The study aims to apply several methods towards machine learning interpretation to study the representation and inference in NLU Transformer-based models fine-tuned for Russian language.
- Structural Probing
- Feature Attribution
- Intervention-Based Analysis
- Visualization
- Language theory