MedFabric and EtHER: A Data-Centric Framework for Word-Level Fabrication Generation and Detection in Medical LLMs
For medical LLM applications, this work provides a more realistic fabrication dataset and a detection method that significantly improves accuracy, addressing a critical safety concern.
The authors propose a data-centric pipeline to generate realistic word-level fabrications in medical LLMs, creating the MedFabric dataset, and introduce ETHER, a modular detector that improves factual alignment. ETHER outperforms state-of-the-art detectors by over 15% on word-level fabrication benchmarks.
Large Language Models exhibit strong reasoning and semantic understanding capabilities but often hallucinate in domains that require expert knowledge, among which fabrications, the generation of factually incorrect yet fluent statements, pose the greatest risk in medical contexts. Existing medical hallucination datasets inadequately capture fabrication phenomena due to limited fabrication coverage, stylistic disparities between human and LLM-authored texts, and distributional drift during hallucinated sample synthesis. To address this, we propose a data-centric pipeline to generate realistic and word-level fabrications that preserve syntactic and stylistic fidelity while introducing subtle factual deviations, resulting in MedFabric. Building upon this dataset, we introduce ETHER, a modular word-level fabrication detector integrating Text2Table Decomposition, Word Masking and Filling and Hybrid Sentence Pair Evaluation to enhance factual alignment. Empirical results demonstrate that MedFabric outperforms state-of-the-art detectors by over 15% on word-level fabrication benchmarks while maintaining consistent performance across structural similarities, offering a comprehensive framework for reliable and domain-specific factuality detection.