Multilingual Natural Language Processing on Aggregate 234 languages
54.1ScoreTIPA
Evaluation results
| Method | Links | |
|---|---|---|
| TIPAaggregation=Typologically weighted parameter aggregation, base_model=XLM-RoBERTa base2026.01 | 54.1 | |
| No Train but Gainmethod=Additive combination of English adapter and closest available adapter, base_model=XLM-RoBERTa base2026.01 | 50.1 | |
| MAD-Xadapter=Target-language or typologically closest, base_model=XLM-RoBERTa base2026.01 | 45.4 | |
| English-only fine-tuningtraining=English task data, inference=Zero-shot, base_model=XLM-RoBERTa base2026.01 | 45 | |
| Parameter averagingaggregation=Unweighted parameter averaging, base_model=XLM-RoBERTa base2026.01 | 43.1 |