面向词元级幻觉检测的时序多信号融合

摘要

词元级幻觉检测器基于单一信号独立地对每个词元打分，因此在生成模型自信地出错时恰恰会失效。本文转而将幻觉视为时间上延展的片段，并通过序列标注来检测它：每个词元的评分来自一个33维特征流，该特征流融合了文本统计量、自然语言推理（NLI）蕴含关系以及语言模型惊异度，且无法访问模型内部信息。基于这些特征的双向门控循环单元（BiGRU）在RAGTruth上达到0.840的AUC（10个随机种子），比独立的逻辑回归基线提升11个百分点（p = 0.002，Wilcoxon符号秩检验）。受控分解表明，大部分增益来自时间顺序而非模型容量：证据在片段内从置信位置传播到模糊的邻近位置。相同的0.845上限在循环、状态空间（Mamba）和注意力架构中反复出现，这表明瓶颈在于特征集而非模型。由于检测器只读取生成的文本和外部信号，它适用于闭源模型，并且对于训练时从未见过的语言模型生成的文本仍然有效，AUC损失低于4%。

English

Token-level hallucination detectors score each token independently from a single signal, and fail exactly when the generating model is confidently wrong. This paper instead treats hallucination as a temporally extended span and detects it by sequence labeling: each token is scored from a 33-dimensional feature stream that fuses text statistics, Natural Language Inference (NLI) entailment, and language model surprisal, with no access to model internals. A Bidirectional Gated Recurrent Unit (BiGRU) over these features reaches an AUC of 0.840 on RAGTruth (10 seeds), an 11-point gain over an independent logistic-regression baseline (p = 0.002, Wilcoxon signed-rank). A controlled decomposition attributes most of the gain to temporal order rather than model capacity: evidence propagates from confident positions to ambiguous neighbors within a span. The same 0.845 ceiling recurs across recurrent, state-space (Mamba), and attention architectures, locating the bottleneck in the feature set rather than the model. Because it reads only the generated text and external signals, the detector works on closed-source models, and it keeps working on text produced by language models it never saw during training, losing under 4% AUC.