I think what I don't understand is the concept: "and so just a few letters are enough to recognize whole word".
This isn't expressed in the example. The rest of the word isn't hidden, it's still there in my peripheral vision (?). It is basically the same as I normally read.
If the unboldened parts of the words were actually discarded, I am not sure if I would be able to recognize all the original words, at least not efficiently and faster than the original.