Detect, Remask, Repair: Diffusion Editing for Faithful Summarization of Evolving Contexts

2026-06-10

AI Summary

This work addresses the challenge of factual inconsistency in abstractive summaries following event evolution, where full regeneration risks losing valid content and struggles to precisely localize necessary updates. The paper proposes the first local refinement framework based on masked diffusion language models, which operates in three stages: detecting outdated segments, dynamically re-masking, and locally repairing only the untrustworthy portions while preserving the rest of the reliable content. This approach enables a controllable trade-off among fidelity, efficiency, and content retention, and further supports post-hoc correction of outputs from autoregressive systems. Evaluated on DialogSum and a newly introduced StreamSum benchmark, the method achieves sub-0.5-second single-step repair latency, significantly improves factual consistency of early summaries, and avoids unnecessary full rewrites.

Abstract

Summaries of real-world events can become outdated as contexts evolve and new information arrives.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 We study localized faithfulness repair: updating outdated spans in an existing summary while preserving supported content. We propose DETECT-REMASK-REPAIR, a diffusion-based framework that identifies, remasks, and repairs outdated regions with masked diffusion language models.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.

Problem

• faithful summarization
• evolving contexts
• localized repair
• outdated information
• summary updating

Innovation

• diffusion editing
• faithful summarization
• localized repair
• masked diffusion language models
• evolving contexts