Abstract
Aims:
This meta-analysis aimed to quantify the overall within- and cross-linguistic relationships between morphological awareness (MA) and vocabulary knowledge and identify moderators that influence their associations in monolingual and bilingual learners.
Methodology:
We conducted a meta-analysis using 109 primary studies that met our inclusion criteria. Data from 79 L1 studies (involving 14,172 monolingual learners) and 30 L2 studies (involving 4,288 bilingual learners) were extracted.
Data and analysis:
Studies were coded for learner characteristics, such as age, language background, and L1–L2 distance, as well as assessment characteristics, including task modality (e.g., oral vs. written), task content (e.g., derivation and compound), and response type (production vs. reception).
Findings/conclusions:
The results indicated that (a) in monolingual research, there was a moderate correlation between MA and vocabulary (r = .43); (b) in bilingual research, there was a significant positive correlation between L1 and L2 MA (r = .33), and between L1 MA and L2 vocabulary (r = .26); (c) the magnitude of the association between MA and vocabulary increased with age in monolingual learners, while that of the correlation between L1 and L2 MA decreased with age in bilingual learners; and (d) significant variations were observed depending on the specificity of the language(s) involved, as well as several assessment-related moderators.
Originality:
This study contributes to the existing literature by systematically exploring the degree of association between MA and vocabulary within and across languages and examining whether these associations are influenced by learner and assessment characteristics.
Significance/implications:
Our results support MA as a multidimensional, transferrable construct and underscore its important role in vocabulary for monolingual and bilingual learners. Several key considerations are discussed for the selection of MA and vocabulary assessment methods.
Introduction
Acquiring the extensive vocabulary necessary for effective communication and language proficiency can be a challenging task for both first-language (L1) and second-language (L2) learners (Schmitt, 2010). An important skill for addressing this vocabulary learning challenge is morphological awareness (MA; e.g., Carlisle, 2000; McBride-Chang, Wagner, et al., 2005; Sparks & Deacon, 2015). MA is a multidimensional construct defined as “the ability to reflect on and manipulate morphemes and employ word formation rules in one’s language” (Kuo & Anderson, 2006, p. 161). Increasing evidence supports that MA is associated with vocabulary knowledge in monolingual learners (e.g., McBride-Chang et al., 2003) and that MA is a sharable resource that can be transferred to enhance vocabulary knowledge in bilingual learners (e.g., Lam & Chen, 2018).
However, there exists considerable variation in the magnitude of the within- and cross-linguistic correlations between MA and vocabulary in both L1 and L2 contexts. Studies also differed substantially in learner (e.g., age, writing system) and measurement characteristics, such as task content (e.g., derivation/compound/inflection, vocabulary breadth vs. depth), and response types (e.g., reception vs. production). Unfortunately, we are less clear about whether such methodological variation would impact their correlational strengths.
A recent meta-analysis by Lee et al. (2023) addressed the role of MA in L1 literacy development and revealed a robust association between MA and vocabulary. However, they did not systematically examine test modalities (e.g., oral vs. written, receptive vs. expressive vocabulary) that may influence the difficulty of morphological and vocabulary tasks and their relationship (e.g., Deacon et al., 2008; Pan et al., 2023; Spencer et al., 2015). In two meta-analyses of L2 learning, Ke et al. (2021) examined the transfer of MA in biliteracy development (decoding and reading comprehension), whereas Yang et al. (2017) focused specifically on the sharing of L1 and L2 MA between Chinese and English. However, neither study directly addressed the cross-linguistic association between L1 MA and L2 vocabulary knowledge in bilingual settings.
This study aimed to systematically quantify the overall relationships between MA and vocabulary knowledge and to identify important moderators of these associations through a meta-analysis. Specifically, we extended previous meta-analyses by (a) examining the strength of within- and cross-linguistic relationships between MA and vocabulary knowledge and by (b) investigating the moderating effects of learner characteristics (e.g., age, L1–L2 distance, orthographic depth) and assessment characteristics (e.g., MA task content and modality). By doing so, we aimed to deepen our understanding of MA as a multidimensional, transferable construct and illuminate key factors that impact its within- and cross-linguistic relationships with vocabulary knowledge in monolingual and bilingual learners.
Review of literature
Theoretical background
MA and vocabulary knowledge
MA concerns the understanding and recognition of the smallest units of meaning (i.e., morphemes), including the ability to manipulate morphemes to form new words and segment words into their meaningful components (Kuo & Anderson, 2006; H. Zhang, 2016). Theoretically, MA may support vocabulary by facilitating word meaning retrieval and integrating the properties of word form and syntax (Nagy et al., 2014)—both of which are essential for efficiently organizing the mental lexicon (Sandra, 1994). MA allows the manipulation of component morphemes and represents properties of phonology and syntax (Mahony et al., 2000). Such awareness can help discern the meanings of unfamiliar words (e.g., Levesque et al., 2017) and ultimately expedite vocabulary learning (e.g., Bowers & Kirby, 2010). Moreover, what is triggered by MA may expand into the activation of a word’s form and its syntactic use. As posited by the binding agent theory (Kirby & Bowers, 2017), morphology is an essential mechanism that strengthens the retrieval of highly integrated lexical representations, featuring the semantic, orthographic, and phonological aspects of words. Overall, MA is an integrated metalinguistic skill (Kuo & Anderson, 2006) that may facilitate different facets of vocabulary knowledge.
Empirical studies on monolingual learners have found a positive association between MA and vocabulary knowledge, yet the magnitude of this correlation is inconsistent ranging from low (e.g., Li et al., 2017; Rothou & Padeliadu, 2015) to high (e.g., Gottardo et al., 2018; Kieffer et al., 2016). Factors related to the learner (e.g., age, writing system) and assessment variables (e.g., MA tasks) may prove to be crucial for understanding these discrepancies. In particular, Lee et al. (2023) found in their meta-analysis that MA correlated more strongly with vocabulary in upper elementary grades than in primary grades. In addition, they identified moderating effects related to the nature of MA tasks, with stronger correlations for productive MA than for receptive MA, and for derivational and compound MA than for inflectional MA.
Cross-linguistic transfer of MA and vocabulary
The transfer facilitation model offers a framework for understanding how MA transfers in bilingual settings (Koda, 2008), with transfer defined as “an automatic activation of well-established first-language competencies, triggered by second-language input” (p. 78). According to this model, metalinguistic competencies (e.g., MA) that have been well-developed to an automatic level in L1 may be transferred to improve L2 reading skills such as vocabulary—a process known as the facilitation effect. Yet, automaticity may not be a necessary antecedent for transfer to occur (Chung et al., 2019). For instance, research suggests that L1 learners may exhibit differing levels of MA competence (e.g., Verhoeven & Perfetti, 2011), and that transfer may occur in children of varied L1 proficiency levels (D. Zhang et al., 2024). Clearly, systematic reviews are needed to clarify which factors may affect the cross-linguistic transfer of MA in bilingual learners.
A growing body of evidence supports that transfer of MA may occur at the construct level (i.e., L1 and L2 MA) or feed into L2 subskills such as decoding and vocabulary (i.e., crossover effects, see the studies by Hipfner-Boucher & Chen, 2016; D. Zhang et al., 2024). Research on bilingual learners has reported small (e.g., Gottardo et al., 2020) or moderate-to-high correlations between L1 MA and L2 vocabulary (e.g., Bae & Joshi, 2018; Marks et al., 2022), although nonsignificant correlations have also been observed (e.g., D. Zhang & Koda, 2014). In general, biliteracy researchers agree that transfer is an interactive process depending on multiple factors such as L1-L2 proficiency and language complexity (Chung et al., 2019).
To the best of our knowledge, two meta-analyses are particularly relevant to this study. Yang et al. (2017) examined the transfer of MA between Chinese and English and identified a small yet significant correlation between L1 and L2 MA (r = .37). Ke et al. (2021) investigated the cross-linguistic effects of L1 MA in L2 biliteracy skills (decoding and reading comprehension) across nine languages and four writing systems. They found that L1 MA was significantly associated with L2 MA (r = .30), and that age, task content, or the L1–L2 writing system distance did not moderate this cross-linguistic correlation. These findings thus support the cross-linguistic shareability of MA in L2 learners’ reading development. However, the crossover effects of L1 MA on L2 vocabulary remain underexplored, and assessment-related factors, such as MA modalities and response types, have not been systematically examined through meta-analysis.
Moderating effects of learner and assessment characteristics
To examine the observed variability in the within- and cross-language correlations between MA and vocabulary, the following section reviews the most relevant factors that could act as moderators, including learner (age, writing system, orthographic depth, L1–L2 language family and orthographic distance) and assessment characteristics (task content, modality, and response type).
Learner characteristics
Prior studies have identified the moderating effect of age or grade. MA appears to play an increasingly important role in monolingual reading acquisition (Berninger et al., 2010). With development, children encounter more morphologically complex vocabulary. Yet, evidence on the developmental relationship between MA and vocabulary is mixed. Some studies have identified a persistent and developmentally increasing correlation between MA and vocabulary (e.g., Carlisle, 2000; Ku & Anderson, 2003; Singson et al., 2000). For example, Ku and Anderson (2003) found that MA was a robust predictor of vocabulary in Chinese- and English-speaking students in the second, fourth, and sixth grades. However, other studies have reported a restricted link across grade levels (e.g., Deacon et al., 2014). Apart from the variations in L1 reading, less is known about this developmental pattern in bilingual learners. One of the goals of our study was to examine whether age moderated the within- and cross-linguistic correlation between MA and vocabulary knowledge.
Another learner-related moderator is the linguistic and orthographic features of the language. For example, studies have reported that orthographic depth may influence the role of MA in reading (Mousikou et al., 2020). By offering regularities in deep orthographies like English, morphological units might be prominent in supporting reading skills (Levesque et al., 2021; Rastle, 2019). Readers of deep orthographies may rely heavily on MA to acquire morphologically complex vocabulary, particularly from Grade 4 onwards, when a large portion of the new vocabulary is learned through reading (Nagy & Anderson, 1984). In contrast, research suggests that the role of MA in reading appears less pronounced in orthographically shallow languages where phoneme-grapheme mappings are more consistent (Lee et al., 2023). Among bilingual readers, the cross-linguistic strength of MA transfer may vary with L1–L2 distance (e.g., D. Zhang & Koda, 2012; Sun et al., 2022). For instance, Sun et al. (2022) reported that the MA transfer impact was twice as strong in Spanish-English bilinguals relative to Chinese-English bilinguals. Overall, we tested the moderating effects of learners’ writing system, orthographic depth, and L1–L2 language distance.
Assessment characteristics
Existing studies have also differed in the characteristics of tasks used to assess both MA and vocabulary knowledge. Deacon et al. (2008) underscored the importance of examining a taxonomy of morphological tasks such as MA modality (e.g., oral vs. written), task content (e.g., derivation vs. compound), and response type (e.g., reception vs. perception). These tasks may pose varying demands and difficulty levels, which is of particular consideration for understanding the reported age differences in the relationship between MA and vocabulary. For example, compared to inflectional awareness, awareness of derivations and compounds was reported to develop throughout elementary years and may have a stronger correlation with vocabulary (Kuo & Anderson, 2006; Lee et al., 2023). In addition, the oral modality is considered to be less challenging than the written format and is typically used with younger children (e.g., Cho & McBride, 2022). Furthermore, the type of response required—MA tasks involving judgment or production of an answer—can influence test difficulty and affect its correlation with vocabulary (Apel, 2014; Deacon et al., 2008).
Vocabulary measures also exhibit different features across studies. Prior studies suggest that different vocabulary measures may capture distinct aspects of vocabulary, such as receptive or expressive vocabulary, which are presumed to develop along distinct trajectories (Henriksen, 1999). These variations may, in turn, affect their relationship with MA (e.g., Ouellette, 2006; Pan et al., 2023), especially in consideration of development. For example, Rothou and Padeliadu (2015) found that MA was more associated with expressive vocabulary than receptive vocabulary among Greek-speaking children across Grades 1–3. Key features of vocabulary measures include the dimensions of vocabulary knowledge (breadth vs. depth), modality (oral vs. written), and response type (receptive vs. expressive). Vocabulary breadth refers to the number of words known (Nation, 2001) and is assessed with tasks like word naming or checklist tests (Sterpin et al., 2021). In contrast, vocabulary depth reflects the deep understanding of a word’s properties across contexts (Anderson & Freebody, 1983) and is typically measured through word definition tasks or polysemy tasks (Binder et al., 2017). Another important factor is the distinction between receptive vocabulary—the ability to recognize the form of a word or understand its meaning—and expressive vocabulary, which includes the production of a word to convey its appropriate meaning in context (Laufer et al., 2004). Receptive vocabulary measures may include matching words to pictures or selecting definitions, while expressive vocabulary measures often involve producing synonyms or definitions (Jeon & Yamashita, 2014). Therefore, one of our goals was to test whether assessment-related variables may influence the strength of the correlation between MA and vocabulary knowledge.
The present study
The purpose of the present study was to examine the magnitude of the overall within- and cross-linguistic correlations between MA and vocabulary knowledge in monolingual and bilingual learners and investigate potential moderating effects of learner and assessment characteristics. Specifically, the present study aims to answer the following questions: (a) In monolingual learners, is MA correlated with vocabulary knowledge? If so, do learner and assessment characteristics affect their correlational strength? (b) In bilingual learners, is MA in one language correlated with MA in another language? If so, do learner and assessment characteristics affect their correlational strength? (c) In bilingual learners, is MA in one language correlated with vocabulary knowledge in another language? If so, do learner and assessment characteristics affect their correlational strength? However, it should be noted that we assessed correlations rather than causations. The current study could not determine the causal relationship or directionality between MA and vocabulary, despite some evidence supporting that vocabulary growth may foster MA and that increased MA may further contribute to vocabulary expansion (e.g., Kieffer & Lesaux, 2012).
Research design
Literature search
The literature search was conducted on five databases (ERIC, PsycINFO, PubMed, ProQuest, and Google Scholar) for publications from 2000 to March 2023. We used combinations of several key terms: “morphological awareness,” “vocabulary/vocabulary knowledge,” “first language,” “second language,” “monolingual,” and “bilingual” separated by the Boolean operators “AND” and “OR.” To complement the electronic search process, we checked additional sources from previous reviews (e.g., Lee et al., 2023) and relevant book chapters (e.g., Lü, 2019). As a result, 1,002 records were identified and included in the screening phase (screening of abstracts and full-texts).
For inclusion in our first research question (examining the relationship between MA and vocabulary in monolingual learners), studies should meet the following inclusion criteria: (a) report data for samples from mainly monolingual backgrounds (e.g., as opposed to the multilingual context); (b) report sample sizes and Person’s r correlation between MA and vocabulary; (c) provide explicit descriptions of MA and vocabulary assessment characteristics; and (d) assess MA and vocabulary at the same time point.
As for our second and third research questions (examining the cross-linguistic transfer of MA and its crossover effects with vocabulary), studies to be included must (a) focus on learners with bilingual backgrounds (simultaneous, sequential/emergent bilinguals, or L2 learners); (b) report sample sizes and Person’s r correlations between L1 and L2 MA or between L1 MA and L2 vocabulary; (c) describe participant and assessment characteristics of MA or/and vocabulary knowledge tasks; and (d) assess MA and vocabulary at the same time point. In addition, we used three exclusion criteria for this study: (a) duplicate reports, (b) studies reporting data from participants older than 18 years or with learning disabilities, and (c) studies with unavailable full texts.
Based on the inclusion and exclusion criteria, 109 studies were included for review (Figure 1). Overall, 79 studies reported correlations between MA and vocabulary in monolingual learners, and 30 studies reported cross-linguistic transfer of MA or its associations with vocabulary in bilingual learners.
Flow diagram of the literature search and inclusion of studies.
Coding schemes
Studies were coded based on study features (author, year), sample sizes, moderators, and effect sizes. Detailed coding information is attached in the Supplemental Material Appendices S1 and S2. When multiple measures were reported within the same study sample (e.g., assessing MA with two different tasks), we coded individual correlations and averaged them for overall calculations to avoid effect-size multiplicity (López-López et al., 2018). For longitudinal studies, we included only the data from the first time point (Melby-Lervåg & Lervåg, 2011).
Our moderator analyses examined three factors involving a total of 11 variables (Table 1): (a) learner characteristics (age, orthographic depth, writing system, language family, and orthographic distance); (b) MA assessment (MA task content, MA modality, and response type); and (c) vocabulary assessment (vocabulary modality, dimension, and response type). Studies that used multiple measures of MA but reported only the overall correlation between MA and vocabulary (rather than separate correlations) were excluded from the moderator analysis.
Moderator coding for this meta-analysis.
Note. VOC = vocabulary knowledge; N/A, not applicable. Detailed summaries for all the studies included in our meta-analysis can be found in Supplemental Appendices S1 and S2.
Data analysis
Effect-size calculations
All analyses were conducted using the Comprehensive Meta-Analysis Version 3.0 (CMA) software (Borenstein et al., 2014). First, effect sizes were computed based on Pearson’s r correlation coefficients. These coefficients were converted into Fisher’s Z and then automatically transformed back to Pearson’s r for reporting (Borenstein et al., 2021). Second, we addressed the dependency of effect sizes following the standard meta-analytic practices (Borenstein et al., 2021). When multiple effect sizes were reported within the same study, they were averaged into a single aggregated effect size to represent the overall correlation. However, for studies reporting correlations from multiple independent samples, each sample was treated as an individual unit, contributing a unique effect size. Third, given the expected variability across included studies, we used random-effects models to calculate the combined effect sizes and their 95% confidence intervals. Finally, in interpreting the strength of effect sizes, we adopted empirical benchmarks to account for the specificity of research data, as recommended by Hill et al. (2008). The following thresholds proposed by Plonsky and Oswald (2014) from language learning research were used, with Pearson’s r of .25, .40, and .60 representing small, medium, and large effects, respectively.
Outlier analysis and publication bias
To examine the impact of outliers on the overall mean correlation range, sensitivity analysis was performed using the “one-study removed” approach. In addition, we investigated the risk of publication bias (i.e., studies with nonsignificant results are less likely to be published) through several visual and statistical methods (Borenstein et al., 2009). First, the visual inspection funnel plot was checked (see the funnel plot section in Supplemental Appendix S3 of the Supporting Information), with asymmetry in this plot indicating publication bias. The trim-and-fill method for the random-effects model (Duval & Tweedie, 2000) was conducted to adjust this publication bias. Second, we checked the classic fail-safe N to determine the number of additional null effects that would be required to overturn the observed effect size. Following Rosenthal’s (1991) guideline, we considered the overall effect size robust if the fail-safe N value exceeded 5k + 10 (where k is the number of observed studies). In addition, we used the regression intercept of Egger et al. (2003) to evaluate publication bias, with a nonsignificant value indicating a lack of bias.
Heterogeneity of variance
To quantify the heterogeneity of variance across effect sizes, we conducted the Q test and the I2 statistics (Hedges & Olkin, 2014). The Q test of difference at p values of <.05 indicates the presence of between-study variability. The I2 statistic shows the magnitude of true heterogeneity observed across included studies (Borenstein et al., 2009), with values of 50%–75% suggesting a moderate amount of variance, enough to examine moderator effects (Higgins et al., 2003).
Moderator analyses
Moderator analyses were performed using the shifting unit of analysis approach (Cooper, 2010). As noted earlier, many of our included studies reported multiple outcomes from the same population (e.g., using several MA and vocabulary measures). When calculating overall correlations and analyzing participant-related moderators (e.g., age, writing system, orthographic depth), we aggregated effect sizes at the sample level to ensure that each sample was represented by a single effect. However, for moderator analyses by assessment types (e.g., MA and vocabulary modalities), the analysis unit was shifted from sample to incorporating individual effect sizes to analyze outcomes separately. This approach is a compromise between avoiding the loss of data while meeting the assumption of statistical independence to the extent possible (Cooper et al., 2010). Similar strategies have been employed in prior meta-analyses (e.g., Araújo et al., 2015; Kim et al., 2019). In total, we examined five learner characteristics and six assessment characteristics related to MA and vocabulary tasks. Specifically, age was the only continuous moderator and was assessed using a method of moments meta-regression analysis with random effects. All other categorical variables (e.g., receptive vs. expressive vocabulary) were examined with between-group Q statistics using mixed-effects models.
Results
RQ1: correlation between MA and vocabulary in monolingual learners
Our first research question synthesized 79 L1 studies and 99 unique samples comprising a total of 203 correlations between MA and vocabulary knowledge. The independent samples included a total of 14,172 monolingual learners from 11 languages (Arabic, Chinese, Dutch, English, French, German, Japanese, Greek, Hebrew, Korean and Persia) with a mean age of 8.36 years (ranging from 4.90 to 13.79 years). The sample size of the individual studies ranged from 30 to 686. As shown in Table 2, the overall association between MA and vocabulary knowledge was moderate, r = .43, 95% CI [.40, .46], and significant, z (98) = 22.70, p < .001. A sensitivity analysis indicated that the overall correlation remained consistent, ranging from r = .421 (95% CI [.391, .450]) to r = .432 (95% CI [.400, .463]). The heterogeneity in correlations among effect sizes was significant and large, Q (98) = 524.05, p < .001. I2 statistic suggested that 81.30% of the observed variation was due to between-study differences (see also Figure 2). Therefore, further moderator analyses are required to examine the heterogeneity.
Overall analysis and the weighted correlations for the three outcomes.
Note. VOC = vocabulary knowledge; N = number of primary studies included in the analysis; k = number of unique samples; n = total sample size; 95% CI = 95% confidence interval; r = Pearson correlation; Z = Fisher’s Z; I2 = proportion of the observed variance that shows real differences across studies.
p < .05; **p < .01; ***p < .001.
Forest plot of the relationship between L1 MA and L1 vocabulary.
We examined publication bias with several methods as previously noted. First, the funnel plot showed an asymmetry, with studies missing on the right of the mean. The trim and fill method imputed 20 studies on the right, yielding an increased adjusted correlation of r = .47 (95% CI [.44, .50]). This adjusted effect size suggests that our computed correlation between MA and vocabulary might be underestimated. Nevertheless, the Egger’s regression test (t = 0.12, p = .90) showed an absence of publication bias. The classic fail-safe N (N = 69067) suggested that our observed effect size is unlikely to be nullified by a substantial number of missing studies. Therefore, these results demonstrated little concern regarding publication bias and supported the robustness of our findings.
Moderating effects of learner and assessment characteristics
We examined whether the correlation between MA and vocabulary knowledge was moderated by three learner-related variables and six assessment features. Following previous meta-analytic procedures (Ke et al., 2021), we excluded subgroups with fewer than two data points: “syllabary” (k = 1) from the writing system category and “oral and written” (k = 1) from the vocabulary modality category. In assessment moderator analyses, five studies were excluded because they did not provide correlations between separate tests of MA and vocabulary (Apel & Thomas-Tate, 2009; Diamanti et al., 2017; Dulay et al., 2021; McCutchen et al., 2008; Tibi et al., 2019). The results of the categorical moderator analyses are summarized in Table 3.
Moderator analyses in the relationship between MA and vocabulary knowledge.
Note. VOC = vocabulary knowledge; Qbetween = the index of between-groups variability.
p < .05; **p < .01; ***p < .001.
Age
A meta-regression showed that age had a significant and positive impact on the correlation between MA and vocabulary knowledge (β = .02, p < .05), explaining 5% of the total between-study variance. As shown in Figure 3, the effect sizes increased with age.
Meta-regression plot of L1 MA and L1 vocabulary knowledge on age.
Writing system
Writing system was found to be a significant moderator (Q(2) = 17.23, p < .001). The effect size was larger in alphabetic languages (r = .49) than in logographic (r = .37) and abjad languages (r = .37).
Orthographic depth
Orthographic depth significantly moderated the magnitude of effect sizes (Q(1) = 3.89, p = .048). The effect size for orthographically deep languages (r = .44) was larger than that for orthographically shallow languages (r = .38). Of note, we excluded one Japanese sample by Inoue et al. (2017) because they did not provide which types of scripts (e.g., Kanji, hiragana) were used in MA tasks. Still, results should be interpreted with caution due to the imbalanced data across this category: 79 independent studies focused on opaque orthographies while only 19 examined transparent orthographies.
MA assessment
MA task content (Q(3) = 12.65, p < .01) and modality (Q(2) = 13.40, p < .01) were found to significantly moderate the correlation between MA and vocabulary knowledge. MA tasks that tested mixed morphology showed the highest effect size (r = .49), followed by tasks that tested derivation (r = .46), compound (r = .40), and inflection (r = .37). Follow-up pairwise tests revealed that the effect size with mixed morphology was significantly greater than the other measures, except for derivation morphology. No significant differences were found between compound morphology and inflectional morphology. As for the MA modality, the combined modality (r = .50) had a significantly larger effect size than the oral (r = .43) and written modality (r = .38). Pairwise comparisons revealed no significant differences between oral and written MA modalities. Regarding the response type, the average effect for receptive and productive MA tasks did not significantly differ (Q(1) = 2.64, p = .10).
Vocabulary assessment
Vocabulary dimensions (Q(1) = 6.37, p < .05) and response types (Q(1) = 7.20, p < .01) were significant moderator variables that influenced the correlation between MA and vocabulary knowledge. Studies that tested vocabulary depth (r = .40) had a smaller effect size than studies that assessed vocabulary breadth (r = .46). Similarly, the effect size was significantly smaller in studies that used expressive (r = .40) vocabulary tasks than in studies that used receptive vocabulary tasks (r = .46). Yet, the average effect for oral and written vocabulary modality did not significantly differ (Q(1) = 0.94, p = .33). The results are summarized in Table 3.
RQ2: correlation between L1 and L2 MA in bilingual learners
The second research question synthesized 30 L2 studies and 34 unique samples comprising a total of 124 correlational coefficients between L1 and L2 MA. The independent samples included 4,288 bilingual learners with various L1 and L2 backgrounds (Arabic, Chinese, English, French, Korean, Malay, and Spanish) with a mean age of 8.44 years (ranging from 5.76 to 13.47 years). The sample size of the independent studies ranged from 35 to 349. As summarized in Table 2, the overall weighted effect size was small, r = .33, 95% CI [.26, .39]) yet significant, z (30) = 9.09, p < .001. A sensitivity analysis showed a consistent correlation between L1 and L2 MA across the included studies, ranging from r = .32 (95% CI [.26, .37]) to r = .34 (95% CI [.27, .40]). The heterogeneity analysis revealed significant variations between studies (Q(30) = 179.81, p < .001, I2 = 81.65), suggesting the presence of moderators (see also Figure 4).
Forest plot of the relationship between L1 MA and L2 MA.
As for the publication bias, the funnel plot indicated that studies were missing on the right side of the mean. In the trim and fill analysis, eight studies were imputed, and the adjusted overall effect size increased to r = .38 (95% CI [.32, .44]). However, Egger’s regression test showed no evidence of publication bias (t = 0.44, p = .66), and the classic fail-safe N (N = 3904) suggested that our findings were robust against publication bias.
Moderating effects of learner and assessment characteristics
Age
Meta-regression analysis revealed that age significantly moderated the correlation between L1 and L2 MA, Q(1) = 3.92, β = -.03, p = .047, with the effect size decreasing with age (Figure 5). In other words, our result suggests a greater cross-language relationship between L1 and L2 MA for younger students.
Meta-regression plot of L1 MA and L2 MA on age.
Orthographic distance and language family
Concerning bilingual learner’s language background, neither orthographic distance, Q(1) = 0.01, p = .98, nor language family, Q(1) = 0.02, p = .90, had a significant impact on the weighted correlation between L1 and L2 MA. The moderator analysis results are displayed in Table 4.
Moderator analyses in the relationship between L1 MA and L2 MA.
Note. N = number of studies; Qbetween = the index of between-groups variability.
p < .05; **p < .01; ***p < .001.
Regarding MA assessment, we should note that although few independent samples (k = 4) reported on inflectional morphology, we did not exclude them from the moderator analyses because inflectional morphology is a conceptually important variable of MA task content. Overall, our results revealed that only MA response type was a significant moderator, Q (2) = 35.23, p < .001. Studies that used perception tests (r = .22) had a smaller effect size than studies that used the production (r = .32) or combined tests (r = .34). Follow-up pairwise tests revealed that the effect size for perception tests was significantly weaker than the other measures. The differences between production and combined tests were not significant. However, no significant moderator effect was found on the effect size for either MA modality (Q(2) = 2.87, p = .24) or MA features (Q(3) = 1.26, p = .74).
RQ3: correlation between L1 MA and L2 vocabulary in bilingual learners
Our third research question synthesized 26 L2 studies, 30 unique samples, and a total of 45 correlational coefficients addressing the cross-language relationship between L1 MA and L2 vocabulary. The independent samples included 3,543 bilingual learners with a mean age of 8.29 years (ranging from 5.76 to 13.5 years). The sample size of the independent studies ranged from 35 to 349. As shown in Table 2, our analyses yielded a small yet significant correlation between L1 MA and L2 vocabulary, r = .26, 95% CI [.20, .31], z (29) = 9.02, p < .001. A sensitivity analysis showed the correlation between L1 MA and L2 vocabulary was consistent across the included studies, ranging from r = .25 (95% CI [.20, .29]) to r = .27 (95% CI [.22, .32]). The heterogeneity analysis found significant variations, Q(29) = 78.00, p < .001, I2 = 62.82% (see also Figure 6).
Forest plot of the relationship between L2 MA and L2 vocabulary.
The inspection with funnel plot analysis revealed a symmetry distribution. In the trim and fill analysis, two studies were imputed, and the adjusted overall effect size was r = .28, 95% CI [.22, .33]. A publication bias is improbable given that the fail-safe N suggested 1,680 missing studies would be needed to nullify the observed effect size, and that the result of Egger’s regression test of the intercept was nonsignificant (t = 0.04, p = .96).
Moderating effects of learner and assessment characteristics
Meta-regression analysis found that age was not a significant moderator for the cross-language correlation between L1 MA and L2 vocabulary, Q(1) = 0.00, p = .98. Regarding learner’s language background, moderator analyses showed that language family (Q(1) = 4.15, p < .05) and orthographic distance (Q(1) = 3.93, p < .05) were significant moderators. Specifically, studies with orthographically similar languages (e.g., both L1 and L2 used alphabets) produced a higher effect size (r = .32) than those with orthographically dissimilar languages (r = .21). The effect size was significantly higher for languages within the same language family (e.g., both L1 and L2 were Indo-European; r = .35) than for languages from different language families (r = .22). The moderator analyses’ results are displayed in Table 5.
Moderator analyses in the relationship between L1 MA and L2 vocabulary knowledge.
Note. VOC = vocabulary knowledge; Qbetween = the index of between-groups variability.
p < .05; **p < .01; ***p < .001.
In the moderator analyses of assessment features, one study was excluded for lacking correlations between separate tests of L1 MA and L2 vocabulary (Xie et al., 2022). Regarding MA assessment, we did not find any statistically significant moderator effects of MA task content (Q(3) = 6.38, p = .10), modality (Q(2) = 2.48, p = .29), or response type (Q(1) = 0.26, p = .61). As for the vocabulary assessment, no moderator analyses for vocabulary modality and dimension were performed due to insufficient data. In addition, there were no significant differences found between the effect size for studies testing receptive vocabulary and expressive vocabulary (Q(1) = 0.05, p = .82).
Discussion
The relationship between MA and vocabulary in monolingual learners
Our first research question explored the overall association between MA and vocabulary knowledge, as well as the possible moderating effects of learner-related and assessment-related factors. Meta-analysis results revealed that the aggregated effect size was r = .43, indicating a medium-sized relationship between MA and vocabulary in monolingual learners. The adjusted effect size we found of .47 was close to .50 reported by Lee et al. (2023) in their recent meta-analysis. The overall positive correlation between MA and vocabulary aligns with the previous studies and theoretical perspective, reinforcing the significant role of morphology in the development of vocabulary knowledge (e.g., Ku & Anderson, 2003; Kuo & Anderson, 2006; McBride-Chang, Cho, et al., 2005). In addition, our results showed a high degree of heterogeneity (I2 = 81.30%), and the moderator analyses identified seven factors that may account for the observed variations in effect size.
The first important finding is that age significantly moderated the relationship between MA and vocabulary, with the correlation increasing as participants get older. In this review, age was analyzed as a continuous variable to provide a thorough examination of its moderator effect (Melby-Lervåg & Lervåg, 2011). The mean ages of the participants ranged from 4.9 to 13.79 years. Of the 99 independent samples, the majority focused on children at the initial (5–8 years, 44.90%) and intermediate (8–12 years, 52.04%) stages of reading development, while a smaller proportion examined older participants (3.06%). Our findings confirm previous empirical studies that MA exerts a developmentally increasing impact on vocabulary in monolingual reading acquisition (e.g., Carlisle, 2000; Ku & Anderson, 2003). Our results are also consistent with the study by Lee et al. (2023) who reported a stronger association between MA and vocabulary in upper elementary grades than in primary grades. Developmentally, children may hone morphological skills as they encounter increasingly complex words with exposure, such that older children who can effectively employ morphological cues are better equipped to learn morphologically unfamiliar vocabulary. There is also evidence that MA and vocabulary are bidirectionally related (e.g., Dulay et al., 2021), suggesting that the expansion of vocabulary may in turn improve the ability to identify and manipulate morphemes. However, it is worth noting that, despite a reliable correlation between MA and vocabulary, this meta-analysis of correlational studies did not reveal a causal relationship. Future meta-analytic research based on experimental designs could help elucidate the directionality of their association.
Another important finding is the significant moderation of the writing system and orthographic depth. Our results found that the weighted correlation tended to be higher for deep orthographies (.44) than for shallow orthographies (.38), and that MA was more strongly related to vocabulary knowledge in alphabetic writing systems (.49) than in non-alphabetic languages (.37). Morphology offers regular information in languages with inconsistent grapheme-phoneme mappings (Rastle, 2019). In alphabetic languages like English, morphology is essential for learning the meanings of vocabulary, as it encourages students to acquire the meanings of vocabulary by discriminating and analyzing the constituent parts and morphological structures of vocabulary (McBride, 2016). Overall, these findings converge to support a consistent association between MA and vocabulary in monolingual learners across cultures (e.g., McBride-Chang, Cho, et al., 2005), while also highlighting the influence of writing system variations on this relationship. Our results also align with empirical studies suggesting that orthographic consistency influences the extent to which MA is involved in reading development (e.g., Mousikou et al., 2020). However, we note that the majority of studies tested alphabetic (55.56%) and logographic writing systems (40.40%), whereas very few studies tested abjad (k = 3) and syllabary writing systems (k = 1). Studies with greater diversity (e.g., Arabic, Hebrew) are required to examine the moderator effect of writing system variations in the relationship between MA and vocabulary, and future research may incorporate a more nuanced analysis of multiple writing system complexity (Daniels & Share, 2018).
The third set of findings provides clues as to whether the observed correlations vary with MA and vocabulary assessment features. We found that the association between MA and vocabulary varied significantly with the content of MA tasks. The largest effect sizes were observed when MA tasks tested mixed morphology (.49) and derivation (.46), followed by tasks that tested compound (.40) and inflection (.37). These results align with the prior meta-analysis of Lee et al. (2023), which found that derivational morphology was more strongly related to vocabulary than inflectional morphology. Our findings also support previous theories that MA is a multidimensional construct, reflecting various aspects of morphological content (e.g., Kuo & Anderson, 2006; Nagy et al., 2014). A mixed measure that combines different types of morphemes might better capture MA and serve as a stronger correlate of vocabulary.
Of particular interest is the largest effect size observed for MA measures that combined written and oral format (r = .50), followed by oral (r = .43) and written MA format (r = .38). These results indicated that MA measure that combined the oral and written modality was a better correlate of vocabulary knowledge. While this finding seems to contrast with the unidimensionality of MA (Spencer et al., 2015), our results align with the theoretical taxonomy by Deacon et al. (2008). Due to the limited exposure to print, young children’s performances in written tasks are potentially hampered by confounding factors like orthographic knowledge and decoding skills. It is important to note that a large portion of the studies included in our analysis (44 out of 99) involved young children in the early stages of reading. Although the effect of oral-written separation may diminish with age (Tibi & Kirby, 2017), our findings highlight the need to take into account the modality of MA tasks, particularly for younger children.
Finally, we found larger correlational strengths in studies testing receptive and vocabulary breadth than in those testing expressive and vocabulary depth. MA may be particularly beneficial in facilitating the form-meaning linkage in vocabulary acquisition, whereas vocabulary depth entails a rich understanding of lexical entries and may require other word-learning abilities such as lexical inference (e.g., Wang & Zhang, 2024). Moreover, since receptive vocabulary precedes expressive vocabulary (Schmitt, 2019), we could expect that the role of MA plays out earlier in receptive vocabulary. However, findings should be interpreted with caution due to the unequal distribution of some assessment features. In sum, together with previous research (e.g., Cho & McBride, 2022; Ouellette, 2006), our preliminary evidence calls attention to the moderation of both MA and vocabulary assessment-related factors and urges for additional future research to unravel the impact of assessment features on the relationship between MA and vocabulary knowledge.
Cross-linguistic transfer of MA and vocabulary in bilingual readers
Regarding our second and third research questions, our findings supported a small but significant positive association between L1 and L2 MA (r = .33), and between L1 MA and L2 vocabulary (r = .26) in bilingual learners. The moderate correlation between L1 and L2 MA (.33) was similar to the effect size (.30) obtained in a recent meta-analysis (Ke et al., 2021). Our findings also align with previous empirical studies (e.g., Ramirez et al., 2010; Tong et al., 2018; D. Zhang et al., 2017), supporting that MA is a transferable metalinguistic resource that can facilitate MA or vocabulary knowledge in L2. In addition, the homogeneity tests of effect sizes revealed significant variations, and moderator analyses identified four learner- and assessment-related factors that may explain these variations in the transfer of MA across studies.
An unexpected finding was the moderation effect of age. A significant negative effect was observed on the correlation between L1 and L2 MA, suggesting a stronger correlation among younger learners. The mean age of participants in this analysis ranged from 5.76 to 13.5 years. As age may be analyzed as a proxy for L1 proficiency, our findings support the availability of cross-language transfer of MA in learners with varied proficiency in L1, consistent with previous theoretical reviews (e.g., Chung et al., 2019) and empirical studies (e.g., Lam & Chen, 2018; Luo et al., 2014). Furthermore, results suggest that language proficiency seems to weaken the cross-language transfer effect of L1 and L2 MA. Yet, our results contrast with those of the meta-analysis by Ke et al. (2021), which reported a nonsignificant effect of age on the transfer of L1 and L2 MA. This discrepancy may be attributed to the inclusion of additional recent independent samples not considered in their study, as well as our use of a meta-regression approach that treated chronological age as a continuous variable. A similar negative moderating effect of age has been observed in the study by Yang et al. (2017). They reported that the cross-language correlation between L1 and L2 decoding was stronger in Pre-K children than in primary-grade children. They attributed results to the mediation of L1 linguistic skills, indicating that bilinguals who are at the initial state of L2 learning may employ L1 translation equivalents to access the meaning of L2 words (Kroll & Stewart, 1994). However, we did not observe the moderator effect of age on the cross-language relationship between L1 MA and L2 vocabulary. The underlying mechanism in this context remains unclear. Echoing Chung et al. (2019), we call for future work to examine the threshold for transfer effect and clarify the role of language proficiency or automaticity in the cross-language transfer mechanism of MA.
Another noteworthy finding is that language family and script distance significantly moderated the crossover correlation between L1 MA and L2 vocabulary. Results showed higher correlations when L1 and L2 were from the same language family (e.g., both Indo-European) or shared similar scripts (e.g., both alphabetic), compared to L1-L2 pairs with greater language or script distance. In other words, the association between L1 MA and L2 vocabulary was stronger when the language and script distance between L1 and L2 was shorter. One possible explanation is to consider the shared lexical and grammatical knowledge (e.g., cognates) in languages from the same family (Ringbom, 1992). For example, Ramirez et al. (2013) observed a facilitative role of Spanish derivational awareness in the learning of English cognate words. These results align with a previous meta-analysis on similar topics, which found that the transfer effect of L1 vocabulary was influenced by L1 L2 language distance (S. Zhang & Zhang, 2022). However, we did not find significant moderating effects of language distance on the construct-level correlations for MA (i.e., between L1 and L2 MA). Similarly, Ke et al. (2021) reported the nonsignificant moderation of the L1 L2 writing system on the relationship between L1 and L2 MA. They concluded that the transferability of L1 and L2 MA was not subject to linguistic constraints. Of note, while we followed previous meta-analytic studies in classifying language distance (Melby-Lervåg & Lervåg, 2011), their analysis employed a more nuanced criterion that captured eight combinations of writing systems. Yet, more research is needed to clarify how language distance affects the construct-level correlation for MA as well as its crossover effect on L2 vocabulary. Overall, this meta-analysis supports the notion that MA can be transferred to facilitate both L2 MA and vocabulary, and that the crossover correlation between L1 MA and L2 vocabulary is influenced by the L1 L2 language distance (Koda, 2008).
Finally, MA task content (i.e., derivation, inflection, compound) was not found to significantly moderate the relation of MA to L2 vocabulary knowledge or L2 MA. This result is in line with the finding of Ke et al. (2021) that the cross-linguistic sharing of MA between two languages is not conditioned by the linguistic constraint of word formation rules. Yet, we observed a significant moderation of MA response types in the correlation between L1 MA and L2 MA, with the effect size for perception MA tests (.22) significantly weaker than the production (.32) or combined (.34) tests. Evidence also suggested that L2 learners perform differently in production tasks and comprehension tasks in bilingual acquisition (e.g., Bratlie et al., 2022; Gibson et al., 2012). Yet, as we did not examine more nuanced features of MA tasks, we were not clear about whether this finding was induced by the use of different MA measurement tasks (e.g., morphological relatedness and sentence analogy tasks).
Limitations and suggestions for future studies
There are several limitations in the current meta-analysis. First, this study is limited by the availability of studies, some of which did not report detailed assessment characteristics and were therefore excluded from our analysis. Another limitation is that we did not consider more nuanced characteristics of MA assessments that may influence the correlation with vocabulary, such as the type of word stimuli (e.g., real words vs. pseudowords) due to time constraints. In addition, our meta-analytic design focused on correlations, which did not allow us to address issues of causality. In addition, while we employed the shifting unit of analysis approach (Cooper, 2010) to handle multiple correlations, future research could adopt more recent methods, such as multilevel modeling and robust variance estimation, to better address data dependencies (e.g., Liu et al., 2024). Finally, given the complexity of cross-linguistic transfer, future studies may explore how socio-cultural factors, such as educational settings and immigration experience, affect the transfer of MA in bilingual contexts (Chung et al., 2019).
Conclusion and implications
Our findings yielded several important conclusions. First, we found a consistent overall association between MA and vocabulary in monolingual learners with varied linguistic backgrounds. We also found a significant positive construct-level correlation between L1 and L2 MA, as well as a positive crossover correlation between L1 MA and L2 vocabulary in bilingual learners across a wide age range. Results indicated that some of these relations differed by learners’ age, the properties of the languages and writing systems involved, L1–L2 language distance, and some assessment-related variables. This study emphasizes the important role of MA in vocabulary and its cross-language shareability in bilingual contexts. These results also highlight the need for a multifaceted assessment approach that incorporates various types of morphemes, with task modality and response type also deserving consideration, particularly for young children. Researchers are recommended to report more detailed learner and assessment characteristics for a nuanced and precise understanding of how MA is associated with vocabulary knowledge both within- and cross-linguistically.
Supplemental Material
sj-docx-1-ijb-10.1177_13670069241311029 – Supplemental material for Morphological awareness and vocabulary knowledge in monolingual and bilingual learners: A meta-analysis
Supplemental material, sj-docx-1-ijb-10.1177_13670069241311029 for Morphological awareness and vocabulary knowledge in monolingual and bilingual learners: A meta-analysis by Xi Cheng, Li Yin and Haomin Zhang in International Journal of Bilingualism
Footnotes
Availability of data and material
Available upon request.
Code availability
The coding scheme is attached.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was sponsored by the National Social Science Research Grant of China (Grant No. 21CYY047).
Ethical statement and informed consent
The study was exempt from the ethics review in researchers’ institution given that the study did not involve human subjects and that all data sources were anonymized. No individual’s indefinable information was used in the analysis.
Supplemental material
Supplemental material for this article is available online.
Author biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
