Translate First Reorder Later: Leveraging Monotonicity in Semantic Parsing

TL;DR

This paper introduces TPOL, a modular two-step semantic parsing method that first translates sentences monotonically and then reorders them, significantly enhancing compositional generalization over traditional seq2seq models.

Contribution

The paper proposes a novel two-step approach, TPOL, leveraging monotonic translation and reordering to improve semantic parsing performance.

Findings

TPOL outperforms seq2seq models on semantic parsing datasets.

Monotonic translation enables learning reliable lexico-logical patterns.

TPOL significantly improves compositional generalization.

Abstract

Prior work in semantic parsing has shown that conventional seq2seq models fail at compositional generalization tasks. This limitation led to a resurgence of methods that model alignments between sentences and their corresponding meaning representations, either implicitly through latent variables or explicitly by taking advantage of alignment annotations. We take the second direction and propose TPOL, a two-step approach that first translates input sentences monotonically and then reorders them to obtain the correct output. This is achieved with a modular framework comprising a Translator and a Reorderer component. We test our approach on two popular semantic parsing datasets. Our experiments show that by means of the monotonic translations, TPOL can learn reliable lexico-logical patterns from aligned data, significantly improving compositional generalization both over conventional seq2seq models, as well as over other approaches that exploit gold alignments.