作者: Yuji Matsumoto , Taku Kudo , Tetsuji Nakagawa
DOI:
关键词:
摘要: The accuracy of part-of-speech (POS) tagging for unknown words is substantially lower than that known words. Considering the high rate up-to-date statistical POS taggers, account a non-negligible portion errors. This paper describes prediction using Support Vector Machines. We achieve in tag substrings and surrounding context as features. Furthermore, we integrate this method with practical English tagger, 97.1%, higher conventional approaches.