Using a dataset of nearly 9 billion tweeted words to identify the new American vocabulary and geolocate its origins