Geonames has a good dataset of locations, and OpenStreetMaps has the regional boundary data.
So given something like "Boston, Massachusetts, USA" would take all the locations within the boundary box of "USA" and find the closest match to "Massachusetts" (using something like soundex), and then within Massachusetts find the closest match to "Boston" and return the long/lat.
Geonames has a good dataset of locations, and OpenStreetMaps has the regional boundary data.
So given something like "Boston, Massachusetts, USA" would take all the locations within the boundary box of "USA" and find the closest match to "Massachusetts" (using something like soundex), and then within Massachusetts find the closest match to "Boston" and return the long/lat.