turkishenglish.com - Comparison of relative clauses in ESL students



            Any literature review related to the study of relative clauses in second language acquisition must cover an extensive amount of territory.  This is necessary because of the various distinct sub-disciplines related to language acquisition.  There has been a substantial amount of work done on relative clauses from a number of different perspectives, but relatively little has been done on exactly how it is that second language learners use relative clauses in their production of English as a second language.  Nevertheless, a great deal can be learned by taking a careful look at the work that has been done.  This chapter will review literature from each of the following areas:  subordination in general, the function of relative clauses in English discourse, second language acquisition theory, and specific studies, both contrastive and otherwise, dealing with relative clauses.  


Subordination in General

            'Subordination' is a linguistic category that has generated a lot of controversy.  It has been somewhat unclear what constitutes 'subordination' and even more importantly how subordination relates to discourse structure.  Haiman and Thompson (1984) attempt to establish some criteria for what the linguistic characteristics of subordination are in an attempt to define subordination more precisely.   Matthiessen and Thompson (1988) distinguish two types of subordinate structures: those they consider to be grammatically embedded and those which they see as examples of clause combining.  They further divide the latter into examples of parataxis and hypotaxis.   These types of subordination typically relate propositions in different ways.

            Subordination has been used as one of the linguistic signals distinguishing foreground information from background information in discourse.  Tomlin (1985) claims that the correlation between foreground and independent clauses, on the one hand, and between background and dependent clauses, on the other, is valid for English narrative.  But, he does acknowledge exceptions, some of which can be accounted for by principles of discourse organization.  Furthermore, it has been noticed that this distinction between foreground and background information, while it may be grossly correct (Givon 1987:176-177), is probably made up of more subtle distinctions than is normally recognized, suggesting these polar opposites be viewed as a continuum instead of a binary system.  Furthermore, Wald (1987:486) questions to what extent the traditional correlation between background information and subordination is cross-linguistically valid.  He notes that in many languages it is difficult to decide whether to use a main clause or a relative clause when translating a particular predication.  Indeed, some pieces of information, which are grammatically coded as subordinate in one language, cannot be subordinated grammatically in another (e.g., Zhao 1989).   

            Subordination is not a phenomenon that is utilized with the same frequency in every language.  Chafe (1982) notes that 34% of the idea units in spoken English discourse are subordinate and in written text the number was even higher - 46%.  Mithun (1984) illustrates the low frequency with which some unrelated polysynthetic languages use subordination in oral narrative.  She found that the amount of subordination in Gungwinggu, Kathlamet, and Mohawk ranged from 7% to 2% - quite a contrast with the English usage.


Relative Clauses

            The surface structure of relative clauses is not the focus of this thesis, so relatively little space will be allotted to structural description.  Languages of the world differ along the following parameters with respect to the surface structure of the relative clause: (1) position with respect to the Head Noun, (2) adjacency to Head Noun, (3) occurrence of pronominal form marking coreferential NP, (4) pronominal morphology, (5) repetition of pronominal reflex, (6) finiteness of verb in embedded clause (cf. Keenan 1985).

            An exhaustive treatment of the functions of relative clauses in English discourse has not been done to the best of my knowledge.  Generally, it has been assumed that realtive clauses (RCs) encode presuppositional background information whose function is to uniquely identify referents (Givon 1979, 1985).  There is, however, some evidence in the literature which suggests this type of subordination is not easy to classify as to its function in discourse.  Some of this evidence comes from cross-linguistic examinations of the phenomenon which have indicated that focused information is not always handled by the same syntactic structures in every language.  Other evidence comes from studies of English discourse.  We will turn to this evidence after a brief excursion to the NP Accessibility Hierarchy.

            Keenan (1975), a study of relativization in discourse, is related directly to Keenan and Comrie (1977) and their now famous 'NP Accessibilty Hierachy.'  This principle basically states that there is a hierarchy of grammatical functions and that the closer one gets to the left side of it the more accessible to relativization the NP becomes.  The grammatical relations are Subject (S), Object (O), Indirect Object (IO) Object of Preposition (OPREP), Genitive (GEN) and Object of Comparative (OCOMP).

                        S > O > IO > OPREP > GEN > OCOMP

This means that it is somehow easier to make a subject the coreferential element in a relative clause than it is to make a Genitive the coreferential element in a relative clause.  (Throughout this thesis I will refer to this coreferential NP in the relative clause as NPRel).  This entails, he supposes, that relative clauses whose NPRel functions as subject will be far more frequent than any other grammatical function.  

            To substantiate the validity of the AH, Keenan (1975) does a text count and finds that indeed 46.16% of the relative clauses in his written corpus have an NPRel which is a subject.  He interprets this preponderance of S relatives as support for the AH since those positions on the hierarchy that are easier to relativize should be more frequent.  Unfortunately, the study only considered RCs with definite Heads, and since RCs with indefinite Heads make up almost half of all the RCs in a text (De Haan 1987), this may not be a very good characterization of the distribution of the grammatical relations (GR) of the NPRel in text.  It would be more convincing if it could be demonstrated that the percentage of Subjects that were relativized is higher than the percentage of Indirect Objects relativized since every RC has a subject and hence there is more opportunity for such relativization.  Fox (1987) argues that for this reason Keenan’s study does not do justice to relative clause usage in discourse.  In fact, she goes even further by suggesting that the NP Accessibility Hierarchy be changed to reflect the Absolutive Hypothesis, whereby intransitive subjects (S relatives) and patients (O relatives) precede transitive subjects (A relatives) in the hierarchy.  She makes this claim because intransitive subjects and objects make up the vast majority of NPRels, while Agent relatives are fairly rare. Ziv (1975)  discusses what types of information may be instantiated by the relative clause.  In particular, she posits a relationship between the position of an RC and its communicative function.  She deals only with indefinite Head Nouns in this study because there is a restriction on the extraposition of RCs with definite heads, which has been noted by others as well (Givon 1990).  

            (1)  The man who was standing on the corner fell down.

            (2)  *The man fell down who was standing on the corner.

RCs with indefinite Head Nouns serve some sort of identifying or backgrounding function when unextraposed, but they may be the main assertion of the sentence when extraposed.  The following examples are from Ziv (1975):


            (3a)  A man who was wearing very funny clothes just came in.

            (3b)  A man just came in who was wearing very funny clothes.


            (4a)  A letter which practically cancels the deal just got to my office.

            (4b)  A letter just got to my office which practically cancels the deal.


In the above sentences, Ziv contends that the predication in the extraposed relative clauses is the main assertion.  To test this, she uses a question tag to determine what the main assertion is.


            (5a)  A man who was wearing very funny clothes just came in, didn't he, Mary?

            (5b)  *A man who was wearing very funny clothes just came in, wasn't he, Mary?

            (6a)  A man just came in who was wearing very funny clothes, wasn't he, Mary?

            (6b)  ?A man just came in who was wearing very funny clothes, didn't he, Mary?


Unfortunately, Ziv’s study was not based on real discourse, but it seems intuitively that (6a,b) are somewhat more acceptable than (5b), which strikes a native speaker as totally ungrammatical.  That RCs should be used in this way to encode the main assertion of a sentence should not strike us as completely foreign since many linguists have commented on the tendency of new or important information to appear in sentence final position. This is the familiar theme/rheme or topic/comment description of information flow, and Ziv uses it as a final piece of evidence for her thesis which is that extraposed relative clause serve to highlight information.

            Ziv posits that if extraposed RCs can convey important or highlighted information, then they should not be amiable to propositions which do not lend themselves to this status.  She uses the term "rhematicity" to refer to propositions encoding new, important, or surprising information.


            (7a)  A man just came in who was wearing very funny clothes.

            (7b)  ?A man was wearing very funny clothes who just came in.


            (8a)  A girl is studying with me who has an IQ of 200.

            (8b)  ?A girl has an IQ of 200 who is studying with me.


The second of each of these pairs of sentences strike us as a little odd, and according to Ziv (1975: 573) this is because "... one of the propositions, due to its content, will function as background, while the other proposition will express the main intention the speaker wishes to convey" (emphasis in original).   In these sentences, "having a high IQ" and "wearing funny clothes" are more unexpected or unusual while "studying" and "coming in" are less surprising and seem to be background information.  Thus, putting this background information into an extraposed RC, which is normally reserved for highlighted information, causes a pragmatic clash.  Of course, what she is arguing for is a match between the type of information conveyed and the syntactic structure which instantiates that information. In her examples, the RC is still performing the semantic function of restriction, but it is pragmatically focused.

            Du Bois (1980), working on referential tracking in discourse, notes that the content of the relative clause is important in other respects as well.  In his analysis of the pear stories, he found that not all first mentions of participants or props were indefinite and that one of the most common types of definite initial mentions was to have a definite noun phrase followed by a relative clause.


            (9)  and she knocks the hat he's wearing [1st] off on the ground, (Speaker 7)

                                                                                                (Du Bois 1980:222)

In this example, the information in the relative clause makes it possible to use the definite NP in this initial mention because it provides enough information to uniquely identify the referent.  The information in the relative clause is unremarkable since people who wear hats generally only wear one and it is not extremely abnormal to wear a hat.  However, according to Du Bois, there were no examples of relative clauses which introduce a new referent with new and important information because this information cannot be presupposed and is too important to be presented in an RC.  In (10), the RC encodes important new information and so RCs of this type are not used to introduce a referent.


            (10)  The woman Bill married last night was nasty to him.


            He calls this the principle of “new information presupposition.”  Important new information is asserted and since a restrictive RC usually encodes presupposed information it is unsuitable for the task.  He says an RC may only be used to introduce a referent if the information in the RC is not too important.  The relative clauses used to introduce a referent are called "file establishing relatives."  They set up a cognitive file in the mind of the listener. 

            One of the first attempts to go beyond the traditional distinction between restrictive and non-restrictive relatives was Bernardo (1979).  Using transcripts of oral narratives, he distinguished two basic types of relative clauses - informative and non-informative.  The informative relative clauses assert information and can be replaced by a simple independent clause.  Bernardo called this the “separability test.”

            (11a)  There is a woman in my class who is a nurse.

            (11b)  There is a woman in my class.  She is a nurse.


The non-informative relatives, on the other hand, cannot be replaced by a independent clause.

            (12a)  She knocked the hat he was wearing off on the ground.

            (12b)  *She knocked the hat off.  He was wearing it.

A strong piece of evidence in support of this distinction between informative and non-informative relatives is that there are relatives which express a "next" event.  This information obviously could not be used to identify a referent since an identificatory relative links a referent to known information.  A next event is not known, but brand new.   Non-informative relatives are also divided into two different categories in this study: identificatory and specificatory.  Identificatory relatives are used to identify a referent which has already been introduced in the discourse.  Specificatory relatives are those which identify a referent not previously known to the listener.  Macaulay (1990) includes specificatory relatives with informative relatives since they do encode new information.  I take her view to be the correct one.  The reason that the specificatory relatives fail the separability test (see 12) set up by Bernardo is that the content of these RCs is presupposed, but still new.

            Another very important study on relative clauses from a discourse perspective is Fox & Thompson's (1990) analysis of conversational data (also Fox 1987).  Some surprising facts resulted from this study.  First, unlike Keenan (1975), they found that the ratio of subject relatives to object relatives was exactly 1:1.  They included RCs with indefinite Heads.  Keenan did not include these RCs and this may have caused the difference.  Second, the number of RCs used to identify a previously introduced referent was very small.  This means that RCs must serve some other discourse function.   Third, the number of relative clauses in which the NPRel functioned as the subject (A) of a transitive proposition was very small.  The bulk of the tokens (90%) were S relatives and O relatives.   The question is what were these being used for? 

            The S and P relatives had very different functions.  The S-relatives describe or characterize the referent.  This is reflected in the type of verb used.   Of these RCs, 43% used be as the main verb, and the head noun was typically non-definite (68%).


            (13)  She's married to this guy who's really quiet.

            The P relatives seemed to serve a somewhat different function.  They tend to

"anchor" the referent, that is, make it relevant to the discourse at the point at which it is introduced.  This is a type of grounding (Prince 1981) in which one referent is linked to another and the NP to which it is linked is GIVEN (Chafe 1976).

            There are actually three types of grounding described by Fox and Thompson.  The first is main clause grounding in which the main clause locates a referent in the ongoing discourse.  In this way, a referent such as the object is linked to another referent.


            (14)  He's got a spring [that comes, way up]


In this example, spring is linked to he and is in this way related to the discourse.  Fox and Thompson note that the matrix verbs are often "semantically neutral" such as have or have got.  This type of construction leaves the relative clause free to make an assertion about the referent.  Comparing the two propositions in the above example, it is difficult to say that the RC is simply background unimportant information.

            The second type of grounding is  anchoring.  Prince (1981:236) defines it this way: "A discourse entity is anchored if the NP representing it is LINKED, by means of another NP, or "Anchor", properly contained in it, to some other discourse entity."


            (15) - (talking about upkeep on houses)

            But uh - the original price of it, eh - you can't even (inaud.) the original price,


            just that little screen porch alone is five hundred dollars,


            The air condish - the uh heater thing [we put in] I think was a hundred uh five six hundred dollars  (Fox and Thompson 1990:300)

Here we see that the heater thing is grounded and made relevant by the pronoun we.  That it is the pronoun which makes it relevant can be demonstrated by replacing it with a possessive pronoun. 


            (16)  Our air condish - our heater thing I think was a hundred uh five six hundred


 Fox and Thompson equate this type of relative with what Du Bois (1980:223) called a "file-establishing" relative clause, and like Du Bois and Ziv, they distinguish between RCs which have a grounding function, and hence are presupposed (as in 17),  and those which actually make assertions (18). 


            (17)  This man [who I have for linguistics] is really too much.

            (18)  There's a woman in my class [who's a nurse].


According to Prince’s (1981) definition of anchoring, there is no difference between the two types of grounding discussed by Fox and Thompson.  They are both instances of a New NP being linked to a Given NP, but different syntactic devices are used to accomplish this with ramifications as to which part of the sentence is carrying the important new information.  In main clause grounding, it is the RC which is now free to make assertions, while with RC grounding the assertion is in the matrix clause.   

            The third type of grounding discussed by Fox and Thompson is propositional linking.  They define it as, “...an entity is linked to Given referents by means of frames invoked in earlier discourse” (Fox and Thompson 1990:301). 


            (19)  The mother’s sister is a real bigot.  Y’know and she hates anyone [who isn’t a Catholic.]


In this example, the NP anyone who isn’t a Catholic, is grounded by the previous mention of bigot. 

            Fox and Thompson distinguish two functional types of RCs.  Their distinction is a function of the status of the Head Noun.  Some relative clauses characterize or describe a New referent. 


            (20)  There's a woman in my class [who's a nurse]


Other RCs identify a Given referent previously known to the listener.


            (21)  and then  the one  [that's bigoted], she's married to this guy


            Fox and Thompson use these principles to explain some of the distributional tendencies of RCs.  For example, 77% of Nonhuman Subject Heads have Object relatives.  They argue that this is so because the Subject of the sentence is typically ungrounded when it is heard since in English the predicate follows the Subject, and one of the most obvious ways to ground an object would be to link it with a human who owns or uses it.  An object relative does the job perfectly.


            (22)  probably the only thing [you'll see] is like the table

Notice that the main clause elements do not ground the referent and this sentence is uninterpretable without the RC, which seems to be saying, 'Upon walking into the room you will only see a table.'  In other words, it sets up a frame for interpretation.

            To see the converse of this, Nonhuman Object Heads, unlike Nonhuman Subject Heads, have an almost equal number of P-relatives and S-relatives.  The reason is that in real time processing the Object is typically towards the end of the clause and by that time the listener has already heard the main clause subject and the verb, and this has provided the grounding necessary to link the Object to its context, so there is no need for the P-relative and the anchoring function it is so well suited to.  As a result, the RC is free to function in other ways; in fact, the most common function is characterization.  In the words of Fox and Thompson (1990:306),  "...the main clause provides the grounding and the relative clause provides the New information."  To support their claim that these RCs are actually conveying New information, they show that in these instances the main clause has the semantically neutral verb has or has got 43% of the time.  Thus, the main clause is not doing the asserting but rather the RC is. 

            Fox and Thompson go on to discuss the distribution of RC types with Existential Heads and discover that S-relatives are most frequently used here because the subjects of existentials are like subjects in that they are specific, relevant to the discourse, and usually human, but since they are unidentifiable, they are introduced with the existential construction.  Why then are S-relatives more likely to occur with Existential Heads?  It is because humans are not usually grounded by reference to another referent (anchoring), but rather by linking them with their activities or characterizing them. 

            An important part of Fox and Thompson’s study was the way the researchers took into account the various factors impinging on RC use such as Humanness, Definiteness, Grammatical Function, Grounding, Function of RC and the Information Status of the NP.  It demonstrates that the motivations for various linguistic decisions are not always obvious, and that future research must be look for explanations in places and combinations that may seem unlikely.

            De Haan (1987) talks about presupposition being an important part of the formation of relative clauses.  Givon (1990) has pointed out that there are instances where the predication in the RC cannot be said to be presupposed, for example, when the Head Noun is REF-Indefinite.  In this same vein, De Haan argues further that it is the content of the RC which determines definiteness.  He offers this invented discourse (1987:178):

            (23a) a car is a wreck

            (23b) John drives the (that) car

He, consequently, derives two sentences which differ with respect to which proposition has been embedded in the relative clause. 

            (24a) The car which John drives (for he drives a car, as you know) is a wreck.

            (24b) John drives a car which is a wreck.

In (a), the content of the RC is presupposed and the Head Noun car must be preceded by a definite article since this information is sufficient to identify the referent, and it is unsurprising enough to be encoded in a relative clause.  This is Du Bois' (1980) "principle of new information presupposition."  The information embedded in the RC in (b) is not presupposed and does not allow the introduction of the Head Noun with a definite article.  Presupposition is simply a function of the status of the information in a given proposition; the difference here, though, seems to have less to do with presupposition and more to do with the ability of a proposition to uniquely identify a referent. 

            In De Haan's study, 19.3% of the Head Nouns were subjects and 74% of these were definite.  He did not distinguish between other types of NPs and so had only two categories: subject and non-subject.  Of the Non-subject Head Nouns, 49.2% were indefinite.

            In addition, 77.3% of all the relative clauses were in sentence final position, and, as expected, more of the indefinite Head Nouns than the definite ones were in final position.  This provides additional support for Fox and Thompson's contention that RCs are used in discourse for more than simple identification and background information.   

            Givon (1990:645) says, " Most typically, a relative clause is used when the speaker assumes that the referent's identity is accessible to the hearer - i.e. definite - but not easily accessible."  This is the identificatory relative discussed by Bernardo (1979).   The picture painted by Fox and Thompson and Bernardo is quite different.  They do not see identification as the major function of RCs.  Givon does, though, go on to discuss a further distinction.  Relative clauses with definite heads encode presupposed information, that is, information assumed to be known to the listener.  Relative clauses on indefinite heads cannot encode such information since a new referent is being introduced about which nothing can be presupposed.  Givon calls this last type an instance, not of "referential identification," but of "referential coherence".  By this, Givon means that the listener is not being instructed to search his memory for some such referent but rather to set up a file which describes the referent as a type (Givon 1990:647).


            (25)  A man I once knew said there was no sense being gloomy on a sunny day.


He maintains that these RCs are not presupposed nor are they asserted because they still seem to be backgrounded.  He notes that this "backgrounding" cannot correspond to either presupposition or "old information" since this type of relative clause is neither.  This is Bernardo's specificatory relative.  Although these are two of the functions of RCs, others have demonstrated that the encoding of new information is an important part of the function of RCs. 

            Macaulay (1990)  analyzed a wide range of genres both spoken and written to quantify the differences with regard to the use of various syntactic structures.  She found that relative clauses were used more often in writing than in speaking (15% more) and that the primary function of relative clauses was that of conveying new information.  Ninety-three percent of the clauses were informative.  The oral texts used far more coordinate structures.  She sees this as a processing tendency related to Pawley and Syder's (1983) “one clause at a time” constraint which says that the production of clauses is limited by the real time, linear, processing constraint resulting in fewer subordinate structures in speech.  Bernardo (1979: 543) gives examples of how the “one clause at a time” constraint works in oral narrative where the speaker says something with what starts out like a relative clause and then changes it to an independent clause.



            (26)  he's driving along this road that's UH it's not paved,  it's just sort of a dirt road.


This explains the additive nature of oral text since the speaker has less time than the writer to integrate information.   Chafe (1982) supports this conclusion since the number of relatives in his written text outnumbered relatives in the oral text by about 75%.  He characterizes written text as being more 'integrated.'    The high number of informative relatives in Macaulay's study can be accounted for by appealing to this notion of integration.  Relative clauses allow for propositional density with regard to some referent.   That is, given a referent about which several things are to be predicated, an RC allows for this to be done in a less redundant, denser fashion and one which relates the propositions in some discoursal pragmatic system. 

            Macaulay used sentences as a means of measuring the quantity of her text and found that procedural discourse used the fewest number of relative clauses and that descriptive used the most.  These findings should be taken with a large grain of salt, however, since only a small number of texts were compared.  The average number of RCs per sentence is as follows:



Table 1.  Relative clause production frequency, by genre.



# of sentences

# of relative clauses






















                                                                                    (adapted from Macaulay 1990)


Exposition, which is what this thesis will be concerned with, exhibits about one RC for every five sentences.


Second Language Acquisition Theory

            Theories about second language acquisition have changed a great deal in the last forty-five or so years.  Beginning with Fries (1945) and followed by Lado (1957), second language acquisition was seen as mostly behaviouristic and as being heavily influenced by the habits of the L1.  This became know as Contrastive Analysis (CA).  Fries states the position succinctly in the forward to Lado's (1957) Linguistics Across Cultures:

            Learning a second language, therefore, constitutes a very different task from learning the first language.  The basic problems arise not out of any essential difficulty in the features of the new language themselves but primarily out of the special "set" created by the first language habits.

            With Chomsky's generative revolution and its emphasis on mentalism and the rule governed nature of language, this view of language acquisition was brought under close scrutiny (Corder 1971, Nemser 1971, and Richards 1971).  Error Analysis (Richards 1971) and Interlanguage (IL) (Selinker 1972) were the result of these new emphases.  Error Analysis emphasized analysis of actual L2 production with special attention given to learners’ errors.  The second one is akin to this in that it views the errors as evidence for the development of a transitional linguistic system not related solely to the TL or the native language (NL).  These approaches to L2 acquisition downplayed the influence of the L1 in the language acquisition process. Richards acknowledges that the mother tongue plays a significant role in second language acquisition, but argues that there are errors which cannot be attributed to the mother tongue, but which should be viewed as a result of developmental strategies of acquisition, such as simplification, or as a result of the complexity of some aspect of the target language (TL).  This shift in emphasis was supported by studies like Dulay and Burt (1974), which claimed that only about 5% of the errors in the production of L2 learners could be traced to the influence of the mother tongue, and Cook (1973), which argued that L1 and L2 acquisition are actually very similar processes since, in their study, native children and adult L2 learners made similar mistakes on imitation tasks involving relative clauses.   As a result of studies like these, researchers began to pay more attention to learners’ errors and IL data, and less time on contrastive analysis.   About the only acceptable view of CA became the weak version put forward by Wardhaugh (1970).   This stated that CA could be used to determine if the errors students made were due to L1 influence (aposteriori confirmation), but could not predict areas of difficulty (apriori prediction). 

            As the new views on second language acquisition became more popular and were subjected to more tests, some researchers began to question the systematic nature of IL altogether.  Corder (1981) explains how the dynamicity of the language acquisition process causes instability in the learner's idiolect.  He adds that this does not mean there is not a system in the learner's grammar;  it simply highlights the inherent faultiness of our ability to get at what is in the mind of the speaker.   In other words, the L2 system may not be readily apparent.  Bertkau (1973) explored to what degree the errors of learners were systematic as opposed to idiosyncratic.  

            Bertkau (1973) demonstrated, specifically for relative clauses, that learners all seem to have an idiolect; that is, given a number of possible errors even learners from the same NL do not consistently make the same types of errors although some errors can be attributed to negative transfer and it is these errors which are the most consistent.  Following Corder (1971), who argued that all L2 learners have a system composed of some rules that are peculiar just to them,  she argues that there is no identifiable system, for the most part, in the errors of a group of learners.  Rather, each learner's language is an idiolect composed of aspects of the TL, the NL, and the learner’s own rules which are normally the result of simplification processes.  If this is true then we don't expect to find any recurring patterns in the present study.  I will, however, reject this view because, as we shall see, there are studies which have discovered very definite patterns in the IL of learners. 

            Contrastive Analysis was put back in vogue by one such study.  Schacter (1974) called into question the validity of both the EA hypothesis and the weak version of the CA hypothesis by demonstrating that although Chinese and Japanese learners made fewer morphosyntactic errors in RC production than Persian and Arabic students, the production of RCs by the Chinese and Japanese students was still much lower.  She argued that this was due to avoidance and that EA could never have discovered this area of difficulty simply by looking at mistakes in production.

            Language transfer is not a simple matter and there are several variables which must be taken into account.  Not only are there explanations for learner difficulty other than the mother tongue, but  Lado's early explanation of how the L1 affects L2 acquisition turns out to be oversimplified.  He says, "Those elements that are similar to his native language will be simple for him, and those elements that are different will be difficult"  (Lado 1957: 2).  This is not always the case and even those who think the Contrastive Analysis Hypothesis is a useful tool are looking for ways to make it more appropriate to the data.


Studies of L2 Relative Clause Acquisition

            The research on acquisition of relative clauses in both L1 and L2 has focused almost exclusively on structural aspects, that is, word order, morphology, and grammatical function.  Three hypotheses related to the syntax of RCs have been put forward.  Kuno (1974) predicts that center embedded RCs will be more difficult to process based on the anti-interruption hypothesis put forth by Slobin (1986), which states that language resists the interruption of linguistic units because it is more difficult in terms of the load that is placed on the short-term memory.   Sheldon (1972) predicts that RCs in which the grammatical function of the coreferential element is different from the grammatical function of the Head NP (SO, OS) will be more difficult to process.  She calls this the parallel function hypothesis.  Keenan (1975) argues that the NP Accessibility Hierarchy makes it easier to relativize on NPs higher on the proposed relativization hierarchy.  Some studies which represent much of what has been done from both an L1 and L2 perspective with respect to these hypotheses are Clancy et al. (1986), Schumann (1980), and Ioup and Kruse (1977).  The problem with these studies, and many others, is that they often produce conflicting results.  While all of the studies support Kuno's hypothesis that center embedding is more difficult, the studies come up with conflicting views on Sheldon's parallel function hypothesis. 

            Ioup and Kruse maintain that the difficulty L2 learners have with RCs  is not dependent upon their native language, but rather upon the structural complexity of the different types of relative clauses.  Indeed, they provide convincing evidence that the influence of the L1 is negligible with respect to structural errors and because of this, they rule out CA, at least in its present form, as an explanation for learner error and difficulty.  They go on to say that,


studies which have not demonstrated a transfer effect have restricted their investigation to these structural parameters, while other studies which claim evidence of interference have extended the scope of their research to include semantic, pragmatic and stylistic variables.  (Ioup and Kruse 1977:57)

In reality, however, their attempt to inveigh against CA and the influence of the L1 has strengthened its case.  Language is not simply structure and any analysis of real language data must take into account pragmatics and semantics.   Lado (1957: 2) implies as much:


individuals tend to transfer the forms and meanings, and the distribution of forms  and meanings of their native language and culture to the foreign language and culture - both productively when attempting to speak the             language and to act in the culture, and receptively when attempting to grasp and understand the language and the culture as practiced by natives. (emphasis mine)

If Lado is right, the form/function relationship is at the heart of CA.  Any look at form without a look at its function in discourse is not as enlightening as an analysis which takes it into account; indeed, it may have missed the mark entirely.  

            Studies relating to the use and non-use of different syntactic structures begin with Schacter's 1974 article entitled "An Error in Error Analysis," which has achieved a sort of 'classic' status with respect to Contrastive Analysis and the phenomenon of avoidance.  Schacter was writing in response to the Error Analysis advocates who had either dismissed Contrastive Analysis as a viable tool in the study of second language acquisition, or had relegated it to an aposteriori role of after-the-fact explanation of learner difficulty - the weak version of CA which denied the predictive power of the theory (Wardhaugh 1970). 

            Schacter examined the relative clauses in four sets of 50 non-native English compositions by intermediate and advanced students from typologically diverse backgrounds and one control group of American compositions.  The results are tabulated below (Schacter 1974: 209):



Table 2.  Relative clause production figures and error rates.






% of Errors

























                                                                                                            (Schacter 1974)


            The most noticeable thing about the results is the difference in the number of relative clauses actually produced.  The Arab and Persian students produced far more relative clauses; in fact, their production was very similar to that of the American control group,  but they also had a much higher error rate.  The Chinese and Japanese produced far fewer RCs, but also had a lower error rate.  Schacter interpreted the data as avoidance and claimed that the avoidance was due to a structural difference: namely, that the RC in Chinese and Japanese is prenominal while in English it is post-nominal.  In essence, the absence of relative clauses was considered a type of error.  This seems intuitively reasonable since the goal of language study is to increasingly approximate the structures, functions, and distributions of the TL syntax.  She argued that simple error analysis and its reliance upon contrastive analysis for explanation would not be able to detect or predict the trouble that Chinese and Japanese students have with the English RC since these students made very few surface structure errors.  It would assume that these groups have no difficulty with this construction, but this is difficult to demonstrate convincingly if Schacter's data is taken seriously.  

            There are several problems with this study, however.  First, it does not do a very good job of quantifying how many RCs were produced per some unit of text.   All we know is that there were 50 compositions, but if the Chinese and Japanese wrote shorter compositions, that would explain in part their lower production rate.  More exact and possibly revealing results might have been obtained by quantifying the RCs with respect to the number of sentences or T-units (see chapter 4). We also do not know what the topics of the essays were and what effect this might have on the number of relative clauses used.  Macaulay (1990) has clearly demonstrated the potential for genres to affect the use of syntactic structures like relative clauses in a significant way.   Nor do we know what effect proficiency had on the production of RCs.  Was the production of RCs by the Chinese and Japanese students consistently lower even as proficiency in the target language increased?  Or, was the difference in production a result of the Chinese and Japanese learners acquiring the RCs at a slower pace than the others?  If they all ended up with the same production rates at higher levels, then this is a problem that resolves itself given time and, while it is interesting that they take longer to acquire the structure, it is not as pressing a problem for the learner.  If, however, these students never actually acquire a native-like use of the RC, then it may deserve some classroom attention beyond that being given.

            Eckman's (1977) Markedness Differential Hypothesis is, in part,  an attempt to account for the differences in the error rate related to pronoun retention in Schacter's study.  He bases his hypothesis on Greenbergian implicational typological universals, and says that elements of the TL which are more marked than the NL will be more difficult and that the difference in relative degree of typological markedness also results in greater difficulty.  He argues that pronoun retention becomes more marked the further up on the NP Accessibility Hierarchy the phenomenon occurs.  Persian has pronoun retention for every position on the Hierarchy except for the subject, thus the difference between English and Persian is greater than the difference between English and all the other languages in this respect.  He claims that this higher relative degree of difficulty explains the higher error rate of the students.  This does not explain the avoidance of the Chinese and Japanese students, though.  

            Gass (1979, 1980) studies the performance of ESL students on three different tasks involving RC production.  In the first, students gave responses as to the grammaticality of several English sentences among which were both correct and incorrect tokens.  The incorrect sentences modeled errors that students might be expected to make because of their NL; for example, pronoun retention by Arab students.  In the second task, students were asked to combine sentences, and the instructions to the students precluded everything but the production of relative clauses.  The third task was a free composition.  Gass is here testing the claims made by Schacter (1974) that avoidance is a manifestation of processing difficulty due to the prenominal position of RCs in Japanese and Chinese.  Speaking of Schacter’s claim, she says, “In order to test the validity of these claims, one would have to  determine directly whether one can in fact correlate lack of use with actual difficulty in production.” 

            In the sentence combining exercise, which was structured so that students would have opportunity to relativize on all parts on the Accessibility Hierarchy, the students clearly avoided positions lower on the hierarchy.  This could not be correlated with the language background of the student, suggesting that this is a universal principle of language transfer and supporting the validity of Keenan and Comrie's proposed NP Accessibility Hierarchy.  The composition reflected this as well since most of the NPs which were relativized were Subjects (76%) and Objects (15%). 

            What was not found in the study was any significant relationship between accuracy on the sentence combining task and frequency of use in composition.  In other words,  the students who used fewer numbers of relative clauses in the free composition did not make more grammatical mistakes; they demonstrated no more difficulty in production than the other students.  In addition, there was no relationship between avoidance of difficult positions on the Accessibility Hierarchy and frequency of use. 

            In summary, it seems that there is still no explanation for the infrequent use of relative clauses by students with a certain language background, except for the structural contrast of prenominal versus postnominal position.  There was no correspondence between accuracy on the combining task or ability to relativize positions on the NP Accessibility Hierarchy and the avoidance mentioned by Schacter.   A drawback of this study was that it was not discourse oriented.  It only looked at the structure of isolated sentences and not how those sentences functioned in the discourse.

            To isolate exactly how the different variables affect RC acquisition, Chiang (1980) did a statistical analysis of several variables relating to relative clause production.  He demonstrated that when Target Language Proficiency, Language Background, and Relativized Position were taken into account, they influenced correct production in that order.  The first one is not surprising since we expect proficiency in the TL to be a major factor in the production of relative clauses.  What is surprising, given the studies by Gass (1980), Ioup and Kruse (1977), and others, is that, even in a simple structural type of study, language background should be such an important factor.  One difference between this study and others, though, is that the data was collected via a question/answer type format, and so, there are discourse considerations which were not accounted for, and which may have affected the results, such as, what type of information was being sought. 

            Other studies of the acquisition of relative clauses by second language learners are Ghadessy (1980), Abbott (1980), and Kharma (1987).   All of these focus only on  surface structure and never on function, although, Kharma (1987), who analyzes the writing of students whose native language is Arabic, does say that all of the errors were errors of form, never of function.  Kharma does not, however, refer to any studies related to the function of RCs in discourse, so one is left to guess as to how this judgement is made.

            All of these studies have one thing in common which is that except for the frequency count in compositions there is virtually no attention paid to discourse elements.  All of the focus is on structure and grammatical function, but not on information packaging and discourse function.  These studies tend to use isolated sentences in their research and this obscures many of the facts about language.  For example, these types of studies could never uncover the fact that A relatives are uncommon, as Fox (1987) did.  Nor could they demonstrate how relative clauses are related to information packaging.


Discourse Level Contrastive Analyses

            The next set of studies are very similar in that they all compare the function of RCs in three different languages, Korean, Japanese, and Chinese, with English RCs.  All of these languages except English have prenominal RCs and this turns out to be an important factor.  These studies will tie in the first section of the literature review on discourse and information packaging and the second section on studies of RC acquisition by L2 learners. 

            Hwang (1990) examines relative clauses in narrative discourse in English and Korean from two perspectives: clause structure and discourse function.   She then asks a pertinent question: "...do RC's give information about participants or information about happenings and plot structure?" (374).  She goes on to say that, though a distinction between the two may be difficult to draw, it is important because, if RCs are more related to NPs, then we expect RC to serve a mainly referential function, that is, they identify NPs.   If, on the other hand, RCs provide more information about "happenings", then we can expect that NEW and important information may well be encoded in RCs.  Hwang’s study demonstrated that basic word order constraints influence the discourse functions of RCs.  The function of RCs in Korean and English narratives were compared and there were some marked differences. 

            The first difference concerns the reporting of sequential events.  Because Korean, an SOV language, has prenominal RCs, this structure is not used to report sequential events related to the Head Noun.  The reason is that since the RC precedes the Head Noun, the chronological order of information would be reversed.  Therefore, a coordinate structure is used in Korean, but a nonrestrictive RC in English.  (All of the following examples are from Hwang (1990, in press (b)):


            (27)  Slowly he walked along the aisle and up the steps to the choir, [where he handed the plate to the priest, [who blessed the gifts the then reverently placed them on the altar]].


            Hwang (in press (b)) analyzes the translation of an expository text from English into Korean.  She noticed that the restriction on temporally sequential events encoded by RCs also holds for at least some logically sequential events.  Where the relation between the main clause and the RC is logical, e.g.,  reason-result, condition-consequence, etc.  a clause chain is used in Korean instead of an RC.  An example of an English RC which is translated as a clause chain in Korean follows:



            (28)  The word big has several lexical meanings [which are not easy for the       beginner to differentiate]  (Hwang in press (b):6)


            Another difference between the discourse function of English as opposed to Korean RCs is the introduction of related participants.  An RC is not normally used in Korean to introduce a participant, while it seems quite common in English and, in fact, the example below is taken from a well-known story which is not about the mother at all, but about the three little pigs.  These crucial participants are introduced in an RC.


            (29)  There was a mother pig [who had three little pigs]. 


The functional equivalent in Korean is a coordinate NP.


            (30)  mommy pig and three baby pigs existed.


            Yet a third difference in function is that while in Korean RCs frequently have a backreferencing function, English seems to prefer preposed adverbial clauses.  What is striking about this is that the structural differences between Korean RCs and preposed participial clauses in English are almost non-existent.  Both structures are prenominal, non-finite, and evidence a gapping strategy by which the co-referential NP is deleted. 


            (31a) SHIM CHUNG's father Shim Bongsa was waiting for daughter to come back.

            (31b) [(who) was tired of WAITing] Shim Bongsa, {to see if DAUGHTER was coming-in.order.to}, Ø lean over a CANE, Ø went outside in the DARK.


            '{Tired of waiting}, Shim bongsa went out in the dark leaning over his cane, {to see if his daughter was coming}.'

Since Korean is verb-final, the last word in (a) is 'wait', and it is repeated at the beginning of (b).  It is not possible to achieve this type of cohesion in English even with a non-restrictive RC because the head noun interrupts the information flow, but a preposed participial clause maintains the natural flow of information.

            Hwang (1990) also reports a tendency for Korean to encode thematic material in RCs.  In narratives with morals, the RC often has a didactic function.

            Japanese, an SOV language like Korean with prenominal RCs, also seems to have at least some distributional differences from English with regard to the function of RCs.1  In Japanese, the number of RCs which have a personal name or pronoun as the Head Noun is much higher than in English (Collier-Sanuki in press).  Collier-Saunki (in press) counted the number of relative clauses in a book which was translated in both English and Japanese and examined the function of these relative clauses following the framework developed by Fox and Thompson (1990).   In English, there are very definable information flow constraints that govern the type of relative used, and one of these is grounding (Prince 1981).  Japanese also has similar principles at work, but because of the difference in basic word order these principles are realized in different ways.  Japanese relative clauses on human referents seem to ground the Head Noun by setting up a circumstantial frame and are usually translated as adverbial clauses or even independent clauses in English.   The following schema and examples of Japanese RCs are from Collier-Sanuki (1991):

            Subject head:  [relative clause] NP Verb

            Object head :  (Subject) [relative cl.] NP verb


            (32)  [who had at last made it to a Japanese school which was in Shinjyuku ward,] Ms. L sensed that it was not a very easy thing to meet the expectations of her mother and older sister. 

            (33)  [who had decided that she must first resolve the housing problems above everything else,]  Ms. L tried hard at her "wisdom of women."


The information in the RCs in the above examples are best paraphrased in English by adverbial clauses, preposed participial clauses or simple independent clauses.  For instance, (a) and (b) above could be paraphrased as the following respectively.



            (32a)  When she arrived at a Japanese school in Shinjyuku,  Ms. L realized that it was ...

            (33a)  She realized that she must first resolve the housing problem..., and she did her best...


            The Collier-Sanuki study also found that not only were personal names with relative clauses more infrequent in English than in Japanese, but they also represented information which was less crucial than that in similar Japanese RCs.  Though she doesn't make this point, all the examples of English RCs given in the study were non-restrictive.  Over 71% of these RCs were stative.  However, the percentage of Japanese RCs that were stative was only 15%.  This seems to support her characterization of Japanese RCs as encoding more important information and their function of grounding the Head Noun by establishing a sort of circumstantial/temporal frame.

            Another difference is that, like Korean, RCs in Japanese cannot encode events subsequent to that in the main clause since this would reverse the iconic ordering of events. 

            A number of recent studies have compared the use of RCs in translations of a single text in two or more languages.  These studies are a type of contrastive analysis on the discourse level.  Their findings are in general agreement with the information flow principles outlined by Fox and Thompson (1990) and Hwang (1990 and in press).

            The results of these studies were rather surprising.  Collier-Sanuki (in press) found that Japanese uses almost twice as many relative clauses as English does.  The Japanese corpus had 613 RCs, while the English corpus had 341.  Furthermore, when a closer look was taken at the individual predications in Japanese and how they were encoded in English, it was found that only 172 of the predications were encoded by RCs in both languages.  Following Hwang (1990) and Fox and Thompson (1990), she attributes this difference to the position of the RC and the attending difference in the functional load RCs bear in the two languages. Kamimoto et al. (1992) found exactly the same thing in their analysis of a text translated in both Japanese and English.   

            In these studies, Collier-Sanuki in particular, there is an overwhelming tendency for Japanese relatives to have Subject Heads (72%) while only 17% have Object Heads.   The English text had a little more than twice the number of Object Heads (36%) and Subject Heads only accounted for 45% of the total.  Clancy et al. (1986) found that Japanese children had difficulty with center embedded relative clauses much like English children.  Unlike English, though, in Japanese it is an RC with an Object Head which is more likely to produce a center embedded relative, so the infrequent occurrence of RCs with Object Heads may be caused partly by more universal processing strategies.  In Japanese, S-relatives are extremely frequent (64%) and since the subject still needs grounded it has this function, but not anchoring like Fox and Thompson's O relatives, but PROPOSITION LINKING whereby a circumstantial frame is established.   

            Bley-Vroman and Houng (1988) and Zhao (1989) are studies like Collier-Sanuki's and are based on translations of a text in both Chinese and English.  The Bley-Vroman and Houng study is seriously flawed in that it does not consider RCs without a relative pronoun.  Since Object relatives and OPREP relatives can optionally delete the relative pronoun, their count could be substantially lower than the actual occurrence of RCs in the text.  Besides this, they did not do a complete count of the RCs used in the Chinese translation, but only of the RCs in the English translation and how they were translated into Chinese.  This tells us if there are RCs in English which cannot be translated into Chinese, but it does not tell us if there are RCs in Chinese that cannot be translated into English.

            Zhao's study mirrored the other studies in that she found different distributions of RCs in the two translated texts.  Her findings are tabulated below (Zhao 1989:107):



Table 3.  Number of RCs in the translation and the amount of overlap.


RCs in English

RCs in Chinese

RC in English = RC in Chinese

In English only

In Chinese only



59 (48%)



                                                                                                                     (Zhao 1989)


Only a little less than half of the English RCs correspond to Chinese RCs.  Zhao makes a list of some of the different ways in which RCs are used in English, but which cannot be RCs in Chinese.  I offer here an abbreviated list. 

            As we saw in Ziv (1975) and Fox and Thompson (1990), RCs can be used in English to convey important, new, and focused information, especially in clause final position.  When this happens, Chinese uses two independent clauses (Example from Zhao 1989):


            (34a)  A man just came in who was wearing very funny clothes.

            (34b)  Jin    lai      le       ge  ren,    ta chuandai  qiguai

                       in    come  PFV   one man  he   wear       funny


She gives several examples of this type from the translation.  Chinese is an SVO language like English and the tendency is to put important information in clause final position.  The prenominal position of the RC in Chinese circumvents this to some degree leading to a preference for two independent clauses in many cases, the first of which functions as a topic and the second as the comment.  Other examples are the have + Head Noun +RC and the existential There + be constructions.  Both of these are similar in that important information related to the Head Noun is sentence final. They differ in that they are utilized to perform different discourse functions.  Other differences are RCs in English that refer to place or time, and RCs that encode logical relations such as reason/result or concession. 

            Zhao included non-restrictive RCs in the count as well, but Chinese does not have these and all English tokens are translated as two independent clauses.

            All of the examples in which Chinese used a relative clause but English did not turn out to be participial phrases in English which can be analyzed as reduced relative clauses.  The only difference is that there is no finite verb and no relative marker.   


            (35)  The man standing on the corner was smoking a cigarette.


            Based on these facts, Zhao interprets the findings of Schacter, not as avoidance, but as transfer on the discourse level.  Zhao characterizes Chinese relativization as having only one function - restriction of the reference of the head noun.  It appears from this one study that Chinese relativization, unlike Japanese and Korean, manifests only a subset of the English RC functions.  This difference may be due to the fact that Chinese is more of a mixed bag typologically speaking since it has SVO word order, but prenominal RCs.  This brings different information flow principles to bear on the form/function relationship of the language.



            This literature review has uncovered a number of important facts with respect to the usage of relative clauses.  In the section dealing with information flow, we saw that relative clauses are, maybe contrary to our expectations, often used in English to convey new and even important information (82% of the time), and that identification is not their primary role (only 7%).   We also saw that the function of the relative clause is uniquely tied up with the grammatical properties of the sentence in which it is located, i.e.,  indefinite heads are more likely to present new non-presupposed information in the relative clause, as are existentials and predicate nominals and clauses in which have is the main verb.  Definite Heads, on the other hand, are more likely to present presupposed information which uniquely specifies a referent.  In fact, it is the content of the relative clause that determines the definiteness of the head in many cases.  Basically, the relative clause can make an NP relevant to the discourse in several general ways: by giving new information which is crucial to the understanding of the text that follows, by grounding or anchoring,  and by identifying.  That these functions of relative clauses are not always cross-linguistic is demonstrated by the disocurse level contrastive analyses summarized in this review.

            All of these studies involved differences between English and a language which had prenominal relative clauses.   There are basically two situations where an RC is used in English but not in another language: particular grammatical constructions, like RCs with existential heads, and new asserted information.  In addition, relative clauses do not seem to be used in OV languages to encode subsequent events or logical relations.  At least one of the languages, Japanese, seems to use RCs that would be translated as adverbial clauses or independent sentences in English.  These tend to be semantically non-restrictive RCs which provide circumstatial information which frames a main predication.

            In research on second language acquisition of relative clauses, most of the focus has been on the problems students have with regard to the surface structure form of the English RC.  Those studies which have focused on the production of RCs by the students have noted the low production rates of learners whose native language differed from English with respect to the position of the RC with respect to the Head Noun, and have offered two different hypotheses.  The first is that the syntactic processing difficulty involved in the switch from a left-branching language to a right-branching language is to blame for the lower production.   The second explanation is that the functions of relative clauses in these students’  NL is different from the functions of relative clauses in English and that the lower production is simply a result of the difficulty the students have adjusting to the new form/function relationship.  The research that has been done points toward the second hypothesis because there have not been any studies which could link avoidance with processing difficulty.




            1Though this thesis does not focus on the issue, there is some discussion in current literature about exactly how to compare syntactic structures across languages.  Matsumoto (1988) claims that in Japanese relative clauses and complement clauses are instances of the same syntactic construction.  Matsumoto calls these adnominal clauses.  The same is true for Korean (Hwang 1993).  Comrie and Horie (1992) note the same thing for Khmer.  It seems that some languages have a general strategy for associating subordinate clauses with a noun.  These may be relative clauses, complement clauses or neither in English.  The following examples are taken from Hwang (1993).


            Adnominal clause with RC interpretation:


            [wi-eso      po-ko.it-ton]     wonsungi-tul

             top-at       see-Prog-rM      monkey-Pl

            ‘the monkeys [(who) were watching for the (tree) top]’


            Adnominal clause with CC interpretation:


            [ilokhe        ppajyotta-n]     mal       iyo. 

            like.this       fell.into-prM    word    is

            ‘(I) mean that (I) fell like this.’  (Lit: (It) is (my) word that I fell like this.)


            Adnominal clause which is neither RC nor CC:


            [holangi       ulpujin-nun]     soli

            tiger              roar-prM         sound

            ‘the sound of tiger roaring’

Next: Statement of the Problem