Big is beautiful Bootstrapping a PoS tagger for Swedish
Eva Forsbom · 2006
A statistical part-of-speech tagger trained on a one-million word Swedish corpus with validated tags was used to tag two considerably larger untagged corpora (≈ 78 and 20 million words, respectively) to bootstrap new, improved, tagger models. The new taggers all showed better accuracy both for seen and unseen words, and the best tagger had 97.02% overall accuracy evaluated on the original corpus (using 10-fold cross-validation).