Exploiting structure of a keyboard...

Can we use the structure of a keyboard to improve spelling correction? Where will it work and where will it fail? The thought of exploiting the structure of a keyboard is very simple and intuitive. If you abstract it at a higher level then it boils down to - Just as you exploit the structure and syntax in language to enhance semantics, why not exploit the structure and syntax of a keyboard to enhance semantics. Also the vocabulary (or structure/syntax) of the keyboard is universal. I am not sure how keyboards in other languages are; but as we are dealing with English, I think the "universality" assumption is fair enough.

Actually taking this at one more level of abstraction, what we are actually trying to do is exploiting structure/syntax in the medium (device) that is used in communication of typed text. So the syntax will change if we used a cellphone keypad to send SMS or if we used a telephone with a rotating number dial (like how we used to have old phones in india) to type in a phone number. So, the rules or language (device) of communication might change but the underlying principle is still the same!

Just a thought - How about recommending (suggesting) a correct phone number given the history of user’s calls as well as the size and structure of the phone keypad! – we’ll probably have fewer wrong numbers :). To this we can add information related to the particular cell phone plan. People tend to make long distance calls when weekend and night minutes are free as opposed to using day time minutes, but you would not care if you had more day times as opposed to a cheaper plan which gives you fewer minutes!.

hmmm... sounds interesting to me... any takers?