Arg-LLaDA: Argument Summarization via Large Language Diffusion Models and Sufficiency-Aware Refinement

  • 2025-07-25 09:07:52
  • Hao Li, Yizheng Sun, Viktor Schlegel, Kailai Yang, Riza Batista-Navarro, Goran Nenadic
  • 0

Abstract

Argument summarization aims to generate concise, structured representationsof complex, multi-perspective debates. While recent work has advanced theidentification and clustering of argumentative components, the generation stageremains underexplored. Existing approaches typically rely on single-passgeneration, offering limited support for factual correction or structuralrefinement. To address this gap, we introduce Arg-LLaDA, a novel large languagediffusion framework that iteratively improves summaries via sufficiency-guidedremasking and regeneration. Our method combines a flexible masking controllerwith a sufficiency-checking module to identify and revise unsupported,redundant, or incomplete spans, yielding more faithful, concise, and coherentoutputs. Empirical results on two benchmark datasets demonstrate that Arg-LLaDAsurpasses state-of-the-art baselines in 7 out of 10 automatic evaluationmetrics. In addition, human evaluations reveal substantial improvements acrosscore dimensions, coverage, faithfulness, and conciseness, validating theeffectiveness of our iterative, sufficiency-aware generation strategy.

 

Quick Read (beta)

loading the full paper ...