Automatic Evaluation of Assistive Interfaces: IUI 2008

Comments
not yet...

Summary
In Automatic Evaluation of Assistive Interfaces, the researchers wanted to take existing HCI user modeling programs and extend them to simulate the actions of disabled users. HCI modelling programs are used to evaluate interfaces by providing a simulated "user" that performs optimally in the interface. But, there is no current model that simulates a disabled user of the system. The researchers wanted to present a new model that would simulate the disabled user as well as a normal user in order to evaluate an "assistive interface" (one that helps disabled users with the system) without having to find lots of disabled people to test the system.

The system was built to simulate many different things:

• Simulating the Practice Phase: Assuming that the user had no idea how to use the system, and could not read the buttons but knew where they were (a blind person and the 'tab' key), the model could try out different options and learn from the feedback.
• Visual Simulation: Using the actions of the keyboard and positions of the mouse, the interface can track the "visual location" of the eyes in order to change interaction and assist the user.
• Motor Simulation: Using a variety of specified disabilities, the model was able to simulate the time it would take for the user to select an option or use the mouse, allowing the interface to be evaluated and new interaction paradigms developed.

In general, the model that the researchers developed was able to accurately simulate a variety of disabilities in order to evaluate assitive interfaces without the need for lots of participants.

Discussion
While this seems useful for developers who want to save time developing applications, it seems that we would still need to do user studies with actual participants that are disabled and get their input into how they would like an interface to work. In general, models are an approximation to reality, and it doesn't seem that you could produce an accurate model to exactly simulate a disabled person (except if you have a model that would only navigate through audible feedback, possibly).