Prompt Learning for Domain Adaptation in Task-Oriented Dialogue

TL;DR

This paper introduces a prompt learning approach that uses canonical forms and soft prompts with large language models to improve domain adaptation and reduce data collection efforts in task-oriented dialogue systems.

Contribution

It proposes a novel method of using prompt tuning with canonical forms for intent classification, enabling effective zero- and few-shot domain adaptation.

Findings

Canonical forms improve intent classification accuracy.

Prompt tuning with large language models enables zero-shot domain transfer.

Method reduces data collection and development effort.

Abstract

Conversation designers continue to face significant obstacles when creating production quality task-oriented dialogue systems. The complexity and cost involved in schema development and data collection is often a major barrier for such designers, limiting their ability to create natural, user-friendly experiences. We frame the classification of user intent as the generation of a canonical form, a lightweight semantic representation using natural language. We show that canonical forms offer a promising alternative to traditional methods for intent classification. By tuning soft prompts for a frozen large language model, we show that canonical forms generalize very well to new, unseen domains in a zero- or few-shot setting. The method is also sample-efficient, reducing the complexity and effort of developing new task-oriented dialogue domains.