This hit home for me as a couple of months ago we faced the choice between continuing to develop a parser-based approach to location extraction from free text,

This hit home for me as a couple of months ago we faced the choice between continuing to develop a parser-based approach to location extraction from free text, or moving to an entity extraction and search approach, i.e. geotagging or geocoding. Notably we were just trying to get city and state, and even limiting ourselves to the U.S. the combinations we were seeing made parsing seem like a game of increasing complexity delivering diminishing returns. We ultimately went with a search-base approach and it's been working much better and is more tolerant of format variations.