The effects of generative AI on foreign language emotions: A meta-analysis

Despite increasing attention to generative artificial intelligence (GenAI) in second language education, its effects on foreign language (FL) emotions remain unclear due to the lack of systematic evidence. In the current meta-analysis study, 43 effect sizes from 23 experimental and quasi-experimental studies (N = 3461) were synthesized to explore the effects of GenAI-assisted instruction on FL emotions. A three-level random-effects model revealed a large and statistically significant overall effect (g = 0.788, p < .001), indicating that GenAI interventions substantially improve learners' emotional experiences by reducing negative emotions and enhancing positive emotions compared with traditional instruction. Emotion valence significantly moderated the effects, with larger improvements observed for negative emotions (anxiety and boredom) than for positive emotions (enjoyment), suggesting that GenAI may be particularly effective in alleviating negative emotional states in language learning. Moderator analyses grounded in the usefulness framework generally yielded limited evidence for stable differences across technological, pedagogical, and social affordances. Although among pedagogical moderators, intervention frequency initially showed a significant advantage for lower session frequency (1-2 sessions per week) over higher frequency, this effect became non-significant in sensitivity analyses that excluded studies with unspecified frequency information, indicating that the finding should be interpreted cautiously. These findings underscore the emotional potential of GenAI-assisted language learning, showing improvements in both positive and negative emotions, with stronger effects observed in alleviating negative emotional experiences, and highlighting the value of emotionally supportive AI-mediated FL instruction.