Did you try entering one of those scrambled sentences into the tokenizer? It's not tokenizing words exactly- the doc says roughly ~100 tokens => 75 words.