Hi, I'm guessing that this is the right list for this stuff. Please advise, if this is not the case.
I've been sketching different on-screen keyboard layouts and while doing that I had this idea: What if, instead of simply recording a glyph, with each keypress the input system would record the exact on-screen position (or area) which was pressed. This position/area would then be compared to the positioning of keys on the on-screen keyboard, and a probability distribution of keys/glyphs would be generated. These probability distributions could then be used to correct/predict the entered text on a word-by-word basis. If this were to work in any reliable way, we could use comparatively small on-screen keys even for thumb input, and use this technique to fix some, if not most, of the resulting typos. -- Jani-Matti Hätinen