用于事实性核查与幻觉检测的分解式蕴含关系
Decomposed Entailment for Factuality Checking and Hallucination Detection
- CEA LIST
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
本研究提出轻量级黑箱框架HallDetect，将生成内容分解为原子主张并经对比式蕴含模型验证，在多基准上优于同资源基线，可定位幻觉错误。
AI中文摘要：
大型语言模型（LLMs）的可靠性常因事实不一致性受损，其中包括幻觉——即生成内容未得到底层来源支持的情况。我们提出HallDetect，这是一个轻量级、无参考且黑箱式的幻觉检测框架，我们不仅在摘要任务上对其进行评估，还在更广泛的基于来源的生成场景中开展评估。HallDetect基于分解式事实性评估：将生成内容分解为原子主张，每个主张通过基于编码器的紧凑蕴含模型，借助多尺度来源块库上的对比公式进行验证，并通过非对称分数聚合，其中单个被明确反驳的主张即标记该响应存在问题。在受控协议下，所有方法共享相同的4位量化骨干和消费级硬件预算，HallDetect在四个基准中的三个上，优于资源相当的生成式和基于嵌入的基线，同时在骨干系列间保持稳定，并生成主张到片段的审计轨迹，可定位每个错误。
英文摘要：
The reliability of Large Language Models (LLMs) is often compromised by factual inconsistencies, including hallucinations---cases where generated content is not supported by the underlying source. We present HallDetect, a lightweight, reference-free, and black-box framework for hallucination detection that we evaluate not only on summarization but across a broader range of source-grounded generation settings. HallDetect builds on decomposition-based factuality evaluation: generated content is decomposed into atomic claims, each verified by a compact encoder-based entailment model through a contrastive formulation over a multi-scale library of source chunks, and aggregated with an asymmetric score in which a single confidently contradicted claim flags the response. Under a controlled protocol in which all methods share the same 4-bit quantized backbones and consumer-grade hardware budget, HallDetect outperforms comparably resourced generative and embedding-based baselines on three of four benchmarks while remaining stable across backbone families, and yields a claim-to-span audit trail that localizes each error.
