Machine Translation on Sumerian-English
44.14BLEUVanilla Transformer (+5xFT)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Vanilla Transformer (+5xFT)Training Protocol=Semi-Supervised, Number of shards for FT=52021.05 | 44.14 | 2.504 | |
| Vanilla Transformer (+7xFT)Training Protocol=Semi-Supervised, Number of shards for FT=72021.05 | 42.95 | 2.367 | |
| Vanilla Transformer (AllComp)Training Protocol=Supervised, Data Configuration=AllComp2021.05 | 42.23 | 2.431 | |
| Vanilla Transformer (+3xFT)Training Protocol=Semi-Supervised, Number of shards for FT=32021.05 | 41.98 | 2.358 | |
| Vanilla Transformer (AllSeg)Training Protocol=Supervised, Data Configuration=AllSeg2021.05 | 37.01 | 2.36 | |
| Vanilla Transformer (UrIIISeg)Training Protocol=Supervised, Data Configuration=UrIIISeg2021.05 | 36.32 | 2.202 | |
| Vanilla Transformer (UrIIIComp)Training Protocol=Supervised, Data Configuration=UrIIIComp2021.05 | 33.45 | 2.242 | |
| XLM + Data Augmentation (BERT+CharSwap)Training Protocol=Unsupervised, Augmentation=BERT+CharSwap2021.05 | 30.1 | — | |
| XLM + Data Augmentation (BERT)Training Protocol=Unsupervised, Augmentation=BERT2021.05 | 29.5 | — | |
| XLM + Data Augmentation (CharSwap)Training Protocol=Unsupervised, Augmentation=CharSwap2021.05 | 29.04 | — | |
| XLM + Data Augmentation (WordNet)Training Protocol=Unsupervised, Augmentation=WordNet2021.05 | 28.57 | — | |
| XLM + Data Augmentation (BERT+WordNet)Training Protocol=Unsupervised, Augmentation=BERT+WordNet2021.05 | 26.57 | — | |
| XLM (Orig)Training Protocol=Unsupervised, Data=Orig2021.05 | 24.64 | — | |
| XLM (Mixed)Training Protocol=Unsupervised, Data=Mixed2021.05 | 21.23 | — | |
| XLM (MLM, Orig)Training Protocol=Unsupervised, Pre-training=MLM, Data=Orig2021.05 | 15.04 | — | |
| XLM + Data Augmentation (BERT+WordNet)Training Protocol=Supervised, Augmentation=BERT+WordNet2021.05 | 13.34 | — | |
| XLM + Data Augmentation (BERT+CharSwap)Training Protocol=Supervised, Augmentation=BERT+CharSwap2021.05 | 13.23 | — | |
| XLM (Mixed)Training Protocol=Supervised, Data=Mixed2021.05 | 13.08 | — | |
| XLM + Data Augmentation (WordNet)Training Protocol=Supervised, Augmentation=WordNet2021.05 | 13.08 | — | |
| XLM + Data Augmentation (BERT)Training Protocol=Supervised, Augmentation=BERT2021.05 | 13.06 | — | |
| XLM + Data Augmentation (CharSwap)Training Protocol=Supervised, Augmentation=CharSwap2021.05 | 12.92 | — | |
| XLM (Orig)Training Protocol=Supervised, Data=Orig2021.05 | 12.73 | — | |
| XLM (MLM, Orig)Training Protocol=Supervised, Pre-training=MLM, Data=Orig2021.05 | 4.49 | — | |
| XLM + Data Augmentation (BERT+WordNet)Training Protocol=Semi-Supervised, Augmentation=BERT+WordNet2021.05 | 1.46 | 1.666 | |
| XLM + Data Augmentation (BERT)Training Protocol=Semi-Supervised, Augmentation=BERT2021.05 | 1.32 | 1.704 | |
| XLM (Orig)Training Protocol=Semi-Supervised, Data=Orig2021.05 | 1.294 | — | |
| XLM + Data Augmentation (WordNet)Training Protocol=Semi-Supervised, Augmentation=WordNet2021.05 | 1.269 | 1.69 | |
| XLM (Mixed)Training Protocol=Semi-Supervised, Data=Mixed2021.05 | 1.104 | — | |
| XLM (MLM + TLM, WMT)Training Protocol=Supervised, Pre-training=MLM + TLM, Data=WMT2021.05 | 0.94 | — | |
| XLM + Data Augmentation (BERT+CharSwap+WordNet)Training Protocol=Semi-Supervised, Augmentation=BERT+CharSwap+WordNet2021.05 | — | 1.757 |