Analysing Inconsistencies and Errors in PoS Tagging in two Icelandic Gold Standards
Steinþór Steingrímsson, Sigrún Helgadóttir, Eiríkur Rögnvaldsson · DSpace repository (University of Tartu) · 2015
This paper describes work in progress.We experiment with training a state-of-the-art tagger, Stagger, on a new gold standard, MIM-GOLD, for the PoS tagging of Icelandic.We compare the results to results obtained using a previous gold standard, IFD.Using MIM-GOLD, tagging accuracy is considerably lower, 92.76% compared to 93.67% accuracy for IFD.We analyze and classify the errors made by the tagger in order to explain this difference.We find that inconsistencies and incorrect tags in MIM-GOLD may account for this difference.