Effectiveness of Text, Acoustic, and Lattice-based representations in Spoken Language Understanding tasks

TL;DR

This paper evaluates various representations for intent classification in Spoken Language Understanding, comparing text-based, lattice-based, and multimodal systems, highlighting the benefits of richer ASR outputs and crossmodal approaches.

Contribution

It introduces a comprehensive benchmark of SLU systems using different representations, including a novel multimodal approach, and analyzes their performance under various conditions.

Findings

Richer ASR outputs improve SLU performance by 5.5%.

Crossmodal learning achieves 17.8% relative improvement over 1-best transcripts.

Multimodal approaches match oracle performance, overcoming transcript limitations.

Abstract

In this paper, we perform an exhaustive evaluation of different representations to address the intent classification problem in a Spoken Language Understanding (SLU) setup. We benchmark three types of systems to perform the SLU intent detection task: 1) text-based, 2) lattice-based, and a novel 3) multimodal approach. Our work provides a comprehensive analysis of what could be the achievable performance of different state-of-the-art SLU systems under different circumstances, e.g., automatically- vs. manually-generated transcripts. We evaluate the systems on the publicly available SLURP spoken language resource corpus. Our results indicate that using richer forms of Automatic Speech Recognition (ASR) outputs, namely word-consensus-networks, allows the SLU system to improve in comparison to the 1-best setup (5.5% relative improvement). However, crossmodal approaches, i.e., learning from acoustic and text embeddings, obtains performance similar to the oracle setup, a relative improvement of 17.8% over the 1-best configuration, being a recommended alternative to overcome the limitations of working with automatically generated transcripts.