Reproduce “German BERT Model for Legal Named Entity Recognition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Harshil Darji, Jelena Mitrović, Michael Granitzer
Researchers at the University of Passau fine-tuned a German BERT model for Legal Named Entity Recognition (LER), establishing a new benchmark by outperforming previous state-of-the-art BiLSTM-CRF+ models on German legal texts, with significant F1-score gains across various legal entity types. The fine-tuned model has been made publicly available.
Key insights
- Transformer-based models, when fine-tuned on domain-specific data, can significantly advance performance in specialized NLP tasks like Legal NER.
- Prior exposure to domain-specific data during the pre-training phase of a language model (e.g., German BERT with OpenLegalData) provides a substantial advantage for subsequent fine-tuning.
- Performance disparities for low-resource entity classes (e.g., 'Country', 'Brand', 'Landscape') persist even with advanced models, highlighting the importance of dataset balance or specialized techniques for such categories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.