SUBS: Subtree Substitution for Compositional Semantic Parsing

TL;DR

This paper introduces a subtree substitution data augmentation method for semantic parsing, significantly improving compositional generalization performance on benchmarks like SCAN and GeoQuery.

Contribution

It proposes a novel subtree substitution approach for data augmentation that enhances compositional generalization in semantic parsing models.

Findings

Achieved state-of-the-art results on GeoQuery compositional split.

Significantly improved performance on SCAN dataset.

Demonstrated effectiveness of subtree-based augmentation over rule-based methods.

Abstract

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.