KG-Guided Multi-Task Pre-training for German Traffic Law Reasoning

This approach integrates Knowledge Graph (KG) guided clustering and paragraph classification into Transformer-based encoders to capture hierarchical regulatory constraints. By encoding structured legal knowledge directly into model weights, the method enables zero-shot legal reasoning without requiring explicit graph traversal during inference.