Rule-Based Dialog Management for Voice Assistents in Automotive Environments


We describe a rule-based voice dialog system for in-vehicle infotainment. The system is specified by a declarative model comprising the dialog definition, dialog management logic, and domain knowledge, which are backed by an RDFS ontology. The approach simplifies coping with dialog phenomena such as anaphora or implicit confirmation, that can be addressed without writing code. We introduce the concept of a virtual knowledge base to accommodate the diversity of dialog-relevant knowledge sources in a car. It unifies the knowledge access for the dialog logic, no matter whether an internal knowledge base or an external source like a web service is being used. The virtual KB thus supports the model’s portability and re-usability. Dialog models can be exported and deployed to on-market voice assistants like Amazon Alexa and Google Assistant, which essentially means to extend their built-in dialog management capabilities. The dialog management run-time system is deployed on a separate cloud server and acts as a back-end for the respective voice assistant.

Year: 2020
In session: Poster
Pages: 185 to 192