MWE identification F1 of the parsing models vs. the mwetoolkit baseline (test set, sentences ≤ 40 words). FactLex* uses gold morphological analyses at test time.
| Model . | Arabic F1 . | French F1 . |
|---|---|---|
| mwetoolkit (baseline) | 15.0 | 37.3 |
| PAPCFG | 13.2 | 46.0 |
| SplitPCFG | 67.4 | 64.2 |
| Berkeley | 74.8 | 71.4 |
| DP-TSG | 81.9 | 71.3 |
| Stanford | 77.5 | 70.5 |
| Stanford+FactLex | 77.8 | 70.5 |
| Stanford+FactLex* | 78.2 | 87.3 |
| Model . | Arabic F1 . | French F1 . |
|---|---|---|
| mwetoolkit (baseline) | 15.0 | 37.3 |
| PAPCFG | 13.2 | 46.0 |
| SplitPCFG | 67.4 | 64.2 |
| Berkeley | 74.8 | 71.4 |
| DP-TSG | 81.9 | 71.3 |
| Stanford | 77.5 | 70.5 |
| Stanford+FactLex | 77.8 | 70.5 |
| Stanford+FactLex* | 78.2 | 87.3 |