Empirical Multi-Artifact Knowledge Modeling for Dialogue Systems

This paper presents a knowledge modeling approach to improve domain-independency in Spoken Dialogue Systems (SDS) architectures. We aim to support task oriented dialogue management strategies via an easy to use interface provided by an adaptive Domain Knowledge Manager (DKM). DKM is a broker that centralizes the knowledge of the domain using a Knowledge Integration Process (KIP) that merges on-the-fly local knowledge models. A local knowledge model defines a semantic interface and is associated to an artifact that can be a household appliance in a home domain or a cinema in a ticket-selling domain. We exemplify the reuse of a generic AmI domain model in a home domain and in a ticket-selling domain redefining the abstractions of artifact, class, and task. Our experimental setup is a domain simulator specially developed to reproduce an Ambient Intelligence (AmI) scenario.