Revision of Part-of-Speech Tagging in Stockholm Umeå Corpus 2.0

Many parsers use a part-of-speech tagger as a first step in parsing. The accuracy of the tagger naturally affects the performance of the parser. In this experiment, we revise 1500+ proposed errors in SUC 2.0 that were mainly found during work with schema parsing, and evaluate tagger instances trained on the revised corpus. The revisions turned out to be beneficial also for the taggers.Samarbete med Eva Forsbom, Uppsala universite