Dependency Parsing on 30 unseen languages split (test)
42.3Average LASworst-case-aware automated curriculum learning
Evaluation results
| Method | Links | |
|---|---|---|
| worst-case-aware automated curriculum learningphi=0.5, PLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 42.3 | |
| worst-case-aware automated curriculum learningphi=1, PLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 42.3 | |
| worst-case-aware automated curriculum learningphi=0, PLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 42.1 | |
| size-proportional samplingPLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 41.9 | |
| smooth-samplingPLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 41.7 | |
| uniform samplingPLM=XLM-R, evaluation_protocol=zero-shot2022.03 | 41.4 | |
| worst-case-aware automated curriculum learningphi=0, PLM=mBERT, evaluation_protocol=zero-shot2022.03 | 36.4 | |
| worst-case-aware automated curriculum learningphi=0.5, PLM=mBERT, evaluation_protocol=zero-shot2022.03 | 36.1 | |
| worst-case-aware automated curriculum learningphi=1, PLM=mBERT, evaluation_protocol=zero-shot2022.03 | 36.1 | |
| smooth-samplingPLM=mBERT, evaluation_protocol=zero-shot2022.03 | 35.2 | |
| uniform samplingPLM=mBERT, evaluation_protocol=zero-shot2022.03 | 35.2 | |
| size-proportional samplingPLM=mBERT, evaluation_protocol=zero-shot2022.03 | 35 |