Leveraging AI to Uncover Linguistic Profiles and Inform Instruction for K–12 Multilingual Learners

• In Event: AI Innovations in Language and Literacy Education for K-12 Multilingual Learners

Abstract

Multilingual students in elementary schools bring richly varied linguistic and literacy repertoires shaped by their dynamic use of multiple languages across home and school contexts (de Bot, Lowie, & Verspoor, 2007; García & Wei, 2014). However, classroom assessments often fail to capture the full complexity of these learners’ abilities, limiting educators’ capacity to tailor instruction to their diverse developmental needs (Cummins, 2000; Lantolf & Thorne, 2006). This study aimed to leverage AI-powered diagnostic assessment to identify latent linguistic profiles among multilingual learners and examine instructional approaches that best support students within each profile. By integrating automated language processing (Burstein et al., 2020; Hirschberg & Manning, 2015) with literacy and oral language data, the study sought a more nuanced and actionable understanding of multilingual development in K–12 settings.

Participants included 423 multilingual students in Grades 3–6 enrolled in urban elementary schools in Canada. Students completed a series of oral language and literacy tasks—including oral reading fluency, story retell, picture description, reading comprehension, and narrative writing—administered through BalanceAI, an AI-driven diagnostic platform. The platform uses natural language processing and machine learning to provide real-time assessment feedback on key aspects of students' language development. To account for cognitive variability, students also completed short-term memory and fluid reasoning tasks. Background data on home language use, length of residence, and classroom learning behaviors were collected via surveys and teacher reports.

Latent Profile Analysis (Collins & Lanza, 2010), controlling for cognitive covariates, revealed three distinct learner profiles. One group demonstrated strong reading and writing skills but limited oral language proficiency, shaped by print-rich school environments and academic language exposure. Another group showed balanced oral language and literacy performance, frequently associated with moderate lengths of residence and cross-modal engagement. A third group exhibited strong spoken language and emerging literacy, often reflecting oral-rich home language practices and limited English print exposure. Further regression analyses identified home language use, years in Canada, and learning orientation as significant predictors of profile membership, illustrating the interplay of cognitive, linguistic, and sociocultural dimensions in shaping multilingual development. To explore instructional implications, classroom observations and teacher interviews were conducted in parallel. Findings highlight differentiated instructional strategies that align with each profile—such as leveraging oral scaffolds to support literacy in oral-dominant learners, embedding academic discourse practices for literacy-dominant students, and integrating home language resources to support balanced bilinguals.

By mapping language development patterns and corresponding pedagogical needs, this study demonstrates how AI-driven diagnostic profiling can bridge assessment and instruction. It offers a model for equity-oriented, data-informed teaching that affirms the assets multilingual learners bring and addresses their diverse pathways of development. In doing so, this research contributes to the growing field of AI in education by showing how technological tools can support more linguistically responsive, culturally sustaining approaches to K–12 language instruction.

Author

• Eunice Eunhee Jang, University of Toronto - OISE