Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GuRE:Generative Query REwriter for Legal Passage Retrieval

Daehee Kim, Deokhyung Kang, Jonghwi Kim, Sangwon Ryu, Gary Geunbae Lee

TL;DR

This work tackles vocabulary mismatch in Legal Passage Retrieval by introducing GuRE, a Generative query Rewriter that trains an LLM to produce rewritten queries from legal context and passages. GuRE is retriever-agnostic, improving retrieval performance across sparse and dense baselines and data pools by converting queries into forms with lower lexical gap to the target passages. Through analysis across data scarcity, model backbones, and training objectives, the authors show that GuRE often outperforms direct retriever fine-tuning, especially in long-tail legal corpora. The approach holds practical promise for real-world legal information systems, enabling more efficient and accurate extraction of relevant passages while maintaining domain-specific knowledge in the rewriting process.

Abstract

Legal Passage Retrieval (LPR) systems are crucial as they help practitioners save time when drafting legal arguments. However, it remains an underexplored avenue. One primary reason is the significant vocabulary mismatch between the query and the target passage. To address this, we propose a simple yet effective method, the Generative query REwriter (GuRE). We leverage the generative capabilities of Large Language Models (LLMs) by training the LLM for query rewriting. "Rewritten queries" help retrievers to retrieve target passages by mitigating vocabulary mismatch. Experimental results show that GuRE significantly improves performance in a retriever-agnostic manner, outperforming all baseline methods. Further analysis reveals that different training objectives lead to distinct retrieval behaviors, making GuRE more suitable than direct retriever fine-tuning for real-world applications. Codes are avaiable at github.com/daehuikim/GuRE.

GuRE:Generative Query REwriter for Legal Passage Retrieval

TL;DR

This work tackles vocabulary mismatch in Legal Passage Retrieval by introducing GuRE, a Generative query Rewriter that trains an LLM to produce rewritten queries from legal context and passages. GuRE is retriever-agnostic, improving retrieval performance across sparse and dense baselines and data pools by converting queries into forms with lower lexical gap to the target passages. Through analysis across data scarcity, model backbones, and training objectives, the authors show that GuRE often outperforms direct retriever fine-tuning, especially in long-tail legal corpora. The approach holds practical promise for real-world legal information systems, enabling more efficient and accurate extraction of relevant passages while maintaining domain-specific knowledge in the rewriting process.

Abstract

Legal Passage Retrieval (LPR) systems are crucial as they help practitioners save time when drafting legal arguments. However, it remains an underexplored avenue. One primary reason is the significant vocabulary mismatch between the query and the target passage. To address this, we propose a simple yet effective method, the Generative query REwriter (GuRE). We leverage the generative capabilities of Large Language Models (LLMs) by training the LLM for query rewriting. "Rewritten queries" help retrievers to retrieve target passages by mitigating vocabulary mismatch. Experimental results show that GuRE significantly improves performance in a retriever-agnostic manner, outperforming all baseline methods. Further analysis reveals that different training objectives lead to distinct retrieval behaviors, making GuRE more suitable than direct retriever fine-tuning for real-world applications. Codes are avaiable at github.com/daehuikim/GuRE.
Paper Structure (47 sections, 5 equations, 9 figures, 13 tables)

This paper contains 47 sections, 5 equations, 9 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Method: GuRE
  3. Experiments
  4. Task Description
  5. Baselines
  6. Query Expansion.
  7. Fine-Tuning
  8. Dataset
  9. Models
  10. Results
  11. Analyses
  12. Rewritten Query Evaluation
  13. Generalizability under Data Constraints
  14. Which Model Should We Train?
  15. Conclusion
  16. ...and 32 more sections

Figures (9)

  • Figure 1: (a) Retriever fails to retrieve the target passage using an original query. (b) GuRE rewrites the query before retrieval. Overlapping context between the "rewritten query" and the target passage is in yellow.
  • Figure 2: Instruction prompt for GuRE.
  • Figure 3: nDCG@10 with 99% confidence intervals (shading) for GuRE and a fine-tuned retriever across sampling thresholds. Higher thresholds yield more unique samples, while lower ones favor frequent samples. Retriever for this experiment is ModernBert.
  • Figure 4: Target passage frequency distribution across different dataset versions (Log Scale)
  • Figure 5: Q2D prompt
  • ...and 4 more figures