Possible Improvements

• Gazetteer: add campionato + [tedesco, spagnolo, ...]
• giocare come difensore = another frame?
• Feature: tag n-gram with ontology class, otherwise keep ENT
• Gazetteer: run it over the whole input (sentence), currently at token level
• Training: use sentence splitter to get 1-sentence examples
• Evaluation: use n-grams instead of links
• Supervised: if a sentence is not in the gold standard, the classifier should discard it (abstention)
• Unsupervised: is the SportsEvent mapping harmful?
• esordire may also trigger Partita
• Gold: add missing words due to errors in the wikiextractor (third-party lib)