Building Legal Reward Models for Grounding and Abstention

• URL: http://arxiv.org/abs/2609.14739v1
• Date: Sun, 13 Sep 2026 18:59:58 GMT
• Title: Building Legal Reward Models for Grounding and Abstention
• Abstract summary: We introduce a framework for transforming existing legal QA datasets into contextual preference data.<n>We find that contextual DPO improves grounded evaluation, but performance is sensitive to preference-data construction.<n>We also find evidence of cross-jurisdiction transfer: models contextually refined primarily on Victorian criminal-law data improve grounded evaluation on external US legal benchmarks.
• Score: 19.598261101266463
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models are increasingly used in high-stakes domains such as law, where systems must ground their reasoning in retrieved evidence and abstain when that evidence is insufficient. However, existing reward models are largely optimised for general preferences rather than contextual grounding, limiting their ability to evaluate these behaviours in retrieval-augmented generation (RAG) settings. We introduce a framework for transforming existing legal QA datasets into contextual preference data and use it to construct LegalRewardBench (LRB), a benchmark for evaluating grounded legal generation under noisy and insufficient retrieval conditions. Across general and legal contextual evaluation, we find that contextual DPO improves grounded evaluation, but performance is sensitive to preference-data construction. Length-balanced augmentation substantially improves grounded legal evaluation, with the strongest configuration combining length-balanced legal and general contextual preference data and improving performance by up to $\mathbf{+25.6}$pp over baseline. We further find evidence of cross-jurisdiction transfer: models contextually refined primarily on Victorian criminal-law data improve grounded evaluation on external US legal benchmarks, including a $\mathbf{+16.2}$pp improvement on \textsc{Housing Statute QA}. Together, these results provide a reproducible foundation for constructing and evaluating grounded legal reward models in retrieval-augmented settings.

Related papers

• Citation-Closure Retrieval and Per-Rule Attribution for Real-World Regulatory Compliance Question Answering [54.97384993676565]
  We formalize Regulatory Compliance QA with RegOps-Bench, a novel benchmark featuring an Operational Knowledge Graph derived from complex national R&D regulations.<n>RefWalk traverses cross-document citations, fuses multi-view candidates via max-based aggregation, and enforces per-rule attribution to explicitly map claims to sources.
  arXiv  Detail & Related papers  (2026-05-28T10:38:38Z)
• Enhancing Judgment Document Generation via Agentic Legal Information Collection and Rubric-Guided Optimization [30.089203895815384]
  Existing approaches, typically relying on standard Retrieval-Augmented Generation and Supervised Fine-Tuning, often suffer from insufficient evidence recall, hallucinated statutory references, and flawed legal reasoning.<n>We propose Judge-R1, a unified framework designed to logically enhance LLM-based judgment document generation by jointly improving legal information collection and judgment document generation.<n>Experiments on the JuDGE benchmark demonstrate that Judge-R1 significantly outperforms state-of-the-art baselines in both legal accuracy and generation quality.
  arXiv  Detail & Related papers  (2026-05-03T18:32:10Z)
• FRESCO: Benchmarking and Optimizing Re-rankers for Evolving Semantic Conflict in Retrieval-Augmented Generation [73.22935457705057]
  We introduce FRESCO, a benchmark for evaluating re-rankers in temporally dynamic contexts.<n>By pairing recency-seeking queries with historical Wikipedia revisions, FRESCO tests whether re-rankers can prioritize factually recent evidence while maintaining semantic relevance.<n>Our evaluation reveals a consistent failure mode across existing re-rankers: a strong bias toward older, semantically rich documents, even when they are factually obsolete.
  arXiv  Detail & Related papers  (2026-04-14T17:04:25Z)
• JUÁ -- A Benchmark for Information Retrieval in Brazilian Legal Text Collections [3.6360100966162787]
  Legal information retrieval in Portuguese remains difficult to evaluate systematically because available datasets differ widely in document type, query style, and relevance definition.<n>We present JU, a public benchmark for Brazilian legal retrieval designed to support more reproducible and comparable evaluation across heterogeneous legal collections.
  arXiv  Detail & Related papers  (2026-04-07T17:10:54Z)
• Enhancing Legal LLMs through Metadata-Enriched RAG Pipelines and Direct Preference Optimization [0.3823356975862005]
  Large Language Models (LLMs) perform well in short contexts but degrade on long legal documents.<n>In the legal domain, where precision is critical, such errors undermine reliability and trust.<n>We identify two failure modes: retrieval errors due to lexical redundancy in legal corpora, and decoding errors where models generate answers despite insufficient context.
  arXiv  Detail & Related papers  (2026-02-25T06:39:37Z)
• LegalOne: A Family of Foundation Models for Reliable Legal Reasoning [54.57434222018289]
  We present LegalOne, a family of foundational models specifically tailored for the Chinese legal domain.<n>LegalOne is developed through a comprehensive three-phase pipeline designed to master legal reasoning.<n>We publicly release the LegalOne weights and the LegalKit evaluation framework to advance the field of Legal AI.
  arXiv  Detail & Related papers  (2026-01-31T10:18:32Z)
• LegalMALR:Multi-Agent Query Understanding and LLM-Based Reranking for Chinese Statute Retrieval [10.997604609194033]
  Statute retrieval is essential for legal assistance and judicial decision support.<n>Real-world legal queries are often implicit, multi-issue, and expressed in colloquial or underspecified forms.<n>We present LegalMALR, a retrieval framework that integrates a Multi-Agent Query Understanding System with a zero-shot large-language-generated reranking module.
  arXiv  Detail & Related papers  (2026-01-25T04:44:56Z)
• Evidence-Augmented Policy Optimization with Reward Co-Evolution for Long-Context Reasoning [52.144281362465996]
  We propose EAPO (Evidence-Augmented Policy Optimization) to apply Reinforcement Learning to long-context scenarios.<n>We first establish the Evidence-Augmented Reasoning paradigm, validating via Tree-Structured Evidence Sampling.<n>We then introduce a specialized RL algorithm where a reward model computes a Group-Relative Evidence Reward.<n>To sustain accurate supervision throughout training, we further incorporate an Adaptive Reward-Policy Co-Evolution mechanism.
  arXiv  Detail & Related papers  (2026-01-15T11:40:57Z)
• ClaimGen-CN: A Large-scale Chinese Dataset for Legal Claim Generation [56.79698529022327]
  Legal claims refer to the plaintiff's demands in a case and are essential to guiding judicial reasoning and case resolution.<n>This paper explores the problem of legal claim generation based on the given case's facts.<n>We construct ClaimGen-CN, the first dataset for Chinese legal claim generation task.
  arXiv  Detail & Related papers  (2025-08-24T07:19:25Z)
• Augmented Question-guided Retrieval (AQgR) of Indian Case Law with LLM, RAG, and Structured Summaries [0.0]
  This paper proposes the use of Large Language Models (LLMs) to facilitate the retrieval of relevant cases.<n>Our approach combines Retrieval Augmented Generation (RAG) with structured summaries optimized for Indian case law.<n>The system generates targeted legal questions based on factual scenarios to identify relevant case law more effectively.
  arXiv  Detail & Related papers  (2025-07-23T05:24:44Z)
• Enhancing Pre-Trained Language Models with Sentence Position Embeddings for Rhetorical Roles Recognition in Legal Opinions [0.16385815610837165]
  The size of legal opinions continues to grow, making it increasingly challenging to develop a model that can accurately predict the rhetorical roles of legal opinions. We propose a novel model architecture for automatically predicting rhetorical roles using pre-trained language models (PLMs) enhanced with knowledge of sentence position information. Based on an annotated corpus from the LegalEval@SemEval2023 competition, we demonstrate that our approach requires fewer parameters, resulting in lower computational costs.
  arXiv  Detail & Related papers  (2023-10-08T20:33:55Z)
• WiCE: Real-World Entailment for Claims in Wikipedia [63.234352061821625]
  We propose WiCE, a new fine-grained textual entailment dataset built on natural claim and evidence pairs extracted from Wikipedia. In addition to standard claim-level entailment, WiCE provides entailment judgments over sub-sentence units of the claim. We show that real claims in our dataset involve challenging verification and retrieval problems that existing models fail to address.
  arXiv  Detail & Related papers  (2023-03-02T17:45:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.