AI-assisted task-based language teaching: EFL learners’ acquisition of collocations through ChatGPT and movie language

This study investigates the potential of generative artificial intelligence (AI), specifically ChatGPT primed with data from the American Movie Corpus (AMC), to support the acquisition of high- frequency spoken collocations in a general English context. Thirty-four B2-level Italian university students were randomly assigned to guided AI, unguided AI, or non-AI control conditions and completed a ten-week task-based language teaching (TBLT) programme. Eight target collocations (e.g., good thing, little problem, get home) were selected based on frequency and statistical association measures (t-score > 2; MI > 3) in the AMC. Learner output was analysed using a linear mixed-effects model to compare collocation frequency at pretest, posttest and delayed posttest. While no statistically significant differences emerged between instructional conditions, the findings point to important trends that suggest potential benefits of AI-mediated tasks. These results are interpreted with caution, given small, unequal groups and a conservative bigram operationalisation that may underestimate more flexible patterns. While the integration of corpus-informed input into AI-mediated tasks remains promising, sustained exposure, stronger focus-on-form, and broader pattern coding may be required to detect gains.