While the increase of electronic devices brings greater flexibility and control to our everyday environments, task-specific configuration of the myriad of devices becomes increasingly burdensome to the user. Ideally, interfaces for collections of such devices in our everyday environment should be calm, easy to learn and use, support control of favorite task specific configurations and allow the use of familiar names for configurations without learning. We propose a user-extensible natural language spoken interface for environment and device control. In order to support user customization in a natural language interface, the system must simultaneously learn the configurations and names for those configurations. We call this the simultaneous naming and configuration (SNAC) problem.

In this thesis, we identify and describe an approach to the simultaneous naming and configuration problem, and present a specific realization of our approach to the SNAC problem called Illuminac. We find that this kind of user-extensible speech interface is quite effective for some types of users and thus an important component in a complete solution to device configuration. In particular, we observe that users are willing to train the system, able to remember their commands, and willing to use voice commands in many cases.




Download Full History