Dialogue Discourse Parsing as Generation: A Sequence-to-Sequence LLM-based Approach
Chuyuan Li, Yuwei Yin, Giuseppe Carenini · 2024
Discourse analysis studies the sentence organization within a document, aiming to reveal its underlying structural information.Existing works on dialogue discourse parsing mostly use encoder-only models and sophisticated decoding strategies to extract structures.Despite recent advances in Large Language Models (LLMs), applying directly these models on discourse parsing is challenging.To fully leverage the rich semantic and discourse knowledge in LLMs, we propose to transform discourse parsing into a generation task using a text-to-text paradigm.Our approach is intuitive and requires no modification of the LLM architecture.Experimental results on STAC and Molweni datasets show that a sequence-tosequence model such as T0 can perform reasonably well.Notably, our improved transitionbased sequence-to-sequence system achieves new state-of-the-art performance on Molweni.Furthermore, our systems can generate richer discourse structures such as graphs, whereas previous methods are mostly limited to trees. 1