Speakers
Description
This presentation will examine the current ability of a Large Language Model, ChatGPT 4, to analyze Discourse Completion Task responses for politeness strategies. The responses to the eight situation DCT meant to elicit 'refusals' were used to create two small corpora: one of University EFL learners and one of University EFL instructors. The instructors' responses were manually coded for use of speech act strategies and used as a baseline to test the LLM. Strategy definitions, examples, and prompts were carefully tested and retooled leading to an eventual accuracy rating of 99.91 percent, when combined with minimal human monitoring. The LLM’s ability to accurately recognize and tabulate the use of speech act strategies can aid researchers and teachers in the analysis of large data sets and identify gaps in strategy use by learners as targets for instruction. The utility of assessing strategy use for pragmatic competence, the process used to prompt the LLM, and the results of the research will be presented.
| Keywords | Pragmatics, Corpus Linguistics, Instructed Pragmatics, Speech ACts |
|---|
