THE ROLE OF GEMINI AI IN FACILITATING NOTICING AND RETENTION OF LANGUAGE FEATURES: A DESCRIPTIVE QUANTITATIVE ANALYSIS

Abstract

This study investigates how Google Gemini, a multimodal large language model (LLM), facilitates conscious
noticing and retention of vocabulary, collocations, and phrases among beginner language learners (A1-A2).
Grounded in Schmidt’s Noticing Hypothesis, the research employed a mixed-methods design involving 20
Indonesian university students aiming to assess (1) The effectiveness of Gemini-generated texts in directing
learner attention to target language features, (2) The relationship between initial noticing and retention over a
three-day interval, and (3) Learner perceptions of Gemini’s usability and pedagogical value. Quantitative
findings revealed a strong positive correlation between post-task noticing and delayed recall (r = .73, p < .01),
offering empirical support for the Noticing Hypothesis in AI-mediated learning contexts. Notably, A2 learners
demonstrated significantly higher retention rates (72.8%) than A1 learners (38.5%), highlighting the importance
of proficiency-sensitive content. While collocations (e.g., to join a club) were retained with high accuracy (95%),
abstract phrases (e.g., to volunteer) showed markedly lower retention (45%). Perception survey results (M ≥
4.3/5) confirmed Gemini’s utility in generating accessible, level-appropriate input, though learners expressed
moderate confidence in its retention efficacy (4.3/5). The study concludes that Gemini effectively scaffolds
noticing and retention for A2 learners but requires adaptive enhancements (e.g., grammatical scaffolding, bias
mitigation) to support A1 beginners. This study contributes to ongoing discussions about the role of AI in
second language acquisition (SLA) theory and practice, offering evidence-based insights and practical
recommendations for educators, developers, and researchers seeking to optimize AI-human collaboration in
language_education.