Non-Autoregressive Semantic Parsing for Compositional Task-Oriented Dialog

TL;DR

This paper introduces a non-autoregressive, convolutional neural network-based semantic parsing model that significantly reduces latency and model size while maintaining competitive accuracy across multiple datasets.

Contribution

It presents a novel non-autoregressive architecture combining CNNs for semantic parsing, enabling faster inference suitable for real-time conversational systems.

Findings

Achieves up to 81% latency reduction on TOP dataset

Reduces parameter size compared to RNN models

Maintains competitive performance on multiple datasets

Abstract

Semantic parsing using sequence-to-sequence models allows parsing of deeper representations compared to traditional word tagging based models. In spite of these advantages, widespread adoption of these models for real-time conversational use cases has been stymied by higher compute requirements and thus higher latency. In this work, we propose a non-autoregressive approach to predict semantic parse trees with an efficient seq2seq model architecture. By combining non-autoregressive prediction with convolutional neural networks, we achieve significant latency gains and parameter size reduction compared to traditional RNN models. Our novel architecture achieves up to an 81% reduction in latency on TOP dataset and retains competitive performance to non-pretrained models on three different semantic parsing datasets. Our code is available at https://github.com/facebookresearch/pytext