Data Recombination for Neural Semantic Parsing

TL;DR

This paper introduces data recombination, a framework that injects prior logical regularities into neural semantic parsing models by inducing a context-free grammar and training with an attention-based copying mechanism, leading to improved accuracy.

Contribution

The paper proposes a novel data recombination framework that incorporates prior knowledge into neural semantic parsing models through grammar induction and specialized training.

Findings

Achieved state-of-the-art results on GeoQuery dataset.

Improved neural semantic parsing accuracy across three datasets.

Demonstrated effectiveness of grammar-based data augmentation.

Abstract

Modeling crisp logical regularities is crucial in semantic parsing, making it difficult for neural models with no task-specific prior knowledge to achieve good results. In this paper, we introduce data recombination, a novel framework for injecting such prior knowledge into a model. From the training data, we induce a high-precision synchronous context-free grammar, which captures important conditional independence properties commonly found in semantic parsing. We then train a sequence-to-sequence recurrent network (RNN) model with a novel attention-based copying mechanism on datapoints sampled from this grammar, thereby teaching the model about these structural properties. Data recombination improves the accuracy of our RNN model on three semantic parsing datasets, leading to new state-of-the-art performance on the standard GeoQuery dataset for models with comparable supervision.