(corpora) Tag: sourceedit |
(add ref) Tag: sourceedit |
||
| Line 14: | Line 14: | ||
* Unlabeled A5achment Score (UAS): % of tokens with correct HEAD • |
* Unlabeled A5achment Score (UAS): % of tokens with correct HEAD • |
||
* Label Accuracy (LA): % of tokens with correct DEPREL |
* Label Accuracy (LA): % of tokens with correct DEPREL |
||
| + | |||
| + | == References == |
||
| + | <references/> |
||
[[Category:Dependency parsing| ]] |
[[Category:Dependency parsing| ]] |
||
Revision as of 20:48, 26 January 2016
Two main streams:
- Transition-based
- Non-monotonic ~: http://www.aclweb.org/anthology/D/D15/D15-1162.pdf
- An important reference: Goldberg & Nirve (2013) they come close to reinforcement learning
- Graph-based
Corpora
Papers mainly use WSJ of Penn Treebank (TODO: confirm this). Although there is more in the treebank, only WSJ has been patched with gold NP-bracketing (Vadas & Curran, 2007)[1].
Evaluation
- Label Attachment Score (LAS): % of tokens for which a system has predicted the correct HEAD and DEPREL
- Unlabeled A5achment Score (UAS): % of tokens with correct HEAD •
- Label Accuracy (LA): % of tokens with correct DEPREL
References
- ↑ Vadas, D., & Curran, J. R. (2007). Adding noun phrase structure to the Penn Treebank. 45th Annual Meeting of the Association of Computational Linguistics, (June), 240–247. Retrieved from http://acl.ldc.upenn.edu/P/P07/P07-1031.pdf