Graph-Based Decoding for Task Oriented Semantic Parsing

TL;DR

This paper proposes a graph-based decoding approach for task-oriented semantic parsing, demonstrating competitive performance and enhanced data efficiency compared to traditional sequence-to-sequence methods, especially with limited or partial data.

Contribution

It introduces a novel graph-based decoding paradigm for semantic parsing, contrasting the dominant sequence-to-sequence approach, and evaluates its effectiveness across various data conditions.

Findings

Competitive with sequence decoders on standard datasets

Significantly better with limited training data

Outperforms in partially-annotated data scenarios

Abstract

The dominant paradigm for semantic parsing in recent years is to formulate parsing as a sequence-to-sequence task, generating predictions with auto-regressive sequence decoders. In this work, we explore an alternative paradigm. We formulate semantic parsing as a dependency parsing task, applying graph-based decoding techniques developed for syntactic parsing. We compare various decoding techniques given the same pre-trained Transformer encoder on the TOP dataset, including settings where training data is limited or contains only partially-annotated examples. We find that our graph-based approach is competitive with sequence decoders on the standard setting, and offers significant improvements in data efficiency and settings where partially-annotated data is available.