There is no proof that this approximates the Levenshtein distance. Might be interesting to show this is true. (or somehow characterize all the bad cases)