To me, it seems to be the same as the Norvig algorithm, except that only deletes are considered. Which allows them to pre-calculate misspellings with modest storage cost.

Am I mistaken?