András Kuba, László Felföldi, András Kocsor

Research Group on Artificial Intelligence,
University of Szeged

In this paper we will briefly survey the key results achieved so far in Hungarian POS tagging and show how classifier combination techniques can aid the POS taggers. Methods are evaluated on a manually annotated corpus containing 1.2 million words. POS tagger tests were performed on single-domain, multiple domain and cross-domain test settings, and, to improve the accuracy of the taggers, various combination rules were implemented. The results indicate that combination schemas (like the Boosting algorithm) are promising tools which can significantly degrade the classification errors, and produce a more effective tagger application.