in greater detail. First, the subjects’ performance in the experimental tests is

Chapter Four Results and Discussion

This chapter presents the major findings of the present study and discusses them

in greater detail. First, the subjects’ performance in the experimental tests is

reported and analyzed with statistical instruments. Then, the results are further

considered in relation to the research questions raised earlier in Chapter Three.

4.1 Results

In this section, the data collected from the RC judgment, context translation, and

passage-rewriting tests are statistically analyzed to answer the research questions,

which concern: (1) acquisition of non-restrictive RCs; (2) use of RCs in different

pragmatic/discourse contexts, including identifying, characterizing, presentative, and

parenthetical; and (3) use of RCs in writing as a useful backgrounding device.

Analyses of each research question incorporate performance differences between the

two groups of subjects, i.e. the first- and third- graders.

4.1.1 Acquisition of Non-restrictive RCs

The first research question of the present study deals with the acquisition of

NRRCs by Taiwanese EFL learners of senior high school in reference to (1) the

tendency to overuse RRCs; and (2) the use of NRRCs in different NP contexts. To

answer this research question, the data collected from the RC judgment test (see

Appendix A for further details) were scored and analyzed, in which the subjects were

to judge several RCs as either restrictive or non-restrictive on the basis of context.

During the data-analysis session, elimination of a small number of test items was

necessary due to the subjects’ failure to complete all the items in time, to provide a

brief explanation for judging a given RC as non-restrictive, or to render

non-restrictive judgments solely based on the degree of referential accessibility of

head NPs in context (as opposed to other irrelevant factors), as could be determined

from their explanations. Accordingly, the subjects’ accuracy rates in this test were

obtained by dividing the total correct items by the total number of applicable items.

Table 5 summarizes the subjects’ overall performance in the RC judgment test:

Table 5: Overall accuracy rate of each RC type in the RC judgment test RC type

Frequency Restrictive Non-restrictive

Total applicable items 840 1332

No. 719 670

Correct

Rate 85.60% * 50.30% *

* p < .05

For items expected to be restrictive, 719 out of 840 were given a correct judgment,

with an accuracy rate of 85.60%; for those expected to be non-restrictive, 670 out of

1332 were answered correctly, with an accuracy rate of 50.30%. Apparently, the

accuracy rate of NRRCs was extremely low, as opposed to that of RRCs. To

determine whether the two accuracy rates differed to a significant level, a Chi-square

test was conducted, and the difference was found to be statistically significant

(p= .000). The results indicate that generally, the subjects did not yet completely

acquire NRRCs, since they still encountered considerable difficulties using them

correctly.

To investigate whether the subjects tended to overuse RRCs, the mean

percentages of their restrictive and non-restrictive responses to items in the RC

judgment test were calculated in Table 6:

Table 6: Percentages of the subjects’ restrictive and non-restrictive responses in the RC judgment test

Total

applicable items Restrictive responses Non-restrictive responses

2172 1381 63.58% * ⁷⁹¹ 36.42% *

* p < .05

Of a total of 2172 RCs, irrespective of RC types, 36.42% were given a non-restrictive

judgment, while 63.58%, a restrictive one; through a Chi-square test, the percentage

of restrictive responses was proven to be significantly higher than that of

non-restrictive ones (p= .000). These RRC-dominated responses simply indicate a

strong tendency for the subjects to use RRCs persistently in most contexts, even in

non-restrictive ones, as shown in Table 5, in which half of the NRRCs in the test were

incorrectly given a restrictive judgment. Based on their undue use of RRCs, it can

be argued that to the subjects, the distinction between RRCs and NRRCs still

remained so blurred and unclear that when using RCs, they fell short of properly

distinguishing between the two by overusing the more prototypical, common, familiar

one, i.e. RRCs.

Since the variable of English-learning experience may very well factor into the

subjects’ performance in the RC judgment test, the accuracy rates of both RC types

for the first- and third-graders are presented in Table 7:

Table 7: Group accuracy rates of each RC type in the RC judgment test Group

RC type 1 ^st graders 3 ^rd graders

Correct 340 379

Restrictive

Total

Rate

420 80.95% *

420 90.24% *

Correct 312 358

Non-restrictive

Total Rate

665 46.92% *

667 53.67% *

* p < .05

The two groups performed similarly in that compared with the 80.95% and 90.24%

accuracy of RRCs, their accuracy of NRRCs was relatively low, with the first-graders

getting 46.92% (312 out of 665) correct and the third-graders, 53.67% (358 out of

667). As revealed by a Chi-square test, the third-graders significantly outperformed

the first-graders not only in RRCs (p= .000) but also in NRRCs (p= 0.014). The

results indicate that with more years of learning, the third-graders did make some

progress in properly using the more marked form of RCs, though their accuracy of

NRRCs was still very low.

In view of the fact that there are different contexts which particularly call for the

use of NRRCs, it is necessary to investigate whether performance differences exist

among these contexts in the use of NRRCs. To address the issue, the contexts of all

the NRRCs in the RC judgment test were categorized into the following eight types

according to the kind of referentially accessible NPs to which their antecedents

pertained: (1) linguistically definite NPs; (2) situationally definite NPs; (3) narrowly

specified NPs; (4) whole-referring NPs; (5) generic NPs; (6) proper NPs; (7) personal

pronouns; and (8) one-of-a-kind NPs. Table 8 gives the overall accuracy rates of

these non-restrictive contexts in the RC judgment test:

Table 8: Overall accuracy rate of each non-restrictive context in the RC judgment test

Context*

Frequency (1) (2) (3) (4) (5) (6) (7) (8)

Total

applicable items 107 108 221 118 206 235 114 223

No. 19 21 44 18 112 175 90 191

Correct

Rate 17.76

%

19.44

%

19.91

%

15.25

%

54.37

%

74.47

%

78.95

%

85.65

%

*(1): linguistically definite NPs (2): situationally definite NPs (3): narrowly specified NPs (4): whole-referring NPs (5): generic NPs (6): proper NPs

(7): personal pronouns (8): one-of-a-kind NPs

As displayed in Table 8, large discrepancies existed among the eight non-restrictive

contexts, with the accuracy rates ranging from 15.25% to 85.65%. Ranking highest

in accuracy was “one-of-a-kind NPs” (85.65%), followed by “personal pronouns”

(78.95%) and “proper NPs” (74.47%). Accuracy began to decrease by about 20% in

“generic NPs” (54.37%), and then dropped drastically to under 20% in “narrowly

specified NPs” (19.91%), “situationally definite NP” (19.44%), “linguistically definite

NPs” (17.76%), and “whole-referring NPs” (15.25%). From these figures, it is clear

that the subjects’ performance in using NRRCs indeed varied from context to context.

The performance disparities among different non-restrictive contexts basically

reflect an inadequate understanding on the part of the subjects of when to use NRRCs,

in the sense that they applied their use of NRRCs only to a few, not all, referentially

accessible NPs. To be more precise, they tended to limit their use of NRRCs to NPs

whose referential accessibility stemmed from their lexical connotation of

“uniqueness,” or “being the only one” (e.g. one-of-a-kind NPs, personal pronouns,

and proper NPs), to the virtual exclusion of those referentially accessible due to such

non-lexical factors as prior discourse (linguistic definite NPs), speech situation

(situationally definite NPs), additional qualifications (narrowly specified NPs),

entire-class referring (whole-referring NPs), and particular-class referring (generic

NPs). Table 9 makes a comparison of accuracy rates between the two groups of

referentially accessible NPs, with the former termed as “lexically accessible NPs,”

and the latter, “non-lexically accessible NPs”:

Table 9: Comparison of accuracy rates between “lexically accessible NPs” and

“non-lexically accessible NPs” in the RC judgment test Context

Frequency Lexically accessible NPs Non-lexically accessible NPs

Total applicable items 572 760

No. 456 231

Correct

Rate 79.72% * 30.39% *

* p < .05

As confirmed by a Chi-square test, the subjects’ performance in using NRRCs

significantly differed between the two NP contexts (p= .000). In lexically accessible

NP contexts, the subjects were usually able to use NRRCs in preference to RRCs,

with the accuracy rate reaching as high as 79.72%, whereas in non-lexically

accessible NP contexts, their use of NRRCs was sporadic, which yielded an extremely

low accuracy of 30.39%.

A close look at the subjects’ explanations for judging a given RC as

non-restrictive further reveals that their propensity to confine the use of NRRCs to

lexically accessible NPs has something to do with their associating NRRCs with

uniqueness-referring NPs. As found in their explanations in the test, the subjects’

non-restrictive judgments were often based on whether head NPs were one-of-a-kind

NPs or proper NPs, i.e. whether head NPs themselves connoted the sense of being

“the only one.” The finding bears the implication that for the subjects, the

overriding factor that determined the employment of NRRCs was not so much the

referential status of head NPs (i.e. whether their referents are readily established in

context) as the existence of specific head NPs, including personal pronouns, proper

NPs, and one-of-a-kind NPs. Given their association of NRRCs with

uniqueness-referring, rather than referential accessibility, the subjects’ inability to

properly use NRRCs with referentially accessible NPs, except for those with the sense

of “uniqueness,” was to be expected.

As regards performance differences between groups, Table 10 provides the

accuracy rates of the eight non-restrictive contexts for the first- and third-graders:

Table 10: Group accuracy rates of each non-restrictive context in the RC judgment test

Group

Context* 1 ^st graders 3 ^rd graders

Correct 8 11

(1) Total Rate

53 15.09%

54 20.37%

Correct 14 7

(2) Total Rate

53 26.42%

55 12.73%

Correct 24 20

(3) Total Rate

109 22.02%

112 17.86%

Correct 8 10

(4) Total Rate

58 13.79%

60 16.67%

Correct 43 69

(5) Total Rate

106

40.57%

*

25.59%

100

69.00%

*

35.17%

Correct 83 92

(6) Total Rate

118 70.34%

117 78.63%

Correct 41 49

(7) Total Rate

56 73.21%

58 84.48%

Correct 91 100

(8) Total Rate

112 81.25%

75.17%

111 90.09%

84.27%

* p < .05

*(1): linguistically definite NPs (2): situationally definite NPs (3): narrowly specified NPs (4): whole-referring NPs (5): generic NPs (6): proper NPs

(7): personal pronouns (8): one-of-a-kind NPs

Generally speaking, the first- and third-graders performed alike. Both groups got the

highest accuracy in lexically accessible NP contexts; the accuracy rates of

“one-of-a-kind NPs,” “personal pronouns,” and “proper NPs” were 81.25%, 73.21%,

and 70.34% for the first-graders, and 90.09%, 84.48%, and 78.63% for the

third-graders. In non-lexically accessible NP contexts, the accuracy rates were all

well under 27% for both groups, apart from that of “generic NPs,” which was 40.57%

and 69.00% for the first- and third-graders, respectively. It may seem that the

third-graders outperformed the first-graders in all the non-restrictive contexts except

“situationally definite NPs” and “narrowly specified NPs.” However, statistical

analysis by a Chi-square test revealed a significant group difference only in “generic

NPs” (p= .000), thereby suggesting that overall, the performance by both groups was

equally satisfactory in lexically accessible NP contexts, and equally disappointing in

non-lexically accessible NP contexts, with the exception of “generic NPs,” where the

third-graders performed significantly better.

To sum up, the above analyses of the data in the RC judgment test suggest that

even at the senior high school stage, the subjects, by and large, were far from having

completely acquired NRRCs. On the one hand, the subjects were observed to use

RRCs consistently in most contexts, even in those where NRRCs should be preferred.

In the light of their strong tendency to overuse RRCs, it seems that the subjects may

have entirely ignored or not fully grasped the difference between RRCs and NRRCs,

thereby failing to make a good distinction between the two when using RCs. On the

other hand, the subjects exhibited great performance differences among various

referentially accessible NP contexts in their use of NRRCs, which essentially reflects

their incomplete knowledge of when to use NRRCs. More specifically, they were

able to properly employ NRRCs with NPs whose lexical sense connoted

“uniqueness,” including personal pronouns, proper NPs, and one-of-a-kind NPs, but

showed a total disregard for other referentially accessible NPs, such as linguistically

or situationally definite NPs, narrowly specified NPs, whole-referring NPs, and

generic NPs. As attested in the subjects’ explanations for their non-restrictive

judgments, their propensity to limit their use of NRRCs to lexically accessible NPs in

particular appeared to be the direct result of their relating NRRCs with

uniqueness-referring, instead of referential accessibility. These two observed

phenomena i.e. the tendency to overuse RRCs in any given context and the propensity

to use NRRCs only with lexically accessible NPs, both contributed to the subjects’

poor performance in correctly employing NRRCs in the RC judgment test.

4.1.2 Use of RCs in Different Pragmatic/Discourse Contexts － － － － Identifying, Characterizing, Presentative, and Parenthetical

The second research question focuses on the extent to which Taiwanese EFL

learners of senior high school are able to employ RCs appropriately in different

pragmatic/discourse contexts to (1) identify a referent as a known entity; (2)

characterize a referent as a particular type; (3) present a topically important referent

into the discourse; and (4) interpolate parenthetical assertions about a referent. To

answer this research question, the data gathered from the context translation test (see

Appendix B for further details)—in which the subjects were asked to provide

appropriate English equivalents for several Chinese sentences in the four contexts

aforementioned—were closely examined to determine the subjects’ preferred structure

in English for each context. In data analysis, frequencies of RC use in each context

were tallied and the main criterion for identifying RC occurrences was this: all errors

related to RC formation were dismissed, except for erroneous deletion of relative

markers, which was considered an instance of the non-use of RCs as it could be a

result of calquing, i.e. the act of translating Chinese directly and literally into English.

Since the main concern of the second research question is the functional aspects

of RCs, the subjects’ grammatical errors in RC formation were not further analyzed

but only identified for reference. Table 11 lists occurrences of five common types of

RC-related errors committed in the context translation test:

Table 11: Occurrences of RC-related errors in the context translation test Group

Type of RC errors 1 ^st graders 3 ^rd graders Overall (1) Wrong placement of head NPs in

relation to their RCs 18 7 25

(2) Wrong deletion of head NPs 19 7 26

(3) Wrong deletion of relative markers 6 7 13

(4) Wrong choice of relative markers 10 6 16

(5) Wrong use of resumptive pronouns 5 3 8

Total number of RC errors 58 30 88

Although the first-graders committed nearly two times more RC errors than did the

third-graders, the number of RC errors by each group can be regarded as quite small,

suggesting that the majority of the subjects already had a good grasp of the structural

complexities of RCs.

Turning now to the subjects’ performance in the pragmatic/discourse functions of

RCs, Table 12 describes the mean percentage of RC use for each context in the

context translation test:

Table 12: Overall frequency rate of RC use for each context in the context translation test

Context

Frequency Identifying Characterizing* Presentative Parenthetical*

Total applicable items 240 240 240 240

No. 230 220 117 139

RC use

Rate 95.83% 91.67% 48.75% 57.92%

*The test included three question items for this context, i.e. items 5, 9, and 10. Nevertheless, in calculating the frequency of RC use for “characterizing context,” item 10 was excluded, because the majority of the subjects somehow misinterpreted item 10 “會在公共場所講話大聲的人很令人討厭” to be

“在公共場所講話大聲很令人討厭,” producing sentences with dummy it like It is very annoying to talk loudly in public places, It is very annoying that people talk loudly in public places, or Talking loudly in public places is very annoying, instead of the target translation People who talk loudly in public places are very annoying. Had item 10 been included, the overall frequency rate of RC use for “characterizing context” would have been much lower (80.83%), which could lead to the wrong conclusion that the subjects seemed to use less RCs in “characterizing context” than in “identifying context.”

*Strictly speaking, in “parenthetical context,” the RC in the target translation should be non-restrictive, i.e.

without any comma separating the head NP and the relative marker. Nonetheless, since the main focus was on whether the subjects could use RCs accordingly to add extra information, in determining the RC frequency for “parenthetical context,” the distinction between restrictive and non-restrictive RCs was disregarded.

Apparently, in “identifying context” and “characterizing context,” RCs were used

with high frequency, almost 100% of the time, while in “presentative context” and

“parenthetical context,” occurrences of RCs were sporadic, with only 48.75% and

57.92% frequency, respectively. It should be noted that the RC frequencies in

presentative and parenthetical contexts could have been even lower if there had been

more distracter items (i.e. items whose target translation involved the use of non-RC

structures). Due to time constraints, the context translation test included only two

distracters (items 6 and 8); therefore, the number of RC occurrences in presentative

and parenthetical contexts may have been raised as a consequence of the shadow

effect, in which the small number of distracters made the subjects aware that it was

RCs that were being targeted, thus causing them to consciously employ more RCs

throughout the test. Despite this shadow effect, the RC frequency rates in

“presentative context” and “parenthetical context” still remained distinctly low.

Put under the scrutiny of a Chi-square test, the RC frequency rates of the four

contexts were shown to differ from one another to a marked degree. To pinpoint

where significant differences lay, a posterior comparison of RC frequencies among

the four contexts was carried out, as illustrated in Table 13:

Table 13: Posterior comparison of frequency rates of RC use among the four contexts in the context translation test

Context Context

Identifying (95.83%)

Characterizing (91.67%)

Presentative (48.75%)

Parenthetical (57.92%)

Identifying (95.83%) * *

Characterizing (91.67%) * *

Presentative (48.75%)

* p < .05

As can be seen in Table 13, significant differences in RC use were found between

“identifying” and either “presentative” or “parenthetical,” and between

“characterizing” and either “presentative” or “parenthetical.” These results indicate

that although committing very few gross grammatical errors in RC formation, as

previously demonstrated in Table 11, the subjects, in general, failed to acquire a full

understanding of the pragmatic/discourse functions of RCs. That is, the distribution

of their RC use usually concentrated on the functions of furnishing a referent with

information of identification or characterization, and seldom extended to the less

prototypical functions of presenting a topical referent into the discourse or inserting

parenthetical assertions.

Regarding performance differences between groups, Table 14 displays the

frequency rates of RC use for each context by the first- and third-graders:

Table 14: Group frequency rates of RC use for each context in the context translation test

Group

Context 1 ^st graders 3 ^rd graders

RC use 112 118

Identifying

Total Rate

120 93.33%

120 98.33%

RC use 100 110

Characterizing

Total Rate

120 83.33%

120 91.67%

RC use 54 63

Presentative

Total Rate

120 45.00%

120 52.50%

RC use 65 74

Parenthetical

Total Rate

120 54.17%

120 61.67%

From Table 14, we can see that for both groups, it was in “presentative context” and

“parenthetical context” that RCs were used the least frequently, with their frequency

rates ranging from 45.00% to 61.67%. The group frequency rates for each context

were submitted to a Chi-square test, and no significant difference was found between

the two levels for each context (p > .05). These results indicate that when it came to

the more sophisticated use of RCs in introducing a thematically important referent or

interpolating additional information, the third-graders, despite more years of exposure

to RCs in their English study, still performed as poorly as did the first-graders.

To investigate what other structure the subjects seemed to prefer in presentative

and parenthetical contexts, their answers in items 1, 2, and 3 were further examined.

It was found that with relevant contextual information given, many subjects still

thought of independent clauses, the same structure used in Chinese, as appropriate

structural equivalents in English for presentative and parenthetical contexts. For

example, in questions 1 and 3, the target translation preferably involved the use of

presentative and parenthetical RCs, respectively, as in (77) and (78):

(77) Question 1a: There was a king who loved his daughter very much.

(78) Question 3: My favorite animal is the elephant, which is considered the

biggest land animal in the world

However, the translation of these questions by the subjects often involved the use of

two independent clauses strung together with either the coordinator and or a comma,

as in (79) and (80):

(79) There was a king and/, he loved his daughter very much.

(80) My favorite animal is the elephant and/, it is considered the biggest land

animal in the world.

Table 15 presents occurrences of relative and independent clauses in presentive and

parenthetical contexts in percentile terms:

Table 15: Percentages of relative clauses and independent clauses in presentative and parenthetical contexts in the context translation test

Group Context

1 ^st graders (Total = 120)

3 ^rd graders (Total = 120)

Overall (Total = 240)

RCs % 54 45.00% 63 52.50% 117 48.75%

Presentative

ICs * % 66 55.00% 57 47.50% 123 51.25%

RCs % 65 54.17% 74 61.67% 139 57.92%

Parenthetical

ICs % 55 45.83% 46 38.33% 101 42.08%

*ICs: independent clauses

From Table 15, it appears that in either context, the subjects used relative and

independent clauses with nearly the same frequency. However, the frequency of

their RC use should be interpreted with caution, because it may have been increased

as a result of the shadow effect, an experimental weakness mentioned previously, in

which insufficient distracters caused the subjects to consciously use more RCs in their

translation. Regardless of such a shadow effect, the recurring use of independent

clauses underlines the very fact that in the subjects’ repertoire of English structure for

presenting a topical referent or supplying extra information, the simple construction of

independent clauses to a certain extent remained as a common alternative to the

complex one of RCs.

Lastly, to determine whether there was a tendency for the subjects to misuse the

definite article the with characterizing RCs, whose antecedents are normally indefinite,

all the characterizing RCs produced by the subjects in questions 5, 9, and 10 were

categorized into two types according to the kind of article used with the antecedent,

i.e. definite or indefinite. Table 16 shows the percentages of characterizing RCs with

the indefinite article and those with the definite in the context translation test:

Table 16: Percentages of characterizing RCs with the indefinite article and those with the definite article in the context translation test

Group Context

1 ^st graders (Total = 134)

3 ^rd graders (Total = 157)

Overall (Total = 291) IA* % 76 56.72% 103 65.61% 179 61.51%

DA* % 42 31.34% 49 31.21% 91 31.27%

Characterizing

NA* % 16 11.94% 5 3.18% 21 7.22%

*IA: indefinite article, i.e. a(n) (with singular countable head NPs) or Ø (with plural countable or uncountable head NPs)

* DA: definite article, i.e. the.

* NA: no article present due to inappropriate deletion of either articles or head NPs.

For the first-graders, 134 instances of characterizing RCs were found; 56.72% were

preceded by the indefinite article and 31.34%, the definite. For the third-graders,

157 instances of characterizing RCs were identified; 65.61% were headed by the

indefinite article and 31.21%, the definite. Though the percentage of characterizing

RCs with the definite article did not appear significant for both groups, it revealed a

certain tendency on the part of the subjects, across groups, to misuse the definite

article with RCs that were mainly to characterize a new referent (i.e. an indefinite

head) rather than identify a known referent (i.e. a definite head).

To recapitulate, the results of the context translation test point to the fact that

showing a good command of RC form, the subjects did not necessarily gain complete

mastery of RC function. As revealed in their RC performance in the four pragmatic/

discourse contexts, the subjects’ use of RCs was mainly for tracking down a referent

as either a familiar entity (i.e. identifying) or particular type (i.e. characterizing);

rarely did they make good use of RCs as a topic-presenting device or an

information-adding interpolator. In presentative and parenthetical contexts, they

were inclined to replace RCs with independent clauses. Furthermore, it was found

that the subjects seemed to have a certain tendency to misuse the definite article with

RCs in a characterizing context.

4.1.3 Use of RCs in Writing as a Useful Backgrounding Device

The third research question inquiries into the extent to which Taiwanese EFL

learners of senior high school are capable of utilizing RCs appropriately in writing to

background supportive materials. To answer this research question, the data

garnered from the passage-rewriting test (see Appendix C for further details)—in

which the subjects were to rewrite a descriptive passage in a more coherent way by

rearranging its idea units with proper conjunctions—were scrutinized with respect to

the subjects’ repertoire of clause-linking strategies in a piece of descriptive writing.

In data analysis, conjunctions used for each pair of sentences to be combined were

identified, with frequencies of their use tallied; mechanical errors, such as spelling,

tense, and agreement, were dismissed and so were RC-related errors.

Table 17 on page 92 summarizes the subjects’ performance in the

passage-rewriting test. Overall, the subjects were quite competent at making good

use of RCs to package background information in their rewriting: for each pair of

sentences in the passage, more than half of the subjects (56.67% to 80.83%) employed

relative markers, like who, which, that, to subordinate the second one, which often

carries amplifying information of description for the ongoing discourse, to the first as

an RC. Take for example the pair, sentences 12 and 13, which the majority of the

subjects rewrote as (81) below:

(81) They also feel criticized by other people.

Other people think that all women should stay at home with their children.

→ They also feel criticized by other people who think that all women should

stay at home with their children.

Table 17: The subjects’ performance in the passage-rewriting test Group

Pair of sentences

1 ^st graders (Total = 60)

3 ^rd graders (Total = 60)

Overall (Total = 120) who/which/that % 45 75.00 % 52 86.67 % 97 80.83 %

and % 5 8.33 % 0 0.00 % 5 4.17 %

because % 9 15.00 % 8 13.33 % 17 14.17 % Ss1-2

No conjunction % 1 1.67 % 0 0.00 % 1 0.83 % who/which/that % 32 53.33 % 39 65.00 % 71 59.17 % and % 12 20.00 % 7 11.67 % 19 15.83 % because % 8 13.33 % 8 13.33 % 16 13.33 % Ss4-5

No conjunction % 8 13.33 % 6 10.00 % 14 11.67 % who/which/that % 33 55.00 % 50 83.33 % 83 69.17 % and % 6 10.00 % 4 6.67 % 10 8.33 % because % 17 28.33 % 6 10.00 % 23 19.17 % Ss7-8

No conjunction % 4 6.67 % 0 0.00 % 4 3.33 % who/which/that % 30 50.00 % 38 66.33 % 68 56.67 % and % 12 20.00 % 14 23.33 % 26 21.67 % because % 1 1.67 % 0 0.00 % 1 0.83 % Ss9-10

No conjunction* % 17 28.33 % 8 13.33 % 25 20.83 % who/which/that % 31 51.67 % 40 66.67 % 71 59.17 % and % 7 11.67 % 5 8.33 % 12 10.00 % because % 16 26.67 % 11 18.33 % 27 22.50 % Ss12-13

No conjunction % 6 10.00 % 4 6.67 % 10 8.33 % who/which/that % 30 50.00 % 44 73.33 % 74 61.67 % and % 10 16.67 % 5 8.33 % 15 12.50 % because % 0 0.00 % 0 0.00 % 0 0.00 % Ss14-15

No conjunction* % 20 33.33 % 11 18.33 % 31 25.83 % who/which/that % 32 53.33 % 47 78.33 % 79 65.83 % and % 7 11.67 % 5 8.33 % 12 10.00 % because % 19 31.67 % 8 13.33 % 27 22.50 % Ss17-18

No conjunction % 2 3.33 % 0 0.00 % 2 1.67 % who/which/that % 32 53.33 % 37 61.67 % 69 57.50 % and % 6 10.00 % 5 8.33 % 11 9.17 % because % 0 0.00 % 0 0.00 % 0 0.00 % Ss19-20

No conjunction* % 22 36.67 % 18 30.00 % 40 33.33 %

*The reason why a great number of subjects failed to use conjunctions in these pairs of sentences is that they tended to rewrite the sentences as More and more people share this attitude toward women today (for sentences 9 and 10);

This belief makes some women feel a lack of balance in their lives (for sentences 14 and 15); and These three examples show how it can be difficult for women to achieve balance between jobs and family responsibilities (for sentences 19 and 20).

When examined across groups, the subjects’ performance in using RCs to background

information exhibited some differences: the third-graders apparently outstripped the

first-graders in the performance by 8% to 28%.

Table 17 further reveals a certain tendency by the subjects to use independent

clauses or adverbial clauses as alternative structures for RCs to code the same

information: for each pair of sentences, a small number of subjects, across groups

(1.67% to 31.67% for the first-graders; and 6.67% to 23.33% for the third-graders),

used and or because instead to have the second one coordinated with or subordinated

to the first as an independent or adverbial clause, as can be illustrated in (82), an

alternative for some subjects to (81) in combining sentences 12 and 13:

(82) They also feel criticized by other people and / because other people (they)

think that all women should stay at home with their children.

In summary, based on the quantitative and qualitative analysis of their

performance in the passage-rewriting test, the subjects can largely be said to have

gotten to grips with the use of RCs in discourse grounding. The extent of their RC

use in writing, nevertheless, is often contingent not only upon their years of learning

but also upon their tendency to employ either independent and-clauses or adverbial

because-clauses in place of RCs for packaging the same piece of information.

4.2 Discussion

In this section, the results reported in the three elicitation tests are further

discussed. For each research question under discussion, a comprehensive review of

the major findings is offered in terms of plausible explanations for and possible

implications of the observed phenomena.

4.2.1 Acquisition of Non-restrictive RCs

The results of the RC judgment test highlight two major problems interfering

with the subjects’ use of NRRCs: (1) the tendency to overuse RRCs in any given

context; and (2) the tendency to employ NRRCs only with lexically accessible NPs

(i.e. uniqueness-referring NPs) and disregard other referentially accessible NPs.

These problems respectively point to inadequacies in the subjects’ acquisition of

NRRCs: (1) a failure to draw a clear distinction between RRCs and NRRCs in their

mental grammar of English RCs; and (2) a failure to gain a full understanding of

when to use NRRCs. Their “flawed” acquisition of NRRCs is explicable in terms of

L1 transfer, cognitive complexity, input frequency, and instructional effects.

First of all, the subjects’ acquisition of NRRCs may have been affected by the

negative transfer of their mother tongue, Chinese. In English, RCs are further

categorized into two subtypes according to whether they delimit the domain of

reference of their antecedents (i.e. restrictive RCs) or simply provide additional

information about them (i.e. non-restrictive RCs). In Chinese, nevertheless, such

differentiation does not exist in that the main and sole function of Chinese RCs is to

identify or define their head NPs, that is, Chinese RCs are purely restrictive. In the

light of this functional difference in RCs between the two languages, it can be posited

that the subjects may have unconsciously transferred the exclusively restrictive nature

of RCs in Chinese into their mental grammar of English RCs, becoming oblivious of

the dichotomous RC distinction maintained in English. Under the influence of their

native language, the subjects seldom sought to further distinguish between RRCs and

NRRCs when using English RCs, thereby showing a tendency to overuse RCCs in

nearly all contexts.

As valid and important a factor in the subjects’ acquisition of NRRCs is the

cognitive complexity inherent in the target structure itself—more specifically, their

failed acquisition of NRRCs is attributable to their inability to fully grasp the

context-based concept of referential accessibility, closely related with the use of

NRRCs. Contrasting with their restrictive counterparts in the referential status of the

antecedent, NRRCs are used only when the referents of their heads are well

established in context. Apparently, in order to acquire NRRCs, one needs to well

understand the notion of referential accessibility. This, however, can be a

cognitively difficult task for learners, most of whom are so accustomed to learning

English at a sentential level that they may not be able to discern how context, be it

lexical, linguistic, or situational, renders a given NP identifiable and prefers NRRCs

over RRCs. It is likely that for lack of a deep appreciation of the notion of

referential accessibility, the subjects still had a hard time establishing a clear

distinction between RRCs and NRRCs in their mental grammar of English RCs, and

thus ended up mixing up the two by overusing the prototypical, unmarked type of

RCs, i.e. RRCs.

Furthermore, the subjects’ unsatisfactory acquisition of NRRCs may be

accounted for by their having less exposure to this particular construction. Since

instructional materials are often the main source of L2 input for learners, to explore

the relationship between acquisition of NRRCs and input frequency, English

textbooks published by Far East (Books 1-6) ¹ , one of the versions used by senior high

schools in Taiwan, were examined to calculate frequencies of RRCs and NRRCs in

reading texts. Table 18 shows occurrences of both RC types in the six textbooks:

1

The rationale for selecting the Far East version in particular for the examination is based on its wide

use not only by the subjects’ schools (i.e. National Chia-Yi Girl’s Senior High School, National Hu-Wei

Senior High School, and Taipei Municipal Xisong Senior High School) but also by the vast majority of

senior high schools in Taiwan.

Table 18: Frequency counts of RRCs and NRRCs in Far East English textbooks of senior high school

Type Book Book1 Book2 Book3 Book4 Book5 Book6 Books1-6

Restrictive 45 44 58 74 70 117 408

Non-restrictive 7 6 6 20 23 23 85

Total RCs* 52 50 64 94 93 140 493

* RCs counted here included fully-fledged RCs (with relative pronouns and adverbs), reduced RCs (with participial phrases), headless RCs (with such relatives as what, whoever), and cleft RCs; occurrences of non-finite RCs (with to-phrases) were dismissed.

As can be seen from Table 18, throughout the three years of senior high school, RCs

to which learners are exposed are predominantly restrictive, with 408 occurrences

observed, as opposed to 85 occurrences of non-restrictive ones. It may follow from

the low frequency of NRRC input in text ² that the total amount of NRRC input

received at the senior high school stage still was not enough for the subjects to readily

consolidate and internalize this “idiosyncratic” structure into their abuilding mental

grammar of English RCs. Accordingly, the subjects were prone to underuse NRRCs,

with which they were less familiar, and overuse RRCs, with which they were more

acquainted. Moreover, this insufficient exposure to NRRCs in text may partly

explain why the accuracy of NRRCs obtained by the third-graders in the present study

still remained very low (only 53.67%, as indicated in Table 7), though proven

2

The result is basically in accordance with the theory of markedness. NRRCs are, by their very

nature, a marked structure, and with this markedness comes the implication that not only is this

construction more difficult to acquire due to its cognitive complexity but it is also less frequent in the

discourse distribution.

significantly better than that of the first-graders.

Finally, instructional effects play a role in the subjects’ acquisition of NRRCs.

In view of potential interference from L1, cognitive complexity of referential

accessibility, and low frequency of NRRC input in text, it becomes all the more

important for classroom instruction to underscore the cross-linguistic difference in

NRRCs between learners’ mother tongue and English, assist learners in better

comprehending the concept of referential accessibility, and enhance the saliency of

features of NRRCs in textual input. To investigate instructional effects, instructional

materials for English, including textbooks of junior and senior high school and

grammar books available on the market, were scrutinized with regard to their

accounts of NRRCs, examples for illustrating how NRRCs are used, and exercises for

practicing NRRCs. The following are some observations made from a survey of

instructional materials in terms of their presentation of English NRRCs.

On the whole, instructional materials have overlooked usage of NRRCs. Most

material writers tend to devote much of the text to expounding how to construct RRCs

with various relative markers, such as relative pronouns (including simple, like which,

who and compound, like what, whoever), relative adverbs, and quasi-relative

pronouns (like as, than, but). Assuming similarities between RRCs and NRRCs in

their formation with relative markers, they are inclined to think it unnecessary to

further explicate usage of NRRCs. Under such circumstances, the presentation of

NRRCs is often neglected and treated as a dispensable, fragmented adjunct to that of

RRCs, merely cramming learners with supplementary facts about how RRCs contrast

with NRRCs in form and meaning. Little information is provided for learners on

how to use NRRCs correctly and appropriately in context.

In addition, instructional materials rarely, if ever, make an attempt to raise

learners’ awareness of the close relationship between NRRCs and referential

accessibility, i.e. the fact that it is essentially the referential status of head NPs, not

other factors, that determines the use or nonuse of NRRCs.

For one thing, instructional materials more often than not present NRRCs merely

with itemized prescriptive rules, common among which are those listed in (83):

(83) Rule 1: There should be commas or other punctuations like parentheses, dashes, setting off the main clause from the NRRC.

Rule 2: The relative marker that cannot be used in an NRRC.

Rule 3: Relative pronouns in the object position of an NRRC cannot further be deleted.

Rule 4: When the antecedent is a personal pronoun, proper NP, or one-of-a-kind NP, an NRRC should be used to modify it.

All these prescriptive rules do little in making learners acutely aware of the

correlation between NRRCs and referential accessibility. Rules 1-3 focus learners’

attention solely on syntactic restrictions with the formation of NRRCs. While rule 4

identifies for learners certain NP contexts for using NRRCs, it does not offer any

explanation for why NRRCs should be used in these NP contexts. When no account

is given of the fact that it is essentially the referential accessibility of these NPs that

obligates the use of NRRCs, rule 4 could give learners the impression that the use of

NRRCs is decided solely upon the presence of such uniqueness-referring NPs as

personal pronouns, proper NPs, one-of-a-kind NPs, rather than the referential status of

NPs.

For another, in introducing NRRCs, instructional materials tend to rely heavily

on context-reduced examples with both restrictive and non-restrictive interpretations,

as in (84):

(84) a. Sam has a daughter who studies in college. (Meaning: Sam has more

than one daughter, and one of them studies in college)

b. Sam has a daughter, who studies in college. (Meaning: Sam has only

one daughter, and she studies in college)

Although examples with parallel sentences like (84) can readily be utilized to make a

direct comparison of RRCs and NRRCs in their meanings, they to a great extent

desensitize learners to the relationship between NRRCs and referential accessibility in

that there is not enough context to accentuate the referential status of the antecedent in

question. These decontextualized examples may very well mislead learners into

thinking that NRRCs are used only with NPs characterized by the connotation of

“being the only one.”

As for exercises for NRRCs, instructional materials, in the main, fail to design

context-embedded exercises to help learners practice using NRRCs according to

whether their antecedents are referentially accessible in context. This is because the

prevalent exercises for NRRCs are sentence-combining, which tends to be

context-reduced, as exemplified in (85):

(85) The storm caused a lot of damage.

Nobody had expected the storm.

(Ans.: The storm, which nobody had expected, caused a lot of damage)

At best, exercises like (85) merely acquaint learners with how to combine two

sentences into an NRRC using appropriate relative markers. They fall short of

equipping learners with knowledge of how to use NRRCs in accordance with context,

as they do not furnish any contextual clues for learners to perceive the referential

accessibility of a given NP and see a need to use an NRRC in preference to an RRC.

To put it another way, instead of sensitizing learners to the relationship between

NRRCs and referential accessibility, sentence-combining only serves to familiarize

them with those prescriptive rules stated in (83), particularly rules 1-3, regarding

structural aspects of NRRCs.

The above inadequacies in the presentation of NRRCs by instructional

materials—neglect of usage of NRRCs, failure to highlight the relationship between

NRRCs and referential accessibility, and lack of contextualized exercises for using

NRRCs with reference to context—may partly be responsible not only for the

subjects’ great confusion between RRCs and NRRCs in their mental grammar of

English RCs, evinced by a tendency to overuse the former, but also for their

insufficient knowledge of when to employ NRRCs, evinced by a propensity to use

NRRCs only with uniqueness-referring NPs to the exclusion of other referentially

accessible NPs.

What is more, instructional effects may even account for why generic NPs were

the only non-lexically accessible NP context that displayed a significant performance

difference between the first- and third-graders, as indicated in Table 10. With more

years of exposure to instruction, the third-graders may have been more familiar with

the prescriptive rule that emphasizes the co-occurrence of NRRCs with

uniqueness-referring NPs (e.g. personal pronouns, proper NPs, and one-of-a-kind

NPs), and thus more able to generalize the use of NRRCs by analogy to generic NPs,

which share a similarity with uniqueness-referring NPs in their semantic meaning of

“particularness”: the former refers to particular classes or categories, and the latter,

particular entities.

4.2.2 Use of RCs in Different Pragmatic/Discourse Contexts － － － － Identifying, Characterizing, Presentative, and Parenthetical

In the context translation test, it was observed that the subjects’ employment of

RCs were more bound up with the prototypical functions of identifying or

characterizing a referent, and less associated with the atypical functions of presenting

a topic or interpolating parenthetical information. The subjects’ rare use of RCs for

the latter two functions is understandable, considering that generally, EFL instruction

makes few attempts to well acquaint learners with such advanced RC uses.

On the one hand, learners may not be familiar with the presentative use of RCs,

given that the teaching of RCs is often taken up at a sentential level, with little

discourse context to help learners attend to such use. With great emphasis on form,

traditional grammar teaching commonly presents language in isolated, decontexulized

sentences. Despite its merit in helping learners readily discern structural features in

RC formation, such a sentence-based approach to grammar is not conducive to

shedding light on the presentative use of RCs, a function best understood only within

connected stretches of discourse. As an illustration of this point, consider (86):

(86) Each of us can imagine difficulties that could result from the introduction of

a cloned child.

To learners, it would seem that the only function that the underlined RC performs is to

characterize the indefinite head NP difficulties with new information; there is nothing

“presentative” about the RC. Nonetheless, when the same RC is analyzed in larger

discourse, as in text (87) below, the discourse function of RCs is made more explicit

and easier for learners to perceive:

(87) Each of us can imagine difficulties that could result from the introduction of a cloned child. For example, it would be very difficult for the parent that was cloned not to have specific ideas about how the “copy”

(the cloned child) should act and develop. Seeing their own image would make it awfully difficult not to impose expectation on their cloned child.

Conversely, how would the cloned teenager react to the parent, seeing their physical future ahead?

--Ian Wilmu, “To Clone or not Clone: Implications of Human Cloning”

(from Chen, 2004: 24) From text (87), it is clear that besides characterizing, the underlined RC also serves to

present the new NP difficulties into the discourse as a persistent and important referent.

Occurring in the beginning position of a discourse unit, the RC helps mark the new

NP as the topic for the ongoing discourse by furnishing it with a salient initial

description that facilitates subsequent reference. Moreover, it contributes to the

development of this discourse topic by coding information about it that is going to be

further discussed, as can be seen from the remaining sentences of the text, all of

which elaborate on how difficulties could result from child cloning. In short, the

above examples, (86) and (87), simply demonstrate that in the absence of a

discourse-oriented approach, traditional grammar teaching may very well be

responsible for learners’ failure to fully appreciate the presentative function of RCs

and to make good use of RCs in written discourse as an effective strategy for topic

construction.

On the other hand, learners may not be cognizant of the parenthetical use of RCs,

for the presentation of NRRCs often focuses not so much on the function of NRRCs

as on the formal and semantic differences between RRCs and NRRCs. As pointed

out in Section 4.2.1, because of their undue stress on usage of various relative markers

in the formation of RRCs, instructional materials tend to give a rather perfunctory and

simplistic account of NRRCs. At best, their introduction of the construction is

nothing more than a list of features whereby RRCs contrast with NRRCs; they

scarcely ever go further to draw learners’ attention to the information-adding function

of NRRCs themselves. With the under-teaching of NRRCs in EFL instruction,

learners are merely crammed with facts about how RRCs and NRRCs differ from

each other, and they may not have the slightest idea of when to use NRRCs.

Unaware of such a device, more sophisticated than independent clauses, at their

disposal to provide supplementary facts, remarks, or explanations, learners are very

likely to end up underusing NRRCs in expressing their stream of thoughts.

In addition to their under-learning of RC functions, the subjects’ sporadic use of

RCs in presentative and parenthetical contexts, as revealed by a close examination of

their translation in the test, also has something to do with their propensity to use

independent clauses in place of RCs for such contexts. The finding implicates that

when expressing themselves in English, the subjects still had a tendency to transfer

the same structure as used in their mother tongue, Chinese, into English, wrongly

assuming that Chinese and English employed the same syntactic strategy to achieve a

particular communicative function. In fact, because of their distinctive pragmatic

features, Chinese and English differ from each other in their preferred structure for

presentative and parenthetical contexts. A discourse-oriented language, Chinese is

characterized by a topic chain, which is syntactically realized as a series of

independent clauses, with the first clause serving as a common topic and the rest

making comments about it. Under this topic chain, the structure preferred in

Chinese for presentative and parenthetical contexts is often a pair of independent

clauses strung together, with the first one being commented on by the second (Lin,

2002: 108-109). By contrast, English, a sentence-oriented language, typically

utilizes different sentence structures to accommodate different kinds of information.

In English, there is nothing wrong with using independent clauses for presentative and

parenthetical contexts; however, in terms of information packaging (as well as

stylistic considerations), RCs are more preferred: RCs used in a presentative context

serve to “spotlight” (i.e. foreground) the important information about a topical

referent that is going to be further developed in the subsequent discourse, while those

used in a parenthetical context help to “sidelight” (i.e. background) the supplementary

information in order to distinguish it from the main assertion, coded in an independent

clause. Probably because such a cross-linguistic difference in preferred structure is

too subtle to perceive, nearly half of the subjects in the study still viewed independent

clauses, the same structure used in Chinese, as appropriate structural counterparts in

English for presenting a topical referent or inserting additional information; seldom

did it cross their mind that in English, there exists an alternative structure, i.e. RCs,

that is used much more commonly for presentative and parenthetical contexts.

Finally, with regard to their use of articles with RCs in a characterizing context,

the subjects were found to exhibit a certain tendency to misuse the definite article

with characterizing RCs. The finding can partially be seen to provide empirical

support for the researcher’s observation concerning the influence of false instruction,

which motivates the second part of the second research question. As briefly pointed

out in Chapter three, quite often, especially in learning when to use the definite article,

learners have been told to use the definite article before NPs with RCs. This

prescriptive rule in fact makes some sense, since as a post-nominal modifier, RCs add

extra information to NPs being modified, and the more information is given, the more

likely these NPs are accessible (i.e. definite) to the hearer. Nevertheless, the rule is

not infallible in that it cannot account for RCs with indefinite NPs, i.e. those which

function mainly to characterize NPs as a particular category or class of people or

things, as opposed to identifying them as familiar entities. Under such false

instruction, emphasizing the co-occurrence of the with RCs without any reference to

their function, learners are very likely to get into the habit of overusing the with RCs,

even in a characterizing context.

As plausible an explanation for the subjects’ tendency to misuse the definite

article with characterizing RCs is RC input received during their English study. By

glancing through the presentation of RCs in instructional materials, one would get the

impression that RCs which he/she comes across are predominantly ones modifying

definite NPs, with only a few—or in some extreme cases, “zero”—instances of RCs

modifying indefinite NPs. To corroborate this assertion, a survey was conducted

with one English textbook and four grammar books concerning occurrences of

identifying and characterizing RCs in the examples and exercises of their grammar

units for RCs. Table 19 displays the number of each type of RRCs in the five

instructional materials:

Table 19: Occurrences of RRCs modifying definite and indefinite head NPs in five instructional materials

RRCs Material* TB GB1 GB2 GB3 GB4

RRCs with definite head NPs 25 18 40 43 27

RRCs with indefinite head NPs 0 8 8 14 15

* Instructional materials under investigation here included one English textbook, published by Joy (Units 7 and 8 in Book 5), one of the various versions used by junior high schools in Taiwan, and four grammar books on the market: 我的第一本文法書(GB1), 無敵英文文法寶典(GB2), Step by Step 搞 定英文文法(GB3), 豎起耳朵學文法(GB4).

From Table 19, it is clear that in these instructional materials, there is a disproportion