Addressing Resource and Privacy Constraints in Semantic Parsing Through Data Augmentation

Abstract

We introduce a novel setup for low-resource task-oriented semantic parsing which incorporates several constraints that may arise in real-world scenarios: (1) lack of similar datasets/models from a related domain, (2) inability to sample useful logical forms directly from a grammar, and (3) privacy requirements for unlabeled natural utterances. Our goal is to improve a low-resource semantic parser using utterances collected through user interactions. In this highly challenging but realistic setting, we investigate data augmentation approaches involving generating a set of structured canonical utterances corresponding to logical forms, before simulating corresponding natural language and filtering the resulting pairs. We find that such approaches are effective despite our restrictive setup: in a lowresource setting on the complex SMCalFlow calendaring dataset (Andreas et al., 2020), we observe 33% relative improvement over a nondata-augmented baseline in top-1 match.

Compositional Generalization for Kinship Prediction through Data Augmentation

“…In our first strategy, we explore the utility of data augmentation towards compositional generalization. Recent works have shown data augmentation to be an effective strategy in improving model performance on different NLP tasks such as Neural Machine Translation (Fernando and Ranathunga, 2022), semantic parsing (Yang et al, 2022), and text summarization (Wan and Bansal, 2022). Our data augmentation strategy focuses on improving a model's ability to extract relations that are explicitly mentioned in the text.…”