DecomposeRL: Learning to Ask Useful, Informative, and Diverse Questions for Semi-Supervised, Traceable Claim Verification
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916053122023424 |
|---|---|
| author | Dipta, Shubhashis Roy Padia, Ankur Ferraro, Francis |
| author_facet | Dipta, Shubhashis Roy Padia, Ankur Ferraro, Francis |
| contents | Claim verification splits between end-to-end classifiers that are accurate but yields no inspectable traces, and decomposition-based methods produce inspectable traces but lag performance on benchmark datasets. We propose DecomposeRL an accurate claim-verifier that produce inspectable traces. DecomposeRL frames decomposition as an RL policy trained with GRPO and a multi-faceted reward ensemble, enabling both fully supervised and semi-supervised learning from unlabeled claims. DecomposeRL addresses the prohibitive training cost of GRPO with a data-curation funnel that distills 115K fact-verification claims into a compact, learning-signal-dense subset of 5K claims. We show that a DecomposeRL-7B policy trained with full supervision on only ~5K curated claims achieves 86.3 in-domain and 69.8 out-of-domain balanced accuracy across 11 claim-verification benchmarks containing biomedical, political, scientific, and general-domain claims. Despite being 4x smaller, it matches 32B baselines and GPT-4.1-mini, and it further outperforms baselines in a semi-supervised setting with only 10% labeled claims data. Code, data, and models are available at https://dipta007.github.io/DecomposeRL |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_27858 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | DecomposeRL: Learning to Ask Useful, Informative, and Diverse Questions for Semi-Supervised, Traceable Claim Verification Dipta, Shubhashis Roy Padia, Ankur Ferraro, Francis Computation and Language Artificial Intelligence Machine Learning Claim verification splits between end-to-end classifiers that are accurate but yields no inspectable traces, and decomposition-based methods produce inspectable traces but lag performance on benchmark datasets. We propose DecomposeRL an accurate claim-verifier that produce inspectable traces. DecomposeRL frames decomposition as an RL policy trained with GRPO and a multi-faceted reward ensemble, enabling both fully supervised and semi-supervised learning from unlabeled claims. DecomposeRL addresses the prohibitive training cost of GRPO with a data-curation funnel that distills 115K fact-verification claims into a compact, learning-signal-dense subset of 5K claims. We show that a DecomposeRL-7B policy trained with full supervision on only ~5K curated claims achieves 86.3 in-domain and 69.8 out-of-domain balanced accuracy across 11 claim-verification benchmarks containing biomedical, political, scientific, and general-domain claims. Despite being 4x smaller, it matches 32B baselines and GPT-4.1-mini, and it further outperforms baselines in a semi-supervised setting with only 10% labeled claims data. Code, data, and models are available at https://dipta007.github.io/DecomposeRL |
| title | DecomposeRL: Learning to Ask Useful, Informative, and Diverse Questions for Semi-Supervised, Traceable Claim Verification |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2605.27858 |