Tense Accuracy and Student Emotions in AI ‐Supported Translation Tasks: A Classroom‐Based Investigation

ABSTRACT Background Generative artificial intelligence (GenAI) is increasingly used in higher‐education language instruction, but classroom evidence combining tense accuracy with task‐related emotions in translation remains limited. Objectives This study examined changes in English tense accuracy from pretest to posttest following ChatGPT‐supported translation tasks and investigated students' enjoyment, curiosity, boredom, and anxiety during the tasks. Methods Thirteen female third‐year English majors completed a seven‐week mixed‐methods study comprising a pretest, four ChatGPT‐supported Polish‐to‐English translation tasks, a posttest, emotion grids, and individual interviews. After each task, participants retrospectively rated the four emotions at the beginning, middle, and end of the activity. Pretest and posttest translations were scored independently by two researchers. Quantitative analyses included a paired‐samples t ‐test and Friedman tests with Bonferroni‐corrected Wilcoxon follow‐ups; interviews were analysed thematically. Results Tense‐accuracy scores increased from pretest ( M = 20.38, SD = 5.99) to posttest ( M = 24.54, SD = 2.44), mean increase = 4.15; t (12) = 2.43, p = 0.032, Cohen's d z = 0.67. Across tasks, only anxiety differed significantly, although no Bonferroni‐corrected pairwise comparison was significant. In Task 1, enjoyment and curiosity exceeded boredom and anxiety; in Task 3, both exceeded boredom, but only enjoyment exceeded anxiety. Participants reported greater metalinguistic awareness, strategic use of ChatGPT for checking and revision, adaptation to time pressure, and concerns about overreliance and contextual mismatches. Conclusions Students achieved higher posttest scores and reported mainly enjoyment and curiosity during the instructional sequence. Because the study involved a small intact class and no control group, the increase cannot be attributed specifically to ChatGPT.