Semantic Parsing in Task-Oriented Dialog with Recursive Insertion-based Encoder

TL;DR

This paper presents RINE, a recursive insertion-based encoder for semantic parsing in task-oriented dialog, achieving state-of-the-art accuracy and faster inference by incrementally building semantic parse trees.

Contribution

The introduction of RINE, a novel recursive insertion-based encoder that improves semantic parsing accuracy and efficiency in task-oriented dialog systems.

Findings

Achieves state-of-the-art exact match accuracy on TOP benchmark.

Performs on par with SOTA in nested named entity recognition.

Runs 2-3.5 times faster than sequence-to-sequence models.

Abstract

We introduce a Recursive INsertion-based Encoder (RINE), a novel approach for semantic parsing in task-oriented dialog. Our model consists of an encoder network that incrementally builds the semantic parse tree by predicting the non-terminal label and its positions in the linearized tree. At the generation time, the model constructs the semantic parse tree by recursively inserting the predicted non-terminal labels at the predicted positions until termination. RINE achieves state-of-the-art exact match accuracy on low- and high-resource versions of the conversational semantic parsing benchmark TOP (Gupta et al., 2018; Chen et al., 2020), outperforming strong sequence-to-sequence models and transition-based parsers. We also show that our model design is applicable to nested named entity recognition task, where it performs on par with state-of-the-art approach designed for that task. Finally, we demonstrate that our approach is 2-3.5 times faster than the sequence-to-sequence model at inference time.