This is a repository for code used in the paper: On the Role of Negative Precedent in Legal Outcome Prediction
- baseline_positive
- baseline_negative
- mtl
- claim_outcome
- joint_model
- claims
To train the Claim-Outcome model, first train a baseline_positive and claims model and provide a path to them using the --pos_path and --claim_path arguments. You must also set the --inference flag.
Create a conda environment with the envirionment.yml file.
To preprocess the datasets, first download from the links below:
- Outcome corpus: https://drive.google.com/file/d/1znbSf0vLJD-CxqpyzslxFw-vEe4qXOxw/view?usp=sharing
- Chalkidis et. al. corpus: https://drive.google.com/file/d/11ZvQf--QPb6Ut78YuNDq0sZNBVdI36by/view?usp=sharing
Make a new 'ECHR' directory and copy the Outcome corpus files into a 'ECHR/Outcome' sub-directory. Similarily, copy the Chalkdis et al. files into a 'ECHR/Chalkidis' sub-directory.
You can now run 'preprocess_data.py' to create the tokenized files.