It has been controversial whether Japanese has verb-stranding VP-ellipsis (VSVPE) (Funakoshi 2016; Landau 2020: among many others). The Adjunct-Inclusive (AI) interpretation with null adjuncts has been used as a tool to diagnose whether elliptical objects are derived via argument ellipses or VSVPE (Oku 1998; Landau 2023; Simpson 2023: inter alia). Against this backdrop, Tanaka (2023) has recently claimed that Japanese lacks AI readings, and that adjuncts are therefore unelidable. In response, this squib argues that adjuncts are indeed elidable in Japanese either through VSVPE or adjunct deletion operations. I discuss the replicability of AI interpretations and argue that research on null adjuncts requires careful and rigid control of test sentences and discourse properties.

The rest of this paper is organized as follows. Section 2 reviews general characteristics of null adjuncts and the observations from Tanaka (2023). Section 3 demonstrates that Japanese does allow AI readings with null adjuncts, and thus, adjuncts are elidable, consistent with Oku (2016), Landau (2023), and Tanabe & Kobayashi (2024a). I then propose an interpretative constraint on coordinated sentences with multiple ellipsis sites. Section 4 argues against Tanaka’s (2023) analysis of inter-speaker variation in AI readings based on Right-Dislocation (RD). Finally, Section 5 concludes the paper.

Whether Japanese elliptical sentences allow for an AI interpretation with null adjuncts has been debated since Oku (1998). Consider the examples in (1), where both the object and adjunct are missing from the elliptical clause.

Oku (1998: 172) notes that it is nearly impossible to obtain an AI interpretation with examples such as (1); that is, the data in (1b) lacks the reading “John didn’t wash the car in a careful manner.” Instead, (1b) is unambiguously interpreted as “John didn’t wash the car (at all).”

In this context, Funakoshi (2016) argues that (1b) can have an AI reading. He slightly modifies (1), as shown in (2), and generalizes the behavior of null adjuncts in (3).

Clearly, the AI interpretation in (2) is easier to obtain when the object and adjunct are elided. Based on this observation, Funakoshi (2016) claims that VSVPE derives the AI reading, as schematically illustrated in (4).

In response, Tanaka (2023) has recently argued that AI readings are generally unavailable, citing examples such as (5). He claims that the AI interpretation is absent because the first conjunct in (5b) contradicts the second. If VSVPE were available, (5) should allow an AI reading; therefore, he concludes that Japanese lacks VSVPE and null adjuncts.

Tanaka further supports his claim with novel data involving the degree adverb tyuutohanpani ‘halfheartedly’ in (6). When the adverb is unpronounced, the sentence in (6) means “Taro doesn’t study math at all,” whereas with the overt adverb, it means “Taro studies math enthusiastically.” As the AI reading is not available in (6), he concludes that Japanese does not allow null adjuncts and that VSVPE is not a viable operation.1

As shown above, the debate over the availability of AI interpretations with null adjuncts in Japanese remains unresolved. While Tanaka’s (2023) observations suggest that AI readings are unavailable in certain contexts, Funakoshi’s (2016) data indicate that AI interpretations are possible in others. In this context, I argue that Japanese does allow AI readings, based on a close examination of Tanaka’s observations and novel supporting evidence. Accordingly, this paper points to the possibility that VSVPE is available (Funakoshi 2016) or that Adjunct Ellipsis, which directly deletes adjuncts, exists in Japanese (Oku 2016; Kobayashi 2020; Landau 2023; Tanabe & Kobayashi 2024a).

I demonstrate that one cannot simply argue whether there is (or is not) VSVPE in Japanese based on the availability of the AI interpretation with only a limited range of examples. Two arguments are in order against Tanaka’s (2023) observations.

Tanaka (2023: 46) attributes the variability in judgments of AI interpretations to the marginal availability of remnant ellipsis in RD. He analyzes (5b) as having the bi-clausal RD structure in (23) (cf. Tanaka 2001), in which clausal ellipsis applies to [the bracketed clause].

Subsequently, Tanaka (2023: 16) claims that the dislocated remnant adjunct (i.e., teineini) undergoes ellipsis. Tanaka assumes that this covert RD is only marginally available to some speakers, which explains why AI readings are generally difficult (or even impossible) to obtain.13

I argue that RD analysis has both conceptual and empirical problems. First, although Tanaka (2023) attributes the inter-speaker variation in the AI interpretation to the availability of covert RD, this reasoning essentially restates Oku’s (1998) assumption that adjuncts are unelidable in Japanese. The covert RD analysis states that AI readings are absent because dislocated remnant adjuncts are unelidable for most speakers, which is basically the same as what has been widely assumed since Oku (1998).

Furthermore, this reasoning is empirically inaccurate, particularly regarding island sensitivity. Tanaka (2023) discusses (24), which is a follow-up to (5a): Taro-wa kuruma-o teineini arat-ta ‘Taro washed the car carefully.’ He claims that (24) is degraded due to Coordinate Structure Constraints (CSC: Ross 1967).14

As already noted in (15), Tanaka (2023) overlooks the fact that the data in (24) contains another gap in the second conjunct. Therefore, it is possible for the adjunct to undergo across-the-board movement. The current analysis predicts that the gaps in each conjunct in (24) contain either (i) kuruma-o ‘the car’ and teineini ‘carefully’ or (ii) only kuruma-o ‘the car’. In either case, (24) results in a contradiction: #Hanako didn’t wash the car (carefully), but she did wash the car (carefully). Thus, one cannot conclude that the unacceptability of (24) is due to a CSC violation, which significantly undermines Tanaka’s (2023) RD analysis.

Next, we examine further evidence against the covert RD analysis from Tanaka (2023), focusing on the Complex NP Constraint (CNPC). Tanaka (2023: 21) argues that the sentence in (25) is degraded due to CNPC violation. However, this expression is unacceptable regardless of whether the adjunct is dislocated or pronounced in-situ, as shown in (26), which is a follow-up to (25a).

I suspect that the unacceptability of (25) stems not only from a potential CNPC violation but also from semantic oddness; it does not make sense (unless supplemented by some unnatural and peculiar contexts) that the dirty car “also” became clean without being carefully washed. The function of additive -mo ‘also’ in these examples remains unclear.

Let us now consider more natural sentences, both logically and contextually. (27) and (28) are uttered in response to the following question: Which subject did Taro and Hanako study halfheartedly? The data in (28) are especially noteworthy: The expression is acceptable regardless of whether the adjunct tyuutohanpani ‘halfheartedly’ is pronounced in-situ, unlike in (26). Nevertheless, the RDed counterpart in (27b) is unacceptable, despite the AI reading being available in (28). This is not predicted by Tanaka’s (2023) covert RD analysis. The fact that the acceptability of (27b) does not correlate with that of (28) undermines Tanaka’s (2023) covert RD analysis of the AI interpretation. This finding indicates that covert RD is irrelevant to AI interpretations.15

Finally, I present two pieces of direct counterevidence to the covert RD analysis of AI interpretations. First, Tanaka’s (2023) reasoning for the contrast in (29) is self-contradictory. While he attributes the unacceptability of (29a) to a contradiction between the antecedent clause and the elliptical clause (Tanaka 2023: 18), he overlooks the fact that the same contradiction is mirrored in (29b) as well. The antecedent clause “Hanako does not wash the car at all” contradicts the elliptical clause “Hanako washed the car, but not in a careful manner.”

Thus, Tanaka’s RD analysis incorrectly predicts that the oft-cited example from Oku (1998) in (29b) should be unacceptable due to inconsistency, which is contrary to fact.16

Second, if covert RD were the source of AI readings, then the contrast between (6) (VN+light verb) and (7) (native verb), as discussed in Section 3 should also be observed in their RDed counterparts. However, no such contrast is found. Example (30) with manab- ‘study’ is just as unacceptable as (30) with the VN+su benkyoo-s- ‘study-do’.17

The contrast between (6) and (7) does not carry over to the pair in (30), further undermining the RD analysis.18

In this section, I have demonstrated that Tanaka’s (2023) RD analysis of AI interpretations is untenable. Therefore, I have suggested two alternative analyses of null adjuncts: VSVPE and Adjunct Ellipsis. While Funakoshi (2016) and other advocates of the head-stranding ellipsis analysis (Sato & Hayashi 2018; Sato & Maeda 2021: among many others) argue that null adjuncts are derived via string-vacuous overt verb-raising and the subsequent remnant phrasal ellipsis, Oku (2016), Kobayashi (2020; 2025), Landau (2020; 2023), and Tanabe & Kobayashi (2024a), inter alia, argue against such an approach and propose an alternative analysis based on argument ellipsis with no recourse to syntactic verb-raising. Further research is needed to determine which of these two accounts more adequately captures the nature of null adjuncts and the AI interpretation. This remains an open question for future research.

In this squib, I have argued against Tanaka’s (2023) claim that adjuncts are generally not elidable in Japanese. It is inconclusive to claim that the language lacks VSVPE or Adjunct Ellipsis based solely on observations of the apparent lack of AI reading. I demonstrated that careful control of test sentences and discourse factors is necessary to guarantee the replicability of AI interpretations with null adjuncts. The observations, especially those in (7) and (12), provide further evidence of VSVPE (Funakoshi 2016) and/or Adjunct Ellipsis (Collins 2015; Oku 2016; Kobayashi 2020; 2025; Tanabe & Kobayashi 2024a). Nevertheless, the Adjunct Ellipsis analysis can be suggested to have broader empirical coverage. Previous studies such as Kobayashi (2020; 2025) and Tanabe & Kobayashi (2024a) provide empirical evidence against the VSVPE analysis of AI interpretation. Furthermore, I have mentioned in Footnote 3 that the observation from the previous literature in (2) that adjuncts cannot be elided independently does not support the conclusion that VSVPE is preferable to the Adjunct Ellipsis analysis. However, various other elliptical phenomena must be considered to determine which analysis is empirically and conceptually superior.

In addition to the open question of which of the two analyses is more appropriate, this paper leaves an important issue unresolved: Why is there individual variability in the judgment of AI interpretation?19 Although I cannot fully discuss this issue in this squib, this paper at least demonstrates that the RD analysis of Tanaka (2023) does not sufficiently address this question. Based on the observations in this paper, I speculate that the inter-speaker variation in AI reading is not syntactic but instead due to pragmatic/discourse factors. This is because the availability of AI interpretation within a single speaker changes depending on the control of extra-syntactic factors, as seen in (6)/(7), (5)/(18), and (12), which is unexpected if inter- and intra-speaker variations are attributable to the availability of specific syntactic structures.20 Previous studies, such as Tanabe & Kobayashi (2024a; b), have argued that recovering adjunct meaning in ellipsis sites requires rigorous control of discourse properties, such as the Question Under Discussion, owing to the optional nature of adjuncts in the argument structure. Others, such as Landau (2023), claim that the AI interpretation is derived via Pragmatic Enrichment. In this pragmatic process, the elliptic site is enriched with an adjunct meaning that is recoverable from the preceding contexts. It is also conceivable that both are available in the human language. Regardless, there is no doubt that null adjuncts require heavy discourse/contextual clues and specific prosodic patterns that evoke them to obtain an AI reading (Kobayashi et al. 2024), possibly because of the optionality of adjuncts in contrast to arguments. Further investigation into this issue is left for future research. Nevertheless, this squib has undoubtedly raised questions regarding commonly held assumptions, and I believe that this opens up new avenues for future research on the hotly debated issues of null adjuncts, verb-raising in Japanese, and the ellipsis phenomena in general.