Comparing spoken discourse marker use in NS and NNS speech: A corpus-based study using ChatGPT

This study investigates the frequency and functional use of spoken discourse markers(DMs) by native speakers (NS) and non-native English speakers (NNS) using spokenmonologue data from the ICNALE corpus. The dataset includes English learners fromnine different L1 backgrounds at two proficiency levels, along with NS speakers. Employing an AI-assisted corpus analysis approach using ChatGPT and regex-basedpattern matching, the study identifies 32 spoken discourse markers categorized into 14functional types. Results reveal clear developmental trends: while NNS-low learnersshowed narrow and formulaic use, NNS-high learners demonstrated more frequent andflexible use, approaching NS patterns. However, both NNS groups underusedpragmatically rich markers (e.g., well, I mean, you know) and overused basic ones likeI think, so and and. NSs showed greater variety and more implicit functional use. Thestudy highlights the value of explicit instruction targeting underused DMs and showshow generative AI can support scalable, context sensitive discourse analysis. Findingsoffer implications for EFL pedagogy and the integration of AI tools into appliedlinguistics research.