EVENT DOSSIER
On September 7, 2026, arXiv cs.AI published the paper “Attributable by Construction: Claim-Anchored Provenance for Multi-Document Summarization”, which proposes a new method to ensure traceability of facts in multi-document summaries by constructing claim anchors. This method aims to address the common issue of unclear attribution of facts in multi-document summaries, enabling the generated summaries to be clearly associated with specific paragraphs or sentences in the original documents, thereby enhancing the credibility and verifiability of the information.
CAMS 框架提出将归因作为生成结构属性而非下游预测，通过分解源文档为原子声明并确定性地解析其来源，实现了从逐字引用到词元跨度的精准溯源。该方法在 MultiNews、DiverseSumm 及 WCEP 零样本测试中，将多源归因准确率从 38% 提升至 64%，同时保持引用数量不变，并将人工验证每个声明的时间缩短 3.4 倍。CAMS 在总结质量上与强基线持平，显著提升了忠实度和引用精度，并明确了归因作为独立于模型准确性的不变量这一特性。研究团队已开源代码及约 32 万条声明 - 引用 - 跨度标注数据。