States are easier since you have a well-defined list (50 names + 50 postal abbreviations + a handful of alternate abbreviations (Mass, Miss, ...)). For cities or anything smaller, the only thing I can think of is catching capitalized words (proper nouns). You then need some lexical flag to differentiate locations from people's names - perhaps prepositions like on, in, at?

Update: On further consideration, this is definitely a job for AI. I note a number of possibilities with search terms like AI, Bayes, and neural net, though a lot of it is labeled alpha.