DiactTOD: Learning Generalizable Latent Dialogue Acts for Controllable Task-Oriented Dialogue Systems

TL;DR

DiactTOD introduces a novel latent space model for dialogue acts that enables controllable, interpretable response generation in task-oriented dialogue systems, achieving state-of-the-art results across various data regimes.

Contribution

The paper proposes DiactTOD, a latent dialogue act model that generalizes across datasets and tasks, enabling controllable response generation without explicit act annotations.

Findings

Achieves state-of-the-art performance on MultiWOZ in zero-shot, few-shot, and full data settings.

Demonstrates effective control of dialogue responses via latent act representations.

Outperforms existing methods that rely on explicit annotations or reinforcement learning.

Abstract

Dialogue act annotations are important to improve response generation quality in task-oriented dialogue systems. However, it can be challenging to use dialogue acts to control response generation in a generalizable way because different datasets and tasks may have incompatible annotations. While alternative methods that utilize latent action spaces or reinforcement learning do not require explicit annotations, they may lack interpretability or face difficulties defining task-specific rewards. In this work, we present a novel end-to-end latent dialogue act model (DiactTOD) that represents dialogue acts in a latent space. DiactTOD, when pre-trained on a large corpus, is able to predict and control dialogue acts to generate controllable responses using these latent representations in a zero-shot fashion. Our approach demonstrates state-of-the-art performance across a wide range of experimental settings on the MultiWOZ dataset, including zero-shot, few-shot, and full data fine-tuning with both end-to-end and policy optimization configurations.