GenIE uses a sequence-to-sequence model that takes unstructured text as input and autoregressively generates a structured semantic representation of the information expressed in it, in the form of (subject, relation, object) triplets, as output.