A Multilevel Model of Lexical-Phonological Interaction in L1-L2 Shift

Thesis Abstract

This study investigates how lexical representations and phonological processing interact across language shift from L1 to L2, situating this interaction within a multilevel framework that integrates psycholinguistic and sociolinguistic factors to explain variability in lexical access, phonological encoding, and cross-language transfer. The problem addressed is the incomplete accounting of multilevel determinants—ranging from individual cognitive architecture to discourse-level exposure—that shape L1-L2 lexical-phonological dynamics during bilingual competence development. The aim is to develop and validate a comprehensive multilevel model that specifies mechanisms by which lexical representations in L1 influence L2 phonological encoding and how phonological adjustments feedback to lexical access under conditions of L1 attrition and L2 proficiency growth. Specific objectives are (1) to quantify the relationship between L1 lexical organization and L2 phonological encoding across proficiency bands; (2) to evaluate the role of phonotactic constraints and phonological neighbourhood density in cross-language activation; (3) to examine bilingual exposure, language use, and sociolinguistic identity as moderators of lexical-phonological interaction; (4) to test the multilevel mediation of cognitive control, working memory, and typological distance on L1-L2 interaction; and (5) to propose a formalized model with testable hypotheses for predicting transfer Direction and magnitude in lexical-phonological processing. The study adopts a mixed-methods design combining quantitative experimental tasks with qualitative interview data. The population comprises Mandarin–English bilingual adults (N = 120) across intermediate to advanced L2 proficiency (CEFR B1–C1), balanced for age, education, and language exposure. A stratified sampling approach yields three proficiency strata (40 participants each). Data collection instruments include a lexical decision and masked priming paradigm to measure cross-language activation, a naming task to assess L2 phonological encoding efficiency, a corpus-derived lexical neighborhood analysis, and a structured interview protocol to capture language use, identity, and exposure variables. Additional measures include a working memory span task (operationalized via the reading span), a cognitive control task (Stroop paradigm), and a sociolinguistic questionnaire. Validity and reliability are addressed through pilot testing (n=20), established task paradigms, and internal consistency checks (Cronbach’s alpha > .80 for attitudinal measures). Data collection occurs in a controlled laboratory setting, with tasks counterbalanced to mitigate order effects. Analytical procedures employ hierarchical linear modeling (HLM) to capture within-subject and between-subject variance across lexical-phonological outcomes, along with structural equation modelling (SEM) to test the proposed mediation and moderation pathways integrating cognitive control, working memory, and sociolinguistic variables. Lexical decision and naming task data are subjected to mixed-effects regression analyses to assess fixed effects of language of instruction, lexical similarity, phonotactic probability, and cross-language overlap, with random intercepts for participants and items. Neural-network-inspired representation analyses are explored via multilayer perceptron models to simulate cross-language transfer and to compare predicted versus observed transfer magnitudes. Thematic analysis of interview data provides qualitative triangulation for identity, exposure, and usage patterns, informing the sociolinguistic moderators in the multilevel model. Expected findings include robust evidence for bidirectional lexical-phonological interaction, wherein L1 lexical organization and phonological constraints predict L2 encoding efficiency, and L2 phonological adjustments reciprocally influence L1 lexical access under high immersion and frequent code-switching; cognitive control and working memory are anticipated to mediate these effects, while sociolinguistic factors such as language dominance and daily usage moderate transfer intensity. The study contributes to knowledge by operationalizing a formal multilevel model that integrates lexical, phonological, cognitive, and sociolinguistic determinants of L1-L2 shift, enabling precise predictions about transfer direction and magnitude in bilingual lexical processing. It also advances methodological integration by combining psycholinguistic experimentation with sociolinguistic profiling and machine-learning simulations. The main conclusion will articulate the viability and limitations of the proposed multilevel framework in accounting for observed variability in L1-L2 lexical-phonological interaction, with recommendations for language-teaching interventions, assessment of bilingual competence, and directions for future research. Implications include refining models of bilingual lexical access, informing diagnostic tools for L1 attrition and L2 phonological intervention, and guiding curricula that acknowledge cross-language interference patterns and cognitive-linguistic moderators.