Task-Oriented Dialogue with In-Context Learning

TL;DR

This paper challenges the dominant intent-based NLU paradigm in task-oriented dialogue by combining in-context learning with a deterministic, flow-defined DSL to drive business logic. Dialogue Understanding uses LLMs to generate a small set of commands that are executed deterministically via a developer-defined flow stack, while Conversation Repair handles non-ideal dialogues out of the box. The approach yields high pass rates and significantly lower development effort than a traditional NLU baseline, demonstrating scalability and explainability across multiple tasks. By making the implementation open-source, the authors aim to accelerate practical adoption and further study of scalable, model-agnostic task-oriented dialogue systems.

Abstract

We describe a system for building task-oriented dialogue systems combining the in-context learning abilities of large language models (LLMs) with the deterministic execution of business logic. LLMs are used to translate between the surface form of the conversation and a domain-specific language (DSL) which is used to progress the business logic. We compare our approach to the intent-based NLU approach predominantly used in industry today. Our experiments show that developing chatbots with our system requires significantly less effort than established approaches, that these chatbots can successfully navigate complex dialogues which are extremely challenging for NLU-based systems, and that our system has desirable properties for scaling task-oriented dialogue systems to a large number of tasks. We make our implementation available for use and further study.