Statistical named entity recognition for Hungarian

In this paper, we present decision tree based statistical Named Entity recognizer system for Hungarian. The model was trained and tested on a segment of the Szeged Corpus, containing short business news articles, collected from MTI (Hungarian News Agency, www.mti.hul. We applied C4.5 for classificaton, and examined the accuracy of the system using training sets of different sizes. For this task we used only numerically encodable information (we excluded the word form itself), which contained some orthographical rules specific to Hungarian, but we trained for the recognition of foreign language proper nouns appearing frequently in business news as well. During the experiments the best results showed an accuracy of 89.6% F measure.