Word completion happens after the fact, i.e. based on if I've typed "drea", it can predict "dreams" or "dreary" or whatever. The other type of prediction happens before the fact, i.e. based on if I've typed "drea", it predicts that pressing "m" or "r" is than for instance "n" ("drean?"), and therefore updates... ... you can think of it updating the size of the "m" key to be virtually much larger than the "n" key, so if you hit the edge of the "n" key it corrects it (initially) to the "m" key. Think of like an invisible button on top of every real button that you see. The sizes of these invisible buttons change after every key press.