ABSTRACT Title of Dissertation: PHONOLOGICAL REPRESENTATION AND PROCESSING IN BILINGUAL LEARNERS: INVESTIGATING CROSS- LANGUAGE PHONOLOGICAL ACTIVATION AND ACCURACY OF PHONOLOGICAL REPRESENTATION Nan Zhang, Doctor of Philosophy, 2025 Dissertation directed by: Dr. Min Wang, Department of Human Development and Quantitative Methodology Phonological activation and processing constitute a pivotal area of inquiry within second language (L2) research. The intersection and distinctiveness of phonological representation across languages leads to cross-linguistic phonological activation and challenges in L2 phonological processing. A large volume of previous evidence suggests that cross-linguistic phonological activation occurs through homophone and cognate priming in processing of isolated words. However, investigations into sentence-level processing remain sparse. Moreover, the significance of cross-linguistic phonological priming and its consistency across various languages, task modalities, and priming directions have yet to be fully elucidated. Concerning the challenges faced by L2 learners in phonological processing, previous research suggests that the difficulties may lie in an inaccurate, imprecise phonemic representation and perception in both auditory and visual word recognition. The primary objectives of this dissertation are: (1) to investigate cross-linguistic phonological activation between Chinese and English in both isolated word and sentence processing contexts (Study 1); (2) to assess the significance and consistency of phonological activation in bilinguals in a meta-analytic review (Study 2); (3) to examine whether and how Chinese ESL learners show fuzzy phonological representation in both auditory and visual word recognition tasks and the role of proficiency in moderating the accuracy (Study 3). The findings from these studies advance our understanding of the organization in bilingual lexicon and interaction across languages, with a particular focus on phonological representation and processing across Chinese and English, and the nuances specific to learning English L2. Phonological representation and processing in bilingual learners: investigating cross- language phonological activation and accuracy of phonological representation by NAN ZHANG Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2025 Advisory Committee: Professor Min Wang, Chair Professor Patricia Alexander Professor Donald J. Bolger Professor Kira Gor Professor Nan Jiang (Dean’s Representative) © Copyright by Nan Zhang 2025 ii Acknowledgements The completion of this dissertation would not have been possible without the unwavering guidance, support, and encouragement of many individuals to whom I am deeply indebted. First and foremost, I owe an immeasurable debt of gratitude to Dr. Min Wang, whose intellectual leadership has illuminated every step of this dissertation. Her faith in my potential never wavered—she steadied my resolve whenever my own confidence flickered and refused to let me settle for less than my best. Dr. Wang pored over countless iterations of manuscripts, fellowship statements, and grant proposals with a surgeon’s precision, offering incisive comments that sharpened my arguments and elevated my methodology. Her strategic guidance on awards and funding opportunities opened doors I had scarcely imagined, while her relentless insistence on clarity, rigor, and organization reshaped the way I think, write, and conduct research. Through her example I have learned not only how to craft meticulous studies but also how to mentor others with the same blend of patience, high standards, and keen analytical vision that defines her own scholarship. I am thankful to Dr. Hong Jiao, my research-assistantship supervisor. Her keen insights into measurement and assessment, along with her generous support of my professional development, provided an indispensable foundation for the work presented here. I am especially fortunate to have had an inspiring dissertation committee. Professor Patricia Alexander welcomed me first as a student and later as her teaching assistant; her warmth and steady encouragement reminded me daily of the value of iii intellectual generosity. Professor Donald J. Bolger was one of the very first mentors I met at UMD and has remained a model for the kind of researcher I aspire to become. Professor Kira Gor—whose sharp research vision I came to admire during two unforgettable courses—played a pivotal role in honing the design of this dissertation. Professor Nan Jiang, serving as the Dean’s Representative (and sharing my first name), has been unfailingly supportive both inside and outside the classroom; his clarity in explaining theory and brilliance in experimental design first sparked the ideas that ultimately grew into this study. Their collective guidance, probing questions, and high expectations elevated this work immeasurably. My heartfelt thanks go to my peer and friend Zhiyi Wu. Our countless discussions—whether about methodology, coding quirks, or the inevitable highs and lows of graduate life—helped transform obstacles into opportunities and made this journey far less solitary. I also wish to acknowledge my undergraduate friends Yining Ma, Chunlin Li, and Mengyao Liu. Their steadfast friendship and enthusiastic moral support—despite the geographic distance and the passage of time—reminded me that genuine camaraderie endures and uplifts. Finally, I owe my deepest gratitude to my family. To my mother Aiping Li and my father Jun Zhang, whose boundless love, quiet sacrifices, and steadfast belief in my potential have sustained me at every turn—thank you for being my constant anchor. I also honor the cherished memory of my late grandparents, Wenrui Zhang and Naiying Li. The curiosity, resilience, and generosity they cultivated in me from childhood through my undergraduate years remain an inexhaustible source of strength iv whenever my resolve falters. This dissertation is, in no small part, a tribute to the values they instilled and the example they set. To all of you, and to everyone else who has cheered me on behind the scenes, thank you. This accomplishment is as much yours as it is mine. v Table of Contents Acknowledgements ....................................................................................................... ii Table of Contents .......................................................................................................... v Chapter 1: Introduction ................................................................................................. 1 Overview ................................................................................................................... 2 The Interaction of Phonological Representation and Processing in Dual Languages Bilingual Lexical Activation Models ........................................................................ 3 Cross-Language Phonological Activation in Single Word Processing ..................... 4 Masked Cross-Language Phonological Priming ................................................... 4 Factors Moderating Masked Cross-Language Phonological Priming .................. 5 Cross-language Phonological Activation During Sentence Processing .................... 7 The Distinctiveness in L2 Phonological Representation .......................................... 8 Fuzzy Phonological Representation in L2 Word Recognition ................................. 9 The Role of L1 Orthography ................................................................................. 9 Semantic Processing ........................................................................................... 10 Language Proficiency ......................................................................................... 10 The Current Dissertation ......................................................................................... 10 Summary ................................................................................................................. 12 References ............................................................................................................... 12 Chapter 2: Cross-Language Phonological Activation in Bilingual Visual Word Recognition: A Meta-Analysis ................................................................................... 18 Chapter 3: Automatic Phonological Access Among Bilinguals With Cross-Script Languages ................................................................................................................... 48 Chapter 4: Fuzzy Phonological Representation Among Chinese ESL Learners: The Effect on Auditory and Visual Word Recognition and the Role of L2 Proficiency ... 68 Abstract ................................................................................................................... 69 Introduction ............................................................................................................. 70 Fuzzy Phonological Representation in L2 Auditory Word Processing .............. 70 Fuzzy Phonological Representation in L2 Visual Word Processing .................. 73 The Role of L2 Proficiency ................................................................................. 76 The Current Study ............................................................................................... 78 Experiment 1: Auditory Semantic Categorization .................................................. 80 Participants .......................................................................................................... 80 English L2 Language Proficiency Test at Word Level ....................................... 81 Design and Materials .......................................................................................... 82 Procedure ............................................................................................................ 85 Results and Discussion ........................................................................................... 85 RT ....................................................................................................................... 85 Accuracy ............................................................................................................. 88 Discussion ........................................................................................................... 89 Experiment 2: Visual Semantic Categorization Task ............................................. 91 Participants .......................................................................................................... 91 Design and Materials .......................................................................................... 92 vi Procedure ............................................................................................................ 92 Results and Discussion ........................................................................................... 92 RT ....................................................................................................................... 92 Accuracy ............................................................................................................. 93 Discussion ........................................................................................................... 94 Comparing Experiments 1 and 2: The Difference of Fuzzy Phonological Representation in the Auditory and Visual Task .................................................... 95 Results and Discussion ........................................................................................... 95 RT ....................................................................................................................... 95 Accuracy ............................................................................................................. 96 Discussion ........................................................................................................... 97 General Discussion ................................................................................................. 98 Fuzzy Phonological Representation in L2 Learners ........................................... 99 The Role of Orthographic Similarity ................................................................ 100 Implications ....................................................................................................... 101 Chapter 5: Conclusion ............................................................................................... 129 Summary of Key Findings .................................................................................... 129 Automatic Phonological Activation .................................................................. 131 Phonological Fuzziness ......................................................................................... 131 Semantic Processing ......................................................................................... 132 The Role of L1 Orthography ............................................................................. 133 Implications ........................................................................................................... 133 Limitations and Future Directions ........................................................................ 134 1 Chapter 1: Introduction 2 Overview To elucidate the intricacies of bilingual language acquisition and processing, it is imperative to study the organization of phonological, orthographic, and semantic representations within the mental lexicon (Jiang, 2023). Phonological representation, in particular, is fundamental to the processing of words in both their spoken and written forms across alphabetic languages like English and logographic languages such as Chinese. This is because phonological processing is not only essential for auditory word recognition but also crucial for the connection between orthography and semantics in the recognition of written words (refer to Frost, 1998; Meade, 2020 for comprehensive reviews). The bilingual lexicon may exhibit both integration and separation, sparking a debate on whether the two languages are stored together or separately. This discussion is further complicated by the presence of shared phonemes across languages, as well as those unique to each language. For example, phonemes such as /a:/, /i/, and /s/ are common to both Mandarin Chinese and English, whereas phonemes like /æ/, /v/, and /θ/ are unique to English. The presence of language-specific phonemes introduces challenges in representation and distinguishing phonemic contrasts between the two languages. This dissertation comprises three studies that dissect the dual aspects of commonality and uniqueness in cross-language activation and processing. The first two studies focus on the commonalities, investigating cross-linguistic co-activation and positing that the bilingual lexicon is organized integratively, with the activation of the two languages being non-selective. Study 1 (Chapter 2) offers a meta-analytic review, while Study 2 (Chapter 3) is an empirical investigation. The third study (Chapter 4) pivots to the uniqueness of L2 phonological representation and processing, incorporating two experiments on spoken word recognition and 3 visual word recognition to examine the relationships between the precision of phonological representation and semantic processing, as well as the role of L2 proficiency in moderating the accuracy of phonological representation. The Interaction of Phonological Representation and Processing in Dual Languages Bilingual Lexical Activation Models A pivotal question within the domain of integrated lexical storage concerns whether bilinguals automatically and non-selectively activate lexicons in both languages simultaneously. This inquiry posits that when bilingual individuals read in one language, they concurrently activate corresponding word candidates in the other language (Brysbaert, 2003; Dijkstra & Kroll, 2005; van Heuven & Dijkstra, 2023). To elucidate the mechanism of bilingual phonological activation during visual word recognition, the Bilingual Interactive Activation Plus Model (BIA+; Dijkstra & van Heuven, 2002) and the Multilink Model (Dijkstra et al., 2019) have been proposed. Both frameworks advocate for an integrated lexicon and parallel activation, suggesting that orthographic, phonological, and semantic representations of two languages coexist within a unified mental lexicon. A language code is instrumental in determining the language membership of visual inputs, enabling non-selective activation of all lexically and semantically related words across languages in response to a specific stimulus (e.g., presenting the French word "fille" simultaneously activates the English word "fee"; Friesen et al., 2020). Task schema, embodying a sequence of cognitive operations aimed at achieving a predetermined goal, plays a crucial role in mediating lexical processing and behavioral responses, tailored to the contextual demands and specific task requirements. For instance, tasks that heavily rely on phonological processing would necessitate a heightened activation of phonological representation (Lam & Dijkstra, 2010). 4 Cross-Language Phonological Activation in Single Word Processing Investigation into cross-language phonological activation has been carried out using both isolated word processing and sentence processing paradigms (Friesen et al., 2020; Yan et al., 2023). Within the realm of visual word recognition research, masked priming paradigm has been used widely and considered one of most stringent methodologies in studying cross-language phonological activation (e.g., Brysbaert et al., 1999; Dimitropoulou et al., 2011; Duyck et al., 2004; Kim & Davis, 2003; Nakayama et al., 2012; Zhou et al., 2010). The findings from this line of research underscore the intricate dynamics of phonological activation across languages, supporting for the integrated and parallel nature of lexical processing in bilinguals. Masked Cross-Language Phonological Priming Masked priming tasks involve the target word's presentation following a mask—a string of symbols—and a briefly displayed prime. The mask's role is to inhibit participants' conscious recognition of the prime, thus mitigating strategic and biased processing of these primes (Foster et al., 2003), providing robust evidence for non-selective activation. Researchers probe phonological activation across languages by manipulating the phonological relatedness between primes and targets (related vs. unrelated) and examining differences in reaction times (RT) and accuracy in participants’ responses. Faster and more accurate responses to targets preceded by phonologically related primes suggest that phonological representation is shared and automatically activated across languages. For example, Kim and Davis (2003) employed masked priming word naming paradigm with Korean-English bilinguals. In the phonologically related condition, primes were interlingual homophones of the targets (e.g., 풀—pull, meaning grass-pull), while in the unrelated condition, primes differed phonologically from the targets (e.g., 각-pull, meaning each-pull). The primes in 5 the related condition revealed a facilitative phonological priming effect through shortened reaction time (RT) in the related condition. Other studies have measured the facilitation of phonological primes through higher accuracy in the related condition across different language pairs (Dutch-French: Duyck et al., 2004; Hebrew-English: Gollan et al., 1997). However, findings regarding the significance and magnitude of phonological priming effects vary. For instance, lexical decision-based phonological priming was not significant among Korean-English bilinguals (Kim & Davis, 2003), yet showed significant facilitation among Japanese-English bilinguals (Ando et al., 2014; Nakayama et al., 2012). These discrepancies likely arise from differences in task demands, priming direction, script distance, stimulus onset asynchrony (SOA), inter-stimulus interval (ISI), participant numbers, and item counts, which will be reviewed subsequently. Factors Moderating Masked Cross-Language Phonological Priming Priming Direction. The varied priming effects in different directions (from L1 to L2 or L2 to L1) suggest that there may be differences in the quality of lexical representation across languages. Given the better quality of representation and faster access to L1 phonological representation, priming effects are expected to be larger from L1 to L2. However, results from previous studies have shown complex patterns, with the L1 to L2 priming greater (e.g., Chinese– English, Zhou et al., 2010), equal to (e.g., Dutch–French, Van Wijnendaele & Brysbaert, 2002), or smaller than (e.g., Chinese–English, Xu et al., 2021) the L2 to L1 priming. Script Distance. Script distance is the visual distinctiveness between languages' writing systems, which is associated with orthographic overlap. Divergent views are held by different researchers on whether a large script distance enhances phonological priming by reducing orthographic inhibition, or results in suppression due to reduced orthographic similarity when 6 scripts differ, thus cross-language phonological priming observed in same-script languages may be weakened in cross-script languages. Task Type. The BIA+ and Multilink models suggest task type can influence lexical processing by differentially prioritizing phonological, orthographic, and semantic codes. Task variations, such as lexical decision, language decision, word naming, semantic categorization, or perceptual identification, provide varying insights into automatic cross-language activation. Word naming and lexical decision tasks, in particular, have been instrumental in examining phonological involvement in visual word recognition within bilingual contexts. Stimulus Onset Asynchrony (SOA) and Inter-Stimulus Interval (ISI). The timing between the prime and target presentation (SOA), including the prime duration and ISI, impacts the visibility of the prime and, consequently, the magnitude of phonological priming effect. Adjusting the SOA can either offset the prime's influence or, if too long, weaken phonological priming due to orthographic dissimilarity overshadowing shared phonological representation. The Number of Participants and Items. The number of participants and stimuli affects the statistical power and, therefore, the interpretability of results. Variation in sample size has led to differing outcomes in studies, pointing to the importance of adequate participant numbers in detecting significant phonological priming. In summary, while numerous studies have explored cross-language phonological priming, discrepancies are present, particularly in cross-script language pairs. Addressing these inconsistencies necessitates a comprehensive review of existing research and a detailed investigation into potential moderating factors. Hence, Study 1 focuses on refining experimental techniques to examine Chinese-English phonological priming in isolated word-level tasks including lexical decision and word naming tasks as well as in a sentence level task. Study 2 7 employs a meta-analytic approach to systematically assess the magnitude of effect and its potential moderators. Cross-language Phonological Activation During Sentence Processing While the nonselective activation hypothesis of the bilingual lexicon has been investigated in the context of visual single word recognition, real-world reading predominantly occurs within sentence contexts. The dynamics of word processing in sentences may diverge from those in isolated words. The sentence contexts in a monolingual mode provide strong language-specific constraints and thus potentially inhibit co-activation of the lexicon in the non- target language (Assche et al., 2012). Among the sparse research on sentence-level cross-language phonological activation, a notable study by Friesen, Whitford, et al. (2020) employed a homophone error disruption paradigm to assess whether phonological information is activated during sentence reading. The study monitored the eye movement of English–French bilinguals while reading English sentences silently. In the homophone error condition, a critical word within each originally correct sentence was substituted with its interlingual homophone—i.e., a word phonologically similar to the translation equivalent in French. Conversely, the control condition involved replacing the critical word with an orthographic neighbor of the homophone. For instance, in the sentence “Kristina’s new haircut made her look very pretty for her graduation,” “pretty” was identified as the critical word. The French translation for “pretty” is “belle,” whose interlingual English homophone is “bell.” Thus, in the homophone error condition, “pretty” was replaced by “bell,” while in the control condition, it was replaced by an orthographically similar word like “bent.” Their results revealed shorter total reading times for critical words in the homophone condition compared to the control condition, suggesting cross-language semantic facilitation 8 mediated by phonology. The unseen French homophone's meaning, mediated by shared phonology, facilitated sentence comprehension. This pattern was corroborated by another study (Yan et al., 2023) using the same paradigm, which utilized within-script language pairs (Cantonese L1 and Mandarin L2). Their results demonstrated that native Cantonese readers activate phonological representation of both Cantonese and Mandarin when reading in either language. Both studies by Friesen, Whitford, et al. (2020) and Yan et al. (2023) examined cross- language phonological priming focusing on language pairs using the same script. It remains uncertain whether their findings extend to bilinguals navigating languages with differing scripts. Hence, the first study in this dissertation sought to bridge this gap by investigating cross- language phonological activation among Chinese-English bilinguals during sentence reading in an English L2 context, in addition to the first two experiments testing the activation in a single word context. The investigation of sentence-level activation contributes to our understanding of the extent to which cross-language phonological activation occurs in a more complex reading context and across different scripts. The Distinctiveness in L2 Phonological Representation Adapted models of BIA+ acknowledge phonological units unique to one language that lack close approximations in the other language. For Chinese-English bilinguals, specific phonemes like the diphthong /ɔɪ/ and the voiced dental fricative /ð/ in English are absent in Chinese. Given the critical period hypothesis, for instance, the line regressing second-language attainment on age of immigration would be markedly different on either side of the critical-age point (Hakuta et al., 2003). Achieving a very high level of proficiency in the target language is often difficult for post-pubertal learners, although such an achievement cannot be ruled out 9 (Azieb, 2021). Therefore, acquiring precise phonological representation becomes challenging post this period, a common scenario for many Chinese L2 learners of English. According to the fuzzy phonological representation hypothesis (Gor et al., 2021), L2-specific phonemes may be represented imprecisely in the L2 mental lexicon, leading to challenges in listening and reading. Fuzzy Phonological Representation in L2 Word Recognition In listening, auditory signals are converted into phonological representation, a process known as phonological encoding. Imprecise representation hinders this process and affect listening comprehension. Studies show that L2 learners often treat L2 minimal pairs as homophones if the phonemic distinction does not exist in their L1, as a result of inaccurate phonological representation of language-specific phonemes (e.g., Cook et al., 2016; Pallier et al., 2001). The impact of fuzzy phonological representation on visual word processing is less studied compared to auditory word processing, especially among Chinese-English bilinguals. Similar to auditory processing, inaccurate phonological representation can impair phonological recoding during visual word processing, impeding semantic access and comprehension. Research with Japanese and Spanish learners of English L2 showed that phonological fuzziness negatively affects L2 visual word processing, due to difficulties associated with phonemic contrasts absent in L1 (Ota, 2010). The Role of L1 Orthography Interestingly, L1 orthography seems to influence the degree of phonological fuzziness in reading. For instance, Spanish learners of English, unlike their Japanese counterparts, did not struggle with phonemic contrasts that are not present in Spanish (Ota et al., 2010). This suggests that sharing a script between L1 and L2 might enable better differentiation of L2 minimal pairs 10 in writing, even if the phonemic contrast is absent in L1. With different scripts across Chinese and English, Chinese ESL learners may show a strong effect of phonological fuzziness in their L2. Study 3 will address this question. Semantic Processing The Logogen model (Morton, 1964; 1969) posits that phonological, orthographic, and semantic representations are crucial in the mental lexicon, with the first two considered lexical forms. Fuzzy phonological representation can impede semantic processing, as seen in studies across languages, such as English-Russian and Japanese-English and tasks, including semantic categorization and translation judgment tasks (e.g., Cook et al., 2016; Ota et al., 2009, 2010). Like bilinguals with other language pairs, Chinese-English bilinguals will also demonstrate the effect of phonological fuzziness on semantic processing, a hypothesis to be tested in Study 3 using semantic categorization tasks. Language Proficiency Language proficiency plays a moderating role in the impact of fuzzy phonological representation. Studies indicate that L2 learners of lower language proficiency exhibit more pronounced phonological fuzziness, while learners of higher proficiency are less affected by similar sounds (Cook et al., 2016; Darcy et al., 2013). Hence, language proficiency will be considered as a variable in Study 3, further elucidating its role in moderating phonological representation accuracy. The Current Dissertation The main objective of the current set of studies is to address existing gaps in the literature concerning cross-language phonological activation and its implications for L2 learning. By 11 adopting a multi-faceted research approach, this work seeks to illuminate the complex interplay between phonological representation and bilingual word recognitions. Study 1 (Chapter 2) utilized a meta-analytic approach to systematically review existing literature on cross-language masked phonological priming. The study sought to ascertain the overall significance and magnitude of the priming effect in bilingual visual word recognition. It investigated whether RTs and accuracy in phonologically related conditions are significantly shorter than those in unrelated conditions and examines how various factors—such as priming direction, script distance, task type, and procedural aspects (SOA and ISI), along with statistical power associated with the number of participants and items—moderate the priming effect. Study 2 (Chapter 3) was an empirical investigation that delves into cross-language phonological activation. This study examined the phenomenon both with isolated single words and within sentences. Employing lexical decision and word naming tasks, it studied priming effects from Chinese to English and vice versa (Experiments 1 and 2). Using a homophone error paradigm, the cross-language phonological priming was examined in the sentence reading context (Experiment 3). The aim was to understand how phonological representation and processing in one language can influence that in another, providing insights into the interconnectedness of bilingual lexicons. Study 3 (Chapter 4) investigated the representation and processing of phoneme contrasts that are difficult for Chinese-English bilinguals, using a semantic categorization paradigm in both auditory (Experiment 1) and visual (Experiment 2) modalities. This study aimed not only to replicate the impact of fuzzy phonological representation on auditory and visual word processing observed in other bilingual populations with Chinese-English bilinguals, but also to take a step further to assess its effect on visual word recognition by systematically examining the effect of 12 orthographic similarity and comparing the phonological fuzziness in listening and reading tasks. The role of language proficiency was also examined via comparing bilinguals of high versus low language proficiency. The first two studies focused on the interaction between phonological representations in two languages, while the third study focused on the distinguishment between the two languages’ phonological system. Summary Phonological representation across languages is posited to be stored together. On one hand, the shared phonemes across languages lead to cross-language phonological priming, supported by the accelerated response to a stimulus in the target language. However, empirical findings on this topic vary, potentially due to factors such as task type, script distance, priming direction, SOA, ISI, the number of participants, and the number of items per condition. Therefore, examining cross-language phonological priming in different settings can provide new empirical evidence. A meta-analytic synthesis of existing evidence could also enhance our understanding of the underlying moderators of cross-language activation. On the other hand, the language-specific phonemes in the integrated lexicon present challenges in phonemic discrimination and representation for L2 learners, affecting both auditory and visual word recognition. Examining the phonological fuzziness in different modalities of language processing and comparing them sheds light on our understanding of the crucial role of accurate phonological representation in language and reading development in general. References Ando, E., Jared, D., Nakayama, M., & Hino, Y. (2014). Cross-script phonological priming with Japanese Kanji primes and English targets. Journal of Cognitive Psychology, 26(8), 853– 870. https://doi.org/10.1080/20445911.2014.971026 13 Assche, E. V., Duyck, W., & Hartsuiker, R. J. (2012). Bilingual word recognition in a sentence context. Frontiers in Psychology, 3. https://doi.org/10.3389/fpsyg.2012.00174 Azieb, S. (2021). The critical period hypothesis in second language acquisition: A review of the literature. International Journal of Research in Humanities and Social Studies, 8(4), 20- 26. Boets, B., Vandermosten, M., Poelmans, H., Luts, H., Wouters, J., & Ghesquiere, P. (2011). Preschool impairments in auditory processing and speech perception uniquely predict future reading problems. Research in developmental disabilities, 32(2), 560-570. Brysbaert, M. (2003). Bilingual visual word recognition. Masked priming: The state of the art, 323-44. Brysbaert, M., Van Dyck, G., & Van de Poel, M. (1999). Visual word recognition in bilinguals: Evidence from masked phonological priming. Journal of Experimental Psychology: Human Perception and Performance, 25(1), 137–148. https://doi.org/10.1037/0096- 1523.25.1.137 Chung, K. K., McBride-Chang, C., Wong, S. W., Cheung, H., Penney, T. B., & Ho, C. S. H. (2008). The role of visual and auditory temporal processing for Chinese children with developmental dyslexia. Annals of dyslexia, 58, 15-35. Cook, S. V., Pandža, N. B., Lancaster, A. K., & Gor, K. (2016). Fuzzy nonnative phonolexical representations lead to fuzzy form-to-meaning mappings. Frontiers in psychology, 7, 188532. Darcy, I., Daidone, D., & Kojima, C. (2013). Asymmetric lexical access and fuzzy lexical representations in second language learners. The mental lexicon, 8(3), 372-420. 14 Dijkstra, T., & Kroll, J. F. (2005). Bilingual visual word recognition and lexical access. Handbook of bilingualism: Psycholinguistic approaches, 178, 201. Dijkstra, T., Peeters, D., Hieselaar, W., & van Geffen, A. (2023). Orthographic and semantic priming effects in neighbour cognates: Experiments and simulations. Bilingualism: Language and cognition, 26(2), 371-383. Dijkstra, T., & Van Heuven, W. J. B. (2002). The architecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition, 5(3), 175–197. https://doi.org/10.1017/S1366728902003012 Dimitropoulou, M., Duñabeitia, J. A., & Carreiras, M. (2011). Phonology by itself: Masked phonological priming effects with and without orthographic overlap. Journal of Cognitive Psychology, 23(2), 185–203. https://doi.org/10.1080/20445911.2011.477811 Duyck, W., Diependaele, K., Drieghe, D., & Brysbaert, M. (2004). The size of the cross-lingua masked phonological priming effect does not depend on second language proficiency. Experimental Psychology, 51(2), 116–124. https://doi.org/10.1027/1618-3169.51.2.116 Forster, K. I., Mohan, K., Hector, J., Kinoshita, S., & Lupker, S. J. (2003). The mechanics of masked priming. Masked priming: The state of the art, 3-37. Friesen, D. C., Ward, O., Bohnet, J., Cormier, P., & Jared, D. (2020). Early activation of cross- language meaning from phonology during sentence processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(9), 1754. Frost, R. (1998). Toward a strong phonological theory of visual word recognition: True issues and false trails. Psychological bulletin, 123(1), 71. https://doi.org/10.1037/0033- 2909.123.1.71 15 Gollan, T. H., Forster, K. I., & Frost, R. (1997). Translation priming with different scripts: Masked priming with cognates and noncognates in Hebrew–English bilinguals. Journal of Experimental Psychology: Learning, memory, and cognition, 23(5), 1122. Gor, K., Cook, S., Bordag, D., Chrabaszcz, A., & Opitz, A. (2021). Fuzzy lexical representations in adult second language speakers. Frontiers in Psychology, 12, 732030. Hakuta, K., Bialystok, E., & Wiley, E. (2003). Critical evidence: A test of the critical-period hypothesis for second-language acquisition. Psychological science, 14(1), 31-38. Kamhi, A. G., & Catts, H. W. (1986). Toward an understanding of developmental language and reading disorders. Journal of Speech and Hearing Disorders, 51(4), 337-347. Kim, J., & Davis, C. (2003). Task effects in masked cross-script translation and phonological priming. Journal of Memory and Language, 49(4), 484–499. https://doi.org/10.1016/S0749-596X(03)00093-7 Lam, K. J., & Dijkstra, T. (2010). Word repetition, masked orthographic priming, and language switching: bilingual studies and BIA+ simulations. International Journal of Bilingual Education and Bilingualism, 13(5), 487-503. Meade, G. (2020). The role of phonology during visual word learning in adults: An integrative review. Psychonomic Bulletin & Review, 27(1), 15–23. https://doi.org/10.3758/s13423- 019-01647-0 *Nakayama, M., Sears, C. R., Hino, Y., & Lupker, S. J. (2012). Cross-script phonological priming for Japanese-English bilinguals: Evidence for integrated phonological representations. Language and Cognitive Processes, 27(10), 1563–1583. https://doi.org/10.1080/01690965.2011.606669 16 Ota, M., Hartsuiker, R. J., & Haywood, S. L. (2010). Is a FAN always FUN? Phonological and orthographic effects in bilingual visual word recognition. Language and speech, 53(3), 383-403. Pallier, C., Colomé, A., & Sebastián-Gallés, N. (2001). The influence of native-language phonology on lexical access: Exemplar-based versus abstract lexical entries. Psychological science, 12(6), 445-449. Perfetti, C. (2007). Reading ability: Lexical quality to comprehension. Scientific studies of reading, 11(4), 357-383. van Heuven, W. J., & Dijkstra, T. (2023). Cross-language influences in L2 visual word processing. Cross-language Influences in Bilingual Processing and Second Language Acquisition, 16, 102. van Wijnendaele, I., & Brysbaert, M. (2002). Visual word recognition in bilinguals: Phonological priming from the second to the first language. Journal of Experimental Psychology: Human Perception and Performance, 28(3), 616–627. psyh. https://doi.org/10.1037/0096-1523.28.3.616 Wagner, R. K., & Torgesen, J. K. (1987). The nature of phonological processing and its causal role in the acquisition of reading skills. Psychological bulletin, 101(2), 192. Xu, G., Lin, J., & Dong, Y. (2021). Cross-script phonological activation in Chinese–English bilinguals: The effect of SOA from masked priming. Canadian Journal of Experimental Psychology/Revue Canadienne de Psychologie Expérimentale, 75(4), 374–386. https://doi.org/10.1037/cep0000262 17 Yan, M., Luo, Y., & Pan, J. (2023). Monolingual and Bilingual Phonological Activation in Cantonese. Bilingualism: Language and Cognition, 1-11. https://doi.org/10.1017/S1366728923000123 Zhou, H., Chen, B., Yang, M., & Dunlap, S. (2010). Language nonselective access to phonological representations: Evidence from Chinese-English bilinguals. The Quarterly Journal of Experimental Psychology, 63(10), 2051–2066. https://doi.org/10.1080/17470211003718705 18 Chapter 2: Cross-Language Phonological Activation in Bilingual Visual Word Recognition: A Meta-Analysis Vol.:(0123456789) Psychonomic Bulletin & Review https://doi.org/10.3758/s13423-025-02692-8 THEORETICAL/REVIEW Cross‑language phonological activation in bilingual visual word recognition: A meta‑analysis Nan Zhang1 · Zhiyi Wu2 · Min Wang1 Accepted: 14 March 2025 © The Author(s) 2025 Abstract Numerous studies have investigated whether phonological activation in the bilingual lexicon is selective or non-selective, using the classic masked priming paradigm that manipulates the phonological relatedness between primes and targets across two languages. The priming effects, however, are mixed: some studies reported reduced reaction times due to the homophone primes, while others observed non-significant priming. In this meta-analysis, we sought to systematically examine whether there is indeed cross-language phonological priming and to identify the factors that may moderate its magnitude. Analyzing 75 effects from 23 articles, we observed a significant, facilitative phonological priming effect (standardized mean difference Hedge’s g = 0.45, SE = 0.07, p <.0001, 95% CI = [0.32, 0.58]), hence supporting the hypothesis of non-selective activation. The moderators examined included priming direction (L1-to-L2 vs. L2-to-L1), task type (lexical decision vs. word nam- ing), script distance (within- vs. cross-script), stimulus-onset-asynchrony (SOA), inter-stimulus interval (ISI), number of participants, as well as number of items per condition. Results revealed a significant effect of task type in cross-script studies. Specifically, the word-naming task produces a smaller priming effect than the lexical decision task. Moreover, the prim- ing effect increases as the number of items in a condition increases. These results collectively suggested that phonological activation in the bilingual lexicon is nonselective, and the effect size is dependent upon task demands and statistical power, essential to a dual-language setting and in cross-language studies. Keywords  Bilingual word processing · Phonological priming · Meta-analysis Introduction To understand bilingual language acquisition and process- ing, it is crucial to study how the two languages are rep- resented and organized in the mental lexicon. One impor- tant question in this area is whether the two languages are stored integratively or separately, and a related question is whether bilinguals automatically activate the lexicons in both languages simultaneously (Brysbaert, 2003; Dijkstra & Kroll, 2005; van Heuven & Dijkstra, 2023). In the mental lexicon, phonological representation is one of the primary determinants (van Heuven & Dijkstra, 2023). Clarifying the organization of phonological information in two languages improves our understanding of bilingual lexical representa- tion and processing. In the line of research on visual word recognition, studies have shown that phonology plays an important role in mediating the path from orthography to semantics (for a review, see Frost, 1998; Meade, 2020). This mediating role of phonology has been supported in a myriad of monolingual empirical studies (for a review, see Rastle & Brysbaert, 2006) and computational models (e.g., Dual Route Cascaded model, Coltheart et al., 2001; for a review, see Norris, 2013). Cross-language phonological activation in bilingual visual word recognition, however, was largely over- looked until the publication of Gollan et al. (1997), Dijkstra et al. (1999), and Brysbaert et al. (1999). Since then, auto- matic cross-language phonological activation has been tested in both non-priming word recognition tasks (e.g., Carrasco- Ortiz et al., 2021; Comesaña et al., 2015; Dijkstra et al., 2010; Friesen et al., 2014; Lemhöfer & Dijkstra, 2004; Peleg et al., 2020; Sáchez-Casas et al., 1992; Schwartz et al., 2007) * Min Wang minwang@umd.edu 1 Department of Human Development and Quantitative Methodology, University of Maryland, College Park, MD 21072, USA 2 Graduate Program of Second Language Acquisition, School of Languages, Literatures, and Cultures, University of Maryland, College Park, MD, USA Psychonomic Bulletin & Review and priming tasks, especially the masked priming tasks (e.g., Ando et al., 2014; Brysbaert et al., 1999; Dimitropoulou et al., 2011; Duyck et al., 2004; Kim & Davis, 2003; Lim & Christianson, 2023; Nakayama et al., 2012; Wijnendaele & Brysbaert, 2002; Zhou et al., 2010). The current meta- analysis focused on evidence obtained from masked priming tasks where the target is presented following a mask (i.e., a string of symbols) and a brief presentation of the prime. In the masked priming paradigm, the primes are automatically processed, which prevents strategic responses, thus provid- ing compelling evidence for non-selective activation. A vast number of masked priming studies adhering to this standard design have been carried out to study lexical representation and processing during visual word recognition in the past decades. In our meta-analysis, we investigated the robustness of cross-language phonological priming under various con- ditions and identified factors that may moderate the strength of this effect. Theoretical models of bilingual lexical activation Various models have explained the representation and pro- cessing of two languages among bilingual speakers (e.g., Sense Model, Finkbeiner et al., 2004; Revised Hierarchical Model, Kroll & Stewart, 1994; Kroll et al., 2010). Among them, the Bilingual Interactive Activation Plus Model (BIA +; Dijkstra & Heuven, 2002), a verbal model, and the Mul- tilink Model (Dijkstra et al., 2019), a computational model, are two leading models contributing to our understanding of bilingual phonological activation during visual word recognition. The BIA + model proposes an integrated lexicon and par- allel activation, where the orthographic and phonological representations (collectively, lexical representations) of two languages are stored in a unified mental lexicon at both lexi- cal and sublexical levels. Additionally, semantic information is integrated across the two languages, and a language code decides the language membership of the visual input. The orthographic, phonological, and semantic representations, plus the language code, jointly form the word identification system. In this model, a specific stimulus non-selectively activates all related words in both languages that share lexi- cal and semantic representations with the stimulus (e.g., when the French word fille is given, the English word fee will be activated simultaneously; Friesen et al., 2020). Con- text and task requirements can mediate lexical processing and behavioral responses by assigning different weights to the different types of lexical representation. A task schema incorporates a series of cognitive operations or actions that lead to a pre-specified goal, for example, correctly perform- ing a word-naming task (Lam & Dijkstra, 2010). In this phonologically demanding task, the activation of phonologi- cal representation would be more necessary. Recently, the localist-connectionist Multilink model has been introduced (Dijkstra et al., 2019). Similar to the struc- ture of the BIA + model, the Multilink model incorporates a symbolic lexical network that contains layers of orthog- raphy, phonology, semantics, and language membership, as well as task/decision process. Going beyond the verbal explanation in BIA + model, it simulates the recognition of three- to eight-letter words including cognates of different lengths. The current Multilink model does not include the sublexical level, instead, focuses primarily on the interac- tions between lexical forms and their meanings, and pho- nological activation linked directly to these forms. In the BIA + model or Multilink model, there exists a feedforward from phonology and orthography to language nodes. How- ever, there is no top-down influence from language nodes to phonology and orthography. Furthermore, the Multilink model takes into account the frequency of usage, length, and cross-linguistic similarity of words in the two languages, which helps explain the effects of language proficiency and cross-language differences in bilingual lexical processing. Some of the critical assumptions made in the Multilink model are as follows: first, there are no lateral inhibitory effects between words within or between languages; second, language access is non-selective, and word neighbors are activated parallelly. In other words, upon receiving a visual input, the word itself and its interlingual homophone will be automatically activated without inhibition. Furthermore, the Multilink model emphasizes that the activation of competi- tors directly depends on the orthographic overlap between the input word and stored lexical representation. The greater orthographic overlap, the stronger activation of competitors. Cross‑language masked phonological priming In the studies that used the masked priming paradigm to investigate cross-language phonological activation in visual word recognition, researchers manipulated the phonological relatedness between primes and targets (related vs. unre- lated) and compared the reaction time (RT) and accuracy of participants’ responses to the targets in the two condi- tions. The faster and more accurate responses to the targets preceded by phonologically related primes indicate that phonological representation is shared across languages and automatically, non-selectively activated in both languages. To avoid strategic and biased processing of the primes, researchers use a mask to prevent participants from con- sciously recognizing the prime, ensuring subliminal process- ing (for a review, see Forster et al., 2003). Psychonomic Bulletin & Review Two types of phonologically contrasting conditions have been employed depending on whether primes and targets are semantically related. The first type involves the comparison between interlingual homophones and a set of graphemic control words, with no semantic overlap between primes and targets. The graphemic control words have no pho- nological overlap with the targets, but the orthographic overlap is similar between control words and targets and between homophone primes and targets (e.g., Dutch-French: Brysbaert et al., 1999; Duyck et al., 2004; Greek-Spanish: Dimitropoulou et al., 2011; Korean-English: Kim & Davis, 2003; Chinese-English: Zhou et al., 2010). For instance, in Brysbaert et al. (1999) where Dutch–French bilinguals were tested on a masked priming lexical decision task, the primes are interlingual homophones to the targets in the phonologically related condition (e.g., wie-OUI; translation: who-YES), whereas in the unrelated condition, the primes are phonologically different from the targets (e.g., jij-OUI; translation: you-YES). The primes in both the phonologi- cally related and unrelated conditions share the same letters at the same positions with the target words (in the phono- logically related example wie-OUI, the overlapping letter is i in the second position; in the control example jij-OUI, the overlapping letter is also i in the second position). A facilitative phonological priming effect was revealed by the higher accuracy in the related condition. Some other prim- ing studies measured the facilitation of phonological primes via faster RT in the related condition (Korean-English: Kim & Davis, 2003; Chinese-English: Zhang et al., 2024; Zhou et al., 2010). The second approach manipulates the phonological similarity of translation counterparts. Researchers either compare cognates (translation counterparts that have form overlap) with non-cognate translations while controlling for the degree of orthographic overlap between the cognate pairs (e.g., Chen et al., 2022; Gollan et al., 1997; Nakayama et al., 2012) or compare cognate pairs that are more phonologi- cally similar with those that are less phonologically similar (e.g., Nakayama et al., 2014). For example, Gollan et al. (1997) compared participants’ response latency to a target primed by its cognate (e.g., רטליפ/feelter/–FILTER) with that to a target primed by its non-cognate translation (e.g., armon/–CASTLE). Since the two conditions were/ןומרא not matched on word frequency, two groups of control words were created for the cognate and non-cognate conditions respectively (e.g., for cognates: ריגרג/gargir/, meaning berry; for non-cognates: תילגיס/sigalit/, meaning violet). The translation priming effect in the cognate condition was larger than in the non-cognate condition in terms of RT, pro- viding evidence for cross-language phonological priming. To date, although findings have been relatively consistent regarding the facilitative role of phonological overlap in the simultaneous activation across languages (for a review, see Jiang, 2023), the magnitude of the phonological priming effect as well as its moderating factors have not been sys- tematically examined. For example, Kim and Davis (2003) failed to observe a priming effect in a lexical decision task among Korean-English bilinguals, whereas Nakayama et al. (2012) and Ando et al. (2014) both showed a significant masked priming effect in Japanese-English bilinguals’ lexicality judgment. The discrepancy could be due to lan- guage pairs and the varied choice of masking and differ- ent priming procedures, including the type and duration of the mask, duration of the prime, and interval between the prime and the target. Taken together, there is a clear gap in the literature in clarifying the underlying factors that may impact phonological priming. In the following sections, we reviewed multiple potential factors, including those related to the stimuli (priming direction and script distance), task type, task procedure (SOA and ISA), and statistical power (number of items and participants). The extent to which these factors may moderate the cross-language phonologi- cal priming was evaluated. Moderators Stimuli Priming direction  Priming direction (from L1 to L2 or from L2 to L1) is a stand-out factor that has received ample attention in cross-language activation research. The varying priming effects in different directions may point to varying strengths of links between L1 and L2 lexical representa- tion and the varying quality of representation in the two languages. According to the Fuzzy Lexical Representation Hypothesis (Gor et al., 2021), phonological representation in L1 is stored in a higher quality and accessed more rapidly compared to that in L2, it is expected that priming would be larger in the direction of L1 to L2 than the other way around. Zhou et al. (2010) showed evidence supporting this hypoth- esis, with a larger priming effect from L1 to L2 (12 ms in word naming and 21 ms in lexical decision) than from L2 to L1 (8 ms in word naming and 18 ms in lexical decision). However, other studies, such as Xu et al. (2021), reported an opposite pattern, with a significant priming effect shown from L2 to L1 (14 ms) but not from L1 to L2 (9 ms) in a masked word-naming task under a similar SOA condi- tion (43 ms). Earlier findings in a Dutch–French study (van Wijnendaele & Brysbaert, 2002) added further nuance to this pattern, as the phonological priming shown from Dutch L1 to French L2 was similar to that from L2 to L1 (7 ms vs. 6 ms). Thus, the current meta-analysis sought to systemati- cally examine to what extent priming direction influences phonological priming. Psychonomic Bulletin & Review Script distance  A script is defined as an expression of the visual appearance of a written language (Perfetti & Dunlap, 2012). There are different types of scripts representing dif- ferent languages in the world, for example, European scripts (Roman, Cyrillic, Greek), Indian scripts (Hindi, Bengali, Tamil), Semitic scripts (Arabic, Hebrew), and Chinese/ Japanese traditional scripts. When two languages used by a bilingual share the same script category, they are consid- ered within-script. For example, English and Spanish, as well as Dutch and French, use the Roman alphabet and thus are classified as within-script. Conversely, if two languages employ different script categories, such as Hebrew (Semitic script) and English (Roman script), they are classified as cross-script. Languages employing different scripts tend to exhibit a reduced degree of orthographic overlap compared to those using the same script (Dijkstra et al., 2023; Miwa et al., 2014; Mountford, 2002). Researchers hold different viewpoints concerning the influence of orthographic distance on cross-language pho- nological priming. In favor of the selective lexical activa- tion hypothesis, Gollan et al. (1997) believed that when the scripts are different between the two languages utilized in a priming study, readers may use orthographic cues to intentionally access the specific lexicon of the prime and target. This allows for prompt access to the correct lexicon, increasing the chance of priming. In addition, a large script distance lowers the orthographic inhibition exerted on the words in the other language and enhances cross-language co- activation, thus leading to significant priming (Dimitropou- lou et al., 2011; Kim & Davis, 2003). However, some other researchers support the language non-selective hypothesis (e.g., Nakayama et al., 2012) and postulate that phonologi- cal representation and activation may be suppressed due to low orthographic similarity when the two languages do not share the same script. Consequently, the cross-language pho- nological priming observed in same-script languages may be reduced or even absent in different-script languages. Findings concerning cross-language phonological prim- ing are relatively consistent in within-script studies. For instance, Brysbaert et al. (1999) used a word perceptual identification task and found a significant phonological priming effect from Dutch L1 to French L2. In this task, a 42-ms display of a prime (either a real word or a pseudoword in Dutch) was followed by a target word (presented for 28 ms or 42 ms) and a postmask (“XXXX”). Participants were instructed to type the word they perceived. The primes were either phonologically similar (e.g., wie-OUI, meaning who- you in English) or unrelated (e.g., jij-OUI, meaning yes-you in English) to the target French real words. Each prime in the unrelated condition and its corresponding prime in the simi- lar condition had the same overlapping letter with the target (e.g., the letter i in the example) to minimize differences in orthographic similarity between the prime and the target. The phonological priming effect was calculated by subtract- ing the accuracy rate in the orthographic control condition from that in the phonologically similar condition. Results revealed a facilitative effect. This finding was replicated by van Wijnendaele and Brysbaert (2002) and Duyck et al. (2004) who examined phonological priming from French L2 to Dutch L1. In contrast to the consistent findings in within-script studies, evidence regarding phonological priming is mixed in cross-script studies. Significant phonological priming has been observed in various patterns: only from L1 to L2 (Hebrew-English: Gollan et al., 1997; Chinese-English: Zhang et al., 2024), only from L2 to L1 (Chinese-English: Xu et al., 2021), or in both directions (Chinese-English: Zhou et al., 2010). For instance, Zhou et al. (2010) showed significant bidirectional phonological priming among Chi- nese-English bilinguals in a word-naming task, whereas Xu et al. (2021) found no priming effect from Chinese L1 to English L2 using the same task. Comparing within- and cross-script studies, the orthographic characteristics of a language may play a role. However, the magnitude of the effect due to varying orthographic overlap has rarely been directly compared, except by Dimitropoulou et al. (2011), who observed a larger cross-language phonological prim- ing effect among Greek-Spanish bilinguals when the ortho- graphic overlap between the prime and the target was higher compared to when the orthographic overlap was lower. A systematic, quantitative comparison of phonological prim- ing between within-script and cross-script studies can help understand the extent to which the orthographic distance mediates cross-language phonological activation. Task type According to the BIA + and the Multilink models, task type in visual word recognition studies can modulate lexical pro- cessing by weighing the phonological, orthographic, and semantic codes differently based on specific task require- ments. These tasks include deciding the lexicality of a word (lexical decision) or the language of a word (language deci- sion), reading a word aloud (word naming), categorizing words based on semantics (semantic categorization), or writ- ing down words after a very brief presentation (perceptual identification). Empirical data from various tasks yielded different conclusions concerning whether automatic cross- language activation is present. Among these tasks, word naming and lexical decision have been the most frequently used word recognition tasks (Ferrand et al., 2011). These two paradigms are used to study the extent to which pho- nology is involved in visual word recognition (Katz et al., 2012) in both monolingual studies (Coltheart, 1985; Gao et al., 2016; Kinoshita & Norris, 2012; Zhang et al., 2018) Psychonomic Bulletin & Review and in bilingual studies (e.g., De Groot, 2011; De Groot et al., 2002; Haigh & Jared, 2007; Jared & Kroll, 2011; Jiang & Pae, 2020; Peleg et al., 2020; Tiffin-Richards, 2024; Xu et al., 2021). The contrast between these two most common tasks provides a better understanding of the nature of form activation in bilingual lexical processing. The difference in task demands between these two tasks stems from whether phonological information is necessary to complete the tasks. In a lexical decision task, where participants decide whether the target is a word, they could utilize phonological, orthographic, or semantic information to complete the task, making access to phonological representation optional. In contrast, a word-naming task requires participants to pro- nounce the target word, necessitating access to its precise phonological representation. Thus, comparing these two tasks could reveal whether phonological priming is modu- lated by task requirements related to phonological informa- tion. In addition to lexical decision and word naming, the perceptual identification task was used in the early phase of this line of research (e.g., Brysbaert et al., 1999; Brysbaert & Van Wijnendaele, 2003; Duyck et al., 2004; Van Wijnen- daele & Brysbaert, 2002). The current meta-analysis com- pared phonological priming effects across these three tasks: lexical decision, word naming, and perceptual identification. Task procedure Stimulus onset asynchrony (SOA) and inter‑stimulus interval (ISI)  The SOA refers to the time interval between the onsets of the prime and the target, consisting of the prime duration and the ISI (i.e., the interval between the offset of prime and the onset of target). The length of SOA determines the time allowed for participants to perceive and process the prime words in priming studies, which in turn affects the prime’s visibility and the priming effect size. The insertion of an ISI is known to increase the visibility of the prime (Forster et al., 2003). The relationship between SOA and the magnitude of the priming effect is complex. Firstly, a short SOA reduces the prime’s visibility, but an excessively brief SOA could lead to partial or null phonological prim- ing. For example, Brysbaert et al. (1999) found a facilita- tive effect of homophone primes with an SOA of 42 ms but not when it was 27 ms. Secondly, within a specific range, the phonological priming effect becomes larger as the SOA increases, because a longer SOA grants participants more time to access and identify the prime. Forster (1999, 2003) revealed that the magnitude of priming effects increases lin- early with the growth of SOA when the SOA is longer than 20 ms and shorter than 50 ms in repetition and orthographic priming studies, while the priming effect grows nonlinearly as the SOA exceeds that range (e.g., 60 ms). Thirdly, an extremely long SOA might decrease the phonological prim- ing effect, since the unmatched orthography suppresses the shared phonological representations. For instance, Lukatela and Turvey (1994) observed that an English target word was primed by a homophone of its semantically associated word (e.g., frog by toad-towed) at an SOA of 57 ms, but this prim- ing effect vanished when the SOA reached 250 ms. However, phonological priming was still shown at long SOAs with pseudoword primes (cf. Drieghe & Brysbaert, 2002). The effects of SOA and ISI were supported by a meta- analysis on the masked non-cognate translation priming effect (Wen & van Heuven, 2017), where SOA and ISI accounted for 26.25% of the variance in translation priming effects from L1 to L2. Another meta-analysis on the same topic (Davis & Kim, 2021) showed that studies using SOAs longer than 60 ms had significantly larger translation prim- ing effects than those shorter than 60 ms. However, no pre- vious systematic review has examined how cross-language phonological priming is moderated by SOA and ISI. In the current meta-analysis, we included them as two separate moderators in the regression analysis to study their indi- vidual influence on phonological priming. Statistical power The number of participants and the number of stimuli deter- mine the statistical power of an analysis. For instance, in a classic study on cross-script phonological priming (Kim & Davis, 2003), researchers failed to show homophone priming from Korean to English, possibly due to the small number of participants (N = 18). Adopting a similar set of materi- als, Lim and Christianson (2023) increased the sample size of participants to 60 and identified a significant phonologi- cal priming effect. These different findings suggest that the number of participants may play an important role in the significance of statistical tests. We sought to systematically examine whether this factor indeed affects phonological priming across a wide range of studies. Previous meta‑analyses on phonological priming Previous meta-analyses have examined semantic activa- tion in monolingual studies (van den Bussche et al., 2009) and non-selective activation of semantics (Davis & Kim, 2021; Lauro & Schwartz, 2017; Wen & van Heuven, 2017) in bilingual studies. However, cross-language phonological activation has only been synthesized in narrative reviews (e.g., Brysbaert, 2003; Desmet & Duyck, 2007; Jiang, 2018, 2023). The first narrative review on automatic cross- language phonological activation in visual word recogni- tion (Brysbaert, 2003) examined the evidence for three hypotheses based on two masked perceptual identification Psychonomic Bulletin & Review studies (in total three experiments; Brysbaert et al., 1999; van Wijnendaele & Brysbaert, 2002). These hypotheses state that (1) phonology plays a similarly crucial role in mediating L2 visual word recognition as it does in L1, (2) L1 words can prime L2 homophones, and (3) L2 words can prime L1 homophones. This review provides a detailed historical per- spective on research concerning phonological processing in both L1 and L2. While the scope of studies included is lim- ited to Dutch and French, reflecting the available research at the time, it meticulously compiled and analyzed the findings from the available literature to offer valuable insights into the field. After Brysbaert’s (2003) work, there were more narrative reviews to synthesize subsequent studies on cross- language phonological activation (e.g., Desmet & Duyck, 2007; Jiang, 2018, 2023). However, these reviews covered a wide aspect of bilingual processing, such as orthography, phonology, and morphology, with phonological activation being a small section. In addition, these reviews did not con- duct a systematic search or perform a quantitative synthesis of previous research on phonological activation. Albeit the lack of meta-analyses specifically focused on cross-language phonological activation, the previous meta- analytic reviews of monolingual phonological processing, particularly those focusing on masked priming tasks, can provide us with valuable insights. One such meta-analysis investigated the robustness of masked phonological prim- ing in monolingual English speakers (Rastle & Brysbaert, 2006). This review systematically scrutinized the literature on masked priming and categorized 20 eligible studies (107 effect sizes) into five groups based on task and procedure: forward-masked perceptual identification, backward-masked perceptual identification, forward-masked reading aloud, forward-masked visual lexical decision, and text reading tasks. Within each group, the researchers found an overall statistically significant homophone priming effect across the studies. However, the effect sizes across the five groups of studies were not compared with each other. Additionally, the potential moderators that may help explain the hetero- geneity in priming were not examined, such as the priming paradigm (i.e., forward- vs. backward-masked), task demand (i.e., perceptual identification, reading aloud, vs. visual lexi- cal decision), or context (i.e., isolated word vs. text reading). Furthermore, this meta-analysis solely examined monolin- guals whose English was L1, making it difficult to generalize the findings to other languages and bilingual phonological activation. More recent meta-analyses have examined cross-language activation with a focus on masked semantic priming (van den Bussche et al., 2009) and masked non-cognate trans- lation priming (Davis & Kim, 2021; Wen & van Heuven, 2017). These studies compared the pooled priming effects against zero and investigated potential moderators via meta-regression. Wen and van Heuven (2017) analyzed non-cognate translation priming effects in 64 masked prim- ing lexical decision experiments from 24 studies and showed a significant non-cognate translation priming effect both from L1 to L2 (standardized mean difference d = 0.86) and from L2 to L1 (d = 0.31). Seven moderators were examined, including the number of participants, the number of items per cell, the prime duration (in ms), the ISI (in ms), the SOA (in ms), the overall response speed (average reaction time across experimental and control conditions), and the script type (same- vs. different-script). Both ISI and SOA were sig- nificant predictors for L1-to-L2 priming, whereas the num- ber of items per cell was a significant predictor for L2-to-L1 priming. Davis and Kim (2021) synthesized masked prim- ing studies (76 effects) on non-cognate translation activation in two priming directions. A significant priming effect was found for both directions, with a significantly larger effect from L1 to L2 than from L2 to L1 (L1 to L2: d = 0.79, L2 to L1: d = 0.29). The priming effects were then regressed on three moderators: script type (same- vs. different-script), SOA (≤ 60 ms vs. > 60 ms), and reading experience (high vs. mid-low). The three moderators combined significantly predicted the effect sizes from L1 to L2, but not from L2 to L1. Findings from the above meta-analytical reviews of semantic and translation priming provide valuable insights into the potential moderators of cross-language phonological activation in the current study. The current meta‑analysis The following questions guided our meta-analysis: (1) Is there an overall significant masked cross-language phono- logical priming effect in bilingual visual word recognition? In other words, are RTs1 in the phonologically related con- dition significantly reduced compared to those in the pho- nologically unrelated condition? (2) Is the priming effect moderated by characteristics of stimuli (priming direction and script distance), task type, task procedure (SOA and ISI), and statistical power (numbers of participants and items in each condition)? We also investigated if there are signifi- cant interactions among priming direction, script distance, and task type. Based on previous literature, we hypothesized an overall significant masked phonological priming in RT. Second, we predicted that the priming effect would be influenced by the following moderators: stimuli, task type, task procedure, and statistical power. For example, 1  In this meta-analysis, we examined the priming effect with RT as the primary outcome measure. However, for the six studies that only reported accuracy, we incorporated accuracy as an alternative out- come metric. Psychonomic Bulletin & Review the priming effect may be larger from L1 to L2 than that from L2 to L1. Furthermore, we hypothesized that the effects of priming direction and task type might be differ- ent between within-script studies and cross-script stud- ies. In cross-script studies, phonological priming from L1 to L2 may be larger than that from L2 to L1, while this asymmetry may not be present in within-script studies. This hypothesized discrepancy is based on the potential suppression of phonological activation due to low ortho- graphic similarity in cross-script languages. The extent of this suppression might vary depending on how much L1 and L2 differ in their orthography-phonology linkage, that is, the possibility and strength of activation from orthog- raphy to phonology. When L1 and L2 share the script, the orthographic similarity between the prime and the target is relatively high, leading to a comparable strength of the orthography-phonology links in both languages. When L1 and L2 employ different scripts, the lower prime-target orthographic similarity may lead to a differential strength of the link between orthographic form and phonological representation in the two languages. Thus, the phonologi- cal activation from the more proficient L1 is likely to be stronger than that from the less proficient L2, resulting in an effect of priming direction. Methods Literature search and inclusion criteria We included 23 studies (k1 for the number of studies hereinafter) that yielded 75 effects (k2 for the number of effects hereinafter) in the meta-analysis. We searched and screened the literature following the suggestion by Pre- ferred Reporting Items for Systematic Reviews and Meta- Analyses statement (PRISMA 2020 version; Page et al., 2021). A flow chart illustrating the procedure of the litera- ture search and screening is presented in Fig. 1. Literature Search: ERIC (via EBSCO), APA PsycINFO (via EBSCO), Linguistics and Language Behavior Abstract (LLBA; via ProQuest), ProQuest Global Dissertations and Theses (PGDT), PubMed, and Web of Science Records after duplicates removed (n = 4839) Abstracts selected according to keywords (cross- language, phonological, or priming) (n = 196) Abstracts excluded (n = 4407) Inclusion Criteria 1) Reported original empirical data for cross-language phonological priming; 2) Used word naming/lexical decision task; 3) Participants: adults with normal reading abilities; 4) Used homophones as stimuli, at the lexical level; 5) Peer-reviewed or thesis written in English. Full-text articles assessed for eligibility (n = 57) Full-text articles excluded (n = 171) Reasons: 1) Did not manipulate phonological similarity as a variable; 2) Not a priming study; 3) Not a cross-language study; 4) Used auditory stimuli or picture stimuli; 5) Overlapping samples with other studies. Hand searched (n = 3) Studies included in meta-analysis (n = 23) Fig. 1   Flow diagram for the literature searching and inclusion criteria Psychonomic Bulletin & Review Search strategy A comprehensive search was conducted in six online data- bases including ERIC (via EBSCO), APA PsycINFO (via EBSCO), Linguistics and Language Behavior Abstract (LLBA; via ProQuest), ProQuest Global Dissertations and Theses (PGDT), PubMed, and Web of Science. Our search strategy involved using the keywords bilingual OR cross- language OR"cross language"OR cross-script OR"cross script"OR second-language OR"second language"AND pho- nolog* OR homophone* AND priming OR prime* in the full text of all databases except for PGDT, where we searched within abstracts only. This search strategy returned 4,839 articles published between 1970 and August 2023. We used Rayyan (http://​rayyan.​qcri.​org; Ouzzani et al., 2016) to fur- ther screen the article manually. After removing duplicates, 4,407 records were retained. Then, 57 potentially eligible articles were retained based on the full text. Using the inclu- sion criteria which are discussed in the following section, we retained 20 articles for coding. After this round of full-text screening, we hand-searched reference lists and tracked for- ward citations of the included articles to identify additional studies that may have been missed and found three more studies. In total, 23 articles (k2 = 75) were included. Unpub- lished theses and dissertations were included to address the publication bias. Among the total 75 effects, 12 effects were from unpublished research (three dissertations). Eligibility criteria The included studies are limited to peer-reviewed journal articles, manuscripts in preparation, and master’s or doctoral theses written in English and met all the following criteria: 1. Participants were bilingual adults with normal read- ing abilities and aged between 18 and 60 years. Studies with a focus on populations with dyslexia, hearing, or visual impairments were excluded, so were studies that involved sign languages. 2. Studies tested a masked priming effect, with primes briefly displayed and masked (less than 100 ms; see the meta-analysis by van de Bussche et al., 2009 where the same criterion was used). Two studies were excluded due to a long ISI (Singh et al., 2022, of which the ISI was 550 ms) or a long prime duration (Lee et al., 2005, in which the SOAs were 140 ms and 250 ms; see Jiang, 2023, for the justification that primes are visible under a long SOA). 3. The prime and the target were in two different languages. Studies that looked at two scripts of the same language (e.g., Rao et al., 2011, which used Roman-script-tran- scribed Urdu to prime Urdu) were excluded. 4. Studies manipulated the phonological similarity between primes and targets. That is, primes and targets had either full or partial phonological overlap in the phonologically related condition and little phonological overlap in the unrelated condition. For instance, among the included articles, Jouravlev et al. (2014) and Timmer et al. (2014) examined the masked onset priming effect, where primes and targets shared the onset in the phonological-related condition; Nakayama et al. (2012) examined the homo- phone priming effect, where primes and targets share the syllable in the phonological-related condition. 5. Eligible studies controlled for the influence of ortho- graphic overlap, word frequency, and word length on the phonological priming effect. Since languages with the same script tend to have orthographic overlap among homophone pairs, only effects where orthographic simi- larity was controlled for were included in our analysis. For instance, in Brysbaert et al. (1999), the homophone pair “wie”–“OUI” has one letter “i” in common, so they added a graphemic control word “jij” that has the same letter “i” in the same position as in “wie”. In studies that compared cognates (phonologically similar) and noncognate translations (phonologically unrelated), the two types of words were matched with two groups of orthographic control words respectively. Furthermore, cognate and noncognate words as targets were matched on word frequency and length. 6. Studies used a lexical decision, a word-naming, or a per- ceptual identification task with written visual stimuli to investigate bilingual visual word recognition. Studies incorporating auditory stimuli or picture stimuli were excluded. 7. Studies provided either conditional means and standard deviations, or t or F statistics with the degree of freedom of numerator as 1 (e.g., Rosenthal, 1995). In cases where a study met other inclusion criteria but failed to provide necessary statistics for calculating the synthesized effect, the first author contacted the authors of that study for more information. Out of the four inquiries made, two (Li, 2021; Liu et al., 2023) supplied the necessary infor- mation for our current meta-analysis, while the other two studies (Lee-Kim et al., 2021; Voga & Grainger, 2007) were excluded due to insufficient statistical data for effect size calculation. Coding process In line with previous meta-analyses on masked priming studies (e.g., Rastle & Brysbaert, 2006; Van den Bussche et al., 2009; Wen & van Heuven, 2017), we developed a coding manual that encompassed the identification num- ber (ID) of each effect (e.g., the Chinese-English phono- logical priming in lexical decision in Zhou et al., 2010), Psychonomic Bulletin & Review moderator, and the size of priming effect (e.g., 10 ms for the first effect in Zhou et al., 2010). The first and the sec- ond authors independently carried out the coding process for all included studies and cross-checked the results, adhering to Wilson’s (2009) recommendation for sys- tematic coding. The inter-rater reliability for all studies yielded a high level of agreement, with a 94.76% con- sensus between the two coders. All discrepancies were resolved through discussion until consensus was achieved. Four types of moderators related to stimuli, task type, task procedure, and statistical power were included in the cod- ing and analysis. Stimuli  The first category of the moderators, stimuli, includes priming direction and script distance. Priming direction was coded as L1-to-L2 or L2-to-L1. The script distance was coded as within-script or cross-script depend- ing on whether the primes and targets use the same script. The languages involved in the meta-analysis are Chinese, English, French, Dutch, Hebrew, Korean, Japanese Kanji, Japanese Katakana, Russian, and Urdu. For alphabetic lan- guages, English, French, Dutch, and Spanish use Roman script; Russian uses Cyrillic script; Hebrew uses Arabic script; Urdu uses Persian script which was derived from Arabic (Mountford, 2002). Three Asian languages, Chi- nese, Japanese (only its Kanji and Katakana writing systems were utilized in the included studies), and Korean, all have their unique scripts. Therefore, the script distance of the 11 language pairs included in the meta-analysis was coded as follows: within-script pairs are Dutch–English, Dutch– French, and English–French; Cross-script language pairs include Chinese–English, English–Hebrew, Greek–Span- ish, Japanese Kanji–English, Japanese Katakana–English, Urdu–English, Russian–English, and Korean–English. The script type and a sample of written words of each language are provided in Table 1. Task type  The task type was coded as a categorical vari- able. Within the 23 included articles, there are three types of tasks, i.e., the lexical decision task, the word-naming task, and the perceptual identification task. Fifty effects from 16 studies used the lexical decision task, 19 effects from ten studies used the word-naming task, and six effects from four studies used the perceptual identification task. Task procedure  The task procedure includes SOA and ISI, which were treated as continuous variables. Statistical power  The final category of the moderators includes the number of participants as well as the number of items in the phonologically related and control conditions respectively. Only items included in the data analysis were considered. To compute the effect size, we incorporated conditional means, standard deviations, and t or F statistics. Procedure Effect size calculation The outcome of interest is the difference in RT between the phonologically related and the unrelated conditions. The phonological priming effect can be obtained in two ways. The first method is to compare the RT in the phonologi- cally related condition with that in the control condition. The calculation of 59 effect sizes followed this approach. Among these, the prime in the experimental condition in 55 effects had full phonological overlap with the target. In the other four effects (Jouravlev et al., 2014; Kim & Davis, 2003; Timmer et al., 2014), the prime and the target shared the onset in the phonologically related condition. The second method is to compare the cognate priming with the noncognate translation priming effects, applied to the remaining 16 effects. In these effects, the stimuli in the phonologically related condition were not matched for word frequency and length with those in the unrelated condition (or in high vs. low phonological similarity con- ditions, as in Nakayama et al., 2014). This discrepancy occurred because the examination of the phonological priming was not their primary focus. Instead, such lexical properties of the stimuli in each experimental condition were matched with those in two different control groups, respectively. For both approaches, a positive value of the coded effect size indicates a facilitatory phonological priming effect. More details for effect size calculation can be found in the following section. A positive value means a facilitatory effect. Note that for studies that did not report RT as the outcome (i.e., Brysbaert et al., 1999; Duyck et al., 2004; Haigh, 2008; van Wijnendaele & Brysbaert, 2002), the error rate (ER) was used instead. We used the following formulas to calculate the stand- ardized mean difference (SMD), Hedge’s g, which corrects the sample size bias by multiplying a correction factor J by Cohen’s d. J is calculated as follows: in which N is the number of participants in each effect. Depending on the available statistical data in each article, four equations were used to calculate Cohen’s d, reported below. If means and standard deviations of the phonologically related and control conditions are reported, we calculated dz as Cohen’s d via J = 1 − 3 4N − 9 , Psychonomic Bulletin & Review in which dz is the standarzed difference score for a within- subject design, obtained from the raw difference between two conditions divided by the SD of difference (Cohen, 1988; Jané et al., 2024). MRT1i is the mean RT in the con- trol condition, MRT2i is the mean RT in the phonologically related condition, and SDpi is the pooled standard deviation of the two groups which is obtained through: in which n is the number of participants, SD1i is the stand- ard deviation of RT in the control condition, and SD2i is the standard deviation of RT in the phonologically related condition, ri represents the correlation between RT in the two conditions. Given that correlations between the related and control conditions are rarely reported in priming studies, we assumed a moderate correlation between the two condi- tions (r = 0.5). To ensure the robustness of our results, we conducted a sensitivity analysis by varying r across different values (i.e., r = 0.1, r = 0.9; for the rationale, see Borenstein et al., 2009). After some algebraic approximation procedures, dz can also be calculated based on t statistic: or F statistic: in which n is the number of participants, and the t-value and F-value are from the comparison between conditions by participants (see also Lucas, 2000; Van den Bussche et al., 2009; Wen & van Heuven, 2017). Therefore, for studies that did not report means and SDs of two conditions for effect size calculation while t- or F- statistics are available (i.e., Ando et al., 2014, 2015; Brysbaert et al., 1999; Chen et al., 2022, Experiment 1; Choi et al., 2010; Dimitropoulou et al., 2011; Kim et al., 2020; Nakayama et al., 2012), dz is calcu- lated in this way. Note that only t with a degree of freedom (df) of 1 or F with a denominator’s df of 1 was used to obtain Cohen’s d. For non-significant results where t or F statistics were not reported, we used the p-value to estimate the t or F values. If p was not provided in non-significant results (i.e., Experiment 3 of Gollan et al., 1997), we assumed p = 0.5 (Follmann et al., 1992; Frühauf et al., 2013). The variance of dz was calculated via: dz = MRT1i −MRT2i SDpi , SDpi = √ SD2 1i + SD2 2i − 2ri × SD1iSD2i, dz = t √ 1 n , dz = √ F n , One study compared the cognate priming effect and non- cognate priming effect to obtain the phonological priming effect. The F/t value did not have a df of 1. Therefore, we used the descriptive data of every participant to get the cor- relation between conditions. We used the following formula to get the priming effect: where drm is another type of standardized difference score for a within-subject design (Lakens, 2013), obtained from the raw difference divided by the within-subject SD. MRT1i is the mean RT in the control condition, MRT2i is the mean RT in the noncognate condition, and MRT3i is the mean RT in the phonologically cognate condition. SD is the pooled standard deviation, which is calculated via the following formula: where sgain is the standard deviation of gain scores, and r is the correlation between the cognate and noncognate priming effects. The variance of drm was obtained through: Subgroup analysis  We performed three subgroup analyses to examine the differences in priming effects between the two priming directions (from L1 to L2 and from L2 to L1), among the three tasks (lexical decision, word naming, and perceptual identification), and between the two types of script distance (within- and cross-script). Additionally, since the perceptual identification task only reported accuracy as the outcome, we conducted a subgroup analysis comparing the priming effects based on RT as the outcome to those calculated by accuracy. Meta‑regression analysis The synthesized effects generally exhibit variability other than sampling errors, namely cross-effect heterogeneity. This heterogeneity affects the generalizability of our con- clusions, and detecting its source can offer more insights into the mechanisms underlying the synthesized outcome (Deeks et al., 2019). We constructed a random-effects model to examine the role of various moderators. Since 16 studies reported multiple effects, there is a likelihood of depend- ency among them, a robust variance estimation (RVE; vi = 1 n + dz 2 2n . drm = (MRT1i −MRT2i) − (MRT1i −MRT3i) SD , SD = √ sgain 2(1 − r) , vi = 2 ( 1 − r1 ) n1 + 2 ( 1 − r2 ) n2 + drm 2 2 ( n1 + n1 ) . Psychonomic Bulletin & Review Hedges et al., 2010; for a review, see Fernández-Castilla et al., 2020) was employed to obtain unbiased nested-effects variance estimates for meta-analyses with a small number of included studies (less than 50; Moeyaert et al., 2017). We first examined the main effects of the seven moderators. Among them, three categorical variables, including priming direction, script distance, and task type, were included in the model as dummy-coded fixed effects, with lexical decision task, cross-script, and priming from L2 to L1 as reference groups. Note that the perceptual identification task was not included in the meta-regression analysis as a type of task due to a scarcity of effect sizes available (k2 = 6). The number of participants, the number of items, SOA, and ISI2 were con- tinuous moderators in the random-effects model. We visu- ally observed interaction patterns for all three categorical variables, i.e., the priming direction, task type, and script distance, when plotting their conditional means. However, no interaction pattern was found among the four continuous variables. Subsequently, we ran another model including the interaction terms between the three categorical moderators. Given the small number of within-script studies, we carried out the regression analysis with priming direction, task type, SOA, ISI, the number of participants, and the number of items as predictors in only cross-script studies. The pooled cross-language phonological priming effect was to analyze the overall sample size regardless of condi- tions. The main effects model examined the role of each moderator without considering the interaction across mod- erators. The interaction effects model examined the interac- tion between task type and priming direction in addition to other factors since the interaction plot showed a trend of such interaction. The sensitivity analysis was to examine whether there were outlying effects or influential effects. Sensitivity analysis  To test the robustness of the conclusion of this meta-analysis, we ran a group of sensitivity analyses. First, we left the outlying and influential effect sizes out one at a time (i.e., leave-one-out method, for an example, see Spence et al., 2021) to examine whether excluding each potentially outlying or influential effect leads to a different pattern of regression results. We identified effect size outliers and influential cases by calculating the external standardized residuals, hat values (h), Cook’s distance (Cook’s D), and DFBETAS using the metafor package (Viechtbauer, 2010), following the guidance in Viechtbauer and Cheung (2010). The external studentized residuals (also known as studentized deleted residuals) indicate the deviation of the observed effect size from the predicted value for the ith study based on the model without it. A large external studentized residual (e.g., exceeding an absolute value of 1.96) would indicate an outlying case. Regarding influential points, hat values indicate the influence of a case on the actual effect size, Cook’s distance measures the effect of deleting the ith effect on the fitted values of all k2 studies, and DFBE- TAS reveals the change of coefficients of each predictor once removing an effect. Then, we used the leave-one-out analysis to investigate the influence of potential outliers and influential effects on the regression outcome. We compared the between- study variance (τ2), the proportion of τ2 changed in the deleted model (R2), the variability among regression residuals (QE), and the reduced QE after deleting one effect in the regression (ΔQE) of each deleted model (Hedges & Olkin, 2014; Viech- tbauer & Cheung, 2010) to determine whether the effect is influential. Since the correlation between related and control con- ditions is not known precisely when imputing the pooled standard deviation of the two groups, we tested with a range of plausible correlations and used a sensitivity analysis to see how different correlations impact estimating effect size dz (Borenstein et al., 2009). Effect sizes were re-calculated and regression models were rerun after assigning different values (i.e., 0.1, 0.9) to correlation r. Publication bias To mitigate issues arising from potentially disproportionately reported findings (e.g., more significant facilitative phonologi- cal priming effects were reported in publications), we evalu- ated the possibility of publication bias in the included studies. We first utilized a funnel plot to visually inspect for small- study effects, which suggests that studies with a smaller sam- ple size tend to exhibit larger effect sizes (Sterne & Egger, 2005). In the funnel plot (Fig. 2a), the residuals of a full mul- tilevel meta-regression model were plotted against the sample sizes (Sterne and Egger, 2005; for an example, see Van den Bussche et al., 2009). Then, we executed a statistical test to further identify any evidence of publication bias. We used the multilevel meta-regression method (Nakagawa et al., 2022) in which the square root of the inverse sample size (  √ 1 n  ) was included as a predictor in the full multilevel meta-regression model