A dynamic multi-application dialog engine for task-oriented voice user interfaces

摘要

This thesis introduces the Dymalog framework for spoken language dialog systems, which separates the applications from the actual dialog system. It facilitates the control of a plurality of applications through a single dialog system, changeable during run time. This is achieved by application-independent knowledge processing inside the dialog system, based on a hierarchical representation of obtained information (o²I -Trees). The approach enables the realization of generic dialog functionalities. Dymalog is composed of a collection of components; each serves mainly a single purpose. It fosters the generation of competing hypotheses during the processing of the user input in order to derive an optimal interpretation at a certain stage in the processing.The Marvin dialog system puts Dymalog into practice. We discuss selected interactions with various applications enabled for the operation through the system. The parameterized hypothesis selection process is considered in detail, especially the parameter estimation algorithm Grail, and the same holds for the development process in the generation of competing hypotheses for the user input.