This model is a fine-tuned version of ModernBERT-large for
Czech legal argument detection. It was introduced in the paper
Mining Legal Arguments to Study Judicial Formalism.
The model is part of the
MADON project, which focuses on detecting and classifying judicial reasoning in Czech court decisions. This specific model corresponds to
Task 1 in the paper: detecting whether a paragraph in a legal decision is argumentative or non-argumentative.
The model was adapted to the Czech legal domain through continued pretraining on a corpus of over 300,000 court decisions and fine-tuned on the MADON dataset. In the paper's evaluation, this model achieved a Balanced F1 score of 82.6% for argument detection.
You can use this model for presence classification of Czech legal arguments using the transformers library:
1from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
2
3model = AutoModelForSequenceClassification.from_pretrained("TrustHLT/ModernBERT-large-madon-arg-detection")
4tokenizer = AutoTokenizer.from_pretrained("TrustHLT/ModernBERT-large-madon-arg-detection")
5
6pipe = pipeline("text-classification", model=model, tokenizer=tokenizer)
7
8text = "This is a legal paragraph" # Replace with Czech legal text
9
10print(pipe(text))
1@article{madon2025,
2 title={Mining Legal Arguments to Study Judicial Formalism},
3 author={Anonymous},
4 journal={arXiv preprint arXiv:2512.11374},
5 year={2025}
6}
