READING COMPREHSNION IN ADOLESCENT BILINGUALS LEARNING … · Two studies examined reading comprehension (RC) among adolescent bilinguals learning English as a second language (L2).
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
READING COMPREHSNION IN ADOLESCENT BILINGUALS LEARNING ENGLISH AS A SECOND LANGUAGE
by
Adrian Dominic Kenneth Pasquarella
A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy
Graduate Department of Applied Psychology and Human Development Ontario Institute for Studies in Education
Chapter 2: Study 1 - Models of RC in Spanish-English and Chinese-English bilinguals: Universal and Language Specific Aspects of Comprehension in the L1 and L2.... 4
Theoretical Models of RC ……..…….……………………………..……………. 5
Typological characteristics of English, Spanish, and Chinese….……………..…. 7
Orthography and phonology………………………………………………….. 7
Morphology…………………………………………………………………... 8
Syntax……………………………..………………………………………….. 9
Oral Language Proficiency and RC………………………………………………. 10
The Role of Word-level Reading Within and Across languages..…….………. 76
The Role of Vocabulary Within and Across Languages………………………. 78
The Role of RC Across Languages…………………………….…………….... 79
Conditions that Support Cross-language Relationships...……………………... 81
Limitations and Future Directions………………..……………………….…… 81
Implications………………………………………………………………….… 83
Chapter 10: General Conclusions…………………………………………………………….. 85
References……………………………………………………………………………………. 87
viii
Lists of Tables
Table 1. Descriptive statistics and Analysis of Variance of age, length of residence, English and L1 variables.....…………………………………………………..……. 102
Table 2. Intercorrelations among English variables for Chinese-English bilinguals and Spanish-English bilinguals……………………………………………..…………... 103
Table 3. Intercorrelations among Spanish and control measures…………………………….. 104
Table 4. Intercorrelations among Chinese and control measures…………………………….. 105
Table 5. Coefficients for structural equation model of English RC………………………….. 107
Table 6. Coefficients of bootstrapping for indirect effects on English RC.………………….. 108
Table 7. Comparison of fit statistics by constraining parameter estimates to be equal across groups……………………………………………………………….………..…….. 109
Table 8. Coefficients for structural equation model of first language RC……...……………. 111
Table 9. Coefficients of bootstrapping for indirect effects L1 RC…………………………… 112
Table 10. Descriptive statistics and Analysis of Variance for Chinese EFL learners, Recent Immigrants, and Long-term Immigrants for age, months in Canada, English and Chinese variables…..………………..………....………………………………….. 115
Table 11. Intercorrelations among Chinese and English variables separated by group..…….. 116
Table 12. Hierarchical linear regression predicting English RC, separate for EFL, Recent Immigrant, and Long-term Immigrant groups..………………………………….. 117
Table 13. Hierarchical linear regression predicting Chinese RC, separate for EFL, Recent Immigrant, and Long-term Immigrant groups……………….…………………….. 118
Table 14. Commonality analyses beta weights, structural coefficients, proportions of total, common and unique variance for EFL, Recent Immigrant and Long-term Immigrant groups when predicting English RC…….……………………………… 119
Table 15. Summary of unique and common variance for independent variables when predicting English RC...…………..……………………….……………………….. 120
Table 16. Commonality analyses beta weights, structural coefficients, proportions of total, common and unique variance for EFL, Recent Immigrant and Long-term Immigrant groups when predicting Chinese RC………………………………….... 121
Table 17. Summary of unique and common variance for independent variables when predicting L1 RC….……………………………………………..….……………… 122
ix
Lists of Figures Figure 1. Theoretical structural equation model of RC in English, Spanish and Chinese…… 106
Figure 2. Structural models of English RC showing direct and indirect effects…...………… 110
Figure 3. Structural model for Spanish RC (Model 3) showing direct and indirect effects….. 113
Figure 4. Structural model for Chinese RC (Model 4) showing direct and indirect effects …. 114
x
Chapter 1: Introduction
Reading comprehension (RC) is critical for school success as it acts as a gateway to
academic and cultural knowledge. RC enables self-directed learning, and contributes to the
development of the mental lexicon (Cunningham & Stanovich, 2001). Bilingual children and
adolescents often experience delays in developing oral language and RC proficiency when
compared to native speakers of English (August, Carlo, Dressler & Snow, 2005; August &
(Proctor et al., 2012) or sentence completion tasks (Chik et al., 2012; Low & Siegel, 1995;
Yueng et al., 2011). It is likely that the different formats of syntactic awareness tasks require
different cognitive and linguistic skills. The current study uses implicit judgment tasks as they
require less vocabulary knowledge than other syntactic awareness tasks. Previous research has
suggested that syntactic awareness may share a unique relationship with RC, but this relationship
was weakened once vocabulary was factored into the equation (e.g., Zhang, 2012). Research is
needed to understand how the relationships between vocabulary, morphology and syntax jointly
influence RC in adolescent bilinguals to gain a clear understanding of each component’s unique
and shared influence on comprehension. Furthermore, examining the contributions of these
skills to RC in English, Spanish, and Chinese outlined similarities and differences of predictors
of RC across languages.
The current study created comparable measures of syntactic awareness to evaluate its
contribution to RC in English, Spanish and Chinese. For each language a judgment task that
required students to determine whether an orally presented sentence was grammatically correct
was used to assess syntactic awareness. The syntactic awareness tasks contained multiple
syntactic features that were representative of the languages understudy. Some of the features
were unique to a particular language, while other features were common to either Spanish and
English or Chinese and English. The goal was to create measures that assessed the participant’s
syntactic awareness regarding multiple representative features found within that language (e.g.,
tense, number, the use of articles or function words, word order, etc.), instead of focusing on one
particular element (e.g., word order).
19
Objectives
The current investigation examined models of RC for bilingual adolescents from two
different L1 backgrounds: Spanish and Chinese. The primary objective was to examine the
relationships of vocabulary, morphological awareness and syntactic awareness with RC.
Comparisons were made across English, Spanish and Chinese to identify how the components
varied across languages. Specifically, four different models were compared: (1) English RC for
Spanish-English bilinguals, (2) English RC for Chinese-English bilinguals, (3) Spanish L1 RC,
and (4) Chinese L1 RC. Identifying relationships across Spanish, Chinese, and English can
reveal language-universal and language-specific aspects of RC.
I hypothesized that vocabulary knowledge would be consistently related to RC in English,
Spanish and Chinese, and would therefore be universally important for RC. On the other hand,
language-specific relationships may also be present. For example, derivational morphological
awareness may be related to RC in Spanish and English but not Chinese, whereas compound
morphological awareness may demonstrate a stronger relationship with Chinese RC than with
Spanish or English RC. The salience and frequency of morphologically complex words
encountered in a language may have implications for how these skills are utilized to comprehend
different languages. Considering syntactic awareness is a predictor of English and Chinese RC
(Chik et al., 2012; Proctor et al., 2012; Yeung et al., 2011; Zhang, 2012) it may also be a
significant predictor for Spanish RC.
20
Chapter 3: Method
Participants
Participants included 117 Chinese-English bilinguals (84 Females; 33 Males) with a
mean age of 17.99 years old (SD = 3.4 years) and 71 Spanish-English bilinguals (43 Females; 28
Males) with a mean age of 19.65 years (SD = 3.36 years) enrolled in high school or in
undergraduate university programs in southern Ontario, Canada. All Chinese-English bilinguals
spoke Mandarin as their L1. The majority were from mainland China (80%), 10% were from
Taiwan, and the remaining 10% did not indicate a country of origin. On average, the Chinese-
English bilinguals had been living in Canada for 6.88 years (SD = 4.34 years). All Spanish-
English bilinguals spoke Spanish as their L1. On average, this group of students had been living
in Canada for 12.04 years (SD = 6.98 years). Ten percent of Spanish-English bilinguals were
from Colombia, 10% were from Mexico, 5% were from Ecuador, and 40% of participants were
distributed among Argentina, Chile, Cuba, Peru, Philippines, Spain, or the US. Additionally,
there were approximately 20% who were born in Canada and another 15% did not provide a
country of origin.
With respect to parental education, approximately 80% of the Chinese-English
bilingual’s parents had completed college or university at the undergraduate or graduate level,
while about 20% of the participant’s parents had completed high school or less. For the Spanish-
English bilinguals, approximately 70% of the participant’s parents had completed college or
university at the undergraduate or graduate level, while about 30% completed high school or less.
For the Chinese-English bilinguals, approximately 74% of the sample reported speaking
to their parents only in Chinese. Another 26% reported speaking with their parents frequently in
Chinese and occasionally in English. The majority of Chinese-English bilinguals (83%) reported
21
conversations with their friends occurring mostly in English. The other 17% reported speaking
with their friends in both English and Chinese to a similar degree. For the Spanish-English
bilinguals, approximately 25% percent of the sample reported speaking to their parents only in
Spanish. Another 50% reported speaking with their parents frequently in Spanish and
occasionally in English. The remaining 25% reported speaking to their parents mostly in English
and rarely in Spanish. About 75% Spanish-English bilinguals reported conversations with
friends occurring mostly in English and occasionally in Spanish. The other 25% reported
speaking with their friends in English and Spanish to a similar degree.
For the Chinese-English bilinguals, 40% of the sample were enrolled in a Chinese
language class. The majority of students who were not enrolled in heritage language classes had
completed high school in China and came to Canada for the purpose of attending university. For
the Spanish-English bilinguals, only 8 students were enrolled in Spanish language classes. For
both groups, L1 languages classes lasted about three hours per week (the range was 2.5 to 3.5
hours per week).
Procedures
Participants completed measures in both their L1 and English. In total, testing took
approximately 4 hours to complete, which was 2.5 hours in English and 1.5 hours in the L1.
Generally, all the English measures were completed in one session and the L1 measures were
completed in another session, with the remainder of English or L1 measures completed in a third
session for a few participants. Tasks within each testing were administered by a trained research
assistant. Typically the English measures were administered before the L1 measures. Measures
of RC, vocabulary, morphological awareness and syntactic awareness in English and the L1 were
completed in groups. Additionally, the demographic and language use questionnaire was also
22
administered during group testing. The measures of word reading, non-verbal reasoning and
working memory were administered individually.
Measures
Reading Comprehension. The Second Canadian Edition of the Gates MacGinitie RC
test (MacGinitie, MacGinitie, Maria, Dreyer, & Hughes, 1992) was used to assess RC in English,
Chinese and Spanish. Levels E and F were used for high school and undergraduate students
respectively to ensure that students worked through the test most appropriate for their age and
ability. Form 3 of Levels E and F were translated into Spanish and Chinese. Form 4 of Levels E
and F were used for English RC. Each test consisted of a series of short passages with multiple
choice questions. In total, each test had 48 multiple choice questions each with 4 or 5 response
options. Students were given 45 minutes to complete this task in each language. Reliabilities1 in
English for Forms E and F were .84 and .93 for the Chinese-English bilinguals, and .83 and .91
for the Spanish-English bilinguals. Reliabilities, based on translated Form 3 Levels E and F,
were .89 and .93 for the Chinese RC test and.87 and .89 for the Spanish RC test
Vocabulary. The Second Canadian Edition of the Gates MacGinitie Vocabulary
(MacGinitie et al., 1992) was used to assess vocabulary knowledge in English, Spanish and
Chinese. Form 4 of Levels E and F were used for high school and undergraduate students
respectively to ensure the participants worked through the test most appropriate for their age and
ability. Form 3 of Levels E and F were translated into Spanish and Chinese. To ensure that the
difficulty of items was similar across languages, word frequencies were tabulated based on the
SUBTLex databases for English (Brysbaert & New, 2009), Spanish (Cuetos, Glez-Nosti, Barbón,
& Brysbaert, 2011) and Chinese (Cai, & Brysbaert, 2010). First, word frequencies were
1 Cronbach’s alpha reliability ratings of internal consistency were calculated based on item-level data from the participants within the current study for all measures in the L1 and L2.
23
calculated for the English tasks to be translated. Then the word frequencies for the translated
tasks were cross-referenced with the English version. For situations where differences between
frequencies were noted, the translated word was replaced with another word in the database that
had a similar frequency to the original target item. T-tests revealed non-significant differences in
the frequency ratings (using LOG frequency per million values) across languages for the
different forms, suggesting the L1 measures are comparable in difficulty to their English
counterparts (p range was .55 to .95). Each test consisted of target words with multiple choice
responses. Each test had 45 target words with 4 or 5 potential definitions. Students were given
35 minutes to complete this task in each language. Reliabilities in English for Form 4 Levels E
and F were .94 and .89 for the Chinese-English bilinguals and .90 and .91 for the Spanish-
English bilinguals. Reliabilities, based on translated Forms E and F, were .90 and .90 for
Spanish vocabulary, and .89 and .94 for Chinese vocabulary
Morphological awareness.
Compound awareness. Experimental compound awareness measures were created in
English, Spanish and Chinese following the same instructions and format. The tasks involved
the generation of the name (or label) of a novel noun based on a short description. The measures
contained non-recursive (two word compounds) and recursive compounds (more than two word
compounds) to increase the difficulty of the measures. A sample English item is: “A factory that
recycles sweaters is called a (sweater recycling factory)”. The task was administered with an
oral recording of the instructions and items. Participants followed along with a written copy and
provided a written responses for the answers. There were six practice trials for non-recursive
items followed by 15 test items. Then there were 4 practice items for recursive compounds
followed by another 10 test items for a total of 25 test items in both the English and Chinese
24
tasks. The Spanish task was the same as the English and Chinese tasks except there were 5
recursive test items (instead of 10), tabulating to 20 test items in total. Participants were
encouraged to complete all items. Cronbach’s alpha reliabilities for the English task for
Chinese-English and Spanish-English bilinguals were .80 and .81, respectively. Cronbach’s
alpha reliabilities were .86 for the Spanish task, and .86 for the Chinese task.
Derivational awareness.
Derivational production. The derivational production tasks were adapted from Carlisle
(2000) for both English and Spanish (Ramirez, Chen, Geva & Kieffer, 2010). Participants were
presented with a root word (target word) and an incomplete sentence via an audio recording, and
followed along with a written copy. The participants were asked to write a written response by
modifying the target word to correctly complete the sentence. An example in English is: “Glory
(target word). The view from the hilltop was (glorious).” There were 31 items for the English
measure and 26 items for the Spanish measure. Participants were asked to complete all items.
Reliabilities of the English task for the Chinese-English and Spanish-English bilinguals were .84
and .90, respectively. Reliability for the Spanish task was .92.
Derivational structure. The derivational structure tasks (adapted from Ramirez et al.,
2010) were created for English and Spanish. The participants were asked to identify the
morphologically complex pseudoword that correctly completed the sentence. Participants
listened to a recording of the sentences with the multiple choice options. The participants then
circled the correct response. In each item, all the choices had the same root but different suffixes.
An example item in English was: “The young baby was sleeping so… a) terbious b) terbiously c)
terbiouses d) terbiousal”. There were 21 items on the English task and 21 items on the Spanish
25
task. Reliabilities for the English task for Chinese-English and Spanish-English bilinguals
were .73 and .79, respectively. The reliability for the Spanish task was .90.
Chinese homographic awareness. An experimental homographic awareness task was
created in Chinese to examine derivational and compound awareness. In this task, participants
were shown a two-character word with a target morpheme [e.g., 商量 (discuss measure: discuss),
with 商 (discuss) as the target morpheme] followed by three two-character options containing
either the target morpheme [e.g., 商定 (discuss decide: decide)] or a morpheme that was a
homonym [e.g., 商店 (business shop: store) and 商人 (business person: businessman)]. The
participants were instructed to pick the option that contained the target morpheme. The task
contained two practice items and 40 test items. There were 20 derivational and 20 compound
items in the task. The test and practice items were recorded and read to participants as they
worked through the written test. The reliability of this task was .87.
Syntactic awareness. Experimental measures of syntactic awareness were constructed
for English, Spanish and Chinese. Participants were asked to judge sentences as being
grammatically correct or incorrect. Sentences were recorded and orally administered to keep
pronunciation and prosody consistent. Students were not shown sentences in the written form.
Students circled “T” if the sentences sounded grammatically correct or “F” if they sounded
grammatically incorrect. The task was modelled after Johnston and Newport’s (1989)
grammatical judgment task. Measures of the task in each language comprised multiple sentences
with representative grammatical features (e.g., third person singular, past tense, uninflected
modals) from that language. For each grammatical feature a short and long sentence (based upon
syllable length) was created, as well as a correct and incorrect version of each. Then all items
were randomized and counterbalanced across two alternate forms to create a balance of long vs.
26
short and incorrect vs. correct items within each form. Participants received all items from both
forms, for a total of 72 items in English, 78 items in Chinese and 56 items in Spanish, with two
practice items in each language. An example of a correct item in English was, “Ann is the one
whose husband is wearing that funny hat”, and the incorrect counterpart was, “Ann is the one
who husband is wearing that funny hat”. Reliabilities for the English task for Chinese-English
and Spanish-English bilinguals were .84 and .87 respectively. Reliability was .80 for the Spanish
task, and .85 for the Chinese task.
Word reading and decoding. For English, the Word Identification subtest of the
Woodcock Reading Mastery Test–Revised (WRMT–R; Woodcock, 1987) was used to assess the
participants’ word reading accuracy. The test involved participants reading aloud a list of words
of increasing length and difficulty. The reliabilities of the Word Identification were .91 and .92
for the Spanish-English and Chinese-English bilinguals respectively. The Word Attack subtest
of the WRMT–R was used to assess pseudoword reading. The test required the participants to
read aloud a list of pseudowords of increasing complexity. The Word Attack subtest was used to
assess proficiency with English phonological decoding. A stop rule of six consecutive words
read incorrectly was used for both tasks. The reliability ratings of the Word Attack subtest
were .79 and .78 for Chinese-English and Spanish-English bilinguals respectively.
The Identification de letras y palabras subtest of Woodcock Language Proficiency
Battery-Revised Spanish Edition (WLPB-S, Woodcock & Munoz-Sandoval, 1995) was used to
assess Spanish word identification skills. This task was analogous to the English version. The
participants were encouraged to read aloud words that increased in length and difficulty. The
same stop rule for the English tasks was used for the Spanish tasks. The reliability for the
Spanish word reading task was .86.
27
Considering that no standardized test of Chinese word reading was available, an
experimental test was created (based on the test in Luo, Chen, Deacon, Zhang, and Yin, 2013)
that included 240 items presented in order of decreasing word frequency based on the corpus
study conducted by Shu et al. (2003). The majority of items were selected from 12 volumes of
the Elementary School Textbooks (Elementary Education Teaching and Research Center, 1996)
used from grades 1 to 6 in Mainland China. Additional low frequency words were selected from
the Chinese SUBTLex database (Cai, & Brysbaert, 2010) to increase the difficulty of the task.
The test was discontinued when 10 consecutive characters were read incorrectly. The reliability
of this task was .99.
Non-verbal reasoning. The Matrix Analogies Reasoning Test (Naglieri, 1985) was
administered to measure Non-verbal reasoning. To shorten administration time, only subtests 2
and 4 were presented. Each subtest contained 16 items leading to a potential maximum score of
32. For each item, the participants were shown a pattern with one section missing and were
asked to choose one of six options that correctly completed the pattern. A stop rule of 4
consecutive incorrect items was used for each subtest. Reliabilities for the Chinese-English and
Spanish-English bilinguals were .76 and .86, respectively.
Working memory. The working memory task was adapted from Daneman and
Carpenter (1980) (see also Gottardo, Stanovich & Siegel, 1996). In this task, participants
listened to sets of pre-recorded statements containing familiar information (e.g., Cars have four
wheels; Fish swim in the sky). The participants indicated whether each sentence was true or
false. Additionally, the participants were required to recall the final word of each statement
(wheels, sky) at the end of a set. A set contained either two, three, or four statements. The
number of final words remembered was calculated, giving a possible maximum score of 42. The
28
reliabilities of this task for the Chinese-English and Spanish-English bilinguals were .90 and .91
respectively. Only an English measure of working memory was developed and administered.
Demographic questionnaire. Participants were asked to fill out a questionnaire to
collect information on their age, grade, country of origin, age of immigration, and length of
residence in Canada. Also, the questionnaire asked students to indicate their parents’ level of
education. The questionnaire also asked to students to report how often they spoke English and
their L1 with family and friends, using a 5 point Likert-type scale (never-rarely-sometimes-
often-frequently). Finally, students reported whether they were currently enrolled in L1
language class (with a yes or no response) and how many hours each week they spent in the class.
Measurement comparability across languages
Central to the design and interpretation of the current study is the comparability of
measures across languages. To create measures that were parallel across languages different
techniques were employed for the various constructs. For RC and vocabulary, an alternate form
of a standardized task was translated into Spanish and Chinese. Translation tasks of RC and
vocabulary were used to ensure that the format and the method of questioning were comparable
across languages. To ensure accurate and authentic translation of the tasks, native speakers of
Chinese or Spanish who were highly proficient in English completed the translations.
Experimental measures were created for morphological awareness and syntactic awareness. To
ensure comparability across languages, the same instructions and format were used in all
languages when possible. Finally, standardized measures of decoding and word reading were
available for Spanish and English and were therefore used. An experimental Chinese character
reading task was adapted from a previously published study (Chen et al., 2009) by adding more
difficult items to ensure it was appropriate for adolescent Chinese speakers.
29
Chapter 4: Results
Mean Scores and Group Differences
Table 1 displays minimum and maximum scores, means, and standard deviations for age,
length of time in Canada, and all measures for the Spanish-English and Chinese-English
bilinguals. Skewness and kurtosis values were also examined for all measures. If the values fell
outside the range of +/- two standard errors then transformations to improve normality were
performed (as per Tabachnick & Fidell, 2007). Transformation was performed on measures of
non-verbal reasoning, word reading in the L1 and English, English derivational production,
English derivational structure for both groups, and the Chinese homographic awareness task2 for
the Chinese-speaking bilinguals. However, subsequent analysis revealed transforming data did
not change patterns of correlations or results of structural equation models. Therefore, only raw
score analyses are reported here3.
Standardized scores were calculated for English measures when norms were available.
For English word identification and word attack, both groups performed close to the standardized
mean. For English vocabulary and RC, grade equivalent scores are presented. Vocabulary and
comprehension grade equivalent scores were between a 9th and 10th grade. On average, the
bilingual students were performing approximately 2 grade levels below on these measures.
A multivariate analysis of variance (MANOVA) was conducted to test for differences
among the Spanish-English and Chinese-English bilinguals for age, months in Canada, and the
English variables. There was a statistically significant difference between the two groups on
2 The correlation between the derivational and compound items was .91 (p < .001). The combined score of all items was used in the current and subsequent analyses. 3 As an additional check to ensure administering both levels E and F within the same sample did not influence the results, subsequent analyses (i.e., t-tests, correlations, and structural models) were conducted using raw scores and extended scale scores. Extended scale scores are typically used to allow for different versions of the tasks to be combined for analysis (MacGinitie et al., 1992). The two methods produced identical findings. Therefore, to avoid redundancy only raw score analyses are presented below.
30
these variables, Wilks’ λ = .61, F (12, 175) = 9.17, p < .001. One-way univariate analysis of
variance (ANOVA) by group (Chinese-English vs. Spanish-English bilinguals) conducted for
English measures, age, and length of residence are reported in Table 1. Scores on L1 measures
were not compared as measures were administered in different languages and contained different
items to reflect language specific characteristics important for Spanish or Chinese. The one-way
ANOVAs revealed that age was significantly different between the two groups. On average
Chinese-English bilinguals were about 15 months younger than the Spanish-English bilinguals.
Additionally, the Chinese-English bilinguals had been in Canada a significantly shorter amount
of time than the Spanish-English bilinguals. The Chinese-English bilinguals scored significantly
higher on the MAT but significantly lower on the English word identification and vocabulary
tasks than the Spanish-English bilinguals. Scores on all other measures, including English RC,
were not significantly different between groups.
Correlations
Table 2 presents the correlations among age, length of time in Canada, and all English
measures for the Chinese-English bilinguals above the diagonal and the Spanish-English
bilinguals below the diagonal. To safeguard against Type I error correlations significant at the
level of p < .01 and lower were considered meaningful. All English measures were positively
correlated with RC for both groups. Additionally, age was negatively correlated, and length of
residence in Canada was positively correlated, with RC for the Chinese-English bilinguals only.
For both groups, measures of English word reading, vocabulary, morphological awareness,
syntactic awareness, and word reading were all correlated.
Table 3 reports correlations among age, length of time in Canada, and all Spanish
measures for the Spanish-English bilinguals. All measures except for length of time in Canada,
31
working memory, and word reading were significantly correlated with RC. Age was positively
correlated with the Spanish measures of vocabulary, morphological awareness, syntactic
awareness and RC. Length of residence in Canada was negatively related to vocabulary,
derivational production, compound awareness and syntactic awareness, but not RC, derivational
structure or word identification. Additionally, measures of vocabulary, morphological awareness
and syntactic awareness were all significantly correlated.
Table 4 displays correlations among age, length of residence in Canada, non-verbal
reasoning, English working memory and all Chinese measures for the Chinese-English bilinguals.
All variables except for non-verbal reasoning and working memory, were correlated with RC.
Length of residence in Canada was negatively correlated with RC and all other Chinese variables,
and age was positively correlated with all Chinese variables. Chinese word reading, vocabulary,
morphological awareness, and syntactic awareness were positively correlated with RC.
Additionally, measures of Chinese word reading, vocabulary, morphological awareness, and
syntactic awareness were all positively correlated with each other.
Specification of Structural Models across Languages
The relationships between RC and the underpinning constituent components were
examined with equivalent statistical models for the Spanish-English and Chinese-English
bilinguals’ L1 and L2. To this end, four models were compared to identify similarities and
differences among predictors of RC. The four models were: (1) English RC for Spanish-English
bilinguals, (2) English RC for Chinese-English bilinguals, (3) Spanish RC, and (4) Chinese RC.
Direct and indirect relationships were examined as previous research has shown that
morphological awareness was often highly associated with vocabulary, with documented
evidence of the mediation of morphological awareness on RC through vocabulary in Spanish-
32
English bilingual children (Kieffer & DiFelice Box, 2013) and adult Chinese-speaking EFL
learners (Zhang & Koda, 2012). Furthermore, syntactic awareness has been indirectly related to
RC (Cain, 2007) and directly related to RC in Geva & Farnia’s (2012) longitudinal study of
elementary school children learning English as a second language. Additionally, other known
cognitive, linguistic, and socio-cultural predictors of RC were incorporated into the statistical
models to rule out the possibility that results could be due to spurious variables. Specifically,
decoding/word reading, non-verbal reasoning and working memory were added into the models
as various studies have shown these variables to be related to RC (e.g., Cain, 2007; Daneman &
To examine within- and cross-language relationships, measures of word-level reading,
vocabulary and RC were administered in both English and Chinese. Hierarchical linear
regression and commonality analyses were used to identify the unique and shared aspects of RC
in the L1 and L2. The advantage of conducting a commonality analysis in addition to the linear
regressions is that it can help understand common and unique contributions of L1 and L2 skills
to RC.
60
Chapter 7: Method
Participants
Participants comprised 57 English majors in Beijing, China (Chinese EFL learners), 60
Chinese-English bilinguals who were recent immigrants to Canada (recent immigrants), and 57
Chinese-English bilinguals who were long-term immigrants to Canada (long-term immigrants).
Chinese-English bilinguals who had lived in Canada for six years or less were classified as
recent immigrants; those who had lived in Canada for more than 6 years were classified as long-
term immigrants. Previous research has indicated that, on average, it takes approximately 6
years of schooling for most immigrants to reach average levels on standardized measures of
academic achievement and to gain proficiency in academic English (Collier, 1987). Therefore, a
cut-score of 6 years was chose to help distinguish groups of immigrants. Both immigrant groups
included students enrolled in high school or in undergraduate university programs in southern
Ontario.
The Chinese EFL learners had a mean age of 20.20 years (SD = 0.66 years), and the
majority (91%) were female. All students spoke Mandarin as their first language and had been
living in a Chinese-speaking society for their entire lives. Approximately half of the sample had
parents who had completed a high school education or less, another 40% had completed college,
and the remaining 10% completed professional or post-graduate degrees. The recent immigrants
had a mean age of 19.22 years old (SD = 2.16) and 80% were females. The recent immigrants
had been living in Canada for approximately 3.42 years (SD = 1.58 years). All students spoke
Mandarin as their first language, and 86% were from China, 10% from Taiwan, and the
remainder were from Indonesia and Malaysia. Approximately 60% of the group had parents
who had completed college or university, 15% of parents had completed only high school or less,
61
and the remaining 25% of parents had completed a post-graduate or professional degree. For the
long-term immigrants, the average age was 17.36 years old (SD = 1.88 years) and 63% were
female. On average, this group had spent 10.53 years in Canada (SD = 3.12 years). All students
spoke Mandarin as their L1, and 90% were originally from China with the other 10% were from
Taiwan. Fifty-five percent of the group reported parents who had completed college or
university, 20% who had completed high school or less, and 25% who had completed a
professional or post-graduate degree.
The use of Chinese varied across the three groups. For the Chinese EFL learners,
approximately 96% of the group reported always speaking to their parents in Chinese. The other
4% reported some conversations in English, but the majority occurred in Chinese. Eighty-eight
percent of the sample reported always speaking to their friends in Chinese. The other 12%
reported speaking to their friends occasionally in English, but the majority of conversations
occurred in Chinese. For the recent immigrants, 82% of the group reporting always speaking to
their parents in Chinese. The other 18% reported frequently speaking to their parents in Chinese,
and occasionally in English. Peer interactions, on the other hand, were more balanced in this
group. The majority (60%) reported speaking to their peers in both English and Chinese. Only
20% reported always speaking with their peers in Chinese, while almost 20% reported always
speaking with their peers in English. Finally, for the long-term immigrants, 67% reported
speaking with their parents only in Chinese. Another 25% reported frequently speaking with
their parents in Chinese. The remainder of the group (8%) reported rarely speaking with their
parents in Chinese. Interactions with peers, on the other hand, occurred predominately in
English, with 72% of the sample reported speaking with their friends only in English. Another
62
23% reported frequently using English and occasionally using Chinese, while the remaining 5%
reported using both Chinese and English some of the time.
Measures
The same Chinese and English measures described in Study 1 were used in Study 2.
Reading Comprehension. The Second Canadian Edition of the Gates MacGinitie RC
(MacGinitie et al., 1992) test was used to assess RC in English, and the alternative form
translation was used for Chinese.
For this study, the Chinese EFL learners received Level F of the Chinese translation of the Gates
MacGinitie and Level E for the English test. Pilot testing revealed Level F of the English task
was too difficult as the majority of students responded at the level of chance. For the recent and
long-term immigrants, both levels E and F were used for the Chinese and English tasks to ensure
that students worked through the test most appropriate for their age and ability (Level E was used
for students in High school and Level F was used for students in University). For the English
RC measures, Cronbach’s alpha was .71 for the Chinese EFL learners, .84 for the recent
immigrants and .93 for the long-term immigrants. Cronbach’s alpha calculated for the Chinese
comprehension task was .70 for the Chinese EFL learners, .89 for the recent immigrants, and .93
for the long-term immigrants.
Vocabulary. The Second Canadian Edition of the Gates MacGinitie Vocabulary test
(MacGinitie et al., 1992) was used to assess vocabulary knowledge in English. As with the
comprehension tests, the EFL students received Level F for Chinese and Level E for English. For
the recent and long-term immigrants, both levels E and F were used to ensure that students
worked through the test most appropriate for their age and ability. For the English vocabulary
measures, Cronbach’s alpha was .77 for the Chinese EFL learners, .89 for the recent immigrants,
63
and .94 for the long-term immigrants. Cronbach’s alpha for the Chinese vocabulary task for the
EFL group was .76, and .89 for the recent immigrants, and .93 for the long-term immigrants,
respectively.
Word-level reading. To measure English decoding, the pseudoword reading subtest,
Word Attack from the WRMT-R (Woodcock, 1987), was administered. Cronbach’s alpha
was .77 for the EFL group, .72 for the recent immigrant group, and .86 for the long term
immigrant group.
To measure Chinese character reading a 240 item experimental task was used.
Cronbach’s alpha was .76 for the EFL group, and .99 for both the recent immigrant and the long-
term immigrant groups.
Demographic questionnaire. Participants were asked to fill out a questionnaire to
collect information on their age, grade, country of origin, age of immigration, and length of
residence in Canada. Also, the questionnaire asked students to indicate their parents’ level of
education. The questionnaire also asked students to report how often they spoke English and
their L1 with family and friends, using a 5 point Likert-type scale (never-rarely-sometimes-
often-frequently). Finally, students reported whether they were currently enrolled in L1
language class (with a yes or no response) and how many hours each week they spent in the class.
Procedures
The participants completed the Chinese and English measures in two or three testing
sessions. The vocabulary and RC tests and demographic questionnaire, were given as part of a
larger battery of group-administered measures. The English decoding and Chinese character
reading tasks were given as part of a larger individually-administered testing session. For the
EFL group, the Chinese measures were administered first, followed by the English measures.
64
The English measures were administered before the Chinese measures to the two immigrant
groups. The total testing time was 2 hours, including about 1 hour for English and 1 hour for
Chinese. All measures were administered by a trained research assistant.
65
Chapter 8: Results
Descriptive Statistics and Group Differences
Table 10 displays minimum and maximum scores, raw means and standard deviations for
age, length of residence in Canada, English and Chinese word-level reading, vocabulary and RC.
Skewness and kurtosis values were examined for all measures. Chinese word reading was
negatively skewed for the EFL group. Additionally, Chinese vocabulary was negatively skewed
for the long-term immigrant group. All other variables were normally distributed. Corrections
to normality, as per Tabachnick and Fidell (2007), did not change patterns of correlations,
regressions or commonality analyses. As such, subsequent analyses were conducted on
untransformed scores.
To check whether administering both levels E and F to the recent and long-term
immigrant groups influenced the results, analyses (i.e., t-tests, correlations, regressions and
commonality analyses) were conducted using both raw scores and extended scale scores. These
analyses produced identical findings. Therefore, for simplicity only raw score analyses are
presented below.
A multivariate analysis of variance (MANOVA) was conducted to test for differences
among the three groups (EFL students, recent immigrants, long-term immigrants) in age, months
in Canada, English and Chinese word reading, vocabulary and RC. There was a statistically
significant difference between the three groups on these variables, Wilks’ λ = .074, F (16, 328) =
55.09, p < .001. Univariate ANOVAs and Tukey’s post-hoc comparisons (see Table 10)
revealed that age was significantly different across the three groups. On average, the Chinese
EFL learners were the oldest by approximately one year. The long-term immigrants were almost
two years younger than the recent immigrants and three years younger than the Chinese EFL
66
learners. Length of residence was significant different across the three groups. As expected, the
recent immigrants had been in Canada significantly shorter time that the long-term immigrants.
It is important to note that length of residence was normally distributed for both groups, and a
few participants (n = 10 in total) were centered on the cut-point of 6 years. Regarding English
decoding, vocabulary and RC, the long-term immigrants scored highest, followed by the recent
immigrants and then the Chinese EFL learners. For Chinese character reading and vocabulary
measures, the Chinese EFL learners scored highest, followed by the recent immigrants and then
the long-term immigrants. For Chinese RC, the Chinese EFL learners and the recent immigrants
scored similarly, and both outperformed the long-term immigrants.
Correlations
Table 11 presents correlations for age, English and Chinese word reading, vocabulary and
comprehension for the Chinese EFL learners in the top panel, the recent immigrants in the
middle, and the long-term immigrants in the bottom panel. For the Chinese EFL learners,
English RC was correlated with both English decoding and vocabulary. Chinese RC was
correlated with Chinese character reading and vocabulary. All Chinese measures were positively
correlated with English RC. Age was not correlated with any variable in English or Chinese.
For the recent immigrants, English RC was related to English decoding and vocabulary.
Chinese RC was related to Chinese character reading and vocabulary. English RC was
positively related to Chinese RC. Additionally, English vocabulary was negatively related to
Chinese character reading. Age was positively correlated with Chinese character reading,
vocabulary and comprehension.
For the long-term immigrants, English RC was related to English decoding and
vocabulary. Likewise, Chinese RC was related to Chinese character reading and vocabulary.
67
English RC was not related to Chinese RC. Instead, English comprehension was negatively
related to Chinese character reading and vocabulary. Additionally, English vocabulary was
negatively related to Chinese character reading, vocabulary and comprehension. Age was
negatively related to English vocabulary and comprehension, and positively related to Chinese
character reading, vocabulary and RC.
Hierarchical Linear Regressions and Commonality Analyses
Tables 12 and 13 present hierarchical regressions which were used to identify within- and
cross-language predictors of RC in English (Table 12) and Chinese (Table 13) for the three
groups of bilinguals. For each regression, within-language relationships of word-level reading
and vocabulary were entered before cross-language predictors.4 For example, when English RC
acted as the outcome variable, English pseudoword reading and vocabulary were entered into the
first and second steps, respectively. Chinese character reading, vocabulary and RC were entered
in the third, fourth and fifth steps, respectively. Table 12 displays unstandarized coefficients,
standard errors, and standardized coefficients when English RC acted as the outcome variable.
For the Chinese EFL learners, 41% of the variance was explained by the regression model.
English vocabulary was a significant within-language predictor and Chinese vocabulary and RC
were unique cross-language predictors. For the recent immigrants, 73% of the variance was
explained by the regression model. English decoding and vocabulary were significant within-
language predictors and Chinese RC was a significant cross-language predictor. For the long-
4 Initially, age was added as a control variable in the first step within the regressions. However,
it was always a non-significant predictor across the three groups and did not influence any other
predictor-outcome relationships. Therefore, it was removed from the subsequent analyses to
simplify the regressions and commonality analyses.
68
term bilingual immigrants, 71% of the variance was explained by the regression model. Only
English vocabulary knowledge was a significant predictor of English RC.
Table 13 reports unstandarized coefficients, standard errors, and standardized coefficients
when Chinese RC acted as the outcome variable. For the EFL and recent immigrant groups,
43% and 77% of the variance was respectively explained by the regression models. Also for
both groups, Chinese vocabulary was a significant within-language predictor, and English RC
also positively predicted Chinese RC. On the other hand, English vocabulary was a negative
predictor of Chinese RC for both groups. Considering there was no significant zero-order
correlation between English vocabulary and Chinese RC, the relationships reported in the
regression need further clarification as variable suppression may be occurring (Zientek &
Thompson, 2010). For the long-term immigrant group, 77% of the variance was explained by
the regression models. Chinese vocabulary was the only significant predictor of Chinese RC.
To gain a better understanding of the cross-language relationships, commonality analyses
were conducted separately for the three groups of bilinguals to clarify aspects of shared and
unique variance. Examining both beta weights and structural coefficients – where structural
coefficients are Pearson correlation coefficients between a predictor variable and the predicted
outcome score (i.e., Ŷ) - is a more useful data analytic technique than just examining one or the
other (Mood, 1971; Newton & Spurrell, 1967; Pedhazur, 1997; Nimon & Reio, 2011; Zientek &
Thompson, 2010). The commonality analyses contained the same variables as the regression
models described previously.
Table 14 displays beta weights, structural coefficients, total variance, common variance
and unique variance for the within- and cross-language predictors of English RC. For the
Chinese EFL learners, the beta weights were largest for English vocabulary, Chinese vocabulary
69
and Chinese RC, and structural coefficients were moderate-to-large for all within- and cross-
language variables. For the significant cross-language predictors, variance was mostly shared
with other predictors. For the recent immigrants, beta weights were largest for English
vocabulary and Chinese RC, and moderate for English decoding and Chinese vocabulary.
However, the beta weight for Chinese vocabulary and English RC was negative. Given that the
structural coefficient was near-zero, Chinese vocabulary appeared to act as a suppressor variable
when predicting English RC for the recent immigrants. For the long-term immigrants, English
vocabulary knowledge shared a near perfect relationship with English RC.
Table 15 displays the variance coefficient (proportion of variance explained) and the
percentage of explained variance for each predictor and all combinations of predictors when
English RC acted as the dependent variable. I was most interested in the unique aspect of the
cross-language predictors and the combinations of common variance components among the
independent variables that included the cross-language predictors. Therefore, I focused on cross-
language vocabulary and RC because these relationships varied across groups.
For the Chinese EFL learners, Chinese vocabulary and RC uniquely explained 8% and
14% of the variance in English RC. Additionally, approximately 13% of the common variance
was shared with Chinese vocabulary, 16% with Chinese comprehension, and 40% with the
combination of Chinese vocabulary and Chinese RC and the within-language predictors. For the
recent immigrants, Chinese vocabulary and RC accounted for approximately 2% and 14% of the
variance in English RC respectively. Twenty-four percent of the common variance was shared
with Chinese vocabulary and 27% with Chinese comprehension. Interestingly, the combination
of shared variance accounted by Chinese vocabulary and Chinese RC was -26%. Considering
that Chinese vocabulary was acting as a suppressor variable and explained negative variance
70
when combined with Chinese RC, it seemed that Chinese vocabulary suppressed the relationship
between Chinese and English RC for the recent immigrants. Further analyses revealed that not
including Chinese vocabulary in the regression predicting English comprehension decreased the
strength of the association between English and Chinese RC, however this cross-language
relationship was still significant5. For the long-term immigrants, Chinese vocabulary and RC
explained virtually no unique variance. The amount of common variance among the cross-
language predictors was also low at 19% for vocabulary, 7% for comprehension, and -4% for the
combination of L1 vocabulary and RC.
Table 16 reports beta weights, structural coefficients, total variance, common variance
and unique variance for the within- and cross-language predictors of Chinese RC. For the
Chinese EFL learners, the beta weights were largest for Chinese vocabulary and English RC.
English vocabulary produced a negative beta weight with Chinese comprehension. Structural
coefficients were moderate-to-large for all within- and cross-language variables, except for a
weak negative coefficient for English vocabulary. For the significant cross-language predictors,
English vocabulary and English RC, variance was mostly shared with other predictors. For the
recent immigrants, beta weights were largest for Chinese vocabulary and English RC. Similar to
the Chinese EFL learners, English vocabulary produced a negative beta weight with Chinese RC.
Given that the structural coefficient was near-zero for both the Chinese EFL learners and recent
immigrants, English vocabulary acted as a suppressor variable when predicting Chinese RC for
both groups. Further analyses also revealed that not including English vocabulary in the
regression predicting Chinese comprehension underestimated the strength of the association
5Given that it is strongly suggested to retain suppressor variables in the regression analyses (e.g., Padey & Elliott, 2010; Thompson, 2006; Zientek & Thompson, 2010), omitting Chinese vocabulary when predicting English reading comprehension would lead to underestimating the relationship between reading comprehension across languages for the recent immigrants.
71
between English and Chinese RC, however this cross-language relationship was still significant
for both groups. For the recent immigrants, there was more unique than shared variance
explained by the cross-language predictors. Finally for the long-term immigrant group, Chinese
vocabulary shared the strongest relationships with Chinese comprehension.
Table 17 displays the variance coefficient (proportion of variance explained) and the
percentage of explained variance for each predictor and all combinations of predictors when
Chinese RC acted as the dependent variable. For the Chinese EFL learners, English vocabulary
and English RC uniquely explained 16% and 14% of the variance in Chinese RC, respectively.
The common variance shared with English vocabulary, and the combination of English
vocabulary and English RC was negative (-1% and -10% respectively), whereas a large amount
of common variance was shared with English comprehension (46%). There appeared to be
considerable overlap between L1 and L2 comprehension skills. Alternatively, English
vocabulary knowledge acted as a suppressor variable when predicting Chinese RC for the EFL
group. The suppression effect resulted in underestimation of the relationship between Chinese
and English RC if English vocabulary was not included in the model.
For the recent immigrants, English vocabulary and English RC explained approximately
2% and 11% of the unique variance in Chinese RC. Additionally, the common variance shared
with English vocabulary was approximately 17%, and 21% for English RC. Like the Chinese
EFL learners, the common variance for the combination of English vocabulary and English RC
was negative (-21%), suggesting that English vocabulary was acted as a suppressor variable for
the recent immigrant group. For the long-term immigrants, English vocabulary and RC
explained virtually no unique variance with Chinese RC. English vocabulary shared 35% of the
72
variance with other predictors, English comprehension shared 13% and the combination of
English vocabulary and comprehension was -12%.
73
Chapter 9: Discussion
The purpose of the current study was to identify how the within- and cross-language
contributions of word-level reading, vocabulary and RC in Chinese, the L1, and English, the L2,
varied as a function of the bilingual students’ language learning environment. Comparable
measures in the L1 and L2 were administered to three groups of bilinguals with distinctly
different experiences regarding exposure to and use of the L1 and L2. The Chinese EFL learners
experienced L2 learning within the confines of instruction only, where the L1 was the societal
language and was almost solely used to communicate with family and peers. The recent
immigrants received most of their education in China and moved to Canada during high school
or to receive an education at a Canadian university. Generally, these students were integrated
into an English education system during adolescence. The long-term immigrants moved to
Canada during elementary school or earlier and received most, if not all, of their education in an
English system. They came from families where the home language was Chinese.
Performance on word-level reading, vocabulary and RC across groups generally followed
trends that would be expected given the differences in the groups’ language learning
environments. The long-term immigrants were most proficient in their English skills and least
proficient in their Chinese skills. Also as expected, the Chinese EFL learners performed the
highest on all Chinese skills and scored the lowest on all English skills (with the exception of the
Chinese EFL learners and the recent immigrants having comparable performance on Chinese
RC). The recent immigrants’ performance was typically in-between the other two groups,
barring the exception noted above.
The regressions and commonality analyses produced similarities and differences among
predictors of English RC across groups. For the Chinese EFL learners, English RC was
74
positively predicted by English vocabulary, Chinese vocabulary and Chinese RC. When
predicting English RC for the recent immigrants, English decoding, English vocabulary and
Chinese RC were positive predictors. No other significant relationships were identified. Finally,
for the long-term immigrants, the only significant predictor of English RC was English
vocabulary.
The results for Chinese RC also produced similarities and differences across groups. For
the Chinese EFL learners, significant positive predictors of Chinese RC were Chinese
vocabulary and English RC. English vocabulary was a significant negative predictor of Chinese
RC. However, the supplemental commonality analyses revealed that English vocabulary was a
suppressor variable in the regression. For the recent immigrants, Chinese vocabulary and
English RC were positive predictors of Chinese RC. Similar to the Chinese EFL learners,
English vocabulary was a negative predictor of Chinese RC. The commonality analyses also
revealed a suppression effect for the recent immigrants. For both suppression results, not
including English vocabulary in the model underestimated the strength of the relationships
between Chinese and English RC. For the long-term immigrants, Chinese vocabulary was the
only significant predictor of Chinese RC.
The SVR (Gough & Tunmer, 1986; Hoover & Gough, 1990) is a useful framework for
understanding the within- and cross-language predictors of RC in adolescent bilinguals coming
from different backgrounds. At the same time, the results suggest that models of RC are
distinctly different across groups of Chinese-English bilinguals in diverse language learning
environments, and these groups of bilinguals utilized their L1 and L2 skills differently to
comprehend text. The following sections will outline the within- and cross-language
relationships of word-level reading, vocabulary and RC for the three groups of bilinguals.
75
The Role of Word-level Reading Within and Across Languages
English word-level reading was only a significant within-language predictor of English
RC for the recent immigrants. The results appear to suggest that recent immigrants were still
mastering decoding and individual differences were predictive of RC performance. Word-level
reading was not a unique predictor of RC in English or Chinese for the Chinese EFL learners or
long-term immigrants. The lack of relationships between English decoding and English RC may
be interpreted differently for the two groups. For the Chinese EFL learners, English education
often does not include phonics instruction (Keung & Ho, 2009). Therefore, the Chinese EFL
learners may be relying on a whole-word strategy during the RC task, which would result in
decoding skills being less related to English RC. On the other hand, the long-term immigrants
may have mastered decoding to a point where it was no longer predictive of RC performance.
Chinese word-level reading was a significant within-language predictor of Chinese RC
for the recent immigrants. However, word-level reading was not a predictor of Chinese RC for
the Chinese EFL learners or the long-term immigrants. The significant relationship for the
recent immigrants and the lack of relationships for the Chinese EFL learners and the long-term
immigrants may be understood by considering the scores obtained on the Chinese word-reading
task. For the recent immigrants there was substantial variability in the measure and no evidence
of floor or ceiling effects. On the other hand, the Chinese EFL learners did very well and there
was hardly any variability in performance which would have prevented the measure from
significantly predicting RC. Alternatively, the long-term immigrants did poorly, and there was
large variability in performance on the Chinese word reading task. Therefore, a floor effect may
have prevented the measure from significantly predicting RC for the long-term immigrants.
76
Consistent with other research regarding cross-language relationships of word-level
reading skills in Chinese and English (e.g. Gottardo et al., 2001), there were no instances of
cross-language relationships within the regression results. Considering there is a large degree of
linguistic distance between reading Chinese logographic characters and reading English words, it
is not surprising that there is no relationship across the L1 and L2. However, examination of
structural coefficients across groups revealed different patterns. Moderate to strong structural
coefficients between word-reading skills and comprehension across languages were evident for
the Chinese EFL learners. Specifically, a strong structural coefficient (.63) was noted for
Chinese word-level reading and English RC, while the structural coefficient between English
word-level reading (a decoding task) and Chinese comprehension was moderate (.36). At the
same time, near zero (or negative) coefficients were evident for the recent and long-term
immigrants. Perhaps the Chinese EFL learners are utilizing more familiar skills, such as a
reliance on a whole word activation strategy, for reading both Chinese and English. While this
strategy is useful for Chinese word-level reading, it is not as helpful when reading English
pseudowords. This interpretation may help clarify the Chinese EFL learners’ poorer
performance on the English word-level reading task in comparison to the other groups, and the
moderate relationship observed between English word-level reading and Chinese RC. This
interpretation is tentative and requires further investigation. Most importantly, a trend emerged
showing that as integration into an L2 system occur word-level reading skills in Chinese and
English become less related. Perhaps, as integration into a L2 system increases, L1 and L2
reading separate and become modularized and more language specific.
77
The Role of Vocabulary Within and Across Languages
Vocabulary was a within-language predictor of RC across groups. Within-language
semantic knowledge was critically important for L1 and L2 comprehension across diverse groups
of language learners. Vocabulary appears to be a stable predictor of comprehension and
consistently important in the L1 and L2 throughout the process of integration into a L2 society.
However, cross-language associations of vocabulary knowledge were far less consistent.
Patterns of cross-language associations differed when either the L1 or L2 acted as the outcome
variable, and also differed across groups.
Chinese vocabulary and English RC were positively related for the EFL group, which
replicated and extended the work of Li et al. (2012). The positive cross-language association in
the current study suggested that the Chinese-English EFL learners were able to utilize semantic
knowledge in their L1 to help comprehend English. The Chinese EFL learners had poor
performance in English vocabulary and superior performance in Chinese vocabulary when
compared to the other Chinese-English bilingual groups. L1 semantic skills appear to be
important in L2 reading during early stages of L2 acquisition. Considering that Chinese and
English do not have cognates, it may be semantic knowledge in the L1 that is related to reading
in the L2. Students who have well-developed semantic knowledge in the L1 may be more
proficient at producing L2 to L1 translations when reading the L2 than students with less
developed L1 semantic knowledge. Additionally, students who have well developed semantic
knowledge in the L1 may also have an easier time understanding a similar concept encountered
in the L2. However, given that only the current study and the Li et al. (2012) study have
reported positive associations between Chinese vocabulary and English RC, further investigation
is necessary to understand the nature of these cross-language associations. No positive
78
associations between L1 vocabulary and L2 RC were found for the other groups. It appears that
L1 semantic knowledge facilitates L2 comprehension only for Chinese EFL learners. Perhaps
the language learning environment may have influenced the cross-language associations. It may
be that English instruction in China emphasizes translations and using well-developed L1
knowledge to build L2 vocabulary and reading skills, while in Canada there may be less reliance
and instruction on using translation knowledge to build L2 reading and language skills.
Within this study, positive associations between L1 vocabulary and L2 comprehension
and negative associations between L2 vocabulary and L1 comprehension were demonstrated.
While the differences in these relationships confirm that asymmetrical relationships between
vocabulary and RC across languages may exist, the negative association is interpreted as
statistical suppression. The suppression effect of English vocabulary underestimated the strength
of the association between Chinese and English comprehension. Since within-language
vocabulary skills are an important predictor of within-language comprehension skills it appears
to be necessary to consider these two constructs together when examining cross-language
relationships.
The Role of RC Across Languages
The current study reported bidirectional associations of RC across Chinese and English
for the Chinese EFL learners and recent immigrants. Cross-language associations, however,
were not found among the long-term immigrants, who had immigrated to Canada at a young age.
The results of the current study showed cross-language relationships of comprehension for
students in the early stages of learning English who were proficient in their L1. These results
replicate and extended the work of Li et al. (2012), who reported bidirectional cross-language
relationships between L1 and L2 RC in younger Chinese-English bilinguals.
79
The bi-directional transfer of comprehension skills for the Chinese EFL learners and the
recent immigrants indicated that L1 and L2 comprehension was closely related for learners who
were developing proficiency in their L2. L1 RC predicted unique variance for the Chinese EFL
learners and recent immigrants but not for the long-term immigrants. Likewise, shared variance
among English and L1 predictors was substantial for all groups except the long-term immigrants.
However, as residency in an L2 educational system and culture increased, comprehension
appeared to become more language specific. The finding that L1 and L2 comprehension for the
long-term immigrants was only predicted by within-language vocabulary skills supported the
idea that L1 and L2 RC became a language specific skill. Additionally, proportions of unique
and common variance explained by cross-language RC were virtually non-existent for the long-
term immigrants. These results suggest that as exposure in an L2 environment increases L1 and
L2 skills become less related.
What is the nature of the cross-language relationships between L1 and L2 comprehension
for the Chinese EFL learners and recent bilingual immigrants? Van Gelderen et al. (2007)
proposed that the locus of transfer is most likely metacognitive in nature, when explaining the
relationship between L1 and L2 comprehension skills of Dutch-English bilinguals. Higher-level
comprehension skills such as strategy use and the ability to construct meaning from text are
skills that may transfer and explain why comprehension is related in the L1 and L2. It is highly
plausible that comprehension strategies also transfer between Chinese and English, and this
transfer explains the results in the current study. Bilingual students with relatively low L2
proficiency may rely heavily upon metacognitive skills to comprehend an unfamiliar language.
However, as L2 proficiencies develop, within-language skills become much better predictors of
L2 performance. Chinese EFL learners, recent, and long-term immigrants may adopt different
80
strategies and utilize different skills to read the L1 and L2. Future studies should examine the
use of comprehension strategies in Chinese-English bilinguals to examine whether a strategy
learned in one language can be used during reading in another language. Additionally,
examining how L1 and L2 proficiency interacts with transfer of strategies across languages
would improve our understanding regarding the nature of the relationships between RC in the L1
and L2.
Conditions that Support Cross-language Relationships
Patterns of cross-language relationships were substantially different across groups of
Chinese-English bilinguals from different language-learning environments. The patterns of
cross-language relationships that emerged in the current study demonstrates that transfer is more
likely to occur for students who are proficient in their L1 and in the early stages of learning an
L2. Even though there is considerable linguistic distance (Koda, 2007) between the L1 and L2,
cross-language relationships were still evident for students in the relatively early stages of L2
acquisition. Furthermore, cross-language relationships varied as a function of the skill under
examination. Strongest patterns of cross-language relationships were found for RC while no
cross-language relationships were found for word-level reading. These patterns suggest that
higher-level skills may transfer more easily than lower-level reading skills in cases where there is
a large degree of linguistic distance between L1 and L2 orthography.
Limitations and Future Directions
A limitation of the current study may be the specific cut-point of length of residence used
to help determine the groups. It is possible that the immigrant groups could have been divided
differently. Using a latent class analysis to sub-divide the immigrant participants would have
been helpful to inform specific cut-points, however the modest sample size of the current study
81
prevented this from being an option. Also, the age differences between the groups and the
unbalanced proportions of males and females within each group is another limitation of the study.
Future research should attempt to compare groups of bilinguals who are more similar in age.
Additionally, having a balanced proportion of males and females would allow us to rule out
differences due to gender in terms of performance or predictors of RC. An additional limitation
is the use of experimental translation tasks to assess L1 vocabulary and RC. While great care
was taken to ensure the translations were as accurate as possible, the authenticity of these
measures is not validated. Future research should attempt to develop and standardize Chinese
vocabulary and RC measures that are appropriate for adolescents with a wide range of language
and reading proficiency in Chinese. Furthermore, the Chinese word-level reading task had a
floor effect for the long-term immigrants and did not capture enough variance for the Chinese
EFL learners. Further development of this task should include higher frequency words when
used with long-term immigrants or students with low levels of proficiency. Alternatively,
additional lower frequency words should be included for students with high levels of Chinese
proficiency. Another suggestion is to add measures of word-reading fluency, as they may
capture more variance and help prevent floor effects in low-proficiency speakers. A future
direction is to include other groups of bilinguals who spoke different L1s. Including bilinguals
with an alphabetic L1 who also varied in terms of differences in their language-learning
environments would have improved the generalizability of the current study.
Given that there are stark differences between patterns of within- and cross-language
contributions to literacy skills in different learning environments, the next step is to investigate
the ways in which these different environments influence models of RC and cross-language
associations. For example: how do acculturation, engagement, and language use interact on an
82
individual level with within- and cross-language predictors of reading performance, and more
importantly reading development? Understanding individual differences in socio-cultural and
motivational factors and how they interact with cognitive and linguistic models of RC is a
natural extension of the current findings presented within this study. A longitudinal design for
the research would increase the clarity of L1 and L2 language and literacy development across
different cultural contexts and conditions of language use and exposure.
Implications
The results from the current study have theoretical and practical importance. Instances of
cross-language associations appear to be influenced by the language-learning environment and
the context for learning an L2. Failure to acknowledge these differences when designing
empirical studies will lead to difficulties improving our understanding of L2 acquisition.
Additionally, these differences should not be neglected when trying to provide the best
educational opportunities for bilinguals. Educators should be mindful of these differences and
structure educational practices accordingly. Different strategies to improve L2 (or even L1)
comprehension can occur for students with different immigration experiences. For example,
students who are learning English in a non-English speaking environment can be taught how to
best utilize existing vocabulary to increase L2 vocabulary and comprehension skills.
Additionally, students in the early stages of L2 acquisition can continue to develop higher-order
comprehension strategies in their L1, while learning L2 decoding skills and vocabulary.
Alternatively, instruction should focus primarily on teaching within-language vocabulary skills
to improve L1 and L2 comprehension for long-term immigrants. Another implication is that
assessment of RC in the L1 can inform future potential in L2 proficiency and development, or
reveal the possibility of difficulties in developing L2 comprehension. However, the results also
83
suggest that assessing L1 proficiency is most useful for students who intend to immigrate or
recently immigrated to an English speaking society.
Additionally, research must increase understanding of the integration between L1 and L2
semantic knowledge as a method to increase L2 vocabulary knowledge. L2 vocabulary
acquisition is a formidable challenge in second language acquisition because it is difficult and
time consuming to improve (e.g., Carlo et al., 2004; Farnia & Geva, 2011). While only the EFL
group could capitalize on L1 vocabulary to improve L2 comprehension, this result does point to
the possibility of students accessing their existing semantic knowledge to increase word learning
experiences in the L2 in the initial stages of second language learning. Across the groups,
comprehension appears to become a language specific skill as time spent in a L2 environment
increases. The results suggest that we can initially support L2 acquisition with L1 skills but
intensive and systematic instruction in the L2 is necessary for students to achieve high levels of
proficiency in a new language.
84
Chapter 10: General Conclusions
What emerged from Study 1 was a more refined understanding of the constituent
components necessary for RC in adolescent bilinguals, and the degree to which these
components were consistent predictors across the L1 and L2 for Spanish-English and Chinese-
English bilinguals. Vocabulary and morphological awareness emerged as consistently important
predictors of comprehension in English, Spanish and Chinese. Specifically, the results suggest a
central role of word knowledge, in terms of semantics and morphological awareness, as a critical
and universal aspect of comprehension across languages (Perfetti, 2007; Perfetti & Hart, 2001,
2002). Fostering word knowledge is essential to achieve reading proficiency in alphabetic and
logographic languages alike. Additionally, English decoding was only a significant predictor of
English RC for the Chinese-English, not the Spanish-English, bilinguals. Overall, there was a
large degree of universality among the within-language predictors of RC across Spanish, Chinese
and English. The common predictors across languages (e.g., vocabulary) were much stronger
predictors of RC performance than the language specific predictors.
While strong instances of universal components of RC were found across two distinctly
different groups of bilinguals, this should not lead to the suggestion that all bilinguals are alike.
In addition to the universal aspects of comprehension, there were also multiple language specific
findings. Understanding how the two work together, for different groups of bilinguals, is
necessary to develop a comprehensive understanding of how people comprehend text in different
languages. Additionally, these studies outline what skills should be the focus of evidence-based
instructional programs for students from different linguistic and cultural backgrounds.
Study 2 identified the degree to which within- and cross-language predictors varied as a
function of changes in the Chinese-English bilinguals’ language learning environment. In the
85
early stages of L2 acquisition, comprehension was highly related across languages, and skills in
the L1 were related to comprehension in the L2. As bilinguals became more integrated into a L2
educational system and culture, reading across languages was less related and became reliant
upon within-language skills. The interconnections between reading skills in the L1 and L2 are
sensitive to educational experiences and the language learning environment, and are a dynamic
process that is redefined as skills and knowledge within each language are developed or
forgotten.
Predictors of comprehension also appear to vary as a function language-learning
experiences. This observation suggests that we should consider the socio-cultural context of
language learning as an important methodological factor that influences how empirical studies
are constructed, how results are interpreted, and how groups of bilinguals are defined. Even
more important are the accompanying educational implications. The research presented in these
two studies strongly suggests that all bilinguals would benefit from common instructional
practices, such as enhancing vocabulary and word study instruction. However, a one-size-fits-all
approach would be a gross overgeneralization of these findings. Not only do bilinguals with
different language backgrounds require a different balance of instruction, but instructional
practices should vary based on their prior educational experiences. Understanding how and
when to capitalize on L1 language and literacy skills and how L2 instruction should be promoted
at different stages of L2 learning is critical. This will ensure we are engaged in the best practices
to reach the societal and moral obligations to ensure all bilingual students are equipped with the
language and literacy skills needed to succeed.
86
References
Anglin, J. M. (1993). Vocabulary development: A morphological analysis. Monographs of the
Society for Research in Child Development, 238, 58.
August, D., Carlo, M., Dressler, C., & Snow, C. (2005). The critical role of vocabulary
development for English language learners. Learning Disabilities Research & Practice,
20, 50–57.
August, D., & Shanahan, T., (2006). Developing Literacy in Second-Language Learners: A
Report of the National Literacy Panel on Language-Minority Children and Youth. NJ:
Erlbaum.
Braze, D., Tabor, W., Shankweiler, D. P., & Mencl, W. E. (2007). Speaking up for vocabulary:
Reading skills differences in young adults. Journal of Learning Disabilities, 40, 266-243.
Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen
and J. S. Long (Eds.) Testing Structural Equation Models. CA: Sage.
Brysbaert, M., & New, B. (2009). Moving beyond Kučera and Francis: A critical evaluation of
current word frequency norms and the introduction of a new and improved word
frequency measure for American English. Behavior Research Methods, 41, 977-990.
Cai, Q., & Brysbaert, M. (2010). SUBTLEX-CH: Chinese word and character frequencies based
on film subtitles. PLoS ONE 5(6) e10729.
Cain, K. (2007). Syntactic awareness and reading ability: Is there any evidence for a special
Table 2. Intercorrelations among English variables for Chinese-English bilinguals (above the diagonal) and Spanish-English bilinguals (below the diagonal)
Table 7. Comparison of fit statistics by constraining parameter estimates to be equal across groups
Notes. df = degrees of freedom, CFI = Comparative Fit Index, RMSEA = Root Mean Square Error of Approximation, AIC = Akaike Information Criterion, BCC = Brown-Cudeck Criterion + p < .06, * p < .05, ** p < .01, *** p < .001
Figure 2. Structural models of English RC showing direct and indirect effects Spanish-English bilinguals (Model 1) Chinese-English bilinguals (Model 2)
Notes. Bolded paths refer to parameters estimates that are significantly different between groups; dotted paths denoted significant indirect effects on RC.
110
Table 8. Coefficients for structural equation model of first language RC
Figure 3. Structural model for Spanish RC (Model 3) showing direct and indirect effects
Notes. Dotted paths denoted significant indirect effects on RC
113
Figure 4. Structural model for Chinese RC (Model 4) showing direct and indirect effects
Notes. Dotted paths denoted significant indirect effects on RC
114
Table 10. Descriptive statistics and Analysis of Variance for Chinese EFL learners, Recent Immigrants, and Long-term Immigrants for age, months in Canada, English and Chinese variables.
Min. Max. Mean SD Min. Max. Mean SD Min. Max. Mean SD F (2,171)
Partial η2
Pairwise Comparisonsa
Age in Months 209 270 242.42 7.88 170 286 230.62 25.92 177 286 208.35 22.58 40.75*** .19 F > R > L Length of Residence 0 0 0 0 6 72 40.63 18.73 84 204 126.59 37.45 410.93*** .83 L > R > F English Decoding 22 42 32.35 5.20 21 44 34.23 4.86 19 43 36.84 4.96 11.57*** .12 L > R > F English Vocabulary 3 32 18.00 6.16 8 45 22.32 8.20 7 42 31.42 9.05 42.91*** .33 L > R > F English RC 16 38 26.51 5.92 14 48 30.70 7.99 11 46 35.30 8.20 19.84*** .19 L > R > F Chinese Word Reading 209 238 222.18 6.48 0 214 187.57 50.69 30 288 98.35 73.28 87.69*** .51 F > R > L Chinese Vocabulary 54 77 69.40 5.42 7 54 48.43 8.69 11 55 30.74 16.58 169.97*** .67 F > R > L Chinese RC 25 44 36.39 4.59 1 41 34.67 8.18 10 45 23.00 11.14 43.02*** .34 (F = R) > L
Notes. Min = Minimum, Max. = Maximum, SD = Standard Deviation * p < .05, ** p < .01, *** p < .001 a Equal sign indicates nonsignificant difference, and greater-than sign indicates p < .05
115
Table 11. Intercorrelations among Chinese and English variables separated by group.
* p < .05, ** p < .01, *** p < .001
1 2 3 4 5 6 7
Chinese EFL learners 1. Age in months 1 .-11 .05 -.27 -.23 -.19 -.07 2. English Decoding 1 .24 .36** .32* .19 .24 3. English Vocabulary 1 .30* .25 .17 -.08 4. English RC 1 .41** .51*** .48*** 5. Chinese Word Reading 1 .71*** .39** 6. Chinese Vocabulary 1 .55*** 7. Chinese RC 1
Recent Immigrants 1. Age in months 1 -.03 -.03 .02 .35** .27* .26* 2. English Decoding 1 .41*** .53*** -.23 -.23 -.01 3. English Vocabulary 1 .73*** -.30* -.21 -.03 4. English RC 1 -.08 .01 .33** 5. Chinese Word Reading 1 .80*** .65*** 6. Chinese Vocabulary 1 .80*** 7. Chinese RC 1
Long-term Immigrants 1. Age in months 1 -.25 -.43*** -.28* .43*** .42*** .35** 2. English Decoding 1 .56*** .53*** -.25 -.23 -.07 3. English Vocabulary 1 .81*** -.60*** -.59*** -.44*** 4. English RC 1 -.34* -.33* -.17 5. Chinese Word Reading 1 .93*** .83*** 6. Chinese Vocabulary 1 .86*** 7. Chinese RC 1
116
Table 12.
Hierarchical linear regression predicting English RC, separate for EFL, Recent Immigrant, and Long-term Immigrant groups
Chinese EFL learners Recent Immigrants Long-term Immigrants
Step Variable Final B SE (B) β Final B SE (B) β Final B SE (B) β 1 English Decoding .186 .129 .19 .394 .131 .24** .084 .156 .05 2 English Vocabulary .268 .115 .24* .568 .078 .58*** .822 .103 .91*** 3 Chinese Word Reading -.082 .111 -.04 -.003 .019 -.02 .010 .023 .09 4 Chinese Vocabulary .347 .148 .28* -.229 .142 -.25 -.003 .111 -.01 5 Chinese RC .400 .171 .30* .553 .122 .57*** .123 .115 .17
Notes. B = Unstandardized Beta, SE = Standard Error, β = Standardized Beta * p < .05, ** p < .01, *** p < .001
117
Table 13.
Hierarchical linear regression predicting Chinese RC, separate for EFL, Recent Immigrant, and Long-term Immigrant groups
Chinese EFL learners Recent Immigrants Long-term Immigrants
Step Variable Final B SE (B) β Final B SE (B) β Final B SE (B) β 1 Chinese Word Reading .068 .086 .01 .011 .018 .07 .042 .027 .28 2 Chinese Vocabulary .292 .115 .41* .659 .105 .70*** .425 .120 .63*** 3 English Decoding .118 .102 .11 -.022 .134 -.01 .245 .185 .11 4 English Vocabulary -.269 .087 -.28* -.206 .100 -.21* -.080 .185 -.07 5 English RC .243 .104 .30* .499 .110 .49*** .178 .167 .13
* p < .05, ** p < .01, *** p < .001
118
Table 14. Commonality analyses beta weights, structural coefficients, proportions of total, common and unique variance for EFL, Recent Immigrant and Long-term Immigrant groups when predicting English RC
Unique to English Decoding .030 7.20 .044 6.01 .002 0.24 Unique to English Vocabulary .050 11.89 .259 35.12 .367 51.75 Unique to Chinese Word Reading .001 0.14 .000 0.02 .001 0.14 Unique to Chinese Vocabulary .034 8.06 .013 1.71 .000 0.00 Unique to Chinese RC .059 14.16 .010 13.54 .007 0.93 Common with Chinese Vocabulary (without Chinese RC) .054 12.81 .176 23.83 .134 18.90
Common with Chinese RC (without Chinese Vocabulary) .067 15.87 .200 27.19 .048 6.70
Common with Chinese Vocabulary and Chinese RC .171 40.71 -.188 -25.53 -.025 -3.53
120
Table 16. Commonality analyses beta weights, structural coefficients, proportions of total, common and unique variance for EFL, Recent Immigrant and Long-term Immigrant groups when predicting Chinese RC
Unique to Chinese Word Reading .000 0.03 .001 0.18 .011 1.37 Unique to Chinese Vocabulary .073 17.12 .165 21.38 .056 7.31 Unique to English Decoding .010 2.31 .000 0.02 .008 1.03 Unique to English Vocabulary .066 15.58 .018 2.27 .001 0.11 Unique to English RC .059 13.84 .086 11.14 .005 0.67 Common with English Vocabulary (without English RC) -.005 -1.22 .134 17.30 .271 35.08
Common with English RC (without English Vocabulary) .196 45.99 .166 21.48 .102 13.20
Common with English Vocabulary and English RC -.043 -10.23 -.165 -21.37 -.090 -11.60