mlsanityData to 7 Oct 2026

The Alignment Bottleneck in Decomposition-Based Claim Verification

arXiv 2602.10380 ↗First posted 11 Feb 2026cs.CL cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Structured claim decomposition is often proposed as a solution for verifying complex, multi-faceted claims, yet empirical results have been inconsistent. We argue that these inconsistencies stem from two overlooked bottlenecks: evidence alignment and sub-claim error profiles. To better understand these factors, we introduce a new dataset of real-world complex claims, featuring temporally bounded evidence and human-annotated sub-claim evidence spans. We evaluate decomposition under two evidence alignment setups: Sub-claim Aligned Evidence (SAE) and Repeated Claim-level Evidence (SRE). Our results reveal that decomposition brings significant performance improvement only when evidence is granular and strictly aligned. By contrast, standard setups that rely on repeated claim-level evidence (SRE) fail to improve and often degrade performance as shown across different datasets and domains (PHEMEPlus, MMM-Fact, COVID-Fact). Furthermore, we demonstrate that in the presence of noisy sub-claim labels, the nature of the error ends up determining downstream robustness. We find that conservative "abstention" significantly reduces error propagation compared to aggressive but incorrect predictions. These findings suggest that future claim decomposition frameworks must prioritize precise evidence synthesis and calibrate the label bias of sub-claim verification models.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 42 of its 51 references that are in the map. First posted on arXiv 11 Feb 2026.

Builds on

42 of 51 references.

References · 42

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,462
  3. 2023FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generationcited by 1,259
  4. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 965
  5. 2021A Survey on Automated Fact-Checkingcited by 376
  6. 2016Rationalizing Neural Predictionscited by 281
  7. 2024MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documentscited by 251
  8. 2023Fact-Checking Complex Claims with Program-Guided Reasoningcited by 146
  9. 2021Automated Fact-Checking for Assisting Human Fact-Checkerscited by 144
  10. 2023AVeriTeC: A Dataset for Real-world Claim Verification with Evidence from the Webcited by 144
  11. 2020Generating Fact Checking Explanationscited by 132
  12. 2019Interpretable Neural Predictions with Differentiable Binary Variablescited by 126
30 more
  1. 2023WiCE: Real-World Entailment for Claims in Wikipediacited by 118
  2. 2023Complex Claim Verification with Evidence Retrieved in the Wildcited by 110
  3. 2017Detection and Resolution of Rumours in Social Media: A Surveycited by 103
  4. 2021COVID-Fact: Fact Extraction and Verification of Real-World Claims on COVID-19 Pandemiccited by 100
  5. 2023Towards LLM-based Fact Verification on News Claims with a Hierarchical Step-by-Step Prompting Methodcited by 97
  6. 2023Factcheck-Bench: Fine-Grained Evaluation Benchmark for Automatic Fact-checkerscited by 86
  7. 2022Generating Literal and Implied Subquestions to Fact-check Complex Claimscited by 66
  8. 2015Analysing How People Orient to and Spread Rumours in Social Media by Looking at Conversational Threadscited by 62
  9. 2022Missing Counter-Evidence Renders NLP Fact-Checking Unrealistic for Misinformationcited by 45
  10. 2024A Closer Look at Claim Decompositioncited by 40
  11. 2022CHEF: A Pilot Chinese Dataset for Evidence-Based Fact-Checkingcited by 37
  12. 2023Explainable Claim Verification via Knowledge-Grounded Reasoning with Large Language Modelscited by 37
  13. 2024Decomposition Dilemmas: Does Claim Decomposition Boost or Burden Fact-Checking Performance?cited by 34
  14. 2023Self-Checker: Plug-and-Play Modules for Fact-Checking with Large Language Modelscited by 31
  15. 2023The Perils & Promises of Fact-checking with Large Language Modelscited by 31
  16. 2022Empowering the Fact-checkers! Automatic Identification of Claim Spans on Twittercited by 19
  17. 2025Resolving Conflicting Evidence in Automated Fact-Checking: A Study on Retrieval-Augmented LLMscited by 16
  18. 2021Generating Fluent Fact Checking Explanations with Unsupervised Post-Editingcited by 15
  19. 2023Improving Probability-based Prompt Selection Through Unified Evaluation and Analysiscited by 13
  20. 2024TELLER: A Trustworthy Framework for Explainable, Generalizable and Controllable Fake News Detectioncited by 10
  21. 2024Assessing the Reasoning Capabilities of LLMs in the context of Evidence-based Claim Verificationcited by 9
  22. 2025BiDeV: Bilateral Defusing Verification for Complex Claim Fact-Checkingcited by 8
  23. 2020Estimating predictive uncertainty for rumour verification modelscited by 6
  24. 2022PHEMEPlus: Enriching Social Media Rumour Verification with External Evidencecited by 6
  25. 2025MMM-Fact: A Multimodal, Multi-Domain Fact-Checking Dataset with Multi-Level Retrieval Difficultycited by 5
  26. 2024Navigating the Noisy Crowd: Finding Key Information for Claim Verificationcited by 3
  27. 2024A Claim Decomposition Benchmark for Long-form Answer Verificationcited by 2
  28. 2023Mastering the ABCDs of Complex Questions: Answer-Based Claim Decomposition for Fine-grained Self-Evaluationcited by 1
  29. 2024Generating Zero-shot Abstractive Explanations for Rumour Verificationcited by 1
  30. 2024Less for More: Enhanced Feedback-aligned Mixed LLMs for Molecule Caption Generation and Fine-Grained NLI Evaluationcited by 1

Built on by

Not cited yet.