The D-Linking Effect on Wh-Extractions from Islands and Non-Islands in L1 Speakers and L2 Learners of English

This study examines the effect of discourse-linked (d-linked) wh -phrases on acceptability of wh - extractions from islands and non-islands. Native speakers of English ( n = 27) and Najdi Arabic learners of English ( n = 31) rated on a 7-point scale their acceptability of wh-questions with bare and d-linked wh-extractions from islands and non-islands (e.g., *what i /*which movie i does she wonder [why he might hate __ i ]?). Learners, like native speakers, showed sensitivity to island constraints as reflected in their low acceptability ratings of wh-extractions from islands. Learners, like native speakers, were also more sensitive to strong (universal constraints) than to weak islands (language-specific constraints), as predicted by Belikova and White’s (2009) proposal. Moreover, both native speakers and learners exhibited a d-linking effect on wh-extractions from wh-islands, rating d-linked higher than bare wh-extractions. As for the source of this d-linking effect, the results of native speakers rather than learners, who could have been misled by the semantic cues of d-linked wh-phrases, suggest that this d-linking effect is more likely caused by the d-linked wh -phrase’s denotation of discrete individuals as claimed by the semantic account of d-linking effect.

Sensitivity to island constraints on wh-movement is a topic of interest in second language acquisition and native psycholinguistics literature.In English, for example, whquestions involve wh-movement (Chomsky, 1981(Chomsky, , 1986)).In (1), the wh-phrase ("what") originates in the object position after the verb ("see") and moves to the beginning of the sentence, leaving a trace. (1) What i did you see __ i ?
( WH-ISLAND English native speakers give low acceptability judgments to ungrammatical sentences that violate island constraints as in (2), suggesting sensitivity to island constraints (e.g., Sprouse et al., 2012).
Second language (L2) research has focused on the acquisition of island constraints to argue for or against L2 learners' access to Universal Grammar (UG).Previous studies that tested L2 acquisition of island constraints did not show consistent results.Some studies argued that sensitivity to island constraints is possible for L2 learners regardless of L1 (e.g., Li, 1998;Martohardjono, 1993).Other studies argued that sensitivity to island constraints is possible only for L2 learners whose L1 instantiates overt wh-movement (e.g., Hawkins & Chan, 1997;Johnson & Newport, 1991).Belikova and White (2009) pointed out that, although previous studies argued for or against island sensitivity, further examination of their results by island type showed that L2 learners were more sensitive to specific types of islands (i.e., adjunct clauses, RCs, sentential subjects) than others (i.e., complex NPs, wh-islands).To account for variations in L2 learners' sensitivity to island types, Belikova and White (2009) proposed, based on Huang's (1982) revised Condition on Extraction Domains, that L2 learners are expected to be more sensitive to strong islands (universal constraints) than to weak islands (language-specific constraints) if they have access to UG.
Although native speakers of English are sensitive to islands, their sensitivity is affected by the linguistic properties of the extracted wh-phrase.Following terminology introduced in Pesetsky (1987), discourse-linked (d-linked) wh-phrases (e.g., "which movie") arguably weaken island effects and increase the acceptability of wh-extractions from islands (e.g., Hofmeister & Sag, 2010).In (3a), the extracted wh-phrase (e.g., "what") is a bare whphrase, and the sentence is expected to receive low acceptability.However, when the bare wh-phrase (e.g., "what") is replaced by a d-linked wh-phrase (e.g., "which movie") as in (3b), the sentence is expected to receive higher acceptability.The dlinking effect on acceptability of wh-extractions from islands is surprising, and there is currently debate in psycholinguistics about the source of this d-linking effect (e.g., Alexopoulou & Keller, 2013;Goodall, 2015;Hofmeister & Sag, 2010).Under the complexity account (e.g., Hofmeister & Sag, 2010), this d-linking effect is caused by the semantic and structural complexity of the d-linked wh-phrase.Under the semantic account (e.g., Szabolcsi & Zwarts, 1993, 1997), however, this d-linking effect is caused by the d-linked wh-phrase's semantic denotation of discrete individuals.
To further investigate island sensitivity and d-linking effect on wh-extractions from islands, in the present study I tested English native speakers and Najdi Arabic learners of English to answer four questions.The first question is whether Najdi learners can show sensitivity to island constraints on wh-movement in English.If so, are they more sensitive to strong (universal constraints) than to weak islands (language-specific constraints), as predicted by Belikova and White (2009)?If L2 learners show island sensitivity as English natives do, this introduces the third question: Is this island sensitivity weakened by d-linking?If so, this raises the fourth question: Is this d-linking effect caused by the complexity of the dlinked wh-phrase as claimed by the complexity account or by the d-linked wh-phrase's denotation of discrete individuals as claimed by the semantic account.

Literature Review
I first give an overview of wh-movement and its island constraints in Najdi Arabic.Next, I review acceptability studies that examined L2 acquisition of island constraints.Then, I review two accounts of d-linking effect and the studies that tested the d-linking effect on wh-extractions.Finally, I discuss the details of the present study.

Linguistic Facts in Najdi Arabic
Wh-questions in Arabic dialects are formed via a variety of strategies.In Modern Standard Arabic, for example, wh-questions can be formed by moving the wh-phrase to the beginning of the sentence (4) or by a strategy of resumption ( 5) 1 (e.g., Alotaibi & Borsley, 2013;Aoun et al., 2010).

1
The examples in (4) and ( 5) are from Aoun et al. (2010), p. 132.  2 The example in ( 6) is from Tucker et al. (2019), p. 54.  3 The examples in ( 7), ( 8) and ( 9) are from Albaty (2013) Chomsky (1973) proposed the subjacency principle to account for all types of island constraints, which states that a wh-phrase cannot cross more than one bounding node, IP or DP, in each single movement.In ( 12

WH-ISLAND
Under this version of island constraints, L2 learners are expected to treat all types of islands similarly if they have access to UG.However, previous L2 studies (e.g., Johnson & Newport, 1991;Li, 1998;Schachter, 1990) that adopted this version of island constraints showed that learners were more sensitive to specific types of islands (i.e., adjunct clauses, RCs, sentential subjects) than others (i.e., complex NPs, wh-islands).Martohardjono (1993), for example, examined sensitivity to island constraints by testing Italian learners of English.In Italian, as in Najdi Arabic, wh-questions are formed via wh-movement, which is sensitive to island constraints.The results showed that Italian learners, like English native speakers, treated island types differently, being more sensitive to adjunct islands and RC islands than to complex NP islands and wh-islands.
To account for variations in L2 learners' performance on types of islands, Belikova and White (2009) adopted an alternate version of island constraints.This version is a revised version of Huang's (1982) Condition on Extraction Domains (CED), under which extraction from non-complements is universally impossible.Therefore, extraction from strong islands (i.e., adjunct clauses, RCs, and sentential subjects) is not possible universally because strong islands are non-complements.However, this entails that the ungrammaticality of extraction from weak islands (e.g., wh-islands, complex NPs) can be attributed to parametric variation.Based on Huang's revised CED, Belikova and White (2009) proposed that learners should be more sensitive to strong than to weak islands because strong islands are universal constraints on extraction while weak islands are language-specific constraints.
Although native speakers of English are sensitive to islands, sensitivity is affected by the type of extracted wh-phrase.For example, d-linked wh-phrases (e.g., "which movie") as in (13b) compared to bare wh-phrases (e.g., "what") as in (13a) have been argued to weaken island effects and increase the acceptability of wh-extractions from islands (e.g., Hofmeister & Sag, 2010).In (13a), the wh-phrase ("what") that is extracted from a wh-island is a bare wh-phrase and the sentence is expected to receive low acceptability.However, when the bare wh-phrase is replaced by a d-linked wh-phrase ("which movie") that prompts an answer that can be inferred from referents already introduced into the discourse as in (13b), the sentence is expected to receive higher acceptability.The d-linking effect on wh-extractions from islands presents an interesting puzzle, and it is not clear how d-linking interacts with syntactic constraints and increases acceptability of wh-extractions from islands.Many accounts have been proposed to explain the source of d-linking effect on wh-extractions from islands.The next section reviews two accounts of d-linking effect on wh-extractions from islands and the studies that tested these two accounts.

Accounts of D-Linking Effect on Wh-Extractions
Under the complexity account (e.g., Hofmeister & Sag, 2010), the d-linking effect on wh-extractions from islands is caused by the complexity of the extracted wh-phrase.This account claims that semantically and structurally more complex wh-phrases (e.g., "which movie") have stronger mental representations compared to bare wh-phrases (e.g., "what") and are thus easier to retrieve from working memory at the gap site (the subcategorizing verb).This ease of processing is claimed to result in higher acceptability.
To support the complexity view of d-linking effect, Hofmeister and Sag (2010) manipulated the type of wh-phrase in a self-paced reading task to show that complex whphrases compared to bare wh-phrases can facilitate processing of wh-extractions from whislands and improve their acceptability.Native speakers of English first read a declarative background sentence, and then read either a question with a bare wh-extraction from a whisland (14a), a question with a d-linked wh-extraction from a wh-island (14b), or a baseline question with a bare wh-extraction from a that-clause (14c)5 .( 14)

BACKGROUND SENTENCE
Albert learned that the managers dismissed the employee with poor sales after the annual performance review.c.Who did Albert learn that they dismissed after the annual performance review?
The results showed faster reading times for the complex wh-phrase condition (14b) than for the bare wh-phrase condition (14a) at the three regions (e.g., "after the annual") that follow the embedded verb (e.g., "dismissed") inside the island, where the retrieval of wh-phrase from working memory is expected to take place.Hofmeister and Sag (2010) argued that this suggests that processing of wh-extractions from wh-islands can be facilitated when the complexity of wh-phrase is increased.
Like Hofmeister and Sag (2010), Goodall (2015) also claimed the d-linking effect is caused by the complexity of the extracted wh-phrase.Goodall tested the d-linking effect on wh-extractions from islands and non-islands by manipulating the type of wh-phrase (bare vs. complex) and the type of the structure in which the gap was located (complex NP vs. whclause vs. that-clause) in six conditions using a 2 × 3 design, as in (15) 6 .Goodall (2015) tested the complexity account, which claims that complex wh-phrases increase the acceptability because they are easier to retrieve at the gap site regardless of whether the gap was located inside an island or non-island structure.If this claim is right, there should be an increase in acceptability of d-linked wh-extractions from both islands (i.e., complex NP islands, wh-islands) and non-islands (that-clauses) as compared to their bare counterparts.Using a 7-point rating scale, with 7 being very good, English native speakers, as predicted, rated d-linked wh-extractions from both islands and non-islands higher than their bare counterparts, supporting the complexity account of d-linking effect.
Unlike Hofmeister and Sag (2010) and Goodall (2015), Alexopoulou and Keller (2013) argued the d-linking effect is driven by semantic factors.Under the semantic account (Szabolcsi & Zwarts, 1993, 1997), which views sensitivity to weak islands as a semantic phenomenon, d-linked wh-phrases increase the acceptability of wh-extractions from weak islands because they denote discrete individuals that can participate in the semantic Boolean operations (e.g., intersection, union, complementation) necessary for the interpretation of weak islands.However, bare wh-phrases are not good extractors because they usually denote kinds that cannot be collected into sets that form the semantic Boolean operations.
To support the semantic view of d-linking effect, Alexopoulou and Keller (2013) examined the d-linking effect on wh-extractions from islands and non-islands as in (16) 7 .The example in ( 15) is from Goodall (2015), p. 4. 7 The example in ( 16) is from Alexopoulou & Keller (2013) Szabolcsi and Zwarts (1993).However, they did not find a d-linking effect on wh-extractions from non-island structures, namely main clauses (16b) and embedded that-clauses (16c) because these structures do not involve a scope island for which the denotation of the d-linked wh-phrase is crucial for its interpretation.They argued that these results support the semantic account of d-linking effect.Goodall (2015) and Hofmeister and Sag (2010) claimed the d-linking effect is caused by the complexity of d-linked wh-phrase, while Alexopoulou and Keller (2013) argued this effect is caused by the d-linked wh-phrase's denotation of discrete individuals as claimed by the semantic account.The source of d-linking effect is still a controversial issue, and this study further investigates the issue by testing the predictions of the complexity account and semantic account of the d-linking effect.

The Present Study
Unlike previous studies (e.g., Alexopoulou & Keller, 2013;Goodall, 2015;Hofmeister & Sag, 2010) that tested only English native speakers to examine island sensitivity and the d-linking effect on wh-extractions, this study also tests L2 learners, a population that is particularly interesting to examine from this perspective, to shed light on whether island sensitivity and d-linking effect are similar in the two populations.

Research Questions
The primary goal of this study is to examine whether island sensitivity and the dlinking effect on wh-extractions are similar in nature in L2 learners and native speakers.The first step is to examine whether Najdi Arabic learners of English show sensitivity to island constraints on wh-movement in English.If Najdi Arabic learners, like native speakers, show sensitivity to islands, this prompts the second question of whether they are more sensitive to strong islands (universal constraints) than to weak islands (language-specific constraints) as predicted by Belikova and White (2009).A third question can also be raised of whether this island sensitivity exhibited by natives and learners is weakened by d-linking effect.If so, this raises the fourth question of whether this d-linking effect is caused by the complexity of the d-linked wh-phrase as claimed by the complexity account or caused by the d-linked whphrase's denotation of discrete individuals as claimed by the semantic account.

Participants
Thirty-one Najdi Arabic learners of English voluntarily participated in the study.The Arabic learners (30 males, mean age = 27.4)started learning English as adults in public schools.All learners completed the Michigan Listening Comprehension Test to assess their English proficiency.The test consisted of 45 listening comprehension questions that targeted various grammatical constructions.The learners' scores ranged from 37 to 44 out of 45 possible correct answers (M = 40.05,SD = 2.31).They were all tested at Imam University, Riyadh, Saudi Arabia.Twenty-seven monolingual native speakers of English (13 females, 14 males; mean age = 40.3)also participated in the study.Twenty-four of them were from the United States and were tested at the University of Kansas, USA, and three participants were from the United Kingdom and were tested at Imam University.Each participant was paid $15 for participating.

Materials
The stimuli in this study were designed to test the effects of two island types: whislands (weak islands) and RC islands (strong islands).To test each of the two island types, the wh-extraction site and the wh-phrase type were manipulated in four conditions as in ( 17).The wh-extraction is either from an island structure, as in (17a) and (17b), or from a nonisland structure, as in (17c) and (17d).The wh-phrase is either a bare wh-phrase (e.g., what), as in (17a) and (17c), or a d-linked wh-phrase (e.g., which movie), as in (17b) and (17d).The first two conditions are ungrammatical because of wh-extraction from an island, while the last two conditions are grammatical because of wh-extraction from a non-island structure and they serve as controls.An example of one set to test the effects of RC islands is shown in (18) 8 .To test wh-islands, I used 16 sets of sentences.They included four sets with the wh-word why, four sets with the wh-word "how," four sets with the wh-word "where" and four sets with the wh-word "when" heading the subordinate clause.To test RC islands, I also used 16 sets of sentences: eight sets with the head of the RC in subject position (four sets with the relative pronoun "who," four sets using "that") and eight sets with the head of the RC in object position (four sets with the relative pronoun "who," four sets using "that").
The sentences from the 32 sets were distributed among four lists using a Latin square design, such that every participant was presented with only one sentence from every set.Each list had 32 sentences that included four sentences for each of the four conditions in wh-

8
The experimental sentences were created by the author, while the fillers were taken from Hawkins & Chan (1997) with some modification.islands and RC islands.Because all experimental sentences were wh-questions, 32 declarative filler sentences were added to each list.Thus, the total number of sentences in each list was 64, including 32 experimental sentences (half grammatical, half ungrammatical) and 32 filler sentences (half grammatical, half ungrammatical).The sentences in each list were presented in four blocks.Each block included eight experimental sentences (four grammatical, four ungrammatical) and eight filler sentences (four grammatical, four ungrammatical).The sentences were randomized in each block.All experimental materials are in Appendix A.

Acceptability Judgment Task
I conducted an acceptability judgment task (AJT), using the experimental control software Paradigm (Tagliaferri, 2005).In each experimental trial, a sentence appeared on the computer screen.The participant then judged, with no time limits, whether the sentence sounded natural or unnatural in English, using a 7-point rating scale displayed underneath the sentence.The rating scale ranged from totally unnatural to perfectly natural.The participants could choose I do not know if they could not make a judgment.The test began with six practice trials to familiarize participants with the task.

Procedure
Native speakers and Najdi learners were tested individually, using a computer.They signed a consent form and completed a background questionnaire.Before Najdi learners took the AJT, they were asked to complete the Michigan Listening Comprehension Test to assess their English proficiency.

Sensitivity to island constraints on wh-movement
As shown in the literature review, wh-questions in Najdi Arabic can be formed via wh-movement, which is sensitive to islands, as is the case in English.This suggests that Najdi Arabic learners have wh-movement and island sensitivity in their L1.Therefore, Najdi Arabic learners, like English native speakers, are predicted to make a distinction between ungrammatical and grammatical wh-extractions, rating ungrammatical wh-extractions from islands lower than grammatical wh-extractions from non-islands.Belikova and White (2009) claimed that L2 learners, like native speakers, are expected to be more sensitive to strong (universal constraints) than to weak islands (language-specific constraints) if they have access to UG.If this claim is right, Najdi learners will rate wh-extractions from RC islands (strong islands) lower than wh-extractions from whislands (weak islands).

D-Linking Effect and Its Source
Much of the literature on d-linking (e.g., Phillips, 2013;Szabolcsi, 2006) claims that d-linking has a greater effect on wh-extractions from weak islands than on wh-extractions from strong islands.If correct, native speakers and learners will exhibit a greater d-linking effect on wh-extractions from wh-islands (weak islands) than on wh-extractions from RC islands (strong islands).
As for the source of d-linking effect, the complexity account argues the d-linking effect is caused by the semantic and structural complexity of the d-linked wh-phrase.This account claims that d-linked wh-phrases, which are complex, are easy to retrieve from working memory at the gap site and this ease of processing leads to an increase in acceptability regardless of whether the wh-extraction is from an island or non-island structure.If correct, native speakers and learners will show an increase in acceptability of dlinked wh-extractions from both island and non-island structures.
However, the semantic account that is relevant only for weak islands argues the dlinking effect is caused by semantic factors 9 .This account claims that d-linked wh-phrases increase acceptability of wh-extractions from weak islands because they denote discrete individuals that can participate in the semantic Boolean operations necessary for the interpretation of weak islands.If correct, native speakers and learners will show an increase in acceptability of d-linked wh-extractions from wh-islands (weak islands) but not of d-linked wh-extractions from RC islands (strong islands) or non-islands because the denotation of the d-linked wh-phrase is not crucial for the interpretation of strong island and non-island structures.

Results
In this section, I first present the results of whether native speakers and learners distinguished ungrammatical from grammatical wh-extractions.Then I present the results of whether native speakers and leaners are more sensitive to strong than to weak islands and the results of the d-linking effect on wh-extractions from islands.Finally, I present the results of the d-linking effect on wh-extractions from non-islands.Before analysis, each participant's acceptability ratings were converted into z scores to eliminate the possibility that participants may vary in their use of the range of the 7-point rating scale used in the AJT.

Ungrammatical vs. Grammatical Wh-extractions
To examine whether native speakers and learners distinguished ungrammatical from grammatical wh-extractions and whether the distinction is affected by wh-phrase type and island type, I conducted a mixed four-way repeated measures ANOVA for acceptability ratings with Grammaticality (grammatical vs. ungrammatical wh-extraction), Wh-Phrase Type (bare vs. d-linked) and Island Type (wh-island vs. RC island) as within-subjects factors and Group (native speakers vs. learners) as the between-subjects factor.Figure 1 summarizes the results for native speakers; Figure 2 summarizes the results for learners.
The analysis showed a main effect of Wh-Phrase Type [F (1,56) = 9.067, p =.004] and an interaction between Wh-Phrase Type and Group [F(1,56) = 9.995, p = .003].There was also an interaction between Grammaticality and Wh-Phrase Type [F(1,56) = 7.716, p = .007]and no three-way interaction with Group [F(1,56) = 2.251, p = .139].This indicates that the distinction in grammaticality is more pronounced in bare than in d-linked wh-extractions for both native speakers and learners and no difference between the two groups.

9
The semantic account is relevant for weak islands because it views them as a semantic phenomenon.The semantic account, however, is not relevant for strong islands because they are typically taken to be a syntactic phenomenon, and the semantic denotation of the extracted wh-phrase is not crucial for their interpretation.
The analysis also showed a main effect of Island Type [F(1,56) = 37.011, p = .000]but did not show an interaction between Island Type and Group [F(1,56) = 0.001, p = .974].Moreover, there was not an interaction between Grammaticality and Island Type [F(1,56) = .553,p = .460]and no three-way interaction with Group [F(1,56) = 1.442, p = .235].This indicates that the distinction in grammaticality was the same in wh-island and RC island sentences for both native speakers and learners, with no difference between the two groups.There was an interaction between Island Type and Wh-Phrase Type [F(1,56) = 11.772,p = .001],but there was no three-way interaction with Group [F(1,56) = 1.442, p = .235].There was no three-way interaction among Grammaticality, Island Type, and Wh-Phrase Type [F(1,56) = .137,p = .713]and no four-way interaction with Group [F(1,56) = 1.186, p = .281].This indicates that the distinction in grammaticality for bare and d-linked wh-extractions was the same in wh-island and RC island sentences for both native speakers and learners, with no difference between the two groups.

Ungrammatical Wh-extractions From Islands
In this section, I present the results of whether native speakers and learners are more sensitive to strong than to weak islands and the results of the d-linking effect on whextractions from islands.I conducted a mixed three-way repeated measures ANOVA for acceptability ratings of wh-extractions from islands with Wh-Phrase Type (bare vs. d-linked) and Island Type (wh-island vs. RC island) as within-subjects factors and Group (native speakers vs. learners) as the between-subjects factor.
The results of ANOVA revealed a main effect of Island Type [F(1,56) = 30.754,p = .000]but no interaction between Island Type and Group [F(1,56) = .859,p = .358].This indicates that both native speakers and learners rated wh-extractions from RC islands (M = −1.00)lower than wh-extractions from wh-islands (M = −0.72)and no difference between the two groups, being more sensitive to strong islands (universal constraints) than to weak islands (language-specific constraints).To examine whether native speakers and learners were more sensitive to strong than to weak islands for both d-linked and bare wh-phrases, I conducted pairwise comparisons between the wh-island/d-linked condition and the RC-island/d-linked condition (Natives: t(26) = 4.555, p = .000;Learners: t(30) = 3.257, p = .003)and between the wh-island/bare condition and the RC-island/bare condition (Natives: t(26) = 1.822, p = .080;Learners: t(30) = 3.148, p = .004).The pairwise comparisons indicate that learners were more sensitive to strong than to weak islands for both d-linked and bare wh-phrases.
Native speakers were also more sensitive to strong than to weak islands for d-linked whphrases.In the case of bare wh-phrases, they tended to show a similar pattern of results and this was marginally significant.
Crucially, there was an interaction between Wh-Phrase Type and Island Type [F(1,56) = 6.974, p = .011]but no three-way interaction with Group [F(1,56) = .331,p = .567].This suggests that the d-linking effect is greater on wh-extractions from wh-islands than on whextractions from RC islands for both native speakers and learners and no difference between the two groups.

Grammatical Wh-extractions From Non-Islands
In this section, I present the results of the d-linking effect on wh-extractions from nonislands.Because native speakers patterned differently from learners with respect to the dlinking effect on wh-extractions from non-islands as shown in Figure 1 and Figure 2, I conducted a separate analysis for each group to examine whether d-linking increases acceptability of wh-extractions from non-islands and whether this is affected by Structure Type.For native speakers, I conducted a two-way repeated measures ANOVA for ratings of wh-extractions from non-islands with Wh-Phrase Type (bare vs. d-linked) and Structure Type (that-clause vs. main RC) as within-subjects factors.
The results of ANOVA revealed a main effect of Wh-Phrase Type [F(1,26) = 7.919, p = .009],which indicates that native speakers did not exhibit a d-linking effect on whextractions from non-islands, rating bare wh-extractions (M = 0.65) higher than d-linked whextractions (M = 0.49).The analysis also showed a main effect of Structure Type [F(1,26) = 8.032, p = .009],which indicates that native speakers rated wh-extractions from that-clauses (M = 0.71) higher than wh-extractions from main RCs (M = 0.44).There was no interaction between Wh-Phrase Type and Structure Type [F(1,26) = .451,p = .508],indicating that dlinking effect is the same in wh-extractions from that-clauses and wh-extractions from main RCs.
For learners, the results of ANOVA revealed a marginally significant main effect of Wh-Phrase Type [F(1,26) = 3.610, p = .067],which indicates that learners, unlike native speakers, tended to exhibit a d-linking effect on wh-extractions from non-islands, rating dlinked wh-extractions (M = 0.41) higher than bare wh-extractions (M = 0.22).The analysis also showed a marginally main effect of Structure Type [F(1,26) = 4.037, p = .054],which indicates that learners tended to rate wh-extractions from that-clauses (M = 0.40) higher than wh-extractions from main RCs (M = 0.23).There was also a marginally significant interaction between Wh-Phrase Type and Structure Type [F(1,26) = 3.465, p = .073],which suggests that the d-linking effect was greater on wh-extractions from that-clauses than on whextractions from main RCs.eliminate island effects and restore the questions with wh-extractions from islands to full acceptability.Interestingly, similar results were found in Alexopoulou and Keller (2013), Goodall (2015), and Sprouse et al. (2016).To conclude, then, and in answer to the third question of this study, d-linking weakened island effects, and its effect was greater on whextractions from wh-islands (weak islands) than on wh-extractions from RC islands (strong islands).
The fourth study question explored whether this d-linking effect on wh-extractions from wh-islands is caused by the complexity of the d-linked wh-phrase (e.g., Hofmeister & Sag, 2010) or by the d-linked wh-phrase's denotation of discrete individuals (Szabolcsi & Zwarts, 1993, 1997).The complexity account claims that d-linked wh-phrases, which are semantically and structurally complex, are easy to retrieve from working memory at the gap site, and this ease of processing leads to an increase in acceptability.If this claim is correct, then I would expect an increase in acceptability regardless of whether the wh-extraction is from an island or non-island structure.However, the semantic account, which is relevant only for weak islands, claims that d-linked wh-phrases increase the acceptability of wh-extractions from weak islands because they denote discrete individuals that can participate in the semantic operations necessary for the interpretation of weak islands.If this claim is correct, then I would expect an increase in acceptability of wh-extractions from wh-islands (weak islands).However, I would not expect an increase in acceptability of wh-extractions from RC islands (strong islands) or non-island structures because the denotation of the extracted whphrase does not matter for the interpretation of RC islands and non-island structures.
Consistent with the predictions of the semantic account, native speakers' results showed that d-linked wh-phrases increased the acceptability of wh-extractions from whislands but not of wh-extractions from RC islands or non-island structures (main RCs and that-clauses).These results support the semantic account of d-linking effect.Both Alexopoulou and Keller (2013) and Sprouse et al. (2016) found similar results, showing a dlinking effect on wh-extractions from weak islands (whether islands, complex NP islands) but no d-linking effect on wh-extractions from strong islands (subject and adjunct islands) or non-island structures (main RCs and that-clauses).
However, Goodall (2015) found a d-linking effect not only on wh-extractions from islands, but also on wh-extractions from non-island structures (that-clauses).Goodall (2015) criticized the results of Alexopoulou and Keller (2013) that showed no d-linking effect on wh-extractions from non-island structures, arguing that Alexopoulou and Keller did not detect a d-linking effect because their experiment could not distinguish among sentences at the very high end of the acceptability scale.Specifically, Goodall (2015) claimed that Alexopoulou and Keller's (2013) results suggest the presence of a ceiling effect because they could not show a difference even in acceptability between sentences with short wh-dependencies and sentences with long wh-dependencies, for which many studies found a very clear difference in acceptability.
In this study, however, the non-effect of d-linking on wh-extractions from non-islands cannot be attributable to a ceiling effect in the scale.Interestingly, two conditions of grammatical filler sentences were rated higher by native speakers than the four experimental conditions of wh-extractions from non-island structures.These fillers were declarative sentences including a RC with a gap in subject position (M =1.15) as in ( 19), or with a gap in object position (M = 1.03) as in ( 20).
(19) The young man who ___was driving fast had an accident.(20) The patient who I visited ___ yesterday was very sick.
If there were indeed a d-linking effect on wh-extractions from non-island structures, no ceiling effect in the current experiment would affect the ability to detect it.The criticisms of Goodall (2015) against Alexopoulou and Keller (2013) cannot be raised against this study.However, it is not clear why Goodall (2015) found a d-linking effect on wh-extractions from that-clause structure as in ( 21) while this study that tested the same structure as in ( 22) did not 10 .It is noteworthy that there is no difference in terms of structure between Goodall's stimuli and the stimuli of this study, and neither was preceded by a context.Goodall observed a d-linking effect on wh-extractions from non-islands probably because he used a different type of d-linked wh-phrase (which of the N). ( 21) What / which of the cars do you believe that he might buy___?( 22) What / which movie does he think that she might hate ___?
In the case of Najdi learners, d-linked wh-phrases increased the acceptability of whextractions from wh-islands and RC islands.They also increased the acceptability of whextractions from one non-island structure (i.e., that-clause structure) but not the acceptability of wh-extractions from the other non-island structure (i.e., main RC structure).Except for whextractions from main RC structure, these results support the complexity account, which claims that d-linking increases not only the acceptability of wh-extractions from islands but also the acceptability of wh-extractions from non-islands.
However, I believe this pattern of results is not driven by the complexity of d-linked wh-phrases because Najdi learners showed two unusual findings related to d-linking.The first is the increase in acceptability of d-linked wh-extractions from RC islands (strong islands).If complexity of d-linked wh-phrases led to an increase in acceptability of these wh-extractions, why didn't native speakers show an increase in acceptability of these wh-extractions?Interestingly, native speakers rated both bare and d-linked wh-extractions from RC islands almost the same, suggesting no d-linking effect on wh-extractions from strong islands as shown by Sprouse et al. (2016), who also found no d-linking effect on wh-extractions from subject and adjunct clause islands (strong islands).
I believe learners exhibited this pattern of results because some were affected by the semantic cues of d-linked wh-phrases that misled them when they were processing whquestions with wh-extractions from RC islands.Specifically, the semantic cues of d-linked wh-phrases initially misled learners by making them tend to incorrectly associate the whphrase (e.g., which article), which was ungrammatically moved from the RC island, with the verb (e.g., wrote) as its argument because of the semantic plausibility match between the whphrase and the verb as shown in ( 23).
(23) *Which article did the author who wrote ____win the prize?However, after learners unconsciously felt that it was not grammatically possible to associate the wh-phrase ("which article"), which was moved from the island, with the verb ("wrote"), they realized that they needed to revise their initial analysis of the structure and consequently 10 The example in ( 21) is from Goodall (2015), p. 4. rejected the sentence.Thus, the increase in acceptability of d-linked wh-extractions from RC islands is more likely caused by learners' initial misanalysis of the sentence structure 11 .
The second unusual finding shown by learners is that d-linking increased the acceptability of wh-extractions from that-clause structure but not the acceptability of whextractions from the main RC structure.Again, I believe these results are driven by the effect of semantic cues of d-linked wh-phrases.D-linking increased acceptability of wh-extractions from that-clause structure because d-linked wh-phrases (e.g., "which movie"), as opposed to bare wh-phrases (e.g., "what"), are semantically more plausible arguments of the verb (e.g., "hate") in the embedded that-clause as shown in ( 24) and ( 25). ( 24) What does he think that she might hate ___? (25) Which movie does he think that she might hate___?
For wh-extractions from the main RC structure, I believe the semantic plausibility match between the extracted wh-phrase and the main clause verb did not help increase acceptability due to processing difficulty.Processing of wh-extractions from the main RC structure is more difficult than processing of wh-extractions from that-clause structure (e.g., Kluender & Kutas, 1993).
To answer the fourth question in this study, my conclusion is based on the results of native speakers rather than the results of learners who could have been misled by the semantic cues of d-linked wh-phrases at their initial processing of wh-dependencies.To conclude, then, and in answer to the fourth question, the d-linking effect on wh-extractions from wh-islands is more likely caused by the d-linked wh-phrase's denotation of discrete individuals that can facilitate the semantic operations necessary for the interpretation of weak islands, as claimed by the semantic account of d-linking effect.However, one must be cautious about generalizing the results of the d-linking effect in this study because only one type of d-linked wh-phrase (which N) was tested, and other types of d-linked wh-phrases (e.g., what N or which of the N) could also be tested.

Conclusion
This study makes two important contributions to the L2 literature on island sensitivity and d-linking effect on wh-extractions.First, this study provides further evidence that island sensitivity is similar in nature in L2 learners and native speakers.Najdi learners patterned similarly to English natives in terms of the strength of their sensitivity to strong versus weak islands, suggesting that Belikova and White's 2009 proposal is on the right track.Second, this study provides evidence that the d-linking effect on wh-extractions from islands is also similar in nature in L2 learners and native speakers.Consistent with previous research on dlinking, both natives and learners exhibited a greater d-linking effect on wh-extractions from wh-islands (weak islands) than on wh-extractions from RC islands (strong islands), and this d-linking effect is more likely caused by the d-linked wh-phrase's denotation of discrete individuals as claimed by the semantic account of the d-linking effect (Szabolcsi & Zwarts, 1993, 1997).

11
Although the semantic information of d-linked wh-phrases led to an increase in acceptability of whextractions from RC islands, these wh-extractions were still rated very low (M = −0.80)compared to grammatical control wh-extractions (M = 0.26).Interestingly, no one could argue that learners' distinction between grammatical and ungrammatical wh-extractions in this study is driven by semantic rather than syntactic cues because reliance on semantic cues cannot help learners to correctly reject ungrammatical wh-extractions from islands, as explained in example (23).

Bio
Saad M. Aldosari is an Assistant Professor of Linguistics at the College of Languages and Translation, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia.He completed his MA and PhD in Linguistics at the University of Kansas, USA.His research examines the acquisition and processing of syntax by second language learners, focusing on the cognitive and linguistic factors that impact development and ultimate attainment.
does he wonder [why she might hate __]? b. *Which movie does he wonder [why she might hate __]?
__ i ʔal-muħaamii fii-l-maktab]? which briefcase worry.2ms[if forgot.3msthe-lawyer at-the-office] 'Which briefcase i do you worry [if the lawyer forgot__ i at the office]?' does he wonder [why she might hate __]? b. *Which movie does he wonder [why she might hate __]?
Who did Albert learn whether they dismissed __ after the annual performance review?WHICH CONDITION b. *Which employee did Albert learn whether they dismissed __ after the annual performance review?BASELINE CONDITION NP ISLAND a. *What / *Which of the cars do you believe the claim that he might buy ___?UNGRAMMATICAL / WH-ISLAND b. *What / *Which of the cars do you wonder who might buy ___?GRAMMATICAL / NON-ISLAND c.What / Which of the cars do you believe that he might buy ___?
What/*What movie/*Which movie/*Which of the movies does Jean wonder whether they will watch ___ at the cinema?6 a. *What does he wonder why she might hate ___?ISLAND/BARE b. *Which movie does he wonder why she might hate ___?ISLAND/D-LINKED c.What does he think that she might hate ___?NON-ISLAND/BARE d.Which movie does he think that she might hate ___?NON-ISLAND/D-LINKED What did the author who wrote ___win the prize?ISLAND/BARE b. *Which article did the author who wrote ___win the prize?ISLAND/D-LINKED c.What did the author who wrote the article win ___?NON-ISLAND/BARE d.Which prize did the author who wrote the article win ___?NON-ISLAND/D-LINKED
i did she wonder [where he found __ i ]?

What i did [ IP she wonder [ CP where [ IP he found __ i ]]]?
), the wh-questions are ungrammatical because the whphrase "what" crosses more than one bounding node, DP, or IP in each movement.

/What movie/Which movie/Which of the movies will
, p. 18. they watch ___ at the cinema?
THAT-CLAUSE EXTRACTION c. What/

What movie/Which movie/Which of the movies
Alexopoulou and Keller (2013)ch ___ at the cinema?Alexopoulou and Keller (2013)tested native speakers of English who exhibited a d-linking effect on wh-extractions from whether islands (16a), with which N condition (e.g., "which movie"), being rated higher than what condition (e.g., "what").They attributed this d-linking effect to the critical property of the distinction between kind denoting wh-phrases (e.g., "what") and individual denoting wh-phrases (e.g., "which movie").They argued that d-linked wh-phrases facilitate processing of wh-extractions from weak islands because they denote discrete individuals that can participate in the semantic Boolean operations necessary for the interpretation of weak islands as proposed by