Spoken dialogue systems and ChatGPT for second language pragmatics research
Tetyana Sydorenko, Judit Dombi, Ameeta Agrawal, Steven L. Thorne, Jung In Lee, Yufei Tao · 2024
This chapter describes two recent technologies, Spoken Dialogue Systems (SDSs) and ChatGPT, and their affordances and challenges for second language pragmatics research. Since role-play is a commonly used task in pragmatics research, we discuss ways that the two technologies can be used for administering a role-play task. While SDSs are unable to deviate from a pre-programmed dialogue structure and thus cannot accommodate participants’ linguistic creativity, they are nevertheless systematic and allow for comparisons across participants. ChatGPT, on the other hand, is similar to a human in being able to accommodate participants’ creativity. As such, SDSs lend themselves to administering closed role-plays, while ChatGPT can be used to collect open role-play data. Nevertheless, ChatGPT’s responses are not fully natural compared to human responses and most often violate Grice’s Maxim of Quantity (i.e., they are too verbose). Future directions, such as multimodality and modifying ChatGPT’s behavior via prompt architecture, are discussed.