Processing relative clauses in Chinese as a second language

Second Language Research http://slr.sagepub.com/ Processing relative clauses in Chinese as a second language Yi Xu Second Language Research 2014 30: 439 originally published online 8 July 2014 DOI: 10.1177/0267658313511485 The online version of this article can be found at: http://slr.sagepub.com/content/30/4/439 Published by: http://www.sagepublications.com Additional services and information for Second Language Research can be found at: Email Alerts: http://slr.sagepub.com/cgi/alerts Subscriptions: http://slr.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav >> Version of Record - Sep 23, 2014 OnlineFirst Version of Record - Jul 8, 2014 What is This? Downloaded from slr.sagepub.com by guest on September 23, 2014 511485 SLR0010.1177/0267658313511485Second Language ResearchXu research-article2014 second language research Article Processing relative clauses in Chinese as a second language Second Language Research 2014, Vol. 30(4) 439–461 © The Author(s) 2014 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0267658313511485 slr.sagepub.com Yi Xu University of Pittsburgh, USA Abstract This project investigates second language (L2) learners’ processing of four types of Chinese relative clauses crossing extraction types and demonstrative-classifier (DCl) positions. Using a word order judgment task with a whole-sentence reading technique, the study also discusses how psycholinguistic theories bear explanatory power in L2 data. An overall preference for DClfirst structures and an advantage of DCl-subject relative clauses over the other three structures were found. Results were largely compatible with the filler-gap domain theory and indicated a weak subject-gap advantage. These motivations are subject to influences from other factors, and a multi-constraint proposal was proposed. Keywords canonical word order, Chinese relative clauses, demonstrative-classifier, filler-gap, linear distance, perspective shift, psycholinguistic theories, second language processing, structural distance, subject preference I Introduction Relative clauses (RCs) have always been of great interest to scholars in the psycholinguistic and language acquisition field. In languages with head-initial RCs, psycholinguistic studies consistently report that subject-extraction relative clauses (SRs) are easier than object relative clauses (ORs) when both argument nouns are animate (e.g. Ford, 1983; Frazier, 1987; Holmes and O’Regan, 1981). Acquisition studies on head-initial RCs using a variety of tasks, including aural comprehension, elicited production, Corresponding author: Yi Xu, Department of East Asian Languages and Literatures, University of Pittsburgh, 3942 O’Hara Street, 113C OEH, Pittsburgh, PA 15260, USA. Email: xuyi@pitt.edu Downloaded from slr.sagepub.com by guest on September 23, 2014 440 Second Language Research 30(4) acceptability judgment, longitudinal data, etc., also find that SRs are easier for both first language (L1) children and second language (L2) adults (e.g. Doughty, 1991; Eckman et al., 1988; Gass, 1979; Keenan and Hawkins, 1987). Recently, much attention has been directed to the acquisition of RCs in East Asian languages, but the asymmetry appears to be much less consistent in those head-final structures (e.g. Jeon and Kim, 2007; Ozeki and Shirai, 2007; Yip and Matthews, 2007). Chinese RCs are unique in its rare combination of the head-final structure and its basic SVO word order in simple sentences. Further, Chinese being a classifier language, relativization structures can be either preceded or followed by demonstrative-classifier (DCl) strings. These features of Chinese might allow us to disentangle the effects from different psycholinguistic motivations that are thought to cause the SR/OR asymmetry in the English type of RCs. The present study investigates the complexity of different types of RCs for learners of Chinese as a foreign language (CFL) and discusses the explanatory powers of psycholinguistic motivations in L2 data. II Background In L1 psycholinguistic studies on Chinese RCs, conflicting results are reported: Kuo (2006), Li et al. (2010), and Lin and Bever (2006), and Vasishth et al. (2013) found an SR preference, while Chen et al. (2008), Gibson and Wu (2013), and Hsiao and Gibson (2003) claimed that ORs were easier. Several L1 acquisition studies suggested that SRs were acquired better (e.g. Cheng, 1995; Hsu, Hermon, and Zukowski, 2009), but quantitative results were not robust in some other research (e.g. Su, 2004). The interaction of DCl position with the extraction type (SR versus OR) recently caught researchers’ attention. DCl-modified RCs in Chinese can have a DCl-first sequence, exemplified by (1a) and (1c) respectively for SR and OR, or can have a DClsecond sequence, as in (1b) and (1d). These four structures are referred to (a), (b), (c), and (d) in the article; de is the RC marker/Complementizer in Chinese (e.g. Ning 1993), and Cl stands for classifiers. (1) a. DCl + SR + N (DCl-SR) [NP zhe ge [NP[S _ i xihuan Xiao Lin de] reni]] hen shuai this Cl like Xiao Lin DE person very handsome ‘The person that likes Xiao Lin is very handsome.’ b. SR + DCl + N (SR-DCl) [NP[S __i xihuan Xiao Lin de] [NP zhe ge ren]i] hen shuai like Xiao Lin DE this Cl person very handsome ‘The person that likes Xiao Lin is very handsome.’ c. DCl + OR + N (DCl-OR) [NP zhe ge [NP[S Xiao Lin xihuan __i de] reni]] hen this Cl Xiao Lin like DE person very ‘The person that Xiao Lin likes is very handsome.’ shuai handsome Downloaded from slr.sagepub.com by guest on September 23, 2014 441 Xu d. OR + DCl + N (OR-DCl) [NP[S Xiao Lin xihuan __i de] [NP zhe ge ren]i ] hen shuai Xiao Lin like DE this Cl person very handsome ‘The person that Xiao Lin likes is very handsome.’ The difference between a DCl-first and DCl-second sequence is sometimes considered as a difference between restrictive and non-restrictive RCs in Chinese (e.g. Chao, 1968), while no consensus has been reached. It is, however, well acknowledged that RC modifies the N in (a) and (c), while it modifies the complete DCl-N in (b) and (d), where the demonstrative is anaphoric (Huang et al., 2009: 217–18). In L1 studies, Wu et al. (2009) reported that speakers read the main clause after the complex noun faster in the DCl-SR condition than in the DCl-absent condition. Wu et al. (2006) also showed that the classifier in the DCl-OR sequence, when not matched with the noun immediately afterwards, caused a ‘lexical disruption’ or a ‘semantic clash’ that slowed down reading. There were only a few existing L2 studies on Chinese RCs. Packard (2008) conducted a self-paced reading task and claimed that SRs were read more slowly than ORs. But animacy (e.g. Traxler et al., 2002) appeared to be a confounding factor in Packard’s study. In Chen’s (1999) grammaticality acceptance and word ordering tasks, she suggested that SRs were easier than ORs in a DCl-first sequence, and the pattern was reversed in a DCl-second sequence. However, Chen did not clearly state her statistical methods. Xu (2013a) reported a listening comprehension task. For DCl-absent RCs with two animate NPs, SRs had a higher accuracy rate than ORs, but for RCs followed by a DCl, no asymmetry between the two extraction types were found. III Processing theories and DCl-modified Chinese RCs As existing L1 and L2 studies produced mixed results, the present research aims to follow the tradition to investigate L2 Chinese learners’ RC processing. In addition to making observations, I will discuss potential motivations responsible for the observed effects. Specifically, the article addresses the following questions: •• Is there an asymmetry between L2 learners’ processing of Chinese SRs and ORs? •• How does DCl position affect Chinese L2 learners’ RC processing? •• Among the four structures in (a–d), is there a clear ranking of difficulty for learners? •• Considering the existing RC processing theories, can the observed L2 data be explained by motivations that purportedly underlie L1 patterns? To address the last question, I take the position that L1 and L2 processing rely on fundamentally the same mechanisms unless proven otherwise. Whereas native and nonnative speaker differences are well documented, Dekydtspotter et al. (2006) and Dekydtspotter et al. (2008) argued against Clahsen and Felser’s (2006) Shallow Structure Hypothesis and showed evidence that L1 and L2 parsing deploys similar principles, and observed differences are caused only by a general slowness in L2 processing. On the Downloaded from slr.sagepub.com by guest on September 23, 2014 442 Second Language Research 30(4) Table 1. Theories tested in L1/L2 Chinese relative clause (RC) studies. Study Nature Theories tested Theories supported Hsiao and Gibson (2003) Kuo (2006) L1 processing Storage cost Gibson and Wu (2013) L1 processing Hsu et al. (2009) Packard (2008) L1 acquisition L2 processing Storage cost; LDT; CWO; AH; PS AH; PS; LDT and storage cost; Experience-based account (CWO and RC type frequency) LDT; Experience-based account CWO; SDT; LDT AH; CWO; LDT L1 processing RC type frequency; AH LDT SDT LDT; CWO Notes. AH = Accessibility Hierarchy; CWO = Canonical Word Order; LDT = Linear Distance Theory; PS = Perspective Shift. topic of RC processing in particular, O’Grady et al. (2003) argued that L1 and L2 learners’ RC processing involves essentially similar operations and representations. Thus, it is meaningful to assess to what extent L2 behavior might conform to predictions made by psycholinguistic hypotheses. Five major theories are evaluated. Among them, the filler-gap domain (Hawkins, 2004), the structural distance (O’Grady, 1997, 1999), and the linear distance theory (Gibson, 1998, 2000) are representative of the working memory accounts.1 The canonical word order (Bever, 1970; Slobin and Bever, 1982) considers the speakers’ experience with word order in simple sentences. A different kind of experience-based account considers the RC structure type frequency in corpus and is based on the notion of expectation and probability (Hale, 2001; Levy, 2008). I will discuss why the RC-type frequency account does not directly apply to the current L2 data in Section V. According to Gibson and Wu (2013), in addition to these major proposals, the perspective shift (MacWhinney, 1977, 1982) constitutes another theory from the pragmatic perspective. It is important to note that the above five theories are not necessarily mutually exclusive, and structural complexity and acquisition difficulty could be the result of several factors (Hawkins, 2007; Lee-Ellis, 2011; O’Grady, 2011). These theories were tested in some recent studies on Chinese RCs, and results are summarized in Table 1. 1 Structural Distance Theory (SDT)2 O’Grady (1997: 136) proposes that the complexity in RCs be measured by the number of XP categories ‘between a gap and the element with which it is associated’. SRs are always easier than ORs because in SRs, e.g. (2a), only the S category intervenes between the gap and the head noun, whereas in ORs, e.g. (2b), both the VP and the S intervene between the gap and the filler. (2) a. [NP The personi [S that __i b. [NP The personi [S that Xiao Lin [VP likes Xiao Lin]]] [VP likes ___i]]] Downloaded from slr.sagepub.com by guest on September 23, 2014 443 Xu The theory predicates that Chinese SRs are easier than ORs, just like English, because regardless of head direction, the object gap is always under VP, and is thus more deeply embedded than the subject gap, which is outside VP. The DCl modification does not affect the prediction. In (3a) and (3b), I show the diagrams for the structure (a) and (b). The diagrams are abbreviated according to the original SDT illustration (O’Grady, 1997: 136).3 (3) a. NP DCl NP S [–NP] gapi Ni VP V NPO NO zhe ge this Cl ___ xihuan Xiao Lin (de) like Xiao Lin de ren person b. NP NPi S [–NP] gapi VP V DCl N NPO NO ___ xihuan Xiao Lin(de) zhe ge ren like Xiao Lin de this Cl person In (3a), DCl-SR, the only XP that has a [–NP] feature is the S. There is no NP missing in the lower VP, and the [–NP] feature is checked off when S merges with the RC head. In (3b), SR-DCl, the only XP category that has a [–NP] feature is also the S, and then the [–NP] is matched with the filler ‘this-Cl person’ when S merges with the RC head/NPi. In other words, the only thing that matters in the SDT is the filler-gap structural distance. The internal structure of the coindexed element – i.e. whether it is an N or a DCl-N – does not affect its predictions. Thus, the theory predicts an SR advantage in Chinese, with the ranking of the four structures from ‘easier’ to ‘more difficult’ to be (a)/(b)>(c)/(d). Downloaded from slr.sagepub.com by guest on September 23, 2014 444 2 Second Language Research 30(4) Filler-gap domain (FGD) Hawkins (2004: 175) proposed that the processing difficulty in relativization is determined by the filler-gap domain, defined as ‘the smallest set of terminal and non-terminal nodes dominated by the mother of a filler and on a connected path that must be accessed for gap identification and processing.’ The number of nodes required for (English) SRs is five, including V, VP, S, NS, and NPS, and that the number for an OR is seven, including V, VP, NS, NPS, S, N, and NP (Hawkins, 1999: 255). That is, ORs should have NS and NPS nodes in their FGDs, but not vice versa, since objects are asymmetrically dependent on the presence of subjects. While that results in an SR preference in most languages, the rare SVO head-final structure makes the FGD in SR and OR the same in Chinese, if the verb is transitive (Hawkins, 2004: 180). That is because the FGD will go beyond the minimal nodes specified above when an RC contains optional elements in the path from the filler to V in SR and OR. For instance, in (4a–b), Xiao Lin is in the path from xihuan ‘like’ to ren ‘person’. Thus, both DCl-absent SRs and ORs (with transitive verbs) in Chinese involve seven nodes, including V, VP, S, N, NP, plus NO and NPO in SR, or plus NS and NPS inj OR. The comparison (4a) and (4c) can illustrate the point. (4) a. DCl-SR NP DCl zhe-ge NP S Ni gapi VP V NPO NO . zhe ge __ this Cl xihuan Xiao Lin(de) ren like Xiao Lin de person b. SR-DCl NP S gapi NPi VP V DCl N NPO NO __ xihuan Xiao Lin (de) zhe ge like Xiao Lin de this Cl ren person Downloaded from slr.sagepub.com by guest on September 23, 2014 445 Xu c. DCl-OR NP NP DCl S NPS Ni VP NS V gapi zhe-ge Xiao Lin xihuan __ (de) ren this-Cl Xiao Lin like de person’ In (4a) and (4c), DCl-first SR and OR, the FGD has the same size, since in (4a) NPO and NO intervene between the link from V to the head N; in (4c), NPS and NS are needed because an object RC obligatorily requires a subject. The circle indicates the size of the FGD. In (4a) and (4c), the coindexation or the filler-gap relation is configured completely within the circle. Thus, for a DCl-first sequence, DCl is not needed for gap identification.4 The FGD predicts that DCl-second RCs are more difficult. In (b) SR-DCl, illustrated above as (4b), DCl is embedded within the NP that is coindexed with the gap. Thus, for one to correctly interpret the gap’s content, the DCl elements must be accessed in processing. The FGD in SR-DCl therefore includes the complete complex NP in (4b). Similarly, for gap identification in OR-DCl, DCl needs to be parsed, and structure (d) would also have a bigger FGD. While the number of nodes within the FGD shown in (4b) is nine, the DCl-second structures will potentially be more complex, as only an abbreviated DCl-N structure is shown above. Thus, the predictions of the FGD regarding the four structures are (a)/(c)>(b)/(d). 3 Linear Distance Theory (LDT) The integration distance proposed in Gibson (1998, 2000) is construed linearly. In this theory, in DCl-absent RCs, ORs are more favorable than SRs in Chinese because in linear ordering there are no intervening elements except for the RC marker de in ORs, while V and NO intervene in SRs. When it comes to DCl-modified RCs, the integration is not limited to filler-gap relations, but includes that of head-dependent relations in general (Gibson, 1998, 2000). Thus, one must consider both the filler-gap coindexation and the association between DCl and N. These two types of integration distance in (a–d) are as follows: (5) a. DCl j [CP __i V NO de] NS i/j: a long-distance filler-gap integration plus a long-distance DCl-N association; b. [CP __ i V NO de] [DCl NS]i: a long-distance filler-gap integration; c. DCl j [CP NS V __i de] NO i/j: a short-distance filler-gap integration plus a long distance DCl-N association; d. [CP NS V __i de] [DCl NO]i: one short filler-gap integration. Downloaded from slr.sagepub.com by guest on September 23, 2014 446 Second Language Research 30(4) Multiple integrations taking place at the same time is particularly costly (Gibson, 2000). Thus, (a) is the most complex structure and (d) the simplest. The prediction of the LDT is thus: (d)>(b)/(c)>(a). 4 Canonical Word Order (CWO) Bever (1970) and Slobin and Bever (1982) suggested that a sequence corresponding to the unmarked word orders in simple sentences in that language is processed more easily. The CWO predicts that (6d) is the easiest, since the word order corresponds best to simple Chinese active sentences that have a NVN form. (6) a. b. c. d. DClS V NO (de) NS (DCl-SR) V NO (de) DClS NS (SR-DCl) DClO NS V (de) NO (DCl-OR) NS V (de) DClO NO (OR-DCl) None of the other structures can be considered as having an internal unit corresponding to dominant word orders in Chinese. In (6a), the head NS is separate from the DCl that modifies it by VNO. In (b), an initial VNO sequence is only permitted under a restricted context that allows pro-drop. For (c), assuming the NS is a proper noun (e.g. Xiao Lin, as in our example) or a personal pronoun, DCl-NS is generally forbidden unless for the rhetorical effect of emphasis. Although the SVO order in Chinese is less consistent than that in English, the CWO theory allegedly works on languages that allow word order flexibilities (Slobin and Bever, 1982). Also, the definition of the ‘canonical form’ is that they ‘are not the most frequent’, but they require ‘the least […] implicit discourse presuppositions’ or the listeners’ ‘background knowledge’ (Slobin and Bever, 1982: 231). In Chinese, only the SVO word order is basic, while all other potential orders are ‘semantically and pragmatically marked in special ways’ (Li et al., 1993: 172–73). Thus, the prediction of the CWO is (d)>(a)/(b)/(c). 5 Perspective Shift (PS) Perspective is the ‘starting point’ in the production and the comprehension process of a sentence (MacWhinney, 1977: 152), and a reader’s perspective is oriented to the sentential subject by default. In RC comprehension, the perspective shifts depending on whether the subject of the RC is the same as the subject of the main clause. In the Chinese structure of (a–b), the perspective remains constant since the subject of the RC, zhe ge ren ‘this person (who likes Xiao Lin)’, is also the subject of the main clause. In (c), the perspective starts with the DCl that modifies the head N, zhe ge (ren) ‘this Cl (person)’, and then shifts to Xiao Lin, the subject of the RC, and shifts back to the perspective of ‘the person’ when the main clause is processed. In (d), there is one perspective shift, as the perspective starts with Xiao Lin and then shifts to ‘this person’. Examples (7a–b) shows the perspective changes in (c) and (d) structures. Downloaded from slr.sagepub.com by guest on September 23, 2014 447 Xu (7) a. zhe ge this Cl 1st perspective Xiao Lin Xiao Lin 2nd perspective b. Xiao Lin xihuan de Xiao Lin like de 1st perspective xihuan de ren hen shuai like de person very handsome back to the 1st perspective zhe ge ren this Cl person 2nd perspective hen very shuai handsome In addition to the above fundamental differences, a pre-RC DCl can have a facilitating ‘cue marking’ effect in (a) and (c) because grammatical cues that signal delays in referential commitment help perspective tracking (MacWhinney, 2008: 381). In both (a) and (c), the initial DCl acts as a deictic perspective that is later on cataphorically fleshed out a bit more with ren ‘person’ (MacWhinney, 2012, personal communication). Thus, in the comparison of (a) and (b), the DCl in (a) establishes a starting point, while the initial verb in (b) cannot be attached to any existing reference, involving a little more representational load. The same initial deictic perspective helps the processing of (c) too, but because the deictic perspective reference is a weaker effect than the basic perspective shifting effect, (c) is still the most difficult among all. The overall prediction of the PS is (a)>(b)>(d)>(c). IV 1 Methodology Participant and materials Below, I report an experiment in which L2 Chinese participants with an English L1 background completed a self-paced word-order judgment task. Forty-nine adult CFL learners in their third semester at the Defense Language Institute (DLI) participated in the experiment, and data from 32 of them were used. (For justification, see Section V.1). Because the DLI regularly conducts qualifying exams, and students exiting the third semester generally receive a score between 2 and 2+ on the Interagency Roundtable Language scale upon graduation, participants had approximately the same level of proficiency. A counterbalanced within-participant design was used. Forty-eight quadruplets of sentences (in correct word order) were constructed, typed in simplified Chinese characters. Each list had 12 sentences in each of the four conditions in (1a–d). In order for the DCl to be correctly associated with the head noun and not the other NP in the RC, the other NP in the RC is always a person’s name or a personal pronoun. (All personal names used in items are specified in the instruction sheet given to the participants before the experiment.) All RCs modify the subject in the main clause, the predicate of which does not contain embedded complex structures. Plausibility norming surveys modeling after Hsiao and Gibson (2003) were conducted to ensure that these sentences describe events that are equally natural in the real world. All items were matched for their plausibility in the SR and OR versions. Each list also contains 48 ungrammatical sentences with incorrect word orders (which if ordered correctly would contain an RC). To prevent a situation in which participants correctly reject those ungrammatical sentences based on readings of the initial one or two words, word sequences encourage participants to read on. Representative items of incorrect word orders are shown in (8a–c) in ‘pseudo-Chinese’. VN, NV, and DCl-N could all be potentially grammatical initial strings. Downloaded from slr.sagepub.com by guest on September 23, 2014 448 Second Language Research 30(4) (8) a. b. c. Support Lao Wang people very DCl de excited. Xiao Lin like very much play games de neighbor invite. DCl person recommend de Xiao Zhang our teacher is. In addition, there are 24 filler sentences, 12 with a correct word order and 12 with an incorrect word order. The stimuli were randomized for each participant by the experiment software during the experiment; for the experiment items, see Appendix 1. 2 Procedure Because Chinese RCs are complex for L2 learners, with previous studies reporting challenges of collecting usable data (Dai, 2010), a whole sentence reading technique instead of a word-by-word reading technique was used to avoid causing excessive burdens on participants. Previous L2 studies have used grammaticality judgment with a wholesentence reading technique (e.g. Juffs and Harrington, 1995). In the present study, ungrammaticality is always induced by changes in the correct word order, and this particular technique has been used in earlier psycholinguistic studies since Forster and Olbrei (1973). Because the only way for a participant to make a correct judgment is to comprehend the sentence, reaction time should reflect the total processing time required for that sentence. Specifically to validate the sensitivity of this task, Nicol et al. (1997) included English SRs and ORs as part of their experiment material and replicated the well-acknowledged SR/OR asymmetry in English in their reaction time (RT) and error rate data using this technique. DMDX software (Forster and Forster, 2003) was used in this experiment. Participants completed the task on PC laptops. Instructions were given in English on an instruction sheet, and brief instructions in Chinese were shown on the computer screen before the experiment started. In the main experiment, a string of words was presented in full on the computer screen, as in normal text. Participants determined whether a sequence contained the correct word order to form a grammatical sentence, and pressed a key to indicate their decision. Participants were instructed to read as fast as they could and to make a judgment as accurately as they could. They had 12 seconds to make a decision before the program moved on to the next item. While participants’ instructors confirmed that all words used in the experiment had been taught in classes, a word recognition survey was conducted immediately after the main task, so that participants could report their familiarity with the words used. A hundred and eight essential vocabulary items used in the main task were included. There were also five pseudo-Chinese words included in the list to ensure that the participants read the words for meaning. The participants were asked to circle the words that they did not know. V Results and discussion 1 Data exclusion criteria In the word recognition task, most participants circled all five pseudo-words to indicate that they did not know them, confirming that they paid close attention to the list and were reporting their vocabulary knowledge in a faithful way. The following participants’ data Downloaded from slr.sagepub.com by guest on September 23, 2014 449 Xu Table 2. Mean RT and error rate in different conditions. Sentence type (a) DCl-SR (b) SR-DCl (c) DCl-OR (d) OR-DCl RT mean in milliseconds (SD) Error rate mean (SD) 48-item 32-item 48-item 32-item 5956 (904) 6101 (1051) 6227 (1083) 6215 (857) 5758 (1030) 6032 (1206) 6093 (1182) 6187 (996) 0.14 (0.11) 0.16 (0.12) 0.10 (0.09) 0.13 (0.12) 0.11 (0.11) 0.18 (0.15) 0.09 (0.10) 0.12 (0.15) Notes. OR = object relative clauses; DCl = demonstrative-classifier; SR = subject-extraction relative clauses. were excluded: participants whose experiment sessions were interrupted, participants who reported L1s other than English in a language background survey, participants who reported more than 20 unknown words, and participants who had more than 23% judgment errors on all items in the main task. To balance the number of participants per group, one participant who reported the largest number of unknown words in his group was also excluded. After applying these exclusions, data from a total of 32 participants were analysed. For those participants, the mean number of words indicated to be known was 101 (SD = 3.6). In addition to a 48-item analysis, another round of analysis was conducted based on 32 items, with items containing words reported by six or more participants to be ‘unknown’ taken out, thus increasing the mean number of known words to 106.9 (SD = 1.2). The cutoff point was set to guarantee that the participants knew almost all the words while leaving enough items to analyse. Since unfamiliar vocabulary could influence results, the 32-item analysis reflects learners’ processing difficulty more accurately. 2 Two-way ANOVA results Descriptive statistics for each condition in both the 32-item and 48-item analyses are reported in Table 2. Two-way ANOVA was performed to answer the first two research questions: the effect of extraction and DCl position as well as their potential interactions. I report results in the 32-item analysis first. a RT. In the by-participant analysis, the DCl and extraction interaction was not significant (F(1,31) = 1.14, p = 0.29). DCl as a main factor reached marginal significance (F(1,31) = 2.88, p = 0.10), and extraction as a main factor was significant (F(1,31) = 6.80, p = 0.01). The by-item analysis did not reveal significance in interaction (F(1,31) = 0.72, p = 0.40) and did not reveal any significant main effect (DCl factor: F(1,31) = 1.93, p = 0.18; extraction factor: F(1,31) = 1.42, p = 0.24). b Error rate. In the by-participant analysis, the DCl and extraction interaction was not significant (F(1,31) = 0.69, p = 0.41). DCl as a main factor was significant (F(1,31) = 4.46, p = 0.04), and extraction as a main factor was significant (F(1,31) = 4.22, p = 0.05). By-item analysis revealed no significant interaction (F(1,31) = 0.79; p = 0.38). DCl was Downloaded from slr.sagepub.com by guest on September 23, 2014 450 Second Language Research 30(4) a significant factor (F(1,31) = 6.86; p = 0.01) and extraction was also a significant factor (F(1,31) = 4.78, p = 0.04). In the 48-item analysis, there were no significant interactions in either RT or error rate. Extraction was a significant factor in RT (F(1,31) = 7.14, p = 0.01), and extraction was also close to significance in error rate (F(1,31) = 3.70, p = 0.06). There were no other significant results. 3 Preliminary discussion For the DCl factor, DCl-first RCs were read faster than DCl-second RCs with marginal significance, and there were also fewer errors associated with DCl-first RCs. These patterns conform to the FGD predictions. For the extraction factor, RTs and error rates point to preferences in different directions: whereas SRs were read faster (in both the 32- and 48-item analyses), there were also more errors associated with SRs. While an interpretation cannot be easily drawn here, pairwise comparisons below indicate that the higher error rate might result from the difficulty in (b) alone instead of from SR structures in general. The RT results thus offer tentative evidence for an SR preference. The LDT predicts DCl-first RCs to be harder than DCl-second RCs, and that SR extraction to be harder than OR. These predictions were not congruent with the results. 4 Pairwise comparisons A one-way ANOVA was conducted to test for any differences among the four condition means; if the one-way ANOVA indicated the presence of a difference among the means, a Tukey’s honestly significant difference (HSD) test was conducted to assess which means differ from each other. The Tukey’s test holds the overall alpha for these six comparisons to the selected alpha (Howell, 2002: 399) and is a recommended test for pairwise comparisons of more than three groups of equal size (Cohen and Lea, 2004: 117). The Tukey’s HSD test is conducted by calculating the HSD statistic from this formula: HSD = critical q − value Mean square error ; Number of observations per group any difference in means greater than HSD is considered statistically significant at the level used to determine the critical q-value. a RT. In the 32-item by-participant analysis, a one-way ANOVA revealed significant differences among the conditions (F(3,93) = 3.68, p = 0.02). For pairwise comparisons, the Tukey’s HSD test indicated that the difference between condition (a) and condition (c) was close to significance (p < 0.10) and that the difference between (a) and (d) was significant (p < 0.05). In the 48-item analysis, a one-way ANOVA indicated significant differences among the four conditions (F(3,93) = 2.77, p = 0.05). In the Tukey’s HSD test, differences between (a) and (c) and between (a) and (d) were found to be marginally significant (p < Downloaded from slr.sagepub.com by guest on September 23, 2014 451 Xu Table 3. Significant and near significant differences in pairwise comparisons in the 32-item analysis. Pairwise comparisons Reaction time by-participant (a) > (b) (a) > (c) (a) > (d) (c) > (b) Error rate by-item by-participant by-item † † * * * Note. * indicates significance at a p < 0.05 level; † indicates marginal significance where 0.05 < p < 0.1. ‘>‘ is interpreted as ‘has preference over’, reflected by faster reaction time or lower error rate. 0.10). The differences among the four conditions did not reach significance in the byitem analysis (32 items: F(3,93) = 1.35, p = 0.26; 48 items: F(3,141) = 0.96, p = 0.41). b Error rate. The by-participant analysis indicated differences among the four conditions (F(3,93) = 3.40, p = 0.02), and in the Tukey’s test the difference between (b) and (c) was significant (p < 0.05). The by-item analysis also showed differences among conditions (F(3,93) = 3.56, p = 0.02), and the Tukey’s test revealed a near significant difference between (a) and (b) (p < 0.10) and a significant difference between (b) and (c) (p < 0.05). With the 48 items, the differences among the four conditions did not reach significance either in the by-participant or by-item analysis. Table 3 summarizes results that were statistically significant or near significant in the 32-item analysis: Condition (a) was more preferable than the other three conditions, reflected by lower error rate in (a) versus (b) comparisons, and by faster RT in (a) versus (c) and (a) versus (d) comparisons; (c) also had lower error rate than (b). 5 Further discussion a DCl position and the FGD. The aforementioned two-way ANOVA analysis revealed a DCl-first advantage, consistent with the FGD, and in pairwise comparisons, the DCl-first advantage was maintained in SRs. The DCl-SR > SR-DCl preference conforms to frequency patterns in the L1 corpora (Ming and Chen, 2010; Wu et al., 2009), but it is unlikely that the pattern observed in this study is driven by RC-type distributional frequencies in one’s exposure, because statistical information in naturalistic L1 production cannot be a reliable indicator of L2 learners’ language experience. Table 4 summarizes the tokens reported in the aforementioned corpus studies and compares them to tokens found in L2 participants’ textbooks and grammar books in units dealing specifically with relativization structures. First, the corpus frequencies exhibited a ranking of (a)>{(b)/(d)}>(c), and that does not correspond to L2 participants’ learning materials. Second, DCl-modified RCs are generally very rare (Wu et al., 2009), and L2 participants in a CFL program were unlikely to have significant input containing these structures outside their classrooms. Also, a pure frequency-based account could not explain why DCl-OR was not the most difficult structure for participants, despite its Downloaded from slr.sagepub.com by guest on September 23, 2014 452 Second Language Research 30(4) Table 4. Structure frequencies in L1 corpus and L2 textbooks. (a) DCl-SR (b) SR-DCl (c) DCl-OR (d) OR-DCl Wu et al. (2009) from Chinese Tree bank 5.0 Ming and Chen (2010) from the LCMC L2 participants’ textbooks 44 18 5 28 124 38 11 32 3 3 1 7 Note: Tokens above are instances of classifier-modified RCs, which can be a DCl string or a numeral-classifier (NCl) string (Ming and Chen, 2010). Processing mechanisms potentially relevant to DCl-modified RCs affect NCl-modified RCs in the same way. extreme rarity. Scholars have argued that probability or distributional frequency alone cannot adequately account for comprehension without considering where the frequency comes from (Vasishth et al., 2013). Thus, it is more plausible that the corresponding preference for DCl-SR in corpora and in this study resulted from the same underlying motivation, which affects processing and production in a shared framework (e.g. Gennari and MacDonald, 2009). Aside from (a)>(b), other comparisons show an initial DCl advantage. Specifically, (a) had a significantly faster RT than (d), and (c) had a significantly lower error rate than (b). Numerically, (c) was also read faster and had a lower error rate than (d), although differences did not reach significance. These results are in accordance with the FGD theory. As to the lack of more significant comparisons between (c) and (d), an explanation might be provided from the ‘semantic clash’ proposal (Wu et al., 2009). Wu et al.’s reaction time experiment examined L1 speakers’ reading of DCl-ORs such as na-wei jushi zazhong de laoshi (‘that Clhuman boulder hit DE journalist’), in which the classifier was mismatched with its adjacent subject NP within the RC. They found that, at the RC verb position, DCl-ORs were read more slowly than DCl-absent ORs. This slow-down effect was not present in DCl-SRs. Evidence for the same argument was reported in Wu et al.’s (2006) eye-tracking study. Note that DCl-OR is also the rarest type among the four structures in corpus studies in Table 4, indicating inherent complexities potentially associated with mismatching classifiers. In the present experiment in (c), the classifier was always followed by a proper noun (e.g. DCl Xiao Lin), which is generally unacceptable in simple sentences in Chinese. Therefore, a similar semantic clash as reported by Wu et al. can be expected in condition (c). That is, the semantic clash as a competing factor could have weakened the FGD effect in (c) and (d) comparisons. b Extraction type and the SDT. Regarding extraction type, pairwise comparisons offer a possible explanation as to why the RT and error rate appeared to indicate preferences in different directions in the two-way ANOVA analysis: the higher error rate in SRs appears to be largely induced by errors in condition (b) alone, instead of by SR structures in general. Specifically, (a) did not have a higher error rate than (c) or (d). Meanwhile, (b) had the highest error rate, and its differences with (a) or (c) conditions were either significant or near significant. Note that the higher error rate in (b) compared to (a) and (c) can be explained by the FGD. Downloaded from slr.sagepub.com by guest on September 23, 2014 453 Xu Meanwhile, reaction time in the two-way analysis suggests an SR preference over OR, a pattern predicted by the SDT. An SR advantage was also evident in DCl-first structures, shown by faster reaction time in (a) than (c). This pattern agrees with previous L2 studies. For instance, Chen (1999) found that, in DCl-first RCs modifying subject positions, SRs were more easily accepted than ORs. In a sentence completion task reported in Xu (2013b), L1 native speakers and participants were recruited from the same student pool as those in this study (i.e. same proficiency, same institute, but recruited separately); given a DCl-first sequence, both L1 and L2 speakers produced significantly more SRs than ORs. It should be acknowledged that the semantic clash in (c) might have colluded with the SDT in the contrast of (a) versus (c). The SR advantage was not obvious in the comparison of (b) versus (d). Seen from the literature, the lack of a clear asymmetry in SR/OR comparisons when DCl-modification is absent or in DCl-second RCs is generally characteristic of Chinese RCs (e.g. Kuo, 2006; Su, 2004; Xu, 2013b). Some earlier studies, when explaining the mixed SR/OR findings in East Asian languages, suggest that a universal SR advantage could be offset or overridden by word order factors (Lee-Ellis, 2011). In the current experiment, potential involvement from the canonical word order factor would favor (d) instead of any other structures. If similarity to simple Chinese sentences facilitates reading, the overall complexity of (b) and (d) would be balanced, since the structural distance motivation and the word order preference pull in different directions. Although the CWO involvement proposal is hypothetical here, the potential interaction of the CWO with the SDT could explain why findings on SR/OR differences have been mixed in previous Chinese RC (e.g. Lin and Bever, 2006; see also Hsiao and Gibson, 2003) and East Asian RC studies (Lee-Ellis, 2011; Yip and Matthews, 2007). Note that due to word order differences, both the CWO and the SDT favor SRs in head-initial RCs in SVO languages such as English. Overall, there appears to be some weak SR preference, indicated by the RT in the twoway analysis. In DCl-first RCs, it is likely that the increased cost in overcoming the mismatching classifier in DCl-OR have augmented the preference for (a) over (c); in DCl-second RCs, a CWO factor might have weakened the advantage of (b), despite its shorter filler-gap distance than (d). As (a) was apparently easier than other structures, overall results were not consistent with the LDT. The findings also did not confirm the PS theory: although the advantage of (a) over other structures was compatible with the theory, the observed advantage of (c) over (b) ran contrary to its predictions. Several other PS predictions, such as (b)>(d) and (d)>(c), were not borne out. VI Conclusions To answer our research questions, first, DCl-position was a significant factor affecting L2 learners’ Chinese processing, with DCl-first RCs being easier than DCl-second RCs. Second, there appeared to be a weak SR preference in general, although in pairwise comparisons the SR/OR asymmetry was obvious only in a DCl-first sequence, when the asymmetry was potentially amplified by the independent motivation of semantic clash in DCl-OR. Third, DCl-SR was the easiest structure among the four for L2 learners. Finally, among the theories considered, data were most compatible with FGD predictions and Downloaded from slr.sagepub.com by guest on September 23, 2014 454 Second Language Research 30(4) were also partly in agreement with the SDT. Classifier–noun mismatch and a word order factor might be additional motivations that interact with the FGD and the SDT. The above suggests that the overall difficulty in processing these structures could be affected by several factors. Previous studies have proposed that RC processing is multifactorial (Diessel and Tomasello, 2005: 902; Kuo, 2006; Levy, 2008: 1166, etc.). I follow such an approach and suggest that filler-gap domain size and filler-gap structural distance interact with semantic parsing and word order factors in L2 Chinese learners’ RC processing. Costs predicted by the FGD and the SDT are based on hierarchical phrase structures, so findings indicate that L2 speakers project the syntactic structure necessary for accommodating gaps and for achieving gap identification. The current proposal is therefore in line with Dekydtspotter et al.’s (2006, 2008) argument that L2 processing is structure based. The preference for DCl-SR over DCl-OR, which corresponds to both L1 and L2 production patterns (Xu, 2013b) and L1 corpus frequencies, also suggests that learners may obey similar principles in parsing as native speakers. Although the SR advantage in this study is relatively weak, such findings offer preliminary evidence for the important hypothesis that the underlying factor favoring SRs, i.e. the structural distance effect, might indeed be ‘universal’, in that it is not absent in languages such as Chinese. The lack of more robust findings for an SR advantage in this research and in previous studies on Chinese RCs can be explained by the multi-constraint mechanism proposed here. When Hawkins (2007) summarized several previous studies on the L2 acquisition of RCs in different languages, he observed that the SR advantage holds true as a universal principle when all other things are equal. In reality, as Hawkins pointed out, other things are usually not equal. The present study, along with previous research (e.g. Lee-Ellis, 2011), offers tentative evidence for Hawkins’ interpretation that the universal SR advantage is susceptible to influences from other mechanisms. That is, aside from structure-based processing, learners might also use lexical-semantic information and their experience with simple structures in the target language. These factors, being unequal in different languages, could interact with the filler-gap dependency constraint, strengthening the SR advantage in certain contexts, such as in DCl-first structures, and canceling the advantage in other environment, such as in DCl-second structures Only a few studies have investigated the role of DCl position, and the current experiment is the first to produce Chinese data suggesting evidence for the FGD. In addition to predicting increased complexity in RCs followed by DCls, FGD was argued by Hawkins (2004) to be a factor that offers advantage to SRs in most languages including English and other East Asian languages. Since support for the FGD comes from the DCl factor in this study, the theory’s strength in explaining an SR/OR asymmetry in other languages needs to be tested in future studies. If both the FGD and the SDT apply to processing cost, as suggested by the current study, the SR advantage in other languages would be stronger than that in Chinese. The study used psycholinguistic experiment methodologies to investigate L2 learners’ difficulty in Chinese RC processing, an area where research is under-represented. Given that previous studies on Chinese L1 adults and children, as well as on East Asian RCs, have yielded inconclusive findings, these new results offer valuable data for psycholinguistic and acquisition studies on head-final RC structures in general. But several Downloaded from slr.sagepub.com by guest on September 23, 2014 455 Xu limitations of the study need to be addressed. First, although the whole-sentence reading technique was appropriate for participants’ proficiency in this experiment, it does not directly tackle particular regions of difficulty. Future studies using a word-by-word reading technique should consider the potential involvement of some other mechanisms such as storage cost in online comprehension.5 Second, the experiment involved a small number of L2 learners in a homogenous L1 group, and it remains to be determined if the outcome might vary due to proficiency differences and L1 transfer. Finally, the multiconstraint proposal in its current version is preliminary. Working out the specifics of such a proposal, such as investigating different factors’ varying strengths in different contexts, is not trivial, requiring a large amount of empirical evidence (Hawkins, 2007; Levy, 2008). While large-scale studies should be conducted to confirm the tentative conclusions presented here, it is hoped that the current proposal could offer a starting point in the investigation in the L2 processing of Chinese RCs in this promising direction. Acknowledgements I thank the students, instructors and coordinators at the Defense Language Institute for their participation and assistance. I am grateful to Kenneth Forster, Heidi Harley, John A. Hawkins, Fenghsi Liu, Brian MacWhinney, Janet Nicol, and William O’Grady for their helpful communication with me during various stages of the project. I also thank the editors and three anonymous reviewers for their valuable suggestions. I am solely responsible for all errors. Declaration of conflicting interest The author declares that there is no conflict of interest. Funding This research originates from part of my dissertation project, supported by the Language Learning Dissertation Grant from Language Learning, the journal, in 2008, and the Social and Behavioral Science Research Institute Small Grant from the University of Arizona in 2008. Notes 1. 2. 3. Another theory under the working-memory based account is the storage cost theory (Gibson, 2000). This theory is not under consideration here because its predictions need to be tested by examining reading time in specific regions within a sentence, a measure deliberately excluded in this study. The SDT predicts that an SR advantage is universal. In some studies, a ‘universal SR advantage’ hypothesis is referred to as an Accessibility Hierarchy (AH) effect. The AH is originally a typological generalization: Keenan and Comrie (1977) observed that the difficulty of relativizing different positions follows the hierarchy of Subject > Object > Indirect Object > Object of Preposition, whereas > means ‘more accessible’. Although the AH effect is later extended to language acquisition and processing, I use the SDT here because the AH itself does not clearly postulate the causes for a universal SR advantage. Because both the SDT and the FGD afterwards use Simpler Structure terms such as S instead of CP/TP, S is used here. Thus, the syntactic node for de is not shown. The DCl-N structure is simplified here without intermediate nodes, because the internal structure of the DCl-N does not affect the comparative complexity of DCl-modified RCs in either the SDT or FGD. Downloaded from slr.sagepub.com by guest on September 23, 2014 456 4. 5. Second Language Research 30(4) A reviewer pointed out that in (a) and (c), the DCl appear to be further away from the head N. This is true from a linear distance point of view. On a hierarchical structure, the distance between the DCl and the N should also be considered for the overall complexity of the structures, but not as part of the FGD, since the DCl is on a higher structure that does not enter into the filler-gap coindexation relation. Instead, the cost in DCl and head N association should be considered as part of the phrasal combination domain (PCD). In (a) and (c), the initial DCl can be considered as an early constructor for the mother NP, and thus the PCD encompasses all items illustrated in (4a) and (4c). In (b) and (d), the early verb or the subject noun will construct a clause/S, which contains a gap that needs to be co-indexed with a head N for the PCD to be complete. Therefore, the PCD in (b) and (d) also comprise all items dominated by the mother NP. All four structures thus have the same PCD size, while (a) and (c) have a simpler FGD (John A. Hawkins, 2013, personal communication). This study did not directly address the possibility of reanalysis. It is sometimes argued that reanalysis might be involved in structure (d), since it follows a sequence similar to that in simple Chinese sentences and might be initially misread as a main clause (Lin and Bever, 2011). To my knowledge there were no documented theories of reanalysis in (b) and (c). Because the only significant comparisons involving (d) was the advantage of (a) over (d), potential reanalysis in (d) could not explain the overall results. Still, future research using a word-by-word reaction time task should investigate whether reanalysis applies at all. I thank an anonymous reviewer for identifying this issue. References Bever TG (1970) The cognitive basis for linguistic structures. In: Hayes JR (ed.) Cognition and the Development of Language. New York: John Wiley, 279–360. Chao YR (1968) A grammar of spoken Chinese. Berkeley/Los Angeles, CA: University of California Press. Chen B, Ning A, Bi H, and Dunlap S (2008) Chinese subject-relative clauses are more difficult to process than object-relative clauses. Acta Psychologica 129: 61–65. Chen C-Y (1999) The second language acquisition of Chinese relative clauses. The World of Chinese Language 94: 59–76. Cheng SY-Y (1995) The acquisition of relative clauses in Chinese. Unpublished MA thesis, National Taiwan Normal University, Taiwan. Clahsen H and Felser C (2006) Grammatical processing in language learners. Applied Psycholinguistics 27: 3–42. Cohen BH and Lea RB (2004) Essentials of statistics for the social and behavioral sciences: Volume 3. Hoboken, NJ: Wiley. Dai Y [戴运财] (2010) 汉语作为第二语言的关系从句习得难度调查 [An investigation of the relative clause acquisition by learners of Chinese as a second language] 中国海洋大学学报 （社会科学版. Journal of Ocean University of China (Social Sciences) 6: 85–91. Dekydtspotter L, Schwartz BD, and Sprouse RA (2006) The comparative fallacy in L2 processing research. In: O’Brien MG, Shea C, and Archibald J (eds) Proceedings of the 8th Generative Approaches to Second Language Acquisition Conference (GASLA 2006). Somerville, MA: Cascadilla Proceedings Project, 33–40. Dekydtspotter L, Donaldson B, Edmonds AC, Liljestrand Fultz A, and Petrush RA (2008) Syntactic and prosodic computations in the resolution of relative clause attachment ambiguity by English–French learners. Studies in Second Language Acquisition 30: 453–80. Diessel H and Tomasello M (2005) A new look at the acquisition of relative clauses. Language 81: 882–906. Downloaded from slr.sagepub.com by guest on September 23, 2014 457 Xu Doughty C (1991) Second language instruction does make a difference. Studies in Second Language Acquisition 13: 431–69. Eckman FR, Bell L, and Nelson D (1988) On the generalization of relative clause instruction in the acquisition of English as a second language. Applied Linguistics 9: 1–20. Ford M (1983) A method for obtaining measures of local parsing complexity throughout sentences. Journal of Verbal Learning and Verbal Behavior 22: 203–18. Forster KI and Forster JC (2003) DMDX: A windows display program with millisecond accuracy. Behavior Research Methods, Instruments, and Computers 35: 116–24. Forster KI and Olbrei I (1973) Semantic heuristics and syntactic analysis. Cognition 2: 319–47. Frazier L (1987) Syntactic processing evidence from Dutch. Natural Language and Linguistic Theory 5: 519–60. Gass SM (1979) Language transfer and universal grammatical relations. Language Learning 29: 327–44. Gennari SP and MacDonald MC (2009) Linking production and comprehension processes: The case of relative clauses. Cognition 111: 1–23. Gibson E (1998) Linguistic complexity: Locality of syntactic dependencies. Cognition 68: 1–76. Gibson E (2000) The dependency locality theory: A distance-based theory of linguistic complexity. In: Marantz A, Miyashita Y, and O’Neil W (eds) Image, language, brain. Cambridge, MA: MIT Press, 95–126. Gibson E and Wu HI (2013) Processing Chinese relative clauses in context. Language and Cognitive Processes 28: 125–55. Hale J (2001) A probabilistic Earley parser as a psycholinguistic model. In: Proceedings of the Second Meeting of the North American Chapter of the Association of Computational Linguistics: Volume 2. New Brunswick, NJ: Association for Computational Linguistics, 159– 66. Retrieved from http://acl.ldc.upenn.edu/N/N01/N01-1021.pdf (December 2013). Hawkins JA (1999) Processing complexity and filler-gap dependencies across grammars. Language 75: 244–85. Hawkins JA (2004) Efficiency and complexity in grammars. Oxford: Oxford University Press. Hawkins JA (2007) Acquisition of relative clauses in relation to language universals. Studies in Second Language Acquisition 29: 337–44. Holmes VM and O’Regan JK (1981) Eye fixation patterns during the reading of relative clause sentences. Journal of Verbal Learning and Verbal Behavior 20: 417–30. Howell DC (2002) Statistical methods for psychology. 5th edition. Pacific Grove, CA: Wadsworth Publishing. Hsiao F and Gibson E (2003) Processing relative clauses in Chinese. Cognition 90: 3–27. Hsu C-CN, Hermon G, and Zukowski A (2009) Young children’s production of head-final relative clauses: Elicited production data from Chinese children. Journal of East Asian Linguistics 18: 323–60. Huang C-TJ, Li Y-HA, and Li Y (2009) The syntax of Chinese. Cambridge: Cambridge University Press. Jeon KS and Kim H-Y (2007) Development of relativization in Korean as a foreign language. Studies in Second Language Acquisition 29: 253–76. Juffs A and Harrington M (1995) Parsing effects in second language sentence processing: Subject and object asymmetries in ‘wh’-extraction. Studies in Second Language Acquisition 17: 483– 516. Keenan EL and Comrie B (1977) Noun phrase accessibility and universal grammar. Linguistic Inquiry 8: 63–99. Keenan EL and Hawkins S (1987) The psychological validity of the accessibility hierarchy. In: Keenan EL (ed.) Universal grammar: Fifteen essays. London: Routledge, 60–85. Downloaded from slr.sagepub.com by guest on September 23, 2014 458 Second Language Research 30(4) Kuo K (2006) Predictability of Chinese relative clauses. Unpublished diploma thesis, University of Potsdam, Potsdam, Germany. Lee-Ellis S (2011) The elicited production of Korean relative clauses by heritage speakers. Studies in Second Language Acquisition 33: 57–89. Levy R (2008) Expectation-based syntactic comprehension. Cognition 106: 1126–77. Li P, Bates E, and MacWhinney B (1993) Processing a language without inflections: A reaction time study of sentence interpretation in Chinese. Journal of Memory and Language 32: 169–92. Li Q, Zhang J, and Yue W (2010) Chinese relative clauses processing in supportive context removing ambiguity. Studies in Literature and Language 1: 12–19. Lin C-JC and Bever TG (2006) Subject preference in the processing of relative clauses in Chinese. In: Baumer D, Montero D, and Scanlon M (eds) Proceedings of the 25th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, 254–60. Lin C-JC and Bever TG (2011) Garden path and the comprehension of head-final relative clauses. In: Yamashita H, Hirose Y, and Packard JL (ed) Processing and producing head-final structures. studies in theoretical psycholinguistics: Volume 38. New York: Springer, 277–97. MacWhinney B (1977) Starting points. Language 53: 152–68. MacWhinney B (1982) Basic syntactic processes. In: Kuczaj SA (ed.) Language development: Volume 1: Syntax and semantics. Hillsdale, NJ: Lawrence Erlbaum, 73–136. MacWhinney B (2008) How mental models encode embodied linguistic perspectives. In: Klatzky R, MacWhinney B, and Behrmann M (eds) Embodiment, ego-space, and action. Mahwah, NJ: Lawrence Erlbaum, 369–410. Ming T and Chen L (2010) A discourse-pragmatic study of the word order variation in Chinese relative clauses. Journal of Pragmatics 42: 168–89. Nicol JL, Forster KI, and Veres C (1997) Subject-verb agreement processes in comprehension. Journal of Memory and Language 36: 569–87. Ning C (1993) The overt syntax of topicalization and relativizaiton in Chinese. Unpublished PhD thesis, University of California, Irvine, CA, USA. O’Grady W (1997) Syntactic development. Chicago, IL: University of Chicago Press. O’Grady W (1999) Towards a new nativism. Studies in Second Language Acquisition 21: 621–33. O’Grady W (2011) Relative clause: Processing and acquisition. In: Kidd E (ed.) The acquisition of relative clauses: Processing, typology and function. Amsterdam: John Benjamins, 13–38. O’Grady W, Lee M, and Choo M (2003) A subject–object asymmetry in the acquisition of relative clauses in Korean as a second language. Studies in Second Language Acquisition 25: 433–48. Ozeki H and Shirai Y (2007) Does the noun phrase accessibility hierarchy predict the difficulty order in the acquisition of Japanese relative clauses? Studies in Second Language Acquisition 29: 169–96. Packard JL (2008) Relative clause processing in L2 speakers of Mandarin and English. Journal of the Chinese Language Teachers Association 43: 107–46. Slobin DI and Bever TG (1982) Children use canonical sentence schemas: A cross-linguistic study of word order and inflections. Cognition 12: 229–65. Su Y-C (2004) Relatives of Mandarin children [electronic version]. Unpublished paper presented at Generative Approaches to Language Acquisition –North America 2004, University of Hawaii at Manoa, Hawaii, HI, USA. Retrieved from: http://www.ling.nthu.edu.tw/faculty/ ycsu/GALANA2004.pdf (December 2013). Traxler MJ, Morris RK, and Seely RE (2002) Processing subject and object relative clauses: Evidence from eye movements. Journal of Memory and Language 47: 69–90. Vasishth S, Chen Z, Li Q, and Guo G (2013) Processing Chinese relative clauses: Evidence for the subject-relative advantage. PLoS ONE 8(10), e77006. doi:10.1371/journal.pone.0077006. Downloaded from slr.sagepub.com by guest on September 23, 2014 459 Xu Wu F, Haskell T, and Andersen E (2006) The interaction of lexical, syntactic, and discourse factors in on-line Chinese parsing: Evidence from eye-tracking. Unpublished poster presented at the 19th Annual CUNY Conference on Human Sentence Processing, CUNY Graduate School and University Center, NY, USA. Wu F, Kaiser E, and Anderson E (2009) The effect of classifiers in predicting Chinese relative clauses. In: Grosvald M and Soares D (eds) Proceedings of the 38th Western Conference on Linguistics, 330–39. University of California, Davis, CA. Retrieved from: http://www-bcf. usc.edu/~emkaiser/WuKaiserAndersen-2009-WECOL-Classifiers.pdf (December 2013). Xu Y (2013a) Acquisition of Chinese relative clauses at the initial stage. In: Kecskes I (ed.) Research in Chinese as a second language: Trends in applied linguistics series. Boston, MA/ Berlin: DeGruyter Mouton, 217–36. Xu Y (2013b) CFL learners’ production of relative clauses with demonstratives: From theory to empirical research. Chinese as a Second Language Research 2: 169–92. Yip V and Matthews S (2007) Relative clauses in Cantonese–English bilingual children: Typological challenges and processing motivations. Studies in Second Language Acquisition 29: 277–300. Appendix 1 Experiment items The following is a list of experiment items and their English translations. Items marked by * were used in the 32-item analysis. 1 * 2 * 3 4 * 5 * 6 * 7 8 * 9 10 11 * 12 * 这个重视小李的朋友很热心。 The friend who values Xiao Li is very warm-hearted. 这个请来小张的军官来军队很久了。 The officer who invited Xiao Zhang has been in the army for long. 这个喜欢唐兰的工程师成熟稳重。 The engineer who likes Tang Lan is mature. 这个相信小王的人工作很认真。 The person who believes Xiao Wang works seriously. 这个相信小陈的班长很负责。 The squad leader who believes Xiao Chen is very responsible. 这个支持小王的干部工作很认真。 The cadre who supports Xiao Wang works seriously. 这个推荐小林的人很能干。 The person who recommends Xiao Lin is very capable. 这个不喜欢小林的人经常来这儿。 The person who dislikes Xiao Lin often comes here. 这个看见唐兰的女孩很有礼貌。 The girl who saw Tang Lan is very polite. 这个拜访小陈的邻居家里很穷。 The neighbor who visited Xiao Chen is very poor. 这个教过小张的人现在很有名。 The person who taught Xiao Zhang is now famous. 这个想念小王的老战友经常写信回来。 The old comrade who misses Xiao Wang often writes letter back. (Continued) Downloaded from slr.sagepub.com by guest on September 23, 2014 460 Second Language Research 30(4) Appendix 1. (Continued) 13 * 14 15 * 16 * 17 18 * 19 * 20 * 21 22 * 23 * 24 25 * 26 * 27 * 28 * 29 30 * 31 * 32 33 * 34 * 不喜欢张力的这个人学习不太好。 The person who dislikes Zhang Li does not study well. 欣赏唐兰的这个男同学很讲究衣着。 The male classmate who admires Tang Lan is particular about dressing. 相信小王的这个同事很老实。 The colleague who believes Xiao Wang is very honest 喜欢小张的这个同学人品不错。 The classmate who likes Xiao Zhang has good integrity. 爱上小陈的这个博士读过很多书。 The Ph.D who falls in love with Xiao Chen has read many books. 爱上唐兰的这个人性格很好。 The person who falls in love with Tang Lan has a good character. 请来大卫的这个女生性格内向。 The girl who invited Da Wei has an introvert character. 爱上安妮的这个诗人很内向。 The poet who falls in love with An Ni is very introvert. 鼓励唐兰的这个人中文学得很好。 The person who encourages Tang Lan studies Chinese very well. 陪着小林的这个同学长得很漂亮。 The classmate who accompanies Xiao Lin is very pretty. 想念小王的这个女孩还在加州读书。 The girl who misses Xiao Wang is still studying in California. 解雇老高的这个人人品不好。 The person who fired Lao Gao does not have good integrity. 这个小李害怕的人就住在我们家附近。 The person that Xiao Li fears lives nearby our house. 这个小陈帮助的警察做事很认真。 The policeman that Xiao Chen helps works very seriously. 这个小李支持的人是我们的好朋友。 The person that Xiao Li supports is our good friend. 这个小张认识的军人经常和我们一起玩。 The soldier that Xiao Zhang knows of often plays with us. 这个老高欣赏的班长刚来部队不久。 The squad leader that Lao Gao appreciates just joined the army. 这个唐兰批评的班干部不太受同学欢迎。 The cadre that Tang Lan criticized is not popular among classmates. 这个小王批评的同事和大家关系不好。 The colleague that Xiao Wang criticized does not get along well with others. 这个张力熟悉的同学经常和大家一起看电影。 The classmate that Zhang Li is familiar with often watches movies with everyone. 这个小张喜欢的女孩叫珍妮。 The girl that Xiao Zhang likes is named Zhen Ni. 这个小李不喜欢的女生最近结婚了。 The girl that Xiao Li dislikes recently got married. Downloaded from slr.sagepub.com by guest on September 23, 2014 461 Xu Appendix 1. (Continued) 35 36 37 * 38 39 * 40 * 41 42 43 * 44 * 45 * 46 * 47 * 48 这个我们尊重的领导对大家都很好。 The cadre that we respect treats everyone well. 这个小林雇用的会计师很有经验。 The accountant that Xiao Lin employs is very experienced. 小陈联系的这个警察很有才能。 The policeman that Xiao Chen communicates with is very capable. 小明歧视的这个印度同学很会玩电脑。 The Indian classmate that Xiao Ming discriminates against plays computer well. 小李认识的这个外国朋友性格开朗。 The foreign friend that Xiao Li knows has a cheerful character. 小陈帮助的这个同学很友好。 The classmate that Xiao Chen helped is very friendly. 唐兰联系的这个情报人员很小心。 The intelligence personnel that Tang Lan communicates with is very cautious. 老高拜访的这个朋友很热情。 The friend that Lao Gao visited is very enthusiastic. 小王陪着的这个办公室主任很友好。 The office manager that Xiao Wang accompanies is very friendly. 唐兰照顾的这个邻居很有礼貌。 The neighbor that Tang Lan takes care of is very polite. 唐兰喜欢的这个男人很帅。 The man that Tang Lan likes is very handsome. 唐兰爱的这个男生喜欢弹琴。 The boy that Tang Lan loves likes playing the piano. 小张找的这个人是个年轻女孩。 The person that Xiao Zhang looks for is a young girl. 大卫寻找的这个女孩是我朋友。 The girl that Da Wei looks for is my friend. Downloaded from slr.sagepub.com by guest on September 23, 2014