Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Method for Detecting Legal Article Competition for Korean Criminal Law Using a Case-augmented Mention Graph

Seonho An, Young Yik Rhim, Min-Soo Kim

TL;DR

This work defines Legal Article Competition Detection (LACD), a retrieval task to identify competing legal articles within a law, with specific challenges arising from similar article descriptions and reliance on inter-article references. It introduces CAMGraph, a Case-Augmented Mention Graph, and the CAM-Re2 retriever, which leverages node-enriched representations and mention relationships to improve discrimination among competing articles. CAM-Re2 demonstrates substantial improvements over naïve retrieval approaches, achieving large reductions in false positives and negatives and notable gains in precision at top results, validated on a Korean criminal-law dataset. The study provides a concrete dataset, detailed experimental results, and an open-source implementation to support future work in automated legal reasoning and lawmaking.

Abstract

As social systems become increasingly complex, legal articles are also growing more intricate, making it progressively harder for humans to identify any potential competitions among them, particularly when drafting new laws or applying existing laws. Despite this challenge, no method for detecting such competitions has been proposed so far. In this paper, we propose a new legal AI task called Legal Article Competition Detection (LACD), which aims to identify competing articles within a given law. Our novel retrieval method, CAM-Re2, outperforms existing relevant methods, reducing false positives by 20.8% and false negatives by 8.3%, while achieving a 98.2% improvement in precision@5, for the LACD task. We release our codes at https://github.com/asmath472/LACD-public.

A Method for Detecting Legal Article Competition for Korean Criminal Law Using a Case-augmented Mention Graph

TL;DR

This work defines Legal Article Competition Detection (LACD), a retrieval task to identify competing legal articles within a law, with specific challenges arising from similar article descriptions and reliance on inter-article references. It introduces CAMGraph, a Case-Augmented Mention Graph, and the CAM-Re2 retriever, which leverages node-enriched representations and mention relationships to improve discrimination among competing articles. CAM-Re2 demonstrates substantial improvements over naïve retrieval approaches, achieving large reductions in false positives and negatives and notable gains in precision at top results, validated on a Korean criminal-law dataset. The study provides a concrete dataset, detailed experimental results, and an open-source implementation to support future work in automated legal reasoning and lawmaking.

Abstract

As social systems become increasingly complex, legal articles are also growing more intricate, making it progressively harder for humans to identify any potential competitions among them, particularly when drafting new laws or applying existing laws. Despite this challenge, no method for detecting such competitions has been proposed so far. In this paper, we propose a new legal AI task called Legal Article Competition Detection (LACD), which aims to identify competing articles within a given law. Our novel retrieval method, CAM-Re2, outperforms existing relevant methods, reducing false positives by 20.8% and false negatives by 8.3%, while achieving a 98.2% improvement in precision@5, for the LACD task. We release our codes at https://github.com/asmath472/LACD-public.

Paper Structure

This paper contains 31 sections, 8 equations, 5 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Definitions
  4. Retrieve-then-rerank methods
  5. Methodology
  6. The LACD task
  7. CAMGraph
  8. CAM-Re2 retriever
  9. Experimental settings
  10. The LACD dataset
  11. Implementation of retrievers
  12. Training and testing
  13. Results and analysis
  14. Experiment for Step 3
  15. Experiment for all Steps
  16. ...and 16 more sections

Figures (5)

  • Figure 1: Example of competing legal articles in Republic of Korea, translated from Korean. In Narcotics Control Act Article 60, uses narcotic drugs includes ingests opium, from Article 2 of the same act. Other articles about opium are omitted.
  • Figure 2: The example of CAMGraph. Blue and yellow boxes mean articles and corresponding LLM-generated cases, respectively. All contents are translated from Korean.
  • Figure 3: LACD results of (a) the naïve Re2 retriever and (b) our CAM-Re2 retriever. The query, Act on the Protection Of Children and Youth Against Sex Offenses Article 11-2, is a draft article in September 30, 2024 (i.e., not yet legalized). Both retrievers are implemented using the KoBigBird model jangwon_park_2021_5654154, and $k=3$ is used for top-k selection.
  • Figure 4: Overview of CAM-Re2 (purple is the query vector).
  • Figure 5: Performance across all Steps (Qwen2.0 is used as the cross encoder, and $k=10$ is used for the top-k selection).