AI-driven analysis of English learners’ perception, attitude, and motivation in online learning environments: a corpus-based study

This paper examines the linguistic expression of motivational orientations and affective responses in the context of online English language learning, integrating the L2 motivational self-system (L2MSS) and the technology acceptance model (TAM) into a unified framework. A mixed-methods design with a corpus-based approach involved a set of 500,000 words generated from 2,400 texts written by learners on Coursera, Edmodo, and Moodle between 2022 and 2024. A stratified purposive sampling plan ensured demographic diversification across 18 countries and proficiency levels (CEFR A2-C1). Sentiment analysis (VADER, TextBlob) was employed to analyze the data, while topic modeling (LDA) utilized chi-square tests and spearman correlations to analyze the data. The data was also analyzed using qualitative coding in NVivo. Findings indicate that positive sentiment is more than the rest and there is a considerable difference across platforms. Ideal L2MSS was the most common and closely related to positive affect, whereas ought-to L2MSS was concentrated in exam-related settings and was associated with neutral or negative feelings. The results contribute to the discussion of learner autonomy and extrinsic motivation, substantiating the methodological usefulness and limitations of sentiment analysis in second language acquisition (SLA) research.