Abstract: Transformer-based models are used to solve a va-riety of Natural Language Processing tasks. Still, these models are opaque and poorly understandable for their users. Current approaches to ...