Rethinking L2 academic speaking: Oral communication strategies and discourse markers in ChatGPT-mediated instruction

The introduction of generative artificial intelligence (GenAI) has created new opportunities to support L2 speaking instruction and development. However, empirical evidence is needed to establish its pedagogic value in higher education contexts, particularly for cognitively demanding second-language (L2) speaking tasks. This study investigates the effects of ChatGPT-mediated instruction on the use of oral communication strategies (OCSs) and discourse markers (DMs) during an academic debate session. Using a quasi-experimental design, a total of 56 Hungarian English as a Foreign Language (EFL) undergraduates were assigned to two instructional conditions: a GenAI-assisted experimental group and a non-GenAI-assisted control group. Data were collected through a posttest debate task and analysed using a combination of quantitative and qualitative techniques. The results indicated significant differences in the deployment of OCSs and DMs between the two instructional conditions. In particular, our analysis revealed an interesting contrast between the strong focus on fluency-maintenance and discourse-organisational devices in the control group as opposed to the preponderance of direct-interactional strategies alongside a wide range of DMs in the experimental group. The findings suggest a potentially beneficial role for ChatGPT-mediated instruction in the development of strategic competence and discourse-level awareness. These GenAI affordances, however, presuppose careful teacher mediation within a progressive instructional design that moves from AI-supported rehearsal to peer- and group-based interaction.