Ran into a lot of examples like that when merging two existing georeferenced datasets of plague outbreaks in Europe. Had to do quite some manual checking, and made good use of the damerau-levenshtein distance between two strings to recognize alternative spellings.