The wh-element why in Chinese, weishenme, has two distinct interpretations (Lin, 1992; Tsai, 1994, 1999, 2008; Stepanov and Tsai, 2008; Fujii et al., 2014): “reason weishenme” (henceforth weishenme_R), which typically yields yinwei “because”-answers, as in (1a); and “purpose weishenme” (henceforth weishenme_P), which typically triggers weile “in order to”-answers, as in (1b).

It has long been assumed that the two interpretations of weishenme “why” in Chinese have different internal structures which may contribute to their island-(in)sensitivity (cf. Tsai, 1994): weishenme_R is regarded as simply a wh-adverb without any internal syntactic structure, as in (2a); whereas wei (le) shenme_P is considered a wh-PP consisting of the preposition wei “for” and its nominal part shenme “what,” where the suffix -le can be inserted between (e.g., Tsai, 1994; Stepanov and Tsai, 2008), as in (2b).

In addition to the difference in their internal structures, the two interpretations of weishenme are also claimed to differ in island sensitivity (Lin, 1992; Tsai, 1994; Stepanov and Tsai, 2008; Fujii et al., 2014). Consider examples (3–4).

Tsai (1994): 128 (10–11)

As shown in examples (3) and (4), weishenme_P is licensed inside island structures (i.e., insensitive to islands), whereas weishenme_R is not (i.e., sensitive to islands). There are two main classes of accounts for this asymmetry. ¹ Some attribute it to the Empty Category Principle (ECP) while assuming that covert (LF) movements are not restricted by subjacency (Chomsky, 1981; Stowell, 1981; Huang, 1982). Others claim that in-situ wh-DPs, like the shenme “what” in (2b), do not need to undergo covert movement, and can be licensed in-situ via Unselective Binding (UB; Baker, 1970; Pesetsky, 1987; Nishigauchi, 1990; Aoun and Li, 1993; Tsai, 1994, 1999; Reinhart, 1998; Stepanov and Tsai, 2008). Both classes of accounts predict that wh-DPs are not sensitive to islands, while wh-adverbs are. Given the structural assumption in (2), wei(le)shenme_P should be island insensitive because it is a PP containing a wh-DP, while weishenme_R should be island sensitive because it is a wh-adverb.

This asymmetry in island sensitivity between in-situ wh-DPs and wh-adverbs mentioned above has been established mostly based on informal judgments by syntacticians. However, recent experimental studies on wh-in-situ languages put this generalization into question: Kim and Goodall (2016) found that Korean wh-DPs are sensitive to wh-islands; Omaki et al. (2020) found that Japanese in-situ wh-adverbs are equally insensitive to subject islands as in-situ wh-DPs; Lu et al. (2020) found that Chinese wh-DPs and wh-adverbs are both sensitive to relative clause islands. ² Crucially, Lu et al. (2020) pointed out that the acceptability contrasts between wh-DPs and wh-adverbs inside island structures (e.g., the contrast between (3/4a) and (3/4b)) was due to a penalty of long distance covert movement of wh-adverbs, which has nothing to do with island sensitivity. Similar bans on embedded in-situ wh-adverbs have also been proposed by Heycock (2006) and Jin (2016). If the island sensitivity asymmetry between wh-DP and wh-adverbs are indeed non-existent in Chinese, we should expect the two interpretations of weishenme to be equally sensitive (or insensitive) to island constraints. The current study thus follows Lu et al.’s (2020) general experimental design, and tests experimentally whether the two interpretations of weishenme indeed differ in island sensitivity.

Island violation arises when there are two factors present in a sentence: an island construction (e.g., a relative clause, a clausal adjunct, etc.), and a syntactic dependency that crosses the boundary of the island construction. ³ Given that island constructions and long distance dependencies might independently contribute to acceptability degradation, island violation should be detected as the superadditive effect of having both factors present at the same time (Sprouse et al., 2012; Sprouse and Hornstein, 2013; Sprouse and Villata, 2021). Following this argument, Sprouse et al. (2012) among others suggested that island effect can be probed experimentally using formal acceptability judgment tasks with a 2 × 2 factorial design, manipulating the structure of the embedded clause (island or non-island) and dependency distance (long or short, where long represents that the movement crosses the embedded clause boundary, and short represents that the movement does not cross the embedded clause boundary). An example set of stimuli is shown below in (5). Island effect would be detected as an interaction of embedded structure and dependency distance (i.e., the contrast between (5a) and (5b) is larger than the contrast between (5c) and (5d)).

This design can also be used to probe the island sensitivity of in-situ wh-elements where there is no overt movement (Sprouse et al., 2011; Kim and Goodall, 2016; Lu et al., 2020). In such cases, the dependency distance factor represents whether the dependency between the wh-in-situ and its scope position crosses the embedded clause boundary or not. Similar stimuli sentences as (5) could be used in such a study, except that the gap positions in (5) would be occupied by in-situ wh-elements.

Note that in this paradigm, the existence of an island effect does not depend on the absolute acceptability rating of the island/long extraction condition [example (5a)]. It is possible that sentences like (5a) are rated as acceptable and receive no asterisk in introspective judgments, yet an acceptability judgment experiment might still detect a significant interaction of embedded structure and extraction distance, suggesting the existence of an island effect. ⁴

In this study, we will use the same factorial design to probe the island sensitivity of weishenme_R (the Reason why) and weishenme_P (the Purpose why) in Chinese.

Contrary to our hypothesis, we found significant interactions of Structure and Distance for all three wh-types tested, suggesting that all three types of wh-elements (weishenme_R, weishenme_P, and weileshenme_P) are sensitive to the relative clause island. This challenges previous empirical generalizations that weishenme_P and weileshenme_P are not island sensitive. Note that despite the significant interactions of Structure and Distance, the long extraction/RC structure conditions may not be perceived as unacceptable due to their middle-of-the-scale absolute ratings. The long/RC/weishenme_R condition received an acceptability z-score of-0.054 (SE = 0.12), the long/RC/weishenme_P condition received an acceptability z-score of-0.059 (SE = 0.099), and the long/RC/weileshenme_P condition received an acceptability z-score of-0.10 (SE = 0.095). They were all rated higher than the unnatural fillers, which received an acceptability z-score of-0.91 (SE = 0.022). The high ratings for the long/RC conditions may have contributed to the misguided empirical generalizations in the past literature that were built upon introspective judgments.

Furthermore, no three-way interaction of Structure, Distance, and Wh-type is found when comparing the weishenme_P condition with the other two wh-types. There are two conclusions we can draw from this finding. First, weishenme_P and weileshenme_P behave similarly with regard to island sensitivity, suggesting that our sanity check using weileshenme_P yielded expected results. Second, weishenme_P and weishenme_R are equally restricted by the relative clause island, contrary to previous generalizations. Admittedly, we are arguing from the lack of an effect which could be due to a lack of power. However, we should note that the three-way interaction effect is numerically in the direction that the interaction effect between Structure and Distance is larger in the weishenme_P condition than in the weishenme_R condition, opposite of what previous generalizations predict even if it is a false negative.

In sum, the results of the current study show that both the reason and the purpose interpretations of weishenme “why,” as well as the wh-PP weileshenme “for what” are all restricted by the relative clause island in Chinese. This poses a challenge to various syntactic accounts of Chinese wh-in-situ. Assuming that the weishenme_P and weileshenme_P both have the internal structure in (2b), repeated below as (6), they both contain a wh-DP shenme “what” (Lin, 1992; Tsai, 1994, 1999, 2008; Stepanov and Tsai, 2008; Fujii et al., 2014). Therefore, they should be grammatical when appearing inside an island according to theories that assume that island effects (or to be more specific, subjacency requirements) do not operate at the LF level (Huang, 1982; Pesetsky, 1987; Uriagereka, 1999; Fox and Pesetsky, 2005), and theories that assume in-situ wh-DPs do not need to undergo LF movement to its scope position (Baker, 1970; Pesetsky, 1987; Nishigauchi, 1990; Tsai, 1994; Reinhart, 1998).

However, we saw in the experiment that island violations arise when in-situ weishenme_P and weileshenme_P appear inside relative clauses. This suggests that either the structure in (6) is incorrect and both weishenme_P and weileshenme_P are in fact wh-adverbs just like weishenme_R, or that wh-DPs also undergoes LF movement while island effects restrict such movements. The latter approach is supported by recent experimental findings (Lu et al., 2020) that the in-situ wh-DP shenme “what” is in fact restricted by islands contrary to previous claims, and fits nicely with developments in minimalist syntax that covert and overt movements are essentially the same with the only difference being which copy of the moved element is pronounced (Chomsky and Howard, 1993; Nunes, 1995, inter alia), and should thus be subject to the same set of restrictions.

One interesting observation pointed out by an anonymous reviewer is that the weishenme_P conditions are generally rated to be less acceptable than the weishenme_R conditions. Although we do not have a definitive explanation for this contrast, below are two possibilities. First, in the current experimental design, weileshenme_P is a salient lexical alternative for weishenme_P, while weishenme_R lacks such an alternative. Through Gricean reasoning, when the word weishenme is used, the comprehenders would infer that the intended meaning is more likely a reason interpretation, because the unambiguous weileshenme_P could have been used had the intended meaning been a purpose interpretation. Another possibility is that the purpose interpretation of weishenme is simply less frequent compared to the reason interpretation. This possibility could be tested by further corpus studies.

In this study, we provide experimental evidence that both the reason and purpose interpretations of weishenme “why,” and the wh-PP weileshenme “for what” are all sensitive to island effect in Chinese. Furthermore, we found no evidence suggesting any difference in their island sensitivity. These results challenge the longstanding generalization that nominal wh-in-situ are island insensitive in Chinese.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Sungkyunkwan University. The patients/participants provided their written informed consent to participate in this study.

NK and JL conceived the study, implemented the experiment, conducted the statistical analyses of the data, and supervised the stimuli creation. ZL created the stimuli. All authors contributed to planning the research and participated in writing the article.

This research was supported by the Sungkyunkwan University and the BK21 FOUR (Graduate School Innovation) funded by the Ministry of Education (MOE, South Korea) and National Research Foundation of Korea (NRF).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.