ishii’s analysis. Then the second conjunct of (14) will have the following structure:

(15)

[CP what1-ACC

[TP you-TOP

[prowho Mary-DAT

t1 bought

COMP] think] Q]

In this structure, the embedded subject pro covaries with whatever value who-NOM in the first conjunct takes (cf. (i) in fn. 3). TP then gets deleted under an appropriate identity condition that applies to the ellipsis site against the antecedent clause, i.e., the first conjunct. In this case, the correlate of what-ACC in the antecedent clause is an implicit argument, so it is reasonable to regard this case of ellipsis as an instance of the spouting type of sluicing in the sense of Chung et al. (1995) (CLM), whose typical instance is given below:

(16)	a.	She’s reading. I can’t imagine what.
	b.	This opera was written in the 19th century, but we’re not sure by whom.
		(CLM: 241-242)

In these cases as well, the correlates of the remnant wh-phrases what and by whom are implicit.

It is predicted under this ellipsis analysis that the type of coordinated wh-questions under consideration is island sensitive. This is in fact borne out:

(17)	a.	?*John-wa	[nani-o     ageta] hito-o	sagasiteiru   no,	sosite dare-ni?
		John-TOP	what-ACC gave   person-ACC	is.looking.for Q	and   who-DAT
		‘Lit. John is looking for the person who gave e what, and who?’
	b.	?*John-wa	[Mary-ga    nani-o       ageta node]	okotteiru no,	sosite dare-ni?
		John-TOP	Mary-NOM what-ACC gave  because	is.angry  Q	and    who-DAT
		‘Lit. John is angry because Mary gave e what, and who?’

To the extent that such an ellipsis analysis as presented above is tenable, it undermines Ishii’s (2014) claim that much the same analysis applies to the original type of coordinated wh-questions, since the conjoined wh-phrases in the latter type behave as if they appear in situ, exhibiting island insensitivity, as shown in (12).

Secondly, the original type of coordinated wh-questions shows clause-mate effects that occur between the conjoined wh-phrases. Consider the following examples:⁶

(18)	a.	?*Dare-ga	sosite nani-o	John-ga	Mary-ni	katte-ageta to	Itta no?
		who-NOM	and what-ACC	John-NOM	Mary-DAT	buy-gave COMP	said Q
		‘Lit. Who1 and what2 t1 said that John had bought Mary t2?’
	b.	?*Dare-ga	sosite itu	John-ga	kubi-ni natta	to	omotteiru no?
		who-NOM	and when	John-NOM	was.fired	COMP	think        Q
		‘Who1 and when2 t1 thinks that John was fired t2?’

It is not clear at all how an ellipsis analysis such as Ishii’s (2014) can capture these clause-mate effects. Note that the second wh-phrases of these examples can undergo long-distance scrambling, as shown below:

(19)	a.	Nani-o1	Bill-wa	[John-ga	Mary-ni	t1katte-ageta to]	itta no?
		what-ACC	Bill-TOP	John-NOM	Mary-DAT	buy-gave COMP	said Q
		‘What did Bill say [that John had bought Mary t]?’
	b.	Itu1	Bill-wa	[John-ga	t1	kubi-ni natta to]	omotteiru no?
		when	Bill-TOP	John-NOM		was.fired COMP	think        Q
		‘When does Bill think [that John was fired t]?’

Thus nothing will go wrong if we assume that (18a), for instance, has the following structure, in which the TP of the first conjunct gets deleted under identity with that of the second conjunct:

To make this point stronger, let us compare the examples in (18) with the corresponding ones of the forward sluicing type:

(21)	a.	Dare-ga	[John-ga	Mary-ni	katte-ageta to]	itta no,	sosite nani-o?
		who-NOM	John-NOM	Mary-DAT	buy-gave COMP	said Q	and what-ACC
		‘Lit. Who said that John had bought Mary e, and what?’
	b.	Dare-ga	[John-ga	kubi-ni	natta to]	omotteiru no,	sosite itu?
		who-NOM	John-NOM	was.fired	COMP	think Q	and when
		‘Lit. Who said [that John was fired e], and when?’

The acceptability of these examples clearly shows that the forward sluicing type of coordinated wh-questions exhibits no clause-mate effects between the conjoined wh-phrases. This again undermines a parallel treatment of “forward and backward sluicing” in terms of ellipsis, suggesting that an alternative analysis should be sought for the “backward sluicing” type.

Finally, the conjoined wh-phrases of the “backward sluicing type” behave as if they make constituents. Thus, in (1a), for instance, the conjoined wh-phrases can be clefted, as shown below:⁷

(22)	[Mary-ni	katte-ageta	no]-wa	dare-ga	sosite	nani-o	na	no?
	Mary-DAT	buy-gave	NL-TOP	who-NOM	and	what-ACC	COP	Q
	‘Lit. Who1 and what2 was it that t1 bought t2 for Mary?’

This is totally unexpected under the backward ellipsis analysis, according to which the conjoined two wh-phrases belong to separate clauses.⁸ We have seen above that the original type of coordinated wh-questions shows clause-mate effects that occur between the conjoined wh-phrases, as exemplified in (18). In fact, these examples are acceptable with different readings, namely ones in which the conjoined wh-phrases are interpreted as clause-mates and John-ga ‘John-NOM’ is interpreted as the matrix subject. On these readings, the conjoined wh-phrases can be clefted, as shown below:

(23)	a.	[John-ga	Mary-ni	katte-ageta to itta no]-wa	dare-ga sosite
		John-NOM	Mary-DAT	buy-gave COMP said NL-TOP	who-NOM and
		nani-o	na	no?
		what-ACC	COP	Q
		‘Lit. Who1 and what2 was it that John said that t1 had bought Mary t2?’
	b.	[John-ga	kubi-ni natta to	omotteiru no]-wa	dare-ga sosite
		John-NOM	was.fired COMP	think NL-TOP	who-NOM and
		itu     na	no?
		when COP	Q
		‘Who1 and when2 was it that John thinks that t1 was fired t2?’

Here again, there will be no way for the backward ellipsis analysis to account for this fact.

One reviewer suggests an interesting way out of this apparent problem under the ellipsis analysis, according to which the cleft example (22) has the following structure:

According to this structure, (22) has two cleft constructions in it, one having the presuppositional part [Mary-DAT buy-gave NL]-TOP with who-NOM in the pivot position and the other having the same presuppositional part with what-ACC in the pivot position. The surface form of (22) is derived from (24) by deleting the presuppositional part of the second conjunct. I see at least two problems with this analysis. First, it is expected under this analysis that a major intonational break should fall right after who-NOM, as is the case when (24) is uttered in full. This does not seem to be borne out, however; the most natural way to utter (22) is to do it as if who-NOM and what-ACC make an intonational phrase. Secondly, it has often been noted (cf. Takano 2015, among others) that the Japanese cleft construction is not comfortable with a nominative-marked phrase in its pivot position. Thus, the first conjunct of (24) is not fully acceptable:

(25)	??[Mary-ni	katte-ageta	no]-wa	dare-ga	na	no?
	Mary-DAT	buy-gave	NL-TOP	who-NOM	COP	Q
	‘Lit. Who1 was it that t1 bought e for Mary?’

(22) does not have any such degradation effect. This effect manifests itself more strongly when the nominative-marked phrase in the pivot position is connected to a gap in the embedded clause of the presuppositional part. According to the reviewer’s suggested analysis, the first conjunct of (23a), for instance, will correspond to the following cleft sentence:

(26)	?*[John-ga	Mary-ni	katte-ageta	to	itta no]-wa	dare-ga	na	no?
	John-NOM	Mary-DAT	buy-gave	COMP	said NL-TOP	who-NOM	COP	Q
	‘Lit. Who1 was it that John said that t1 had bought Mary e?’

Here again, the degraded status of (26) does not carry over to (23a). This fact undermines the reviewer’s suggested analysis of cleft sentences such as in (22) and (23).

I conclude from the above discussion that the problems with the backward ellipsis analysis of coordinated wh-questions pointed out in this section are fatal enough to seek an alternative.⁹

3 Proposal: right node raising analysis

I argue that Japanese coordinated wh-questions should be analyzed as involving RNR, that is, ATB raising of a verb or its larger projections. Let us first consider a simple case:

(27)	Dare-ga	sosite	nani-o	nusunda	no?
	who-NOM	and	what-ACC	stole	Q
	‘Lit. Who1 and what2 t1 stole t2?’

In this case, I assume that this sentence has the following underlying structure:

(28)	[CP[TP [VP [VP	dare1-ga	nusun]	sosite	[VPpro1	nani-o nusun]] da] no]
		who-NOM	steal	and		what-ACC steal PAST Q

I basically follow Kasai (2016) in assuming that Japanese coordinated wh-questions involve VP coordination with the further assumption that subjects can stay in situ in this language, though I do not assume the so-called split VP hypothesis (i.e., V-v distinction) just for simplicity, so that subject simply occupies the highest position of VP. As for unpronounced arguments, I follow Kasai (2016) in assuming that the unpronounced subject in the second conjunct of (28) is an instance of pro, which covaries with whatever value dare-ga ‘who-NOM’ in the first conjunct takes, as witnessed by the fact that pro may be replaced by an overt anaphoric item such as soitu-ga ‘the guy-NOM’ (cf. fn. 3). On the other hand, I depart from Kasai (2016) in not positing pro in the first conjunct that corresponds to nani-o ‘what-ACC’ in the second conjunct. In that case, the anaphoric relation would be backward, so it would be inappropriate to maintain that pro covaries with whatever value nani-o takes, as witnessed by the fact that pro cannot be replaced by an overt anaphoric item without causing unacceptability, as noted by Kasai (2016):

(29)	*Dare-ga	sore-o	sosite	nani-o	nusunda	no?
	who-NOM	it-ACC	and	what-ACC	stole	Q

Kasai instead assumes that the pro in question is an unpronounced variety of indefinite pronoun, so that the whole sentence is interpreted as ‘who stole something and what did that person steal?’. Though it has been claimed in the literature (cf. Hoji 1998, among others) that indefinite pro is available to Japanese, it is dubious to posit such a pro in the construction under consideration, as it cannot be replaced with an overt form such as nanika ‘something’, as shown below:

(30)	*Dare-ga	nanika-o	(nusumi)	sosite	nani-o	nusunda	no?
	who-NOM	something-ACC	steal	and	what-ACC	stole	Q

For this reason, I do not assume pro in this construction that would correlate with a wh-phrase backwards, but rather assume that in that case, the unpronounced argument is not syntactically realized and is interpreted implicitly.

Under the backward ellipsis analysis, the output form of (27) can be derived from (28) by simply deleting the verb nusun ‘steal’ in the first conjunct, as shown below:

Alternatively, I propose, along the lines of Koizumi’s (2000) idea, that the output form of (27) is derived from (28) by applying ATB head raising to both instances of nusun ‘steal’ and adjoining them to the above T, as shown below (where the items enclosed with angled brackets represent unpronounced copies):

An obvious difference between these two analyses lies in the fact that whereas the conjoined wh-phrases dare-ga sosite nani-o ‘who-NOM and what-ACC’ make a constituent as conjoined VPs under the RNR analysis, they do not under the backward ellipsis analysis. Thus, contrary to the latter, the RNR analysis immediately explains the fact that the conjoined wh-phrases in Japanese coordinated wh-questions can be clefted, as demonstrated in the preceding section:

(33)	[Nusunda	no]-wa	dare-ga	sosite	nani-o	na	no?
	stole	NL-TOP	who-NOM	and	what-ACC	COP	Q
	‘Lit. Who1 and what2 was it that t1 stole t2?’

Further, note that (27) is a question about a single event rather than complex events, so that it has basically the same meaning as who stole what?. It is not obvious under the backward ellipsis analysis how this fact is captured. Under the RNR analysis, on the other hand, it is natural to assume that this is guaranteed by ATB movement of shared verbs or their larger projections, as stated below:¹⁰

(34)	When V1, V2, … or their projections undergo ATB movement, the resulting chain
	denotes one and the same event or state.

It follows then that in (32), the implicit object of the first occurrence of nusun ‘steal’ denotes whatever nani-o ‘what-ACC’ in the second conjunct does and that pro in the second conjunct necessarily denotes whatever dare-ga ‘who-NOM’ in the first conjunct does.

Let us now consider a slightly more complex case, namely (1a), reproduced below:

(35)	Dare-ga	sosite	nani-o	Mary-ni	katte-ageta	no?
	who-NOM	and	what-ACC	Mary-DAT	buy-gave	Q
	‘Lit. Who1 and what2 t1 bought t2 for Mary?’

In this case, applying ATB head raising of the verb katte-age ‘buy-give’ alone to the following underlying structure will not give rise to the appropriate interpretation of (35):

(36)	[CP [TP [VP [VP	dare1-ga	katte-age]	sosite	[VP pro1 nani-o Mary-ni
		who-NOM	buy-give	and	what-ACC Mary-DAT
	katte-age]] ta]	no]
	buy-give PAST	Q

If this underlying structure is uttered in full, it sounds strange in that the first conjunct presupposes that someone bought something for someone and asks who made a purchase, whereas the second conjunct presupposes that that person bought something for Mary and asks what was bought. Rather, the first conjunct should presuppose that someone bought something for Mary, just like the second conjunct. Thus, the underlying structure of (35) should be something like the following:

(37)	[CP [TP [VP [VP	dare1-ga [V’	Mary-ni	katte-age]]	sosite [VP pro1	nani-o
		who-NOM	Mary-DAT	buy-give	and	what-ACC
	[V’ Mary-ni	katte-age]]]	ta]	no]
	Mary-DAT	buy-give	PAST	Q

I propose that the output form of (35) is derived from (37) by applying rightward movement to the V’ [Mary-ni katte-age] ‘Mary-DAT buy-give’ in an ATB fashion and adjoining it to VP, as shown below:¹¹

Given the standard assumption (cf. Hoji 1985) that indirect object is underlyingly higher than direct object, the V’ of the second conjunct must contain the trace of nani-o ‘what-ACC’, which is scrambled to the left of Mary-ni. Hence, there is an issue about exactly what is the relevant identity condition that makes ATB movement possible. Since it is beyond the scope of this paper to formulate such a condition, I simply stipulate that some sort of non-distinctness is relevant for determining the applicability of ATB movement, so that when an unpronounced copy in one conjunct has no correlate in the other, it is ignorable for applying ATB movement as long as the target phrases are located in parallel positions in the relevant coordination, V’ in VP coordination in this case. In the structure (38), the conjoined wh-phrases dare-ga sosite nani-o ‘who-NOM and what-ACC’ make a constituent as conjoined VPs thanks to the rightward ATB movement of the V’ [Mary-ni katte-age]. Thus, as demonstrated in (22) in the preceding section, they can be clefted:

(39)	[Mary-ni	katte-ageta	no]-wa	dare-ga	sosite	nani-o	na	no?
	Mary-DAT	buy-gave	NL-TOP	who-NOM	and	what-ACC	COP	Q
	‘Lit. Was it who1 and what2 that t1 bought t2 for Mary?’

Under this proposal, let us consider why an instance of coordinated wh-questions such as (7), reproduced below, is unacceptable:

(40)	*Dare-ga	sosite	Mary-ni	nani-o	katte-ageta	no?
	who-NOM	and	Mary-DAT	what-ACC	buy-gave	Q
	‘Lit. Who1 and for Mary what2 t1 bought t2?’

First of all, it is not the case that the conjoined wh-phrases in this construction must be adjacent to each other, as witnessed by the acceptability of the following sentence, which differs from (35) only in that the subject pro in the second conjunct is replaced by an overt item (cf. fn. 9):

(41)	Dare-ga	sosite	soitu-ga	nani-o	Mary-ni	katte-ageta	no?
	who-NOM	and	the guy-NOM	what-ACC	Mary-DAT	buy-gave	Q
	‘Lit. Who (bought e) and what did that person buy for Mary?’

The unacceptability of (40) is naturally attributed to the discrepancy of the presuppositions that the two conjuncts have, just like (36): while the first conjunct presupposes that someone bought something for someone, the second conjunct presupposes that that person bought something for Mary. Note that in this case, Mary-ni ‘Mary-DAT’ and the verb katte-age ‘buy-give’ do not make a constituent, so that they cannot undergo rightward ATB movement. It is then predicted that if Mary-ni appears in the first conjunct rather than in the second, so that pro can be posited in the latter to refer to it, the resulting sentence should be acceptable. This is in fact borne out:¹²

(42)	?Dare-ga	Mary-ni	sosite	nani-o	katte-ageta	no?
	who-NOM	Mary-DAT	and	what-ACC	buy-gave	Q
	‘Lit. Who1 for Mary and what2 t1 bought t2?’

This sentence is derived from the following underlying structure by applying ATB head raising to buy-give and adjoining it to the above T:¹³

(43)	[CP [TP [VP [VP who1-NOM	Mary2- DAT buy-give] and [VP pro1 pro2 what-ACC
	buy-give]]     PAST] Q]

Given that the conjoined wh-phrases can make a constituent as conjoined VPs under the present RNR analysis, it is expected that they may undergo scrambling in the same way as Koizumi (2000) argues with the coordinated structure he discusses (cf. fn. 8). Let us consider Ishii’s (2014) example (3b):

(44)	(Ittai)	nani-o	sosite	doko-de	kimi-wa	tabeta	no?
	(the hell)	what-ACC	and	where	you-TOP	ate	Q
	‘Lit. What1 and where (the hell) did you eat t1?’

In this case, the conjoined wh-phrases precede the subject kimi-wa ‘you-TOP’. If it is assumed that this topic marked subject is located in Spec-TP (or probably in a higher specifier position), (44) is derived from the following structure by applying scrambling to the conjoined VPs and adjoining them to TP:

Let us now consider the examples in (11), reproduced below:

(46)	a.	John-wa	[Bill-ga	(ittai)	dare-ni	sosite nani-o ageta to]
		John-TOP	Bill-NOM	(the hell)	who-DAT	and what-ACC gave COMP
		omotteiru	no?
		think	Q
		‘Lit. John thinks that Bill gave who and what (the hell)?’
	b.	(Ittai)	dare-ni	sosite nani-o	John-wa	[Bill-ga ageta to]
		(the hell)	who-DAT	and what-ACC	John-TOP	Bill-NOM gave COMP
		omotteiru	no?
		think	Q
		‘Lit. Who and what (the hell) does John think that Bill gave?’

Recall that I pointed out in the preceding section that cases such as (46a) as well as those given in (12) in which the conjoined wh-phrases appear within islands are problematic to the backward ellipsis analysis proposed by Ishii (2014). Under the present proposal, (46a) is analyzed as simply involving V’ coordination in the embedded clause with the embedded verb age ‘give’ undergoing ATB head raising into the above T, as shown below:

Exactly the same analysis applies to those cases given in (12), which also involve V’ coordination within island clauses with the embedded verb age ‘give’ undergoing ATB head raising into the above T. These cases clearly show that no special syntactic operations apply to the conjoined wh-phrases in Japanese coordinated wh-questions and that they retain their in-situ properties, exactly like normal wh-phrases in this language. Now (46b) is derived from the structure partially shown in (47) by scrambling the whole conjoined V’s to the top of the sentence.

Finally, let us consider how the clause-mate effects of Japanese coordinated wh-questions are derived under the present analysis of this construction. The relevant examples are reproduced below from (18):

(48)	a.	?*Dare-ga	sosite nani-o	John-ga	Mary-ni	katte-ageta to itta no?
		who-NOM	and what-ACC	John-NOM	Mary-DAT	buy-gave COMP said Q
		‘Lit. Who1 and what2 t1 said that John had bought Mary t2?’
	b.	?*Dare-ga	sosite itu	John-ga	kubi-ni	natta to omotteiru no?
		who-NOM	and when	John-NOM	was.fired	COMP think Q
		‘Who1 and when2 t1 thinks that John was fired t2?’

The ungrammaticality of these sentences can be attributed to the low coordination of the conjoined wh-phrases, namely VP coordination. Under the present proposal, (48a), for instance, will have the following structure:

In this structure, what-ACC in the second conjunct has undergone long-distance scrambling from the embedded object position to the matrix VP-adjoined position. Then, the matrix V’ [John-NOM Mary-DAT (<what-ACC>) buy-gave COMP] say has undergone rightward ATB movement and is adjoined to the matrix VP (note that I have assumed above that an unpronounced copy like <what-ACC> is ignorable for applying ATB movement). With this structure, it is most natural to claim that what is blamed for illegitimacy resides in the landing site of the long-distance scrambling of what-ACC. In Abe (2022), I argue that long-distance scrambling in Japanese necessarily gives rise to a focus chain and that this property is captured by assuming that a phrase to be scrambled long-distance must bear a [Focus] feature and the feature is licensed by moving the phrase to the Spec of Focus Phrase (FP), which is located above TP. As a piece of evidence, I refer to Saito’s (1985) observation that long-distance scrambling cannot land in a clause-medial position:

(50)	??John-ga	sono hon-o1	minna-ni	[Mary-ga t1	motteiru to]	itta (koto)
	John-NOM	that book-ACC	all-DAT	Mary-NOM	have      COMP	said fact
	‘Lit. (the fact that) John, that book1, told everyone that Mary had t1.’				(Saito 1985: 267)

Given that the matrix subject John-ga may be located in Spec-TP, it is most natural to assume that the scrambled phrase sono hon-o ‘that book-ACC’ is adjoined to the matrix VP. Under the assumption that this phrase carries a [Focus] feature, the ungrammaticality of (50) is attributed to the fact that this feature remains unlicensed. Likewise, what-ACC in (49) must carry a [Focus] feature and the feature cannot be licensed in the VP adjoined position it has landed on. Note that this wh-phrase cannot be adjoined to a higher position, since in that case, it would induce a violation of Coordinate Structure Constraint (CSC), which prohibits a phrase from being extracted from one of the conjoined phrases.¹⁴

3.1 Superiority?

Ishii (2014) observes that there is a word order restriction between the conjoined wh-phrases in Japanese coordinated wh-questions:

(51)	a.	(Ittai)	dare-ga	sosite	nani-o	Mary-ni	ageta	no?
		the hell	who-NOM	and	what-ACC	Mary-DAT	gave	Q
		‘Lit. Who1 and what2 (the hell) t1 gave t2 to Mary?’
	b.	?*(Ittai)	nani-o	sosite	dare-ga	Mary-ni	ageta	no?
		the hell	what-ACC	and	who-NOM	Mary-DAT	gave	Q
		‘Lit. What1 and who2 (the hell) t2 gave t1 to Mary?’ (Ishii 2014: 99)

(51a, b) differ only in the order of the conjoined wh-phrases dare-ga ‘who-NOM’ and nani-o ‘what-ACC’. Ishii claims that this contrast follows from the superiority effects that arise in multiple wh-questions. Under his backward ellipsis analysis, the first conjunct of (51b) has the following structure:

In this structure, the lower wh-phrase what-ACC has crossed the higher wh-phrase who-NOM, inducing a superiority effect, just like the following English wh-question:

(53)	*What did who give to Mary?

There are several reasons to maintain that we should reject this way of accounting for the unacceptability of (51b) in terms of superiority. First, as we have seen above, there is no good reason to assume that the conjoined wh-phrases in Japanese coordinated wh-questions undergo overt wh-movement to Spec-CP. Rather they behave just like normal wh-phrases in that they can stay in situ. Note that no superiority effect arises in Japanese when a lower wh-phrase crosses a higher wh-phrase by clause-internal scrambling:

(54)	Nani-o1	dare-ga	Mary-ni t1	ageta	no?
	what-ACC	who-NOM	Mary-DAT	gave	Q
	‘Lit. What did who give to Mary?’

Thus, it is crucial for Ishii’s (2014) analysis to assume that the first conjoined wh-phrase what-ACC in (52) undergoes overt wh-movement to Spec-CP, but there is no good evidence for such an assumption. Second, as Kasai (2016) claims, it is more appropriate to assume that a sentence like (51b) is interpreted as ‘what did someone give to Mary and who was it that gave it to Mary?’, so that no multiple question is involved in each conjunct. If so, superiority effects are simply irrelevant for coordinated wh-questions. Third, such a word order restriction between the conjoined wh-phrases as shown in (51) also holds for what Ishii (2014) calls forward sluicing:

(55)	a.	Dare-ga	Mary-ni	ageta	no,	sosite	nani-o?
		who-NOM	Mary-DAT	gave	Q	and	what-ACC
		‘Lit. Who gave e to Mary and what?’
	b.	?*Nani-o	Mary-ni	ageta	no,	sosite	dare-ga?
		what-ACC	Mary-DAT	gave	Q	and	who-NOM
		‘Lit. What did e give to Mary and who?’

Here again, (55a, b) differ only in the order of the conjoined wh-phrases dare-ga ‘who-NOM’ and nani-o ‘what-ACC’. The unacceptability of (55b) could not be accounted for even under Ishii’s (2014) analysis, since it must be the second wh-phrase dare-ga ‘who-NOM’ that undergoes overt wh-movement to Spec-CP and hence no superiority violation should occur.

I propose that the unacceptability of (51b) and (55b) should be attributed to the lack of subject in the first conjuncts. Under my analysis, the first conjuncts of these sentences have the following structure:

(56)	[CP [TP [VP what1-ACC [V’ Mary-DAT t1 give]] and …

Crucially, I have been assuming that the implicit argument in the first conjunct that correlates with the wh-phrase in the second conjunct is not syntactically present. Hence no subject is present in (56). Given that the subject-predicate relation is a core property of a proposition, it is reasonable to attribute the illegitimacy of (56) to the lack of subject.

In Abe (2019), I demonstrate that the same restriction holds for what I call “predicate ellipsis”. A typical example of this construction is given below:

(57)	Bill-ga	Susan-o	sikatta	ra,	John-ga	Mary-o [Pred e].
	Bill-NOM	Susan-ACC	scolded	and	John-NOM	Mary-ACC
	‘Bill scolded Susan and John Mary …’

One might consider (57) as a case of gapping in Japanese where the missing predicate corresponds to that of the first conjunct, hence interpreted as sikatta ‘scolded’. This is not the case, however.¹⁵ Rather, the elided clause is interpreted as an unidentified event or state in which John and Mary both participate. Thus, the null predicate could be taken as hometa ‘praised’, tataita ‘hit’, etc. but since the speaker did not complete the sentence for some reason, the content of the null predicate is unknown to the hearer. Hence, there is no linguistic antecedent for the null predicate in question. One might consider that the grammatical status of these sentences corresponds to that of an English sentence like I saw, I kissed, etc. with missing objects. Nonetheless, it is not so difficult for Japanese native speakers to imagine that these sentences can be uttered out of the blue in a situation where the speaker tried to say, for example, John hit Mary severely, but he/she could not complete the sentence because he/she was so shocked or perplexed. Since these fragments describe events or states in which John and Mary are both involved, a natural response to these fragments is, what happened to John and Mary or what did John do to Mary, etc. In Abe (2019), I then make the observation that the null predicate in question requires a subject for its licenser. Compare (57) with the following examples:

(58)	a.	?*Bill-ni	Susan-o	sono	sensei-ga	syookaisita	ra,
		Bill-DAT	Susan-ACC	that	teacher-NOM	introduced	and
		John-ni	Mary-o	[Pred e].
		John-DAT	Mary-ACC
		‘That teacher introduced Susan to Bill and Mary to John ...’
	b.	?*Bill-ni	hon-o	sono	sensei-ga ageta ra,	John-ni	zassi-o [Pred e].
		Bill-DAT	book-ACC	that	teacher-NOM gave and	John-DAT	magazine-ACC
		‘That teacher gave a book to Bill and a magazine to John ...’

I basically attribute the unacceptability of these sentences to the lack of subject in the second conjuncts.

In Abe (2019), I further argue that the null predicate in question is licensed not only by the subject-predicate relation but also by the topic-comment relation, so that if one of the non-subject remnant arguments of the null predicate is topicalized, then the resulting sentence improves. Compare the examples in (58) with the following, where the dative arguments of the null predicates bear a topic marker:

(59)	a.	Bill-ni	Susan-o	sono	sensei-ga	syookaisita	ra,
		Bill-DAT	Susan-ACC	that	teacher-NOM	introduced	and
		John-ni-wa	Mary-o	[Pred e].
		John-DAT-TOP	Mary-ACC
		‘That teacher introduced Susan to Bill and Mary to John ...’
	b.	Bill-ni	hon-o	sono	sensei-ga ageta ra,	John-ni-wa	zassi-o [Pred e].
		Bill-DAT	book-ACC	that	teacher-NOM gave and	John-DAT-TOP	magazine-ACC
	‘That teacher gave a book to Bill and a magazine to John ...’

The significant improvement of these sentences suggests that the topic-comment relation is another way of legitimizing propositions. If so, it is predicted that unacceptable sentences of coordinated wh-questions that are accounted for due to the lack of subject, such as (51b) and (55b), should improve when topicalized phrases are added to the first conjuncts. This is in fact borne out:

(60)	a.	Mary-ni-wa	(ittai)	nani-o	sosite	dare-ga	ageta no?
		Mary-DAT-TOP	the hell	what-ACC	and	who-NOM	gave Q
		‘Lit. As for Mary, what1 and who2 (the hell) t2 gave t1 to her?’
	b.	Mary-ni-wa	nani-o	ageta	no,	sosite	dare-ga?
		Mary-DAT-TOP	what-ACC	gave	Q	and	who-NOM
		‘Lit. As for Mary, what did e give to her and who?’

Thus it is reasonable to conclude that the unacceptability of such examples of coordinated wh-questions as (51b) and (55b) has nothing to do with superiority but rather has to do with the lack of the subject-predicate relation.

3.2 Kasai’s (2016) case of possessor-possessed relationship

Kasai (2016) provides an interesting case of Japanese coordinated wh-questions to support his backward ellipsis analysis:

(61)	Taroo-wa	dare-no	sosite	nani-o	nusunda	no?
	Taro-TOP	who-GEN	and	what-ACC	stole	Q
	‘Lit. Whose and what did Taro steal?’

A peculiar property of this example is that it presupposes that Taro stole something and that it asks about one and the same thing whose it is and what it is. Thus, Kasai paraphrases this example as follows:

(62)	Taroo-wa	dare-no	sosite	soitu-no	nani-o	nusunda	no?
	Taro-TOP	who-GEN	and	the guy-GEN	what-ACC	stole	Q
	‘Lit. Whose and his/her what did Taro steal?’

Note that unlike English, Japanese allows a wh-word like nani ‘what’ to be modified by a possessor phrase. Note further that dare-no ‘who-GEN’ cannot appear on its own in a normal wh-question:

(63)	*Taroo-wa	dare-no	nusunda	no?
	Taro-TOP	who-GEN	stole	Q
	‘Lit. Whose (one) did Taro steal?’

Thus, dare-no in (61) is permitted only when it modifies nani-o ‘what-ACC’ in the second conjunct. In order to exclude such a case as (63), let us characterize the relevant condition as a filter of the following sort: *[_NP possessor φ].

Keeping this in mind, let us consider what Kasai posits as the underlying structure for (61):

(64)	Taroo-wa [VP [VP [NP	dare1-no-proIndf.-o] nusumi]	sosite [VP [NP pro1 nani-o]
	Taro-TOP	who-GEN-pro-ACC steal	and            what-ACC
	nusun]]-da no
	steal-PAST Q

Pro₁ in the second VP conjunct takes whatever value dare-no ‘who-GEN’ takes, just as soitu ‘the guy’ does in (62). Pro_Indef. in the first VP conjunct, on the other hand, is an indefinite pronoun that correlates with nani-o ‘what-ACC’ in the second conjunct. Kasai claims that from this underlying structure, dare-no ‘who-GEN’ moves out of NP and is adjoined to the first VP conjunct. The output form of (61) is then derived by deleting the lower VP of the first conjunct. In this derivation, the movement of dare-no ‘who-GEN’ out of NP appears to violate the Left Branch Condition (LBC). Kasai maintains that this violation is remedied by deleting the material inducing the violation, as standardly assumed since Merchant (2001), as shown below:

A problem with this analysis arises from whether the underlying structure given in (64) correctly captures the intended meaning of (61), especially the fact that it is a question about one and the same thing even though it involves conjoined wh-phrases. Let us consider what reading obtains when (64) is uttered in full:

(66)	Taroo-wa	dare1-no-o	nusumi,	sosite	nani-o	nusunda	no?
	Taro-TOP	who-GEN-ACC	steal	and	what-ACC	stole	Q
	‘Lit. Whose one did Taro steal and what did he steal?’

The most natural interpretation of this question is that there were two things that Taro stole and whose was one thing and what was the other, which is the interpretation unavailable to (61). Thus it is doubtful that (64) expresses the right structure for (61). Moreover, Kasai’s analysis of (61) will incorrectly predict that something like the following example is grammatical:

(67)	*Taroo-wa	dare-no	sosite	dare-ni	(sore-o)	ageta	no?
	Taro-TOP	who-GEN	and	who-DAT	it-ACC	gave	Q
	‘Lit. Whose and to whom did Taro give?’

In this example, the two conjoined wh-phrases clearly denote different entities, unlike (61); dare-no ‘who-GEN’ denotes something possessed by someone and dare-ni ‘who-DAT’ denotes someone who got it from Taro. According to Kasai’s analysis, this sentence would have the following underlying structure:

(68)	Taroo-wa [VP proIndef. [NP1	dare-no-proIndf.-o]	age] sosite [VP	dare-ni pro/sore1-o
	Taro-TOP	who-GEN-pro-ACC	give and	who-DAT it-ACC
	age]-ta no
	give]-PAST Q

In this structure, pro_Indef. in the first conjunct correlates with dare-ni ‘who-DAT’ in the second conjunct, and pro/sore-o ‘it-ACC’ in the second conjunct takes whatever value [_NP dare-no-pro_Indf.-o] does. After dare-no ‘who-GEN’ moves out of NP₁ and adjoins to the first conjoined VP, just as dare-no does in (64), the output form of (67) could be derived by deleting the lower VP of the first conjunct under identity with the corresponding part in the second conjunct.¹⁶ This overgeneration problem indicates that Kasai’s analysis fails to capture the fact that the possessor form dare-no ‘who-GEN’ can appear in coordinated wh-questions only when it denotes the same entity as the other wh-phrase does.

Under my RNR analysis, there is a rather natural way of capturing the above facts. First, I modify the underlying structure given in (64) in such a way that dare-no-pro_Indf. ‘who-GEN-pro’ in the first conjunct does not have an accusative case marker attached to it, as shown below:

(69)	Taroo-wa [VP [VP [NP	dare1-no-proIndf.]	nusun]	sosite [VP [NP	pro1 nani-o]
	Taro-TOP	who-GEN	steal	and	what-ACC
	nusun]]-da	no
	steal-PAST	Q

This modification is natural since pro does not have any overt case marker in general. Then, I propose that the output form of (61) is derived from (69) by first adjoining dare-no ‘who-GEN’ and [_NP pro₁ nani-o] ‘what-ACC’ to each of the conjoined VPs, as shown in (70a),¹⁷ and then applying the rightward ATB movement to the lower VP in each conjunct that consists of the direct object NP and V, as shown in (70b):

Here again an issue arises about what identity condition guarantees the applicability of the ATB movement in (70b). It must be the case that [_NP <dare₁-no>-pro_Indf.] and [_NP <pro₁ nani-o>] are regarded as identical under the relevant identity condition on ATB movement. I simply stipulate that as far as unpronounced copies and pro are concerned, they are regarded as identical for applying ATB movement if their internal structures are the same; in this case, they both have N projections with possessor NPs. Recall that we have assumed above the following interpretive rule about ATB movement of verbs and their larger projections:

(71)	When V1, V2, … or their projections undergo ATB movement, the resulting chain denotes one and the same event or state.

Given this, once the two NPs [_NP <dare₁-no>-pro_Indf.] and [_NP <pro₁ nani-o>] are raised by ATB movement, they must be regarded as identical in semantic interpretation as well. This will guarantee that in (61) what Taro stole was one (set of) thing and that the apparently two questions in terms of dare-no ‘who-GEN’ and nani-o ‘what-ACC’ are actually asked about the same thing.

Note that I am following Kasai (2016) in assuming that the LBC violation induced by the movement of dare-no ‘who-GEN’ in (70a) is remedied by wiping out its bottom copy, by way of rightward ATB movement in this case.¹⁸ It follows then that in deriving the output form of (61) from (70a), the RNR strategy adopted in (70b) is the only legitimate option; applying ATB movement just to the verb nusun ‘steal’ in (70a) could also derive the output form of (61) but this cannot remedy the LBC violation in question. One might object that in that case, dare-no would not have to move out of the immediately dominating NP, to begin with. Note, however, that in that case, the resulting structure will violate the filter *[_NP possessor φ], which rules out a sentence like (63).

The unacceptability of (67), reproduced below, follows naturally under the present analysis:

(72)	*Taroo-wa	dare-no	sosite	dare-ni	(sore-o)	ageta	no?
	Taro-TOP	who-GEN	and	who-DAT	it-ACC	gave	Q
	‘Lit. Whose and to whom did Taro give (it)?’

This sentence will have the following underlying structure:

(73)	Taroo-wa [VP [VP [NP1	dare-no-proIndf.]	age]	sosite [VP	dare-ni pro/sore1-o
	Taro-TOP	who-GEN-pro	give	and	who-DAT it-ACC
	age]]-ta	no
	give-PAST	Q

From this structure, we cannot derive the output form of (72) in the same way as shown in (70). First, we could adjoin the two wh-phrases dare-no ‘who-GEN’ and dare-ni ‘who-DAT’ to each of the VP conjuncts, but we could not move onto the next step; that is, we could not apply rightward ATB movement to the VPs that dominate the verbs and the objects since the direct object [_NP1 dare-no-pro_Indf.] in the first conjunct and pro/sore₁-o in the second conjunct do not have the same internal structure. Thus, the only option that could derive the output form of (72) is to apply ATB raising of the verb age ‘give’ to the structure (73), but the resulting structure would violate the filter *[_NP possessor φ].

Note further that (61) becomes unacceptable if the conjoined wh-phrases are swapped in their word order:

(74)	*Taroo-wa	nani-o	sosite	dare-no	nusunda	no?
	Taro-TOP	what-ACC	and	who-GEN	stole	Q
	‘Lit. What and whose did Taro steal?’

Under the present assumptions, this sentence will have the following underlying structure:

(75)	Taroo-wa [VP [VP [NP	nani-o]	nusun]	sosite [VP [NP	dare1-no-proIndf.]
	Taro-TOP	what-ACC	steal	and	who-GEN
	nusun]]-da	no
	steal-PAST	Q

In this case, nani-o ‘what-ACC’ in the first conjunct does not have pro in its Spec position, unlike that in (69), and hence this wh-phrase cannot be regarded as identical to the wh-object [_NP dare₁-no-pro_Indf.] in the second conjunct so as to allow the dominating VPs to undergo rightward ATB movement after the two wh-phrases move out of them. Hence the only way to derive the output form of (74) is to apply ATB head raising to the verb nusun ‘steal’ in the structure (75), but the resulting structure would violate the filter *[_NP possessor φ].^19,20

4 Implications for the cross-linguistic perspectives in analyzing coordinated wh-questions

In Sect. 2, I pointed out mainly three problems to the backward ellipsis analysis of Japanese coordinated wh-questions. Two of them, i.e., the fact that the conjoined wh-phrases behave as if they make a constituent and the fact that they show the normal in-situ properties, just like single wh-phrases, so that they appear in embedded contexts, are under-generation problems. Thus, only with these problems, we cannot conclude that the backward ellipsis strategy is not an option for analyzing Japanese coordinated wh-questions, since it might be the case that the effects of this strategy are covered up due to the availability of the RNR strategy. One more problem, however, i.e., the fact that the conjoined wh-phrases show clause-mate effects, is an over-generation problem for the backward ellipsis analysis, so this problem clearly indicates that this analysis is not an option for Japanese coordinated wh-questions.

There is further evidence for this claim, which comes from what Giannakidou and Merchant (1998) call “reverse sluicing”. A typical English example is given below:

(76)	It’s not clear if and when the police arrested the demonstrators.
	(Giannakidou and Merchant 1998: 234)

Giannakidou and Merchant analyze such an example as involving ellipsis of the complement clause of the Q-complementizer if, and its content is recovered from the following TP, i.e., [_TP the police arrested the demonstrators]. Japanese, on the other hand, does not allow such reverse sluicing:

(77)	*John-wa [CP e	ka dooka]	sosite [CP	itu	Mary-ga	tazunete-kita	ka]
	John-TOP	whether	and	when	Mary-NOM	visit-came	Q
	oboetei-nai.
	remember-not
	‘John does not remember if and when Mary came and visited (him).’

If the backward ellipsis strategy were available to this construction, so that the content of the empty TP in the first conjunct could be supplied by the following TP, we could not account for the unacceptability of (77). In Abe (2015), I demonstrate that ka dooka ‘whether’ can support “forward sluicing”:

Thus the contrast between (77) and (78B) strongly indicates that the backward ellipsis analysis is unavailable to Japanese coordinated wh-questions.

It will not be unreasonable to conjecture from the above discussion that the unavailability of the backward ellipsis strategy is a universal property of coordinated wh-questions, given the standard poverty of stimulus argument: how do Japanese children come to know that the backward ellipsis strategy is unavailable to Japanese coordinated wh-questions? There is one thing that should be made clear at this point. We are not maintaining that the backward ellipsis strategy is never allowed in universal grammar, as it has been observed that backward ellipsis is possible in some ellipsis constructions such as VP ellipsis, sluicing and NP ellipsis:

(79)	a.	If he hasn’t yet [VP e], John should try to climb the Eiffel Tower.       (Jackendoff 1971: 27)
	b.	Although I don’t know why [TP e], John takes LSD.    (Wasow 1972: 90)
	c.	?Because Steve’s [NP e] had been stolen, I borrowed Fred’s diagram of a snake’s eye.        (Jackendoff 1971: 31)

Even Japanese allows for backward ellipsis, as shown in the following sluicing example:

(80)	Dare-to	ka	wakara-nai	keredomo,	Mary-wa	kinoo
	who-with	Q	know-not	though	Mary-TOP	yesterday
	yuuenti-ni		itta.
	amusement park-to		went
	‘Although (I) don’t know with whom, Mary went to an amusement park yesterday.’

Given that all of the examples in (79) and (80) involve backward ellipsis in subordination, it is reasonable to conclude that the unavailability of the backward ellipsis strategy to coordinated wh-questions is due to the fact that this construction involves coordination.^21,22

Let us then hypothesize that the backward ellipsis strategy is unavailable in coordination and see what implications this hypothesis has for the analysis of coordinated wh-questions in general. As briefly mentioned in Sect. 2, there are mainly two types of analysis for this construction: the mono-clausal and the bi-clausal analysis. The mono-clausal analysis is proposed for coordinated wh-questions in multiple wh-fronting languages, according to which the multiple wh-phrases are coordinated during the derivation and the resulting amalgam is moved to the Spec of an interrogative clause. Thus, the backward ellipsis strategy is irrelevant for this type of analysis. The bi-clausal analysis, on the other hand, is proposed for coordinated wh-questions in non-multiple wh-fronting languages such as English and German, which typically show some restrictions on the conjoined wh-phrases. As we saw in (4) Sect. 2, reproduced below, these languages do not allow two wh-arguments to be coordinated:

(81)	a.	*Who and what bought?
	b.	*Who and what gave to Mary?

More generally, Lipták (2011) nicely captures the relevant restrictions as follows: “the grammaticality judgments for CMWQs [= coordinated multiple wh-questions] are fully parallel to the judgments on bi-clausal questions”:

(82)		CMWQs in English
	a.	*What and to who did you give?
	b.	*What and where did you fix?
	c.	What and where did you eat?
	d.	When and why did you leave?

(83)		Bi-clausal questions in English
	a.	*[What did you give] and [to who did you give]?
	b.	*[What did you fix] and [where did you fix]?
	c.	[What did you eat] and [where did you eat]?
	d.	[When did you leave] and [why did you leave]?	(Lipták 2011: 156)

Thus it is reasonable to posit bi-clausal questions such as in (83) for the underlying structures of CMWQs such as those in (82). Then, as Lipták (2011) puts it, the question comes down to how the underlying structures of the first conjuncts of CMWQs are reduced to their surface forms. One of the strategies proposed for doing this job is the backward ellipsis strategy, according to which the material in the first conjuncts of CMWQs is deleted except for the fronted wh-phrases under identity with the corresponding part in the second conjuncts. Our hypothesis, posited above, that the backward ellipsis strategy is unavailable in coordination discourages this option, however.

There are two other strategies proposed in the literature: one is the multi-dominance approach, advocated by Gračanin-Yuksek (2007) and Citko and Gračanin-Yuksek (2013), according to which the repeated material in the two conjuncts shares the relevant structure, i.e., the whole structure of C’, and the other is the RNR approach, advocated by Park (2006) and Haida and Repp (2011), according to which the shared material in each conjunct is right node raised and is adjoined to the coordinated CP. These two analyses are compatible with my analysis of Japanese coordinated wh-questions, so the latter analysis could be regarded as giving indirect support to the former. Nonetheless, I am not sure at this point if the backward ellipsis analysis is actually not an option for analyzing coordinated wh-questions in English and German, given the present stage of understanding according to which there is not really a good independent argument for either the multi-dominance or the RNR analysis. As far as I can determine, Lipták (2011) provides the most powerful argument for the backward ellipsis analysis, which is concerned with what Merchant (2001) calls swiping, as illustrated below:

(84)	a.	*Who from did Mary receive a package?
	b.	Mary received a package, but I don’t know who from.	(Lipták 2011: 160)

Swiping is a phenomenon in which the word order of a preposition and the wh-word that appears in its complement is flipped, and as the contrast between (84a) and (84b) shows, it occurs only in the sluicing construction, a hall-marked ellipsis phenomenon. Lipták (2011) reports that the English native speakers she consulted with allow for swiping in coordinated wh-questions:

(85)	a.	Who from and why did Mary receive a package?
	b.	Who to and when did Chomsky lecture about syntax?	(Lipták 2011: 160)

Though not all English native speakers may allow for this phenomenon (cf. Gračanin-Yuksek 2007), the fact that some do indicates that universal grammar does not prohibit the backward ellipsis strategy from being accessed in coordinated wh-questions.

Haida and Repp (2011) present interesting scope-out phenomena to give support to the RNR analysis. They first note, following Sabbagh (2007), that such phenomena are observed with a typical instance of RNR:

(86a) shows that the shared QP every patient who was admitted last night takes scope over the indefinite some nurse, so that a different nurse may be involved in treating each patient. This is not the case with (86b), in which the QP every patient occupies each original position of the right node raised QP in (86a). Sabbagh (2007) argues that these data support the rightward ATB movement approach to RNR, since it correctly predicts that the shared QP in this construction can be scoped out of the coordinated clauses. Now on the model of this scope-out phenomenon of RNR, Haida and Repp (2011) provide the following paradigm:

(87)	a.	Tell me if every guest arrived.	(IF > ∀, *∀ > IF)
	b.	Tell me when every guest arrived.	(WHEN > ∀, ∀ > WHEN)
	c.	Tell me if and when every guest arrived.	(IF&WHEN > ∀, ∀ > IF&WHEN)
			(Haida and Repp 2011: 384)

Even though the QP every guest takes scope under if in (87a), it can take scope out of the latter when if and when are conjoined, as shown in (87c). Haida and Repp argue that these data support the RNR analysis of coordinated wh-questions in the same way as the data in (86) do: (87c) is derived by applying rightward ATB movement to TP in each conjunct, so that every guest can scope out of the interrogative clauses.

It is not clear whether the paradigm given in (87) should be accounted for on a par with that in (86). For one thing, what mechanism of scope interaction makes it possible that movement of the whole TP out of the above CP enables its subject QP to take scope over that CP? Under the standard assumption that the scope order of two scope-bearing elements is determined in terms of c-command, the scope ambiguity observed in (87b) is naturally captured as a result of the reconstruction effects of fronted wh-phrases; when is base-generated lower than the subject QP every guest and movement of this wh-phrase over the subject makes the scope order of these two phrases ambiguous. In (87a), on the other hand, if is base-generated in the interrogative C head, hence always asymmetrically c-commanding the subject QP. This accounts for why the latter cannot take scope over the former. Given this standard account of scope interaction in terms of c-command, it is not immediately clear how the subject QP every guest in (87c) is able to take scope over if as a result of applying rightward ATB movement to TP in each conjunct.

Furthermore, the paradigm given in (87) can be replicated with what Ishii (2014) calls forward sluicing in Japanese:

(88)	a.	Subete-no	gesuto-ga	tootyakusita	ka	osiete.
		every-GEN	guest-NOM	arrived	Q	tell
		‘Tell (me) if every guest arrived.’ (IF > ∀, *∀ > IF)
	b.	Subete-no	gesuto-ga	itu	tootyakusita	ka osiete.
		every-GEN	guest-NOM	when	arrived	Q tell
		‘Tell (me) when every guest arrived.’ (WHEN > ∀, ∀ > WHEN)
	c.	Subete-no	gesuto-ga	tootyakusita ka	sosite	itu     ka osiete.
		every-GEN	guest-NOM	arrived         Q	and	when Q tell
		‘Tell (me) if every guest arrived and when. (IF&WHEN > ∀, ∀ > IF&WHEN)

(88c) is an instance of forward sluicing and can have the reading in which the embedded subject QP subete-no gakusei ‘every guest’ can take scope over the if- and when-clauses, just like (87c). Recall that I have argued in Section 2 that unlike coordinated wh-questions, forward sluicing is best analyzed as an instance of ellipsis in Japanese. Further, note that it is quite unlikely that such forward sluicing as illustrated in (88c) is derived from some syntactic operation similar to RNR that enables the embedded QP to scope out of the immediately dominating TP or the CP above it. Thus, the availability of the relevant reading to such a forward sluicing case as (88c) strongly undermines Haida and Repp’s (2011) argument for the RNR analysis on the basis of the paradigm given in (87). After all, this latter paradigm may be turned into supporting evidence for the ellipsis analysis of coordinated wh-questions, together with that given in (88), though it remains to be seen how this paradigm is accounted for.

Judging from what we have seen above, it seems hasty to conclude that the backward ellipsis strategy is not an option for reducing the bi-clausal structures posited for coordinated wh-questions in languages such as English and German. As for Japanese coordinated wh-questions, on the other hand, I believe convincing arguments have been provided against the backward ellipsis strategy. More work is necessary to fill the gap between these states of affairs.

5 Further consequences

There is another construction in Japanese, as illustrated below, which has given rise to the debate between the backward ellipsis analysis and the RNR analysis:

(89)	John-ga	Mary-o	sosite	Bill-ga	Susan-o	hometa.
	John-NOM	Mary-ACC	and	Bill-NOM	Susan-ACC	praised
	Lit. John [V e] Mary and Bill praised Susan.’

In this example, the verb home ‘praise’ is missing in the first conjunct. I name this construction “backward gapping” as an analysis-neutral term. It is reasonable to posit (90) as the underlying structure of this example:

(90)	[John-NOM Mary-ACC praise] and [Bill-NOM Susan-ACC praise]+PAST

According to the backward ellipsis analysis, advocated by Abe and Hoshi (1997), among others, (89) will be derived from (90) by deleting praise in the first conjunct. According to the RNR analysis, advocated by Kuno (1978) and Saito (1987), among others, on the other hand, this example will be derived from (90) by applying ATB head raising to praise in each conjunct. Given our conclusion reached above that backward ellipsis is impossible in coordination at least in Japanese, we should opt for the RNR analysis of backward gapping. In this section, I argue that this is the right choice.

Keeping the assumptions made for Japanese coordinated wh-questions, I propose that (89) has the following underlying structure, which involves VP coordination with the subject in each conjunct staying in situ:

(91)	[TP [VP [VP John-NOM Mary-ACC praise] and [VP Bill-NOM Susan-ACC praise]] T]

From this underlying structure, the output form of (89) is derived by applying ATB head raising to the V praise in each conjunct and adjoining it to the above T, as shown below:

Let us now consider a little more complicated example of backward gapping:

(93)	John-ga	Mary-o	sosite	Bill-ga	Susan-o	yuuenti-ni	turete-itta.
	John-NOM	Mary-ACC	and	Bill-NOM	Susan-ACC	amusement park-to	took
	‘Lit. John [V’ e] Mary and Bill took Susan to an amusement park.’

This example is interpreted such that John took Mary to an amusement park and Bill did the same thing to Susan. Analyzing such an example on a par with coordinated wh-questions, I propose that it involves rightward ATB movement of the verb phrase yuenti-ni turete-iku ‘take to an amusement park’, as shown below:

On the other hand, there is a crucial difference between coordinated wh-questions and backward gapping. Recall that we noted in Sect. 3 that coordinated wh-questions express single events and that we attributed this property to ATB movement of shared verbs or their larger projections, as stated in (34), reproduced below:

(95)	When V1, V2, … or their projections undergo ATB movement, the resulting chain denotes one and the same event or state.

Notice that backward gapping, as illustrated in (89) and (93), clearly expresses complex events, unlike coordinated wh-questions. This is naturally attributed to the fact that in this construction, the phrases left in each conjunct after rightward ATB movement is applied are contrastively focused; in both (89) and (93), John-ga and Mary-o are contrastively focused with Bill-ga and Susan-o, respectively. To accommodate this fact, I modify (95) into the following:²³

(96)	When V1, V2, … or their projections undergo ATB movement, the resulting chain denotes one and the same event or state, unless each V takes arguments or modifiers that are contrastively focused.

I simply assume (96) as an interpretive rule for ATB movement of shared Vs or their larger projections, leaving aside the question of why such ATB movement manifests this semantic property.

Recall that one of the advantages of the RNR analysis of Japanese coordinated wh-questions over the backward ellipsis analysis was that the former correctly captures the fact that the conjoined wh-phrases make a constituent. Here again, if the RNR analysis is right for backward gapping, it is predicted that the conjoined phrases behave as if they make a constituent, whereas the backward ellipsis analysis makes the opposite prediction. As Takano (2002) observes, the prediction made by the RNR analysis is borne out:²⁴

(97)	a.	[Hometa] no-wa	John-ga	Mary-o	sosite	Bill-ga      Susan-o           da.
		praised NL-TOP	John-NOM	Mary-ACC	and	Bill-NOM Susan-ACC COP
		‘Lit. It was John [V e] Mary and Bill [V e] Susan that e praised e.’
	b.	[Yuuenti-ni	turete-itta	no]-wa	John-ga	Mary-o sosite Bill-ga
		amusement park-to	took	NL-TOP	John-NOM	Mary-ACC and Bill-NOM
		Susan-o	da.
		Susan-ACC	COP
		‘Lit. It was John [V’ e] Mary and Bill [V’ e] Susan that e took e to an amusement park.’

(97a, b) are the cleft versions of (89) and (93), respectively, in which the coordinated phrases are clefted. These data give strong support to the RNR analysis.

In Abe and Hoshi (1997), we argue that the RNR analysis will not account for the grammaticality of the following example:

(98)	John-ga	Bill	sosite	Mary-ga	Susan-nituite	hanasita.
	John-NOM	Bill	and	Mary-NOM	Susan-about	talked
	‘Lit. John [V e] Bill and Mary talked about Susan.’ (Abe and Hoshi 1997: 111)

In this case, the shared material in the two conjuncts is -nituite hanasita ‘talked about’, which does not make a constituent. Hence in order to derive the output form of (98) correctly, we need to apply leftward movement to Bill and Susan before applying rightward ATB movement to the V’ in each conjunct that consists of the PP headed by -nituite ‘about’ and the V hanasi ‘talk’, as shown below:²⁵

This derivation would not be permissible, since Japanese does not allow for P(ostposition)-stranding by overt movement. However, Takano (2002) points out that in (98), John-ga Bill sosite Mary-ga Susan-nituite ‘John-NOM Bill and Mary-NOM Susan-about’ as a whole behaves as if it makes a constituent, as verified by the clefting test:

(100)	[hanasita	no]-wa	John-ga	Bill	sosite	Mary-ga	Susan-nituite da.
	talked	NL-TOP	John-NOM	Bill	and	Mary-NOM	Susan-about COP
	‘Lit. It was John [V e] Bill and Mary [V e] about Susan that e talked e.’
							(Takano 2002: 281)

As Takano claims, this fact strongly suggests that (98) shares its underlying structure with the following example and that its output form is derived by simply deleting the P -nituite ‘about’ that is attached to Bill in the first conjunct:

(101)	John-ga	Bill-nituite	sosite	Mary-ga	Susan-nituite	hanasita.
	John-NOM	Bill-about	and	Mary-NOM	Susan-about	talked
	‘Lit. John [V e] about Bill and Mary talked about Susan.’

Under the RNR analysis, the output form of (101) is derived by applying ATB head raising to hanasi ‘talk’ in each conjunct, as shown below:

(100) is then derived from this structure by clefting the conjoined VPs and then deleting the P -nituite that is attached to Bill. To the extent that this analysis is tenable, an example like (98) will not pose a problem to the RNR analysis.²⁶

Next, recall that we have seen that Japanese coordinated wh-questions show clause-mate effects. Under the present analysis of backward gapping, it is predicted that the two contrasted phrases in each conjunct should show such effects. Let us consider the following schematic structure:

In this structure, the second NP in each conjunct is moved out of a finite clause. From this structure, an instance of backward gapping could be constructed by applying rightward ATB movement to the matrix V’ in each conjunct that consists of CP and V. Recall, however, that under the assumption that the movement of NP in (103) is an instance of scrambling, it is ruled illegitimate since it is an instance of long-distance scrambling, hence necessarily carrying a [Focus] feature and this feature cannot be licensed in the position the NP has landed on. This is how the clause-mate effects should arise between the two contrasted phrases in each conjunct in backward gapping. This prediction is in fact borne out by the following example, which is cited from Abe and Hoshi (1997: 132) with a slight modification:²⁷

(104)	?*Harry-ga	Mary-ni	sosite	John-ga	Susan-ni [CP	sono sensei-ga t au
	Harry-NOM	Mary-DAT	and	John-NOM	Susan-DAT	that teacher-NOM meet
	to]	omotteiru.
	COMP	think
	‘Lit. Harry [V’ e] Mary and John thinks that that teacher will meet Susan.’

Under the RNR analysis, this sentence could be derived by moving Mary-ni in the first conjunct and Susan-ni in the second out of the embedded CP, exactly as shown in (103), and then applying rightward ATB movement to the matrix V’ in each conjunct that consists of the embedded CP and the V omotteiru ‘think’. The ungrammaticality of this sentence is now attributed to the illegitimate application of long-distance scrambling to Mary-ni and Susan-ni.^28,29

Finally, recall how I analyzed Kasai’s (2016) example (61), reproduced below, in terms of the RNR analysis in Section 3.2:

(105)	Taroo-wa	dare-no	sosite	nani-o	nusunda	no?
	Taro-TOP	who-GEN	and	what-ACC	stole	Q
	‘Lit. Whose and what did Taro steal?’

I assumed that dare-no ‘who-GEN’ and nani-o ‘what-ACC’ undergo scrambling out of each conjunct before the relevant rightward ATB movement is applied, as shown below:

In this derivation, I maintained that the apparent LBC violation induced by scrambling dare-no out of the dominating NP is remedied by “wiping out its trace” by way of applying ATB rightward movement to conjoined VPs. Now given that backward gapping is analyzed on a par with coordinated wh-questions in terms of RNR, it is expected that there should be cases of backward gapping that tolerate LBC violations. This expectation is in fact fulfilled by the following example:

(107)	John-ga	Mary-no	sosite	Bill-ga	Susan-no	kodomo-o	hometa.
	John-NOM	Mary-GEN	and	Bill-NOM	Susan-GEN	child-ACC	praised
	‘Lit. John [V e] Mary’s (child) and Bill praised Susan’s child.’

Under the present analysis, this example is derived in the following way:

In (108a), Mary-GEN in the first conjunct and Susan-GEN in the second are moved out of V’, inducing LBC violations, and in (108b), the lower V’ of each conjunct undergoes rightward ATB movement, which correctly gives rise to the output form of (107). The acceptability of this example then indicates that the LBC violations in question can be remedied by wiping out the traces of Mary-GEN and Susan-GEN.

From the above discussion, it is reasonable to conclude that backward gapping is best analyzed by the RNR strategy. This accords well with the above argument that the backward ellipsis strategy is unavailable to coordination in Japanese.

6 Conclusion

In this paper, I argued for the RNR analysis of coordinated wh-questions in Japanese, according to which verbs or their larger projections are moved rightward across-the-board in the coordinated structures, with the conjoined wh-phrases staying in their original VP domains. I demonstrated that this analysis can properly capture the following properties of this construction: (i) the conjoined wh-phrases retain the in-situ property of wh-phrases in this language; (ii) they behave as if they make a constituent; and (iii) they are sensitive to the clause-mate condition. The most crucial theoretical implication of my arguments for the RNR analysis is that the backward ellipsis analysis is inaccessible to coordination in this language. This is further confirmed by the behaviors of what I call backward gapping, which is also amenable to the RNR analysis. Examining whether this implication holds even cross-linguistically, I reached only the tentative conclusion that it might not accord with what has been found out by the bi-clausal analysis of coordinated wh-questions in other languages.