APPLICATION OF WEIGHTED VOTING TAGGERS TO LANGUAGES DESCRIBED WITH LARGE TAGSETS
Marcin Kuta, Jacek Kitowski, Wojciech Wójcik, Michał Wrzeszcz · 2010
Manuscript received 7 July 2008; revised 2 July 2009Communicated by Milan RuskoAbstract. The paper presents baseline and complex part-of-speech taggers appliedto the modified corpus of Frequency Dictionary of Contemporary Polish, annotatedwith a large tagset. First, the paper examines accuracy of 6 baseline part-of-speechtaggers. The main part of the work presents simple weighted voting and complexvoting taggers. Special attention is paid to lexical voting methods and issues ofties and fallbacks. TagPair and WPDV voting methods achieve the top accuracyamong all considered methods. Error reduction 10.8% with respect to the bestbaseline tagger for the large tagset is comparable with other author’s results forsmall tagsets.Keywords: Part-of-speech tagging, combination tagger, weighted probability dis-tribution voting tagger, TagPair taggerMathematics Subject Classification 2000: 68T50, 68T05, 68T35