Code-switched text synthesis on De-En
32.65BLEUFine-tuned PMMTM on all language pairs (augment-MMT)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Fine-tuned PMMTM on all language pairs (augment-MMT)Type=Supervised2023.05 | 32.65 | 59.96 | |
| Fine-tuned PMMTM on all language pairs (mBART50-MMT)Type=Supervised2023.05 | 32.24 | 59.75 | |
| Gupta et al. (2020)Type=Supervised2023.05 | 24.15 | 30.47 | |
| GLOSS (augment-MMT + prefix)Type=Zero-shot transfer2023.05 | 21.88 | 50.33 | |
| GLOSS (mBART50-MMT + prefix)Type=Zero-shot transfer2023.05 | 20.49 | 48.49 | |
| GLOSS (mBART50-MMT + adapter)Type=Zero-shot transfer2023.05 | 18.63 | 48.28 | |
| GLOSS (augment-MMT + adapter)Type=Zero-shot transfer2023.05 | 14.58 | 40.75 | |
| Fine-tuned PMMTM on available language pairsType=Zero-shot transfer2023.05 | 9.09 | 32.34 | |
| Machine TranslationType=Unsupervised2023.05 | 6.3 | 30.28 | |
| Translate, Align, then SwapType=Unsupervised2023.05 | 5.53 | 27.3 | |
| Copy InputType=Unsupervised2023.05 | 3.29 | 22.76 |