This study investigated how machine translation (MT) and generative AI (GenAI) differentially influence multiple dimensions of L2 writing anxiety among 146 young Japanese EFL learners. Using pre-post data from the second language writing anxiety inventory (SLWAIr), the analyses showed that both MT and GenAI reduced somatic anxiety and avoidance behavior, but neither tool produced group-level differences. GenAI, however, was accompanied by a non-significant increase in appraisal concerns and a significant increase in communication apprehension, whereas MT did not. These contrasting within-group patterns indicate that AI tools do not uniformly alleviate writing anxiety but instead redistribute its internal structure across subcomponents. Regression analyses further revealed that post-test anxiety was most strongly predicted by avoidance behavior and communication apprehension, underscoring the need to examine multidimensional profiles rather than relying on global scores. The findings contribute to a more nuanced understanding of the affective mechanisms underlying AI-assisted writing and highlight the importance of selective, guided integration of AI tools in classroom contexts. Limitations related to the lack of usage data and the short-term design are noted, along with directions for future research.

This study examined the comparative effectiveness of corpus-based data-driven learning (DDL; Linguee) and artificial intelligence (AI)-assisted DDL (ChatGPT) among 69 Japanese university EFL learners. Both approaches produced comparable learning gains, with no significant difference between groups after controlling for pretest performance. However, proficiency emerged as a key moderating factor: intermediate-level learners achieved greater improvements than low-proficiency learners. Learner perceptions, assessed through the technology acceptance model (TAM), indicated higher ratings of perceived ease of use, behavioral intention, and overall preference for AI-assisted DDL. These findings underscore the importance of aligning DDL implementation with learner proficiency and technology acceptance. AI-assisted tools such as ChatGPT offer accessible, engaging alternatives to traditional corpus-based methods, broadening opportunities for inductive, data-driven language learning.