POS Induction (State of the art)
Evaluation
Many-to-1: Map every induced label to a gold standard tag greedily (45 labels to 45 tags of the Penn tag set). Use the mapping to compute tag accuracy on the Wall Street Journal portion of the Penn TreeBank.
Results
Listed in order of decreasing accuracy
| System name | Short description | Main publications | Software | Many-to-1 |
|---|---|---|---|---|
| UPOS | Learning Syntactic Categories Using Paradigmatic Representations of Word Context | Yatbaz et al. (2012) | upos | 80.2% |
| Brown+proto | MRF initialized with Brown prototypes | Christodoulopoulos, Goldwater and Steedman (2010) | 76.1% | |
| Logistic regression with features and LBFGS | Berg-Kirkpatrick et al. (2010) | 75.5% | ||
| Clark DMF | Distributional clustering + morphology + frequency | Clark (2003) | alexc | 71.2%* |
* according to Christodoulopoulos, Goldwater and Steedman (2010)
References
Listed alphabetically.
- Berg-Kirkpatrick, Taylor, Alexandre Bouchard-Cote, John DeNero, and Dan Klein. 2010. Painless Unsupervised Learning with Features. NAACL 2010.
- Christodoulopoulos, Christos, Sharon Goldwater and Mark Steedman. 2010. Two Decades of Unsupervised POS induction: How far have we come? In Proceedings of EMNLP 2010.
- Clark, Alexander. 2003. Combining distributional and morphological information for part of speech induction. In Proceedings of EACL 2003, pages 59–66, Morristown, NJ, USA.
- Yatbaz, Mehmet Ali, Enis Sert and Deniz Yuret. 2012. Learning Syntactic Categories Using Paradigmatic Representations of Word Context. In Proceedings of EMNLP 2012, pages 940–951.