Describir: Exploring attention patterns and neural activations in transformer architectures for sequence classification in context free grammars