Affiliations: Department of Computer Science and CAIR, Korea Advanced Institute of Science and Technology (KAIST), 373-1, Kusong-Dong, Yusong-Ku, Taejon 305-701, Korea, e-mail: jhoon@csking.kaist.ac.kr | Department of Computer Science, Sogang University, 1, Shinsoo-Dong, Mapo-Ku, Seoul, 121-742, Korea, e-mail: seojy@ccs.sogang.ac.kr | Department of Computer Science and CAIR, Korea Advanced Institute of Science and Technology (KAIST), 373-1, Kusong-Dong, Yusong-Ku, Taejon 305-701, Korea, e-mail: gckim@csking.kaist.ac.kr
Abstract: Part-of-speech (POS) tagging is a process of assigning a POS to each word in a sentence. Because many words are often ambiguous in their POSs, POS tagging must be able to select the most proper POS sequence for a given sentence. Recently, probabilistic approaches have shown very promising results to solve such ambiguity problems. Probabilistic approaches, however, usually require lots of training data to get reliable probabilities. To alleviate such restriction, we use fuzzy membership functions instead of probability distributions. Such a POS tagging model is called a fuzzy network POS tagging model. The membership functions are automatically estimated by using probabilities and neural networks with a learning algorithm. Experiments show that the performance of the fuzzy network POS tagging model is much better than that of a hidden Markov model under a limited amount of training data.
