GPT-based simulation of oral Q&A to support students attending first conference

Abstract

Students often need good practice not only for English oral presentation but also for participating in oral Q&A sessions, especially when they present their research results for the first time at international conferences. In this study, we aimed to assist those students in improving their oral skills in Q&A by having ChatGPT read the students’ papers and ask questions on the papers. The study explored six different configurations for generating questions, focusing on two key variables: whether to explicitly upload reference papers to ChatGPT, and whether the questions should be generated based on each keyword, each section, or from the entire paper. Eight master students, who had recently attended their first conferences, and eight Japanese teachers of English evaluated the questions generated for the students' papers. Surprisingly, both students and teachers favored the questions generated by the simplest configuration, where reference papers were not uploaded explicitly and no scopes were specified. Following this result, ChatGPT in the simplest configuration was integrated with a speech interface, ASR and TTS, so that ChatGPT and students can interact orally. After using this environment, all the participants expressed strong preference for using this system before attending their future conferences.