Papers with Hallucination
The Unintended Trade-off of AI Alignment: Balancing Hallucination Mitigation and Safety in LLMs (2026.findings-eacl)
Copied to clipboard
| Challenge: | Hallucination in large language models has been studied, but a side effect remains unrecognized . a new study examines the trade-off between truthfulness and safety alignment . |
| Approach: | They propose a method that disentangles hallucination from hallucinian features using sparse autoencoders. |
| Outcome: | The proposed method preserves refusal behavior and task utility while maintaining safety alignment. |
DelucionQA: Detecting Hallucinations in Domain-specific Question Answering (2023.findings-emnlp)
Copied to clipboard
Mobashir Sadat, Zhengyu Zhou, Lukas Lange, Jun Araki, Arsalan Gundroo, Bingqing Wang, Rakesh Menon, Md Parvez, Zhe Feng
| Challenge: | Hallucination is a well-known phenomenon in text generated by large language models . state-of-the-art LLMs still have a number of weaknesses, including the tendency to generate hallucinatory statements without considering the factuality . |
| Approach: | They propose a dataset that captures hallucinations made by retrieval-augmented LLMs . they propose to use these methods to help detect hallucinosity in QA tasks . |
| Outcome: | The proposed method captures hallucinations made by retrieval-augmented LLMs for QA tasks. |
Developing a Reliable, Fast, General-Purpose Hallucination Detection and Mitigation Service (2025.naacl-industry)
Copied to clipboard
| Challenge: | Hallucination is a problem in large language models that produce incorrect output . authors propose a reliable and high-speed production system to detect and rectify hallucinations . |
| Approach: | They propose a high-speed production system that detects hallucinations in LLMs . they propose NER, natural language inference, span-based detection and a rewriting mechanism . |
| Outcome: | The proposed system detects a wide range of hallucinations in LLM responses. |
Learning Auxiliary Tasks Improves Reference-Free Hallucination Detection in Open-Domain Long-Form Generation (2025.acl-short)
Copied to clipboard
Chengwei Qin, Wenxuan Zhou, Karthik Abinav Sankararaman, Nanshu Wang, Tengyu Xu, Alexander Radovic, Eryk Helenowski, Arya Talebzadeh, Aditya Tayade, Sinong Wang, Shafiq Joty, Han Fang, Hao Ma
| Challenge: | Existing methods for detecting hallucination in long-form tasks focus on limited domains or rely heavily on external fact-checking tools, which may not always be available. |
| Approach: | They propose a new paradigm that augments fine-tuning with an auxiliary task for the model to jointly learn with the main task of hallucination detection. |
| Outcome: | The proposed method outperforms existing methods for detecting hallucination in open-domain long-form generation and is more accurate than random guessing. |
Hermit Kingdom Through the Lens of Multiple Perspectives: A Case Study of LLM Hallucination on North Korea (2025.coling-main)
Copied to clipboard
| Challenge: | Existing solutions to hallucination in large language models (LLMs) focus on aligning models with credible sources or improving how models communicate their confidence in outputs. |
| Approach: | They examine how best-performing multilingual LLMs and specific language-based models generate information about North Korea in three languages spoken in countries with significant geo-political interests. |
| Outcome: | The best-performing models generate information in three languages spoken in countries with significant geo-political interests: English (United States, United Kingdom), Korean (South Korea), and Mandarin Chinese (China). |
Beyond Semantic Entropy: Boosting LLM Uncertainty Quantification with Pairwise Semantic Similarity (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) generate long one-sentence responses that are less effective because they overlook two crucial factors: intra-cluster similarity and inter-c cluster similarity. |
| Approach: | They propose a method that generalizes semantic entropy and uses token probabilities to quantify uncertainty in large language models. |
| Outcome: | The proposed method can be extended to white-box settings by incorporating token probabilities. |
On-Policy Self-Alignment with Fine-grained Knowledge Feedback for Hallucination Mitigation (2025.findings-acl)
Copied to clipboard
Xueru Wen, Jie Lou, Xinyu Lu, Yuqiu Ji, Xinyan Guan, Yaojie Lu, Hongyu Lin, Ben He, Xianpei Han, Debing Zhang, Le Sun
| Challenge: | Large language models exhibit behavior that deviates from the boundaries of their knowledge during response generation. |
| Approach: | They propose a framework that allows large language models to explore their knowledge boundaries and self-correct generation behavior through fine-grained feedback signals. |
| Outcome: | The proposed framework enables LLMs to explore their knowledge boundaries and self-correct generation behavior through fine-grained feedback signals. |
Fine-tuning Large Language Models for Improving Factuality in Legal Question Answering (2025.coling-main)
Copied to clipboard
| Challenge: | Hallucination remains a critical challenge in large language models (LLMs) in high-stake domains such as legal question answering. |
| Approach: | They propose a method to mitigate hallucination in legal question answering by using behavior cloning and a novel Hard Sample-aware Direct Preference Optimization. |
| Outcome: | The proposed method improves non-hallucinated Statute Rate, Statute Relevance Rate, Legal Claim Truthfulness, and traditional metrics. |
Reefknot: A Comprehensive Benchmark for Relation Hallucination Evaluation, Analysis and Mitigation in Multimodal Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing research focuses on object-level or attribute-level hallucinations, neglecting the more complex relation hallucinosities. |
| Approach: | They propose a comprehensive benchmark targeting relation hallucinations comprising over 20,000 real-world samples and a confidence-based mitigation strategy which reduces the halluciation rate by an average of 9.75% across three datasets. |
| Outcome: | The proposed approach reduces the hallucination rate by an average of 9.75% across three datasets, including Reefknot. |
Can We Trust AI Doctors? A Survey of Medical Hallucination in Large Language and Large Vision-Language Models (2025.findings-acl)
Copied to clipboard
Zhihong Zhu, Yunyan Zhang, Xianwei Zhuang, Fan Zhang, Zhongwei Wan, Yuyan Chen, QingqingLong QingqingLong, Yefeng Zheng, Xian Wu
| Challenge: | Hallucination is a critical challenge for large language models and large vision-language models (LVLMs) however, dedicated research on medical hallucinations remains unexplored. |
| Approach: | They provide a unified perspective on medical hallucination for both LLMs and LVLMs, and delve into its causes. |
| Outcome: | The proposed models have demonstrated impressive performance on a variety of medical benchmarks. |
RAGEval: Scenario Specific RAG Evaluation Dataset Generation Framework (2025.acl-long)
Copied to clipboard
Kunlun Zhu, Yifan Luo, Dingling Xu, Yukun Yan, Zhenghao Liu, Shi Yu, Ruobing Wang, Shuo Wang, Yishan Li, Nan Zhang, Xu Han, Zhiyuan Liu, Maosong Sun
| Challenge: | Existing evaluation metrics for RAG systems are lacking due to high costs of data construction and lack of factual accuracy. |
| Approach: | They propose a framework to evaluate RAG systems in specialized scenarios . they propose three new metrics to evaluate LLM-generated responses . |
| Outcome: | The proposed framework outperforms zero-shot and one-shot methods in terms of clarity, safety, conformity, and richness of generated samples. |
On A Scale From 1 to 5: Quantifying Hallucination in Faithfulness Evaluation (2025.findings-naacl)
Copied to clipboard
| Challenge: | Hallucination is a popular topic in natural language generation (NLG). |
| Approach: | They propose to use large language models to evaluate faithfulness of guided NLGs by a rubric template and large language inference models to score the generation on quantifiable scales. |
| Outcome: | The proposed system can provide accurate judgement and explain whether a source and generation are factually consistent. |
MHALO: Evaluating MLLMs as Fine-grained Hallucination Detectors (2025.findings-acl)
Copied to clipboard
| Challenge: | Hallucination remains a critical challenge for multimodal large language models, undermining their reliability in real-world applications. |
| Approach: | They propose a benchmark specifically designed for evaluating MLLMs’ capability in performing token-level hallucination detection (FHD) . they use curated training data to train a specialized model that significantly outperforms existing models. |
| Outcome: | The proposed model outperforms existing models in the evaluation of 9 MLLMs and reaches an average F1IoU of 40.59%. |
Rethinking Evaluation for LLM Hallucination Detection: A Desiderata, A New RAG-based Benchmark, New Insights (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks lack long context and label noise for stress-testing detectors . a new RAG-based HDB that underwent a rigorous human annotation process is developed . |
| Approach: | They propose a desiderata of properties for hallucination detection benchmarks to exhibit . they build a RAG-based HDB that underwent a rigorous human annotation process . |
| Outcome: | The proposed benchmark exhibits all desirable properties of existing HDBs . existing benchmarks lack realistic label noise for stress-testing detectors despite human annotation . |
CaPE: Contrastive Parameter Ensembling for Reducing Hallucination in Abstractive Summarization (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing work suggests that the degree of hallucination depends on factual errors in training data. |
| Approach: | They propose a method to use training data to reduce hallucination by ensembling parameter variations in training data. |
| Outcome: | The proposed method improves on XSUM and CNN/DM datasets on human evaluations and factual metrics. |
Removal of Hallucination on Hallucination: Debate-Augmented RAG (2025.acl-long)
Copied to clipboard
| Challenge: | erroneous or biased retrieval can mislead generation, compounding hallucinations. |
| Approach: | They propose a framework that integrates multi-agent debates into retrieval and generation stages to improve retrieval reliability. |
| Outcome: | The proposed framework improves retrieval reliability, reduces hallucinations and significantly improves overall factual accuracy. |
Hallucination Detection in Structured Query Generation via LLM Self-Debating (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Hallucination remains a key challenge in applying large language models to structured query generation . we propose the Self-Debating framework to enhance detection performance . |
| Approach: | They propose a framework that prompts an LLM to generate contrastive explanations from opposing perspectives . they also propose 'self-debating' framework to enhance detection performance . |
| Outcome: | The proposed framework outperforms LLM-as-a-Judge baselines in hallucination detection . the framework generates contrastive explanations from opposing perspectives . |
Learning What Matters: Dynamic Dimension Selection and Aggregation for Interpretable Vision-Language Reward Modeling (2026.acl-long)
Copied to clipboard
Qiyuan Chen, Hongsen Huang, Jiahe Chen, Qian Shao, Jintai Chen, Hongxia Xu, Renjie Hua, Ren Chuan, Jian Wu
| Challenge: | Existing multimodal reward models are interpretable but slow, while discriminative ones are opaque "black boxes." |
| Approach: | They propose a framework that dynamically decomposes evaluation into granular, interpretable dimensions. |
| Outcome: | The proposed framework outperforms open-source reward models on benchmarks like VL-RewardBench. |
Attention-guided Self-reflection for Zero-shot Hallucination Detection in Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Hallucination is a significant barrier to the effective application of Large Language Models (LLMs). |
| Approach: | They propose an Attention-Guided SElf-Reflection approach for hallucination detection in Large Language Models. |
| Outcome: | The proposed method significantly outperforms existing methods in zero-shot hallucination detection on four widely-used LLMs across three different halluciation benchmarks. |
The Law of Knowledge Overshadowing: Towards Understanding, Predicting and Preventing LLM Hallucination (2025.findings-acl)
Copied to clipboard
Yuji Zhang, Sha Li, Cheng Qian, Jiateng Liu, Pengfei Yu, Chi Han, Yi R. Fung, Kathleen McKeown, ChengXiang Zhai, Manling Li, Heng Ji
| Challenge: | Hallucination is a persistent challenge in large language models where even with rigorous quality control, models often generate distorted facts. |
| Approach: | They propose a new framework to quantify factual hallucinations by modeling knowledge overshadowing. |
| Outcome: | The proposed framework improves model factuality on Overshadow (27.9%), MemoTrap (13.1%) and NQ-Swap (18.3%). |
Where Confabulation Lives: Latent Feature Discovery in LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Despite advances in large language models, hallucination remains a critical failure mode . despite these advances, LLMs are prone to hallucinic outputs that contain illusory information presented as if it were factual or grounded in reality. |
| Approach: | They propose to isolate and analyze confabulation, a foundational aspect of hallucination, where the model fabricates facts about unknown entities. |
| Outcome: | The proposed method reveals that the model can fabricate facts with minimal disruption, shedding light on the inner representations that drive factual and non-factual output. |
LoVeC: Reinforcement Learning for Better Verbalized Confidence in Long-Form Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for hallucination detection are limited to short-form question answering tasks and do not generalize well to open-ended generation. |
| Approach: | They propose a method that trains LLMs to append a numerical confidence score to each generated statement during long-form generation. |
| Outcome: | The proposed method is 20 faster than traditional self-consistency methods while achieving better calibration. |
ART: Attention Replacement Technique to Improve Factuality in LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to mitigate hallucinations in large language models are expensive and require significant resources. |
| Approach: | They propose a training-free method that replaces uniform attention patterns in shallow layers with local attention patterns to reduce hallucinations. |
| Outcome: | The proposed method reduces hallucinations across multiple LLM architectures. |
Hallucinations as Orthogonal Noise: Inference-Time Manifold Alignment via Dynamic Contextual Orthogonalization (2026.findings-acl)
Copied to clipboard
Mingkuan Zhao, Wentao Hu, Tianchen Huang, Yuheng Min, Suquan Chen, Yide Gao, Yanbo Zhai, Shuangyong Song, Xuelong Li
| Challenge: | Hallucinations in Large Language Models persist in critical domains where generated content diverges from contextual facts or logical constraints. |
| Approach: | They propose to generate hallucinations as orthogonal noise relative to the semantic manifold of the residual stream. |
| Outcome: | The proposed method achieves superior contextual faithfulness compared to state-of-the-art methods. |
MARCH: Multi-Agent Reinforced Check for Hallucination (2026.acl-long)
Copied to clipboard
Zhuo Li, Yupeng Zhang, Pengyu Cheng, Jiajun Song, Mengyu Zhou, Hao Li, Shujie Hu, Yu Qin, null Erchao.zec, Xiaoxi Jiang, null Guanjunjiang
| Challenge: | Existing methods to detect hallucinations suffer from inherent confirmation bias, where the verifier inadvertently reproduces the errors of the original generation. |
| Approach: | They propose a framework that enforces rigorous factual alignment by leveraging deliberate *information asymmetry* by combining a pipeline of three specialized agents: a Solver, a Proposer, and a Checker. |
| Outcome: | Extensive experiments across hallucination benchmarks demonstrate that MARCH substantially reduces hallucinism rates. |
VISTA: Verification In Sequential Turn-based Assessment (2026.acl-long)
Copied to clipboard
| Challenge: | Existing metrics evaluate isolated responses or treat unverifiable content as errors, limiting their use for multi-turn dialogue. |
| Approach: | They propose a framework for evaluating conversational factuality via claim-level verification and sequential consistency tracking. |
| Outcome: | The proposed framework improves hallucination detection over existing benchmarks and models. |