Modal Verbs Modality Detector V1
Overview
The "Modal Verbs Modality Detector" project aims to build a machine learning model that detects and classifies the different modalities of French verbs. This model specifically predicts modality for the French verb 'pouvoir'. The labels for token classification are 'sporadicité', 'possibilité matérielle ou capacité', 'éventualité', and 'permission'. The project leverages data augmentation, cross-validation, and other techniques to ensure the model's robustness and accuracy.
Model Details
- Architecture: Based on FlauBERT
- Language: French
- Task: Modality detection for the verb "pouvoir"
- Version: 1.0
Training Hyperparameters
- K-Fold Cross-Validation: k = 5
- Number of Epochs: 7
- Learning Rate: 2e-5
- Batch Size: 5
Performance Metrics
| Class | Precision | Recall | F1-Score | Support | Confusion Matrix (TP, FP, FN, TN) |
|---|
| 0 | 1.00 | 1.00 | 1.00 | 22,083 | (22082, 1, 0, 0) |
| éventualité | 0.93 | 1.00 | 0.96 | 41 | (41, 0, 0, 0) |
| permission | 0.95 | 0.96 | 0.95 | 54 | (52, 2, 0, 0) |
| possibilité matérielle ou capacité | 0.90 | 0.81 | 0.85 | 43 | (35, 3, 2, 3) |
| sporadicité | 0.93 | 0.91 | 0.92 | 44 | (40, 1, 1, 2) |
Contributors
- Diego Rossini - Université Paris Nanterre
- Anna Colli - Université Paris Nanterre, Laboratoire MoDyCo
License
MIT
Upcoming Publication
An article detailing the methodology, data, and training process of the "Modal Verbs Modality Detector" will be published soon. Stay tuned for more information and in-depth insights on this project.
Git
