How lexical quality predicts L2 reading comprehension in early bilingual education

ABSTRACT

This study investigated the impact of second language (L2) lexical quality on L2 reading comprehension in bilingual primary education. The participating children were taught in the L2 English for 30-50% of the time and the remaining time in their L1 Dutch. Seventy-three pupils from four different schools were assessed in the L2 on vocabulary, lexical specificity and phonological awareness in kindergarten, vocabulary, lexical specificity and decoding in grade 1, and reading comprehension in grade 3. L1 reading comprehension was tested in grade 2. Results showed that lexical specificity in kindergarten had a direct effect on L2 reading comprehension in grade 3, and phonological awareness in kindergarten had an indirect effect on L2 reading comprehension via L2 decoding skills in grade 1. These effects remained significant after taking L1 reading comprehension skills in grade 2 into account. The lexical specificity task, measuring specificity of phonological representations, had a more profound impact on reading comprehension than vocabulary. This highlights the importance of examining the specificity of phonological representations when it comes to studying L2 reading comprehension proficiency and implies that teachers should focus on L2 lexical specificity in bilingual kindergarten to build the foundation for L2 reading comprehension in grade 3.

KEYWORDS:

• Reading comprehension
• lexical quality
• lexical specificity
• vocabulary
• bilingual education

Acquiring reading comprehension skills is a fundamental aspect of primary education, with the lexicon playing a pivotal role in this process (Perfetti and Helder 2022). The relation between lexical quality and reading comprehension is exemplified in the lexical quality hypothesis by Perfetti and Hart (2002). This hypothesis claims that the more phonologically, semantically, and orthographically specific words are embedded in one's mental repository, the smoother their retrieval from memory becomes, consequently making it easier to use them in the reading comprehension process. Lexical quality directly relates to word identification, being able to rapidly read a word and access its meaning, which is highly important for comprehension (see Perfetti and Stafura 2014). Young children have a mental lexicon that is rapidly growing and evolving (Metsala and Walley 1998), which requires representations to become more specific to be able to discriminate between them. Indeed, lexical specificity (i.e. the ability to learn phonologically similar words and build specific representations) in children in L2 immersion education has been shown to foster phonological awareness, which in turn facilitated word decoding in the L1 (Krenca et al. 2020). The relevance of lexical specificity for L2 reading comprehension has yet to be uncovered. Therefore, in the current study, we examined the role of lexical specificity, next to well-known predictors (vocabulary, phonological awareness and decoding skills) in reading comprehension development in the L2 within the context of bilingual primary education.

Lexical quality and reading comprehension in the L1

To understand written text, words have to be identified and integrated into the text. In order to identify a word, the reader needs to address its orthographic, phonological and semantic units from memory (Perfetti and Stafura 2014). These components of the word are represented in the reader’s memory: the tighter the connection between these representations (i.e. lexical specificity), the easier they are to retrieve and use (Perfetti and Hart 2002).

Becoming a fluent word decoder in which decoding is an automized skill, is crucial for developing reading comprehension (Verhoeven and Van Leeuwe 2008). At first, decoding is relatively simple with words that children verbally use already, especially in those early years of primary education. Beginning decoding, therefore, relies heavily on phonological awareness and vocabulary in that phase. The relation between phonological awareness (the ability to recognize and manipulate spoken parts of sentences and words) and decoding has been established in previous research (Melby-Lervåg, Lyster, and Hulme 2012). Its importance as a predictor of decoding overrules that of vocabulary in the early stages of learning to read (see e.g. Wagner et al. 1997).

Within the context of the lexical quality hypothesis, children’s vocabulary is strongly associated with their reading comprehension skills (Perfetti 2007). Research has shown that vocabulary, knowing the meaning of words, predicts a child’s reading comprehension abilities, in particular in inference making while reading (Cain and Oakhill 2014; Ouellette 2006). Results of a study in young children (4–8-year-olds) showed that vocabulary in preschool predicted decoding skills, and that both vocabulary and decoding skills independently predicted reading comprehension scores in grade 2 (Kendeou et al. 2009).

The link between phonological awareness and vocabulary as precursors of reading comprehension might be lexical specificity: the ability to learn new detailed phonological representations of phonetically similar words (Janssen et al. 2015; Van Goch, McQueen, and Verhoeven 2014). Lexical specificity differs from phonological awareness in the sense that it requires the ability to detect minimal differences in words without immediately actively manipulating them. Children who are better at detecting phonemes and differentiating and manipulating them, as required in most phonological awareness tasks, are quicker at learning how to decode words from print, even after controlling for intelligence, memory and vocabulary knowledge (Elbro 1996).

The degree of lexical specificity changes over time and especially rapidly during the first couple of years of a child’s life. For example: a young child first learns the difference between words like ‘apple’ and ‘pear’, which differ quite a bit, but later needs to be able to understand the difference between ‘pear’ and ‘bear’, words that are phonologically more similar. Although phonological representations have been found to be highly specific at toddler age already (White and Morgan 2008), they increase in specificity as vocabulary grows. It could, however, also be argued that it works both ways and that vocabulary also grows due to better lexical specificity. When a lexical representation is specific, its storage in memory is done with more detail and its retrieval from memory becomes more reliable (Perfetti 2007).

Lexical quality and reading comprehension in the L2

From the moment someone can understand what they read, vocabulary does not only have to be verbally presented, but is also developed through written presentations. Developing reading comprehension in an L2 builds on the same foundational skills as the ability in the L1 (Wealer et al. 2022). This suggests that the lexical quality hypothesis also applies to L2 reading comprehension and that L2 lexical specificity, besides vocabulary, phonological awareness and decoding skills, is an important precursor of reading comprehension in the L2.

Studies have shown that children who learn to read in a second language show weaker vocabulary knowledge and may therefore longer rely on code-related skills and phonological awareness (Jongejan, Verhoeven, and Siegel 2007). Lervåg and Aukrust (2010) found that differences in vocabulary levels explained variability in initial reading comprehension: L1 learners had better vocabulary and therefore higher reading comprehension scores than L2 learners. No differences were observed between both groups on decoding, however, and only when decoding skills were attained at a certain level did vocabulary explain the different reading comprehension levels. In a study in which L1 English and L2 French pupils in French immersion education in Canada were assessed on lexical specificity in regard to decoding, results showed that L1 lexical specificity predicted L1 decoding, mediated by L1 phonological awareness. This was not found for the L2. Word reading in the L2, however, was predicted by L1 word reading, mediated by L1 phonological awareness (Krenca et al. 2020).

Lexical specificity has not yet been examined in relation to L2 reading comprehension and only very few studies have investigated this within the context of bilingual primary education, in which children experience dual immersion of an L1 and an L2 in which the L2 is not only a foreign language that is not spoken at home or in society, but also first introduced to them at the age of 4 (later than most bilinguals who learn two languages from birth, but earlier than most children who learn a second language in school).

Current study

It is yet unclear what role L2 lexical quality plays in the early years of bilingual primary education. To fill this gap, this study aimed to investigate the impact of L2 lexical quality on L2 reading comprehension in bilingual primary education, while controlling for L1 reading comprehension (Figure 1). The research question asked was: how do L2 (receptive) vocabulary, L2 lexical specificity, L2 phonological awareness and L2 decoding skills in kindergarten and grade 1 impact L2 reading comprehension skills in grade 3, while taking into account L1 reading comprehension skills in grade 2? In the longitudinal design children were assessed at four different time points: kindergarten (see also Kwakkel et al. 2021), grade 1, grade 2 and grade 3. In line with the lexical quality hypothesis we expected that both the L2 lexical skills (receptive vocabulary and lexical specificity) as well as the L2 phonological and decoding skills of the children in kindergarten and grade 1, would predict L2 reading comprehension outcomes in grade 3. Although an effect of L1 reading comprehension was expected, we hypothesized that the L2 lexical quality predictors would keep their predictive value even when L1 reading comprehension was taken into account.

Figure 1. Conceptual model of the current study.

A diagram showing the variables and arrows illustrating their potential relations from kindergarten, via grade 1 and grade 2 to grade 3.

In the Netherlands, Dutch is the main national language. English is spoken by many people and widely available through television and social media, but is not an official language. English is an obligatory subject in schools for all children from grade 5 of primary school onward. Some primary schools have early language programs, meaning that they teach English from a younger age (sometimes age 8 or 9, sometimes age 4 already), for 30 up to 240 min at the most per week. Since 2014, 17 schools in The Netherlands have participated in a pilot of bilingual primary education, in which children are taught in English 30-50% of the educational time (compared to the maximum of 15% mentioned previously for early foreign language programs). These schools have built a bilingual curriculum (dual immersion education), teach through Content & Language Integrated Learning (CLIL) (Dalton-Puffer 2011) and focus specifically on global citizenship and internationalization (Nuffic n.d.). Children in bilingual primary education have much more exposure the L2 English than children in regular Dutch education, as well as a much younger age at which they first start learning the L2. The present study is conducted within the context of Dutch bilingual primary education. This specific type of bilingual education is rather new but will become more widely available to the community in the years to come. In North America, dual language programs are widely implemented but focus mainly on minority students and second language learning (Genesee and Lindholm-Leary 2013). The European Union states in its language policy that every European citizen should master two or more other languages in addition to their mother tongue (European Union 2017). The European implementation of bilingual education is therefore mainly based on the concept of ‘content and language integrated learning’ (CLIL) in which an additional language is used for the learning of content of different subjects and the additional language (Coyle, Hood, and Marsh 2010). There are many differences between bilingual programs worldwide: starting age, time of exposure to the languages, students’ backgrounds, pedagogical and didactical strategies, and choice of additional language(s) (Sánchez-Pérez and Manzano-Agugliaro 2021). Not only will this current research shed light on how two languages are developed alongside from a very young age (four years old) with around 50% exposure in both the L1 (majority language) as well as the L2 at school, it will also help teachers to find focus points and implement a dual curriculum effectively in primary education.

In the bilingual primary educational setting where this study takes place, English is mainly taught orally in the first 3–4 years of school. Children do not receive formal English decoding instruction but learn to read in the L1 (Dutch).

Method

Participants

A total of 73 children, 34 boys and 39 girls, from four different schools in The Netherlands participated. The children had a mean age of 5 years and 10 months at the beginning of this study (in kindergarten) and a mean age of 8 years and 4 months at the end (early grade 3). All schools participated in the bilingual primary education pilot and taught 30-50% of the educational time in the L2 English and the remaining time in the L1 Dutch. Seven children spoke another language (not English) at home besides the L1 Dutch. The other 66 children came from monolingual Dutch families. Children who spoke English at home were excluded from this study.

Materials

Lexical quality

Vocabulary

L2 receptive vocabulary size was assessed using the Peabody Picture Vocabulary, a test that measured the form-meaning link in a receptive way. The task contains 17 sets of 12 items with increasing difficulty (PPVT-IV, Dunn and Dunn 2007). The standardized individual administration was carried out according to the manual. The child’s age determined the starting set. When the child answered more than one item in one set incorrectly, the previous set was administered until a set with a maximum of one error was reached (basal set). The ceiling set was found after continuing the test until a child reached eight or more errors in a set. Scores were calculated according to the manual by subtracting the total number of errors from the highest item the child had completed. This raw score was used for analyses in this study.

Lexical specificity

L2 lexical specificity was assessed by measuring how well children are able to learn phonologically similar words. The task is considered dynamic because it consists of a training phase where feedback is provided and a testing phase at the end. This task was an adaptation from the lexical specificity word learning task designed by Van Goch, McQueen, and Verhoeven (2014) and later used by Janssen et al. (2015). Children saw four pictures on a computer screen and clicked on the correct picture after being presented with an auditory stimulus. Children were also allowed to point to the picture of choice, after which the examiner would click on the corresponding picture with the computer mouse. Reaction time was not used for the analyses.

The assessment consisted of 22 quadruplets of monosyllabic phonetically similar English words (Appendix B) with corresponding pictures. These quadruplets consisted of two unfamiliar target words (e.g. ‘can’, ‘tan’) and two familiar controls (e.g. ‘pan’, ‘man’). Bilingual primary school teachers helped define which words were familiar and which would be unfamiliar to the children. The two target words differed on one acoustic-phonetic feature. The two familiar controls differed from the target words in one acoustic-phonetic feature each. The test started off with a practice phase (five trials), then continued with a training phase (88 trials), and ended with a test phase (22 trials). Each trial started with a fixation cross (500ms) after which four pictures were presented on the computer screen. After 1000ms, the computer played the auditory stimulus: ‘What is … [target word]?’. The child then had to click on the picture that corresponded with that target word. A happy clown would appear when the child answered correctly (Figure 2). No feedback was provided following an incorrect answer.

Figure 2. Sample trial sequence from the first block of the lexical specificity task (English).

A series of five squares displaying a plus sign (only the first square), pictures of a moon and star, cow, can and pan (squares 2, 3 and 4), and a clown with colorful features (square 5), with varying time intervals and a speaker symbol in the third square.

The five trials in the practice phase were used to familiarize the child with the task and to teach the strategy of exclusion (find the unfamiliar target words by ruling out pictures that corresponded to familiar control and filler words). The training phase consisted of two blocks of 44 trials. In the first block, a picture of the unfamiliar target word (e.g. ‘can’) was presented on the screen together with the picture of a familiar control word (e.g. ‘pan’) and two phonologically unrelated filler words (e.g. ‘moon’ and ‘cow’). In the second block, the unfamiliar target word (‘can’) was paired with the other familiar control (‘man’) and two fillers (e.g. ‘dog’ and ‘car’). The same happened with the other unfamiliar target word of the quadruplet (in this case: ‘tan’): it was first paired with one familiar control (‘pan’) and in the second block with the other familiar control (‘man’). Ultimately, in block 3 of the task, the unfamiliar target words (‘can’ and ‘tan’) were presented together with two filler words (e.g. ‘flower’ and ‘fire’). Block 3 was the test phase to check whether the child had learned the meaning of the phonetically very similar unfamiliar target words. The training blocks and the test blocks correlated highly, ranging from r = .438 to .725 with p < .01 in kindergarten and from r = .328 to .632 with p < .01 in grade 1. To retain power and the highest reliability, composite scores of all blocks were obtained and used in the current study. This total number of correct answers (block 1, 2 and 3) also reflected how quickly and well a child learns and remembers new words when trained: a high score indicating correct answers (thus quicker learning). On average, the lexical specificity tasks took about 20 min to be completed. Reliability on both time points was good: Cronbach’s α = .89 (kindergarten) and Cronbach’s α = .82 (grade 1).

Phonological awareness

Phonological awareness was measured by two subtasks: a phoneme blending task, which assessed auditory synthesis skills, and a phoneme segmentation task, which assessed auditory analysis ability. The CTOPP-2 subtest Blending Words (Wagner et al. 2013) was used for the blending subtask. The segmentation subtask had a similar design, based on a Dutch phoneme segmentation task by Verhoeven (1987). Both subtasks comprised of 20 items each. The segmentation tasks started with words with two phonemes for the child to segment (i.e. ‘on’ which the child had to split into ‘o-n’) and increased in difficulty to a maximum of five phonemes in a word (i.e. ‘first’, which needed to be split into ‘f-i-r-s-t’). Children were asked to segment the orally presented words into phonemes after three practice items. In the blending task, children had to combine orally presented phonemes into a word. This task also increased in difficulty; starting with two phonemes in a word (i.e. ‘i-t’ that the child needed to blend into ‘it’) and ending with up to 9 phonemes (e.g. ‘u-n-d-er-s-t-a-n-d’ that had to be blended into ‘understand’). The assessments were ceased after five consecutive errors. The scores of the blending and segmenting subtask were combined into one total score, since both subtasks correlated strongly (r = .727, p < .01). Both subtests displayed good reliability (Cronbach’s α = .91 for segmenting and Cronbach’s α = .94 for blending), as did the sum of both tasks (Cronbach’s α = .95).

Decoding skills

We used the One-minute Reading Test by Transvaal Education Department (1987). Children were presented with a word list of 158 mono-syllabic words and received the instruction to read as many words correctly as possible within one minute. Words had to be pronounced in English (e.g. /a/ in cat had to be pronounced as /æ/ to be correct). Dutch pronunciation of a word led to zero points for that word. Total number of correctly read words was scored and used in the analyses.

Reading comprehension

Dutch reading comprehension scores were obtained from the regular school assessment of reading comprehension required by the Dutch primary school curriculum (Feenstra et al. 2010). These assessments, which are administered twice every year, are standardized tests in which children read 14 short texts and answered 40 multiple choice questions with four answer options. The number of questions per text varied between one and four. The task is divided into two equal parts and is administered on two separate days within the same week. The first task started with three practice texts and five example questions. The actual test consisted of 7 texts and 20 questions. The second task started with three practice texts and four example questions. Part 2 of the actual test consisted of 7 texts and 20 questions as well. For the current study, the midterm raw scores of grade 2 were used. Reliability of this test is good with an estimated reliability score of .86 (Jolink et al. 2015).

English reading comprehension was tested with the Placement Reading subtask of the Primary School Assessment Kit (Integrate Ireland Language and Training 2006). The subtask consisted of 9 blocks of 5 items, each block starting with an example. The blocks increased in difficulty, from linking single words to corresponding pictures in the first and second block, to deciding whether sentences were true or false in the third and fourth block, to putting sentences in the correct order in the fifth and sixth block, to reading short texts and answering multiple choice questions about them in the last three blocks. Two items (items 4 and 5 of the fourth block) were removed from the dataset due to negative item-total correlations after reliability analysis. The sum score of the remaining 43 items was calculated as a score for English reading comprehension. The task had good reliability (Cronbach’s α = .88, p < .01).

Procedure

This study was approved by the Ethics Committee of the Faculty of Social Sciences at our university (ECSW2017-2808-532). All children who participated in this study received active consent from their parents. Tests were administered in a quiet room within the school of the children by several research assistants, all of whom were Educational Sciences bachelor’s or master’s students. There were four time points of testing: midterm kindergarten, midterm grade 1, midterm grade 2 and the first term of grade 3. Lexical quality was measured in two consecutive grades: kindergarten and grade 1. Vocabulary and lexical specificity were assessed to measure semantic and phonological specificity in both kindergarten and grade 1. Phonological awareness was assessed only in kindergarten and decoding skills only in grade 1. Since most bilingual schools start formal reading instruction in Dutch first and later introduce reading in English, reading comprehension scores were obtained in Dutch midterm grade 2 and in English at the start of grade 3. The COVID pandemic happened during data collection. The first lockdown was right after grade 2 testing. The second lockdown was forced right after testing in grade 3.

Results

Descriptive statistics and correlations

Table 1 shows descriptive values for all measured variables. The skewness of L2 decoding was found to be 1.316 (SE = .281), indicating that the distribution was right-skewed. On average, children performed better on vocabulary (t(72) = −7.76, p < .001, Cohen’s d = .75) and lexical specificity (t(72) = −3.89, p < .001, Cohen’s d = .46) after one year. Table 2 shows Pearson’s r correlation between all variables. L2 reading comprehension was significantly correlated with L2 vocabulary, lexical specificity, decoding and L1 reading comprehension.

Table 1. Descriptive Statistics of Vocabulary, Lexical Specificity, Phonological Awareness, Decoding Scores, L1 and L2 Reading Comprehension.

	N	Min	Max	Possible max. score	M	SD
Kindergarten
L2 Vocabulary	73	22	84	204	50.62	16.58
L2 Lexical specificity	73	48	95	110	66.89	12.33
L2 Phonological awareness	73	0	35	40	14.52	10.81
Grade 1
L2 Vocabulary	73	23	107	204	63.71	17.97
L2 Lexical specificity	73	55	91	110	71.85	8.66
L2 Decoding skills	73	3	73	158	20.40	17.44
Grade 2
L1 Reading comprehension	70	15	39	40	28.67	6.31
Grade 3
L2 Reading comprehension	73	17	42	43	34.21	5.78

Note: L2 = English, L1 = Dutch.

Table 2. Correlations for Vocabulary, Lexical Specificity, Phonological Awareness, Decoding Scores, L1 and L2 Reading Comprehension.

	L2 Vocabulary K	L2 Lexical specificity K	L2 Phonological awareness K	L2 Vocabulary G1	L2 Lexical specificity G1	L2 Decoding scores G1	L1 Reading comprehension G2	L2 Reading comprehension G3
L2 Vocabulary K	-	.410**	.151	.654**	.536**	.122	.101	.347**
L2 Lexical specificity K		-	.198	.461**	.509**	.115	.330**	.464**
L2 Phonological awareness K			-	.116	.015	.537**	.316**	.204
L2 Vocabulary G1				-	.635**	.149	.234	.395**
L2 Lexical specificity G1					-	.188	.292*	.498**
L2 Decoding scores G1						-	.458**	.439**
L1 Reading comprehension G2							-	.483**
L2 Reading comprehension G3								-

Note: L1 = Dutch; L2 = English, K = kindergarten, G1 = grade 1, G2 = grade 2, G3 = grade 3.

* p < .05; ** p < .01

Prediction of L2 reading comprehension

As a first step in answering the main research question, a regression analysis was conducted with only the grade 1 and 2 predictors of L2 reading comprehension. Figure 3 shows that L2 lexical specificity and decoding predict L2 reading comprehension, even when L1 reading comprehension is taken into account as a predictor.

Figure 3. Linear Regression Analysis: Effects of L2 Vocabulary, L2 Lexical Specificity, L2 Decoding in grade 1 and L1 Reading Comprehension in Grade 2 on L2 Reading Comprehension in grade 3.

Note: Standardized β coefficients are shown. * p  < .05; ** p  < .01; *** p  < .001.

A diagram showing the variables and arrows illustrating their relations from grade 1 to grade 3, including β-values and asterisks for p-values.

To further examine the effect of L2 lexical quality on L2 reading comprehension, adding the kindergarten measures to the model, a mediation analysis was conducted using the Process add-on in SPSS 25 (Hayes 2013). Mediation analysis is a statistical method to test and explain the mechanisms through which an independent variable influences a dependent variable considering the inclusion of a third, or mediator variable, therefore investigating direct, indirect and total effects. As such, it helps to understand how observed relations between variables are linked together. Hayes’ model 4 was used and run three times alternating between the variables vocabulary, lexical specificity and phonological awareness in kindergarten as independent variable and covariates, making it possible to test all possible total and indirect effects in the model. Vocabulary, lexical specificity and decoding in grade 1 were mediating variables in the model. Bootstrapping was set at 5000 cycles. In the model, direct, indirect and total effects of vocabulary, lexical specificity and phonological awareness in kindergarten (independent variables), vocabulary, lexical specificity and decoding skills in grade 1 (mediators) on L2 reading comprehension in grade 3 (dependent variable) were calculated.

Multicollinearity was not a concern: all variables were only moderately correlated with variance inflation factors (VIF) ranging from 1.050–2.201. The model explained 43% of variance in L2 reading comprehension (R²_total = .434, p < .001). Figure 4 shows the standardized direct effects of the predictors to L2 reading comprehension. Lexical specificity in kindergarten and decoding in grade 1 showed a direct effect and total effect to L2 reading comprehension. An indirect effect was found for L2 phonological awareness on L2 reading comprehension via L2 decoding (95% CI [.12, .32]). Although significantly correlated, vocabulary did not uniquely predict L2 reading comprehension in this model.

Figure 4. Standardized Direct Effects of L2 Vocabulary, Lexical Specificity and Phonological Awareness in Kindergarten, L2 Vocabulary, Lexical Specificity and Decoding in Grade 1 on L2 Reading Comprehension.

Note: Standardized β coefficients are shown. Only significant values are displayed. * p  < .05; ** p  < .01; *** p  < .001.

A diagram showing the variables and arrows illustrating their relations from kindergarten to grade 3, including β-values and asterisks for p-values.

We checked whether vocabulary would take over as a predictor of reading comprehension when lexical specificity in both kindergarten and grade 1 were left out of the model, conducting a mediation analysis without lexical specificity. The effects remained similar for phonological awareness via decoding (β = .53, p < .000) to reading comprehension in grade 3 (β = .42, p < .001) and for vocabulary in kindergarten to vocabulary in grade 1 (β = .65, p < .000). Again, no effect was found for vocabulary in grade 1 to reading comprehension in grade 3 (β = .24, p < .070). Regarding vocabulary, however, an additional total effect was found for vocabulary in kindergarten on reading comprehension in grade 3 (β = .32, p = .005). The model predicted only 14% of the variance (R²_total = .14, p = .004).

Returning to the complete model (Figure 4), we added L1 reading comprehension in grade 2 to check whether the lexical quality predictors would keep their value. All of the effects in the original mediation analysis were still significant once L1 reading comprehension was added to the model (Appendix A). Despite significant predictive value of L1 reading comprehension (β = .221, p = .048) on L2 reading comprehension, L2 lexical specificity and decoding remained significant predictors of L2 reading comprehension.

Discussion

This longitudinal study examined the impact of L2 lexical quality in kindergarten and grade 1 on L2 reading comprehension in grade 3 while taking into account L1 reading comprehension within the novel context of bilingual primary education in the Netherlands. Children were tested in the L2 on vocabulary, lexical specificity and phonological awareness in kindergarten, on vocabulary, lexical specificity and decoding skills in grade 1 and on reading comprehension in grade 3. L1 reading comprehension was tested in grade 2. Results showed that L2 lexical quality impacted L2 reading comprehension, even when L1 reading comprehension was taken into account. More specifically: L2 lexical specificity in kindergarten directly predicted L2 reading comprehension in grade 3, while grade 1 lexical specificity was only significant when leaving out the kindergarten measures. L2 phonological awareness in kindergarten predicted L2 reading comprehension in grade 3 indirectly via L2 decoding skills in grade 1. Vocabulary showed no additional unique predictive value in predicting L2 reading comprehension, but vocabulary in kindergarten showed a significant total effect when excluding lexical specificity from the model.

In accordance with our expectations, we found that lexical quality in the L2 uniquely predicted L2 reading comprehension, which is commensurate with the lexical quality hypothesis (Perfetti and Hart 2002). However, some results were unexpected. First, the effect of vocabulary was not significant in the full model, and second, the effect of lexical specificity was from the kindergarten assessment, and not the grade 1 assessment.

Regarding vocabulary, it should be noted that we did replicate previous research in that decoding and vocabulary together explain variance in reading comprehension (e.g. Kendeou et al. 2009; Verhoeven and Van Leeuwe 2008). The total effect of vocabulary in kindergarten on reading comprehension in grade 3 disappeared when lexical specificity was included in the model. L2 reading comprehension was better predicted by lexical specificity in kindergarten, which was a measure that focused considerably more on the phonological specificity of representations than the vocabulary measure did.

Highly detailed and specific phonological representations seem to be a more important factor in reading comprehension than just the size of vocabulary (Melby-Lervåg, Lyster, and Hulme 2012; Perfetti and Stafura 2014). This also links our findings to the lexical quality hypothesis again: the specificity of phonological representations seems to be more important at this stage than semantic knowledge. Although vocabulary is known to be a predictor of reading comprehension, it has also been shown to be less important in the early years of reading comprehension instruction, since decoding ability is more relevant at that point (Lervåg and Aukrust 2010; Nakamoto, Lindsey, and Manis 2007). Reading comprehension assessments at that age mainly contain high frequency words, relying more on decoding skills than on semantic knowledge or vocabulary. On top of that, the lexical specificity assessment in our study was a dynamic test. It combines measuring current knowledge and learning potential, both of which are important in second language learning (Lantolf and Poehner 2011).

In accordance with our second expectation, we found that L1 reading comprehension in grade 2 significantly predicted L2 reading comprehension in grade 3. The lexical quality predictors kept their predictive value when L1 reading comprehension was taken into account, like we expected. Since educational practice focusses on instruction in decoding and reading comprehension in the L1 first, it makes sense that transfer of that skill of the L1 to the L2 was found. It could be argued that, besides previously found evidence of phonological awareness and knowledge transfers between the L1 and the L2 (Melby-Lervåg and Lervåg 2011), there is also some form of skill-transfer present. We can speculate that the L1 reading comprehension task measures vocabulary, reading and inferencing, skills that might be transferred to L2 reading comprehension.

Limitations

Some limitations of this study should also be addressed. Since the English language has a high status in The Netherlands and is widely available, the children in this study were most likely exposed to English outside instruction or classroom leading to incidental vocabulary learning (Peters 2018; Sundqvist and Sylvén 2016). This was not taken into account in the current study. Future research could therefore include other predictors that could account for individual differences in lexical quality or reading comprehension, such as family factors and exposure to the L2 outside of school. Secondly, generalization of our findings to other contexts or languages should be exercised with caution due to the unique and relatively novel bilingual primary educational setting in which current research was conducted. Even though the Dutch and English orthographies differ with regard to transparency, they share the same alphabetic principles and syllabic structures. The findings reported in this study might not be easily generalized to languages that differ on more linguistic characteristics. Besides that, a study on Dutch children in different stages of learning L2 English at school showed that the PPVT-4 could overestimate vocabulary knowledge of users with limited English proficiency, because the instrument does not represent the actual proportion of cognates within the L1 (Goriot et al. 2021). Even though the children in the current study rarely scored below 24 items correct, the threshold Goriot et al. (2021) set for it being a reliable measure for young Dutch children, it might be the case that the task was not the best fit for our population at that time. On top of that, using the PPVT gave us results on receptive (form-meaning focused) vocabulary skills only. Even though it is a challenge to find an appropriate existing test that measures all aspects of form-meaning knowledge of vocabulary (Laufer and Goldstein 2004), using a broader productive vocabulary test in future research should be explored.

Implications and conclusion

The longitudinal set up of this study gives us a unique insight in direct and indirect relations between first and second language skills of young children in bilingual primary education. The educational situation in the current study is quite exceptional in that children are taught in their L1, which is also their home language and the societal language, for only half of the time, and taught in a foreign language the other half of the time. This curriculum is not considered full immersion, but still children are expected to acquire that second language because of it.

Regarding the outcomes of this study, kindergarten teachers in bilingual primary education should be made aware of the importance of instruction and practice in phonological awareness and lexical specificity, especially since they appear to be so important for reading comprehension later on. Janssen et al. (2019) have already shown that children who received form-focused instruction instead of meaning-focused instruction, showed better development of phonological awareness, semantic knowledge and letter knowledge. The level of lexical specificity also predicted gain in semantic knowledge in this group and not in the group who received training on word meanings. Teaching children in bilingual education might require this focus on word form as well (i.e. phonologically specific minimal pairs within and between the languages). Even though transfer was observed between L1 and L2 reading comprehension in the current study, certain skills in the second language showed direct and indirect effects on reading comprehension. It therefore seems particularly important to attend to lexical quality in both the L1 and the L2 with these young children.

On the other hand, it has been found that training phonological awareness in the L1 is an effective intervention for struggling L2 readers in Canadian French immersion education (Wise, D’Angelo, and Chen 2016). Krenca et al. (2020) have found effects of L1 lexical specificity and phonological awareness on L2 word reading, and stress the importance for teachers to look at both L1 as well as at L2 skills when judging a child’s emerging reading development. We have previously shown that L1 and L2 phonological awareness in kindergarten are highly correlated and interrelated (Kwakkel et al. 2021). L1 and L2 phonological awareness might essentially be the same skill, just measured through different modalities of phonological representations depending on the language. In the Dutch bilingual primary education setting it could therefore possibly be argued that once children understand how decoding works in their L1 (whether through formal instruction or not), they will be able to do the same in their L2 as long as they have enough phonological awareness and vocabulary to make sense of what they read.

To conclude, this study presented data from longitudinal research on bilingual primary education in The Netherlands. Results showed clear evidence that L2 reading comprehension in grade 3 is impacted by L2 lexical quality even after taking into account reading comprehension scores in the L1. Besides L2 decoding scores in grade 1, L2 lexical specificity in kindergarten predicts L2 reading comprehension in grade 3. L2 lexical specificity seems to predict L2 reading comprehension better than vocabulary in the current study. Not the sheer number of words, but their phonological representations (lexical specificity) in the mental lexicon seem to matter most in the early stages of L2 reading comprehension.

Acknowledgements

An expression of gratitude towards the children who participated in this study, the teachers who accommodated data collection within their school and the students and research assistants who helped collect the data.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Nederlandse Organisatie voor Wetenschappelijk Onderzoek: [Grant Number 023.009.050].

Notes on contributors

Hedi Kwakkel

Hedi Kwakkel, MSc. is an external PhD candidate at the Behavioural Science Institute of Radboud University and a primary school teacher in bilingual primary education.

Mienke Droop

Dr. Mienke Droop is assistant professor in Educational Sciences at the Behavioural Science Institute of Radboud University.

Ludo Verhoeven

Dr. Ludo Verhoeven is professor of Communication, Language and Literacy at the Behavioural Science Institute of Radboud University and professor of Language Innovation and Language Policy at University of Curaçao.

Eliane Segers

Dr. Eliane Segers is professor of Learning & Technology at the Behavioural Science Institute of Radboud University and scientific director of the Expertisecentrum Nederlands.

Appendix A

Regression analysis including L1 reading comprehension predicting L2 reading comprehension

Regression analysis coefficients predicting L2 reading comprehension in grade 3 (N = 73)

Table

	R^2	B	SE	β
Model	.499***
Constant		9.380	4.717
L1 reading comprehension G2		.200	.099	.221*
L2 vocabulary K		.059	.044	.167
L2 lexicial specificity K		.114	.052	.245*
L2 phonological awareness K		-.065	.059	-.122
L2 vocabulary G1		.016	.043	.050
L2 lexicial specificity G1		.087	.086	.133
L2 decoding G1		.114	.039	.333**

Note. L1 = Dutch; L2 = English, K = kindergarten, G1 = grade 1, G2 = grade 2.

* p < .05; ** p < .01; *** p < .001

Appendix B

Table of all 24 quadruplets of the lexical specificity task.

Table

Unfamiliar Target A	Unfamiliar Target B	Familiar Control 1	Familiar Control 2
GAD	GAL	CAT	GAS
TUG	TUB	TURN	TONGUE
BUM	BUN	BIRD	BUS
GANG	GAG	CAP	GAS
POP	POCK	PORK	POT
RAP	RAG	RAM	RAT
GALL	CAWL	SJAWL	BALL
BUCK	BUG	BIRD	BUS
GAB	GAP	CAT	GAS
TIN	TIG	THINK	TICK
LIE	RYE	I	PIE
DIP	RIP	LIP	SHIP
DEAD	LEAD	RED	BED
CAN	TAN	MAN	PAN
COX	POX	SOCKS	ROCKS
HEW	FEW	BLUE	NEW
SIGH	FIE	I	PIE
HAKE	SHAKE	BAKE	CAKE
FAN	VAN	PAN	MAN
FOX	VOX	SOCKS	BOX
FOAL	VOLE	MOLE	HOLE
TIP	TUP	TYPE	TOP