SUBS: Subtree Substitution for Compositional Semantic Parsing
This addresses the problem of compositional generalization for semantic parsing systems, which is an incremental improvement over prior data augmentation methods.
The paper tackles compositional generalization in semantic parsing by proposing subtree substitution for data augmentation, where semantically similar subtrees are treated as exchangeable. Experiments showed this approach achieved significantly better performance on SCAN and GeoQuery benchmarks, reaching new state-of-the-art on the compositional split of GeoQuery.
Although sequence-to-sequence models often achieve good performance in semantic parsing for i.i.d. data, their performance is still inferior in compositional generalization. Several data augmentation methods have been proposed to alleviate this problem. However, prior work only leveraged superficial grammar or rules for data augmentation, which resulted in limited improvement. We propose to use subtree substitution for compositional data augmentation, where we consider subtrees with similar semantic functions as exchangeable. Our experiments showed that such augmented data led to significantly better performance on SCAN and GeoQuery, and reached new SOTA on compositional split of GeoQuery.