Correlations between island (in)sensitivity and base positions of (non-)standard wh-in-situ

It has been well-accepted in the literature that the island (in)sensitivity of wh-in-situ falls under the so-called Noun versus Adverb Generalization (NAG), which states that an in-situ wh-phrase is island-free iff it is (or contains) a whnominal (Tsai 1994a,b; Stepanov & Tsai 2008; Fujii et al. 2014). However, we show that the NAG is not sufficient to explain the island behaviors of some (non)standard in-situ wh-phrases in Korean. Alternatively, we suggest that the island (in)sensitivity of in-situ wh-phrases may correlate not with their categorial status but with their base-generated positions: specifically, we assume that an in-situ whphrase that is base-generated in the CP domain (Spec-CP) is island-sensitive, while an in-situ wh-phrase that is base-generated below CP/TP is island-insensitive.

( In both the Chinese and Korean examples, the wh-arguments-shenme in (1a) and mwe-l in (2a)-can occur inside the complex NP island, whereas the reason wh-adjuncts-weishenme in (1b) and way in (2b)-cannot. 1 To account for such asymmetries in island effects, many researchers (e.g., Tsai 1994a,b;Stepanov & Tsai 2008;Fujii et al. 2014) have argued, under an unselective-binding approach, that the island behaviors of wh-in-situ fall under the morphological generalization in (3), which, following Fujii et al. (2014), we refer to as the Noun versus Adverb Generalization. 2

(3) Noun versus Adverb Generalization (NAG):
An in-situ wh-phrase is island-insensitive iff it is (or contains) a wh-nominal. (Fujii et al. 2014: (3)) The NAG gives a straightforward account of the contrasts in (1) and (2): the wh-arguments shenme and mwe-l are island-insensitive since they are wh-nominals, whereas the reason whadjuncts weishenme and way are island-sensitive since they are pure wh-adverbs.
2. Counterexamples to the NAG. Here we show that island (in)sensitivity of in-situ wh-phrases in Korean is not properly constrained by the NAG. Before we proceed, however, we would like to briefly introduce two different but similar types of non-standard wh-question in Korean, whose island behaviors, as we will see, play a key role in testing whether the NAG is empirically correct or not. Consider (4) and (5).
(4) a. Mimi-nun Mimi-TOP The wh-questions like (4a) and (5a) are taken to be non-standard wh-questions in the sense that a non-'why' wh-phrase receives a 'why'-like interpretation. In (4a), the wh-phrase mwe-l, which canonically functions as an argument in a sentence as in (4b), behaves like a wh-adjunct corresponding to way 'why'. In a similar manner, in (5a), the wh-phrase ettehkey, which is standardly used as a manner or instrumental adjunct as in (5b), is interpreted as a reason whadverbial when it occurs with a causative -key toy-construction with past tense. 3 As indicated in (4a) and (5a), throughout the paper, the superscript w is used in glossing the non-standard wh-items (i.e., reason mwe-l and reason ettehkey) and distinguishing them from their standard counterparts (wh-argument mwe-l and manner/instrumental ettehkey). Now let us examine whether or not the NAG correctly predicts island (in)sensitivity of introduce a variable subject to unselective binding and, therefore, relies on local binding with subsequent covert movement to its scope position, in which case it induces island effects when it occurs inside an island, as illustrated in (ib). See Tsai (1994a) and Stepanov & Tsai (2008)  (non-)standard wh-in-situ in Korean. If the NAG is on the right track, then it is expected that manner/instrumental ettehkey is island-sensitive, since it is a wh-adverb; however, as shown in (6), it is island-insensitive in that it can occur inside a complex NP island or an adjunct island (Chung 2000(Chung , 2005 In dealing with the island insensitivity of manner/instrumental ettehkey, Chung (2005) offers an interesting proposal under an unselective-binding approach. On his view, manner/instrumental ettehkey can be decomposed into four sub-parts, [ DP/NP e-tte]-h-key 'Det-CNP-do-adverbializer': the first part e is a determiner which combines with the common noun phrase tte that follows it, the third part h(a) is a transitive verb that takes the preceding DP/NP, and the final part key is an adverbializer. Based on this morphological structure, Chung explains that the insensitivity of manner/instrumental ettehkey to strong islands is because it contains the nominal element tte that introduces a variable subject to unselective binding: that is, its island behavior is captured by the NAG. If Chung's nominal analysis is on the right track, then it is expected that reason ettehkey is island-insensitive, since its morphological form is the same as its standard counterpart; however, that is not the case, as evidenced by the ungrammatical examples in (7) where reason ettehkey cannot occur inside a complex NP island and an adjunct island, just like way.  Contra Chung's proposal, although we treat both manner/instrumental and reason ettehkey as pure wh-adverbs, the NAG still has difficulty accounting for the asymmetry between the two variants of ettehkey in island contexts: since the two variants are wh-adverbs, they both must be island-sensitive according to the NAG, contrary to fact.
The NAG may also have difficulty explaining the asymmetry between reason mwe-l and its standard counterpart (i.e., wh-argument) in island effects. Under the NAG, reason mwe-l is predicted to be island-insensitive, since it is a wh-nominal; however, the prediction is not borne out, as seen in (8) where reason mwe-l is sensitive to a complex NP island, as in (8a), and an adjunct island, as in (8b) 3. Correlations between island (in)sensitivity and base positions of (non-)standard wh-in-situ.
In the previous section we have demonstrated that the NAG fails to capture the island (in)sensitivity of some (non-)standard in-situ wh-phrases in Korean. Here, as an attempt to provide a (potential) alternative view to the NAG, we suggest that island (in)sensitivity of in-situ wh-phrases may correlate with their base-generated positions, not with their categorial status (i.e., noun versus adverb), by showing that an in-situ wh-phrase base-generated in the CP domain (Spec-CP) is island-sensitive, while an in-situ wh-phrase base-generated below CP/TP is island-insensitive.
3.1. TWO DIAGNOSTIC TESTS FOR IDENTIFYING wh-PHRASES BASE-GENERATED IN THE CP DOMAIN. Korean is well-known as exhibiting an asymmetry between way 'why' and other whoperators with respect to the Intervention Effect: unlike the latter, the former does not exhibit the Intervention Effect when c-commanded by a Scope Bearing Element (SBE) like amwuto 'anyone' or man 'only' (Beck & Kim 1997;Beck 2006;Ko 2005Ko , 2006. 4 Consider (9). ahn-ass-ni? not-PST-QUE 'Why did no one read the paper?' As illustrated here, unlike the wh-argument mwe-l and the manner/instrumental ettehkey, the wh-adjunct way can follow the SBE amwuto.
To account for the peculiar behavior of way in terms of the Intervention Effect, i.e., its ability to be preceded by an SBE, Ko (2005) assumes that way in an interrogative clause is externally merged in its checking position, Spec-CP (CP-Modifier Hypothesis), while other whphrases undergo LF movement to Spec-CP for feature checking. This is illustrated in (10). In addition, she proposes the Intervention Effect Constraint, where at LF a wh-phrase cannot move across an SBE to its checking (scope) position, as illustrated in (11)  On Ko's analysis, the ungrammaticality of (9a) and (9b) is simply because the SBE amwuto blocks LF movement of the given wh-phrase to Spec-CP, as illustrated in (12) (9b)) Meantime, the well-formedness of (9c) is because the wh-adjunct way is licensed in its base position (i.e., Spec-CP) and, therefore, does not move across the c-commanding SBE which has undergone overt scrambling over the wh-phrase: As noted by Ko (2005), the external merge of way at Spec-CP is supported by the fact that it always takes wide scope over negation in an interrogative clause: Since the reason wh-adjunct way is base-generated in the CP domain, it is impossible for it to be interpreted under the negation in IP.
Adopting Ko's ideas, we assume that if a wh-phrase (i) does not show the Intervention Effect when c-commanded by an SBE in an interrogative clause and (ii) takes obligatory wide scope over negation, then it is taken to be base-generated in the CP domain (Spec-CP).

EXTERNAL MERGE OF NON-STANDARD wh-IN-SITU IN THE CP DOMAIN.
Reason mwe-l is assumed to originate in the CP domain given that, like way, it can be preceded by an SBE, as in (15a), and takes wide scope over negation in an interrogative clause, as in (15b) Meantime, manner/instrumental ettehkey is assumed to be base-generated below NegP (in IP) in that it is subject to the Negative Island Effect, where negation blocks extraction of certain (wh-)phrases (Rizzi 1990;Shlonsky & Soare 2011). Consider the contrast in (17). As observed in (17a), the reason wh-adjunct way is not sensitive to negation in the clause with which it is construed. This can be explained by Ko's (2005) analysis that assumes that way is base-generated in the CP domain (above NegP). On the other hand, the ungrammaticality of (17b) can be accounted for by assuming that manner/instrumental ettehkey originates in a structurally lower position below negation and that its LF movement to Spec-CP is blocked by the negator, inducing the Negative Island Effect. Notice that, as can be seen in (16), reason ettehkey is exempt from the Negative Island Effect, just like way, which supports the claim that reason ettehkey is base-generated in the CP domain.

ISLAND (IN)SENSITIVITY AND BASE POSITIONS OF OTHER IN-SITU wh-PHRASES.
The other in-situ wh-phrases that we have not examined so far, such as nwukwu 'who' and 'low' wh-adjuncts like encey 'when' and eti 'where', are all island-insensitive. For instance, examples in (18)  All these island-insensitive wh-phrases presented here are known as being base-generated below CP/TP. This is consistent with the proposed view that an in-situ wh-phrase is island-sensitive iff it is base-generated in the CP domain.
4. Summary. The correlations between island (in)sensitivity and base positions of (non-)standard in-situ wh-phrases in Korean we have examined so far are summarized in Table 1 Table 1. Correlations between island (in)sensitivity and base positions of (non-)standard in-situ wh-phrases in Korean The table shows that an in-situ wh-phrase base-generated below CP/TP is island-insensitive, whereas an in-situ wh-phrase base-generated in the CP domain (Spec-CP) is island-sensitive, regardless of whether the given wh-phrase is a wh-nominal (e.g., reason mwe-l) or a wh-adverb (e.g., manner/instrumental ettehkey), a standard wh-phrase or a non-standard wh-phrase. 5 5. How about other wh-in-situ languages?. Our preliminary literature review, which is summarized in Table 2, indicates that the proposed correlation between island (in)sensitivity and base positions of wh-in-situ may also hold for Chinese and Japanese.
lg. wh-phrases island (in)sensitivity base positions references Chinese reason weishenme 'why' sensitive in the CP domain Lin 1992, Ko 2005 purpose wei(-le) shenme 'for what' insensitive below CP/TP Stepanov & Tsai 2008 causal zenme 'how' sensitive in the CP domain , Jin 2016 manner zenme 'how' insensitive below CP/TP Jin 2016, Murphy 2017 other whs ('who ', 'what', 'when', 'where') insensitive below CP/TP Japanese naze 'why' sensitive in the CP domain Ko 2005, (Ochi 2014 donna riyuu-des 'for what reason' insensitive below CP/TP Ko 2005, Fujii et al. 2014 doo (yatte) 'how' insensitive below CP/TP Fujii & Takita 2007 other whs ('who ', 'what', 'when', 'where') insensitive below CP/TP Table 2. Correlations between island (in)sensitivity and base positions of wh-in-situ in Chinese and Japanese 6. Concluding remarks and further work. In this paper we have demonstrated that the NAGthe well-accepted generalization in accounting for island (in)sensitivity of wh-in-situ-is not enough to capture the island behaviors of some (non-)standard in-situ wh-phrases in Korean. Alternatively, we have suggested that the island (in)sensitivity of (non-)standard in-situ whphrases may correlate with their base positions, by showing that an in-situ wh-phrase is islandsensitive iff it is base-generated in the CP domain. However, there remain many important issues to be addressed to confirm the crosslinguistic validity of the generalization and to explain the resulting pattern. We need to do further work to see whether the proposed correlation is applicable over a wide range of wh-in-situ languages and to provide theoretical support. Although we leave the crucial issues unresolved here, we hope the (potential) proposed correlation between island (in)sensitivity and base positions gives researchers new insights into understanding various asymmetries in island effects in wh-in-situ languages.