Let's say my tokenizer is only correct 60% of the time about where it splits words, would that be sufficient enough for basic searching capabilities?