Subjacency effects on overt wh -movement in wh -in-situ languages: Evidence for nominal structure

 Abstract. This paper investigates whether overt wh -movement in Korean, a wh -in-situ language, triggers Subjacency violations in the same set of bounding configurations as English. Yoon (2013) and Jung (2015) showed that Korean wh islands display Subjacency effects, and we ask whether the Complex NP and Coordinate Structure Constraints are also observed. We find that they are not. We propose that this is because Korean nominal expressions need not project DP. Our analysis supports previous accounts of Korean nominal structure (Kim et al. 2010), which suggest (based on optionality of determiners) that “Korean NP structure is non-configurational or lacks the category

wh-islands. In (1a), the movement of who crosses a single IP node in each movement and is acceptable. In each of (1b) and (1c), however, two bounding nodes are crossed in a single step. In (1b), who crosses the DP node dominating books and the matrix IP node in a single step. In (1c), who crosses both IP nodes in a single step, since the lower specifier of CP is already occupied by why. Both (1b) and (1c)  Potentially complicating Subjacency violations and making them worse is the ECP, which demands that traces be properly governed. Proper government of a trace can be satisfied either by its being the semantic argument of a verb (i.e., theta-governed) or by its being anteceded by another co-indexed element that is close enough (i.e., antecedent governed). The contrast between (1b) and (2b) illustrates.
(2) a. In (1b), movement of who across the DP node dominating books and the matrix IP node in a single step violates Subjacency, but the trace of who as object of the verb criticized allows it to be properly (theta) governed by the verb. In (2b), movement of how severely across the DP node dominating the actions and the matrix IP node in a single step also violates Subjacency, but the trace of how severely is not theta-governed by the verb criticized and cannot be antecedent governed by the moved wh-element. It thus also violates the ECP, making (2b) much worse than (1b).
3. Subjacency vs. the ECP in Chinese and Japanese. We can now turn to languages (Chinese and Japanese) that ordinarily have covert wh-movement (i.e., movement at LF). Because ECP violations are incurred at LF and Subjacency violations are triggered at Spell-out, these languages usually show effects of the former and not the latter.
In arguing for LF as a syntactic level of representation, Huang (1982Huang ( , 1995 demonstrates that (i) Chinese wh-expressions undergo movement at LF and are subject to ECP constraints at that level of structure, and (ii) LF movement does not trigger Subjacency violations. Example (3) illustrates a case in which an embedded in-situ wh-element can have perfectly natural main clause question interpretation in a structure that would normally trigger a Subjacency violation in English.
( The structures shown in (5ʹ) illustrate why this is so. In (5aʹ), the trace of shenme is properly (theta) governed by the verb mai 'buy', and the interpretation is available (since Subjacency does not apply). In (5bʹ) the trace of weishenme is not properly governed, since it is not theta governed by the verb and it cannot be antecedent governed by its coindexed antecedent. As regards ECP violations in Japanese, Lasnik & Saito (1992) show that LF movement in Japanese shows the same argument-adjunct asymmetry that we observed for Chinese. In (6), we see two sentences with embedded wh-elements, nani 'what' in (6a) and naze 'why' in (6b). Only the first of these can be interpreted as a main clause wh-operator. In (6a), after LF movement to the matrix Spec,CP, the trace of nani is properly (theta) governed, while in (6b) the trace of naze is not properly governed and violates the ECP.

Overt scrambling vs. wh-movement in Japanese and Korean.
Having illustrated the differential effects of Subjacency and the ECP for wh-in-situ constructions, we turn to the matter of Subjacency effects in Japanese and Korean when embedded wh-elements are overtly moved to the beginning of a matrix clause. In some instances, their movement may involve focus-triggered scrambling while in others it may be pure wh-movement. Saito (1989) provides examples of overt scrambling of wh-expressions in Japanese, shown here in (7). In (7), only the embedded clause is marked as an interrogative, and nani 'what' is interpreted as being within the scope of ka in both (7a) and (7b). Saito assumes that nani in (7b) is scrambled and then reconstructed to its original embedded position at LF.

knows 'John knows what Mary bought.'
Example (8) presents a different, and more complex, problem. In (8), both the matrix and embedded clauses are interrogative, with ka signaling [+Q] on the lower clause and no signaling [+Q] on the matrix clause. Here, either the matrix or the embedded clause can be interpreted as a yes-no question. For the interpretation in (a), it is assumed that nani moves to Spec,CP of the embedded clause and remains within the scope of ka. For interpretation (b), nani would move to the matrix CP and check the Q-feature of no, leaving the embedded ka to signal a yes-no question. Note that in order for (b) to be available, Subjacency must not apply at LF (Takahashi 1993 It is a fact, though, that (8) tends to be more readily interpreted as a main clause yes-no question (i.e., interpretation (a)), and some Japanese linguists (Nishigauchi 1990, Watanabe 1992) have therefore suggested that Subjacency might apply at LF in Japanese. We take no position on this.
Comparing the two structures in (7) and (8), involving respectively (i) overt movement of wh-expressions and (ii) matrix and embedded interrogatives in the same sentence, we turn to what happens when these two factors are combined in Korean. Here, in example (9), we see a wh-phrase nwukwu 'who' overtly moved to the beginning of the clause. Alongside this, both the matrix and embedded clauses are interrogatives, marked with -ci and -ni, respectively.
(9) Nwukwu-lul1 ne-nun [ku-ka partner-lo t1 senthaykha-l-ci] al-ko sip-ni? who-ACC you-TOP he-NOM partner-as choose-FUT-Q wants.to.know-Q (a) 'Do you want to know who he will choose as his partner?' (b) ??'Who1 do you want to know whether he will choose t1 as his partner?' The two interpretations shown here are dependent on whether nwukwu is scrambled into the main clause as in (a), leaving the main clause as a yes-no interrogative and nwukwu to be interpreted within the scope of ci, or whether it is wh-moved into the matrix clause as in (b), leaving the embedded interrogative to get a yes-no question interpretation. 2 The interpretation of (9) has been controversial. While Lee (1993) claims that both interpretations of (9) are available, subsequent research has disputed this (inconclusively). In Park (2010), it is claimed that only (b) is available, while Kim (1998) and Shin (2005) assert that (9) can only be interpreted as in (a). More recent work by Yoon (2013) and Jung (2015) has shed some additional light on the question. In Yoon (2013), an experimental reading test found that native Korean speakers preferred interpretation (a) over (b) by a factor of about 10 to 1 (91.2% to 8.8%). 3 This preponderant preference for scrambling interpretation (a) over wh-movement interpretation (b), led Yoon (and Jung) to tie the results to Subjacency effects, and to suggest that the interpretation of overt wh-movement in Korean is restricted (unsurprisingly) by Subjacency.

CNPC and CSC effects in Korean.
Having seen here above that overt wh-movement can respect Subjacency in Korean, at least as regards wh-islands, we are led to ask whether other Subjacency constraints, such as the Complex NP Constraint (CNPC) and Coordinate Structure Constraint (CSC), are also observed. We find they are not. Examples (10) and (11)   It is important to note that a matrix clause interpretation for the embedded wh-expression nwukwu 'who' is not more available when it is left in situ as when it is overtly moved to the matrix clause.
(i) ne-nun [ku-ka partner-lo nwukwu-lul senthaykha-l-ci] al-ko sip-ni? you-TOP he-NOM partner-as who-ACC choose-FUT-Q wants.to.know-Q (a) 'Do you want to know who1 he will choose t1 as his partner?' (b) ??'Who1 do you want to know whether he will choose t1 as his partner?' 3 A listening test showed that subjects are quite sensitive to sentence intonation. When the target sentence was asked with wh-question intonation, Korean speakers showed preference for a wh-question interpretation (60%) over a yesno question interpretation (37.5%). However, when the target sentence was presented with yes-no intonation, 82.5% of participants interpreted it as a yes-no question, but 17.9% interpreted it as a wh-question.

Hana-ka [banana-wa t1 ] mek-ess-ni? what-ACC Hana-NOM banana-and eat-PST-Q (What1 did Hana eat [a banana and t1]?)
In (10) and (11), we see that overt movement of the wh-element mwues 'what' is perfectly acceptable out of a complex NP in (10b) and out of either the first or the second position within the conjoined noun phrase in (11b) and (11d). These sentences do not display acceptability deficits that would ordinarily be expected of Subjacency violations. 4 It is important to point out here that examples (10) and (11) all involve bare, non-discourselinked (D-linked) wh-expressions. As originally explained in Pesetsky (1987), D-linked interrogative expressions, such as which banana, are distinguished from non-D-linked interrogatives, such as what, in that they (the former) imply "the existence of a context set of familiar entities of the type denoted by the nominal" (e.g., a set of already mentioned bananas). 5 This is important to our observations because D-linked wh-expressions can readily violate locality constraints. That the extraction of bare wh-expressions is grammatical in (10) and (11), in contrast with (9), is quite telling, since if Subjacency could block any objects from extraction, it should be these.
It needs be acknowledged here that left-dislocated, bare wh-expressions in wh-in-situ languages (such as Japanese and Korean) may in fact be D-linked. Following Pesetsky (1987), which asserted a categorical D-linking difference between bare and complex wh-expressions, research on Japanese and Korean wh-movement has argued for left-dislocation of whexpressions in those languages to involve scrambling and for scrambling to involve D-linkage. Thus, according to Miyagawa (2006), scrambled wh-expressions can at some point be D-linked (Miyagawa 2006, Yoon 2013. This is illustrated in Miyagawa (2006), wherein it is suggested for pair-list questions, such as (12), that a scrambled wh-expression induces D-linkage.  Lee (1993), scrambling in Korean can sometime be restricted due to an anti-ambiguity strategy, which functions as a discourse constraint (see Kuno 1980). The examples in (i), structurally identical to (11) (11), which involves subjects and objects of the verb 'eat', the subject and object of 'like' can all be people as in (i), leaving the semantic role of the scrambled elements ambiguous. In these cases, scrambling is confounded by language processing difficulties. 5 http://www.glottopedia.org/index.php/Discourse-linked_interrogative_phrase Regarding (12), Miyagawa states that the scrambled wh-word nani 'what' presupposes that there are a specific set of objects which some people bought. Thus, this question is distinct from its unmoved counterpart who bought what? by adding additional discourse information.
While Miyagawa (and others) are likely correct in their analysis of scrambling as involving D-linkage, the contrast in the effects of Subjacency on wh-islands as in (9) and on CNPC/CSC islands as in (10) and (11) remains and requires an explanation.
6. Explaining wh-island vs. CNPC/CSC island contrasts: Optional projection of DP. In order to adequately understand the difference between wh and CNPC/CSC islandhood, it is important to realize that the bounding node relevant to the latter is DP, rather than NP. This is shown quite clearly in Davies & Dubinsky (2003) (henceforth D&D 2003, wherein subject islands are proven to involve DP projections that are present in English even when the subject is not nominal (i.e., not an NP).
Noting the distribution of emphatic reflexive pronouns, among several other tests, D&D (2003) demonstrate that all subjects involve a DP projection in English. This is illustrated in (13), where only the clausal subject of made in (13a) and not the clausal object of hate in (13b) can host an emphatic reflexive. Having shown that clausal subjects (but not objects) project DP, D&D (2003) go on to demonstrate that the DP-hood of these clausal subjects induces Subjacency violations, as in (15a), in contrast with (15b). Returning to consider the differences between (9) on the one hand and (10)/(11) on the other which might lead to such a contrast, we propose that it is due to Korean, unlike English, not needing to project a DP node for its nominal expressions. In (9), wh-movement crosses two IP (bounding) nodes in one movement and violates Subjacency. There, the wh-object nwukwu 'who' in the embedded clause moves to the matrix clause, crossing the wh-island induced by the interrogative affix -ci in the embedded clause.
In the potential CNPC and CSC case, however, Subjacency violations would not arise if saram 'person' in (10) and mwues-kwa banana 'what and a banana' in (11) are only dominated by NP, and no DP nodes are involved. If the structures in (10b) and (11b) are as shown in (10bʹ) and (11bʹ), rather than as in (10bʹʹ) (Kim et al. 2010) which have proposed (based on the optionality of determiners) "that the Korean NP structure is non-configurational or lacks the category D." It is thus the particular structure of Korean nominal expressions that allows for the differences between wh and CNPC/CSC islandhood to arise.

7.
Optional DP and Subjacency in other languages: Bulgarian. Having attributed Subjacency differences between Korean and English as being due to differences in nominal structure, specifically in the projection of DP, we would note that Korean is not alone in this. Other languages are also found to optionally project DPs in contexts where English demands it, and when they do not, Subjacency effects go away. Bulgarian is one of these. Having shown above that extraction from clausal subjects in English is sensitive to Subjacency and triggered by these clausal subjects projecting DP, we can compare analogous constructions in Bulgarian. Here below, we see for English that wh-extraction from an extraposed clausal subject in (16a) is possible, but not out of an in situ clausal subject in (16b). When analogous Bulgarian constructions are compared, we find wh-extraction out of clausal subjects to be perfectly acceptable, as in (17c) In D&D (2003), these facts along with other supporting evidence are marshalled to show that Bulgarian clausal subjects, unlike those in English, do not project DP. Thus, we can conclude that Korean, like Bulgarian, is another language in which the divergent distribution of DP projections is found to interact with locality constraints on overt movement.

Conclusion.
In this paper, we have confirmed that Subjacency and the ECP are observed in different conditions. That is, Subjacency applies at Spell-out and the ECP applies at LF. In Korean, regarding constraints on wh-movement, the ECP applies to all wh-constructions including those with covert LF wh-movement, whereas Subjacency can only apply to overtly moved wh-elements (similarly to overt wh-movement in Japanese). However, we also found that overt wh-movement (i.e., leftward scrambling of wh-elements) is not blocked out of nominal islands, and that the Complex NP Constraint (CNPC) and Coordinate Structure Constraint (CSC) do not apply to these structures. Our explanation for this apparent incongruity in the application of Subjacency is that Korean nominal expressions need not project DP and thereby need not constitute bounding domains.