User Study of a Collaborative Spoken-Language Interface
We have studied the effects of requiring a user to speak to a collaborative agent using a small, constrained subset of English. Our study used a prototype spoken-language collaborative agent for operating a personal video recorder(PVR), implemented using the COLLAGEN middleware. The user communicated his goals to the software agent in simple spoken English, and the agent asked for more information when necessary.
Background & Objective: Speech recognition technology has matured dramatically in the past few years, with the first generation of products using embedded speech recognition now coming to market. These products typically support only a very small set of commands. Our focus is a second generation of spoken-language interfaces, which will be more collaborative and conversational. One approach to maintaining low speech error rates in such systems is to use a small, constrained subset of a natural language.
Technical Discussion: Our study compared the task success that users had when the constrained subset of English they must use is either always available ("persistent help") or available only when requested ("non-persistent help"). All users were able to complete their tasks in a reasonable amount of time. Persistent-help users had about an 80% success rate in their first session using the system, while non-persistent help users had a 60% success rate. In subsequent sessions, where persistent help was not available, the non-persistent-help users' success rate improved to eventually reach 80%, while the persistent-help users' success remains unchanged. It appears that persistent-help users did not retain knowledge of the constrained subset language as readily as did the non-persistent-help users.
Technology Areas:
Spoken Language Interfaces
Off the Desktop Interaction and Display
Modification Date: September 14, 2007