A Rule-Based Tagger Development Framework

Zoltán Alexin, Péter Leipold, János Csirik, Károly Bibok, and Tibor Gyimóthy. In Lubos Popelínský and Miloslav Nepil, editors, Proceedings of the 3d Workshop on Learning Language in Logic, pages 1--10, Strasbourg, France, September 2001.

Abstract

POS (Part-of-speech) tagging is an important step in natural language processing because identically written words may have different meanings. Part-of-speech tagging is the procedure during which the correct morphological annotation (the correct tag) for an ambiguous word is selected. Computer programs able to do the process automatically are called POS taggers. In this paper the RTDF (a Rule-based Tagger Development Framework) is presented that is capable identifying general tagging rules using different machine learning tools given a suitably large training data set. The framework can combine the learned tagging rules, and evaluate the resulted taggers. The authors are participants of a project for developing a medium sized learning corpus for Hungarian. The corpus contains 1 million words and -- among others -- can serve as a suitable medium on which the previously developed POS-taggers can be tested. During the project the morphological analyzer for Hungarian has been thoroughly investigated and the MSD encoding has been refined. The development of the corpus is going on and will be completed at the end of 2001.

BibTeX entry.

Other publications

Z Alexin, alexin@inf.u-szeged.hu,