Detect, Remask, Repair: Diffusion Editing for Faithful Summarization of Evolving Contexts
For summarization systems dealing with evolving contexts, this work offers a controllable, efficient alternative to full regeneration, enabling faithfulness-speed tradeoffs.
The paper addresses the problem of updating outdated summaries when contexts evolve, proposing a diffusion-based framework (DETECT-REMASK-REPAIR) that repairs only unsupported spans. Experiments show that one-step repair reduces cost to under half a second while improving faithfulness, and the framework can serve as a post-hoc correction for autoregressive systems.
Summaries of real-world events can become outdated as contexts evolve and new information arrives. A common response is to generate a new summary from the updated context, but full regeneration discards the previous draft, can obscure what changed, and may be unnecessary when only a few claims are unsupported. We study localized faithfulness repair: updating outdated spans in an existing summary while preserving supported content. We propose DETECT-REMASK-REPAIR, a diffusion-based framework that identifies, remasks, and repairs outdated regions with masked diffusion language models. To evaluate evolving-context summarization, we introduce StreamSum, a benchmark of synthetic event timelines. Experiments on DialogSum and StreamSum show that localized diffusion repair provides a controllable alternative to full rewriting: faithfulness-steered repair improves early drafts, one-step repair reduces repair cost to under half a second, with the framework enabling faithfulness-speed-preservation tradeoffs across datasets. We also find that the framework can provide a post-hoc correction step that improves faithfulness for autoregressive systems.