Metraipolis › archive › dc_5fba6e76feb8

SOPO: Socratic Guided Policy Optimization for Span-Level Hallucination Detection — arXiv:2610.32212

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.32212
License (as archived)
CC0-1.0
Archived
2026-10-26T19:51:47.883Z
SHA-256 of the archived text
206df6114669dba340335765e4d57768586629091e7ef12c7c4cb166538b8642
Stable URL of this page
https://metraipolis.com/documento/dc_5fba6e76feb8

How to cite

SOPO: Socratic Guided Policy Optimization for Span-Level Hallucination Detection — arXiv:2610.32212 — arXiv (en), https://arxiv.org/abs/2610.32212 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5fba6e76feb8 (Metraipolis archive dc_5fba6e76feb8, sha256 206df6114669dba340335765e4d57768586629091e7ef12c7c4cb166538b8642)
BibTeX
@misc{dc5fba6e76feb8,
  title = {SOPO: Socratic Guided Policy Optimization for Span-Level Hallucination Detection — arXiv:2610.32212},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.32212},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5fba6e76feb8; sha256 206df6114669dba340335765e4d57768586629091e7ef12c7c4cb166538b8642},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.32212. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large Language Models (LLMs) have gained widespread adoption in various natural language processing tasks, but suffer from hallucination issues where they generate unfaithful or inconsistent content. While Reinforcement Learning with Verifiable Rewards (RLVR) has shown promising applications in reasoning tasks, its application to span-level hallucination detection is challenged by localization ambiguity, reward sparsity on hard examples, and uncontrolled length inflation. To address these issues, we introduce Socratic Guided Policy Optimization (SOPO), a multi-turn RL framework for robust hallucination detection. The key innovation of SOPO is a Socratic feedback mechanism, where a teacher model guides the reasoning trajectories during rollout by providing indirect hints inferred from the ground truth. This teacher-guided approach enables the model to generate more effective reasoning trajectories for hard examples, thus alleviating reward sparsity. Furthermore, we incorporate Turn-based Advantage Scaling to penalize verbosity bias, encouraging the model to develop efficient and autonomous reasoning capabilities. Empirical results on RAGTruth demonstrate that SOPO (4B) achieves state-of-the-art performance, surpassing leading proprietary models (e.g., GPT-5) and standard RL baselines, while obtaining higher reported mean scores than GRPO on four out-of-domain benchmarks, with mixed evidence against CAPO. We release our training and evaluation code alongside the 1.7B and 4B SOPO models at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.32212 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-26T19:45:33.398Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).