monotonic alignment networks (monets)

These are used for sequence-to-sequence learning tasks and explicitly model the alignment between input and output sequences.