Religious Bias Evaluation on Multilingual CrowS-Pairs (test)
4.17Bias Score (DE)XLM-R + MD w/ Full Fine-Tune
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| XLM-R + MD w/ Full Fine-TuneBase Model=XLM-R, Debiasing Strategy=Multiple-Debias, Tuning=Full Fine-Tune2026.04 | 4.17 | 6 | 1.24 | 9.34 | 5.19 | |
| mBERT + SDBase Model=mBERT, Debiasing Strategy=Self-Debias2026.04 | 6.67 | 6.77 | 12.24 | 2.52 | 7.05 | |
| XLM-R + MD w/ Prefix TuneBase Model=XLM-R, Debiasing Strategy=Multiple-Debias, Tuning=Prefix Tune2026.04 | 7.5 | 1.67 | 7.09 | 7.65 | 5.98 | |
| XLM-R + MD w/ Adapter TuneBase Model=XLM-R, Debiasing Strategy=Multiple-Debias, Tuning=Adapter Tune2026.04 | 9.17 | 0.86 | 10.51 | 10.19 | 7.68 | |
| XLM-R + MCDABase Model=XLM-R, Debiasing Strategy=Multilingual CDA2026.04 | 11.88 | 6.54 | 14.05 | 13.54 | 11.5 | |
| XLM-R + MD w/ Prompt TuneBase Model=XLM-R, Debiasing Strategy=Multiple-Debias, Tuning=Prompt Tune2026.04 | 13.33 | 10.13 | 7.03 | 8.48 | 9.74 | |
| mBERT + MSDBase Model=mBERT, Debiasing Strategy=Multilingual Self-Debias2026.04 | 15.83 | 8.46 | 15.53 | 3.37 | 10.8 | |
| XLM-R + CDABase Model=XLM-R, Debiasing Strategy=Counterfactual Data Augmentation2026.04 | 16.67 | 16.94 | 4.66 | 3.42 | 10.42 | |
| XLM-R + SDBase Model=XLM-R, Debiasing Strategy=Self-Debias2026.04 | 16.67 | 20.36 | 6.3 | 7.63 | 12.74 | |
| mBERT + MCDABase Model=mBERT, Debiasing Strategy=Multilingual CDA2026.04 | 17.71 | 16.57 | 22.55 | 3.36 | 15.05 | |
| mBERTBase Model=mBERT2026.04 | 19.17 | 18.65 | 22.33 | 4.19 | 16.08 | |
| XLM-R + MSDBase Model=XLM-R, Debiasing Strategy=Multilingual Self-Debias2026.04 | 20 | 19.47 | 2.91 | 10.98 | 13.34 | |
| XLM-RBase Model=XLM-R2026.04 | 21.67 | 22.03 | 17.22 | 1.62 | 15.64 | |
| mBERT + CDABase Model=mBERT, Debiasing Strategy=Counterfactual Data Augmentation2026.04 | 22.5 | 22.07 | 23.21 | 9.25 | 19.26 | |
| mBERT + MD w/ Full Fine-TuneBase Model=mBERT, Debiasing Strategy=Multiple-Debias, Tuning=Full Fine-Tune2026.04 | 22.5 | 11.88 | 19.79 | 2.54 | 14.18 | |
| mBERT + MD w/ Adapter TuneBase Model=mBERT, Debiasing Strategy=Multiple-Debias, Tuning=Adapter Tune2026.04 | 24.17 | 14.42 | 16.43 | 5.92 | 15.24 | |
| mBERT + MD w/ Prompt TuneBase Model=mBERT, Debiasing Strategy=Multiple-Debias, Tuning=Prompt Tune2026.04 | 24.17 | 16.11 | 17.31 | 5.87 | 15.86 | |
| mBERT + MD w/ Prefix TuneBase Model=mBERT, Debiasing Strategy=Multiple-Debias, Tuning=Prefix Tune2026.04 | 26.67 | 19.49 | 15.6 | 8.4 | 17.54 |