Structure-Preserving Data Augmentation via Discourse Parsing
October 15, 2026
This framework uses LLMs and discourse parsing to perform document-level data augmentation by identifying rhetorical relations. It replaces key phrases while preserving logical flow and coherence, improving performance in summarization and question answering tasks.
HOW THIS AFFECTS YOU
●
researcherYou can improve low-resource NLP model performance by maintaining inter-sentential relationships during augmentation.