Describir: Self-Attention-Based Models for the Extraction of Molecular Interactions from Biological Texts