Unlocking Compositional Generalization in Pre-trained Models Using Intermediate Representations

TL;DR

This paper demonstrates that using carefully designed intermediate representations significantly enhances the compositional generalization of pre-trained seq2seq models in semantic parsing tasks, achieving state-of-the-art results.

Contribution

It introduces the use of intermediate representations in pre-trained seq2seq models without altering architecture, leading to substantial improvements in compositional generalization.

Findings

Achieved +14.8 accuracy on CFQ dataset

Improved accuracy by +15.0 to +19.4 on text-to-SQL datasets

Intermediate representations are a key factor for better generalization

Abstract

Sequence-to-sequence (seq2seq) models are prevalent in semantic parsing, but have been found to struggle at out-of-distribution compositional generalization. While specialized model architectures and pre-training of seq2seq models have been proposed to address this issue, the former often comes at the cost of generality and the latter only shows limited success. In this paper, we study the impact of intermediate representations on compositional generalization in pre-trained seq2seq models, without changing the model architecture at all, and identify key aspects for designing effective representations. Instead of training to directly map natural language to an executable form, we map to a reversible or lossy intermediate representation that has stronger structural correspondence with natural language. The combination of our proposed intermediate representations and pre-trained models is surprisingly effective, where the best combinations obtain a new state-of-the-art on CFQ (+14.8 accuracy points) and on the template-splits of three text-to-SQL datasets (+15.0 to +19.4 accuracy points). This work highlights that intermediate representations provide an important and potentially overlooked degree of freedom for improving the compositional generalization abilities of pre-trained seq2seq models.