Document data augmentation keeps story flow for better NLP results
Structure-Preserving Document-Level Data Augmentation via Discourse Parsing
Computation and LanguageArtificial Intelligence
Summary
Many current methods to create more training text change words or sentences without thinking about the whole story or how sentences connect to each other. The authors created a way to add new text examples while keeping the way sentences fit together logically, using big language models and tools that understand document structure. They replace parts of sentences with new but related content that still fits the overall story flow. This helps improve computer systems that summarize texts or answer questions by giving them better examples to learn from.
What this means in practice
- •For nlp engineering teams: Improve training data for summarization and QA systems by augmenting documents without breaking their logical flow.
- •For legal document analysts: Generate variant legal texts that keep original argument structures to train AI models on diverse but coherent data.
Authors
Ruilin Shang, Shilong Hou, Zi Long, Xianghua Fu
Abstract
Data augmentation (DA) plays a vital role in improving model performance in low-resource or limited-supervision scenarios within Natural Language Processing (NLP). Existing DA methods, such as synonym replacement and back-translation, have demonstrated effectiveness on the word or sentence level; however, they frequently neglect discourse-level coherence and logical flow, which are essential for complex tasks dependent on inter-sentential relationships. In this paper, we propose a structure-preserving document-level data augmentation framework based on large language models (LLMs) and document-level discourse structure parsing. Our approach identifies rhetorical relations between sentence pairs and extracts key phrases, which are then replaced with semantically coherent but topically shifted content while preserving the original discourse structure. Experimental results on text summarization and question answering show that the proposed augmented data provides complementary supervision and improves combined-data training across the evaluated tasks. Comparisons with a standard RST parser and controlled ablations further support the proposed discourse representation.