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

https://github.com/DiegoRossini/Modal-verbs-modality-detector.git