No the point is you know exactly where T is just by looking at the dictionary (or at least, you learn this if you use a dictionary a lot).
IOW your prior on the data distribution lets you skip the first 4-5 binary chops.