<p>Manual annotation remains essential for identifying complex pragmatic and discourse-level features in corpus linguistics, particularly the functional components of speech acts. While part-of-speech and semantic tagging can be automated with high accuracy, annotating discourse strategies remains challenging because strategies can appear as a wide range of surface forms that cannot be fully captured by predefined example sets. These limitations hinder the scalability of function-to-form approaches and constrain the development of richly annotated corpora for pragmatics research and instruction. This study investigates whether a large language model (LLM), GPT-4 (accessed via the ChatGPT interface), can assist with the annotation of refusal strategies as they are explicitly realized in written learner responses, rather than inferred or context-dependent meanings. The focus is on identifying clear surface forms that express pragmatic functions. A corpus of written Discourse Completion Tasks completed by Japanese university English learners was analyzed for reliability, GPT-4–human agreement, accuracy, and the model’s ability to generalize beyond the specific examples provided in the prompts. The results suggest that GPT-4 can effectively support this type of form-guided, functional pragmatic annotation and thereby increase scalability and accuracy.</p>