Continual Dialogue State Tracking via Reason-of-Select Distillation

TL;DR

This paper introduces the Reason-of-Select distillation method, enhancing dialogue state tracking models with meta-reasoning to improve continual learning, reduce forgetting, and handle complex multi-domain dialogues effectively.

Contribution

The paper presents a novel meta-reasoning based distillation approach for continual dialogue state tracking, addressing catastrophic forgetting and the Value Selection Quandary.

Findings

Significantly improves DST performance across multiple domains.

Effectively mitigates catastrophic forgetting in continual learning.

Enhances model reasoning with multi-value resolution and semantic contrastive strategies.

Abstract

An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services and confronting catastrophic forgetting, along with a critical capability loss termed the "Value Selection Quandary." To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel 'meta-reasoning' capability. Meta-reasoning employs an enhanced multi-domain perspective, combining fragments of meta-knowledge from domain-specific dialogues during continual learning. This transcends traditional single-perspective reasoning. The domain bootstrapping process enhances the model's ability to dissect intricate dialogues from multiple possible values. Its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Additionally, two novel improvements, "multi-value resolution" strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers' reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method. The source code is provided for reproducibility.