Compositional generalization in a deep seq2seq model by separating syntax and semantics

Open in new window