Parsing (State of the art)
"Standard" measure:
PARSEVAL -- the evalb[1] program.
"Standard" datasets:
- Training: WSJ sections 2-22
- Testing: WSJ section 23
| System Name | Short Description | Main Publications | Software (if available) | Results | Comments (i.e. extra resources used, train/test times, ...) |
|---|---|---|---|---|---|
| Johnson & Charniak's Parser | Lexicalized N-Best PCFG + Discriminative re-reanking | Mark Johnson and Eugene Charniak, ACL 2005 [2] | Available [3] | 91.4 | Works well also on Brown |
| Collins' Parser | Lexicalized PCFG | Collins' Thesis [4], Bikel's Thesis [5] | Dan Bikel's Implementation [6] | ??? | |
| Berkeley Parser | Automatically induced PCFG | Petrov, Barrett, Thibaux and Klein, ACL 2006 [7], Petrov and Klein, NAACL 2007 [8] | Available [9] | 90.1 |
Works well also for Chinese and German |