Learning to Recombine and Resample Data for Compositional Generalization

Flexible neural models outperform grammar- and automaton-based counterparts on a variety of sequence modeling tasks. However, neural models perform poorly in settings requiring compositional generalization beyond the training data -- particularly to rare or unseen subsequences. Past work has found symbolic scaffolding (e.g. grammars or automata) essential in these settings. Here we present a family of learned data augmentation schemes that support a large category of compositional generalizations without appeal to latent symbolic structure. Our approach to data augmentation has two components: recombination of original training examples via a prototype-based generative model and resampling of generated examples to encourage extrapolation. Training an ordinary neural sequence model on a dataset augmented with recombined and resampled examples significantly improves generalization in two language processing problems---instruction following (SCAN) and morphological analysis (Sigmorphon 2018)---where our approach enables learning of new constructions and tenses from as few as eight initial examples.