A multi-layer conversation management approach for information seeking applications

Abstract

The paper describes a new conversation management approach which may apply to a class of mixed initiative information seeking applications. It employs an application-independent conversation model inspired by the conversation theory of Grosz and Sidner [1]. Based on the model, we design a multi-layer conversation manager which employs instance-based learning (IBL) to determine conversation plans, while performs modularized realization of each conversation move in a conversation plan. Because we adopt IBL to determine conversation plans, new conversation behaviors are easy to incorporate. Our realization of conversation moves is also application-independent. We illustrate this method with a real estate application.