Just today I heard about this guy with motor issues who can play competitive fighting games using only his tongue (not directly, using it to push his cheeks/skin) I would imagine combining something like that with voice recognition would be the best approach, something like press X before saying (on your mic) a variable name, or press Y before saying a control structure, such approach would remove a lot of ambiguity because the number of possible interpretations by the speech recognition software would be greatly reduced.