将多模态幻觉检测转为定位任务，提升模型对错误文本的识别能力。
Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data
- 把幻觉检测当作序列标注任务，定位具体错误文本段落。
- 用伪造的错误数据预训练，减少标注依赖，提升小样本效果。
- 适合研究多模态模型可靠性与幻觉检测的开发者使用。
多模态语言模型在输出中常出现幻觉，影响其可信度。自动检测这些错误对缓解问题至关重要，但现有方法未定位幻觉，仅将其视为分类任务。本文首次将多模态幻觉检测定义为序列标注任务，要求模型定位幻觉文本片段，并提出一个强基线模型。由于人工标注成本高，我们通过生成带有错误关联信息的数据来提升模型样本效率：利用短语定位数据，将真实对应内容替换为幻觉内容，生成带幻觉的文本。实验表明，使用此类数据进行预训练可显著提升微调时的样本效率，且原始定位数据提供的学习信号在此过程中起关键作用。
原文摘要 · Abstract (English)
Multimodal language models can exhibit hallucinations in their outputs, which limits their reliability. The ability to automatically detect these errors is important for mitigating them, but has been less explored and existing efforts do not localize hallucinations, instead framing this as a classification task. In this work, we first pose multimodal hallucination detection as a sequence labeling task where models must localize hallucinated text spans and present a strong baseline model. Given the high cost of human annotations for this task, we propose an approach to improve the sample efficiency of these models by creating corrupted grounding data, which we use for pre-training. Leveraging phrase grounding data, we generate hallucinations to replace grounded spans and create hallucinated text. Experiments show that pre-training on this data improves sample efficiency when fine-tuning, and that the learning signal from the grounding data plays an important role in these improvements.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。