# Neural Legal Judgment Prediction in English

**Authors:** Ilias Chalkidis, Ion Androutsopoulos, Nikolaos Aletras  
**arXiv:** [1906.02059](https://arxiv.org/abs/1906.02059)  
**Date:** 2019-06-05  
**Summary page:** https://www.summarizepaper.com/en/paper/1906.02059v1/

## Key points

- The paper addresses the task of legal judgment prediction in English
- Introduces a new English legal judgment prediction dataset from the European Court of Human Rights
- Evaluates various neural models and compares them to previous feature-based models
- Neural models outperform feature-based models in binary violation classification, multi-label classification, and case importance prediction tasks
- Investigates biases towards demographic information and employs data anonymization techniques for fairness
- Proposes a hierarchical version of BERT to overcome length limitations and improve analysis and prediction of legal judgments
- Presents advancements in legal judgment prediction by introducing an English dataset, demonstrating superiority of neural models, exploring biases, and proposing an extended BERT model

## Summary

The paper titled "Neural Legal Judgment Prediction in English" by Ilias Chalkidis, Ion Androutsopoulos, and Nikolaos Aletras addresses the task of legal judgment prediction, which involves automatically predicting the outcome of a court case based on a text describing the case's facts. While previous research on this topic has mainly focused on Chinese language, this study introduces a new English legal judgment prediction dataset comprising cases from the European Court of Human Rights. The authors evaluate various neural models using this dataset and compare their performance against previous feature-based models that relied on techniques like bags of words and topics. The results demonstrate that the neural models outperform the feature-based models in three specific tasks: binary violation classification, multi-label classification, and case importance prediction. Additionally, the paper investigates whether these models exhibit biases towards demographic information by employing data anonymization techniques to ensure fairness and mitigate any potential biases present in the predictions made by these models. As a side contribution, the authors propose a hierarchical version of BERT (Bidirectional Encoder Representations from Transformers), which overcomes BERT's length limitation and allows for more comprehensive analysis and prediction of legal judgments. Overall, this study presents significant advancements in legal judgment prediction by introducing an English dataset and demonstrating the superiority of neural models over traditional feature-based approaches as well as exploring potential biases and proposing an extended BERT model to improve fairness and accuracy in predicting legal outcomes.