Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mind Your Neighbours: Leveraging Analogous Instances for Rhetorical Role Labeling for Legal Documents

T. Y. S. S Santosh, Hassan Sarwat, Ahmed Abdou, Matthias Grabmair

TL;DR

The paper tackles Rhetorical Role Labeling (RRL) for legal judgments, a task challenged by long-range context, interdependent roles, data scarcity, and label imbalance. It introduces neighbor-based strategies at both inference and training time: interpolation with kNN, single and multiple prototypes, and neighborhood-driven losses including contrastive and prototypical objectives, including a discourse-aware variant. Across four Indian-law datasets, the methods yield macro-F1 gains, with kNN interpolation providing strong gains, prototypical learning offering robust cross-domain transfer, and discursive contrastive losses enhancing embedding structure. The work demonstrates that leveraging neighborhood information improves both in-domain performance and cross-domain generalization, contributing scalable, domain-adaptable approaches for legal document processing.

Abstract

Rhetorical Role Labeling (RRL) of legal judgments is essential for various tasks, such as case summarization, semantic search and argument mining. However, it presents challenges such as inferring sentence roles from context, interrelated roles, limited annotated data, and label imbalance. This study introduces novel techniques to enhance RRL performance by leveraging knowledge from semantically similar instances (neighbours). We explore inference-based and training-based approaches, achieving remarkable improvements in challenging macro-F1 scores. For inference-based methods, we explore interpolation techniques that bolster label predictions without re-training. While in training-based methods, we integrate prototypical learning with our novel discourse-aware contrastive method that work directly on embedding spaces. Additionally, we assess the cross-domain applicability of our methods, demonstrating their effectiveness in transferring knowledge across diverse legal domains.

Mind Your Neighbours: Leveraging Analogous Instances for Rhetorical Role Labeling for Legal Documents

TL;DR

The paper tackles Rhetorical Role Labeling (RRL) for legal judgments, a task challenged by long-range context, interdependent roles, data scarcity, and label imbalance. It introduces neighbor-based strategies at both inference and training time: interpolation with kNN, single and multiple prototypes, and neighborhood-driven losses including contrastive and prototypical objectives, including a discourse-aware variant. Across four Indian-law datasets, the methods yield macro-F1 gains, with kNN interpolation providing strong gains, prototypical learning offering robust cross-domain transfer, and discursive contrastive losses enhancing embedding structure. The work demonstrates that leveraging neighborhood information improves both in-domain performance and cross-domain generalization, contributing scalable, domain-adaptable approaches for legal document processing.

Abstract

Rhetorical Role Labeling (RRL) of legal judgments is essential for various tasks, such as case summarization, semantic search and argument mining. However, it presents challenges such as inferring sentence roles from context, interrelated roles, limited annotated data, and label imbalance. This study introduces novel techniques to enhance RRL performance by leveraging knowledge from semantically similar instances (neighbours). We explore inference-based and training-based approaches, achieving remarkable improvements in challenging macro-F1 scores. For inference-based methods, we explore interpolation techniques that bolster label predictions without re-training. While in training-based methods, we integrate prototypical learning with our novel discourse-aware contrastive method that work directly on embedding spaces. Additionally, we assess the cross-domain applicability of our methods, demonstrating their effectiveness in transferring knowledge across diverse legal domains.
Paper Structure (24 sections, 10 equations, 2 figures, 3 tables)

This paper contains 24 sections, 10 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Task, Datasets, Baseline
  4. RQ 1: Leveraging the Neighbourhood at Inference
  5. Methods
  6. Interpolation with kNN
  7. Interpolation with Single Prototype
  8. Interpolation with Multiple Prototypes
  9. Experiments
  10. Implementation Details
  11. Results
  12. RQ 2: Leveraging the Neighbourhood at Training
  13. Methods
  14. Contrastive Learning
  15. Single Prototypical Learning
  16. ...and 9 more sections

Figures (2)

  • Figure 1: Sensitity to hyperparameters - kNN (M-CL) $\lambda$ = 0: interpolation only, $\lambda$ = 1: baseline only
  • Figure 2: t-SNE visualizations of different models on M-CL dataset. Disc.: Discourse, Contr.: Contrastive. head, torso and tail in Disc.-aware Contr. plot indicate the relative position of the sentence in a document.