• translation
• proofreading
• pricing and ordering
• resources
• references
• contact us
• home

CHAPTER V

FINDINGS

            This study analyzed the compositions of Spanish and Korean ESL students to discover how they used relative clauses.  Four hypotheses were tested.  The results of the study with respect to these hypotheses will be discussed first, and then some additional findings will be noted. 

Testing of Hypotheses

Hypothesis #1:  Because of the prenominal position of relative clauses in the NL, the Korean subjects will produce fewer relative clauses than the Spanish students.  The Spanish rate of production will be similar to that of English because their NL has postnominal relative clauses.

            This hypothesis was supported.  Had this been found incorrect for the Korean students, we would have falsified the syntactic processing explanation for the underproduction of RCs by students in previous studies.  As it is, we must remain open to the possibility that this contributes to the underproduction of students whose native language has prenominal RCs.  The Spanish students, however, produced more relative clauses than the American students in Schacter’s study.  This is an issue worth exploring and will be discussed in section three when we consider Hypothesis 3.

            The Korean data is unsurprising.  The production of relative clauses by the Korean students was found to be very similar to the Chinese and Japanese students in Schacter's study.  The Spanish students, on the other hand, produced more relative clauses than even the Arabic, Persian, and American students in Schacter’s study.  In fact, they produced about 50% more RCs than the American students and three times as many as the Koreans.  The following chart compares the production rates of the students in Schacter’s study with the production of the Spanish and Korean students in my study by giving the number of compositions, the total number of RCs, and the number of RCs per composition.  

Table 5.  The number of RCs per composition in Schacter’s study compared with

the present one.

Note:  The present study included NRRCs, but since Schacter's study did not include NRRCs, this table does not include these RCs.

            One difficulty with comparing the results of my research with Schacter's is that I have no indication of what type of essays her students wrote.  As a result, one might think me presumptuous in saying that the difference between the Spanish students and the Arabic, Persian, and American students is significant since essay type might have affected production.  However, when one remembers that all of my essays were exposition, and that expository and procedural discourse have fewer relative clauses than any other according to Macaulay (1990), it becomes much more likely that my numbers are highly significant.  If Schacter's essays were expository, then we are comparing like things.  If her essays were not expository, then the difference between the production of my students and the American students may be even more significant.

Hypothesis #2:  Production and distribution of relative clauses will become greater and more varied as TL proficiency increases. 

            This was expected since one of my postulates is that any two languages will differ to some degree with respect to the distribution and function of a particular piece of syntax, and that a big part of increased proficiency is the student’s ability to mirror the native speaker’s usage.

            This hypothesis was supported.  The Korean students demonstrated greater and more varied production in a number of areas.  The Spanish students’ production rose only slightly, but their usage did show significant shifts, though in different areas than the shifts of the Korean students.  I considered the students to have shown significant change in some area if at least 3 of the 4 groups evidence a particular trend, either up or down, when compared with  students in contiguous levels.  First, I will discuss the differences in Korean production, then I will discuss the adaptations of the Spanish students.

            The Korean students produced more relative clauses as their proficiency in English increased.  The number of RCs per composition practically doubled from level 3 to level 6, and there was a corresponding decrease in the number of T-units per RC. 

Table 6.  Number of RCs produced by the Korean students, by level.

The number of RCs used by the advanced Korean students is approaching that of the American students in Schacter's study.  That is not to say that both groups employ RCs in the same way when constructing a discourse, but it is an indication that the Korean students are approximating the TL structure.  

            The distribution of the grammatical relations (GR) of the coreferential noun in the RC (NPRel) also shifted dramatically.  The Korean students used an overwhelming number of S relatives in the beginning, but this had moderated by the time they reached the advanced levels.  This shift in usage, of course, occasioned a marked increase in other types of relatives.  Hwang (1993) has reported that  Korean uses far more S relatives than any other type, so the frequency with which low intermediate students use S relatives seems to be a clear case of the transfer of NL distribution.  The NPRel distribution from Hwang’s analysis of some Korean texts is given in the following table.  In Korean, adjectives have all the formal characteristics of RCs, so they are included in the following table, but the distribution without adjectives is also given. 

Table 7.  Distribution of GR for NPRel in native Korean text.

Even without figuring in the adjectives, the Korean text still used S relatives 54% of

time.  If we compare this with the findings of Fox and Thompson (1990), we find that S relatives only comprised 32% of the total number of relatives in oral English, a significant difference.  Collier-Sanuki (in press) notes that in Japanese, a language that most consider closely related to Korean both genetically and typologically,  a very high percentage (64%) of the RCs are S relatives.  This seems to be typical of SOV languages and is certainly typical of the Korean students in this study.

Table 8.  Distribution of GR for NPRel in the Korean data, by level.

Given the numbers, it seems clear that these students are still transfering their native language distribution of relatives to their IL in the intermediate stages of language acquisition, but they do begin shifting this distribution as proficiency increases towards that of the native speaker, and while the advanced Korean students still do not have native distribution, there is movement towards that in every area.¹ 

            With respect to the grammatical relation of the Head Noun, it seems that the clearest trend among the Korean students (Table 9) was towards an increased number of Predicate Nominals.  This is also clearly more in line with native speaker use of RCs.  Fox and Thompson (1990) report that Predicate (P) Heads made up 19.6% of the total number of Head NPs.  The number of X (Other) Heads also increased, which suggests that the students have become more comfortable with increasingly complex syntax.  Both Subject and Existential (E) Heads increased according to my evaluation metric.  But, for Existential Heads, the percentage difference between level 3 and level 6 is not that great, and the level 5 students used no S Heads at all, so it is difficult to say that there is a trend toward increased usage of these two Head NPs.

Table 9.  Distribution of Head Nouns in Korean data, by level.

            The occurrence of have as a main verb decreased as the students became more proficient.  Have is frequently used for main clause grounding in spoken English, but the more integrated nature of written text may mitigate against its occurrence in writing.  Unfortunately, I do not have any way to substantiate this by comparing it with native speaker frequency, but the decline in frequency of occurrence suggests that this might be the case since one of my expectations is that part of increasing proficiency is a closer approximation of native language frequency.  As we will see, the Spanish students have the same tendency and this supports my reasoning. 

Table 10.  Percent of matrix clause verbs appearing as have in Korean data, by level

The students’ use of have is sometimes peculiar as in (1-3) below. 

            (1)  While Seoul has many apartments that people usually live there, Arlington has many house that people live in.   (K3.4.3-4)

            (2)  First, Seoul and Arlington had similarities that I could feel and see two places somethings.  (K3.10.3)

            (3)  They got a big problem that is an air pollution.  (K4.4.1)

In (1), the student is comparing the housing situation in Seoul to that in Arlington.  Putting the information in an RC seems awkward because it is encyclopedic knowledge not really restrictive in any sense.  To interpret it as such leads us to the conclusion that there are also apartments in Seoul that people do NOT live in.  It seems that when have is used for grounding in the main clause we expect the RC to provide important or new information, so (1) seems strange.  A more natural English sentence might be (1a), where more innocuous preposition in does the grounding.

            (1a)  While in Seoul people usually live in apartments, in Arlington most of people live in houses.

Sentences (2) and (3) seem similar.  I offer an alternative to (3) below in which the function of the matrix verb have is performed by a possessive modifier.

            (3a)  The city's biggest problem is air pollution.  (or) Air pollution is their biggest problem.

 Other examples seem to be more natural, however.

            (4)  They have the dream which their ancestors had embraced.  (K3.8.1)

            (5)  I have a friend who is living in Japan now.  (K3.17.1)  (K3.19.1 is identical)

The example in (5) is at the very beginning of the essay and performs a

presentative function, that is, the matrix clause introduces a referent and grounds that referent by relating it to the speaker, but it is the information in the RC that is important for the following discourse because it frames the discussion.  Compare this with (1) above where the information in the RC is not important or new.  

            Another clear trend was that the students tended to use personal pronouns as Head NPs more infrequently as their proficiency increased.  I have asked Korean students as to the acceptability of such heads in their own language, and they have indicated that they are acceptable.  Personal pronouns with RCs are widespread in Japanese as well, but this type of relativization is infrequent in English though it is possible in some instances, bookish though it may be, such as when the pronoun has a generic type of interpretation. 

            (6)  You who side with the king forfeit your life. 

The personal pronouns used by the Korean students did sometimes have this sort of interpretation, but still seem unacceptable to my native ears. 

            (7)  When we who is youth live in the world, we can use the generation gap.

                   (K3.15.9)

The interpretation was not always generic, however.

            (8)  I who lived in Seoul for a long time was born and grew in there.  (K4.3.1)

Table 11.  Percentage of RCs with Personal PROs as Head NP in Korean data, by level.

As we can see, the advanced students never used a Personal PRO as a Head NP, implying that the IL system of the Korean student is beginning to converge with that of the TL. 

            There is one more area in which one might want to claim that there has been a clear shift in usage and that is an increased number of NRRCs.  Unfortunately, I have not found any studies which would provide information about the frequency of this type of RC in native Korean.  But, when one considers that in Japanese, again a language closely related to Korean, many of the RCs are semantically nonrestrictive and thus would be translated into English as NRRCs, it might be more a case of transferring their native usage than a case of varying their production to more closely match that of the target language.  This is underscored by the fact that the greatest difference is between level 3 and level 4.  By level 4 the students have better control of the form of the language and so may feel more comfortable forming RCs, thus making it easier for them to transfer native function to the TL form. The other levels do not differ from one another in such a  marked way.  These students tend to use Head NPs that are uniquely identifiable, such as place names like Dallas, Seoul, etc., quite often resulting in NRRCs.²

Table 12.  Percentage of NRRCs in Korean data, by level.

            In summary, the Korean students seem to have varied their production in a number of ways.  Their production of RCs doubled.  Their use of S relatives declined 65% with corresponding increases elswhere.  Their use of personal pronouns as a Head Noun was entirely eliminated by the time they reached the advanced levels.  The occurrence of have as a main verb declined 100%, and their use of Predicate Nominals as Head Nouns increased 550%.  The significance of some of these changes will become more apparent later.

            The Spanish students did not evidence changes in distribution and patterning as pervasive as that of the Korean students.  The number of relative clauses produced by these students did not rise very much.  The number of RCs rose slightly, but nowhere near the doubling of the Korean students.  In addition, the type of relatives used by the Spanish students was practically unchanged from level 3 to 6, closely resembling the distribution of native English.  This seems to suggest that the distribution of RCs in English is similar to that in Spanish, and/or that the Spanish students feel like this is a construction they can tranfer easily.

Table 13.  Spanish production of relative clauses, by level.

            The grammatical function of the Head Noun changed significantly in at least one respect.  The use of existential heads declined steadily and drastically in the Spanish data.³  Again it may be the nature of expository text which causes this decline.  I would like to give one example of an awkward use of an existential in expository text.  Note the second existential.  It is used to frame a hypothetical situation, but the use of the existential seems strange because does not signal a shift to a hypothetical state.

            (9)  On the other hand, to have a good boss is important.  Because in the way   of doing a good boss there is an adequate environment for everybody.  For instance, there is a man who is supervisor in a bank.  If he is not nice with             his subordinates...

The number of Subject Heads also fell slightly, while the number of Predicate Heads rose slightly. 

Table 14.  The grammatical function of the Head Nouns in the Spanish data, by level.

            Like the Korean students, the Spanish students also tended to use have as a main verb less frequently as their proficiency increased. 

Table 15.  Occurrence of have as matrix verb in Spanish data, by level. 

            Another area in which the Spanish students may have evidenced some change is the definiteness of the Head Noun.  With the exception of level 6, there is a steady decline in the number of DEF Heads.  This is closely linked to whether or not the RC is asserted information.  DEF NPs tend to have presupposed information in the RC while indefinite Heads tend to have asserted information.   Both figures are included.

Table 16.  Percentage of DEF Heads and Asserted RCs in Spanish data, by level. 

Notice that although the level 6 students do not follow the trend with respect to the definiteness of the Head Noun, the number of asserted RCs continued to increase.              Besides the decline in definiteness, another factor that contributed to the increase in the percentage of asserted RCs was the decline in the use of the Head Noun thing(s).  This is so because RCs occurring with this construction do not usually pass the separability test since the Head NP has very little semantic content of its own. 

            (9)        I will teach them things that I will like them to know about it.  (S 3.3.5)

            (9a)      ? I will teach them things.  I want them to know about them.

            (10)      The other thing that could help in feeling happiness is having all the basic things that you need in life. (S3.13)

            (10a)    ?* The other thing that could help in feeling happiness is having all the basic things.  You need them in life. 

This turns out to be one of the biggest changes in the way the Spanish students used RCs.  In level 3, over a quarter of the Head Nouns were thing(s),  but this number fell to just 8% by level 6.  Whether this reflects native speaker usage is a matter of further research, but it certainly reflects the teaching in the classroom where students are encouraged not to use thing as a Head NP because it sounds too colloquial and ambiguous.  Students are encouraged to state exactly what sort of thing they are talking about.  In the compositions on the requirements for happiness, the Spanish students often used NPs such as the things we need to be happy.  This could be replaced with the requirements for happiness.   It would certainly make their writing teacher much happier.

Table 17.  Percentage of Spanish Head Nouns filled by thing(s).

            In conclusion, it seems that the hypothesis was strongly supported for the Korean students, who evidenced broad based changes in their use of RCs as their proficiency increased.  With the exception of production, the Spanish data supported the hypothesis as well.  What is interesting is the different ways the two groups adjusted their use of RCs.  These differences lend credence to discourse explanations for the avoidance phenomenon that focus on differences in usage and distribution at the discourse level since two languages can be expected not to have the same differences with respect to a third language, but rather, if you will, different differences.  Korean differs from English rather clearly with respect to the distribution of the GR of the NPRel, and this was one area in which the Korean students varied their production; the Spanish students did not vary their production in this area, suggesting that Spanish is more like English in this regard.  Furthermore, while some of the Korean usage changes are clearly a result of overcoming the influence of the NL ( e.g. type of relative), it is less clear that the Spanish shifts are the same type of change.  Maybe some of the changes of both the Spanish and Korean students are just the results of increasing proficiency and complexity and have nothing to do with NL influence (e.g. the decreasing use of have as the matrix clause verb).  If further study revealed that most students whose native language had postnominal RCs performed similarly to the Spanish students, that is, there were no major changes in the way they used RCs, then this would provide support for the emerging position that elements of language as basic as constituent order can drastically affect the distribution and function of RCs. 

Hypothesis #3:  The two groups will evidence clear differences in the distributional patterns related to the use of relative clauses. 

            This was posited due to evidence which suggested that different discourse functions are performed by different syntactic structures (cf.  Fox and Thompson 1990).   To get a characterization of the typical way in which the Korean and Spanish students used English RCs, all of the data for the two groups were put together regardless of proficiency level. 

            This hypothesis was upheld.  There are three major differences between the two groups: the GR of the Head NP and the NPRel, the information status of the RC, and the presuppositionality of the RC.

Grammatical Relation of the Head NP and the NPRel:

            First, the two groups differed with respect to the distribution of the GR of the NPRel.  There were significant differences in every area except for A relatives.       

Table 18.  Comparison of Spanish and Korean NPRel distribution. 

The large number of S relatives used by the Korean students is very consistent with the distribution of the NL, and they use almost twice as many as the Spanish students.  Several researchers have reported on the paucity of O relatives in Japanese and Korean.  It seems that the anchoring function these relatives often perform in English is not a typical function of the RC in these languages.  Instead, RCs provide circumstantial and semantic framing by characterizing the Head NP.

            Here, we are comparing IL data, the production of students trying to conform to native language forms and function, so we expect some skewing between the distribution of the native language and the IL forms.  To illustrate the degree to which two native languages may differ in this regard and to establish some basis of comparison with native speaker usage, I have provided the following data which  demonstrates the type of relatives used in a translation of a text in both English and Japanese (adapted from Collier-Sanuki in press).

Table 19.  Comparison of English and Japanese NPRel distribution.

            As we can see by comparing Tables 18 and 19, the Spanish distribution is very similar to that of the English text, and the Korean distribution is very similar to that of the Japanese text.  This strongly suggests that language typology may be what determines the distribution of relatives in language.  There has been much discussion about how to interpret the distributions found in different languages and two positions have emerged. 

            One position wants to use the distribution of RCs in discourse to validate Keenan and Comrie's AH.  According to Gass (1980), the grammatical relation of the NPRel in the compositions she analyzed was that of Subject 76% of the time and DO 15% of the time.  These numbers are used to argue for the validity of Keenan and Comrie's AH.  Gass seems to be arguing, following Keenan (1975), that  frequency of usage proves that one position is easier to relativize than another.  Gass claims that one factor in avoidance is the tendency not to relativize on positions low on the AH, and that the AH should be seen as one of the constraints on second language acquisition (SLA).   This is supported by  the Korean data, where S relatives are very common,  but the Spanish data is clearly different from the results of her study because the number of Subject and Object relatives is almost the same.  If the AH is a constraint in SLA, then its influence should be stronger on students at lower proficienciy levels because they do not have a very complete understanding of the TL and hence should rely more on universal principles.  Yet, even in level 3, the Spanish students use fewer S relatives (25%) than O relatives (33.8%).  Here we have a clear counterexample to the claim made by Gass, and it would be difficult to argue for the validity of the AH with this data and her logic.   

            The other position is that of Fox (1987).  She argues, based on her analysis of English, that the AH should be re-interpreted in light of the Absolutive Hypothesis and that S and O should be more prominent than A because they occur with almost equal frequency in text, while A relatives are much rarer.  The Spanish data in this study supports her position, but it is difficult to reconcile with the Korean and Japanese data, since they use far more Subject relatives than any other type.  Since the AH is supposed to be universal in scope, it seems unwise to argue for its revision based on frequency counts when we have clear counterexamples like Korean and Japanese, where S relatives are far more common than O relatives.  In fact, it is very likely that frequency counts are of little value in testing the validity of the AH. 

            The controversy over the AH may not be an issue at all when searching for an explanation for the distribution of RCs in discourse.  Their distribution seems to be much more a function of information flow principles and word order typology than it is a function of proposed universals.  The differences in distribution can definitely be drawn along typological lines.  The correspondences between Spanish and English and between Korean and Japanese support a shift in focus in which discourse distribution and frequency are explained in terms of language typology and information flow and not in terms of the  AH.  This is, of course, easy to test.  If it could be shown that there were languages with postnominal RCs which had as many S relatives as languages with prenominal RCs, then the explanation of RC distribution in terms of information flow and language typology would not be valid.   

            One factor that may contribute to the disparity in NPRel distribution is the different types of Head NPs used by the two groups.  Fox and Thompson (1990) demonstrated that non-human Heads were more likely to have O relatives since inanimate objects are often made relevant to the discourse by being related to humans, who are typically pronominal subjects.  Of all the Head NPs used by Spanish students, 69.5% were non-human, compared to the Korean frequency of 51.1%.   Again, we find that the Spanish distribution is very similar to that of English.  In English, non-human Head NPs are 70% of the total while Human Heads account for 30%.  

            The two groups also evidenced some differences with respect to the GR of the Head NP.  The Korean students tended to use more Subjects and fewer Predicate Nominals than the Spanish students.  In order to allow for some comparison with English, the figures from Fox and Thompson’s analysis of oral English and Hwang’s analysis of Korean text are included in the following table. 

Table 20.  Comparison of the distribution of Head NPs used by the Korean and Spanish students and native English and native Korean distribution. 

            The Korean students use more Subject Heads than the Spanish students and fewer Predicate Nominals.  The overproduction of Subjects by the Korean students seems to be due to the NL distribution.  In Korean, RCs with Subject Heads are the most common.  Why the Korean students should use fewer Predicate Nominal Heads than they do in their native language is unclear, but it may be due to a difference in discourse function between the two languages.  Again, the Spanish students’ pattern of usage is very similar to that of the native speaker suggesting that their NL may resemble English much more closely in this respect than the Korean usage. 

            The Spanish and Korean students have used both Head NPs and Relatives with different frequencies and this brings us to the question of whether the students showed any clear differences in the way in which these two combine.  For example, when the Head NP is a Subject was there any predilection for a certain type of Relative?  The answer is yes. 

            There are several differences in usage, but I will focus on the more salient differences, those more likely to be statistically significant.  The Korean students differed radically from the Spanish students with respect to three of the five types of Head Nouns - Subject, Other, and Predicate Nominal.  The next three charts quantify the differences between the groups.  Table 21 includes the numbers from Fox and Thompson (1990).