CoDial: Interpretable Task-Oriented Dialogue Systems Through Dialogue Flow Alignment

Abstract

Recent advances in Large Language Models (LLMs) have enabled significant progress in building general-purpose dialogue systems. Despite their impressive capabilities, these models face serious challenges in real-world applications, where domain experts need to define the flow of the dialogues and steer the model's behaviour. Teaching specialized, unseen tasks to dialogue systems, such as LLMs, remains challenging due to the high cost of expert knowledge, the scarcity of annotated training data, and the technical barrier involved in specifying model behaviour. To support domain-specific applications, it is essential to build frameworks that enable non-technical experts to define, test, and refine system behaviour with minimal effort. Achieving this requires an interpretable approach that allows dialogue flow definition with high abstraction. In this work, we introduce a novel framework, CoDial (Code for Dialogue), that converts domain knowledge, represented as a novel structured heterogeneous graph, into executable conversation logic. CoDial can be easily implemented in existing guardrailing languages, such as Colang \cite{nvidia2024nemo}, to enable interpretable and user accessible specification of task-oriented dialogue systems. Empirically, CoDial achieves state-of-the-art performance on the STAR dataset for inference-based models and is competitive with similar baselines on the well-known MultiWOZ dataset. We also demonstrate CoDial's iterative improvement via manual and LLM-aided feedback, making it a practical tool for expert-guided alignment of LLMs in high-stakes domains.