EMNLP Findings2024
Molecular Facts: Desiderata for Decontextualization in LLM Fact Verification
Anisha Gunjal, Greg Durrett
Abstract
Automatic factuality verification of large language model (LLM) generations is becoming more and more widely used to combat hallucinations. A major point of tension in the literature is the granularity of this fact-checking: larger chunks of text are hard to fact-check, but more atomic facts like propositions may lack context to interpret correctly. In this work, we assess the role of context in these atomic facts. We argue that fully atomic facts are not the right representation, and define two criteria for molecular facts: decontextuality, or how well they can stand alone, and minimality, or how little extra information is added to achieve decontexuality. We quantify the impact of decontextualization on minimality, then present a baseline methodology 1 for generating molecular facts automatically, aiming to add the right amount of information. We compare against various methods of decontextualization and find that molecular facts balance minimality with fact verification accuracy in ambiguous settings.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
6 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation cites Enabling Large Language Models to Generate Text with Citations
- Conformal Linguistic Calibration: Trading-off between Factuality and Specificity cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- Conformal Linguistic Calibration: Trading-off between Factuality and Specificity cites Long-form factuality in large language models
- Conformal Linguistic Calibration: Trading-off between Factuality and Specificity cites DnDScore: Decontextualization and Decomposition for Factuality Verification in Long-Form Text Generation
- Fine-Grained Human Feedback Gives Better Rewards for Language Model Training cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- Fine-Grained Human Feedback Gives Better Rewards for Language Model Training cites Enabling Large Language Models to Generate Text with Citations
- Long-form factuality in large language models cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- DnDScore: Decontextualization and Decomposition for Factuality Verification in Long-Form Text Generation cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- DnDScore: Decontextualization and Decomposition for Factuality Verification in Long-Form Text Generation cites Long-form factuality in large language models
- Detecting Corpus-Level Knowledge Inconsistencies in Wikipedia with Large Language Models cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- LAQuer: Localized Attribution Queries in Content-grounded Generation cites FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
- LAQuer: Localized Attribution Queries in Content-grounded Generation cites Enabling Large Language Models to Generate Text with Citations
Cited by top-tier papers6
- Conformal Linguistic Calibration: Trading-off between Factuality and SpecificityZhengping Jiang, Anqi Liu, Benjamin Van DurmeNeurIPS 2025 · 22 citations
- DnDScore: Decontextualization and Decomposition for Factuality Verification in Long-Form Text GenerationMiriam Wanner, Benjamin Van Durme, Mark DredzeEMNLP 2025 · 17 citations
- A New HOPE: Domain-agnostic Automatic Evaluation of Text ChunkingHenrik Brådland, Morten Goodwin, Per-Arne Andersen, Alexander Salveson Nossum et al.SIGIR 2025 · 9 citations
- Detecting Corpus-Level Knowledge Inconsistencies in Wikipedia with Large Language ModelsSina J. Semnani, Jirayu Burapacheep, Arpandeep Khatua, Thanawan Atchariyachanvanit et al.EMNLP 2025
- LAQuer: Localized Attribution Queries in Content-grounded GenerationEran Hirsch, Aviv Slobodkin, David Wan, Elias Stengel-Eskin et al.ACL 2025
Builds on21
- Fine-Grained Human Feedback Gives Better Rewards for Language Model TrainingZeqiu Wu, Yushi Hu, Weijia Shi, Nouha Dziri et al.NeurIPS 2023 · 516 citations
- How Language Model Hallucinations Can SnowballMuru Zhang, Ofir Press, William Merrill, Alisa Liu et al.ICML 2024 · 406 citations
- FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text GenerationSewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis et al.EMNLP 2023 · 225 citations
- Long-form factuality in large language modelsJerry Wei, Chengrun Yang, Xinying Song, Yifeng Lu et al.NeurIPS 2024 · 182 citations
- Enabling Large Language Models to Generate Text with CitationsTianyu Gao, Howard Yen, Jiatong Yu, Danqi ChenEMNLP 2023 · 152 citations
Related papers
- Towards Effective Extraction and Evaluation of Factual ClaimsDasha Metropolitansky, Jonathan LarsonACL 2025 · 17 citations
- FactLens: Benchmarking Fine-Grained Fact VerificationKushan Mitra, Dan Zhang, Sajjadur Rahman, Estevam HruschkaACL Findings 2025 · 2 citations
- FaStFact: Faster, Stronger Long-Form Factuality Evaluations in LLMsYingjia Wan, Haochen Tan, Xiao Zhu, Xinyu Zhou et al.EMNLP Findings 2025 · 2 citations
- ContextCheck: Sentence-Level Faithfulness Verification with Context-Aware DisambiguationYueqin Yin, Yaxi Li, Xin Liu, Xun Wang et al.ACL Findings 2026
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free GuaranteesFan Nie, Xiaotian Hou, Shuhang Lin, James Zou et al.ICML 2025
