Argumentation Structure Prediction in CJEU Decisions on Fiscal State Aid

What is it about?

We address argumentative link prediction in decisions by Court of Justice of the European Union on fiscal state aid. We extend our Demosthenes dataset with a new annotation layer aimed at capturing the inferential connection(s) between a set of premises and their outcome(s) We compare two NLP models, experimenting with different thresholds of argumentative distance and oversampling. We obtain the best results with an ensemble of residual networks, that is also robust to class imbalance and less resource-intensive than the transformer model.

Featured Image

Why is it important?

Identifying the relations between arguments is a very challenging task, and it is crucial in legal argument mining, where it could help identify motivations behind judgments or even fallacies or inconsistencies.