1.1 What Is Metalinguistic Awareness?

While it is generally agreed that metalinguistic awareness plays a critical role in language acquisition and development, the definition of metalinguistic awareness varies when researchers examine the link between metalinguistic awareness and X. Both cognitive and developmental scientists who study child first language (L1) development (e.g., Reference GombertGombert, 1992; Reference Tunmer, Pratt and HerrimanTunmer et al., 1984) and applied linguists who study second language (L2) teaching and learning have referred to metalinguistic awareness as explicit or conscious knowledge about language (see a review in Reference Roehr-BrackinRoehr-Brackin, 2018). Researchers who are interested in the cognitive advantages of bilingualism/multilingualism over monolingualism have proposed that metalinguistic awareness involves two components – the analytical ability to reflect upon and manipulate formal properties of language and the attentional control of the mental mechanism that operates language processing (Reference BialystokBialystok, 2001; Reference Bialystok and RyanBialystok & Ryan, 1985). Among researchers of L1 or L2 reading, metalinguistic awareness has been defined as “the ability to reflect on and manipulate the structural features of language” (Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007, p. 53; Reference Nagy and AndersonNagy & Anderson, 1995, p. 2; see also Reference Kuo, Anderson, Koda and ZehlerKuo & Anderson, 2008). Learning to read is fundamentally metalinguistic because learners need to understand how the internal elements of a spoken word relate to units of graphic symbols. Reading in an L2, as compared to reading in an L1, can be even more metalinguistically demanding. Whereas successful adult L2 readers can be well cognizant of sharable metalinguistic resources between two languages and readily apply those resources to facilitate their L2 reading, this process can be a challenging task for children who are learning to read for the first time in an L2.

Metalinguistic awareness, in the context of reading acquisition, is often conceived of as a complex multidimensional and multifaceted construct, involving several related yet distinct components, ranging from segmental understanding (i.e., understanding that words can be segmented into smaller, functionally identifiable units), to structural sensitivity (i.e., the ability to isolate, blend, and combine segmental word information), to functional awareness (i.e., to apply structural understandings for functional purposes such as lexical inferencing)Footnote ¹ (Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Koda, Miller, In and PaeKoda & Miller, 2018; Reference Zhang and KodaZhang & Koda, 2013). Researchers have primarily examined three major facets of metalinguistic awareness: phonological awareness (the ability to reflect upon and manipulate phonological units in a language), orthographic awareness (the ability to form, store, and access orthographic representations of words), and morphological awareness (the ability to reflect on and manipulate the morphemic structures of words). Reading has been theorized as an interactive process of phonology, orthography, and morphology (Reference Plaut, McClelland, Seidenberg and PattersonPlaut et al., 1996; Reference Seidenberg and McClellandSeidenberg & McClelland, 1989). According to Reference PerfettiPerfetti (2003), the universal process of learning to read involves learning how one’s writing system encodes one’s spoken language at various levels, with phonology and morphology at the higher mapping level and inclusive of different languages, and orthography at the lower mapping level entailing more language-specific constraints. In regard to the relationship between metalinguistic awareness and L2 reading development, prior studies have been guided by various reading theories (e.g., the Psycholinguistic Grain Size Theory, Reference Ziegler and GoswamiZiegler & Goswami, 2005; the Repertoire Theory of Literacy Development, Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000; the Structural Sensitivity Theory, Reference Kuo and AndersonKuo & Anderson, 2010; Universal Grammar of Reading, Reference PerfettiPerfetti, 2003) and models/frameworks of transfer (e.g., the Transfer Facilitation Model, Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008; An Interactive Framework of Bilingual Reading Development, Reference Chung, Chen and GevaChung et al., 2019) (see Appendix S2 online: osf.io/4z6mw, which includes the coding of thirteen different frameworks/hypotheses/models/theories cited in the selected studies).

This Element is mainly guided by the Lexical Quality Hypothesis (Reference PerfettiPerfetti, 2007; Reference Perfetti, Hart, Vehoeven, Elbro and ReitsmaPerfetti & Hart, 2002), the Repertoire Theory of Literacy Development (Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000), and the Transfer Facilitation Model (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008). According to the Lexical Quality Hypothesis (Reference PerfettiPerfetti, 2007; Reference Perfetti, Hart, Vehoeven, Elbro and ReitsmaPerfetti & Hart, 2002), successful reading comprehension depends on high-quality lexical knowledge represented by orthography, phonology, morphosyntax, meaning, and the binding of these four features. While the Lexical Quality Hypothesis pertains to the causal relationship between lexical knowledge and reading comprehension and specifies the feature properties such as orthography, phonology, and morphosyntax, it does not directly address how metalinguistic awareness contributes to lexical knowledge or reading ability at the lexical level. In comparison, the Repertoire Theory of Literacy Development (Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000) proposes explicitly that the development of word spelling and reading ability is subject to the application of phonological, orthographic, and morphological awareness, and provides specific guidance for reading instruction. In other words, teachers should not single out any facet of metalinguistic awareness, but rather concurrently promote different facets of metalinguistic awareness. Last but not least, the Transfer Facilitation Model (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008) provides important insights into cross-language transfer in L2 reading by treating metalinguistic awareness as a window of investigation, whereas other theories cited in previous research either were developed for L1 reading purposes only or examined a wide range of transferable reading subskills instead of focusing on metalinguistic awareness.

1.2 Reciprocal Development of Metalinguistic Awareness, Linguistic Knowledge, and Reading Competence

In this Element, we focus on the contributions from metalinguistic awareness to reading-related outcomes (e.g., word decoding, vocabulary knowledge, and reading comprehension) in L2 reading development. It is important, however, to acknowledge that metalinguistic awareness, reading competence (e.g., reading comprehension), and linguistic knowledge (e.g., vocabulary knowledge) (Reference KodaKoda, 2005; Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007; Reference Wagner and MerosWagner & Meros, 2010)Footnote ² are developmentally and reciprocally related. Reference Nagy, Wagner, Muse and TannenbaumNagy (2007) proposed that metalinguistic awareness is the causal mediator between vocabulary knowledge and reading comprehension (see also Reference Zhang and KodaZhang & Koda, 2018). To understand the mechanism underlying this metalinguistic hypothesis, four questions need to be answered: 1) How does metalinguistic awareness support reading comprehension? 2) How does vocabulary knowledge support reading comprehension? 3) How does metalinguistic awareness support vocabulary knowledge? and 4) What are the shared and unique contributions of different facets of metalinguistic awareness to reading development?

How does metalinguistic awareness support reading comprehension? Metalinguistic awareness plays at least two important roles in reading development. At the early stage, metalinguistic awareness enables learners to map the elements of spoken language onto the writing system and is thus fundamentally important for word decoding. Later, when learners develop more refined metalinguistic awareness, including deeper understandings of word-internal phonological, orthographic, and morphological structures, they can apply analytical approaches toward lexical inferencing during reading (Reference KodaKoda, 2005; Reference Nagy, Carlisle and GoodwinNagy et al., 2014). These two distinct yet related roles of metalinguistic awareness in reading development are supported by ample empirical evidence that uncovered indirect contributions of various facets of metalinguistic awareness to L1 or L2 reading comprehension via the mediation of word decoding (e.g., L1 Chinese: Reference Li and WuLi & Wu, 2015; Reference Zhao, Wu, Sun, Xie, Feng and ChenZhao et al., 2019; L1 English: Reference BadianBadian, 2001; Reference Deacon, Kieffer and LarocheDeacon et al., 2014; just to name a few)Footnote ³ or vocabulary knowledge (e.g., L2 Chinese: Reference Ke and KodaKe & Koda, 2019; Reference Leong, Tse, Loh and KiLeong et al., 2011; L2 English: Reference Kieffer, Biancarosa and Mancilla-MartinezKieffer et al., 2013; Reference Nassaji and GevaNassaji & Geva, 1999; Reference Zhang and KodaZhang & Koda, 2012; Reference Zhang and LinZhang & Lin, 2021). Metalinguistic awareness has also been found to contribute directly to reading comprehension. For example, refined morphological awareness in English involves learners’ sensitivity to the distributional properties of derived words, which account for a large proportion of words covered in academic English texts (Reference Nagy and TownsendNagy & Townsend, 2012). This metalinguistic insight helps learners to use the syntactic cues provided by affixes and conduct sentence parsing in word-to-text integration (Reference Levesque, Breadmore and DeaconLevesque et al., 2021; Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007; Reference Perfetti and StafuraPerfetti & Stafura, 2014). Another rationale that supports the direct contribution of metalinguistic awareness to reading comprehension is that metalinguistic awareness is a subset of metacognition, and metacognition is important for reading comprehension (Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007), perhaps more so for L2 learners (Reference Bialystok and RyanBialystok & Ryan, 1985). According to Reference GombertGombert (1992), metacognition refers to one’s awareness and reflections about one’s knowledge, experiences, and learning in general; metalinguistic awareness pertains to the reflections about one’s language use and learning. In previous interventional studies of readers who have a heightened understanding of their own processes for acquiring knowledge (i.e., metacognition) and who are able to consider how and why language is used (i.e., metalinguistic awareness), findings have demonstrated that the introduction of explicit metacognitive and metalinguistic strategy instruction can positively influence reading comprehension (Reference Williams, Atkins, Hacker, Dunlosky and GraesserWilliams & Atkins, 2009; Reference Yuill and McNamaraYuill, 2007; Reference Zipke, Ehri and CairnsZipke et al., 2009). The logic is that the language in written texts is more decontextualized than the language in spoken conversations; as a result, learners need to pay close attention to and analyze the language in written texts, reflect on the analyzed language, and control these cognitive processes during text reading.

How does vocabulary knowledge support reading comprehension? Vocabulary knowledge is often conceptualized to entail vocabulary size/breadth (how many words a learner knows; Reference NationNation, 2001) and vocabulary depth (how well a learner knows the words; Reference Meara, Brown, Malmkjaer and WilliamsMeara, 1996; Reference ReadRead, 1993, Reference Read2000). Both dimensions of vocabulary knowledge directly impact reading comprehension. Previous research indicates a very high vocabulary coverage rate for successful reading comprehension, ranging between 95 percent (Reference Liu and NationLiu & Nation, 1985) and 98 percent (Reference Hu and NationHu & Nation, 2000; Reference NationNation, 2006; Reference Schmitt, Jiang and GrabeSchmitt et al., 2011). This clearly underscores the importance of vocabulary size in reading comprehension. Independent of the effect of vocabulary size, vocabulary depth makes an additional contribution to explaining variation in reading comprehension (e.g., Reference QianQian, 1998; Reference Zhang and YangZhang & Yang, 2016). The direct effect of vocabulary knowledge on reading comprehension has also been supported by a meta-analysis of thirty-seven primary studies on the effects of vocabulary instruction in preK–12th graders whose L1 was English (Reference Elleman, Lindo, Morphy and ComptonElleman et al., 2009). Reference Elleman, Lindo, Morphy and ComptonElleman and colleagues (2009) identified significant, albeit small, improvement in vocabulary and reading comprehension outcomes as a result of vocabulary instruction. In addition, both vocabulary breadth and depth can indirectly contribute to reading comprehension via lexical inferencing while reading (Reference KodaKoda, 2005; Reference NassajiNassaji, 2003, Reference Nassaji2006; Reference Paribakht and WescheParibakht & Wesche, 1999; Reference QianQian, 2005). Successful lexical inferencing, either intentional or unintentional, will subsequently fill the semantic gaps of a learner’s text representation and predict reading comprehension (Reference HatakeyamaHatakeyama, 2012; Reference Koda, Miller, In and PaeKoda & Miller, 2018).

How does metalinguistic awareness support vocabulary knowledge? Metalinguistic awareness and vocabulary knowledge can serve as a bootstrapper for each other. For instance, Reference McBride-Chang, Tardif, Cho, Shu, Fletcher, Stokes, Wong and LeungMcBride-Chang and colleagues (2008) tracked the developmental relationships between morphological awareness and vocabulary knowledge in preschoolers in three languages (Cantonese, Mandarin, and Korean) at two time points, and observed bidirectional bootstrapping effects: At Time 1 where vocabulary knowledge, phonological processing, and reasoning skills were controlled, morphological awareness predicted Time 2 vocabulary knowledge across languages; vocabulary knowledge also predicted subsequent morphological awareness, with Time 1 morphological awareness controlled. As reviewed earlier, morphological awareness and vocabulary knowledge are linked via lexical inferencing during reading comprehension. In other words, the ability to productively combine morphemes is viewed as an important skill in inferring meanings of new words (Reference Wagner and MerosWagner & Meros, 2010). Furthermore, vocabulary-depth knowledge acquired through formal instruction, such as knowledge of word roots, prefixes, and suffixes, serves as the foundation of morphological awareness (e.g., Reference Zhang and KodaZhang & Koda, 2018). One may doubt how the various facets of metalinguistic awareness jointly support vocabulary knowledge development. Regarding oral vocabulary knowledge acquisition, logic would suggest that both phonological and morphological awareness matter because learners need to associate spoken (phonological) forms of words with their meanings (morphemes) (e.g., Reference McBride-Chang, Wagner, Muse, Chow and ShuMcBride-Chang et al., 2005). Notably, orthographic awareness also plays a facilitative role in oral vocabulary knowledge acquisition in alphabetic and nonalphabetic languages as orthography has been found to boost learners’ recall of novel new words verbally over no orthography in previous quasi-experimental studies (e.g., Chinese in Reference Zhang, Lin, Liu and NagyZhang et al., 2020; English in Reference Ricketts, Bishop and NationRicketts et al., 2009). One possible mechanism through which orthographic awareness supports oral vocabulary knowledge development can be explained by the Lexical Quality Hypothesis (Reference PerfettiPerfetti, 2007; Reference Perfetti, Hart, Vehoeven, Elbro and ReitsmaPerfetti & Hart, 2002). Lexical representation that includes phonology, orthography, and semantics is considered of higher quality than a lexical representation that includes only phonological and semantic information (see also Reference Salins, Cupples, Leigh and CastlesSalins et al., 2022). Regarding print vocabulary knowledge learning, according to Reference Apel, Masterson, Hart, Silliman and WilkinsonApel and colleagues (2004), phonological, orthographic, and morphological awareness are integral language components in print vocabulary acquisition measured by spelling. Apel and colleagues’ position is also consistent with stage-based theories that view reading ability as developing from a pre-alphabetic phase to semiphonetic, phonetic, within-word, syllable conjuncture, and derivational constancy phases (Reference Ehri, Snowling and HulmeEhri, 2005; Reference Templeton, Morris, Kamil, Mosenthal, Pearson and BarrTempleton & Morris, 2000). Last but not least, as Reference Nagy, Wagner, Muse and TannenbaumNagy (2007) pointed out, many existing vocabulary knowledge measures are metalinguistic in nature (e.g., word definition skills, Reference KangKang, 2013; Reference Ordóñez, Carlo, Snow and McLaughlinOrdóñez et al., 2002). Therefore, it is not surprising to find a strong association between metalinguistic awareness and vocabulary knowledge.

What are the shared and unique contributions of different facets of metalinguistic awareness to reading development? Phonological, orthographic, and morphological awareness are three interconnected yet distinct facets of metalinguistic awareness. In a recent structural equation modeling study, Reference Tighe, Little, Arrastia-Chisholm, Schatschneider, Diehm, Quinn and EdwardsTighe and colleagues (2019) explored a five-factor model of reading comprehension in struggling adult English readers and the interrelationships among these factors (i.e., phonological awareness, orthographic awareness, morphological awareness, word decoding, and vocabulary knowledge). Tighe and colleagues found that, as separate constructs, none of the metalinguistic skills emerged as uniquely predictive of reading comprehension; yet when the three facets of metalinguistic awareness were loaded to a second-order factor, the second-order factor was found to affect reading comprehension indirectly via word decoding and vocabulary knowledge. Although Tighe and colleagues’ study focused on struggling adult readers instead of child L2 readers, it was among the first to provide empirical evidence supporting readers’ dissociable use of the various facets of metalinguistic awareness for reading comprehension purposes. On the other hand, prior research of English reading comprehension in child and adult readers with or without reading disabilities has identified the unique contribution of morphological awareness to reading comprehension over and above phonological and orthographic awareness (e.g., Reference Apel, Wilson-Fowler, Brimo and PerrinApel et al., 2012; Reference Nagy, Berninger, Abbott, Vaughan and VermeulenNagy et al., 2003; Reference Tighe, Little, Arrastia-Chisholm, Schatschneider, Diehm, Quinn and EdwardsTighe et al., 2019). Similar results have been reported in reading in L2 English (e.g., Reference Bae and JoshiBae & Joshi, 2018) or a nonalphabetic L1 (e.g., Chinese in Reference ZhangZhang, 2017a). As Reference Kirby, Bowers, Cain, Compton and ParrilaKirby and Bowers (2017) put it, morphology is the binding agent of phonological, orthographic, and semantic features of words. There are a few exceptions in L2 reading literature though, suggesting that, out of the three facets of metalinguistic awareness, either phonological awareness or orthographic awareness was among the most important predictors when word decoding instead of reading comprehension was treated as the target reading outcome (e.g., Reference ZhangZhang, 2017b; Reference Zhou, McBride, Leung, Wang, Joshi and FarverZhou et al., 2018). For example, Zhou and colleagues observed that, for Southeast Asian immigrant children learning both English and Chinese as additional languages in Hong Kong primary schools, phonological awareness was the most important factor for word decoding in English, whereas orthographic and morphological awareness were both unique factors of word decoding in Chinese.

In sum, there are multiple direct and indirect paths that connect metalinguistic awareness, reading comprehension, and vocabulary knowledge together. A notable skill that helps to establish the reciprocal relationships between these three competencies (i.e., covarying with metalinguistic awareness, reading comprehension, and vocabulary knowledge) is lexical inferencing (see Reference KodaKoda, 2005; Reference Nagy, Carlisle and GoodwinNagy et al., 2014). In addition, previous research has also found word decoding to modulate the indirect contribution of metalinguistic awareness to reading comprehension (e.g., Reference BadianBadian, 2001; Reference Deacon, Kieffer and LarocheDeacon et al., 2014; Reference Li and WuLi & Wu, 2015; Reference Zhang, Lin, Liu and NagyZhang et al., 2020; Reference Zhao, Wu, Sun, Xie, Feng and ChenZhao et al., 2019). Last but not least, phonological, orthographic, and morphological awareness are the three most-researched facets of metalinguistic awareness in the literature, and they have been found to make both shared and unique contributions to reading development. Therefore, in the scoping review in Section 3, we aim to explore how and to what extent the various facets of metalinguistic awareness are related to word decoding, lexical inferencing, vocabulary knowledge, and reading comprehension, respectively, in prior L2 reading studies. The meta-analysis reported in Section 4, however, only covers the cross-language relationships between metalinguistic awareness and word decoding. Subskills of L2 reading such as lexical inferencing, vocabulary knowledge, and reading comprehension are not included in the meta-analysis. This is because, according to the scoping review results, prior studies that have measured phonological, orthographic, and morphological awareness simultaneously have focused on child L2 readers instead of adult L2 readers. In addition, word decoding was the most studied outcome; vocabulary knowledge and reading comprehension received far less attention; and little child L2 reading research examined lexical inferencing as a reading-related outcome.

1.3 Transfer Facilitation Effects of Metalinguistic Awareness Across Languages

In this Element, transfer is defined as “automatic activation of well-established first language competencies triggered by second language input” (Reference Koda, Koda and ZehlerKoda, 2008, p. 78). Facilitation is the bootstrapping consequence of applying available L1 resources, such as metalinguistic awareness, to L2 reading tasks (Reference Genesee, Geva, Dressler, Kamil, August and ShanahanGenesee et al., 2006; Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008; Reference Riches, Genesee, Genesee, Lindholm-Leary, Saunders and ChristianRiches & Genesee, 2006). Various frameworks have been proposed to inform the understanding of transfer in applied linguistics research in general and in L2 reading research in particular. Examples include:

1. 1. The Contrastive and Typological Framework (Reference LadoLado, 1957), which views transfer as interference in the L2 due to L1 structural properties;

2. 2. The Linguistic Interdependence Hypothesis, which distinguishes between cognitively and conceptually more and less demanding knowledge, and predicts the conditions under which the learner can demonstrate transfer of knowledge crosslinguistically (Reference CumminsCummins, 1979, Reference Cummins1981);

3. 3. The Common Underlying Cognitive Processes Framework, which states that individual differences in L1 and L2 reading skills can be predicted by a common set of underlying cognitive constructs (such as phonological awareness and decoding; e.g., Reference Geva and RyanGeva & Ryan, 1993);

4. 4. The Structural Sensitivity Theory (Reference Kuo and AndersonKuo & Anderson, 2010), which postulates that “having access to two languages renders structural similarities and differences between languages more salient, thus allowing bilingual children to form representations of language structure at a more abstract level” (p. 365);

5. 5. The Transfer Facilitation Model (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008), which was briefly noted earlier.

Notably, Koda’s Transfer Facilitation Model is “the most elaborate theory of transfer to date” (Reference Chung, Chen and GevaChung et al., 2019, p. 158). The model specifies that subskills are transferred in L2 reading acquisition rather than a set of L1 linguistic rules or a holistic construct (like L1 reading ability or L1 proficiency). More importantly, metalinguistic awareness is postulated as an important reading subskill that provides a window for crosslinguistic examinations. The Transfer Facilitation Model also highlights the nonvolitional and automatic nature of transfer, and provides predictions on multiple factors that affect transfer of subskills in L2 reading and the conditions for the transfer, including, for example, the joint influence of L1 metalinguistic awareness sophistication and L2 print input experience, as well as L1–L2 distance, which are in line with the following hypotheses on L2 reading.

1. 1. The Script Dependent Hypothesis (SDH) (Reference Geva and SiegelGeva & Siegel, 2000) suggests that the development of reading component skills is a direct function of orthography transparency; thus, L2 reading efficiency is a direct function of L1 orthography.

2. 2. The Linguistic Coding Differences/Deficit Hypothesis (LCDH) (Reference Sparks, Ganschow and PohlmanSparks et al., 1989, Reference Sparks and GanschowSparks & Ganschow, 1991) indicates that students who do poorly in a foreign language may have language problems in their L1 that interfere with their ability to learn.

3. 3. The Linguistic Threshold Hypothesis (Reference Alderson, Alderson and UrquhartAlderson, 1984) proposes that a threshold level of L2 linguistic knowledge/proficiency is required for L1 reading skills to transfer to L2.

4. 4. The Short-Circuit Hypothesis (Reference ClarkeClarke, 1980) proposes that L2 linguistic knowledge plays a more important role in predicting L2 reading ability development than does L1 reading ability (see also Reference Bernhardt and KamilBernhardt & Kamil, 1995; Reference CarrellCarrell, 1991).

None of the four hypotheses mentioned above, however, attempts to specifically explain the transfer of metalinguistic awareness. Accordingly, no predictions have been made in them regarding to what extent L1 metalinguistic awareness transfers and facilitates L2 reading subskill development and under what conditions transfer facilitation occurs. In contrast, Reference Koda, Koda and ZehlerKoda (2008) provided four pertinent contentions:

1. 1. Shared metalinguistic awareness (e.g., phonological awareness), once developed in one language, is readily available in the early stage of learning to read in another.

2. 2. Language-specific metalinguistic awareness (e.g., orthographic and morphological awareness) reflects the specific ways in which language elements are graphically encoded in the writing system. When transferred, language-specific metalinguistic awareness, closely attuned to L1 properties, promotes the development of corresponding metalinguistic awareness and reading subskills in L2.

3. 3. When transferred, L1 metalinguistic awareness competencies, reflecting L1 properties, are adjusted through print experience in L2. The degree of adjustment as well as the amount of L2 print experience are influenced by L1–L2 distance (i.e., how closely related L1 and L2 are).

4. 4. The resulting L2 metalinguistic awareness and L2 reading subskills vary systematically in learners with diverse L1 backgrounds.

Since the Transfer Facilitation Model was first proposed by Reference KodaKoda (2005), it has been widely tested in empirical studies (for reviews, see Reference Koda and KeKoda & Ke, 2018; Reference Koda and ReddyKoda & Reddy, 2008). The central tenet of the model, which underscores a facilitative transfer effect of L1 metalinguistic awareness, is consistent with the broad conceptualizations of L1 providing resources for L2 reading (Reference Genesee, Geva, Dressler, Kamil, August and ShanahanGenesee et al., 2006). In addition, the association between L1 and L2 metalinguistic awareness as predicted by the model has been supported by substantial empirical studies on child or adult L2 readers across different linguistic and educational settings (e.g., American university learners of Chinese as a foreign language, Reference Ke and KodaKe & Koda, 2017; American university learners of Chinese as a heritage language, Reference Zhang and KodaZhang & Koda, 2021; Chinese heritage children in the United States, Reference Koda, Lü, Zhang, Chen, Wang and LuoKoda et al., 2014; college-level Chinese students who learned English in the United States, Reference Li and KodaLi & Koda, 2022; Chinese children reading English as a foreign language in mainland China, Reference Zhang and KodaZhang & Koda, 2013; Japanese university learners of English as a foreign language, Reference Koda, Miller, In and PaeKoda & Miller, 2018; Kanada-speaking children learning to read English in India, Reference Reddy and KodaReddy & Koda, 2013; and multilingual children in Singapore, Reference ZhangZhang, 2016; Reference Zhang and KeZhang & Ke, 2019), as well as a small number of meta-analytic studies (e.g., Reference Jeon and YamashitaJeon & Yamashita, 2014; Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011). Consequently, we have adopted this model to guide our reviews and meta-analysis in the rest of this Element.

In what follows, we first review the findings of five previous meta-analytic studies that are related to (but nonetheless distinct from) our meta-analysis (Section 2). We then provide a scoping review of empirical studies on metalinguistic awareness and L2 reading (Section 3). In Section 4, we present our meta-analysis of intralingual and interlingual correlations of phonological, orthographic, and morphological awareness with word decoding in L2 readers.

2.1 Review Inclusion Criteria

Five meta-analytic studies (i.e., Reference Jeon and YamashitaJeon & Yamashita, 2014; Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011; Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021; Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018) have been included in the review in this section (as shown in Table 1) because they (1) examined the correlation between at least one facet of metalinguistic awareness (i.e., phonological, orthographic, or morphological awareness) and one reading-related outcome (e.g., word decoding, vocabulary knowledge, or reading comprehension) and (2) either focused on bilingual readers (Reference Jeon and YamashitaJeon & Yamashita, 2014; Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011) or both monolingual and bilingual readers (Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021), or adopted a crosslinguistic perspective comparing correlational relationship(s) between two languages (e.g., Chinese and English in Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018).Footnote ⁴ We tabulated and compared the findings of these five studies. We noted that a few other meta-analyses reported Cohen’s d (mean differences) rather than correlation coefficient r, such as Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie and colleagues (2022) and Reference Goodwin and AhnGoodwin and Ahn (2010, Reference Goodwin and Ahn2013). In addition, Reference Chung, Chen and GevaChung and colleagues (2019) was a critical descriptive review rather than a meta-analysis. The findings of these meta-analyses and critical review were not directly compared. Nevertheless, we consider them when discussing potential moderators of interest below.

Among the five meta-analyses included in the review below, notably four focused on child bilingual readers and only one involved adult L2 readers in their sample populations (i.e., Reference Jeon and YamashitaJeon & Yamashita, 2014). Jeon and Yamashita was also the only meta-analysis reporting a pertinent correlation between L2 orthographic awareness and an L2 reading outcome (i.e., reading comprehension),Footnote ⁵ whereas all of the other four meta-analyses reported correlations between phonological or morphological awareness and a reading outcome (word decoding or reading comprehension). None of the five meta-analyses examined the relationship between any facet of metalinguistic awareness and vocabulary knowledge or lexical inferencing.

2.2 Evidence Supporting Cross-Language Transfer Facilitation of Metalinguistic Awareness

Table 1 summarizes the findings from previous meta-analyses of metalinguistic awareness and reading development. In the table, we present the findings on correlations between L1 and L2 metalinguistic awareness, intralingual correlations between metalinguistic awareness and reading outcomes, as well as interlingual correlations of L1 metalinguistic awareness with L2 reading outcomes. Our interpretation of effect sizes (rs) is based on Reference Plonsky and OswaldPlonsky and Oswald’s (2014) estimates for small, medium, and large correlations (r = .25, .40, .60, respectively).

In regard to the extent to which metalinguistic awareness in L1 and L2 are related, as shown in the upper panel of Table 1, Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg’s (2011) meta-analysis of phonological awareness and L2 reading development suggested a large correlation between L1 and L2 phonological awareness (r = .60), while Reference Ke, Miller, Zhang and KodaKe and colleagues’ (2021) meta-analysis on morphological awareness identified a small correlation between L1 and L2 morphological awareness (r = .30). The finding that the two facets of metalinguistic awareness correlated to different extents in two languages (i.e., higher for phonological awareness and lower for morphological awareness) is consistent with Reference Koda, Lü, Zhang, Chen, Wang and LuoKoda and colleagues’ (2014) position that phonological awareness is a language-general and shared resource across languages (see also Reference Geva and RyanGeva & Ryan, 1993), whereas morphological awareness is a language-specific and shareable resource. To the best of our knowledge, there is no meta-analytic evidence in the literature on cross-language correlations for orthographic awareness. Reference Chung, Chen and GevaChung and colleagues’ (2019) critical review, however, provided an insight into this issue: L1 and L2 orthographic awarenessFootnote ⁶ were significantly correlated between two Roman alphabets (e.g., English–French: Reference Chung, Chen and DeaconChung et al., 2018; Reference Commissaire, Duncan and CasalisCommissaire et al., 2011; Reference Deacon, Wade‐Woolley and KirbyDeacon et al., 2009; Reference Pasquarella, Deacon, Chen and Au-YeungPasquarella et al., 2014; English–Spanish: Reference Sun-Alperin and WangSun-Alperin & Wang, 2011); yet few studies supported any significant correlation of orthographic awareness between different L1 and L2 scripts (e.g., Chinese–English: Reference Gottardo, Yan, Siegel and Wade–WoolleyGottardo et al., 2001; Russian–English: Reference Abu-RabiaAbu-Rabia, 2001).

As the middle panels of Table 1 illustrate, the relative correlations between L1 phonological awareness and L1 reading-related outcomes (i.e., word decoding and reading comprehension) and those between L1 morphological awareness and L1 reading outcomes varied in previous meta-analyses, subject to the influence of the target language (Chinese and English: Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018; Spanish: Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021). The correlations between phonological awareness and reading-related outcomes in L1 Chinese were all small, whereas the magnitudes were moderate in L1 English and ranged from small to moderate in L1 Spanish. In addition, the correlations between morphological awareness and reading-related outcomes in L1 Chinese were higher than those between phonological awareness and reading in L1 Chinese, but still fell within the small range. The correlations between morphological awareness and reading-related outcomes in L1 English, however, were moderate. Regarding intralingual correlations between L2 metalinguistic awareness and L2 reading-related outcomes, regardless of the facet of metalinguistic awareness or reading outcomes under examination, the evidence was consistent across the previous meta-analyses; that is, the correlations were significant and moderate. Taken together, these findings on intralingual correlations demonstrated less variation in the relationship between metalinguistic awareness and reading outcomes in L2 than in L1.

Finally, as shown in the bottom panel of Table 1, interlingually speaking, the findings are consistent across previous meta-analyses – the correlations between L1 metalinguistic (phonological or morphological) awareness and L2 reading-related outcomes (word decoding or reading comprehension) were all small yet significant.

2.3 Evidence Related to Factors Impacting the Transfer Facilitation of Metalinguistic Awareness in L2 Reading

Table 1 also summarizes the (non)significant moderating effects. One of the theory-driven factors is L1–L2 distance, which was, for example, operationalized as L1–L2 writing system type in Reference Ke, Miller, Zhang and KodaKe and colleagues (2021) and Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg (2021) (see also Section 1.3). No significant moderating effect, however, was identified for this factor. This finding suggests that, regardless of L1–L2 distance, L1 metalinguistic awareness not only transfers at the construct level (i.e., significant correlations between corresponding L1 and L2 metalinguistic awareness), but also has crossover facilitation effects on L2 reading-related outcomes (i.e., significant correlations of L1 metalinguistic awareness with L2 reading abilities).

Proficiency in L2 is another theory-driven factor related to L2 print input experience (Reference Chung, Chen and GevaChung et al., 2019; Reference KodaKoda, 2005, Reference Koda2007, Reference Koda, Koda and Zehler2008) that may impact transfer of metalinguistic awareness (see relevant discussion in Section 1.3). Nevertheless, it was not examined in the selected meta-analyses. Although Reference Jeon and YamashitaJeon and Yamashita (2014) did consider the influence of L2 proficiency on the relationship between L1 and L2 reading comprehension, they did not conduct any moderator analysis investigating its impact on the relationship between metalinguistic awareness and reading in L2. Pertinent and indirect evidence of the influence of L2 proficiency may be found in the recent meta-analysis by Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie and colleagues (2022). Bratlie and colleagues compared morphological knowledge between language minority and language majority children, and found that there were moderate to large differences between the two groups. As part of their moderator analysis, the authors found that some continuous variables such as vocabulary, syntactic knowledge, and reading comprehension significantly predicted the group difference. In addition, Bratlie and colleagues did not find any significant differences in morphological knowledge between Chinese L1 children acquiring English, Spanish L1 children acquiring English, and Turkish L1 children acquiring Dutch. The significant gap in language minority children’s morphological knowledge in the societal language or L2 might be attributed to their relatively low proficiency in that language because, as discussed in Section 1.3, a threshold of L2 linguistic knowledge or proficiency may need to be passed for language minority children to benefit from their L1 reading subskills such as metalinguistic awareness.

In addition to theory-driven factors, construct-, learner-, measurement-, and instruction-related factors must also be considered (Reference Chung, Chen and GevaChung et al., 2019; Reference Gottardo, Chen and HuoGottardo et al., 2021; Reference Ke, Miller, Zhang and KodaKe et al., 2021). At the construct level, metalinguistic awareness consists of three major components – phonological, orthographic, and morphological awareness – each being a multidimensional/multifaceted construct itself. Phonological awareness can be further categorized into phonemic awareness, intrasyllabic awareness, and syllabic awareness based on different linguistic grain sizes (Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin et al., 2012; Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021; Reference Ziegler and GoswamiZiegler & Goswami, 2005, Reference Ziegler and Goswami2006). In a meta-analysis of Spanish phonological awareness in Spanish monolinguals and bilinguals whose L1 is mainly English, Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez and colleagues (2021) observed the largest correlation for syllabic awareness. In comparison, in a meta-analysis of bilingual readers of mixed language backgrounds, Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin and colleagues (2012) identified greater cross-language correlations in phonological awareness tasks that measured multiple grain sizes than those that measured syllabic awareness alone. Branum-Martin and colleagues speculated that phonological awareness manipulating multiple grain sizes may involve processes that are more highly related across languages.

The findings about cross-language transfer of orthographic awareness are inclusive. Evidence that supports the shareability of orthographic awareness in two languages was mainly based on lexical orthographic processing in two alphabetic scripts (see a review by Reference Chung, Chen and GevaChung et al., 2019). However, in primary studies of both sublexical and lexical orthographic awareness in two typologically different scripts (e.g., Chinese and English in Reference Wang, Yang and ChengWang et al., 2009, Reference Wang, Lin, Yang, Chen, Wang and Luo2014), there was no significant positive correlation or the correlation was even negative between L1 and L2 orthographic awareness.

Lastly, for the conceptualization of morphological awareness, there have been two different approaches. One approach adopts a continuum view of morphological awareness as more language independent or more language specific. A language-independent facet of morphology involves segmentation of words into sublexical components (referred to as a basic facet by Reference Zhang and KodaZhang & Koda, 2013 or structural awareness by Reference Zhang and KodaZhang & Koda, 2018), whereas a language-specific facet involves mapping morphological information onto graphic symbols (referred to as a “refined facet” by Reference Zhang and KodaZhang & Koda, 2013 or “functional awareness” by Reference Zhang and KodaZhang & Koda, 2018). The other approach is based on word formation rules and categorizes morphological awareness into inflectional, derivational, and compound awareness (e.g., Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie et al., 2022). In a meta-analysis that compared morphological knowledge in language minority and language majority children (mainly readers of English), Bratlie and colleagues observed larger variances in primary studies that measured inflectional awareness than in those that measured derivational or compounding awareness. However, Reference Ke, Miller, Zhang and KodaKe and colleagues’ (2021) meta-analysis of morphological awareness in bilingual child readers did not identify any significant moderating effect of word formation rules.

At the learner level, an often-examined factor was age/grade level. Previous meta-analyses either operationalized factors of this type as lower versus upper grade level in subgroup analyses (Reference Goodwin and AhnGoodwin & Ahn, 2013; Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011) or treated age as a continuous variable in meta-regression analyses (e.g., Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin et al., 2012). No significant age effect was found for the transfer of phonological awareness between L1 and L2 (e.g., Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin et al., 2012; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011). However, significant moderating effects of age/grade level were found in meta-analyses of morphological awareness. For example, Reference Goodwin and AhnGoodwin and Ahn’s (2013) meta-analysis investigated the effects of English morphological intervention in readers of mixed backgrounds (e.g., English-only learners, English language learners, and children with learning disabilities), and found that morphological intervention led to significant improvement in morphological knowledge and reading-related outcomes in a broad range of grade levels from kindergarten to middle school levels. Furthermore, Reference Ke, Miller, Zhang and KodaKe and colleagues (2021) identified a bigger correlation between L2 morphological awareness and L2 reading comprehension in children at the upper elementary grades than children between kindergarten and grade two.

Regarding measurement-related factors, it is generally agreed in the literature that how metalinguistic awareness and reading-related outcomes are measured contributes significantly to the variation in primary studies. Recent studies found that 1) complex phonological awareness tasks correlated more strongly with different reading-related outcomes (Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018); 2) phonological awareness correlated more strongly with nonword reading than with real word reading in bilingual readers (Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021); and 3) a larger variation existed in studies examining phonological awareness and word reading accuracy than those on phonological awareness and word reading fluency (Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018). For morphological awareness, larger correlations were observed for studies that measured oral and expressive/productive morphological awareness than those measuring written and receptive/judgmental morphological awareness (Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie et al., 2022; Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018). Reference Ke, Miller, Zhang and KodaKe and colleagues (2021) identified eight specific types of morphological awareness measured in the literature and found a larger correlation for studies that used the morphological structure task (e.g., Reference McBride-Chang, Wagner, Muse, Chow and ShuMcBride-Chang et al., 2005) than those using the Test of Morphological Structure (Reference CarlisleCarlisle, 2000). For orthographic awareness, due to the small number of primary studies, moderator analysis has barely been conducted in previous meta-analyses. Reference Georgiou, Martinez, Vieira and GuoGeorgiou and colleagues’ (2021) meta-analysis on dyslexic and typical readers (mostly English readers) found no moderating effect of orthographic awareness measurement (i.e., accuracy vs. response time measures). No meta-analysis, however, has been conducted in L2 readers where any moderating measurement effect was tested for orthographic awareness.

Finally, instruction-related factors (e.g., medium of instruction) seems to have been underexamined in previous meta-analyses (except Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011). Melby-Lervåg and Lervåg did not find any significant effect of medium of instruction on the relationship between L1 phonological awareness and L2 word decoding. The overall lack of attention to instruction-related factors could be partially due to the confounding effects of simultaneously examining multiple factors, either intentionally or unintentionally (Reference Gottardo, Chen and HuoGottardo et al., 2021). For example, L2 as the medium of instruction can be applied to different societal settings (e.g., L2 as the societal language in English-as-a-second-language/ESL classrooms with learners of lower socioeconomic status; or L2 as the non-societal language in dual-language immersion programs with learners of higher socioeconomic status).

To sum up, previous meta-analytic reviews of metalinguistic awareness and reading development have mainly focused on phonological and morphological awareness and their within- and cross-language contributions to reading in child bilingual readers. Much less attention has been paid to adult L2 readers. Reading-related outcomes have largely focused on word decoding. There has been little attention to the correlations of metalinguistic awareness with vocabulary knowledge, lexical inferencing, or reading comprehension. Also, while the evidence seems to converge on the shareability of phonological and morphological awareness between two languages, the shareability of orthographic awareness has not been systematically examined. Overall, it remains unclear the extent to which different facets of metalinguistic awareness correlate with different reading-related outcomes intralingually and interlingually. Lastly, inconclusive evidence can be drawn regarding significant factors moderating the within- and cross-language correlations between metalinguistic awareness and L2 reading.

3.1 Literature Search, Inclusion and Exclusion Criteria, and Coding

Literature Search. We conducted a three-step literature search (see Figure 1). First, our search prioritized research exploring orthographic awareness in L2 reading because orthographic awareness has received much less attention than phonological or morphological awareness in previous meta-analyses and other reviews, as indicated in Section 2. Two sets of keywords, including “orthographic knowledge” OR “orthographic awareness” OR “orthographic processing” OR “orthographic skill” OR “ortho-” AND “second language” OR “L2” OR “bilingual” OR “L3” OR “third language” OR “multilingual” OR “additional language” were used in Boolean searches in three electronic databases: ProQuest (including three subdatabases, namely, ERIC, LLBA, and ProQuest Dissertations and Global Theses), PsycINFO, and Web of Science. The database search was first conducted in December 2021 with another search in January 2022. Second, we examined the references of twenty-four meta-analytic and critical reviews of phonological awareness, orthographic awareness, or morphological awareness (see the list in Appendix A), looking for studies that simultaneously measured phonological, orthographic, and morphological awareness in L2 readers. Third, manual searches were conducted among six scholarly journals: Applied Psycholinguistics; Journal of Research in Reading; Language Learning; Reading and Writing; Reading Research Quarterly, and Scientific Studies of Reading. As of January 31, 2022, a total of 2,245 studies, including duplicated reports, were identified. After screening using a priori criteria described below, sixteen studies (N = 2,654) with nineteen independent samples were included for subsequent analysis. The complete list of the included studies can be found in Appendix B. From these search results, the first research trend we noticed is that the number of L2 reading studies simultaneously measuring all three facets of metalinguistic awareness is considerably smaller than the number of those examining one or two facets, as revealed in previous meta-analyses. More specifically, forty-seven studies were identified for phonological awareness by Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg’s (2011) meta-analysis and thirty-four studies were found for morphological awareness by Reference Ke, Miller, Zhang and KodaKe and colleagues’ (2021) meta-analysis.

Figure 1 Literature search.

Inclusion and Exclusion Criteria. Primary studies included for review in this Element met five criteria: 1) they were written in English between 1990 and 2021; 2) they included participants without a learning or reading disability; 3) they presented empirical data based on direct testing of phonological, orthographic, AND morphological awareness either in L1 or L2; 4) they reported data based on direct testing of at least one L2 measure, such as metalinguistic awareness, word decoding, vocabulary knowledge, lexical inferencing, or reading comprehension; and 5) they reported sample size and data for calculating effect sizes, including the correlation coefficient r and/or descriptive statistics with means and standard deviations. We excluded duplication reports as well as studies on L1 or monolingual readers only.

Coding. Detailed procedures of coding effect sizes and moderators are reported in Section 4.2. The complete dataset as well as the coding scheme (as shown in Appendix S1) is available online: osf.io/4z6mw. The first author coded the data twice. For effect size data, the intracoder reliability was .98. For categorical variables and qualitative coding, the first author resolved any inconsistencies via a third check.

3.2 An Overview of the Research Methodologies in the Selected Studies

The meta-analysis reported in Section 4 included sixteen correlational studies published between 2009 and 2021 with data collected from nineteen independent samples of child L2 readers (N = 2,654) in seven major locations (i.e., Canada, Hong Kong SAR China, Israel, mainland China, Singapore, South Korea, and the United States).Footnote ⁷ None of the studies included adult L2 participants. All studies were of a cross-sectional and observational design, except Reference LuoLuo (2013), which adopted a longitudinal design.

Table 2 illustrates the number of independent study samples and mean correlations (r) for the relationships between the various facets of metalinguistic awareness and different reading-related outcomes. These outcomes included word decoding, reading comprehension, and vocabulary knowledge. More specifically, the results indicated that most of the included L2 child reading studies had focused on the associations between metalinguistic awareness and word decoding, followed by the associations between L1 and L2 metalinguistic awareness. Very limited attention has been paid to the relationships between metalinguistic awareness and reading comprehension or vocabulary knowledge. No selected studies examined the effects of metalinguistic awareness on lexical inferencing.

When sample participants’ L1 and L2 were both considered, a total of twelve languages (i.e., Arabic, Chinese, English, French, Hebrew, Hindi, Indonesian, Korean, Nepali, Russian, Tagalog, and Urdu) and four writing systems (i.e., abjad, alphabet, alphasyllabary, and morphosyllabary) were represented. Out of the nineteen independent samples, there were seven types of L1–L2 combinations: L1 English–L2 Chinese (k = 2); L1 Arabic–L2 English (k = 1); L1 Chinese–L2 English (k = 8); L1 Korean–L2 English (k = 3); L1 Hebrew/Russian–L2 English (k = 2); L1 South Asian languages–L2 English (k = 2)Footnote ⁸; and L1 English–L2 French (k = 1). Grades ranged from kindergarten through grade eight, with a mean age of 109 months. Eleven study samples included participants of the language majority group, seven included the minority group, and one sample contained mixed groups. Regarding medium of instruction, five samples included school instruction in the target L2.

3.3 Study Quality Evaluation

To evaluate the quality of the selected studies (after Reference Marsden, Morgan-Short, Thompson and AbugaberMarsden et al., 2018), we have summarized the internal reliabilities (based on Cronbach’s alpha) of pertinent measures reported in primary studies in Table 3. All primary studies reported the internal reliability (Cronbach’s alpha) for at least one instrument. Orthographic awareness tasks seemed slightly less reliable than phonological and morphological awareness tasks, while word decoding tasks appeared to be more reliable than reading comprehension and vocabulary knowledge tasks. In general, the mean reliability (except L2 word decoding tasks, mean Cronbach’s alpha = .93) seemed to be lower than that reported in previous second language acquisition research (the median reliability of second language acquisition research focusing on reading skills was .86 according to Reference Plonsky and DerrickPlonsky & Derrick, 2016). According to Plonsky and Derrick, the 25th, 50th, and 75th percentiles of instrument reliability coefficients for second language acquisition research in general were 74, .82, and .89, respectively. Thus, the mean task reliabilities of the selected studies fall between the 25th and 50th percentiles.

In addition, following Reference Hohn, Slaney and TafreshiHohn and colleagues’ (2019) assessment protocols, we coded and evaluated a total of eleven study quality indices. These indices included the adequacy of conclusions made in the primary study; study design; information found in literature reviews such as theoretical frameworks, the measures used, missing data, and attrition rates; power considerations; the publication status of the primary studies; statistical analyses used; sample-size considerations; reliability estimates; and validity concerns. Due to limitations of space, the complete coding is presented in Appendix S2 online: osf.io/4z6mw.

3.4 Definitions and Measures of Metalinguistic Awareness in the Selected Studies

To recapitulate, just as metalinguistic awareness has been defined in different ways in the literature, there have been different definitions and measures of the three major facets of metalinguistic awareness. This section summarizes the various operationalizations for each facet in the selected studies. Since the number and complexity of metalinguistic awareness measures is a potential moderator on the relationship between metalinguistic awareness and L2 reading (see the meta-analytic results in Section 4), the measures reviewed in this section can also provide methodological implications for future research. Complete coding can be found in Appendix S3 online: osf.io/4z6mw.

Definitions and Measures of Phonological Awareness. Only eleven of the nineteen independent study samples provided an explicit definition of phonological awareness. These definitions can be categorized into four major types:

1. 1. The ability to perceive and manipulate sound units of spoken language (Reference Goswami and BryantGoswami & Bryant, 1990, as cited in Reference Wang, Yang and ChengWang et al., 2009).

2. 2. The ability to identify, isolate, or delete phonemes from a word (Reference RussakRussak, 2020).

3. 3. The ability to process phonemes is typically called phonemic awareness, whereas phonological awareness is a general term referring to the ability to manipulate several units of spoken languages, such as syllables or onsets and rimes in addition to phonemes (Reference Bae and JoshiBae & Joshi, 2018).

4. 4. The sensitivity to the segmental as well as suprasegmental features of Chinese phonology (Reference Halliday, Asher and HendersonHalliday, 1981, as cited in Reference ZhangZhang, 2017b).

The first definition is a generic definition referring to a learner’s awareness of the sound structure of a language; the second type is specifically about phonemic awareness; the third definition, more broadly speaking, includes phonemic, intrasyllabic, and syllabic awareness; and the last is language specific and includes a focus on Chinese tone (a suprasegmental feature) awareness. Influenced by the various types of definitions, phonological awareness measures were operationalized at different levels in the selected studies: at the phoneme level only (k = 6), at phoneme and intrasyllable levels (k = 3), at phoneme and syllable levels (k = 3), at intrasyllable and syllable levels (k = 4), at intrasyllable and suprasegmental levels (k = 2), and at the syllable level only (k = 1).

There were four approaches to measuring phonological awareness:

1. 1. Deletion (e.g., “Say ‘three’ without the sound /r/.” See Reference Abu-Rabia and SanitskyAbu-Rabia & Sanitsky, 2010);

2. 2. Segmentation and counting (e.g., “Identify the number of speech sounds in the word ‘cat.’” /k/, /a/, /t/. See Reference Bae and JoshiBae & Joshi, 2018);

3. 3. Categorization/odd-one-out (e.g., “Choose which one of the three syllables does not share either the onset, rime, or tone with the other two syllables.” See Reference Wang, Yang and ChengWang et al., 2009);

4. 4. Tone discrimination (e.g., “Identify the third syllable [/jiàn/] that differs in tone from the other two syllables [/kě/ and /lěng/].” See Reference ZhangZhang, 2017b).

The selected studies commonly asked participants to respond orally or provide answers to multiple-choice questions in paper-and-pencil packets after listening. Notably, the four approaches reviewed above do not represent an exhaustive list of phonological awareness measures in that a number of measures described in previous reviews (e.g., Reference McBride-ChangMcBride-Chang, 1995; Reference Schatschneider, Francis, Foorman, Fletcher and MehtaSchatschneider et al., 1999; Reference Stahl and MurrayStahl & Murray, 1994) were not used in the selected studies, including, for example:

1. 1. (First-) Sound to word matching (e.g., “Does fish begin with /f/?”);

2. 2. Blending (“What does /f-i-sh/” say?”);

3. 3. Position analysis (e.g., “Say ‘nelf.’ Now what sound comes after the /l/ in this nonsense word?”).

Definitions and Measures of Orthographic Awareness. Reference Apel, Henbest and MastersonApel and colleagues (2019) provided a comprehensive review of orthographic knowledge at two levels (i.e., sublexical and lexical) and emphasized the differences between measuring implicit orthographic knowledge and explicit orthographic awareness. Lexical orthographic knowledge refers to the stored mental representations of known words or word parts; sublexical orthographic knowledge refers to the rules, positions, or patterns for a letter or letter concatenations (in English, for example) (see also Reference ApelApel, 2011). Implicit orthographic knowledge is captured in a setting where learners are not overtly asked to think about or consider what they know about orthography (e.g., spontaneous spelling). In contrast, explicit orthographic awareness requires learners to actively think about or reflect on their knowledge of orthography. A commonly used lexical orthographic awareness task is orthographic word choice, in which participants are asked to choose the correct spelling of a real word while facing the actual correct spelling of the real word and a pseudoword with an incorrect spelling (e.g., “rain–rane”). An often-cited sublexical orthographic awareness task is an orthographic word-likeness task, in which participants are asked to choose from a pair of pseudowords, one following legal orthographic conventions and the other violating orthographic conventions (e.g., “gool–giil”). In the nineteen independent study samples, eleven provided pertinent definitions of orthographic awareness, and two definitions seemed to be aligned with Reference Apel, Henbest and MastersonApel and colleagues’ (2019) conceptualization of orthographic awareness at lexical and sublexical levels, which orthographic processing tasks have traditionally tapped into:

1. 1. Individual word-specific representations or letter patterns that occur across many words (e.g., Reference Deacon, Wade‐Woolley and KirbyDeacon et al., 2009);

2. 2. One’s knowledge of word spelling regularities and permissible letter sequences in a word (e.g., Reference Zhao, Joshi, Dixon and ChenZhao et al., 2017).

The coding results indicated that three samples measured both lexical and sublexical orthographic awareness, five samples measured lexical orthographic awareness only, and eleven samples measured sublexical orthographic awareness only.

Again, there were definitions specific to the Chinese language (e.g., students’ visual–orthographic sensitivity to Chinese characters and their awareness of the representational functions of the semantic radicals, Reference Wong and ZhouWong & Zhou, 2021; or knowledge about Chinese radicals and the sensitivity to their spatial configurations, Reference ZhangZhang, 2017b). In the selected studies on readers of Chinese, sublexical orthographic awareness was captured with tasks consisting of items on strokes, radicals, and individual characters, while lexical orthographic awareness was measured with tasks of two-character words.

Definitions and Measures of Morphological Awareness. Not all study samples provided a clear definition of morphological awareness; only eleven out of nineteen did so, mostly following Reference Carlisle and FeldmanCarlisle’s (1995) general definition of morphological awareness as “conscious awareness of the morphemic structure of words and [the] ability to reflect on and manipulate that structure” (p. 194). A language-specific operational definition was also included in some studies. For instance, in Reference ZhangZhang’s (2017a) study, morphological awareness was defined as learners’ sensitivity to homography and morphological compounding, which are both prevalent in Chinese.

In previous meta-analytic research (e.g., Reference Ke, Miller, Zhang and KodaKe et al., 2021), eight different morphological awareness measures have been identified: 1) affix choice (after Reference Singson, Mahony and MannSingson et al., 2000); 2) morphological structure awareness (e.g., Reference McBride-Chang, Wagner, Muse, Chow and ShuMcBride-Chang et al., 2005); 3) morphological decomposition (e.g., Reference Saiegh-Haddad and GevaSaiegh-Haddad & Geva, 2008); 4) morphological relatedness (e.g., Reference Saiegh-Haddad and GevaSaiegh-Haddad & Geva, 2008); 5) riddle guess (e.g., Reference Berninger and NagyBerninger & Nagy, 1999); 6) sentence analogy (after Reference Nunes, Bryant and BindmanNunes et al., 1997); 7) the Test of Morphological Structure (after Reference CarlisleCarlisle, 2000), and 8) the Wug Test (e.g., Reference Eviatar, Taha and ShwartzEviatar et al., 2018). We produced similar findings based on sixteen of the selected studies for the present meta-analysis. In addition, a Chinese-specific measure (i.e., homophone/homograph discrimination) was identified (e.g., in Reference ZhangZhang, 2017b). For that measure, participants were presented with groups of three words of two characters that shared a homograph. For each group, they were asked to identify the word where the meaning of the target character was different from that in the other two words. For example, in both “商业, /shāngyè/, business” and “商品, /shāngpǐn/, product,” “商, /shāng/” means “commerce”; however, in “商量, /shāngliáng/, discuss,” “商/shāng/” means “discuss” or “consult.”

There was great variation in how the selected studies operationalized the construct of morphological awareness. Take word formation rules, for example. Two independent study samples included inflected words only; two included derived words only; eight included compound words only; one included inflected and derived words; two included derived and compound words; one included inflected, derived, and compound words; and three included compound words and homophones/homographs.

The selected studies also varied in the modality/modalities used for measuring morphological awareness. Unlike phonological awareness measures that mainly assessed spoken languages only, morphological awareness measures often involve a combination of spoken and written modalities. This is not surprising because morphological awareness is considered as a binding agent that strengthens the links between the orthographic, phonological, and meaning representation of morphemes and words (Reference Kirby, Bowers, Cain, Compton and ParrilaKirby & Bowers, 2017; Reference Nagy, Carlisle and GoodwinNagy et al., 2014). Reference Kuo, Anderson, Koda and ZehlerKuo and Anderson (2008) termed morphological awareness measured in the written modality as “grapho-morphological awareness,” namely “the ability to reflect upon how semantic information is encoded in the orthography and how orthography provides cues to meaning” (p. 54) (see also Reference Chen, Ke and KodaChen et al., 2021). Other researchers differentiated oral morphological awareness from written morphological awareness and identified them as “preliterate” versus “postliterate” morphological awareness (e.g., Reference Pan, Song, Su, McBride, Liu, Zhang and ShuPan et al., 2016). In the nineteen independent study samples, six samples measured morphological awareness using oral modality, two were in the written modality, three were based on mixed modalities, and eight samples did not report the measurement modality.

3.5 Summary

To reiterate, a total of sixteen studies or nineteen independent study samples that considered all three facets of metalinguistic awareness (i.e., phonological, orthographic, and morphological awareness) were selected for the scoping review and subsequent meta-analysis. The sample participants were K–8 bilingual children (N = 2,654) of twelve languages and four writing systems when both L1 and L2 were considered. The majority of the samples included English as the target L2. The selected studies were conducted in seven major locations (i.e., Canada, Hong Kong SAR China, Israel, mainland China, Singapore, South Korea, and the United States). Only one study adopted a longitudinal design; the rest were observational and cross-sectional.

The selected L2 reading studies predominantly examined the relationships between metalinguistic awareness and word decoding in L1 and L2. They did not include sufficient evidence for further meta-analysis of other reading-related outcomes, such as vocabulary knowledge, lexical inferencing, and reading comprehension.

We also reviewed the definitions and measures of the three facets of metalinguistic awareness. Only about half (i.e., eleven) of the nineteen independent study samples provided clear construct definitions. Due to the various operational definitions adopted in the selected studies, there was great variation in how researchers measured the three facets of metalinguistic awareness. We also found that, for studies of nonalphabetic languages like Chinese, language-specific definitions and measures were adopted (e.g., Reference Wong and ZhouWong & Zhou, 2021; Reference ZhangZhang, 2017a, Reference Zhang2017b).

Regarding study quality indexed by task reliability, the mean task reliabilities of the selected studies fell between the 25th and 50th percentiles when compared to second language acquisition research in general.

Before we conducted the meta-analytic analysis reported in the next section, we reassessed whether there was sufficient evidence for the various relationship outcomes (as shown in Table 2). Although Reference Valentine, Pigott and RothsteinValentine and colleagues (2010) suggested that a minimum of two studies is needed for any meta-analysis, this suggestion should be evaluated with consideration of the number of independent study samples for subsequent subgroup and meta-regression analyses, which we also did for the present meta-analysis. According to Reference Cuijpers, Griffin and FurukawaCuijpers and colleagues (2021), a subgroup analysis requires three to four times the number of studies needed for the main analysis to have sufficient power. For instance, for a two-level categorical moderator analysis, Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg (2011) included a minimum of seven independent study samples. Regarding meta-regression analysis, a minimum of ten primary studies is expected based on the rule of thumb proposed by Reference Borenstein, Hedges, Higgins and RothsteinBorenstein and colleagues (2009). In this regard, our meta-analysis only focused on the relationships between the various facets of metalinguistic awareness and word decoding in child L2 readers.

4.1 Guided Questions

The following meta-analysis focuses on the cross-language transfer of metalinguistic awareness between two languages, the intralingual contributions of metalinguistic awareness in L1 and L2 word decoding respectively, and the interlingual contributions of L1 metalinguistic awareness to L2 word decoding. Four questions were posed:

1. 1. To what extent do L1 phonological, orthographic, and morphological awareness correlate with L2 phonological, orthographic, and morphological awareness in L2 readers?

2. 2. To what extent do phonological, orthographic, and morphological awareness correlate with word decoding intralingually in L2 readers?

   1. a) To what extent do L1 phonological, orthographic, and morphological awareness correlate with L1 word decoding in L2 readers?

   2. b) To what extent do L2 phonological, orthographic, and morphological awareness correlate with L2 word decoding in L2 readers?

3. 3. To what extent do L1 phonological, orthographic, and morphological awareness correlate with L2 word decoding in L2 readers?

4. 4. To what extent do the relations in Questions 1–3 vary as a function of linguistic-, learner-, measurement-, and instruction-related factors (i.e., L1–L2 writing system type, L2 proficiency, grade level/age, language majority/minority group, the measurement of phonological awareness, orthographic awareness, morphological awareness or word decoding, and medium of instruction)Footnote ⁹?

4.2 Coding of Effect Sizes and Moderator Variables

Effect size coding was based on the zero-order correlation matrix (i.e., Pearson’s r in selected studies). For the primary study of a longitudinal design, only time-one correlations were recorded to avoid interdependence. Because it is common to find multiple measures of the same construct, we used the arithmetic mean of effect sizes (after recent meta-analyses such as Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin et al., 2012; Reference Peng, Lee, Luo, Li, Joshi and TaoPeng et al., 2021; Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018). For moderator coding, details are as follows:

L1–L2 Writing System Type. Whether both L1 and L2 writing systems were alphabet or at least one language was non-alphabet was coded (after Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011).

L2 Proficiency. Participants who were identified as beginning-level students or in the first two years of L2 learning were coded as Basic and the rest as Beyond Basic (after Reference Jeon and YamashitaJeon & Yamashita, 2014).

Grade Level/Age. The exact grade levels and age in months were coded.

Language Group Membership. There were two groups: language minority group whose L1 was not the societal language versus language majority group.

Phonological Awareness, Orthographic Awareness and Morphological Awareness Measurement. For the respective three facets of metalinguistic awareness, we coded whether previous studies used multiple measures or a single measure only.

Word Decoding Measurement. Two moderators were coded: 1) standardized design versus researcher designed; 2) word status (i.e., real words or both real and pseudowords). Since all of the selected studies measured word decoding accuracy only without including word decoding fluency or efficiency, no relevant moderator analysis was conducted for this measurement factor.

Medium of Instruction. Two categories were coded: using L1 as the medium of instruction or using L2 as the medium of instruction.

4.3 Meta-Analytic Procedures

All data analyses were conducted with Comprehensive Meta-Analysis (CMA) software Version 3.0 (www.metaanalysis.com/) and Microsoft Excel. The meta-analysis included seven major steps:

First, to examine the overall correlations, an average of the correlations of all primary studies weighted for the sample sizes was calculated using an estimation of the inverse variance of each effect size (see Reference Borenstein, Hedges, Higgins and RothsteinBorenstein et al., 2009).

Second, data analysis was based on random-effects models, which according to Reference Borenstein, Hedges and RothsteinBorenstein and colleagues (2007) are a more plausible match than fixed-effect models.

Third, to investigate variability among the correlations across primary studies, heterogeneity tests (Q test; Reference Hedges and OlkinHedges & Olkin, 1985) were conducted. A significant value in the heterogeneity test would provide evidence for variability among the correlations (i.e., I² indicates the proportion of the heterogeneity that is real rather than due to chance and the parameter τ² indicates the variance in true effect sizes).

Fourth, given that most of the moderator variables in this study are categorical, a heterogeneity test (Q test) was performed based on mixed-effects modeling, both within and across the subgroups created by the moderator analyses. A statistically significant difference between subgroups would suggest the moderator influences the mean correlation. The analysis of potential moderators indicates whether subgroup membership affects the correlational outcomes. Notably, in addition to the moderator analyses with grade level as a categorical variable via subgroup analyses, we also examined age (in months) as a continuous variable via meta-regression analyses and compared the results. The rationale for comparing grade level and age is that the age range might vary for the same grade among different bilingual and bicultural educational settings.

Fifth, for cases where there was only one sample for a specific subgroup, we conducted a sensitivity analysis and removed one sample at a time from each moderator analysis to determine if the results would be altered. All sensitivity analyses results can be found in Appendix S4 available online at: osf.io/4z6mw.

Sixth, after significant moderator effects were identified, meta-regression analyses were conducted for any relationships with at least ten primary studies (based on the rule of thumb proposed by Reference Borenstein, Hedges, Higgins and RothsteinBorenstein et al., 2009) to identify the proportion of variance explained by the moderator(s).

Last, we examined possible effects of publication bias. Among the nineteen independent samples, sixteen were selected from published journal articles, two were based on book chapters, and one was based on an unpublished doctoral dissertation. Funnel plots for random-effects models were used to determine the presence of retrieval bias, for which the funnel would be asymmetric (see Appendix S5 online: osf.io/4z6mw). The results of Duval and Tweedie’s trim and fill are reported in Tables 4 to 7 when asymmetries were found in the funnel plots.

4.4 Cross-Language Transfer between L1 and L2 Phonological, Orthographic, and Morphological Awareness

As mentioned earlier, the interpretation of effect sizes (rs) is based on Reference Plonsky and OswaldPlonsky and Oswald’s (2014) estimates for small, medium, and large correlations (r = .25, .40, .60, respectively). Regarding the observed variance in primary studies indexed by I², 25 percent, 50 percent, and 75 percent might be considered as low, moderate, and high (Reference Borenstein, Hedges, Higgins and RothsteinBorenstein et al., 2009).

For guided Question 1, Table 4 and Figures 2a to 2c illustrate that the overall mean correlation between L1 and L2 phonological awareness (k = 7, N = 782) was medium and significant, r = .44, 95% CI [.28, .57], z(6) = 5.14, p < .001; whereas the respective mean correlations between L1 and L2 orthographic awareness (k = 8, N = 858) and between L1 and L2 morphological awareness (k = 8, N = 811) were both small yet significant (r = .30, 95% CI [.08, .49], z(7) = 2.65, p =.008; r = .28, 95% CI [.22, .35], z(7) = 8.15, p < .001). The adjusted effect estimate results did not change our conclusions of the correlation strengths (see Table 4).

Figures 2a Forest plot of the relationship between first-language (L1) and second-language (L2) phonological awareness (PA). The results for the overall mean correlation are given in the last line.

Figure 2b Forest plot of the relationship between first-language (L1) and second-language (L2) orthographic awareness (OA). The results for the overall mean correlation are given in the last line.

Figure 2c Forest plot of the relationship between first language (L1) and second language (L2) morphological awareness (MA). The results for the overall mean correlation are given in the last line.

Heterogeneity test results indicated that there was significant and large variability in the correlations between L1 and L2 phonological awareness (Q(5) = 33.54, p < .001, I² = 82.11%, τ² = .05) as well as between L1 and L2 orthographic awareness (Q(6) = 76.44, p < .001, I² = 90.84%, τ² = .10). However, no significant variation was observed in the correlations between L1 and L2 morphological awareness (Q(6) = 5.68, p = .578, I² <.001, τ² < .001).

Thus, the answer to Question 1 is that there was a significant and moderate correlation between L1 and L2 phonological awareness, whereas there were significant and small correlations between L1 and L2 orthographic awareness as well as between L1 and L2 morphological awareness.

4.5 Intralingual Relationships between Metalinguistic (Phonological, Orthographic, Morphological) Awareness and Word Decoding in L1

In this section, we focus on the intralingual correlations between metalinguistic (phonological, orthographic, and morphological) awareness and word decoding in bilingual readers’ L1. This answers guided Question 2a. The results in Table 5 and Figures 3a to 3c indicate that, for all three relationships, the correlations were significant and small. The mean correlation between L1 phonological awareness and L1 word decoding was .27 (k = 11, N = 1,239, 95% CI [.22, .32], z(10) = 9.43, p < .001); the overall mean correlation between L1 orthographic awareness and L1 word decoding was .34 (k = 11, N = 1,239, 95% CI [.15, .50], z(10) = 3.50, p < .001); and the mean correlation between L1 morphological awareness and L1 word decoding was .38 (k = 11, N = 1,239, 95% CI [.30, .45], z(10) = 8.78, p < .001). The adjusted effect estimate results did not change our conclusions of the correlation strengths (as illustrated in Table 5).

Figure 3a Forest plot of the relationship between first-language (L1) phonological awareness (PA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 3b Forest plot of the relationship between first-language (L1) orthographic awareness (OA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 3c Forest plot of the relationship between first-language (L1) morphological awareness (MA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

In addition, no significant variation among primary studies was identified for the relationship between L1 phonological awareness and L1 word decoding (Q(9) = 10.31, p = .414, I²= 2.97, τ² < .001), whereas significant large and moderate variation was found respectively for the relationship between L1 orthographic awareness and L1 word decoding (Q(9) = 78,30, p < .001, I² = 89.78, τ² = .08), as well as the relationship between L1 morphological awareness and L1 word decoding (Q(9) = 23.16, p = .010, I² = 56.81, τ² = .01).

In response to Question 2a, we found that L1 phonological, orthographic, and morphological awareness all correlated significantly with L1 word decoding in bilingual child readers and the mean correlation magnitudes were all small.

4.6 Intralingual Relationships between Metalinguistic (Phonological, Orthographic, and Morphological) Awareness and Word Decoding in L2

This section is guided by Question 2b, that is, “To what extent do L2 phonological, orthographic, and morphological awareness correlate with L2 word decoding in L2 readers?” All three facets of metalinguistic awareness correlated significantly and moderately with word decoding in bilingual readers’ L2: 1) The overall mean correlation between L2 phonological awareness and L2 word decoding was .40 (k = 11, N = 1,153, 95% CI [.33, .46], z(10) = 10.08, p < .001). 2) The mean correlation between L2 orthographic awareness and L2 word decoding was .47 (k = 12, N = 1,374, 95% CI [.37, .56], z(11) = 8.27, p < .001). 3) The mean correlation between L2 morphological awareness and L2 word decoding was .46 (k = 10, N = 1,096, 95% CI [36, .54], z(9) = 8.35, p < .001) (see Table 6 and Figures 4a to 4c).

Figure 4a Forest plot of the relationship between second-language (L2) phonological awareness (PA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 4b Forest plot of the relationship between second-language (L2) orthographic awareness (OA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 4c Forest plot of the relationship between second-language (L2) morphological awareness (MA) and word decoding (WD). The results for the overall mean correlation are given in the last line.

Notably, heterogeneity tests were all significant for the three L2 intralingual relationships. Specifically, the variations were both moderate for the relationship between L2 phonological awareness and L2 word decoding (Q(9) = 20.85, p = .022, I²= 52.04, τ² = .01), as well as between L2 morphological awareness and L2 word decoding (Q(8) = 29.80, p < .001, I²= 69.80, τ² = .02). There was large variation for the relationship between L2 orthographic awareness and L2 word decoding (Q(10) = 54.04, p < .001, I²= 79.64, τ² = .03).

In response to Question 2b, there were significant and moderate correlations between L2 metalinguistic (phonological, morphological, and orthographic) awareness and L2 word decoding.

4.7 Interlingual Relationships between L1 Metalinguistic (Phonological, Orthographic, and Morphological) Awareness and L2 Word Decoding

For guided Question 3, Table 7 and Figures 5a to 5c illustrate the results of the overall mean correlations between L1 metalinguistic (phonological, orthographic, and morphological) awareness and L2 word decoding. All relationships were significant (ps < .05). For the relationship between L1 phonological awareness and L2 word decoding, the mean correlation (r) was .39 (k = 10, N = 1,239, 95% CI [.31, .47], z(9) = 8.54, p < .001). Notably, the adjusted effect estimate was .44 (95% CI [.39, .48]) after trim and fill, which indicated a moderate effect size. The mean correlations between L1 orthographic awareness and L2 word decoding as well as between L1 morphological awareness and L2 word decoding were both small: The respective rs were .34 (k = 9, N =1,050, 95% CI [.15, .50], z(8) = 3.50, p < .001) and .33 (k = 11, N =1,141, 95% CI [.33, .42], z(10) = 6.20, p < .001). The adjusted effect estimate for L1 orthographic awareness and L2 word decoding was .40 (95% CI [.35, .45]) after trim and fill, which indicated a moderate correlation.

Figure 5a Forest plot of the relationship between first-language (L1) phonological awareness (PA) and second-language (L2) word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 5b Forest plot of the relationship between first-language (L1) orthographic awareness (OA) and second language (L2) word decoding (WD). The results for the overall mean correlation are given in the last line.

Figure 5c Forest plot of the relationship between first-language (L1) morphological awareness and second-language (L2) word decoding. The results for the overall mean correlation are given in the last line.

Heterogeneity tests were significant and small for the relationship between L1 phonological awareness and L2 word decoding (Q(8) = 22.18, p < .001, I²= 59.43, τ² = .01); significant and small for the relationship between L1 morphological awareness and L2 word decoding (Q(7) = 34.32, p <.001, I²= 70.87, τ² = .08); and significant and large for the relationship between L1 orthographic awareness and L2 word decoding (Q(9) = 78.30, p < .001, I²= 89.78, τ² = .02).

In response to Question 3, there were significant and moderate correlations (after adjusted estimation) between L1 phonological awareness and L2 word decoding, and between L1 orthographic awareness and L2 word decoding, as well as a significant and small correlation between L1 morphological awareness and L2 word decoding.

4.8 Moderator Analyses: Linguistic-, Learner-, Measurement-, and Instruction-Related Effects

4.8.2 Meta-Regression Analyses

After significant moderator effects were identified, meta-regression analyses were conducted for three relationships in a minimum of ten primary studies (based on the rule of thumb proposed by Reference Borenstein, Hedges, Higgins and RothsteinBorenstein et al., 2009): 1) L2 phonological awareness and L2 word decoding, 2) L2 orthographic awareness and L2 word decoding, and 3) L1 morphological awareness and L2 word decoding.

For all four relationships, we conducted moderator analyses with age as a continuous variable and compared the results with those with grade level as a categorical variable. Grade level was not found to be a significant moderator for any of the relationships examined in Section 4.7; however, as a continuous variable, age in months was a significant moderator for the relationship between L2 phonological awareness and L2 word decoding (β = −0.004, p = .035, R² = 0.37; see Model 1 in Table 11). As Figure 6 illustrates, as age increased, the strength of the relationship between L2 phonological awareness and L2 word decoding decreased. No moderating effect of age on other relationships was found.

Table 11 Meta-regression analysis results with age and L2 proficiency as covariates for the relationship between L2 phonological awareness and L2 word decoding (k = 10).

Model 1
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.804	0.223	0.367	1.242	3.60	< .001
Age	−0.004	0.002	−0.007	0.0003	−1.82	.035	0.37
Model 2
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.480	0.047	0.387	0.572	10.18	< .001
L2 proficiency (beyond basic)	−0.186	0.073	−0.328	−0.044	−2.56	.005	.71
Model 3
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.888	0.176	0.543	1.233	5.04	< .001
Age	−0.004	0.002	−0.007	−0.001	−2.44	.007
L2 proficiency (beyond basic)	−0.181	0.059	−0.297	−3.080	−3.08	.001	1.00
Model 4
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.888	0.176	0.543	1.233	5.04	< .001
L2 proficiency (beyond basic)	−0.181	0.059	−0.297	−0.066	−3.08	.001
Age	−0.004	0.002	−0.007	−0.001	−2.44	.007	1.00

Notes. L2 = second language. For the “L2 proficiency” moderator, there are two categories (basic vs. beyond basic); the basic L2 proficiency group was treated as the reference group. CMA V3 software automatically generated dummy coding for the categorical moderator.

Figure 6 Scatter plot of the moderating effect of age on the relationship between second-language (L2) phonological awareness and second-language (L2) word decoding.

For the relationship between L2 phonological awareness and L2 word decoding, we also identified L2 proficiency (basic vs. beyond basic) as a significant moderator (β = −0.186, p = .005, R² = 0.71, as shown in Model 2 in Table 11). As L2 proficiency increased, the magnitude of the relationship between L2 phonological awareness and L2 word decoding decreased. When both age and L2 proficiency were entered for meta-regression in reverse order, each had an additional effect over the other (as shown in Model 3 and Model 4 in Table 11). The variance was completely explained by the two moderators (R² = 1.00).

Table 12 illustrates the meta-regression analysis results for the relationship between L2 orthographic awareness and L2 word decoding. When language group membership was entered into the meta-regression model, it had a significant effect, thereby explaining the variance of the correlation between L2 orthographic awareness and L2 word decoding (R² = .28). According to Reference Plonsky and GhanbarPlonsky and Ghanbar (2018), an R² less than .10 or .18 is considered as small or moderate, and regression models explaining more than 50 percent of the variance are considered fairly robust. In the case of this study, the moderating effect of language group membership on the relationship between L2 orthographic awareness and L2 word decoding was moderate.

Table 12 Meta-regression analysis results for the relationship between L2 orthographic awareness and L2 word decoding (k = 12).

Model
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.376	0.091	0.197	0.555	4.12	< .001
Language group membership (majority)	0.237	0.119	0.005	0.470	2.00	.023	0.28

Note. For the moderator “language group membership,” there are two categories (minority vs. majority); the language minority group was treated as the reference group.

Finally, we also examined the moderating effect of word status in word-decoding measures on the relationship between L1 morphological awareness and L2 word decoding. As Table 13 indicates, when it was entered into the meta-regression model, “word-decoding word status” had a significant effect explaining the variance of the correlation between L1 morphological awareness and L2 word decoding (respective R² = 0.56), thereby making it a robust finding.

Table 13 Meta-regression analysis results for the relationship between L1 morphological awareness and L2 word decoding (k = 10).

Model
Covariate	Coefficient	Standard error	95% lower	95% upper	Z-value	p	R2
Intercept	0.409	0.049	0.313	0.506	8.34	< .001
word-decoding word status	−0.390	0.127	−0.639	−0.141	−3.06	=.001	0.56

Note. For the moderator “word-decoding word status,” there are two categories (real words vs. real and pseudowords); the real word group was treated as the reference group.

In response to Question 4, “to what extent do the relations above (in Questions 1–3) vary as a function of linguistic-, learner-, measurement- and instruction-related factors (i.e., L1–L2 writing system type, language majority/minority group, L2 proficiency, grade level/age, the measurement of phonological awareness, orthographic awareness, morphological awareness, or word decoding, and medium of instruction),” there were five major findings:

1. 1. The L1–L2 writing system type did not have any significant moderating effects.

2. 2. We also found that the variance in the correlation between L2 orthographic awareness and L2 word decoding was significantly moderated by language group membership. The relationship was stronger for the language majority group.

3. 3. The variance in the correlation between L2 phonological awareness and L2 word decoding can be explained by age and L2 proficiency (R² = 1.00). As age and L2 proficiency increased, the strength of the relationship decreased. However, due to the small independent sample size, we could not analyze any interactional effects between age and L2 proficiency. Notably, we examined potential moderating effects of grade level as a categorical variable (kindergarten to grade two vs. grade three and above) versus age as a continuous variable, and found no significant moderating effect of grade level. In contrast, a significant moderating effect of age on the relationship between L2 phonological awareness and L2 word decoding was observed.

4. 4. We observed a significant moderating effect of orthographic awareness measurement on the relationship between L1 and L2 orthographic awareness based on subgroup analysis, with a larger mean correlation in primary studies using multiple measures. We also found that “word-decoding word status” had a robust effect explaining the variance of the correlation between L1 morphological awareness and L2 word decoding, and the correlation was higher for real word decoding.

5. 5. No significant moderating effects of medium of instruction were detected.

4.9 General Discussion: A Comparison of Results between This Meta-Analysis and Prior Meta-Analytic Reviews

Transfer of Metalinguistic Awareness. The findings of different magnitudes of mean correlations among the three facets of metalinguistic awareness (i.e., a moderate effect size for phonological awareness as well as small effect sizes for orthographic and morphological awareness) seem to be consistent with Reference KodaKoda’s (2005, Reference Koda, Koda and Zehler2008) conceptualizations of phonological awareness as a language-general facet of metalinguistic awareness versus orthographic and morphological awareness as language-specific facets of metalinguistic awareness (see also Reference Chung, Chen and GevaChung et al., 2019; Reference Koda, Lü, Zhang, Chen, Wang and LuoKoda et al., 2014; Reference PerfettiPerfetti, 2003). In previous meta-analyses with larger independent sample pools, Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg (2011), for example, observed a large and significant mean correlation between L1 and L2 phonological awareness (r = .60, 95% CI [.49, .69], k = 16, N = 1,340, z(15) = 58.50, p = 0.01). In addition, Reference Ke, Miller, Zhang and KodaKe and colleagues’ (2021) meta-analysis of morphological awareness in child biliteracy development found a small correlation between L1 and L2 morphological awareness (r = .30, 95% CI [.22, .38], k = 20, N = 1,657, z(19) = 7.14, p < .001). To our knowledge, our meta-analysis is among the first to provide aggregative evidence of the significant and small correlation between L1 and L2 orthographic awareness (r = .30). This has filled a notable gap in previous meta-analyses of metalinguistic awareness and reading.

Intralingual Relationship between Metalinguistic Awareness and Word Decoding in L1. The comparison results are summarized in Table 14, which indicates significant variation in the meta-analytic results for the relationship between L1 phonological awareness and L1 word decoding, ranging from small to moderate for bilingual readers of mixed L1s (as shown by the results of this meta-analysis), L1 Chinese readers (see Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018), or Spanish readers with monolingual and bilingual backgrounds (see Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021). Specifically, for L1 English monolingual readers, the correlation between L1 phonological awareness and word decoding appeared to be higher (above the moderate level or .40) according to Reference Ruan, Georgiou, Song, Li and ShuRuan and colleagues (2018). With respect to the relationship between L1 morphological awareness and L1 word decoding, both this meta-analysis and that of Ruan and colleagues suggested that the magnitude was close to the moderate range (i.e., .40).

Regarding the relationship between L1 orthographic awareness and L1 word decoding, to the best of our knowledge there is only one meta-analytic review in the existing literature (Reference Georgiou, Martinez, Vieira and GuoGeorgiou et al., 2021). In their research, Georgiou et al. analyzed sixty-eight studies, the majority of which were about reading English as the target language. They found that there were significant differences between dyslexic readers against a chronological age control group and a reading level control group, and that the significant heterogeneity in the effect sizes was partly explained by the level of orthographic knowledge (larger effect sizes were observed at the lexical level than at the sublexical level). Based on a relatively small sample size (k = 9), our meta-analytic results suggested that orthographic awareness, like phonological and morphological awareness, correlated significantly with word decoding in bilingual children’s L1 reading and that the effect size was small (r = .34).

Intralingual Relationship between Metalinguistic Awareness and Word Decoding in L2. For comparison purposes, we tried to identify related effect sizes from previous meta-analyses of phonological awareness and word decoding in bilingual/L2 readers (e.g., Reference Branum-Martin, Tao, Garnaat, Bunta and FrancisBranum-Martin et al., 2012; Reference Jeon and YamashitaJeon & Yamashita, 2014; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011). However, the relationship between L2 phonological awareness and L2 word reading was not examined in those meta-analyses. One exception is Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez and colleagues’ (2021) meta-analysis of Spanish phonological awareness in a sample comprised mainly of monolingual Spanish-speaking children and mixed with L2 Spanish-speaking children whose L1s were Aymara, Basque, or English. Míguez-Álvarez and colleagues found a stronger correlation between phonological awareness and word decoding in bilingual children than in monolingual children. Similarly, in the present meta-analysis, we found a larger correlation between phonological awareness and word decoding in bilingual readers’ L2 (r = .46 within the moderate effect size range) than in their L1 (r = .27 within the small effect size range).

Regarding the strength of the relationship between L2 morphological awareness and L2 word decoding, both this meta-analysis (albeit it having a small sample size, k =10) and Reference Ke, Miller, Zhang and KodaKe and colleagues (2021) (k = 34) identified an overall mean (r) of .46. Lastly, we could not identify any meta-analysis of orthographic awareness and word decoding in L2 reading in the existing literature. Our results suggested a moderate correlation between L2 orthographic awareness and L2 word decoding (r =.47, k = 12).

Cross-Language Transfer Facilitation Relationships between L1 Metalinguistic Awareness and L2 Word Decoding. When comparing our effect sizes with those in previous metanalyses (see Table 15), we found that both this meta-analysis and previous ones suggested a moderate range of correlation between L1 phonological awareness and L2 word decoding as well as a small correlation between L1 morphological awareness and L2 word decoding. No comparison could be drawn in previous meta-analyses regarding the correlations between L1 orthographic awareness and L2 word decoding, but this meta-analysis found a medium correlation based on adjusted estimation (r =.40, k = 11).

Linguistic-, Learner-, Measurement-, and Instruction-Related Moderating Effects. To recapitulate, a range of moderator factors was examined in this meta-analysis, including L1–L2 writing system type; L2 proficiency; grade level/age; language majority/minority group; measurements of phonological, orthographic, and morphological awareness or word decoding; and medium of instruction.

Neither this meta-analysis nor previous ones (e.g., Reference Ke, Miller, Zhang and KodaKe et al., 2021; Reference Melby‐Lervåg and LervågMelby-Lervåg & Lervåg, 2011) identified any significant moderating effects of L1–L2 writing system type, thus indicating that the cross-language transfer of L1 metalinguistic awareness is not subject to L1–L2 distance. Once metalinguistic awareness is developed in L1, it is readily exploitable for L2 reading subskill acquisition. This finding applies to both language-general (phonological) and language-specific (orthographic and morphological) metalinguistic awareness. On the other hand, it could be influenced by our general operationalization of L1–L2 writing system type as a two-level categorical variable (i.e., a combination of two alphabetic languages vs. inclusion of at least one nonalphabetic language in L1 or L2). This was mainly because the target L2s for the three relationships (i.e., L1 and L2 phonological awareness, L1 and L2 orthographic awareness, and L1 and L2 morphological awareness) were either English or French in the selected studies. We thus urge future research to analyze the L1–L2 distance factor with more refined categorization. As one of the anonymous reviewers pointed out, the magnitudes of transfer facilitation effects might vary between alphabetic L1–morphosyllabic L2 readers and morphosyllabic L1–alphabetic L2 readers.

We observed significant moderating effects of age and L2 proficiency, specifically for the relationship between L2 phonological awareness and L2 word decoding. As age and L2 proficiency level increased, the correlation became weaker. Additionally, in this meta-analysis, no influence of age or L2 proficiency was observed for any of the relationships associated with orthographic or morphological awareness. Our findings, taken together, suggest that phonological awareness may play a less important role than orthographic and morphological awareness in L2 reading development in the long run. This also seems to corroborate Reference Ke, Miller, Zhang and KodaKe and colleagues’ (2021) finding that there was a larger correlation for L2 morphological awareness and L2 reading at upper grade levels (i.e., grade three and above) than at lower grade levels (i.e., kindergarten to grade two), which suggests a sustainably important role of morphological awareness in L2 reading development.

We also found significant moderating effects of language group membership on the relationship between L2 metalinguistic awareness and L2 word decoding. Larger correlations were observed for the language majority group, which was consistent with the finding of Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie and colleagues’ (2022) meta-analysis that mainly compared English morphological knowledge between language majority and language minority groups. The weaker relationship between metalinguistic awareness and word decoding in the language minority group could be jointly influenced by their relatively weaker foundation in L1 metalinguistic awareness sophistication and less exposure to L2 print input compared to the language majority group (see also Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008).

In regard to measurement-related effects, previous meta-analyses of phonological awareness or morphological awareness (e.g., Reference Bratlie, Brinchmann, Melby-Lervåg and TorkildsenBratlie et al., 2022; Reference Míguez‐Álvarez, Cuevas‐Alonso and SaavedraMíguez-Álvarez et al., 2021; Reference Ruan, Georgiou, Song, Li and ShuRuan et al., 2018) identified larger correlations for studies using complex measures of phonological awareness and included nonword decoding and word decoding accuracy as reading outcomes. Larger correlations were also observed for those that measured oral and expressive/productive morphological awareness compared to those measuring written and receptive/judgmental morphological awareness. To add to the literature, this meta-analysis found a significant moderating effect of orthographic awareness measurement on the relationship between L1 and L2 orthographic awareness based on subgroup analysis, with a larger mean correlation in primary studies using multiple measures compared to studies using only a single measure.

Lastly, medium of instruction did not have any significant moderating effects on any of the focal relationships. This finding is similar to that of Reference Melby‐Lervåg and LervågMelby-Lervåg and Lervåg’s (2011) meta-analysis of phonological awareness only.

4.10 Summary

The meta-analytic analyses above focused on the cross-language correlations between three facets of metalinguistic awareness (i.e., phonological, orthographic, and morphological awareness), as well as the intralingual and interlingual relationships between metalinguistic awareness and word decoding in 2,654 bilingual children with nineteen independent samples from sixteen primary studies. Our guided research questions were not about whether there is any transfer of metalinguistic awareness or how transfer facilitation occurs; rather, we examined to what extent metalinguistic awareness developed in L1 correlates with corresponding L2 metalinguistic awareness and reading subskills. We also aimed to demonstrate more systematic analyses of linguistic-, learner-, measurement-, and instruction-related factors to identify significant moderators.

The findings included: 1) There was a significant and moderate correlation between L1 and L2 phonological awareness, a language-general facet of metalinguistic awareness; in contrast, there were significant and small correlations between L1 and L2 language-specific metalinguistic awareness (i.e., orthographic and morphological awareness), consistent with the predictions of the Transfer Facilitation Model (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008). 2) There were significant and small correlations between L1 metalinguistic (phonological, morphological, and orthographic) awareness and L1 word decoding in child bilingual readers; in contrast, there were significant and moderate correlations between L2 metalinguistic (phonological, morphological, and orthographic) awareness and L2 word decoding. This result is also aligned with the Lexical Quality Hypothesis (Reference PerfettiPerfetti, 2007; Reference Perfetti, Hart, Vehoeven, Elbro and ReitsmaPerfetti & Hart, 2002) and the Repertoire Theory of Literacy Development (Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000), suggesting that all three major facets of metalinguistic awareness (i.e., phonological, orthographic, and morphological awareness) are important for both L1 and L2 word decoding. A unique finding of our meta-analysis was that metalinguistic awareness played a more important role in L2 word decoding than in L1 word decoding in child bilingual readers. 3) Interlingually speaking, there were significant and moderate correlations between L1 phonological awareness and L2 word decoding, and between L1 orthographic awareness and L2 word decoding, as well as a significant and small correlation between L1 morphological awareness and L2 word decoding. 4), Age, L2 proficiency, language group membership (i.e., majority vs. minority), as well as the measurement of orthographic awareness and word decoding had significant moderating effects. Specifically, as age and L2 proficiency level increased, the correlation strength between L2 phonological awareness and L2 word decoding decreased. Additionally, language group membership had a significant influence on the relationship between L2 orthographic awareness and L2 word decoding, and the relationships were weaker for the language minority group. Furthermore, larger correlations were observed for primary studies using multiple measures of orthographic awareness (as opposed to a single measure) and measuring word decoding with real words and standardized testing design. Finally, no significant moderating effects were identified for L1–L2 writing system type or medium instruction.

5.1 Reflection

In this Element, our primary goal was to provide a comprehensive account of the complexity of metalinguistic awareness and its impacts on second-language reading development. To our knowledge, we are among the first to evaluate the roles of the three different facets of metalinguistic awareness (i.e., phonological, orthographic, and morphological awareness) simultaneously via a meta-analytic review of L2 reading studies. We began by providing brief explanations of the theoretical foundations by relating them to the Lexical Quality Hypothesis (Reference PerfettiPerfetti, 2007; Reference Perfetti, Hart, Vehoeven, Elbro and ReitsmaPerfetti & Hart, 2002), the Repertoire Theory of Literacy Development (Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000), and the Transfer Facilitation Model (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008), clarifying the definition of metalinguistic awareness as a multifaceted construct in relation to reading development, that is, “the ability to reflect on and manipulate the structural features of language” (Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007, p. 53; Reference Nagy and AndersonNagy & Anderson, 1995, p. 2), and describing the possible mechanisms through which metalinguistic awareness develops reciprocally with reading competence and lexical knowledge and contributes to reading development. Subsequently, we reported a scoping review and meta-analysis of correlation coefficients with four guided questions: 1) To what extent do L1 phonological, orthographic, and morphological awareness correlate with L2 phonological, orthographic, and morphological awareness in L2 readers? 2) To what extent do phonological, orthographic, and morphological awareness correlate with word decoding intralingually in L2 readers? 3) To what extent do L1 phonological, orthographic, and morphological awareness correlate with L2 word decoding in L2 readers? 4) To what extent do the relations in Questions 1–3 vary as a function of linguistic-, learner-, measurement-, and instruction-related factors?

Based on the evidence of the present meta-analysis and previous ones reviewed in this Element (Section 2), we concluded, for Question 1, that, regardless of the distance between a learner’s first and second languages, there were significant small to moderate correlations between the various facets of metalinguistic awareness in the two languages. In response to Questions 2 to 3, our meta-analysis focused on word decoding only because primary studies yielded insufficient evidence for other reading-related outcomes (e.g., lexical inferencing, vocabulary knowledge, and reading comprehension). Results indicated that there were significant correlations between the three facets of metalinguistic awareness and word decoding in both first and second languages, and metalinguistic awareness played a more important role in L2 reading than in L1 reading (see also Reference BialystokBialystok, 2001; Reference Goodwin and AhnGoodwin & Ahn, 2013). In addition, the meta-analytic evidence seemed to confirm the theorization that phonological awareness is a language-general facet of metalinguistic awareness, whereas orthographic and morphological awareness are language specific (Reference KodaKoda, 2005, Reference Koda, Koda and Zehler2008; Reference PerfettiPerfetti, 2003). Specifically, these conclusions were supported by the evidence that there were moderate correlations between L1 phonological awareness and L2 phonological awareness or word decoding, small correlations between L1 orthographic awareness and L2 orthographic awareness or word decoding, and small correlations between L1 morphological awareness and L2 morphological awareness or word decoding.

Finally, there was significant variation in the selected studies (i.e., sixteen primary studies that measured the three facets of metalinguistic awareness simultaneously in K–8 child bilingual learners representing twelve languages and four writing systems, mostly reading English as the target L2). Among a range of linguistic-, learner-, measurement-, and instruction-related factors tested under Question 4, we have identified five significant moderators: age, L2 proficiency, language group membership (i.e., majority vs. minority), as well as the measurement of orthographic awareness and word decoding. Another notable construct-related factor was how the selected studies defined and operationalized the three facets of metalinguistic awareness, each being a complex and multidimensional construct. Our scoping review has found that only about half the primary studies provided clear definitions of metalinguistic awareness. For studies that did not include this information, the strength of test validity (the extent to which a test measures what it is intended to measure, Reference KerlingerKerlinger, 1999; Reference Messick and LinnMessick,1989) may be questionable.

We acknowledge that several limitations need to be addressed in future research. First, the evidence meta-analyzed was primarily correlational, not causal (e.g., based on empirical studies of longitudinal or interventional designs, Reference Ke and ZhangKe & Zhang, 2021). Second, the independent study sample size was modest and the participants were mainly child learners of L2 alphabetic languages (e.g., English and French), partly because of our strict inclusion criterion that all three facets of metalinguistic awareness should be concurrently measured in a study. There is a need in the future for a (semi-) replicated meta-analysis with a larger pool of independent study samples examining at least one facet of metalinguistic awareness and including both child and adult learners of diverse L2s. Third, due to limited independent study samples, we could not carry out meta-analyses of the relationships between L1 metalinguistic awareness and L2 lexical inferencing, reading comprehension, or vocabulary knowledge. Although our meta-analytic evidence seemed to converge on the cross-language correlations between L1 metalinguistic awareness and corresponding metalinguistic awareness and word decoding in the L2, and we could therefore possibly infer that metalinguistic awareness can indirectly facilitate reading comprehension via the mediating effects of word decoding (e.g., Reference BadianBadian, 2001; Reference Deacon, Kieffer and LarocheDeacon et al., 2014; Reference Li and WuLi & Wu, 2015; Reference Zhang, Lin, Liu and NagyZhang et al., 2020; Reference Zhao, Wu, Sun, Xie, Feng and ChenZhao et al., 2019), it is important to examine the reciprocal relationships, particularly interlingually, among metalinguistic awareness, word decoding, lexical inferencing, vocabulary knowledge, and reading comprehension via primary and meta-analytic research. Finally, caution should be taken regarding the interpretation of the five significant moderators identified in the present meta-analysis because there might be confounding effects among them (Reference Gottardo, Chen and HuoGottardo et al., 2021). To our knowledge, it is rare for existing meta-analyses of reading or applied linguistics research to examine interactional effects of more than one moderator.

5.2 Implications for Applied Linguistics Research: A Research Agenda

There are three cross-cutting themes worthy of more in-depth research in the future:

1. 1. A systematic account of societal/contextual factors in addition to cognitive and linguistic factors in the development of metalinguistic awareness in relation to L2 reading. Ever since Reference KodaKoda (2005, Reference Koda, Koda and Zehler2008) proposed the Transfer Facilitation Model of L2 reading, ample evidence has been generated from studies with L2 learners who usually have established L1 skills. Existing evidence has supported the important contentions proposed in the model (e.g., metalinguistic awareness provides a window for researching transfer; L1 transfer is nonvolitional and automatic; multiple cognitive and linguistic factors modulate transfer facilitation effects such as L1–L2 distance, L1 and L2 proficiency, and language complexity [for reviews, see Reference Koda and KeKoda & Ke, 2018; Reference Koda and ReddyKoda & Reddy, 2008]). Recent reflections on the Transfer Facilitation Model (e.g., Reference Chung, Chen and GevaChung et al., 2019) have called for more attention to societal/contextual factors related to L2 reading. Some examples include, yet are not limited to, program type, teaching methods, and use of L1/L2 as dominant societal and instructional languages (see a discussion in Reference Gottardo, Chen and HuoGottardo et al., 2021). In addition, the Transfer Facilitation Model has mainly dealt with two languages; its explanatory power remains unclear for research of multilingual or multidialectal learners of more than two languages (except an ambitious forthcoming volume entitled Handbook of Literacy in Diglossia and in Dialectal Contexts: Psycholinguistic, Neurolinguistic, and Educational Perspectives edited by Reference Saiegh-Haddad, Lask and McBrideSaiegh-Haddad et al., 2022).

2. 2. The causal effect of metalinguistic awareness on L2 reading development. Two potential approaches can be used to verify the causality relationship between metalinguistic awareness and L2 reading. One is to expand current lines of longitudinal and interventional research of metalinguistic awareness instruction in L2 learners, and examine whether instruction promoting L1 metalinguistic awareness will lead to changes in L2 metalinguistic awareness and reading subskills, ideally in typologically different languages and writing systems (e.g., the transfer of English morphological awareness to Malay as a result of English morphology instruction in Reference ZhangZhang, 2016). In a recent review of L2 English reading research, Reference Ke and ZhangKe and Zhang (2021) suggested that the effects of morphological instruction delivered in English are transferrable to lexical inferencing in English and to reading development in another language. The other approach is to validate the metalinguistic hypothesis (Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007) via meta-analyses of a multivariate nature. With recent advancement in meta-analytic methods (e.g., meta-analytic structural equation modeling, Reference CheungCheung, 2014; Reference Cheung and ChanCheung & Chan, 2005; Reference Jak and CheungJak & Cheung, 2020) and aggregative evidence of L2 reading studies, it is possible to model the reciprocal relationships among metalinguistic awareness, reading competence (including word decoding, lexical inferencing, and reading comprehension), and vocabulary knowledge to test whether metalinguistic awareness mediates the relationship between reading competence and vocabulary knowledge in L2 learners. It should be noted that since prior empirical studies tended to examine word decoding rather than reading comprehension as the focal reading outcome (as shown in Table 2), it might be premature to provide meta-analytic evidence comparing the relative contributions of different metalinguistic skills to word decoding and reading comprehension, respectively. Therefore, more attention needs to be paid to examining L2 reading comprehension as one of the target L2 reading outcomes in future research of metalinguistic awareness.

3. 3. Assessment of metalinguistic awareness in relation to L2 reading. Among the three facets of metalinguistic awareness, phonological and morphological awareness assessment are relatively more established, while what constitutes orthographic awareness is still debatable (Reference ApelApel, 2011; Reference Apel, Henbest and MastersonApel et al., 2019). This may not be surprising because advances in reading psychology in recent decades have led to a focus on the linguistic aspects of reading, particularly the phonological and morphological processes involved (Reference ShareShare, 2021), and, unfortunately, a relative neglect of the visual processes (Reference Berninger and BerningerBerninger, 1994; see a review by Reference LiLi, 2021). However, recent research of self-teaching during reading in L1 and/or L2 readers, including both children and adults, has reignited interest in orthographic learning and awareness (Reference Li, Li and WangLi et al., 2020). Another future direction is the development of timed computerized tests of metalinguistic awareness. Since our meta-analysis sample participants were child bilingual readers, the metalinguistic awareness tests were commonly paper-and-pencil based or administered orally in a face-to-face manner. These tests recorded accuracy rates only. This operationalization that focused on untimed task performance and accuracy seems to limit the examination of metalinguistic awareness to language analysis only, and neglect the attentional control component involved in metalinguistically demanding tasks (Reference BialystokBialystok, 2001; Reference Bialystok and RyanBialystok & Ryan, 1985). Recent adult L1 and L2 reading research has begun to adopt computerized and timed tasks that tap efficiency performance (i.e., both reaction times and accuracy rates; e.g., an English morpheme counting task by Reference Bernstein, Flipse, Jin and OdegardBernstein et al., 2020, and English and Chinese affix shifting tasks by Reference Ke and KodaKe & Koda, 2021). These tasks have the potential to be expanded for assessment and diagnostic purposes in cognitively and linguistically diverse learner populations. Finally, this Element has only focused on three major facets of metalinguistic awareness. It may be worthwhile examining other facets of metalinguistic awareness (e.g., syntactic awareness, the ability to manipulate and reflect on the grammatical structure of language, in Reference Brimo, Apel and FountainBrimo et al., 2017; Reference CainCain, 2007; Reference Sohail, Sorenson Duncan, Koh, Deacon and ChenSohail et al., 2022; and Reference Tong, Kwan, Tong and DeaconTong et al., 2022) in relation to L1 and L2 reading development in future meta-analytic studies.

It would not be possible to execute the above agenda without close collaboration among a team of applied linguists, cognitive- and neuroscientists, educational psychologists, psychometricians, (teacher) educators, and policy makers.

5.3 Implications for Second Language Reading Instruction

The most important practical implication of our review and meta-analysis of metalinguistic awareness, and this Element in general, is perhaps to support teachers in developing an understanding of what metalinguistic awareness is, as well as how metalinguistic awareness functions in learning to read and reading to learn (e.g., lexical inferencing). Specifically, differentiating metalinguistic awareness from tacit linguistic knowledge (e.g., vocabulary size) to foster learning to read subskills in L2 learners will help fill the metalinguistic gap (Reference Nagy, Wagner, Muse and TannenbaumNagy, 2007) in classrooms composed of learners with various L1 backgrounds and L2 proficiency levels. If teachers adopt only tacit vocabulary knowledge instruction, there would be a Matthew effect, that is, learners coming to the classroom with a greater target language vocabulary size in the first place will benefit more than those bringing in a smaller vocabulary size to school, and any previously existing vocabulary gap could widen over time. Recent research has provided support for the effectiveness of metalinguistic (morphological) instruction in both L1 and L2 learners, and L2 learners may benefit more from metalinguistic instruction (e.g., Reference Goodwin and AhnGoodwin & Ahn, 2013). Metalinguistic awareness is also beneficial in promoting an analytical or words-as-tools approach in reading to learn in child or adult L2 learners (Reference Koda and YamashitaKoda & Yamashita, 2018; Reference Nagy and TownsendNagy & Townsend, 2012).

Another notable implication is that metalinguistic instruction, in relation to reading development, should aim to foster a full metalinguistic repertoire in L2 learners instead of any single facet of metalinguistic awareness (see the Repertoire Theory of Literacy Development, Reference Apel, Masterson, Hart, Silliman and WilkinsonApel et al., 2004; Reference Masterson and ApelMasterson & Apel, 2000). Convergent evidence has suggested that phonological awareness is a language-general facet of metalinguistic awareness, whereas orthographic and morphological awareness are language specific. This means that instruction in phonological awareness is less language dependent and an instruction focus on phonological/phonemic awareness can be very limited for bilingual children. Instruction in orthographic and morphological awareness should be very important and will involve more explanations of language-specific features. For example, English is morphophonemic in that the basic grapheme unit is an alphabetic letter, but printed words encode both phonemic and morphemic information, whereas Chinese orthography is morphosyllabic with the character, the basic grapheme, encoding the sound of a morpheme at the syllable level. Teachers should also (be taught to) be aware that the three facets of metalinguistic awareness play different roles from developmental and crosslinguistic perspectives. For instance, phonological awareness is an important predictor of word decoding at lower grade levels and refined morphological awareness does not surface till grade three or above in English readers. In contrast, orthographic and morphological awareness develop earlier in Chinese readers and are a more important predictor of reading comprehension in Chinese than phonological awareness. More recently, morphological awareness instruction in English has also been found to be effective and feasible for kindergartners and first- and second-grade students from low-socioeconomic-status homes (Reference Apel and DiehmApel & Diehm, 2014; Reference Apel, Brimo, Diehm and ApelApel et al., 2013). A possible pathway to promote a metalinguistic repertoire in L2 learners is to promote self-teaching during reading via phonological decoding, high-quality spelling and writing, and morphological decoding and analysis, which reinforce the utilization of phonological, orthographic, and morphological awareness, respectively (e.g., Reference Li, Li and WangLi et al., 2020).

The implications for L2 reading instruction discussed above are based on evidence reviewed in this Element as well as our understanding of the recent development in the science of reading (see the special issue The Science of Reading: Supports, Critiques, and Questions in Reading Research Quarterly, edited by Reference Goodwin and JiménezGoodwin & Jiménez, 2020). Still, there is an urgent need for more collaboration between L2 reading/biliteracy researchers and teachers to conduct interventional research that verifies the causal effects of metalinguistic awareness in L2 reading, and to identify feasible and effective assessments and instructional approaches for cognitively and linguistically diverse multilingual learners.