Spec-TOD: A Specialized Instruction-Tuned LLM Framework for Efficient Task-Oriented Dialogue Systems

TL;DR

Spec-TOD is a new framework that enhances task-oriented dialogue systems by using instruction-tuned LLMs and efficient training strategies, enabling strong performance with limited labeled data.

Contribution

It introduces a specialized end-to-end TOD framework with explicit task instructions and a lightweight training approach for low-resource scenarios.

Findings

Achieves competitive results on MultiWOZ dataset

Reduces labeled data requirements significantly

Demonstrates effectiveness of specialized LLMs in TOD

Abstract

Task-oriented dialogue (TOD) systems facilitate goal-driven interactions between users and machines. While recent advances in deep learning have improved the performance, TOD systems often struggle in low-resource scenarios with limited labeled data. To address this challenge, we propose Spec-TOD, a novel framework designed to train an end-to-end TOD system with limited data. Spec-TOD introduces two main innovations: (i) a novel specialized end-to-end TOD framework that incorporates explicit task instructions for instruction-tuned large language models (LLMs), and (ii) an efficient training strategy that leverages lightweight, specialized LLMs to achieve strong performance with minimal supervision. Experiments on the MultiWOZ dataset, a widely used TOD benchmark, demonstrate that Spec-TOD achieves competitive results while significantly reducing the need for labeled data. These findings highlight the potential of the proposed framework in advancing efficient and effective TOD systems in low-resource settings.