I think the assumption here is that the set of distinct words (hence, also the keys of the frequency map) will be small compared to the overall file.