CLASS: A Novel Method for Chinese Legal Judgments Summarization

Abstract

We propose a novel method to generate abstractive summarization of Chinese legal judgments named CLASS (Chinese LegAl judgmentS Summarization) which exploits the element structure of the legal judgments. Firstly, we extract sentences with high importance from the legal judgments. Secondly, the extracted sentences along with its summaries are split into different source-target element pairs that are used for training an abstractive model to summarize different elements of the judgments separately. Finally, a complete summary is generated by combining the summaries of each element. We conduct comparative experiments on Chinese legal judgments dataset and the results show that CLASS can generate more faithful summaries with less information lost, which shows the effectiveness of CLASS on capturing the deep contextualized information.