Abstract

Abstract

Adaptable multimodal systems are difficult to test. We present a methodology for evaluating parallel multimodal output which is generated in response to a specific set of user, device and situation constraints. Our method involves ranking of many potential output variants using a fitness function, and selecting well-differentiated variants for user testing. We focus on the generation of multiple variants of graphical user interfaces for small-screen devices and natural language voice output, within a system we term UI on the Fly.

[Close window]