When silver glitters more than gold: Bootstrapping an Italian part-of-speech tagger for Twitter

摘要

We bootstrap a state-of-the-art part-of-speech tagger to tag Italian Twitterdata, in the context of the Evalita 2016 PoSTWITA shared task. We show thattraining the tagger on native Twitter data enriched with little amounts ofspecifically selected gold data and additional silver-labelled data scrapedfrom Facebook, yields better results than using large amounts of manuallyannotated data from a mix of genres.