Generating Sequences by Learning to Self-Correct

Presents SELF-CORRECTION, an approach that decouples an imperfect base generator (an off-the-shelf LM) from a separate corrector that learns to iteratively correct imperfect generations.

This approach can improve the accuracy of sequence generation applications that require satisfying semantic constraints. It can be used to correct imperfect outputs from language models even when the corrector is much smaller than the base generator.