Exploring interaction process, L2 motivation and intended effort in chatbot-mediated interactions: evidence from Latent Dirichlet Allocation (LDA) topic modeling

Interaction shapes the learning opportunities and learners' motivation through exposure, feedback and practice. However, despite the growing interest of incorporating chatbots in second language (L2) speaking, few studies have examined how the dynamics of chatbot-mediated interactions trigger motivational changes over time. To remedy the gap, the present study empirically examined the impact of chatbot-mediated interactions on learners' L2 motivational self system (L2MSS) and intended effort, as well as unpacking the interaction process as evidence for these motivational changes. In this study, 232 undergraduate students were recruited and assigned to either the experimental group (EG-CHAT) or the control group (CG). Doubao, a chatbot built on large language models (LLMs), was introduced to deliver chatbot-mediated interactions with LLMs-powered speaking activities for the EG-CHAT group, while FiF Speaking Practice was employed for the CG to offer structured practices such as video watching, repeating and shadowing. After the five-week quasi-experiment, two-way mixed-design ANOVA results revealed significant interaction effects for both the ideal L2 self and intended effort, indicating that the EG-CHAT group exhibited greater motivational changes across the intervention period than the CG. However, while both groups demonstrated significant improvements in the L2 learning experience, no group outperformed the other, and no significant change was observed in the ought-to L2 self from pre- to post-test. The Latent Dirichlet Allocation (LDA) topic modeling results of the dialogic content provide further insight into these motivational changes and the interaction process, as well as identifying insufficient learner behaviors that call for motivational scaffolding and cognitive intervention from teachers.