Text this: Improving Transformer-Based Neural Machine Translation with Prior Alignments