YI XU

University of Pittsburgh

E－mail:xuyi＠pitt.edu

INTRODUCTION

Among the various well－studied syntactic structures of human language，the grammatical complexity and the typological universals of the relative clause（RC）have made this structure of great interest to theoretical linguists as well as scholars in the applied linguistic field including psycholinguists and second language acquisition（SLA）researchers.For instance，the recursive nature of RC structures makes infinite utterances possible，demonstrating a major advantage of the generative approach that represents languages in a hierarchical structure rather than a linear one.Because head directions and different clausal word orders are manifested in RCs in different languages，examination of the structure also provides invaluable information to probing universal processing cost（Gibson and Wu 2011）.In this paper，I review scholarly findings of the relativization structures from an interdisciplinary perspective.While I shall refer to formal linguistic theories and processing models，I focus on how theoretical foundations can inform acquisition studies，and particularly the second language（L2 ）acquisition studies of RCs.This paper takes two important background factors into consideration:First，there has been continuous development in applying formal linguistic theories in the SLA field and more recently in using psycholinguistic methodologies in empirical research（Juffs 2001；Marinis 2003；Roberts 2012）；second，acquisition studies of English RCs have been going on for almost half a century，with Brown（1971），Sheldon（1974）and Gass（1979）as pioneering examples，and recently，agrowing interest is to test whether certain universal mechanisms make particular types of RCs more difficult than others cross－linguistically，with uncertainties especially in East Asian languages.For instance，Studies in Second Language Acquisition published a thematic issue in 2007that reported a range of studies of RC acquisition in Japanese，Korean，and Cantonese.In the recent decade or so，research in the syntax of Chinese as a second／foreign language acquisition burgeoned，drawing global attention（Shi and Wen 2009；Zhao 2011），yet compared to the long－standing history and the more conclusive findings of RC acquisitions in European languages，published studies on the（L2）acquisition of Chinese RCs are scarce.Meanwhile，Chinese RCs are typologically unique（Comrie 2008）:like English RCs，the clausal word order is SVO；yet like Japanese and Korean RCs，it is prenominal／head－final.Dryer（2005）reports that only five languages out of a sample of 756show such a rare combination，and the study of Chinese RC is therefore particularly informative in disambiguating the validity of different psycholinguistic models（Gibson and Wu 2011），and is key to assessing if acquisition difficulties of RCs follow some purported structural markedness or universal orders.With these situations as a backdrop，this review adopts an interdisciplinary approach，and attaches importance to the comparison of studies in English RCs and the current gap in our knowledge of East Asian，particularly Chinese，RC acquisitions.

FORMAL LINGUISTIC AND PSYCHOLINGUISTICS THEORIES OF RCs

Theoretical foundations

Relative clauses may be categorized in several different ways.The following terminologies will be used in this paper.

（a）Based on the head direction，the structure may be head－final（prenominal），exemplified by RCs in Korean and Japanese，or head－initial（postnominal），as in English，F(xiàn)rench，German，etc.

（b）Based on head position，RCs can be externally－h(huán)eaded，i.e.the head noun occurs external to the subordinate clause，as in regular RCs witnessed in Chinese and English，or they can be internallyheaded，i.e.head within the subordinate clause.Both these structures exist in Korean.

The following are important features and variables tested in behavioral studies of relativization:

（c）In terms of relativization strategy，an RC can be formed by using agap，or by using a（resumptive）pronoun（Keenan and Comrie 1977）；

（d）Different position in a sentence can be relativized.Depending on the“relativization／extraction type”，there exist subject RC，direct object RC，indirect object RC，etc.；

（e）An RC can occur in different positions in a main clause:It may be subject－modifying（subjectmod）or object－modifying（object－mod）；

（f）In more recent years，animacy of the argument nouns is frequently thought to be a factor affecting RC ease.

There are many other ways to categorize different RC types.For instance，in semantic interpretation，there are restrictive versus non－restrictive RCs，and in Chinese，the difference is often thought to be reflected by the demonstrative－classifier（DCl）－RC or RC－DCl sequences，although there is no agreement on which is which（e.g.Chao 1968；Tang 1981；Tsai 1994）.Discussions will be restricted to RC properties more often discussed in terms of comparative processing and acquisition difficulties in this paper.

de Vries（2002）refers to two essential properties of relative constructions:A relative clause is subordinated，and is connected to surrounding material by apivot constituent that is semantically shared by the matrix clause and the relative clause.The pivot is the“head”of a relative clause from a syntactic perspective.In English，the relative clause typically contains a gap，and the head appears as an antecedent outside the RC.Assuming Chomsky's（1977）matching／operator analysis，English RCs have the internal structure in（1），with（1a）illustrating a wh－relative，and（1b）athat－relative clause.

In both，the RC head is base－generated external to the relative clause，which is adjoined to it.In（1a），within the RC，awh－pronoun is base－generated in argument position，and acts as an operator and moves from within the TP to the Spec of CP position.In that－relatives or RCs without an overt marker，a nonovert operator moves from the position within TP to the Spec of CP.If that is present，it occupies the complementizer position.

Assuming the matching／operator analysis，a Chinese RC has the structure in（2）.

Extraction positions and the NPAH

An early generalization regarding different relativized positions is the Noun Phrase Accessibility Hierarchy（NPAH，or AH，Keenan and Comrie 1977）.In the following，I will simplify the syntactic tree structure—the simplification does not affect the current discussion，since what is omitted is identical throughout different relativizable positions.（3a－3d）illustrate the relativization of some different positions.The AH is a purported typological universal based on summaries of about fifty world languages.It can be roughly presented as SU＞DO＞IO＞Oblique Object／OPrep＞Genitive＞Object of Comparison，where“＞” means“more accessible”or“easier to relativize”（Keenan and Comrie 1977）.The relativized position is also referred to as the“extraction”position or extraction type，the trace position a“gap”，and the relative head noun（N）a“filler”.（“Extraction”here is not identical to“movement”in formal syntax，as（1）and（2）illustrate that the head noun in the matching analysis does not move；it is an easier way to refer to the filler－gap coindexation relation.）（3a－f）show English RCs in different extraction types.

Relativizing the SU position is easier than relativizing the DO position，followed by relativization of IO，and then Oblique object／Object of Preposition，Genititive of the Subject and Object position，etc.Keenan and Comrie（1977）pointed out that to achieve relativization，a language may use the primary strategy of a gap，or the secondary strategy of a pronoun.In addition，a language must be able to relativize the subject，if it has relativization structures at all，and a relativization strategy has to apply to a continuous segment of the hierarchy.For instance，if language X uses the gap strategy in the IO position，then it must also use the gap strategy to relativize the SU and DO positions；meanwhile，relatization of lower positions（OPrep，Genitive，etc.）may not exist，or the pronoun strategy may have to be used.For instance，Chinese uses the gap strategy for SU and DO，and the pronoun strategy for IO and Oprep relativization.

The AH bears great importance to applied linguistic studies.It has been hypothesized that the Hierarchy is a“psycholinguistic”one in nature（Keenan and Comrie 1977），or reflects processing difficulties（Gass 1979）.The validity of the AH in implying greater processing cost in lower positions was confirmed in Keenan and Hawkins（1987）in which adult and children English L1participants were asked to write down the sentences containing RCs that they heard from a tape:the comparative rankings of correctness of recall adhered to the AH significantly.In acquisition studies，since Eckman（1977）proposed that typological markedness could be a possible explanation or a means of predicting interlanguage（IL）development in SLA，the NPAH is frequently thought to be a markedness universal:extractions lower on the hierarchy are thought to be more“marked”and therefore inherently more difficult to acquire compared to other less marked structures.Indeed，the Hierarchy is found to be robustly predictive of the acquisition difficulty reflected by accuracy rate at particular stages of IL development and the order of acquisition in postnominal RCs in Indo－European languages，mostly in English（Gass 1979，1982；Doughty 1991；Eckman et al.1988）.If psycholinguistic mechanisms are indeed responsible for the observed hierarchy，then it is not surprising that the same motivations would give rise to learners'processing challenges and difficulties in acquisition.

However，it should be noted that although Keenan and Comrie（1977:93）briefly suggested some“speculations”regarding the processing motivations for the AH，they were not decisive on the exact mechanisms responsible for it.Therefore the AH itself was not formulated as a processing theory，but merely ageneralization based on the authors'observations.In the following，I review some major psycholinguistics models that can potentially explain or predict processing cost associated with extraction position and other variables.Among them，the structural distance theory（SDT）explains the AH，although the SDT does not necessarily predict asymmetries between extraction types in some lower positions.

Psycholinguistic models

Because extraction in lower positions such as IO，OPrep，etc.，inevitably involves more words and the parsing of a range of other syntactic and semantic relations，only the difference between SU and DO can be clearly attributed to extraction position when experiment materials control all other factors such as using the same lexical items.Psycholinguistic studies therefore most often focus on the asymmetry between SU and DO extractions，and preference for SU relativization was attested to in a series of studies as early as Wanner and Maratsos（1978）.The range of measures that have been used include eye－movement，e.g.Holmes and O'Regan（1981）for French and Traxler et al.（2002）for English，lexical decision times in Ford（1983），and comprehension accuracy and reaction times in King and Just（1991）in self－paced reading.Crosslinguistic evidence was also found in Dutch（Frazier 1987），German（Mecklinger et al.1995；Schriefers et al.1995），mostly in head－initial RCs.But while the SU preference is consistently reported using a variety of psycholinguistic measures，there is little agreement on what exactly makes an SU easier.

According to Gibson and Wu（2011），there are three classes of theories explaining these attested results.The first major class is an experience－based account，in which more frequently－used or encountered structures are assumed to be easier to process.Several hypotheses can be subsumed under this account including those that are based on corpus statistics and the notion of expectation versus surprisal（Levy 2008；Hale 2001）.While more recent proposals include MacDonald and Christiansen（2002）and Diessel and Tomassello（2005），the Canonical Word Order Hypothesis（Bever 1970；Slobin and Bever 1982）appears to be the one most frequently referred to and tested in existing L1processing and acquisition studies.Note that these variations do not always give identical predictions.For instance，Hale's account（2001），which relies more on RC type frequencies in corpus，differs from the word order account.

Gibson and Wu（2011）considers explanations from the AH perspective to be a version of an experiencebased account too，assuming that“the most plausible potential explanation”for the AH is that extraction from less accessible positions is less frequent than extraction from higher positions.Although Keenan and Comrie（1977）did not really suggest frequency in usage to be responsible for the typological observation，textual frequencies of relativization on different positions corresponding to the AH were reported in Keenan（1975）for English.

The second class is a working memory－based account，including considerations of storage cost and retrieval／integration cost，both referred to in Gibson（1998；2000）.Examinations of the former would crucially rely on online processing experimental methodologies（with the concept of“online”technique explained later）.For the integration cost，there are several different metrics that can realize the exact calculation.In this paper I refer to two interestingly contrastive metrics，the linear distance theory（LDT）proposed by Gibson（1998，2000）and the SDT proposed in O'Grady（1997，1999）to explain acquisition behavioral patterns that are consistent with the NPAH，and independently proposed by Hawkins（1999）to explain the psycholinguistic motivations for the two alternative RC－forming strategies（agap strategy and a pronoun strategy）in natural languages.

The third class is the Perspective Shift Theory（MacWhinney 1977，1982；MacWhinney and Pleh 1988）.In the following，I summarize some of the major theories.

The CWO

This canonical word order（also referred to as the NVN－schema in Bever 1970）hypothesis states that if the word order in a sequence／structure is similar to the word order in canonical sentences of the language，that structure should be easier to process.In English，for instance，the word order in SU relatives，e.g.the dog that bit the man，has the word order of Noun－Verb－Noun（N－V－N），which is similar to the word order in simple English sentences.In DO relative clauses in English，the word order is N－N－V，e.g.the man that the dog bit，and it is different from the canonical word order in English.This theory makes the right prediction that processing and acquisition of SU relatives in English is easier than DOs，and it was also supported cross－linguistically by evidence in acquisition of SOV languages such as Korean（Clancy et al.1986）.In Chinese，SU relatives follow the order of［［＿V N de］N］（i.e.V－N－N，exemplified in（4a）），while DO relatives have the order of［［N V ＿de］N］（i.e.N－V－N，as seen by（4b））.In general，the interpretation of CWO's prediction regarding Chinese is that DO is easier.（See，for instance，Hsiao and Gibson 2003），since a sequence starting with a verb is most often considered to be inconsistent with the canonical word order of Chinese.

An issue with the CWO theory，as critiqued by Kuo and Vasishth（2006），is that it has never been formalized for straight－forward interpretation in different languages.The implementation of the theory in Chinese may be especially hard，since a number of frequently－used structures in Chinese including the baconstruction，the pro－drop properties，etc.，would make the SVO word order inconsistent in Chinese（Juffs 2007）.Therefore，it is much less certain that SVO is a word order as canonical in Chinese as it is thought to be in English.

The LDT

Gibson（1998，2000）proposed the dependency locality theory that considers two mechanisms，a linear integration cost of associating the gap with the filler（linear integration cost，also referred to as retrieval cost），and a memory cost of keeping track of the required syntactic categories to complete the current input string as grammatical sentence（storage cost）.The two mechanisms can be teased apart and tested independently（e.g.Hsiao and Gibson 2003）.I will introduce only the linear integration cost，also referred to as the LDT（see Wu et al.2007；Hsu et al.2009）here，as contrasted with the structural distance theory discussed below.The LDT predicts that longer filler－gap linear distance is associated with greater processing cost.Therefore，in English（3b）is more difficult than（3a）（SU＞DO），because dogand bit intervenes between the filler and the gap，and in Chinese，（4a）is more difficult than（4b）（SU＜DO），due to the intervening of xihuan‘like'and Lisi.

The dependency locality theory is said to provide a unified explanation to a variety of phenomena observed in processing studies aside from RC extraction effect，such as the preference of local rather than distant attachment of an adverb to a verb（Gibson 1998）.However，although Gibson（1998，2000）maintained that the dependency locality theory is supported by cross－linguistic evidence from Japanese，Dutch，and German，its applicability in Chinese RCs is much less certain.In Hsiao and Gibson's（2003）online processing experiment of Chinese RCs，for instance，they found support for the storage cost，but not for the linear integration cost.

The SDT

According to the structural distance theory，processing cost is determined by the structural distances between the gap and the head noun.O'Grady（1999，1997）and similarly Hawkins（1999）propose that the complexity is calculated by the number of XP（VP，CP／S，etc.）categories that intervenes between the gap and the head.The crucial difference between the SDT and the LDT is that the former considers the filler－gap distance on a hierarchical tree structure rather than on linear order:while both correctly predict that SU is easier in English，the SDT predicts that SU is easier in Chinese as well as in all other human languages.This is because the minimal number of nodes（head and maximal projections，considered as the minimal filler－gap domain）required to establish a SU relativization is five，including V，VP，CP，N，and NP，and the number required to establish a DO relativization is seven，including V，VP，NS，NPS，CP，N，and NP.That is because in DO relativization，a subject NP is obligatory，but in SU relativization，whether an object NP exists or not（depending on the transitivity of the verb）does not affect the filler－gap relation.While readers are directed to Hawkins（1999:255）for the exact syntactic illustrations，the minimal filler－gap domain in IO，OPrep，Gen－SU（genitive of subject），and Gen－DO involves nine，nine，nine，and eleven nodes，respectively.—Because IO，OPrep，and Gen－SU involve the same number of nodes，this theory does not directly predict IO＞OPrep＞Gen－SU，but it is generally consistent with and is thought to provide the motivations for the AH.1

The SDT has some obvious advantages.For one，its predictions are not restricted to SU／DO asymmetries，but can explain the markedness observed in other extraction positions on the Accessibility Hierarchy（at least to the extent of explaining SU＞DO＞IO／OPrep／Gen／OComp）.Hawkins'（1999）theory is also appealing in explaining several other aspects of the original AH observation，including the preference for a gap strategy versus the pronoun strategy in higher positions，the application of those strategies on a continuous segment of the hierarchy，as well as the potential overlaps between the two RC－forming strategies:Hawkins（1999:257－258）hypothesized that“gap”is preferred in more accessible positions for its“economy of expression”while“pronouns”are used to make longer－distance filler－gap coindexation easier by explicit（rather than abstract，as in the case of a gap）dependency marking，and thus used in less accessible positions.O'Grady's（1997，1999）proposal is the same as Hawkins'in spirit，and was based on and confirmed by the L2acquisition of Korean head－final RCs（O'Grady et al.2003）.Therefore，the SDT is applicable to language typology，processing，as well as acquisition.

But the theory has some shortcomings as a processing model:Gibson and Wu（2011）critiqued that the SDT was not independently supported by evidence other than the SU／DO asymmetry:neither O'Grady nor Hawkins made any claims regarding the explanatory power of the SDT to other syntactic relations beyond RC－extractions.

Another outcome of the SDT is that since it essentially relies on hierarchical syntactic structure，the particular word order of a language（whether it is SVO or SOV，etc.）no longer matters.Thus，this theory predicates that the ranking of processing difficulty in different extraction positions is“universal”.While postnominal RC studies report results consistent to this，there is a lot ground to be covered up in studies of prenominal RCs for one to assess if such“universality”can hold cross－linguistically.

The PST

A third class of processing theories that can potentially explain the SU／DO asymmetry is the Perspective Shift theory（MacWhinney 1977，1982，2005；MacWhinney and Pleh 1988）.“Perspective”is a pragmatic and functional notion and this theory proposes that a reader's perspective is oriented to the sentential subject by default.Parsing cost depends on whether the same perspective is used in the relative clause as in the main clause.This theory considers the relation of extraction position and modifying（or matrix）position（i.e.whether the RC modifies a subject or an object in the main clause）.（5a），for instance，is an example of subject－modifying，subject－extraction（SS）RC whereas（5b）is an example of subject－mod，object－extraction RC（SO）.（5a）is the easiest among（5a－d）because perspective remains to be the one oriented to the dogthroughout the sentence.For（5b），there is first a shift from the Subject of the main clause（the dog）to the Subject of the RC（the cat），and then another shift back to the perspective of the dogat the encounter of the main clause verb.For（5c），the OO type，the perspective begins with the dog and has to shift to the horse at the encounter of the RC.For（5d），the OS type，one perspective shift also has to occur when readers realize that the cat，Object of the RC，is the Subject of the main clause.That is，the PST would predict the order of difficulty to be SS＞OO／OS＞SO.

（5）a.SS:The dog that chased the cat kicked the horse.—no shift

b.SO:The dog that the cat chased kicked the horse.—two shifts

c.OO:The dog chased the cat that the horse kicked.—one shift

d.OS:The dog chased the cat that kicked the horse.—one shift

MacWhinney（1982）referred to several studies that appeared to corroborate such a prediction.However，a major disadvantage of this theory is that it seemingly predicates that SU and DO relativizations entail no processing differences if they occur in object－modifying positions，as both（5c）and（5d）involve one perspective shift.This was not found to be true in psycholinguistic studies such as Gibson et al.（2005），in which DO was reported to be harder disregarding modifying positions in English.Regarding the PST's prediction for Chinese RCs crossing modifying position and extraction type，the SS type of RCs are predicted to be the easiest，while the other three types are relatively hard.Xu（2012）discusses the PST's prediction for Chinese RCs involving demonstrative－classifiers（DCls）.

The above four models are representative theories that offer distinctively different explanations to the SU／DO asymmetry in postnominal RCs.While some attempts have been made，as summarized in the next section，their applicability to prenominal RCs has not been tested as extensively as in English，and great potential exists for researchers to explore their validity particularly using data from East Asian languages.

BEHAVIORAL STUDIES OF RCs

Psycholinguistic studies

SLA researchers are increasingly aware that acquiring an L2involves not only accumulating grammatical competence but also the ability to apply the knowledge into real－time processing（White 2003；Clahsen and Felser 2006）.On the topic of the L2acquisition of RCs in particular，Juffs（2007）states that existing research mostly adopted offline methods，and the field will benefit from future studies using online，psycholinguistic methodologies.Meanwhile，in Roberts'（2012）summary of psycholinguistic techniques in SLA，he mentioned，among other things，the use of self－paced reading and eye tracking，techniques used the RC L1processing.I will start with L1processing studies on the topic，focusing on self－paced reading experiments，to lay out the theoretical foundations and to establish the basis for L1－L2comparisons.

In self－paced reading，participants most often read word by word on a personal computer，and proceed by pressing a key／button to bring up the next word on the window（i.e.each button－push removes the current word from the screen and replaces it with the next），thus controlling the reading pace themselves.The time it takes for each button－push is recorded by the experimental software as reaction time（RT）.RT is therefore a measure of how easy or difficult it is to integrate each unit into the comprehender's current analysis of the input（Roberts 2012）.Aside from RT，comprehension questions often appear on the screen at the end of each item to ensure that the participants read for meaning in relative clause experiments（e.g.Gibson et al.2005）.Items and data from participants whose error rate is high are often excluded from analysis in L1studies，but Juffs（2001），Juffs and Harrington（1995）cautioned that accuracy／error rate may be significant to conclusions in L2studies.

Gibson et al.（2005）used a self－paced experiment to examine English RCs crossing three factors，modifying position，extraction type，and embeddedness.（An RC is embedded in a matrix clause by the fact that.For instance，a test item of embedded subject－modifying，object－extraction RC is The fact that ［the reporter who the senator attacked on Tuesday ignored the president］bothered the editor.The part within the bracket constitutes a non－embedded counterpart of it.）They tested four hypotheses:the storage cost，the LDT，the CWO，and the PST，and suggested that several factors were simultaneously responsible for their result:the storage hypothesis accounted for the effect of embedding，and the integration cost explained why object－extraction is harder.They also found subject－modification to be easier，a result not predicted by any of the above－mentioned theories.To explain this，Gibson et al.（2005:328－329）proposed an information－flow hypothesis:Old，background information is supposedly comprehended more easily early in a sentence，such as in a position modifying the subject，while new，foreground material is processed more easily later in a sentence，such as in a position in the main predicate of the sentence.Because restrictive RCs are thought to bear old information（Del Gobbo 2003，among others），RC－modifications in the matrix object position is easier than RC－modification of the subject.Gibson et al.'s（2005）experiment 2，using non－restrictive RCs，which supposedly carry new information，confirmed this information－flow hypothesis.

Hsiao and Gibson（2003）used the same experimental technique to test Chinese RCs，examining only the effect of extraction.Among the five theories that they tested，they found that the storage resource theory not only correctly predicts the contrast（DO＞SU），but also the locus of the effect，while the LDT correctly predicted the contrast，but not the locus of this effect；CWO correctly predicted the contrast，but the authors argued that an RC－type frequency－based account cannot be correct，because corpus analysis reveals that SU RCs are in fact more frequent in Chinese（Kuo and Vasishth 2006；Chen et al.2010）.The AH and the PST were rejected in this study.

While SU preference was reported for Japanese in Ishizuka et al.（2003），Miyamoto and Nakamura（2003），Ueno and Garnsey（2008），and in Korean in Kwon et al.（2006），Hsiao and Gibson's（2003）claims of a DO preference in Chinese RCs are especially controversial.Although their results were replicated in Chen et al.（2008）with L1Chinese participants with lower working memory span，as well as in Gibson and Wu（2011），other studies，including Lin（2006），Lin and Bever（2006），Kuo and Vasishth（2006），contended that Chinese was not different from postnominal RC languages and that SU was easier.According to Gibson and Wu（2011），Lin and Bever's（2006）result was confounded by temporary ambiguities of Objectextraction，Object－modifying RCs，2and that the SU preference reported in Lin and Bever was mainly caused by the difference in SU and DO in the Object－modifying RC types.Gibson and Wu（2011）made improvement from previous studies in that they provided participants with discourse contexts so that only particular noun phrases as answers to questions would be expected，thus excluding the potential effect of temporary ambiguity associated with the processing of Object RC.3Gibson and Wu（2011）claimed that their result supported the LDT.However，this result was not replicated in other studies using similar approaches to Gibson and Wu（2011）:Chen et al.（2010）conducted experiments first without supportive context and then with context，and found that the presence of context did not change the SU preference，and the slower reading time in DO cannot be exclusively attributed to garden path／temporary ambiguity effect before the head noun.Li et al.（2010）also used similar online processing tasks in supportive context（although the context provided was different from Gibson and Wu（2011）design）and reported that SU RCs were read faster.Table 1includes a summary of these studies，showing the variables included，and the tested and supported hypotheses.

Table 1:Psycholinguistic experiments on relative clauses

（Continued）

Acquisition studies

In acquisition studies，there is ample evidence that in English RCs，SU relativization is easier than L1 children（Brown 1971；Diessel and Tomasello 2000；Keenan and Hawkins 1987）.As for extraction of other positions，Gass（1979）found that English L2learners'comparative accuracy rate in different RC types correlated remarkably with the AH.According to Eckman's Structural Conformity Hypothesis，universal markedness should be reflected in ILs（Eckman 1984，1991，1996）.A number of later studies illustrate that the acquisition order or inherent difficulty of RCs in a second language，mostly in English but also in other Indo－European languages（e.g.Swedish L2acquisition in Hyltenstam 1984），can generally be predicated by the AH.The implicative power of the AH was also tested in both formal and informal learning context in Pavesi（1986）with positive results.

At the same time，the implication of the AH in SLA is not to be interpreted too literally.Gass（1979:339）mentions that while the AH“represents the order of difficulty in L2acquisition”，it is not a“rigid constraint that must be followed in all instances”.For instance，in the sentence combination task result in Gass（1979）as well as Doughty（1991），while learners'accuracy generally decreased as the position descended on the hierarchy，an exception was found in relativizing the Genitive position，in which condition more correct responses were observed than would have been predicted.Gass（1979）hypothesized that this may have something to do with the uniqueness or saliency of the relative marker whose，or participants may have treated whose＋noun as one unit and in turn processed it as the subject or direct object of an RC structure.

These studies involve a range of tasks including sentence combination，grammaticality judgment，picturecued oral production，etc.（Aarts and Schils 1995；Eckman et al.1988；Doughty 1988，1991；Gass 1979，1982）.Among them，several are pedagogical intervention studies which showed that by instructing learners more marked constructions（i.e.RCs lower on the hierarchy），learners would acquire both the instructed constructions as well as less marked constructions（i.e.extraction from positions higher on the AH than what is instructed）.However，it remains somewhat ambiguous if the implicational generalization is always unidirectional（Gass 1982；Doughty 1988，1991；Eckman et al.1988）.For instance，experimental groups receiving instructions on OPrep RCs only were found to have gains in SU，DO，IO positions（higher on the AH）as well as some gains in Gentive and Object of Comparison positions（lower／more marked than OPrep）in Doughty（1991）.

Aside from the NPAH，two other theories have been referred to and sometimes tested in the SLA field，one being Kuno's（1974）perceptual difficulty hypothesis（PDH ）and the other Hamilton's（1994）SO hierarchy hypothesis（SOHH）.The PDH considers that due to short－term memory limitations，center embedding，which interrupts the processing of the matrix sentence with the RC，is perceptually more difficult than right and left embedding，In the pair of comparison below，right embedding is thought to be easier.

（6）a.Center embedding:The cheese that the rat that the cat chased ate was rotten.

b.Right embedding:The cat chased the rat that ate the cheese that was rotten.

According to the PDH，regardless of extraction type，English sentences with RCs embedded in the matrix subject position is more difficult than sentences with RCs embedded in the matrix object position.The PDH was supported by Ioup and Cruse'（1977）English L2study.Note that although Gibson et al.（2005）admitted that nested constructions are generally thought to be more difficult，the predications of the PDH were contradictory to Gibson et al's（2005）results.At the same time，although it has never been specifically tested，the PDH would predicate that subject－modifying RCs in Chinese are easier than object－modifying ones，as the former constitute left embedding while the latter center－embedding.

The SOHH was formulated by Hamilton（1994）to combine the effect of extraction type and modifying position.Hamilton（1994）considers“processing discontinuity”as the unified motivation for difficulties associated with both，and such discontinuity can be created by the center embedding of an RC，or by phrasal boundaries within the RC that separate the relativized position from the wh－pronoun or non－overt operator（i.e.one discontinuity in subject－extraction and two discontinuities in object－extraction）.

（7）a.SS:The mani［CPwhoi［TPtineeded a job］］helped the woman.—two discontinuity:one within RC，one by center－embedding.

b.SO:The dogi［CPOpithat［TPthe woman［VPowns ti］］］bit the cat.—three discontinuity:two within RC，one by center－embedding.

c.OS:They saw the boyi［CPwhoi［TPtientered the room］］.—one discontinuity within RC

d.OO:A man bought the clocki［CPOpithat［TPthe woman［VPwanted ti］］.—two discontinuity within RC

（Examples adapted from Hamilton 1994:134）

The SOHH also predicates that OPrep is more difficult than SU and DO extractions，because it involves three discontinuities within the RC:three phrasal boundaries PP，VP，and TP are crossed over.

（8）The man［CPwhomi［TPwe［VPthought［PPabout ti］］］

Therefore and the overall predication of the SOHH is OS＞OO／SS＞SO／OOPrep＞SOPrep，where＞means“easier than”.Izumi（2003）used a combination of reception（grammaticality judgment and interpretation test）and production tasks（sentence combination）to test all the three theories including NPAH，PDH，and the SOHH on L2English participants with different L1s.He concluded that results partially supported the NPAH with the exception of the relative difficulty of DO and OPrep；the PDH prediction was supported by all tests，and the SOHH was largely supported by all results with the notable exception of the OPrep.Izumi（2003:311）in turn suggested that discontinuity caused by phrasal boundaries and the one by embedding“may not...carry the same weight”.

These three hypotheses were tested in EFL studies conducted in mainland China as well.Xiao and Lü（2005）used a sentence combination task to test the three hypotheses using a 2（modifying position）×4（extraction type）design.They reported that the PDH was supported.However，since they did not report statistical results of pairwise comparisons，their results merely showed that both extraction type and modifying positions were significant factors，and not necessarily strong support for the SOHH or the AH.Their numerical data suggested a tendency of Gen relativization being easier than DO and OPrep，consistent with the SOHH but not the AH（since discontinuities caused by Gen－SU and Gen－DO are identical to SU and DO extractions），and a tendency of OPrep being easier than DO，a surprising result not predicted by any of the theories.Dai et al.（2008）investigated the topic using a sentence combination task and naturalistic written data from students'compositions.The authors attempted to evaluate the validity of eight processing theories（all included in Table 2and the note）and asserted that different aspects of their results supported the AH，the minimal distance theory（referring ambiguously to the LDT and the SDT），the CWO，and the SOHH.However，the authors relied on tendencies of numerical differences and token numbers to interpret the data，while the only pairwise comparisons that reach statistical significance were SS＝OS＞SIO in the experimental data.Therefore，their conclusions must be taken with caution.Finally，Tang and Xu（2011）used three tasks，sentence combination，grammaticality judgment and Chinese－to－English translation，to investigate the same two factors.They found that the AH was supported by the sentence combination task most，while results from all tasks supported the PDH.

In sum，in behavioral studies of English RC studies，although controversies exist in the comparative difficulty of lower extraction positions（IO，OPrep，Gen，etc.）as well as the interaction of other factors such as modifying position with extraction，the preference of SU over DO is generally acknowledged.Several models can potentially explain the asymmetry.Different hypotheses are summarized in Table 2，while results from earlier studies that representatively show a variation of task types and theories supported are included in Table 3，together with findings in East Asian languages，reviewed in the next section.

Table 2:Major theories regarding structural difficulties associated with relativization

Controversies in East Asian language acquisition

Recently，much attention has been directed to prenominal relative clauses（or head－final RCs）in East Asian languages，with the controversy centering on whether SU is easier than DO in those languages.In L1acquisition studies，preference for SU over DO was reported in Lau（2006）for Cantonese and Cho（1999）for Korean.On the other hand，Clancy et al.'s（1986）study on L1Korean，Japanese，and English children using an act－out task，while reporting evidence supporting theories such as the canonical word order，did not find result that can support Subject preference except in English.In L2acquisition of East Asian languages，extraction type as well as other factors such as animacy，internally－h(huán)eaded（IHRC）versus externally－h(huán)eaded RCs（EHRCs）are considered.For instance，Jeon and Kim（2007）reported that the internally－h(huán)eaded RCs were acquired earlier，and that there was an SU RC preference in externally－h(huán)eaded RCs.However，because in Jeon and Kim's（2007）study，target－SU responses all contained animate head nouns while most target－DO responses contained inanimate heads，they suggested that the preference of SU cannot be attributed solely to the AH and that there was a clear binary association of SU with an animate head and DO with an inanimate head.（Animacy effect was reported in English RC studies too，e.g.Traxler et al.2002.）Ozeki and Shirai's（2007）study includes two components，the first based on naturalistic data from a learner corpus consisting of oral interview transcripts from novice to the superior learners，and the other from a written sentence combination task crossing factors of extraction type and animacy of the head noun.Corpus data showed that learners used SU，DO and OPrep with equal comfort even at early stages of acquisition.The sentence combination task suggested that SU and DO were both easier than OPrep，while accuracy rate between them did not differ.Importantly，Ozeki and Shirai（2007）noted that SU relatives with animate head nouns and DO／OPrep relatives with inanimate head nouns were easier than SU relatives with inanimate heads and DO／OPrep relatives with animate heads.Another study that takes both the extraction type and the semantic cues into consideration is Kanno（2007），in which L2Japanese from different L1 groups participated in a listening comprehension task.The test items include both the ［－R］type（nonreversible RCs with one animate argument and one inanimate argument）and ［＋R］（reversible ones with two animate arguments），and both conditions also have SU and DO extractions.This study found that although L2learners overall found SU easier to process，the preference stopped when there was no semantic clues.The author observed that ［－R］was much easier to comprehend as seen by accuracy rate，and that the usefulness of animacy was evident.However，because Kanno（2007:197）investigated learners with 2 months'exposure to Japanese only，learners'performances in relativization were“poor”and they“vary”when no semantic cue was available in the author's own words，making it impossible to conclusively interpret the result as indicating a structural preference for SU or DO.

In general，all those studies suggest a strong reliance on semantic information other than syntactic relations in L2speakers'processing or production of RCs，but none of them can be interpreted as a direct rebuttal to the markedness of the AH:what AH rules out is apotential learner language exhibiting more marked relativizations（lower on the hierarchy）while less marked relativization is not acquired，and there was no clear evidence of that in the above－mentioned three studies.Yip and Matthews（2007）examined the L1 acquisition of Cantonese and English by bilingual children using longitudinal diary data.They claim that the acquisition of DO RCs occurred earlier than SUs，contrary to the AH.However，it should be noted that their data came primarily from the observation of three bilingual children and since those were recorded naturalistic data，the authors admitted that the“first attestations are likely to lag behind the initial use of each type”（Yip and Matthews 2007:287）.Thus，one cannot be completely certain regarding what RC type really emerged first.Also，one of three children produced SU and DO RCs on the same day，leaving data from only two children who have attested utterances of DO RCs before Subject ones.Futhermore，while Yip and Matthews'（2007）study was particularly interesting for its unique data type，there was some difficulty in interpreting the result directly in line with other L2studies.First，the participants were bilingual children，making their data different from other L1or L2results in nature:potentially，it is possible that the patterns observed may not be exactly the same with the developmental sequence in either L1or L2acquisition of Cantonese or English RCs.Second，Yip and Matthews（2007）suggested that DO（but not SU ）RCs in Cantonese can be analyzed as IHRCs.3If so，then the early emergence of DO in Cantonese cannot be evidence against the AH，but merely confirming that the IHRC emerges earlier than the EHRC，as L1／L2studies of Korean demonstrated.（See Kim 1987and Lee 1991for L1acquisition and Jeon and Kim 2007for L2acquisition.）

Studies of Chinese RCs

The existing few L1acquisition studies of Chinese RCs generally suggest a preference of Subject RCs，but the results are not particularly robust.Cheng（1995）used a picture－description task and reported that in correctly using the gap strategy，children performed better in the targeted SU condition.Su（2004）used an acting－out，verbal description task with L1children and adults，and although she did not find differences between the SU and DO condition in the percentages of RCs and targeted RCs produced，she reported that both older and younger children used more resumptive pronouns in DO than in SU relativization.Hsu et al.（2009）also used an elicited production task using pictures with L1children and adults.They found that children performed significantly worse with DOs than SUs.

L2acquisition studies of Chinese RCs are scarce.Recently，Packard（2008）used an on－line self－paced reading task to examine L2speakers'processing of Chinese RCs and claimed that Subject RCs were read more slowly than Object RCs.This study takes a progressive step of using psycholinguistic techniques in inquiries of L2syntax.However，because the experiment sentences were not counterbalanced and animacy as a confounding factor was not controlled，Packard's（2008）results may need reconsideration.Xu（in press－a）conducted a listening comprehension multiple choice task in which the prompt contained only the complex nouns crossing three variables，animacy，SU／DO extraction types，and the presence／absence of a post－RC DCl.Results in Kanno（2007）were replicated:learners generally performed better in Subjectanimate，Object－inanimate RCs，most likely due to the additional semantic cues.For DCl－absent RCs with two animate argument nouns，SU had a higher accuracy than DOs，but the same asymmetry was not found in DCl－present，both argument animate RCs.

I am only aware of two recent studies that specifically investigated whether the L2acquisition of Chinese conforms to the AH:both Dai（2010）and Xu（in press－b）conducted a written sentence combination task.Dai's（2010）experiment included both extraction type and modifying position as variables，and he claimed that for subject－modifying RCs，the order of relative ease was SDO＞SS＞SOPrep＞SIO，supporting Gibson's LDT，while in object－modifying RCs，the order was OS＞ODO＞OIO＞OOPrep，consistent with the AH.Meanwhile，in two－way ANOVA，the effect of modifying position was not found to be significant and Dai（2010）hypothesized that the ease of SDO and OS balanced out the modifying position effect.Xu（in press－b）found the accuracy rate in the SU and DO conditions to be much higher than in the IO and OPrep conditions.She argued that although accuracy between SU and DO conditions did not differ，most of the errors in the SU condition were not related to the relativization structure，and that learners tended to produce SUs when other types of RCs were targeted，even when extra lexical items and passive bei needed to be inserted.These qualitative analyses pointed to an overall preference for SU over DO.Seemingly surprising was the result that IO relative clauses were produced with better accuracy than OPrep in Xu（in press－b）.The author found that the difference was caused solely by more target－like use of pronoun strategies in the OPrep position and more gaps in IOs；she argued that because the AH generalization was an increasing reliance on pronoun strategies in lower positions，her result in fact supported the implication of AH in learner language，though not in terms of the ranking of accuracy.In fact，in her study，each individual learner's IL in terms of RC－forming strategies（using agap or a pronoun）conformed remarkably to the AH stipulation.Hawkins'（1999）SDT was argued to be the underlying motivation.

In Chinese RCs，the position of the DCl potentially affects processing and acquisition too.DCl can occur either before or after the relative clause，as shown by（9）.

Wu et al.（2007）and Wu et al.（2009）observed from corpus data that when RCs are modified by DCls，SU tend to have the DCl－RC sequence，while DOs are more likely to have the RC－DCl sequence.In behavioral experiments，Wu et al.（2009）reported that L1speakers were able to read the main clause after the complex noun faster in the DCl－SU condition than in the DCl－absent condition and they attributed the effect to the cueing function of classifier to predict the upcoming noun，but this facilitating effect was not obvious in Kuo and Vasishth's（2006）experiment.A few L2acquisition studies of Chinese RCs took the DCl sequence factor into consideration.Chen（1999）used a grammaticality acceptance task to examine SU and DO RCs with the DCl－first and DCl－second sequences，and reported that，in general，for Chinese L1adults as well as for Chinese L2speakers with English and Japanese L1background，in the DCl－RC sequence，SU is easier，whereas in the RC－DCl sequence，DO is easier.But because Chen's（1999）experiments involved several other variables aside from extraction and DCl sequence，including modifying position，animacy，and different L1background，and her statistical tests did not take all factors simultaneously into consideration，the result could not interpreted decisively.Xu（2012）conducted a self－paced word order judgment task in which participants read complete sentences on the screen，with half of the RC－containing items presented in the right word order and the remaining items in wrong word order.With a 2×2design，crossing DClsequence（before or after RC）and extraction（SU vs.DO），she reported that DCl－SU was the easiest among all.Among the processing theories that she tested（LDT，SDT，CWO，and PST），the strongest support was found for the SDT.Xu（2012）concluded that L2learners of Chinese used a processing strategy based on structural integration.

Table 3gives a summary of representative studies that confirmed or rejected some of important hypotheses outlined in this paper，focusing on a variety of target languages，task types，and results with exceptions.

Table 3:Acquisition studies of relative clauses

（Continued）

OPEN ISSUES

As seen by Table 3，in English RCs，while the preference of SU is reported by studies involving a variety of tasks，controversies were more often found in whether extraction in lower positions follows the AH hierarchy.In addition，although the complexity of center－embedding was attested in several studies，Gibson et al.'s（2005）information flow hypothesis suggested that a matrix subject position in English may be easier from a semantic interpretation perspective.At the same time，while a rich number of distinctive theories have been proposed，Diessel and Tomasello's（2000；2005）argued that acquisition patterns are multifactorial.Studies in East Asian languages are particularly valuable in assessing different hypotheses，but little is certain regarding the behavioral patterns of relativization in those languages.

Methodologies issues in interpreting the data in AH implications

Seen from the existing literature，there are some ambiguities in interpreting the implication of the AH in learner language.As Hawkins（1987，2007）points out，some acquisition studies do not take the relationship in the right way（Hawkins 2007:340）.Specifically，the hierarchy does not necessarily predicate that learners would use SU relatives first or make fewer errors relativizing on Subject than Object，but rather that in any given stage of acquisition，the learner language or IL should be like a type of the natural languages，i.e.displaying features of markedness within the constraint of the universal（Hawkins 2007:341）.In some previous studies，accuracy rate is thought to be a key measure to test the applicability of the NPAH ，with the assumption that the same processing ease responsible for the NPAH might be reflected in“greater accuracy”in more accessible positions（Gass 1979:339）.However，while accuracy could be one way in which the AH can manifest in learner language，it is possible that the ease of SU relativiztion could be observed in other aspects.For instance，in Diessel and Tomasello's（2005 ）English and German L1 acquistion study，they found that children in elicited oral imitation tasks often changed IO／OPrep type of RCs into SU RCs.Interestingly，this pattern of changing other RC types into SUs was reported in Ozeki and Shirai（2007）in Japanese L2acquisition，Xu（in press－b）（in Chinese L2），and Xiao and Lü（2005）in English L2studies.Meanwhile，all these studies reported little to no instances in which RC type change went in the other direction，suggesting that SU relativization is indeed preferred in different languages.Also，Xu（in press－b）questions if“target－like responses”can always be directly interpreted as evidence of acquisition when it comes to pronoun vs.gap strategies，since in her study，more target－like pronoun strategies in OPrep was attributed to more processing cost（thus supporting Hawkins'filler－gap domain theory），rather than acquisition.

The other methodological issue also concerns data analysis.Eckman（2007:325－326）critiqued that while several recent published studies relied solely on group data（e.g.Jeon and Kim 2007；Kanno 2007），the applicability of the AH to the learner's interlanguage in RC acquisition would be best performed when based on individual data，because one cannot assume the IL of all learners to be exactly the same.Eckman（2007:323）asked us to imagine a scenario in which learner A produced SU，DO，IO RCs correctly and OPrep，Gen，and OComp incorrectly，and if learner B's responses were exactly the contrary，then one would end up with 50%accuracy rate throughout all positions if conclusions were derived through group data，while one would be blindfolded to the fact that learner A conformed to the AH and B violated it.Although Hamilton（1994）and Doughty（1988，1991）as examples of earlier studies analyzed group performance as well as individual learners'data，and Xu（in press－b）more recently reported individual learners'RC－forming strategies in addition to group tendencies，discussions on individual performance are sometimes overlooked.

Finally，the applicability of the AH to East Asian languages was challenged from a theoretical perspective.Comrie（2002）suggests that putative“RCs”in Japanese may merely be a type of general noun－modifying clause because they do not exhibit the full typological and syntactic characteristics of RCs.Several later studies（e.g.Yip and Matthews 2007；Ozeki and Shirai 2007），while interpreting their data as being inconsistent with the AH，used Comrie's（2002 ）hypotheses to explain away the contrast between acquisition patterns of putative RCs in East Asian languages and well－attested SU＞DO＞IO hierarchy in English RCs.However，two clarifications are needed.First，while generally relying on Japanese examples and suggesting that RCs in East Asian languages may be similar，Comrie（2002）did not specify to what extent this non－RC proposal may extend to languages such as Cantonese or Mandarin Chinese.Second，whether RCs exist in those languages must be explored using a more formal approach such as testing if the“gap”is indeed a proin Comrie's analysis，if syntactic movement constraints are observed，and if typological and structural differences between noun－modifying clauses and RCs exist.Such theoretical analyses are generally unavailable in SLA studies，including those mentioned above.Therefore，SLA researchers should refer to theoretical works if observed results are to be explained by a non－RC proposal.For instance，in Chinese，ample evidence exists that a pro analysis cannot explain all types of relativization，and syntactic constraints such as Subjacency，the Complex Noun Phrase Constraint，the Sentential Subject Condition，and the Adjunct Condition all hold in Chinese（Ning 1993；Aoun and Li 2003；Huang et al.2009），confirming that there is syntactic movement.Therefore，any inconsistency between the AH or other RC hypotheses and processing／acquisition difficulties should not be attributed to the absence of RCs in Chinese.

Task variations

Some of the earliest RC studies were critiqued for shortcomings in design.For instance，Keenan and Hawkins（1987）and Hamburger and Crain（1982）critiqued that earlier L1acquisition studies using prop，actingout tasks did not always meet the felicity conditions:in a prompt sentence of Thehorsethatjumpedover thepigkickedthedog，only one horse was available for children to act out the scenario described in the prompt，making it unnecessary for them to interpret the RC as a syntactic unit.Hamburger and Crain（1982）argued that one must devise a situation in which the RC is indeed needed to communicate to test children's grammatical knowledge，and based on such reasoning，Keenan and Hawkins（1987）rejected Sheldon's（1974）parallel function analysis.

Experimental methodologies in L1and L2acquisition are not always identical.For instance，the actingout task is generally not used in SLA studies，possibly due to its play－like nature.Izumi（2003）found that comprehension／receptive tasks are often used in L1studies while L2studies rely more on production tasks.In his study，the scores on the production test were much lower than those on the two reception tests，and the differences in scores were much more acute in the production test.Hakansson and Hansson's（2000）study on RCs revealed that comprehension and production are different but mutually dependent processes.Several other studies have pointed out that the role of task effect，and results from different tasks might even support different theories of RC processing or acquisition（Prideaux and Baker 1986；Tang and Xu 2011）.Therefore，while the preference for using production tasks may be pertinent to the SLA research's origin in more pedagogical concerns，it is important that future studies use a variety of tasks when appropriate.The present review indicates that SLA studies of RCs predominantly use a written sentence combination task since Gass（1979）.While this task has a number of benefits，one would heed Juffs'（2007）insightful comments that if the AH is indeed motivated by processing issues，evidence should be found not only in offline studies but also in experiments using online techniques.Tang and Xu's（2011）study made significant improvement in design by incorporating a“think aloud”protocol and immediate retrospective interview into this task，so that the underlying reasons for learners'responses can be discussed.But while their qualitative data revealed learners'strategies in coping with the task，such a procedure cannot replace techniques probing real－time processing using more sensitive measures such as reaction time.In this aspect，Packard（2008）and Xu（2012）made some potentially fruitful efforts.But both studies have some limitations，the former with a lack of a complete counter－balancing design，and the latter used a complete sentence rather than wordby－word reading technique due to the constraint of participants'proficiency.While several studies were conducted by scholars in China to examine RC acquisition，research using psycholinguistic measures seems particularly lacking.

In processing studies of RCs using RT measures，a number of issues complicate a more straight－forward interpretation.First，a number of semantic and syntactic factors may interact.For instance，while both intuitively sound，the information flow hypothesis and the perceptual difficulty hypothesis give conflicting predictions regarding subject－modification in English.Other factors such as animacy must also be strictly controlled across extraction conditions for conclusions to be valid.Second，temporary ambiguities exist in several types of RCs.Take Chinese for instance，Kuo and Vasishth（2006）and Chen et al.（2010:7－8）pointed out that the majority of RC－like sequences in Chinese are not RCs.For instance，strings such as tisheng qiye de jingzhengli‘to increase the company's competitiveness'have a V－N－de－N sequence，similar to that of SU RCs，and they occur much more frequently than real RCs in corpus.Thus，if structural frequency affects how readers would initially interpret the string（based on experience），both SU and DO are ambiguous until the head noun is processed，causing slowdowns due to reanalysis at the head noun.Gibson and Wu's（2011）supportive context presents an example of a more rigid design，in which such ambiguities can be ruled out.However，while this approach may be implementable in L1adult processing，such a design may not be appropriate for L2studies，as extensive texts would presumably impose higher demands on learners'proficiency and add more burdens for learners to complete the task successfully.

Other issues of data type in L2studies

Ozeki and Shirai（2007）noted that studies on RC acquisition tended to rely more on experimental rather than naturalistic data，although recent efforts were made in their study as well as in Yip and Matthews（2007）and Dai et al.（2008）with both oral and written data.However，it could be difficult to follow these examples，due to the complexity of the RC structure and the infrequency of usage in natural discourses.For instance，the predication of RCs can generally be expressed by two related simple clauses，as shown by the sentence combination task design.A number of other factors（e.g.animacy）also affect the felicitous conditions of their usage.Therefore，unless there is some unambiguous record of first emergence of RCs in a learner's IL，or unless L2speakers'use of RCs can be compared to that of L1's when corpora of both L1 and L2groups consist of data in comparable genre，style，and topic，one cannot decisively interpret naturalistic data as evidence of（in）acquisition.

In experimental studies，accuracy rate is most often used.However，several studies relied on numerical tendencies to interpret the result as evidence for or against certain theories when statistical pairwise comparisons are not available（e.g.Dai et al.2008；Dai 2010）.Similarly，although pairwise comparisons were conducted in Izumi（2003），several of the author's claims were made based on numerical differences only.A particular theory can be better supported when both the main effect and pairwise comparisons are significant，as in Diessel and Tomasello（2005）and Xu（2012）.On the other hand，Eckman（2007）warns that in interpreting the AH's implications，as long as there is no evidence of IO＞DO，DO＞SU，etc.，results were not really contradictory to the AH generalization.These factors all need to be considered in interpreting statistical data.

To summarize，future studies on RC acquisition can attend to the following issues:

（a）Use a combination of production vs.reception tasks；

（b）Use a combination of offline versus online tasks；

（c）To the extent possible，use both naturalistic and experimental data；

（d）Use caution in quantitative and qualitative data analysis，such as attending to statistical significance，individual data，and analyzing learners'RC－forming strategies separately from other errors；

（e）Make connections to L1acquisition and processing studies.As L2acquisition is undeniably affected by inherent structural difficulties，which might also manifest in children's acquisition as well as adults'processing，such comparisons are needed for researchers to understand the complex mental representations of L2speakers.

Given the nature of SLA studies，it is desirable to have native speaker control groups as in Packard（2008）and／or different proficiency groups of learners as in Izumi（2003），Ozeki and Shirai（2007）.Generally，learner performance that systematic deviates from the target－like form in RC studies are taken to be a reflection of grammatical incompetence while no studies that I am aware of specifically differentiate“mistakes”versus“errors”based on Corder's（1967）distinctions.Still，comparison with an L1adult group performance can be informative in some tasks.In addition，research aiming at finding the developmental patterns of the L2 should include learners in different proficiency groups，as it will not only enable us to investigate acquisition orders，but may also reveal different processing or task－coping strategies.For instance，Tang and Xu（2011）found that the lower proficiency group completed the sentence combination task by imitating examples，while intermediate leaners relied primarily on looking for the relative clause head，and the advanced levels used a variety of techniques.

Finally，a few issues particularly merit attention in the studies of Chinese RCs:

（a）Whether SVO can be considered the canonical word order in Chinese and how“experience”affects Chinese RC processing.Specialized research on corpus frequencies is needed to investigate the validity of different experience－based accounts in Chinese；

（b）To what extent animacy affects Chinese RC processing.While the role of animacy in relativization is already discussed in English，Japanese，and Korean，and Lin and Garnsey（2011）suggest that animacy affects Chinese RCs too，much remains to be explored regarding the interaction of animacy and syntactic relations.

（c）To what extent DCl sequence affects RC processing，and why.While a few studies made attempts，results from Kuo and Vasishth（2006）and Wuet al.'s（2009）remain ambiguous，and Chen（1999）did not discuss any potential motivations although she reported interaction effect of DCl with extraction types.Xu（in press－a）and Xu（2012）included L2participants only.These studies suggest that DCl plays a role in corpus frequency as well as L1and L2processing.

（d）In L2acquisition studies，it is reasonable to hypothesize that different L1background，depending on the L1's head direction and clausal word order，may affect processing and acquisition difficulty.In this sense，Dai's（2010）study is potentially confounded by the uncontrolled factor of different L1s（English，Japanese，and Korean）.In most studies of English RC acquisition，different L1groups were discussed separately（Gass 1979，among others）.It is prudent to control L1RC typology when conditions permit.

CONCLUSION

In this paper，I laid out major theories，studies，and findings in the field of psycholinguistics and L1／L2 acquisition of RCs，with the hope that this review can encourage more researchers，particularly those in China，to engage in studies on the topic.While applied linguistic studies of RCs have a long tradition，recent development favors a cross－linguistic perspective and the use of interdisciplinary approaches and psycholinguistic techniques.For English／EFL RC studies，theoretical knowledge as well as familiarity with more recent experimental methodologies shall prove helpful.For Chinese（L2）studies，apotential challenge for researchers is the lack of proficiency－appropriate participants to carry out experiment with a solid design.But it is encouraging that research in this direction has increased in volume in recent years.One can expect psycholinguistically－oriented research and studies on Chinese RC acquisition to burgeon in the future and to offer enlightening findings for scholars worldwide.

NOTES

1 Hawkins（1999:255）mentions that the number of nodes is only one index，albeit a fundamental one，and correlating properties and additional morpho－syntactic and semantic operations such as case and theta－role assignment bearing other potential processing cost co－occur.Therefore，the identical node quantities of IO，OPrep，Gen－SU extractions are not to be taken as equal processing cost in this model.

2 Context is provided by sentences describing a scenario where two different objects can be distinguished.（e.g.“On a highway，a motorcycle chased a car through heavy traffic；another car saw the situation，and then chased the motorcycle”）.Then，in the scenario，a character said“I heard that a high school student was driving one of the cars and a middle－aged woman was driving the other.Which car was the high－school student driving？”

3 Yip and Matthews（2007）argued that（ⅲ ）can be ambiguous between a main clause or a complex NP with a relative clause.The authors also suggested that the same sequence can be an internally headed Object RC，i.e.（ⅳ）or a head－final RC，i.e.（ⅴ）

（ⅲ ）Alicia waak6go2 di1 je5 Alicia draw DEM Cl things

a.Alicia drew those things

b.［These are］the things that Alicia drew”.

（ⅳ ）［NP／SAlicia waak6go2di1je5］（ⅴ ）［NP［SAlicia waak6 ＿［go2di1je5］］］（Examples from Yip and Mattews 2007:293.）

Meanwhile，SU RCs in Cantonese cannot be IHRCs，due to the initial VO sequence，with an obvious“gap”in the pre－V subject position.

4 Examples of Object－modifying SU and DOs are illustrated below，respectively as（ⅰ ）and（ⅱ）

（ⅰ ）yiyuan zhuangdao－le gouyin yuanzhang de shaonyu.Congressman bumped.into seduce dean Rel young.lady‘The congressman bumped into the young lady that seduced the dean'

（ⅱ ）yiyuan zhuangdao－le yuanzhang gouyin de shaonyu Congressman bumped.into dean seduce Rel young.lady.

‘The congressman bumped into the young lady that the dean seduced'

Gibson and Wu（2011）argued that the reader would interpret the noun after the main verb in（ⅱ ）‘dean'initially as the direct object of the main clause，and this analysis has to be abandoned at the encountering of the RC verb‘seduce'；this causes a garden－path effect.No such temporary ambiguity is involved in（ⅰ ），as the initial N－V－V sequence suggests that the second verb‘seduce'leads to an RC.

Aarts，F(xiàn).and E.Schils.1995.‘Relative clauses，the accessibility hierarchy and the contrastive analysis hypothesis，'IRAL International Review of Applied Linguistics 33／1:47－63.

Aoun，J.and Y.Li.2003.Essays on the Representational and Derivational Nature of Grammar:The Diversity of Wh－Constructions.Cambridge，Mass:MIT.

Bever，T.G.1970.‘The cognitive basis for linguistic structures'in J.R.Hayes（ed.）:Cognition and the Development of Language.NY:J.Wiley，pp.279－360.

Brown，H.D.1971.‘Children's comprehension of relativized English sentences，'Child Development 42／6:1923－36.

Chao，Y.R.1968.A Grammar of Spoken Chinese.Berkeley and Los Angeles:University of California Press.

Chen，B.，A.Ning，H.Bi，and S.Dunlap.2008.‘Chinese subject－relative clauses are more difficult to process than object－relative clauses，'Acta Psychologica 129:61－65.

Chen，Chun－Yin.1999.‘The second language acquisition of Chinese relative clauses，'The World of Chinese Language 94:59－76.

Chen，Z.，Q.Li，K.Kuo，and S.Vasishth.2010.‘Processing Chinese relative clauses:Evidence for the universal subject preference，'Available at:http:∥www.ling.unipotsdam.de／～ vasishth／pdfs／Chen Li Kuo Vasishthsubmitted.pdf（accessed on 9June 2012）.

Cheng，S.1995.The Acquisition of relative clauses in Chinese.M.A.thesis，National Taiwan Normal University.

Cho，S.1999.The acquisition of relative clauses:Experimental studies on Korean.Unpublished Ph.D.Disserta－tion，University of Hawaii at Manoa.

Chomsky，N.1977.‘On wh－movement'in P.Culicover，T.Wasow，and A.Akmajian（ed.）:Formal Syntax.NY:Academic Press，pp.71－132.

Clahsen，H.and C.Felser.2006.‘Grammatical processing in language learners，'Applied Psycholinguistics 27:3－42.

Clancy，P.M.，H.Lee，and M.Zoh.1986.‘Parsing strategies in the acquisition of relative clauses:Universal principles and language－specific realizations，'Cognition24:225－62.

Comrie，B.2002.‘The case of relative clauses'in A.G.Ramat（ed.）:Typology and Second Language Acquisition.Berlin:Mouton de Gruyter，pp.19－37.

Comrie，B.2008.‘Prenominal relative clauses in verb－object languages，'Language and Linguistics 9／4:723－33.Accessed on July 9th，2012from http:∥iacliscll.ling.sinica.edu.tw／eip／FILES／journal／2008.10.29.27180117.0968969.pdf.

Corder，S.P.1967.‘The significance of learner's errors，'International Review of Applied Linguistics in Language Teaching5／4:161－70.

Dai，Y.C.2010.‘An investigation of the relative clause acquisition by learners of Chinese as a second language，'Journal of Ocean University of China（Social Sciences）06:85－91.

Dai，Y.C.，Hu，H.L，＆Zhang，X.W.2008.‘A confirmatory study of the hypotheses on relative clause acquisition in the case of Chinese learners of English，'Journal of PLA University of Foreign Languages 31／5:71－77.

Del Gobbo，F(xiàn).2003.Appositives at the Interface.PhD dissertation，University of California Irvine.

Diessel，H.and M.Tomasello.2000.‘The development of relative clauses in spontaneous child speech，'Cognitive Linguistics 11:131－51.

Diessel H.and M.Tomasello.2005.‘A new look at the acquisition of relative clauses，'Language 81:882－906.

Doughty，C.1988.The Effect of instruction on the acquisition of relativization in English as a second language.Unpublished Ph.D.dissertation，University of Pennsylvania.

Doughty，C.1991.‘Second language instruction does make a difference，'Studies in Second Language Acquisition 13:431－69.

Dryer，M.S.2005.‘Relationship between the order of object and verb and the order of relative clause and noun'in M.Haspelmath，M.S.Dryer，D.Gil，and B.Comrie（eds）:The World Atlas of Language Structures.Oxford ＆ New York:Oxford University Press，pp.366－69.

Eckman，F(xiàn).R.1977.‘Markedness and the contrastive focus analysis，'Language Learning27／2:315－30.

Eckman，F(xiàn).R.1984.‘Universals，typologies，and interlanguage'in W.E.Rutherford（ed.）:Language Universals and Second Language Acquisition.Amsterdam:John Benjamins，pp.79－105.

Eckman，F(xiàn).R.1991.‘The Structural Conformity Hypothesis and the acquisition of consonant clusters in the interlanguage of ESL learners，'Studies in Second Language Acquisition 7:289－307.

Eckman，F(xiàn).R.1996.‘A functional－typological approach to second language acquisition theory，'in W.C.Ritchie and T.K.Bhatie（ed.）:Handbook of Second Language Acquisition.New York:Academic Press，pp.195－211.

Eckman，F(xiàn).R.2007.‘Hypotheses and methods in second language acquisition:Testing the noun phrase accessibility hierarchy on relative clauses，'Studies in Second Language Acquisition29:321－27.

Eckman，F(xiàn).R.，L.Bell，and D.Nelson.1988.‘On the generalization of relative clause instruction in the acquisition of English as a second language，'Applied Linguistics 9／1:1－20.

Ford，M.1983.‘A method for obtaining measures of local parsing complexity throughout sentences，'Journal of Verbal Learning and Verbal Behavior 22／2:203－18.

Frazier，L.1987.‘Syntactic processing evidence from Dutch，'Natural Language and Linguistic Theory 5／4:519－59.

Friedmann，N.and R.Novogrodsky.2004.‘The acquisition of relative clause comprehension in Hebrew:A study of SLI and normal development，'Journal of Child Language 31:661－81.

Gass，S.M.1979.‘Language transfer and universal grammatical relations，'Language Learning29／2:327－44.

Gass，S.M.1982.‘From theory to practice'in M.Hines and W.Rutherford W（ed.）:On TESOL'81:Selected papers of the fifteenth annual Conference of Teachers of English to Speakers of Other Languages. Washington DC:TESOL，pp.129－39.

Gibson，E.1998.‘Linguistic complexity:locality of syntactic dependencies，'Cognition68:1－76.

Gibson，E.2000.‘The dependency locality theory:A distance－based theory of linguistic complexity'in Y.Miyashita，A.Marantz，and O'Neil（ed.）:Image，Language，Brain.Cambridge，MA:MIT Press，pp.95－126.

Gibson，E.and H.l.Wu.2011.‘Processing Chinese relative clauses in context，'Language and Cognitive Processes.Epub ahead of print 07March 2011.DOI:10.1080／01690965.2010.536656.

Gibson，E.，T.Desmet，D.Grodner，D.Watson，and K.Ko.2005.‘Reading relative clauses in English，'Cognitive Linguistics 16／2:313－53.

Hakansson，G.and K.Hansson.2000.‘Comprehension and production of relative clauses:A comparison between Swedish impaired and unimpaired children，'Journal of Child Language 27:313－33.

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（Vol.2）.New Brunswick，NJ:Associate for Computational Linguistics，pp.159－66.

Hamburger，H.and S.Crain.1982.‘Relative acquisition'in S.A.Kuczaj（ed.）Language Development:Syntax and Semantics.Hillsdale，NJ:Lawrence Erlbaum，pp.245－73.

Hamilton，R.L.1994.‘Is implicational generalization unidirectional and maximal？Evidence from relativization instruction in a second language，'Language Learning44／1:123－57.

Hawkins，J.A.1987.‘Implicational universals as predictors of language acquisition，'Linguistics 25:453－73.

Hawkins，J.A.1999.‘Processing complexity and filler－gap dependencies across grammars，'Language 75／2:244－85.

Hawkins，J.A.2007.‘Acquisition of relative clauses in relation to language universals，'Studies in Second Language Acquisition 29／2:337－44.

Holmes，V.M.and J.K.O'Regan.1981.‘Eye fixation patterns during the reading of relative clause sentences，'Journal of Verbal Learning and Verbal Behavior 20／4:417－30.

Hsiao，F(xiàn).and E.Gibson.2003.‘Processing relative clauses in Chinese，'Cognition27／3:3－27.

Hsu，C.，G.Hermon，and A.Zukowski.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.－T.，A.James，Y.－H.Li，and Y.Li.2009.The Syntax of Chinese.London:Cambridge University Press.

Hyltenstam，K.1984.‘The use of typological markedness conditions as predictions in second language acquisition:The case of pronominal copies in relative clauses'in R.Anderson（ed.）:Second Language:A Cross－linguistic Perspective.Rowley，MA:Newbury House，pp.39－58.

lshizuka，T.，K，Nakatani，and E.Gibson.2003.Relative clause extraction complexity in Japanese.Poster presented at The 16th Annual CUNY Conference on Human Sentence Processing，Massachusetts Institute of Technology，Cambridge，MA.

loup，G.and A.Cruse.1977.‘Interference versus structural complexity as a predictor of second language relative clause acquisition'in C.Henning（ed.）:Proceedings of the Second Language Research Forum.Los Angeles，UCLA.

lzumi，S.2003.‘Processing difficulty in comprehension and production of relative clauses by learners of English as a second language，'Language Learning53／2:285－323.

Jeon，K.S.and H.－Y.Kim.2007.‘Development of relativization in Korean as a foreign language:The noun phrase accessibility hierarchy in head－internal and head－external relative clauses，'Studies in Second Language Acquisition 29／2:253－76.

Juffs，A.2001.‘Psycholinguistically oriented second language research，'Annual Review of Applied Linguistics 21:207－21.

Juffs，A.2007.‘Commentary:Second language acquisition of relative clauses in languages of East Asia，'Studies in Second Language Acquisition 29／2:361－65.

Juffs，A.and M.Harrington.1995.‘Parsing effects in second language sentence processing:Subject and object asymmetries in“wh”－extraction，'Studies in Second Language Acquisition 17／4:483－516.

Kanno，K.2007.‘Factors affecting the processing of Japanese relative clauses by L2learners，'Studies in Second Language Acquisition 29／2:197－218.

Keenan，E.L.1975.‘Variation in universal grammar'in R.Fasold and R.Shuy（eds）:Analyzing Variation in English.Washington DC:Georgetown University Press，pp.136－148.

Keenan，E.L.and C.Bernard.1977.‘Noun phrase accessibility and universal grammar，'Linguistic Inquiry8／1:63－99.

Keenan，E.L.and S.Hawkins.1987.‘The psychological validity of the accessibility hierarchy'in E.L.Keenan（ed.）:Universal Grammar:Fifteen Essays.London:Routledge，pp.60－85.

Kim，Y.1987.The acquisition of relative clause in English and Korean:development in spontaneous production.Ph.D dissertation，Harvard University，Cambridge，MA.

King，J.and M.A.Just.1991.‘Individual differences in syntactic processing:The role of working memory，'Journal of Memory and Language 30／5:580－602.

Kuno，S.1974.‘The position of relative clauses and conjunctions，'Linguistic Inquiry5／1:117－36.

Kuo，K.and S.Vasishth.2006.Processing relative clauses:Evidence from Chinese.Accessed on January 10 2007from http:∥www.ling.unipotsdam.de／～vasishth／vasishth2.html.

Kwon N.，M.Polinsky，and R.Kluender.2006.‘Subject preference in Korean'in D.Baumer，D.Montero，M.Scanlon（eds）:Proceedings of the 25th West Coast Conference on Formal Linguistics.Somerville，MA:Cascadilla Proceedings Project，pp.1－14.

Lau，E.2006.The acquisition of relative clauses by Cantonese children:An experimental approach.M.A.Thesis，University of Hong Kong.

Lee，K.1991.On the first language acquisition of relative clauses in Korean:The Universal Structure of Comp.Ph.D.Dissertation，Cornell University，Ithaca，NY.

Levy，R.2008.‘Expectation－based syntactic comprehension，'Cognition106:1126－77.

Li，Q.，J.Zhang，and W.Yue.2010.‘Chinese relative clauses processing in supportive context removing ambiguity，'Studies in Literature and Language 4／1:12－9.

Lin，C.－J.2006.Grammar and parsing:A typological investigation of relative－clause processing.Ph.D.dissertation，University of Arizona.

Lin，C.－J.and T.G.Bever.2006.‘Subject preference in the processing of relative clauses in Chinese'in D.Baumer，D.Montero，and M.Scanlon（eds）:Proceedings of the 25th West Coast Conference on Formal Linguistics.Somerville，MA:Cascadilla Proceedings Project，pp.254－60.

Lin，Y.and S.M.Garnsey.2011.‘Animacy and the resolution of temporary ambiguity in relative clause comprehension in Mandarin'in H.Yamashita，Y.Hirose，and J.L.Packard（eds）:Processing and Producing Head－Final Structures.London／New York:Springer，pp.241－76.

MacDonald，M.C.and M.H.Christiansen.2002.‘Reassessing working memory:A reply to Just and Carpenter and Waters and Caplan，'Psychological Review109／1:35－54.

MacWhinney，B.1977.‘Starting points，'Language 53／1:152－68.

MacWhinney，B.1982.‘Basic syntactic processes'in S.A.Kuczaj（ed.）.Language Development:Syntax and Semantics.Hillsdale，NJ:Lawrence Erlbaum，pp.73－136.

MacWhinney，B.2005.‘The emergence of grammar from perspective'in D.Pecher and R.A.Zwaan（eds）:The Grounding of Cognition:The Role of Perception and Action in Memory，Language，and Thinking.Mahwah，NJ:Lawrence Erlbaum Associates，pp.119－223.

MacWhinney，B.and C.Pleh.1988.‘The processing of restrictive relative clauses in Hungarian，'Cognition 29／2:95－141.

Marinis，T.2003.‘Psycholinguistic techniques in second language acquisition research，'Second Language Research 19／2:144－61.

Mecklinger，K.，H.Schriefers，K.Steinhauer，and A.D.Friederici.1995.‘Processing relative clauses varying on syntactic and semantic dimensions，'Memory and Cognition 23／4:477－94.

Miyamoto，E.and M.Nakamura.2003.‘Subject／object asymmetries in the processing of relative clauses in Japanese'in G.Garding and M.Tsujimura（eds）:Proceedings of the 22nd West Coast Conference on Formal Linguistics.Somerville，MA:Cascadilla Press，pp.342－55.

Ning，C.1993.The overt syntax of relativization and topicalization in Chinese.Unpublished Doctoral Dissertation，University of California Irvine.

O'Grady，W.1997.Syntactic Development.Chicago:University of Chicago Press.

O'Grady，W.1999.‘Towards a new nativism，'Studies in Second Language Acquisition 21／4:621－33.

O'Grady，W.，Lee M.，and M.Choo.2003.‘A subject－object asymmetry in the acquisition of relative clauses in Korean as a second language，'Studies in Second Language Acquisition 25／3:433－48.

Ozeki，H.and Y.Shirai.2007.‘Does the noun phrase accessibility hierarchy predict the difficulty order in the acquisi－tion of Japanese relative clauses？'Studies in Second Language Acquisition 29／2:169－96.

Packard，J.L.2008.‘Relative clause processing in L2 speakers of Mandarin and English，'Journal of the Chinese Language teachers Association 43／2:107－46.

Pavesi，M.1986.‘Markedness，discoursal modes，and relative claus formation in a formal and an informal context，'Studies in Second Language Acquisition 8／1:38－55.

Prideaux，G.and W.Baker.1986.Current Issues in Linguistic theory（Vol.46）:Strategies and Structures:The Processing of Relative Clauses.Amsterdam:John Benjamins.

Roberts，L.2012.‘Psycholinguistic techniques and resources in second language acquisition research，'Second Language Research28／1:113－27.

Schriefers，H.，A.Friederici，and K.Kuhn.1995.‘The processing of locally ambiguous relative clauses in German，'Journal of Memory and Language 34／4:499－520.

Sheldon，A.1974.‘The role of parallel function in the acquisition of relative clauses in English，'Journal of Verbal Learning and Verbal Behavior 13／3:272－81.

Shi，F(xiàn).and B.Wen.2009.‘Overview of the second language acquisition of Chinese，'Journal of Chinese Linguistics 37／1:130－44.

Slobin，D.l.and T.G.Bever.1982.‘Children use canonical sentence schemas:a cross－linguistic study of word order，'Cognition12／3:229－65.

Su，Y.－C.2004.‘Relatives of Mandarin children，' ［Electronic version］.Paper presented in Generative Approaches to Language Acquisition—North America 2004，University of Hawaii at Manoa.Available at:http:∥www.ling.nthu.edu.tw／faculty／ycsu／GALANA2004.pdf（accessed on 10August 2008）.

Tang，C.X.and J.J.Xu.2011.‘The acquisition order of English relative clauses by Chinese senior high school students:a mixed－design study，'Foreign Language Teaching and Research 43／1:96－108.

Tang，T.C.1981.Studies in Chinese Syntax.Taipei:Student Book.

Tavakolian，S.1981.‘The conjoined－clause analysis of relative clauses'in S.Tavakolian（ed.）:Language Acquisition and Linguistic Theory.Cambridge，MA:MIT Press，pp.167－87.

Traxler，M.J.，R.K.Morris，and R.E.Seely.2002.‘Processing subject and object relative clauses:Evidence from eye movements，'Journal of Memory and Language 47／1:69－90.

Tsai，W.－T.D.1994.On economizing the theory of A－Bar dependencies.Unpublished Ph.D.dissertation，MIT.

Ueno，M.and S.Garnsey.2008.‘An ERP study of the processing of subject and object relative clauses in Japanese，'Language and Cognitive Processes 23／5:646－88.

de Vries M.2002.The Syntax of Relativization.Utrecht，Netherland:Lot.

Wanner，E.and M.Maratsos.1978.‘An ATN approach in comprehension'in M.Halle，J.Bresnan and G.Miller（eds）:Linguistic Theory and Psychological Reality.Cambridge，MA:MIT Press，pp.119－61.

White，L.2003.Second Language Acquisition and Universal Grammar.Cambridge:Cambridge University Press.

Wu，F(xiàn).，E.Kaiser，and E.Anderson.2009.The Effect of Classifiers in Predicting Chinese Relative Clauses ［manuscript］.Available at:http:∥www－bcf.usc.edu／～emkaiser／WuKaiserAndersen－2009－WECOLClassifiers.pdf（accessed on 22June 2012）.

Wu，F(xiàn).，E.Kaiser，T.Haskell，and E.Anderson.2007.‘Factors in the use of relative clauses in Mandarin:Behavioral and corpus evidence，'Interdisciplinary Approaches to Relative Clauses（REL07）.University of Cambridge.

Xiao，Y.N.and Lü，J.2005.‘Empirical studies of Chinese students'acquisition of English relative clauses，'Foreign Language Teaching and Research 37／4:259－64.

Xu，Y.2012.Processing Relative Clauses in Chinese as a Second Language.［manuscript］.

Xu，Y.in press－a.‘Acquisition of Chinese relative clauses at the initial stage'to appear in I Kecskes（ed.）:Research in Chinese as a Second Language:The Acquisition of Language and Culture.DeGruyter Mouton.

Xu，Y.in press－b.‘Relativization strategies Chinese as a foreign language learners'interlanguage:Applicability of the Noun Phrase Accessibility Hierarchy'to appear in James Huang and Feng－h(huán)si Liu（eds）:Peaches and Plums.Taipei:Academia Sinica.

Yip，V.and S.Matthews.2007.‘Relative clauses in Cantonese－English bilingual children:Typological challenges and processing motivations，'Studies in Second Language Acquisition 29／2:277－300.

Zhao，Y.2011.‘A tree in the wood:A review of research on L2Chinese acquisition，'Second Language Research27／4:559－72.