Has anyone tried to improve upon the Dasher eye-tracking thing for text composition? Seems like there could be more ideas here that haven't really been explored.

http://www.inference.org.uk/dasher/DasherSummary2.html