Structure-aware Fine-tuning of Sequence-to-sequence Transformers for Transition-based AMR Parsing.