NP Chunking (State of the art)

• Performance measure: F = 2 * Precision * Recall / (Recall + Precision)
• Precision: percentage of NPs found by the algorithm that are correct
• Recall: percentage of NPs defined in the corpus that were found by the chunking program
• Training data: sections 15-18 of Wall Street Journal corpus (Ramshaw and Marcus)
• Testing data: section 20 of Wall Street Journal corpus (Ramshaw and Marcus)
• original data of the NP chunking experiments by Lance Ramshaw and Mitch Marcus
• data contains one word per line and each line contains six fields of which only the first three fields are relevant: the word, the part-of-speech tag assigned by the Brill tagger, and the correct IOB tag
• dataset is available from ftp://ftp.cis.upenn.edu/pub/chunker/
• more information is available from NP Chunking

Table of results

System name	Short description	Main publications	Software	Results (F)
KM00	B-I-O tagging using SVM classifiers with polynomial kernel	Kudo and Matsumoto (2000)	YAMCHA Toolkit (but models are not provided)	93.79%
KM01	learning as in KM00, but voting between different representations	Kudo and Matsumoto (2001)	No	94.22%
SS05	specialized HMM + voting between different representations	Shen and Sarkar (2005)	No	95.23%

References

Kudo, T., and Matsumoto, Y. (2000). Use of support vector learning for chunk identification. Proceedings of the 4th Conference on CoNLL-2000 and LLL-2000, pages 142-144, Lisbon, Portugal.

Kudo, T., and Matsumoto, Y. (2001). Chunking with support vector machines. Proceedings of NAACL-2001.

Shen, H., and Sarkar, A. (2005). Voting between multiple data representations for text chunking. Proceedings of the Eighteenth Meeting of the Canadian Society for Computational Intelligence, Canadian AI 2005.

See also

• State of the art