Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Court View Generation with Knowledge Injection and Guidance

Ang Li, Yiquan Wu, Yifei Liu, Fei Wu, Ming Cai, Kun Kuang

TL;DR

This work tackles Court View Generation (CVG) by addressing domain knowledge gaps in pretrained language models. It introduces Knowledge Injection and Guidance (KIG), combining a knowledge-injected prompt encoder (keyword initialization and label attention) with a generating navigator that guides inference without altering PLM architecture. Empirical results on real-world civil data show that KIG yields significant gains, notably an 11.87% improvement in claim-response accuracy over strong baselines, and human evaluation confirms higher fluency and fitness. The approach is efficient, transfer-friendly, and publicly releasing code and data, with broad implications for practical LegalAI deployment and adaptation to other PLMs and jurisdictions.

Abstract

Court View Generation (CVG) is a challenging task in the field of Legal Artificial Intelligence (LegalAI), which aims to generate court views based on the plaintiff claims and the fact descriptions. While Pretrained Language Models (PLMs) have showcased their prowess in natural language generation, their application to the complex, knowledge-intensive domain of CVG often reveals inherent limitations. In this paper, we present a novel approach, named Knowledge Injection and Guidance (KIG), designed to bolster CVG using PLMs. To efficiently incorporate domain knowledge during the training stage, we introduce a knowledge-injected prompt encoder for prompt tuning, thereby reducing computational overhead. Moreover, to further enhance the model's ability to utilize domain knowledge, we employ a generating navigator, which dynamically guides the text generation process in the inference stage without altering the model's architecture, making it readily transferable. Comprehensive experiments on real-world data demonstrate the effectiveness of our approach compared to several established baselines, especially in the responsivity of claims, where it outperforms the best baseline by 11.87%.

Enhancing Court View Generation with Knowledge Injection and Guidance

TL;DR

This work tackles Court View Generation (CVG) by addressing domain knowledge gaps in pretrained language models. It introduces Knowledge Injection and Guidance (KIG), combining a knowledge-injected prompt encoder (keyword initialization and label attention) with a generating navigator that guides inference without altering PLM architecture. Empirical results on real-world civil data show that KIG yields significant gains, notably an 11.87% improvement in claim-response accuracy over strong baselines, and human evaluation confirms higher fluency and fitness. The approach is efficient, transfer-friendly, and publicly releasing code and data, with broad implications for practical LegalAI deployment and adaptation to other PLMs and jurisdictions.

Abstract

Court View Generation (CVG) is a challenging task in the field of Legal Artificial Intelligence (LegalAI), which aims to generate court views based on the plaintiff claims and the fact descriptions. While Pretrained Language Models (PLMs) have showcased their prowess in natural language generation, their application to the complex, knowledge-intensive domain of CVG often reveals inherent limitations. In this paper, we present a novel approach, named Knowledge Injection and Guidance (KIG), designed to bolster CVG using PLMs. To efficiently incorporate domain knowledge during the training stage, we introduce a knowledge-injected prompt encoder for prompt tuning, thereby reducing computational overhead. Moreover, to further enhance the model's ability to utilize domain knowledge, we employ a generating navigator, which dynamically guides the text generation process in the inference stage without altering the model's architecture, making it readily transferable. Comprehensive experiments on real-world data demonstrate the effectiveness of our approach compared to several established baselines, especially in the responsivity of claims, where it outperforms the best baseline by 11.87%.
Paper Structure (39 sections, 14 equations, 4 figures, 6 tables)

This paper contains 39 sections, 14 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Prompt Learning
  4. Controlled Text Generation
  5. Court View Generation
  6. Problem Formulation
  7. Method
  8. Preliminary
  9. Prompt Tuning.
  10. Knowledge-injected Prompt Encoder
  11. Keyword-based Initialization
  12. Label Attention
  13. Generating Navigator
  14. Principles
  15. Recommendation Score Calculation
  16. ...and 24 more sections

Figures (4)

  • Figure 1: A real case in CVG. BART does not respond to the plaintiff's interest claim and mistakenly responds to the guarantee liability claim.
  • Figure 2: The architecture of KIG. (1) is the frozen pretrained GPT-2, utilized for obtaining input context. (2) is the knowledge-injected prompt encoder, which incorporates claim knowledge to obtain claim-aware context. (3) is the same frozen GPT-2, which is used for calculating the activation of output from the knowledge-injected context. (4) is the generating navigator, guiding the model's inference process.
  • Figure 3: Case study.
  • Figure 4: The upper shows the impact of guiding strength coefficient $\lambda$ and the lower shows the impact of the guidance starting position $k$.