Island-sensitivity of two different interpretations of why in Chinese

Kim, Nayoun; Li, Ziying; Lu, Jiayi

doi:10.3389/fpsyg.2022.1059823

BRIEF RESEARCH REPORT article

Front. Psychol., 20 January 2023

Sec. Psychology of Language

Volume 13 - 2022 | https://doi.org/10.3389/fpsyg.2022.1059823

Island-sensitivity of two different interpretations of why in Chinese

Nayoun Kim¹

Ziying Li¹

Jiayi Lu²^*

¹Department of English Language and Literature, Sungkyunkwan University, Seoul, South Korea
²Department of Linguistics, Stanford University, Stanford, CA, United States

It has been assumed that the wh-element weishenme “why” in Chinese has two distinct interpretations: a reason reading, which typically yields yinwei “because”-answers, and a purpose reading, which typically triggers weile “in order to”-answers. It is claimed that the two interpretations differ in island sensitivity: the reason weishenme is sensitive to islands while the purpose weishenme is not. Assuming that the reason weishenme is a wh-adverb without finer internal structure, while the purpose weishenme is a wh-PP consisting of the preposition wei “for” and a wh-DP shenme “what,” this contrast in island sensitivity can be considered as an instance of a broader generalization: the so-called argument-adjunct asymmetry (or the DP-adverb asymmetry) of wh-in-situ island sensitivity. However, recent experimental studies provided mixed findings on whether the argument-adjunct asymmetry of wh-in-situ island sensitivity actually holds. The current study focuses on the two interpretations of weishenme “why/for what” in Chinese, and provides evidence using a formal acceptability judgment experiment that the two weishenmes are both sensitive to islands, contrary to previous generalizations. Our results provide further empirical challenge to the so-called argument-adjunct asymmetry of wh-in-situ island sensitivity.

1. Introduction

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).

1. A: Xiaoli weishenme jingchang bangzhu Xiaochen?

Xiaoli why/for.what often help Xiaochen

a. “Why does Xiaoli often help Xiaochen?”

B: Yinwei ta xihuan Xiaochen.

because he like Xiaochen

“Because he likes Xiaochen.”

b. “For what purpose does Xiaoli often help Xiaochen?”

B. Weile taohao Xiaochen.

in.order.to please Xiaochen

“In order to please Xiaochen.”

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).

2. a. [_AdvP [weishenme “why”]].

b. [_PP [_P wei(le) “for”] [_DP shenme “what”]].

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).

3. [Women {a. *weishenme, b. wei(le)shenme} nianshu]

we why for what study

cai you yiyi?

just have meaning

a. *“What is the reason x such that it is meaningful [for us to study for x]?”

b. “What is the purpose x such that it is meaningful [for us to study for x]?”

4. Ni bijiao xihuan [[{a. *weishenme, b. wei(le)shenme}

you more like why for what

gongzuo] de ren]?

work REL people.

a. *“What is the reason x such that you like better [people [who work for x]]?”

b. “What is the purpose x such that you like better [people [who work for x]]?”

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.

2. Probing island effects experimentally

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)).

5. Example stimuli for probing relative clause island effects with a 2 × 2 factorial design.

a. Embedded structure is island, long extraction:

What did John see the girl who was eating__?

b. Embedded structure is island, short extraction:

Who __ saw the girl who was eating sushi?

c. Embedded structure is non-island, long extraction:

What did John think that the girl was eating __?

d. Embedded structure is non-island, short extraction:

Who __ thought that the girl was eating sushi?

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.

2.1. Predictions based on previous generalizations

Following the previous generalization by Tsai (1994) among others, weishenme_R needs to undergo covert LF movement and is predicted to show island sensitivity. By contrast, weishenme_P, which, just like weileshenme_P, is a wh-PP and contains a nominal wh-element that can be licensed in-situ without covert movement, should be insensitive to islands. Using the factorial design to probe for island sensitivity as introduced in the previous sections, we expect there to be a significant interaction of embedded structure and dependency distance when the wh-element is weishenme_R but not weishenme_P or weileshenme_P.

2.2. Method

To test our hypothesis, we conducted an acceptability judgment experiment employing a 2 × 2 × 3 design, manipulating the structure of the embedded clause (non-RC vs. RC, where “non-RC” refers to a complement clause structure, and “RC” refers to a relative clause structure), wh-scope dependency distance (short vs. long, where “short” means the wh-scope dependency does not cross the embedded clause boundary, and “long” means the wh-scope dependency crosses the embedded clause boundary), and wh-type (weishenme_R, weishenme_P, and weileshenme_P ‘for what’). Note that we followed Lu et al. (2020) and used relative clauses to probe for island sensitivity. All stimuli are in the form of question-answer pairs. The question sentence for the weishenme_R and the weishenme_P conditions are identical, and the difference in wh-type is achieved through the answer sentences that disambiguate the intended interpretation of weishenme in the question sentence. The weileshenme_P “for what” condition is included as a sanity check: its question sentences are the same as the other two conditions except that the wh-element weishenme “why” is replaced with weileshenme “for what,” forcing the purpose interpretation. Example stimuli are shown below in Table 1.

TABLE 1

Table 1. Example stimuli for the experiment.

The experiment was implemented on PC IbexFarm, a web-based presentation platform (Drummond, 2020). A total of 40 native speakers of Chinese (age range: 20–40, mean age: 24) were recruited to participate in the experiment. Each participant was paid an electronic convenience store voucher with the equivalent value of ₩2000 (approximately $2) after completion of the experiment. All participants were born in mainland China, acquired Mandarin Chinese as their first language, and use Chinese as a dominant language. They participated in the experiment using their own laptop via the experimental link distributed through Prolific.co. Experimental stimuli (Question/Answer pairs) were presented one at a time, and participants were asked to rate the naturalness of the Question/Answer pair (1 = totally unnatural, 7 = totally natural). A total of five practice questions were given prior to the actual experiment. There were 16 critical items in total. Each critical item appears exactly once for each participant, and randomly appears as one of the critical conditions. Also included in each presentation list were 48 filler items irrelevant to the current experimental manipulation. Among the filler items, there are 24 natural question-answer pairs and 24 unnatural ones. The fillers contain yes-no questions and wh-questions (other than weishenme “why”) with no island structures. Each participant saw 64 test items in total in addition to the 5 practice items at the beginning of the experiment.

2.3. Results

We first calculated the by-participant z-score from the raw ratings. Among the filler items, natural fillers received a mean acceptability z-score of 0.67 (SE = 0.021), and unnatural fillers received a mean acceptability z-score of-0.91 (SE = 0.022). The mean acceptability z-score of each critical condition is shown in Figure 1. For each of the three wh-types (weishenme_R, weishenme_P, and weileshenme_P), we analyzed the results using the same linear mixed-effects regression model (Baayen et al., 2008; Barr et al., 2013), predicting acceptability rating from the sum-coded fixed effects of Structure (non-RC vs. RC) and Distance (short vs. long) and their interaction. Also included in each model are the by-participant and by-item random intercepts and random slopes for both fixed effects and their interaction. Island effects, as discussed earlier, is detected as a positive interaction between Structure and Distance.

FIGURE 1

Figure 1. Mean acceptability Z-scores for all conditions.

In the weishenme_R condition, we found significant main effects of Structure (β = 0.12, SE = 0.057, t = 2.20) and Distance (β = 0.16, SE = 0.078, t = 2.10) such that the non-RC condition is more acceptable than the RC condition, and the short condition is more acceptable than the long condition. Furthermore, there is a significant interaction between Structure and Distance (β = 0.11, SE = 0.051, t = 2.24).

In the weishenme_P condition, there is no significant main effect of Structure (β = 0.012, SE = 0.048, t = 0.26), but there is a marginally significant main effect of Distance (β = 0.11, SE = 0.062, t = 1.77) such that the short condition is more acceptable than the long condition. There is also a significant interaction between Structure and Distance (β = 0.16, SE = 0.060, t = 2.65).

In the weileshenme_P condition, there are significant main effects of Structure (β = 0.15, SE = 0.057, t = 2.55) and Distance (β = 0.090, SE = 0.044, t = 2.07) such that the non-RC condition is more acceptable than the RC condition, and the short condition is more acceptable than the long condition. There is also a significant interaction between Structure and Distance (β = 0.15, SE = 0.047, t = 3.29).

To test whether there is any difference between the wh-types in terms of island sensitivity, we pooled the data from all three wh-types together and fit a linear mixed-effects regression model predicting acceptability rating from the sum-coded fixed effects of Structure and Distance, and the dummy-coded fixed effect of Wh-type with reference level set to weishenme_P. The model also includes by-item and by-participant random intercepts and random slopes for all three fixed effects and their interactions. We observed no significant three-way interaction for either weishenme_R (β = 0.027, SE = 0.066, t = 0.41) or weileshenme_P (β = −0.016, SE = 0.066, t = −0.25). An omnibus test (Type III ANOVA with Satterthwaite’s method) on the model shows that there is no significant three-way interaction (F(2) = 0.083, p = 0.92), further confirming that wh-type does not affect the magnitude of islandhood.

3. Discussion

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).

6. [_PP [_P wei(le) “for”] [_DP shenme “what”]].

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.

4. Conclusion

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.

Data availability statement

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

Ethics statement

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.

Author contributions

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.

Funding

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).

Conflict of interest

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.

Publisher’s note

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.

Footnotes

1. ^See Cheng (2009) for a comprehensive review on the various analyses of in-situ wh-elements.

2. ^See Tian et al. (2022) for an argument that Lu et al. (2020) experimental design is confounded.

3. ^The terms "island," "island violations," and "islandhood" are used in this paper in a purely descriptive and theory-neutral way: we remain agnostic about whether such "island violations" are the result of violated grammatical constraints, or the result of processing-level or informational structural constraints. In the current study, we simply probe for the island sensitivity of wei(le)shenme in Chinese with the aim to verify/challenge previous empirical generalizations. For various non-grammar-level proposals for islands effects in general, see Chaves (2021) and Liu et al. (2022), inter alia.

4. ^The high absolute ratings of island/long-extraction sentences might point to processing-level or discourse-level accounts of the island effect. However, for the purpose of discussion and in a purely descriptive sense, we still refer to such sentences as containing an "island violation" in this paper.

References

Aoun, J., and Li, Y. H. A. (1993). Wh-elements in situ: syntax or LF? Linguist. Inq. 24, 199–238.