New pipeline combines extractive and abstractive methods for legal document summarization

Summarizing legal judgments automatically offers significant relief to law professionals grappling with dense legal language, specialized terminology, and voluminous documents. However, recent research has identified limitations in purely abstractive approaches to legal text summarization, prompting the development of hybrid and extractive alternatives along with new domain-specific datasets.

The AugAbEx approach#

Researchers have introduced AugAbEx, a pipeline designed to enhance an existing legal case summarization dataset. The method augments a collection of human-written reference summaries with silver standard extractive summaries generated through a systematic process that incorporates expert judgment. By pairing gold-standard human summaries with machine-generated extractive versions, the enriched dataset becomes more useful for developing and testing novel hybrid and extractive summarization algorithms.

Quality assurance across dimensions#

The team conducted rigorous validation of the augmented extractive summaries, examining them from three perspectives: structural consistency, lexical properties, and semantic meaning. These assessments were performed within a legal domain-specific evaluation framework to guarantee quality. Testing across seven English legal case-summarization datasets showed that the silver standard summaries performed well across all assessment criteria.

Competitive performance#

Comparative evaluation against two baseline approaches and three state-of-the-art methods demonstrated that AugAbEx outperformed all competing techniques. The summaries produced by the proposed pipeline achieved superior quality across evaluation dimensions, positioning it as an effective tool for researchers developing the next generation of legal summarization systems.