On the Licensing Condition on Sluicing: Evidence from Japanese

Abstract

Recent studies on sluicing have been pursuing its licensing condition by examining mismatch phenomena in sluicing. The groundbreaking work was by Rudin, who proposes a syntactic licensing condition on sluicing through investigating mismatch phenomena. This paper aims to critically examine Rudin’s proposal by discussing novel Japanese data regarding mismatches in modality, polarity and verbs. We show that these data challenge Rudin’s proposal both conceptually and empirically and thus suggest that it needs to be re-examined. We then show that Kroll’s semantic licensing condition on sluicing captures the Japanese data in question and thus argue that the condition has a wider empirical coverage compared with Rudin’s proposal.

1. Introduction

Recent research on the licensing condition on sluicing has been advancing by examining mismatch phenomena, where TP in a wh-question is elided under incomplete identity to the antecedent clause. The groundbreaking study was Rudin (2019), who investigated mismatch phenomena and argues to restrict syntactic identity to the eventive core (see Section 2 for more details). Since Rudin (2019), more and more studies have been pursuing the licensing condition on sluicing by focusing on mismatch phenomena: Anand et al. (2025) expands Rudin’s proposal to the Argument Domain, which includes any thematic domain, such as small clauses; Ranero (2021) relaxes identity above the root level to featural non-distinctness; Kroll (2019), focusing on semantic-pragmatic constraints of sluicing, revises the Givenness Condition to account for local context updates that can render propositions beyond the antecedent.

Most of these studies assume, following Rudin’s proposal, that constraints on sluicing involve a division of labor between syntax and semantics/pragmatics. According to this view, mismatch tolerance in sluicing crucially depends on structural identity in a certain syntactic domain. Once such structural conditions are satisfied, interpretive constraints come into play to determine the acceptability of specific instances, depending on the contextual and semantic properties of the elided material.

The aim of this paper is to critically examine the syntactic constraints proposed by Rudin (2019). More specifically, we focus on novel mismatch data from Japanese and show that they challenge Rudin’s proposal. We then argue that these facts are more naturally accounted for under a semantic approach, particularly that of Kroll (2019). Note, however, that this should not be taken as conclusive evidence that semantic accounts are categorically superior to syntactic ones. As Rudin, Kroll, and others have noted, syntactic and semantic constraints are not necessarily in competition but in fact coexist. Thus, the perspective developed in this study is intended to suggest that semantic constraints may play a more decisive role than previously assumed in licensing sluicing constructions.

The structure of this paper is as follows. Section 2 provides an overview of Rudin’s syntactic approach to sluicing. Section 3 examines novel Japanese data and points out both conceptual and empirical issues of his proposal. Section 4 then demonstrates that the problematic data highlighted in the previous section can be accounted for under the semantic approach proposed by Kroll. Section 5 concludes the paper.

2. Rudin (2019): Syntactic Licensing Condition

This section provides a brief overview of the syntactic licensing condition on sluicing proposed by Rudin (2019). His proposal is motivated by the relationship between the possibility of mismatches under sluicing and what he calls the eventive core. The latter notion, which originated from Langacker (1974), refers to the ‘objective content’ of a sentence, which includes the verb and its arguments but excludes such elements as tense, modality and illocutionary force. Two observations are of particular relevance. First, under sluicing, mismatches are not possible for elements within an eventive core. For example, Merchant (2013) observes that voice (i.e., active/passive) mismatches are not allowed under sluicing, as exemplified in (1).1 (Hereafter, ${\langle \cdots \rangle }_A$ and ${\langle \cdots \rangle }_E$ in examples stand for an antecedent clause and elided clause (or ellipsis site), respectively).

(1)	a. *〈Joe was murdered〉A, but we don’t know who 〈murdered Joe〉E.
	b. *〈Someone murdered Joe〉A, but we don’t know by who 〈Joe was murdered〉E.
	(Merchant, 2013, p. 81)

Given that verbs represent the distinction of voices and that voices alter argument structures, it is plausible to assume that voices are associated with eventive cores. Second, under sluicing, mismatches are allowed with elements outside of the eventive core, in contrast to those inside of it. Such elements include those related to finiteness, tense, modality, polarity and illocutionary force; witness (2)–(6), all of which are taken from Rudin (2019, p. 266).2

(2)	Mismatches in finiteness
	〈Sally cooks〉A every night; she learned how 〈Sally to cook〉E from her father.
(3)	Mismatches in tense
	〈Your favorite plant is alive〉A,
	but you can never be sure for how long 〈your favorite plant will be alive〉E.
(4)	Mismatches in modality
	Sally said that 〈customers should be given lower rates〉A,
	but Susie said it’s hard to see how 〈customers could be given lower rates〉E.
(5)	Mismatches in polarity
	〈Either turn in your final paper by midnight〉A
	or explain why 〈you didn’t turn it in by midnight〉E.!
(6)	Illocutionary mismatches
	〈Always save a little from each paycheck〉A. Once you’re older,
	you’ll understand why 〈you should always save a little from each paycheck〉E.

Building on these two contrastive facts (and earlier suggestions by Chung (2013)), Rudin (2019) proposes a syntactic licensing condition for sluicing, with the well-accepted assumption that sluicing involves deletion of TP at PF (e.g., Merchant, 2001).3 This condition and the related notions are exhibited in (7) and (8)–(10), respectively.

(7)	Syntactic condition on sluicing:
	Given a prospective ellipsis site E and its antecedent A, nonpronunciation of the phonological content associated with any head $h\in E$ is licit if at least one of the following conditions holds:
	a. h did not originate within E’s eventive core.
	b. h has a structure-matching correlate $i\in A$.	(Rudin, 2019, pp. 269–270)
(8)	Eventive core (to be revised)
	The eventive core of any TP is its highest vP.	(Rudin, 2019, p. 269)
(9)	Structure matching
	A node n in domain dstructure-matches a node $n^{\prime }$ in domain $d^{\prime }$ iff n and $n^{\prime }$ are dominated by an identical sequence of immediately dominating nodes within d and $d^{\prime }$.
		(Rudin, 2019, p. 258)
(10)	Correlate
	A node n can be a correlate for a head h iff at least one of the following conditions holds:
	a. n is a head and n and h are tokens of the same lexical item.
	b. n is coindexed with h.	(Rudin, 2019, p. 264)

According to (7), the process of sluicing involves checking the possibility of deletion (of TP) ‘head by head.’ The deletion mechanism first checks whether a certain head, or more precisely, the tail of its movement chain, is located within an eventive core, as (7a) states. Rudin maintains, as (8) shows, that an eventive core syntactically corresponds to (the highest) vP, given that it is the domain where a verb and its arguments originate. If a head does not originate within vP, it can be freely elided; if it is, as (7b) states, it must be further checked with regard to its structure-matching correlate, which is defined based on (9) and (10). To put it simply, according to (7b), a head within vP can be elided if the antecedent clause has the same lexical item or a coindexed item in the same syntactic position as that head.

With this proposal, the mismatches under sluicing in (1) are not allowed because the verb and its arguments in the elided clause, which originated within vP, do not have any structure-matching correlate. In contrast, the mismatches in (2)–(6) are possible, with all the mismatched elements being outside of vP.

3. Some Doubts on Rudin’s Syntactic Licensing Condition

This section presents three sets of data that cast doubt on Rudin (2019)’s syntactic licensing condition for sluicing illustrated in the previous section. Since these data build on novel observations from sluicing-like constructions in Japanese, we first provide a brief overview of those constructions.

3.1. Sluicing(-like Constructions) in Japanese

Japanese has sluicing-like embedded wh-questions that consist only of a wh-phrase and the sentence-final question marker ka; for example, see (11a), which is interpreted in the same way as (11b).

(11)

This construction has been argued to involve deletion of a proposition content (at PF) like genuine sluicing.4 However, it is unclear whether it can be analyzed in the same way as genuine sluicing, involving wh-movement plus deletion of TP, since Japanese is a wh-in situ language. Most of the relevant previous works have argued that the construction in question is derived in a way different from genuine sluicing (e.g., Nishiyama et al., 1995; Kim, 1997; Hiraiwa & Ishihara, 2002, 2012; Saito, 2004; Kimura & Takahashi, 2011; Abe, 2015; Fujiwara, 2022; Noguchi, 2022; but see D. Takahashi (1994) and Takita (2011)). While their proposals regarding the derivation of the sluicing-like construction are different, what many of them have in common is that it is derived from a so-called in situ focus construction, which sentence-finally involves a complementizer no and the copula da. For example, (12) involves an embedded wh-question that an in-situ focus construction underlies. For example, the embedded wh-question in (12) has the sentential structure of an in-situ focus construction.

(12)

One piece of evidence for the in situ-focus-construction-based analysis is the fact that the sluicing-like construction can optionally contain the copula da, which appears in in situ focus constructions as well; for instance, see (13), which obtains by adding da to the elliptical wh-question in (11a).

(13)

Following Hiraiwa and Ishihara (2002) and others, we assume that no and da head Fin(ite)P and Foc(us)P (Rizzi, 1997), respectively. Under this assumption, the embedded wh-question in (12) has the structure in (14).

(14)

We further follow Hiraiwa and Ishihara (2002) and assume that the sluicing-like construction is derived by deleting FinP after a wh-phrase undergoes focus movement, or movement to Spec, FocP. It should be noted that we assume this analysis for expository purposes; the discussion would go in the same way even if we posited another analysis in the literature. With this assumption, the elliptical wh-question in (11a)/(13) has the structure in (15), being derived from (14).

(15)

Under the current assumptions, Japanese sluicing-like constructions are similar to genuine sluicing in that a wh-phrase undergoes (A’-)movement. However, the two differ in the projection to be deleted; in Japanese sluicing-like constructions, what is elided is FinP, while it is TP in the case of genuine sluicing. In this paper, we assume that this difference is only trivial and thus that the licensing condition for genuine sluicing applies to Japanese sluicing-like constructions as well. Given this assumption, we will refer to the sluicing-like construction simply as sluicing, to simplify the discussion.

3.2. Three Sets of Data

We now turn to the three sets of data that question Rudin (2019)’s syntactic licensing condition on sluicing. The first set of data, which concerns modals appearing within vP, points to a conceptual problem in Rudin’s proposal. This problem is not so serious, given in particular that Rudin himself remarks on the same problem. However, the other sets of data, which concern mismatches in polarity and verbs, pose more serious empirical problems to Rudin’s proposal and thus suggest the necessity of reconsidering the licensing condition in question.

3.2.1. Modals Within vP

To begin with, we will discuss modal elements that appear within vP. We will first introduce Rudin (2019)’s discussion on modal verbs in German, and discuss his response to the data and its conceptual problem for his proposed licensing condition for sluicing. We will then discuss a novel set of data from Japanese concerning directive sentences, which will turn out to give rise to the same problem as German modal verbs.

Let us first discuss German. In German, modals appear as verbs in sentences, unlike in English, where a modal appears on T (e.g., (4)). This means that in German a modal is accompanied by an independent vP. Recall here that, according to Rudin (2019)’s licensing condition for sluicing, an element cannot be freely deleted under sluicing if it is within the highest vP. It is then predicted that German modal verbs cannot be mismatched under sluicing. Rudin (2019) observes, however, that this prediction is not borne out; mismatches of modal verbs are possible in German, as shown in (16), which he credits to Andreas Walker p.c.

(16)
	‘This was a problem that physics had to solve, but for a long time it wasn’t clear how.’
		(Rudin, 2019, p. 271)

To capture this observation, Rudin revises the definition of the eventive core in (8) and presents (17).

(17)	Eventive core (final definition)
	The eventive core of a clause is its highest vP that is associated with an event-introducing predicate.
	(Rudin, 2019, p. 271)

Rudin notes that modal verbs are not event-introducing predicates, given that they “only communicate something about the modal status of their prejacent proposition [and] thus they are parasitic on the event introduced by the matrix verb of that prejacent” (Rudin, 2019, p. 271). Hence, in (16), the modal verb in the elided clause is outside of the eventive core, or vP associated with the event-introducing predicate lösen ‘solve’, and thus can be freely deleted. This solution is not without any problems, however. Notice in particular that the revised definition of the eventive core (17) draws on the semantic notion “an event-introducing predicate”. This seems to be incongruent with Rudin (2019)’s original intention to propose a syntactic licensing condition for sluicing. In fact, Rudin himself makes a remark on this point in a footnote.5 To add to this conceptual problem, below we will discuss novel data regarding directive sentences in Japanese.

Consider next Japanese directive sentences. In Japanese, there are several typical expressions that show directive speech acts. Here we will focus on two such expressions. One is imperatives, where an imperative morpheme is attached to a verb stem. Ref. (18) shows two examples of Japanese imperatives.

(18)

As (18) shows, the imperative morpheme has two allomorphs: One is -e, which is used when a verb stem ends with a consonant (e.g., (18a)), while the other is -ro, which is appended to vowel-ending verb stems (e.g., (18b)). Another expression that shows a directive speech act is sentences with yooni, which follows a verb with the present tense. We will refer to this type of expression as a yooni-sentence. Ref. (19) shows examples of yooni-sentences.

(19)

To discuss the crucial data involving these two directive expressions, we will now establish the assumptions of their syntax and semantics. Concerning the syntax, notice that the two expressions differ in what element the relevant morphemes follow. On the one hand, imperative morphemes are attached to verb stems, as observed in (18); they cannot be appended to tensed predicates, as (20) shows (Saito, 2015).

(20)

With the plausible assumption that a verb stem composes vP with its arguments, the above morphological fact suggests that an imperative morpheme merges with vP. Specifically, we follow Mihara and Ebara (2012) and Mihara (2015), and assume that an imperative morpheme is an adjunct of vP (cf. Saito, 2015). Under this assumption, the examples in (18) can be structurally represented as in (21).

(21)

On the other hand, yooni follows present-tensed predicates (see (19)), which suggests that the element in question merges with TP. It is then plausible to assume that yooni is the head of CP; I refer the reader to Uchibori (2000) for more discussion on the syntactic status of yooni as a complementizer. With this assumption, (19) can be structurally represented as in (22).6

(22)

Concerning the semantics, we will adopt the modal-based analysis of imperatives proposed by Schwager (2006)/Kaufmann (2012), according to which imperatives involve a special necessity modal called an imperative modal. A more detailed definition of this modal will be provided in the next section. Here, we assume that the directive force of imperatives and yooni-sentences in Japanese derive from the special necessity modal in them and that that modal is encoded in the imperative morpheme and yooni respectively (cf. Ihara & Noguchi, 2018; Ihara, 2020, 2021). Incidentally, under this assumption, if (i) an imperative or a yooni-sentence serves as the antecedent clause of sluicing and (ii) the elided clause is interpreted with the meaning of a necessity modal like must/should, this sluicing can be viewed to involve a matching with respect to a modal element (Noguchi & Ihara, 2020). The relevant example is shown in (23), where the elided clause in (23a), doko-o ka ‘where-acc q’, is interpreted to involve a necessity modal as in (23b).

(23)

With this much background, consider now the predictions that Rudin (2019)’s syntactic licensing condition for sluicing makes with regard to the two directive expressions under consideration. Recall that according to Rudin’s proposal, mismatches under sluicing are allowed for elements outside of vP, but not for those within vP. It is then predicted that the necessity modal in yooni-sentences, which appears on C, can be mismatched under sluicing, while the one in imperatives, which occurs within vP, cannot. To examine whether these predictions are borne out, consider first the baseline example in (24).7

(24)

Notice in particular that in (24) the embedded wh-question in the second sentence cannot be interpreted with the necessity modal; such an interpretation would make the sentence infelicitous under the given context. This means that if sluicing applies to the embedded wh-question in (24) with the first directive sentence being the antecedent clause, a mismatch will be forced with respect to the modal elements. With this baseline in mind, consider now the two crucial cases that obtain by applying sluicing to (24). The first case is displayed in (25), where the first sentence is the yooni-sentence.

(25)	[The context is the same as (24):]

(25) can be felicitously interpreted. This means that a mismatch is allowed in (25) with respect to the modals. With our assumption that the necessity modal in yooni-sentences appears at C, this is correctly predicted from Rudin (2019)’s syntactic licensing condition on sluicing, according to which elements above vP can be unconditionally elided. Consider next the second case in (26), where the imperative is used in the first sentence.8

(26)	[The context is the same as (24):]

(26) can be felicitously construed, as (25) can. This means that a mismatch is allowed with respect to the modals in this case as well. Notice, however, that this felicity is not predicted from Rudin (2019)’s ‘original’ licensing condition for sluicing, according to which elements within the highest vP cannot be freely deleted; under our assumption, the necessity modal of imperatives occurs within vP. In other words, the felicity of (26) brings the same conceptual problem to Rudin’s proposal as modal verbs in German.

3.2.2. Polarity Mismatches in Japanese

The second data that makes dubious Rudin (2019)’s licensing condition on sluicing concerns polarity mismatches. An English example of this mismatch is already shown in (5), repeated in (27). (28) exhibits another example from Kroll (2019).

(27)	Mismatches in polarity
	〈Either turn in your final paper〉A by midnight or explain why 〈you didn’t turn it in by midnight〉E!
		(Rudin, 2019, p. 266)

(28)	I don’t think that 〈California will comply〉A, but I don’t know why 〈California won’t comply〉E.
	(Kroll, 2019, p. 2)

We now observe that polarity mismatches are possible in Japanese as well. See, for example, (29a), whose (possible) interpretation is shown in (29b), with the polarity differing between the antecedent and (prospective) elided clause.

(29)

Recall that according to Rudin’s licensing condition on sluicing, elements outside of vP can be elided without any condition. Hence, his proposal can capture the possibility of mismatches of the polarity, which appears above vP, at least in some languages including English. However, we will show that polarity mismatches in Japanese pose a problem to the licensing condition for sluicing under consideration, in particular by examining the morphology of negation in Japanese.

In Japanese, the negative form of a verb is obtained by attaching -nai ‘not’ to the verb stem. However, if a verb stem ends with a consonant, another exponent -a appears between the verb stem and -nai. See relevant examples in (30).

(30)	a.	Negative forms of verbs with a vowel-ending stem
		tabe-nai ‘not eat’, ne-nai ‘not sleep’, oki-nai ‘not get up’, mi-nai ‘not look’, etc.
	b.	Negative forms of verbs with a consonant-ending stem
		hanas-a-nai ‘not speak’, kik-a-nai ‘not listen’, sitagaw-a-nai ‘not comply’, etc.

Regarding the exponent -a, the recent work by H. Takahashi and Nakajima (2023) claims that it is the head of vP. In terms of semantics, they argue that the head of vP -a encodes the inchoative aspect, indicating ‘emergence’ or ‘existence’. 9 With their claim, a negative sentence in Japanese compositionally denotes the negation of the ‘emergence’ or ‘existence’ of the event denoted by VP. For example, consider the negative sentence in (31a), whose syntactic structure is represented in (31b).

(31)

According to the structure in (31b), (31a) is compositionally interpreted as follows: the event of Ken’s drinking beer did not ‘emerge’ or ‘exist’.

Assuming H. Takahashi and Nakajima (2023)’s claim regarding the negation in Japanese, let us now turn back to (29a), the example of a polarity mismatch under sluicing in Japanese. Suppose that (29a) is the result of applying sluicing to the embedded wh-question in (29b). Then, under the current assumption, TP of the antecedent and elided clause can be structurally represented as in (32a) and (32b), respectively.

(32)

Recall that according to Rudin (2019)’s licensing condition for sluicing, elements within vP cannot be elided unconditionally; to be elided, they must have (structure-matching) correlates in the antecedent clause (see (7)–(10)). With this in mind, notice now that the v, -a, in (32b) does not have a correlate in (32a), whose v is distinct from the one in (32b). It is then predicted that sluicing cannot apply to the embedded wh-question in (29b), which is contrary to fact, as the acceptability of (29a) indicates. This observation thus suggests that Rudin’s licensing condition for sluicing should be reconsidered.

3.2.3. Verb Mismatches

The final data that casts doubt on Rudin (2019)’s licensing condition for sluicing involves the deletion of a verb that lexically differs from but is semantically similar to the one in the antecedent clause. More concretely, we will focus on the two verbs: dokusyosu- ‘read books’ and yom- ‘read’. They are semantically similar in that they share the meaning of reading. However, they are semantically different in whether they include the meaning of books as the object of reading. This difference is reflected in their syntax; dokusyosu- cannot take an object while yom- must, as exemplified in (33).10

(33)

With this syntactic difference in mind, consider now the crucial data in (34a), where sluicing applies to the embedded wh-question in the second sentence.

(34)

According to Rudin (2019)’s licensing condition for sluicing, the verb must have a (structure-matching) correlate in the antecedent clause so that it can be deleted under sluicing (see (7)–(10)). It is then expected that in (34a), the verb that is deleted in the embedded wh-question in the second sentence is dokushyosu- ‘read books’, which is also used in the antecedent clause. This expectation cannot be correct, however, given that the remnant wh-phrase in (34a) is an object wh-phrase; dokushyosu- cannot take an object, as shown in (34b) (and (33a)). Rather, given the interpretation of the elided clause, the verb that is deleted under sluicing in (34a) should be yom- ‘read’, which can take an object as shown in (34b) (and (33b)). Therefore, (34a) can be viewed to exemplify a mismatch of verbs under sluicing, which cannot be captured by Rudin’s licensing condition on sluicing. The condition under consideration thus needs to be re-examined.

To recap this section, under the assumption that the licensing condition on sluicing applies to similar constructions in Japanese as well, we have discussed three data sets from Japanese which suggest that Rudin (2019)’s syntactic licensing condition on sluicing should be reconsidered; the first data, which concern modals that appear within vP, point to a conceptual problem of that condition, while the other two, which concern polarity and verb mismatches, provide more serious empirical issues.

4. Semantic Account

In this section, we argue that a semantic analysis—specifically, the dynamic semantic account proposed by Kroll (2019)—provides an effective explanation for the data discussed in the previous section that were difficult to capture under Rudin (2019)’s syntactic constraints.

It should first be noted, however, that as both Rudin and Kroll suggest, syntactic and semantic (and possibly pragmatic) generalizations are not necessarily in conflict. Thus, although the validity of the present analysis may call for some modification of the syntactic theory, it does not entail the rejection of a syntactic approach. What this section aims to demonstrate is (i) the possibility that semantic constraints, at least for certain phenomena, have broader empirical coverage than Rudin’s, and; (ii) that Kroll’s semantic account can explain not only the polarity mismatch she originally focused on, but also some phenomena in Japanese, which may not be attested in English or some other languages.

In Section 4.1, we first gives an overview of Kroll’s semantic licensing condition on sluicing, based on polarity mismatch phenomena. Section 4.2 then shows that her proposal captures the three Japanese mismatch data discussed in Section 3.2.

4.1. Kroll (2019): Local Givenness

Kroll (2019) introduces a dynamic semantic approach to the constraints on sluicing, empirically motivated by data from polarity mismatches (e.g., (28) repeated in (35)), which could not be predicted by earlier semantic theories of sluicing, including the one by Merchant (2001).11

(35)	I don’t think that 〈California will comply〉A, but I don’t know why ${\langle }_{TP}$California won’t comply${\rangle }_E$.
		(Kroll, 2019, 2)

In the spirit of Schwarzschild (1999)’s GIVENness theory, Kroll proposes a new semantic identity condition called Local Givenness, which argues that sluices are licensed by local contextual entailment in the sense of Heim (1983). In Heim’s Context Change Potentials (CCPs) system, a context is defined as a set of worlds, and crucially, a sentence contributes a function from contexts to contexts. The notion of ‘contexts’ here is distinguished into global contexts, contexts at which the current sentence is interpreted, and local contexts, contexts at which the current clause is interpreted (Karttunen, 1973), with the latter being particularly crucial in Kroll’s analysis.

Here are the basic assumptions she makes: (i) a context c and a proposition p are both defined as sets of worlds. Thus, if a context c entails a proposition p, then $c\subseteq p$. (ii) A context c is updated with a new proposition p by conjoining, or intersecting, with p. That is, if c entails the presuppositions of p, then $c+p=c\cap p$; if c does not entail the presuppositions of p, then c is undefined.

With these assumptions, informally speaking, Kroll proposes that the TP of an interrogative can be elided if and only if the proposition expressed by the TP, modulo existential closure, is entailed by the context in which the sentence expressing the proposition would be uttered. This proposal is summarized in (36), where $ExClo$ stands for existential closure and $c_L$ for a local context of p.12

(36)	Local Givenness.
	A TP $\alpha$ can be deleted iff $ExClo\left({\left[\left[\alpha \right]\right]}^g\right)$ expresses a proposition p such that $c_L\subseteq p$ and p is maximally salient.
		(Kroll, 2019, (39))

Another key aspect of her approach is that, unlike Merchant (2001)’s e-GIVENness analysis, her proposal in (36) does not require bidirectional entailment between the antecedent clause and the local context, thereby constituting a less restrictive condition.13

Now, recall that the polarity mismatch is also possible in Japanese, just as it is in English (cf. (28)/(35)). In what follows, we analyze this polarity mismatch in order to demonstrate how Kroll’s analysis works. (Note, however, that as mentioned earlier, Kroll herself does not deal with Japanese data but analyzes only English examples.) A slightly modified version of example (29) is repeated in (37).

(37)	〈Kariforunia syuu-ga sitagaw-u〉A to-wa omow-anai kedo,
	naze 〈Kariforunia syuu-ga sitagaw-anai〉E ka-wa wakar-anai.
	‘〈I don’t think that California will comply〉A,
	but I don’t know why 〈California won’t comply〉E.’ (cf. (29))

The possibility of the mismatch in the data above is demonstrated through the following steps. Throughout the paper, following Kroll, a function C takes a function F and returns a set such that for any function F of type $\langle s,t\rangle$, $C\left(F\right)=\{w:F(w)=1\}$, and W is a set of all possible worlds. Kroll assumes that the assertion of the antecedent with think (in (29)/(37)), omow ‘think’) combined with the excluded-middle presupposition invoked by this verb entails that the speaker has the belief that California will not comply.14 She also assumes that the matrix think-verb is a communicative assertive verb, which asserts its complement as true in a local context. With this, the pragmatic assertion of (c) creates a local context in which the worlds under consideration are only those in which California does not comply.

(38)	a.	Denotation of A:
		${\left[\left[A\right]\right]}^g=\lambda w.\neg \forall w^{\prime }.[w^{\prime }\in DO{X}_{sp}\left(w\right)\to comply\left(CA\right)\left(w^{\prime }\right)]$, where $DO{X}_{sp}\left(w\right)$ is the set of worlds compatible with the doxastic state of the speaker.
	b.	Excluded Middle Presupposition of A:
		$\lambda w.[\forall w^{\prime }.[w^{\prime }\in DO{X}_{sp}\left(w\right)\to comply\left(CA\right)\left(w^{\prime }\right)]\vee$
		$\forall w^{\prime }.[w^{\prime }\in DO{X}_{sp}\left(w\right)\to \neg comply\left(CA\right)\left(w^{\prime }\right)]]$
	c.	Strengthened Neg-Raised Interpretation of A:
		$\lambda w.\forall w^{\prime }.[w^{\prime }\in DO{X}_{sp}\left(w\right)\to \neg comply\left(CA\right)\left(w^{\prime }\right)]$
	d.	Local Context $c_L$ for E:
		$c_L=W\cap C\left([\lambda w.\neg comply\left(CA\right)\left(w\right)]\right)=W\cap \{w:\lambda w.\neg comply\left(CA\right)\left(w\right)\}$
	e.	Denotation of $ExClo\left({\left[\left[E\right]\right]}^g\right)$:
		$ExClo\left({\left[\left[E\right]\right]}^g\right)=\{w:\lambda w.\neg comply\left(CA\right)\left(w\right)\}$, where ${\left[\left[E\right]\right]}^g=\lambda w.\neg comply\left(CA\right)\left(w\right)$
	f.	Local Givenness:
		$c_L\subseteq ExClo\left({\left[\left[E\right]\right]}^g\right)$

The denotation of A in step (a) asserts that it is not true that the speaker believes that California will comply. The excluded middle presupposition of omow ‘think’ in (b) requires that the speaker either believes that California will comply or believes that California will not comply. Together with (a) and (b), the strengthened neg-raised interpretation is derived in (c). Since the local context in (d) entails the elided clause with existential closure in (e), Local Givenness is satisfied in (f). Kroll’s theory thus predicts the felicitous elision of the TP in (29).

4.2. Application of Local Givenness

Let us now illustrate how Kroll (2019)’s proposal captures the remaining two data discussed in Section 3, namely those regarding directive sentences (Section 4.2.1) and verb mismatches (Section 4.2.2).

4.2.1. Sluicing of Directive Sentences Involves a Restriction on Worlds

As we have already partially introduced in Section 3.2.1, with regard to the semantics of imperatives, we adopt the modal approach proposed by Schwager (2006)/Kaufmann (2012), according to which imperatives semantically encode a necessity modal like should. The modal approach basically relies on a version of Kratzer (1981)’s semantics for modality in a possible world semantics, where modal expressions are interpreted as quantifiers over the sets of possible worlds compatible with a conversational background, specifically a modal base, a set of propositions that describe what is known to an agent, and an ordering source, a set of propositions that describe what is desired by an agent, what the law says, etc.15 This line of approach assumes that imperatives are interpreted as (39), where f is a circumstancial modal base, o is a prioritizing ordering source that represents rules, preferences, or goals, and $Best(f,o,w)$ are the best worlds in $\bigcap f\left(w\right)$ such that no other worlds in $\bigcap f\left(w\right)$ are strictly better than they are according to $o\left(w\right)$.

(39)	a.	$\left[\right[$ Go to bed! $\left]\right]\approx \left[\right[$ You should go to bed! $\left]\right]$
		$=\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to go(addr,bed)\left(w^{\prime }\right)]$
	b.	At all worlds $w^{\prime }$ that are optimal according to modal base f and ordering source o evaluated at w, you go to bed in $w^{\prime }$.

The example (40) (repeated and modified from (23)) refers to the case where the elided proposition has a deontic-modalized meaning. The steps in (41) apply Local Givenness to (40), where the modal is included in the interpretation of E.

(40)	〈Hayaku dokoka-o {hasir-e / hasir-u yooni}〉A.
	Doko-o 〈omae-ga {hasir-anaitoikenai / harir-ubeki / hasir-ebayoi}〉E ka-wa siranai kedo.
	‘〈Run somewhere right away〉A!
	I don’t know where 〈you {must/should/may} run〉E, though.’
(41)	a.	Denotation of A:
		${\left[\left[A\right]\right]}^g=\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)]$
	b.	Local Context $c_L$:
		$c_L=W\cap C\left({\left[\left[A\right]\right]}^g\right)=W\cap C(\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)])$
		$=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)]\}$
	c.	Denotation of $ExClo\left({\left[\left[E\right]\right]}^g\right)$:
		$ExClo\left({\left[\left[E\right]\right]}^g\right)=\{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)]\}$,
		where ${\left[\left[E\right]\right]}^g=\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f,g,w)\to run\left(addr\right)\left(w^{\prime }\right)]$.
	d.	Local Givenness:
		$c_L\subseteq ExClo\left({\left[\left[E\right]\right]}^g\right)=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)]\}\subseteq$
		$\{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to run\left(addr\right)\left(w^{\prime }\right)]\}$.

As can be seen above, since the local context (b) entails the elided proposition (d), the account correctly predicts the empirically available example.

We now turn to the mismatch case where the deontic modal is not included in the interpretation of the elided clause, despite the existence of the modalized antecedent. See the data in (42) that we have been concerned with, repeated from (26).

(42)	〈Kono kadai-wa getumatu-madeni das-e〉A.
	Dooyatte 〈kono kadai-o dasu〉E ka-wa konsyuu-juuni osie-ro.
	‘〈Submit this assignment by the end of this month〉A.
	Tell me how 〈you will submit the assignment〉E by the end of this week.’

The calculation of the applicability of Local Givenness is demonstrated through the following steps. Here, the only difference from the earlier case in (41) is that the elided proposition is not deontically modalized.

(43)	The case where a modal is included in the interpretation of E
	a.	Denotation of A:
		${\left[\left[A\right]\right]}^g=\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to submit(addr,as)\left(w^{\prime }\right)]$
	b.	Local Context $c_L$:
		$c_L=W\cap C\left({\left[\left[A\right]\right]}^g\right)$
		$=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to submit(addr,as)\left(w^{\prime }\right)]\}$
	c.	Denotation of E:
		${\left[\left[E\right]\right]}^g=\lambda w.submit(addr,as)\left(w\right)$
	d.	Denotation of $ExClo\left({\left[\left[E\right]\right]}^g\right)$:
		$ExClo\left({\left[\left[E\right]\right]}^g\right)=\{w:submit(addr,as)\left(w\right)\}$
	e.	Local Givenness:
		$c_L⊈ExClo\left({\left[\left[E\right]\right]}^g\right)$
		$=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f,o,w)\to submit(addr,as)\left(w^{\prime }\right)]\}$
		$⊈\{w:submit(addr,as\left)\right(w\left)\right\}$

Contrary to the fact, sluicing turns out to be impossible in this case; the local context (b) does not entail the elided proposition, as shown in (e).

Resolving the problem requires additional assumptions. What we should point out here is that the interpretation of an antecedent utterance and a consequent utterance in sluicing is similar to that of Conditional Conjunctions (CCs) or Imperative and Declaratives (IaDs) (Schwager, 2006) in that the first conjunct clause gets mapped onto the second clause. Intuitively, as exemplified in (44) and (45), asserting a CC/IaD commits the speaker to the second conjunct only conditionally on the first. In both cases, the first conjunct introduces a hypothetical state of affairs as the topic, relative to which the second conjunct is evaluated.16

(44)	a.	decl1 and decl2. ≈ If decl1, then decl2.	(CC)
	b.	Mary sings another song and John leaves the bar.
		≈If Mary sings another song, John leaves the bar.
(45)	a.	imp and decl. ≈ imp. If you do so, then decl.	(IaD)
	b.	Drink one more beer and I’m out of here.
		≈Drink one more beer. If you do so, I’m out of here.

We assume that sluicing constructions with imperative antecedents exhibit the close dependency found in CCs and IaDs, in that the evaluation of the elided clause of the second conjunct is anchored to the hypothetical world in which the first conjunct clause is interpreted.

(46)	a.	pA! I don’t know  wh〈q〉E, though.
		≈pA! If you do so, I don’t know wh〈q〉E, though.
	b.	〈Go somewhere〉A! I don’t know where 〈wh you will/should go〉E, though.
		≈〈Go somewhere〉A! If you do so, I don’t know where 〈wh you will/should go〉E, though.

This technically means that the domain of evaluation worlds for the proposition at the ellipsis site aligns with that of the best worlds utilized in the interpretation of the antecedent imperative clause, as schematized as follows:

(47)	a.	pA! But I don’t know wh〈q〉E.
		≈pA! But I don’t know wh〈q (at all the best world where p is true) 〉E.
	b.	〈Eat something 〉A! But I don’t know what 〈you will/should eat〉E.
		≈〈Eat something 〉A! But I don’t know what 〈you eat (at all the best world where ‘you eat something’ is true) 〉E.

Note that this analysis is in line with the approaches that explain the constraint of sluicing based on Question under Discussion (QUD, Roberts, 2012) equivalence in an important respect. QUD-equivalence approaches to sluicing (e.g., AnderBois, 2011; Barros, 2014; Weir, 2014) appeal to the intuition that assertions with indefinites and disjunctions make certain QUD salient. For instance, in “John is dancing with someone, but I don’t know who John is dancing with,” both the antecedent John is dancing with someone and the elided clause who John is dancing with share the same QUD who is John dancing with? The sluicing is possible because of this QUD-equivalence. What should be emphasized is that the QUD-equivalence approach requires that the QUD raised by the elided clause be congruent to the one raised by the antecedent. Put intuitively, in a sluicing construction, the elided clause must share the same discourse context created by the antecedent. This reflects the same line of reasoning as in the earlier case in (42)/(26), where the antecedent is a modalized sentence and its truth depends on an ideal set of possible worlds, thereby constraining the truth of the elided proposition to those same ideal worlds.

Keeping this in mind, we now re-examine the data presented earlier. In (48), what distinguishes it from (43) is that the interpretation of the elided clause E is restricted to the best worlds with respect to the background f and o utilized in the antecedent A. The assertion of the antecedent creates a local context in which the worlds under consideration are only those in which the addressee submits the assignment, given in step (c).

(48)	a.	Denotation of A:
		${\left[\left[A\right]\right]}^g=\lambda w.\forall w^{\prime }.[w^{\prime }\in Best(f^{\prime },o^{\prime },w)\to submit(addr,as)\left(w^{\prime }\right)]$
	b.	Local Context $c_L$:
		$c_L=W\cap C\left({\left[\left[A\right]\right]}^g\right)=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f^{\prime },o^{\prime },w)\to submit(addr,as)\left(w^{\prime }\right)]\}$
	c.	Denotation of E:
		${\left[\left[E\right]\right]}^g=\lambda w.[$$\forall w^{\prime }\in Best(f^{″},o^{″},w)$$\to submit(addr,as)($$w^{\prime }$$\left)\right)]$,
		where $f^{″}\mapsto f^{\prime },o^{″}\mapsto o^{\prime }$.
	d.	Denotation of $ExClo\left({\left[\left[E\right]\right]}^g\right)$:
		$ExClo\left({\left[\left[E\right]\right]}^g\right)=\{w:\forall w^{\prime }\in Best(f^{\prime \prime },o^{\prime \prime },w)\to submit(addr,as)\left(w^{\prime }\right)\}$
	e.	Local Givenness:
		$c_L\subseteq ExClo\left({\left[\left[E\right]\right]}^g\right)$
		$=W\cap \{w:\forall w^{\prime }.[w^{\prime }\in Best(f^{\prime },o^{\prime },w)\to submit(addr,as)\left(w^{\prime }\right))]\}\subseteq$
		$\{w:\forall w^{\prime }\in Best(f^{″},o^{″},w)\to submit(addr,as)\left(w^{\prime }\right)\}$

As shown in (48e), thanks to the earlier assumption that the antecedent clause restricts the worlds in which the elided clause is evaluated, the local context $c_L$ entails the elided clause E, and thus the possibility of sluicing is correctly predicted.

4.2.2. Explaining Verb Mismatch

Finally, we demonstrate that the remaining data, namely the verb mismatch, can be explained by Kroll (2019)’s theory. As we saw in Section 3.2.3, one type of Japanese data that cannot be captured by Rudin (2019)’s syntactic analysis involves a mismatch between two verbs that differ in their syntactic properties but are semantically equivalent. The relevant data is (49), repeated from (34a).

(49)	〈Ken-wa kinoo dokusyosi-ta〉A sooda. Nani-o 〈Ken-ga yon-da〉E ka-wa sir-anai kedo.
	‘I heard that 〈Ken read a book yesterday〉A.
	I don’t know what 〈Ken read yesterday〉E, though.’

The mismatch described above poses no problem under the semantic analysis. The steps involved in applying the notion of Local Givenness to (49) are shown in (51) (for simplicity, the temporal adverb kinoo ‘yesterday’ is omitted in the interpretation). Here, we translate the transitive verb yom- ‘read’ as $rea{d}_1(x,y)$ and the intransitive dokusyosu ‘read’ as $rea{d}_2\left(y\right)$. We assume that $rea{d}_1$ and $rea{d}_2$ describe the same eventuality; that is, they are semantically identical, except for the presence/absence of an syntactic object. Intuitively, they both describe the act of y engaging in reading (some book).

(50)	a.	${\left[\left[yom\right]\right]}^g⇝\lambda x\lambda y\lambda w.rea{d}_1(x,y)\left(w\right)$
	b.	${\left[\left[dokusyosu\right]\right]}^g⇝\lambda y\lambda w.rea{d}_2\left(y\right)\left(w\right)$
(51)	a.	Denotation of A:
		${\left[\left[A\right]\right]}^g=\lambda w.rea{d}_2\left(K\right)\left(w\right)$
	b.	Local Context $c_L$ for E:
		$c_L=W\cap C\left([\lambda w.rea{d}_2\left(K\right)\left(w\right)]\right)=W\cap \{w:rea{d}_2\left(K\right)\left(w\right)\}$
	c.	Denotation of $ExClo\left({\left[\left[E\right]\right]}^g\right)$:
		$ExClo\left({\left[\left[E\right]\right]}^g\right)=ExClo\left([\lambda x_e.\lambda w.rea{d}_1(x,K)\left(w\right)]\right)$
		$=\{w:\exists x.rea{d}_1(x,K)\left(w\right)\}$, where ${\left[\left[E\right]\right]}^g=\lambda x_e.\lambda w.rea{d}_1(x,K)\left(w\right)$
	d.	Local Givenness:
		$c_L\subseteq ExClo\left({\left[\left[E\right]\right]}^g\right)=W\cap \{w:rea{d}_2\left(K\right)\left(w\right)\}\subseteq \{w:\exists x.rea{d}_1(x,K)\left(w\right)\}$

As shown in step (d), Local Givenness is satisfied; the local context $c_L$ in (b) where Ken read (a book) in w entails the elided proposition in (c) that asserts Ken read some book in w. Kroll’s theory based on Local Givenness thereofore correctly predicts the verb mismatch in Japanese.

4.3. Summary

As demonstrated in this section, several Japanese data—many of which are novel and have not been previously discussed in the literature—can be fully accounted for under Kroll (2019)’s semantic identity condition.17

However, we would like to emphasize, as reiterated in the conclusion, that our proposal in this paper is not intended to refute syntactic identity conditions of sluicing, including Rudin (2019)’s, nor to modify or advance them. Rather, it only highlights that Kroll’s semantic approach empirically covers a broader range of data, including those not initially intended by Kroll. Regardless of the direction future research may take—be it refining syntactic accounts, positing some language-specific constraints, or arguing for the superiority of semantic analyses—what has become evident is that Rudin’s syntactic account cannot uniformly account for the phenomena observed in Japanese.

5. Conclusions

In this paper, focusing on Japanese, we have presented and discussed novel data that pose a new challenge to the syntactic licensing condition on sluicing proposed by Rudin (2019). More specifically, we have pointed out conceptual and empirical problems in his proposal by investigating Japanese data regarding modal mismatch in the vP domain, polarity mismatches, and verb mismatches. We have then illustrated that those Japanese data can be captured by Kroll (2019)’s semantic licensing condition on sluicing.

The findings of this paper open up multiple avenues for future research: these include revising syntactic constraints while maintaining the division of labor between syntactic and semantic conditions, exploring the possibility of cross-linguistic variation in the strength of syntactic constraints on sluicing, or investigating the potential that the decisive factor in sluicing constraints may be semantics. Regardless of which direction future research may take, the Japanese data presented here offer valuable insights.

Of course, it remains true that there are specific linguistic phenomena that support the utility of syntactic analyses (e.g., Anand et al., 2025), and these too must be examined comprehensively. At the same time, as shown by Ranero (2021)’s work on Kaqchikel (Mayan) and Landau (2023)’s on Hebrew, a broader cross-linguistic investigation of empirical data will be crucial going forward.