SUPPLE: A Dialogue Management Approach Based on Conversation Patterns

Abstract

AbstractWe propose SUPPLE, a new class of dialogue management systems that takes the core concept of a dialogue sequence as its main starting point. SUPPLE is inspired by the conversation patterns from the Natural Conversation Framework (NCF). While NCF primarily provides a design framework, we propose to automate the selection and updating of dialogue sequences as a central component of the dialogue management module, enabling the dialogue system to build a hierarchical dialogue structure at run-time. The conversation patterns are combined with the key concepts of update strategies and agenda adopted from the Information State Update approach. We formally describe the building blocks of our approach, and show how dialogue competencies like sequence expansion and slot-filling can be performed in our approach. These are further illustrated in a cooking assistant scenario.