It will likely be something like imagining typing, or swiping. It won't work at the level of words. You'd have to correlate between 50k and 250k individual words per person, which would be cumbersome and obnoxious. You have to correlate word components, and the easiest way to train is tapping into a feedback loop that already achieves your intent, like typing.

Systems like the palm pilot graffiti interface, or some optimized symbolic gesture system will become prevalent, but imagining typing will probably carry us sufficiently until better systems are worked out.