The BERT-base parsing results for the full ceiling model and the probing model on the PTB Stanford Dependencies (SD) test set and CoNLL 2015 in-domain test set. The metrics and settings are identical to Table 1 except only one seed is used.
| . | PTB SD . | CoNLL 2015 DM . | ||||||
|---|---|---|---|---|---|---|---|---|
| Metrics . | AbsΔ . | RelΔ . | Full . | Probe . | AbsΔ . | RelΔ . | Full . | Probe . |
| LAS/F1 | −13.6 | −14.4% | 94.6 | 81.0 | −23.2 | −24.8% | 93.6 | 70.4 |
| LEM | −35.8 | −73.7% | 48.6 | 12.8 | −39.4 | −91.6% | 43.0 | 3.6 |
| UEM | −44.7 | −74.1% | 60.3 | 15.7 | −42.0 | −91.5% | 45.9 | 3.9 |
| . | PTB SD . | CoNLL 2015 DM . | ||||||
|---|---|---|---|---|---|---|---|---|
| Metrics . | AbsΔ . | RelΔ . | Full . | Probe . | AbsΔ . | RelΔ . | Full . | Probe . |
| LAS/F1 | −13.6 | −14.4% | 94.6 | 81.0 | −23.2 | −24.8% | 93.6 | 70.4 |
| LEM | −35.8 | −73.7% | 48.6 | 12.8 | −39.4 | −91.6% | 43.0 | 3.6 |
| UEM | −44.7 | −74.1% | 60.3 | 15.7 | −42.0 | −91.5% | 45.9 | 3.9 |