1 Introduction

Scrambling refers to the phenomenon whereby, in some languages such as Japanese and Korean, constituents in sentences can occur in various different orders that deviate from the canonical one (Ross 1967). For example, in the Korean sentence (1a), the object DP precedes the subject, deviating from the canonical SOV word order. Scrambling may also be cross-clausal in Korean (i.e., long-distance scrambling): in (1b), the embedded object precedes the matrix subject.

Long-distance scrambling also applies to wh-elements in Korean. Since Korean is a wh-in-situ language, wh-elements do not need to move in the formation of wh-questions, as shown in example (2a). Nevertheless, as shown in (2b), wh-elements can also be fronted through long-distance scrambling, resulting in a surface form similar to wh-questions in wh-movement languages. We shall call this operation wh-scrambling.

Long-distance scrambling in Korean has received two main classes of analyses: base-generation analyses, and movement analyses. Under the base-generation analysis, the scrambled elements are merged directly at their surface positions (Kornfilt & Bayer 1994; Neeleman 1994; Bošković & Takahashi 1998; Fanselow 2001: inter alia). For example, under Bošković & Takahashi’s (1998) proposal, the scrambled element (e.g. ‘this book’ in (1b)) is base-generated in its surface TP-adjoined position (and stays there for PF purposes). Since the scrambled element cannot be θ-marked in its surface position, as a last resort, it undergoes an LF lowering operation to receive its θ-role from the embedded V.

Under the movement-based analyses, scrambling involves movement of the scrambled element from its theta-position to its surface position (Ross 1967; Chomsky & Lasnik 1977; Saito 1982; 1985; Hoji 1985; Kuroda 1988; Fukui 1993; Lee 1993; Miyagawa 1997; Bailyn 2001; Ko 2007; 2014: inter alia).

The base-generation accounts, unlike the movement-based accounts, makes room for the possibility that scrambling may not be island sensitive: it is possible that the operations to get a base-generated scrambled DP θ-marked (e.g., LF lowering) may not respect the same locality constraints that restrict overt A’-movements. In fact, in Korean, it has been repeatedly shown that long-distance scrambling can escape various island domains. Ko et al. (2019) tested both leftward and rightward scrambling in Korean experimentally, and showed that both are insensitive to the adjunct and wh-islands; Kim & Ji (2023) showed that Korean scrambling is insensitive to the Complex-NP island; Kim (2026) showed that Korean scrambling shows no adjunct island effect, but a statistically significant yet non-robust relative clause island effect. These findings largely contrast with the island sensitive nature of other A’-movements in Korean such as relativization and topicalization Han 1992; Na & Huck 1993; Han & Kim 2004; Han 2013; Kim 2019; cf. Kim 2013, and some may consider these findings evidence against analyzing scrambling in Korean as a movement operation.

In this paper, we further examine the nature of Korean long-distance scrambling. Specifically, we present a piece of unexpected finding that goes against the previous generalization that Korean scrambling is island insensitive: we find that the scrambling of wh-DPs in Korean is restricted by temporal clausal adjuncts headed by -ttay (when). This contrasts with previous experimental work by Ko et al. (2019) and Kim (2026) that adopt similar factorial designs to probe for island effects, both of which reported no island effect for the scrambling of non-wh DPs out of clausal adjuncts in Korean. Our finding supports the analysis that the long-distance scrambling of wh-elements involves regular upward syntactic movement in Korean (contra Bošković & Takahashi’s 1998 analysis for Japanese); it also points to the possibility that wh-scrambling and the scrambling of non-wh DPs may be derived via different mechanisms, given the different findings in the current study and Ko et al. (2019) and Kim (2026): non-wh-scrambling may be derived via base-generation of the seemingly displaced elements in the left periphery, whereas this base-generation option is not available for wh-elements.

In this paper, we present results from three experiments. Experiment 1 presents evidence for the adjunct island effect on wh-scrambling in Korean. Experiment 2 shows that the clausal adjuncts tested in Experiment 1 are interpreted as discourse backgrounded. Experiment 3 shows that adjunct islands do not restrict in-situ wh-DPs, in line with previous generalizations that in-situ wh-DPs are not restricted by island effects in Korean (Kim & Goodall 2016; Kim et al. 2023, cf. Cho 2018) and crosslinguistically (Huang 1982; Tsai 1994; 2009; Sprouse et al. 2011; Çakır 2016; 2017; Tian et al. 2022; Jin & Yan 2025, cf. Lu et al. 2020; Chieu 2022; Kim et al. 2023; Chen 2024; Yue & Xiang 2025; Tanaka 2025). Taken together, Experiments 2 and 3 confirm that the results of Experiment 1 reflect a genuine syntactic island effect on movement, as opposed to a discourse-level ban on new-information-seeking questions targeting discourse-backgrounded information (Abeillé et al. 2020; Goldberg 2013; Namboodiripad et al. 2022; Cuneo & Goldberg 2022; 2023; Lu et al. 2025; Ambridge & Goldberg 2008).

2 Experiment 1

2.2 Results and discussion

The mean acceptability rating for each condition is shown in Figure 1. We fit an ordinal regression model predicting the acceptability rating with the sum-coded fixed effects of Structure, Length, and their interaction. Also included were the by-item random intercept and slopes for all the fixed effects, the maximal random-effect structure allowing model convergence. There is a significant main effect of Structure (β = 0.16, SE = 0.079, z = 2.04, p < 0.05) where the island conditions are rated lower than the non-island conditions, and a significant main effect of Length (β = 2.01, SE = 0.020, z = 10.03, p < 0.001) where the short conditions are rated higher than the long conditions. There is also a significant Structure×Length interaction (β = –0.30, SE = 0.069, z = –4.29, p < 0.001), where the length effect is greater for the island conditions than for the non-island conditions. The significant interaction term suggests that there is a temporal adjunct island effect.

Figure 1: Acceptability ratings for Experiment 1.

This result crucially contrasts with the findings of Ko et al. (2019) and Kim (2026), where no adjunct island effect is found for the long-distance scrambling of non-wh DPs. One would be tempted to conclude that the findings of this experiment favors a movement-based analysis of wh-scrambling over a base-generation analysis.² But there is a potential confound: one possible explanation for the difference is that the effect found in the current experiment is in fact not a genuine syntactic island effect on movements, but rather a discourse-level penalty unique to wh-questions. Some suggested that forming wh-questions targeting discourse backgrounded information results in degradedness due to the clash between the fact that wh-questions should be new-information-seeking, and the discourse-old status of backgrounded and presupposed information (Erteschik-Shir 1973; Ambridge & Goldberg 2008; Goldberg 2013; Namboodiripad et al. 2022; Cuneo & Goldberg 2022; 2023; Lu et al. 2025). In the context of the temporal adjunct island effect, a discourse-based explanation goes as follows. Let’s assume, for the moment, that temporal adjunct clauses are interpreted as discourse backgrounded in Korean (we will explicitly verify this assumption in Experiment 2). By virtue of using a temporal adjunct clause in a sentence, the speaker is signaling that the content embedded inside the adjunct clause is discourse backgrounded. At the same time, by virtue of uttering a new-information-seeking wh-question, the speaker signals that the questioned content is foregrounded and at-issue. Under this analysis, wh-questions, regardless of whether there is genuine syntactic movement of the wh-elements, cannot target elements contained inside clausal adjuncts: doing so would give rise to a clash in the discourse-backgrounding function of the adjunct clause and the discourse-foregrounding function of new-information-seeking questions. By contrast, forming questions targeting non-backgrounded constituents (e.g., complement clauses) would not give rise to this clash in discourse function.

To test whether the temporal adjunct island effect we observed reflects a movement constraint or a discourse-level constraint, we further conducted Experiments 2 and 3 to test for the temporal adjunct island sensitivity of in-situ wh-DPs in Korean. Unlike in-situ wh-adverbials that mandatorily undergo covert wh-movement to their scope positions, the quantificational dependency of in-situ wh-DPs can be established through unselective binding without movement, and is thus immune to movement constraints (Pesetsky 1987; Nishigauchi 1990; Tsai 1994; Watanabe 2001).³ If the adjunct island effect observed in Experiment 1 reflects a discourse-level constraint on wh-questions with or without movement, we should expect the same island effect on wh-in-situ. If the temporal adjunct island effect does not constrain wh-in-situ, the island effect in Experiment 1 cannot be reduced to a discourse-level constraint in wh-questions, and instead points to genuine syntactic movement in wh-scrambling.

3 Experiment 2

Under certain discourse-level accounts of island effects (e.g., Erteschik-Shir 1973; Ambridge & Goldberg 2008; Goldberg 2013; Namboodiripad et al. 2022; Cuneo & Goldberg 2022; 2023; Lu et al. 2025), island effects (including the temporal adjunct island effect observed in Experiment 1) could reflect a discourse constraint on question formation that is orthogonal to whether syntactic movement is involved: discourse backgrounded/presupposed content cannot be questioned. Under these accounts, the existence of island effects cannot be taken as evidence for the existence of syntactic movement. Experiments 2 and 3 combined test whether this is a plausible explanation for the island effect observed in Experiment 1. Experiment 2 tests whether the temporal adjuncts used in Experiment 1 are indeed discourse backgrounded, as expected under a discourse account of island effect. Experiment 3 tests whether the temporal adjunct clauses restrict in-situ wh-questions where there is no overt syntactic movement. If Experiment 2 shows that the temporal adjunct clauses are interpreted as discourse backgrounded, but Experiment 3 shows that the temporal adjunct clauses do not restrict wh-in-situ, the results from the two experiments taken together would challenge the discourse account as an explanation for the island effect observed in Experiment 1 (alternative to an explanation where wh-scrambling involves genuine syntactic movement restricted by islands).

3.2 Results and discussion

The negation test responses are shown in Figure 2. We fit an ordinal regression model predicting the responses with the sum-coded fixed effects of Polarity, Structure, and their interaction, as well as the maximal by-item and by-participant random effect structure. There is a significant main effect of Polarity (β = 1.01, SE = 0.14, z = 7.35, p < 0.001) where the not-negated conditions received higher score than the negated conditions. There is a significant main effect of Structure (β = –1.85, SE = 0.21, z = –8.92, p < 0.001), where the complement conditions received lower scores than the adjunct conditions. There is also a significant Polarity×Structure interaction (β = 0.99, SE = 0.14, z = 6.86, p < 0.001), where the Polarity contrast is greater for the complement conditions than for the adjunct conditions. This significant interaction effect shows that the -ttay temporal adjunct clauses used in the stimuli of Experiment 1 are indeed interpreted as discourse backgrounded by the participants.

Figure 2: Negation test responses for Experiment 2.

4 Experiment 3

In Experiment 2, we have shown that the -ttay temporal adjunct clauses in Korean, just like their English counterparts, are interpreted to be discourse backgrounded. In this experiment, we test whether in-situ wh-DPs are sensitive to the -ttay temporal adjunct island in Korean. If the adjunct island effect observed in Experiment 1 reflects a discourse constraint on wh-questions regardless of whether movement is involved, we expect the same island effect to show up for wh-in-situ sentences, given the discourse backgrounded status of adjuncts. If the island effect in Experiment 1 reflects a locality constraint on movement, we do not expect in-situ wh-DPs to show the same effect.

4.2 Results and discussion

The mean acceptability rating for each condition is shown in Figure 3. We fit an ordinal regression model predicting the acceptability rating with the sum-coded fixed effects of Structure, Length, and their interaction. Also included were by-item and by-participant random intercepts and random slopes for the fixed effects and their interaction. There is a significant main effect of Length (β=0.71, SE = 0.12, z = 5.93, p < 0.001) where the short conditions are rated higher than the long conditions. There is no significant effect of Structure (β = –0.072, SE = 0.078, z = –0.93, p = 0.35) or Structure×Length interaction (β = –0.014, SE = 0.080, z = –0.18, p = 0.86).

Figure 3: Acceptability ratings for Experiment 3.

The results suggest that there is no evidence for any temporal adjunct island effect on in-situ wh-DPs, replicating previous findings in the literature (Kim & Goodall 2016). The results also point to a dissociation between discourse backgroundedness and islandhood in Korean: even though clausal adjuncts are discourse backgrounded as shown in Experiment 2, forming wh-questions targeting elements embedded in clausal adjuncts is not necessarily degraded. Therefore, the -ttay temporal adjunct island effect observed in Experiment 1 cannot be reduced to a discourse constraint on question formation, and instead should reflect a genuine syntactic island constraint on movement.

5 General discussion

This squib makes three main contributions to the study of scrambling and island effects in Korean.

First, on the empirical side, we present experimental evidence suggesting that wh-scrambling is sensitive to the -ttay temporal adjunct island in Korean. This contrasts with earlier work demonstrating adjunct island insensitivity for the long-distance scrambling of non-wh DPs (Ko et al. 2019; Kim 2026), as well as with Kim & Goodall’s 2016 and our findings in Experiment 3 that in-situ wh-DPs are also insensitive to the adjunct island. Our finding thus expands the empirical landscape of long-distance scrambling and the adjunct island effect in Korean. Admittedly, we only examined one particular type of clausal adjunct (-ttay clauses), and it is possible that movements may display different sensitivity to different types of clausal adjuncts, as have been demonstrated crosslinguistically (Truswell 2007; Bondevik et al. 2021; Bondevik & Lohndal 2023). We leave it to future studies to further expand the empirical coverage by testing other types of adjunct clauses.

Second, our findings point to the theoretical possibility that wh-scrambling, but not regular scrambling, may be derived through syntactic movement. If wh-scrambling necessarily proceeds via overt A′-movement, the adjunct island effect follows naturally. Meanwhile, the island insensitivity of non-wh scrambling remains consistent with base-generation accounts of scrambling (e.g. Bošković & Takahashi 1998; Fanselow 2001) in which scrambled DPs have the option to be directly merged at their surface positions. An open question, however, is what bars wh-elements from being base-generated at its surface scrambled position. We do not intend to offer a solution in this squib, and will leave this as an important question for future work.

Apart from the aforementioned explanation based on the different derivation paths for wh- and non-wh-scrambling, there are a few alternative explanations for why our findings diverge from previous experiments on non-wh-scrambling in Korean. In Ko et al. (2019), the authors gave two speculative explanations for why they didn’t find island effects on non-wh-scrambling. First, the authors speculate that Korean allows multiple specifiers, and the scrambled phrases in Korean can escape islands by merging as multiple specifiers of C. This explanation is implausible: the availability of multiple specifier positions might account for the lack of wh-island effects in Korean (which is also reported in Ko et al. 2019), but not clausal adjunct island effects, which cannot be straightforwardly attributed to the lack of a CP specifier position at the island edge. A second explanation provided in Ko et al. (2019) is that their experiment might be confounded by a floor effect, which masks an island effect by compressing the acceptability contrasts near the bottom of the scale. However, Kim (2026) replicates Ko et al.’s 2019 lack of adjunct island effect, and the average acceptability Z-scores for all conditions in Kim (2026) are positive and thus not confounded by any floor effect. Finally, a third possible explanation is that the short dependency length conditions in our Experiment 1 are different from the short conditions in Ko et al. (2019), which might have led to different experimental results. Our stimuli involve subject wh-elements in the short conditions and object wh-elements in the long condition, whereas Ko et al.’s stimuli involve object wh-elements in all conditions. Ko et al. observed a pronounced acceptability contrast between their short.island and short.non-island conditions, and this contrast drives a statistically significant sub-additive interaction effect between length and structure, in the opposite direction of an island effect. However, in our Experiment 1, we did not observe any pronounced contrast between the short conditions. One might speculate that the difference in findings of the two studies is driven by the difference in the short conditions. An anonymous reviewer suggests that the contrast between the two studies might reflect a subject-object asymmetry in scrambling in Korean. However, any such asymmetry should be reflected as a main effect in dependency length in Experiment 1, and should not affect the super-additive island effect. Alternatively, there could be a penalty that selectively affects object scrambling from the matrix when an adjunct clause is present. Since both short conditions in Ko et al. (2019) involve the extraction of the indirect object DP from a position that linearly follows the embedded (complement or adjunct) clause, the finding in Ko et al. (2019) could reflect a processing-level penalty that burdens the processing of dependencies linearly spanning adjunct clauses but not ones that linearly span complement clauses.⁴ Nevertheless, one should note that Kim (2026) used stimuli similar to the ones in Ko et al. (2019) where all scrambled elements are objects, but there is no contrast between the two short conditions in Kim (2026). We leave it to future studies to test whether this contrast in the short conditions reported in Ko et al. (2019) is a reliably replicable effect.

A third contribution of this study is that the results of Experiments 2 and 3 inform the broader debates about the status of discourse-based accounts of island effects. Experiment 2 confirmed that the -ttay adjuncts in Korean are indeed interpreted as discourse backgrounded, just like their English counterparts. However, Experiment 3 showed that wh-in-situ questions can target elements inside these -ttay clauses without any island effect. This means that the degradation observed in Experiment 1 cannot be reduced to a pragmatic clash between the new-information-seeking nature of wh-questions and the backgrounded status of clausal adjuncts. On a higher level, given the dissociation between discourse backgroundedness and islandhood we observed, the findings of Experiments 2 and 3 constitute a counter-example to the discourse-based accounts that attribute island effects to the clash between the discourse foregrounding/new-information-seeking function of wh-questions, and the discourse backgrounded status of certain structural domains (Ambridge & Goldberg 2008; Goldberg 2013; Cuneo & Goldberg 2022; Namboodiripad et al. 2022). Nevertheless, this does not mean that all discourse-based accounts of islands are untenable. One recent proposal by Winckel et al. (2025) states that only wh-elements that undergo movement are foregrounded. Our findings are still compatible with this particular account: the island effect in Exp.1 arises because the wh-scrambling construction involves movement of the wh-element, whereas the in-situ argument wh-questions in Exp.3 do not involve (overt or covert) movement and thus do not give rise to any discourse clash. Note that under this account, we can still conclude that wh-scrambling in Korean involves movement: if the sentence-initial wh-elements are base-generated, Winckel et al.’s (2025) account would predict no island effect since no movement is involved, contrary to our finding.

In sum, this squib presents a data point that wh-scrambling, unlike non-wh-scrambling or wh-in-situ, is sensitive to the adjunct island constraint in Korean. This result is compatible with a movement-based analysis of wh-scrambling, and invites a reevaluation of the assumption that Korean scrambling constitutes a single unified operation.

Data availability statement

All stimuli, data files, and analysis scripts can be accessed at https://osf.io/bfwtu/

Ethics and consent

The experiments included in this study have been approved by the Sungkyunkwan University Institutional Review Board, approval number 2021-04-009, and the Northwestern University Institutional Review Board, approval number STU00224884.

Funding information

This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2025S1A5A8008853).

Acknowledgements

The authors are grateful to the three anonymous reviewers at Glossa, Hee Joong Choi, Dongwoo Park, and Nozomi Tanaka for their valuable comments at various stages of this project. All remaining errors are ours.

Competing interests

The authors have no competing interests to declare.