Browse › Natural Language Processing › Machine Translation › WMT2014 English-German
WMT2014 English-German Benchmark (Machine Translation)
Machine translation is the task of translating a sentence in a source language to a different target language.
Approaches for machine translation can range from rule-based to statistical to neural-based. More recently, encoder-decoder attention-based architectures like BERT have attained major improvements in machine translation.
One of the most popular datasets used to benchmark machine translation systems is the WMT family of datasets. Some of the most commonly used evaluation metrics for machine translation systems include BLEU, METEOR, NIST, and others.
The archive carries no text for this table; the description above is the archive's text for the task Machine Translation. archive 2025-07-28
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: BLEU score (higher is better), Number of Params (lower is better). Not inferred (points only, no best-so-far line): SacreBLEU, Hardware Burden, Operations per network pass. Points are placed at the row's paper date; 89 of 91 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Transformer Cycle (Rev) | 35.14 | 33.54 | – | Paper | Code | 2021 | 2 of 2 ran · 0 unverified | report | |||
| 2 | Noisy back-translation | 35.0 | 33.8 | 146G | ✓ | Paper | Code | 2018 | 1 of 1 ran · 0 unverified | report | ||
| 3 | Transformer+Rep(Uni) | 33.89 | 32.35 | – | Paper | Code | 2021 | 1 of 1 ran · 0 unverified | report | |||
| 4 | T5-11B | 32.1 | 11110M | – | Paper | Code | 2019 | 2 of 31 ran · 29 unverified | report | |||
| 5 | BiBERT | 31.26 | – | Paper | Code | 2021 | 2 of 2 ran · 0 unverified | report | ||||
| 6 | Transformer + R-Drop | 30.91 | 49G | – | Paper | Code | 2021 | 2 of 6 ran · 4 unverified | report | |||
| 7 | Bi-SimCut | 30.78 | – | Paper | Code | 2022 | linked, not harvested | report | ||||
| 8 | BERT-fused NMT | 30.75 | – | Paper | Code | 2020 | linked, not harvested | report | ||||
| 9 | Data Diversification - Transformer | 30.7 | – | Paper | Code | 2019 | 3 of 4 ran · 1 unverified | report | ||||
| 10 | SimCut | 30.56 | – | Paper | Code | 2022 | linked, not harvested | report | ||||
| 11 | Mask Attention Network (big) | 30.4 | 215M | – | Paper | Code | 2021 | 0 of 3 ran · 3 unverified | report | |||
| 12 | Transformer (ADMIN init) | 30.1 | 29.5 | 256M | – | Paper | Code | 2020 | 6 of 9 ran · 3 unverified | report | ||
| 13 | PowerNorm (Transformer) | 30.1 | – | Paper | Code | 2020 | linked, not harvested | report | ||||
| 14 | Depth Growing | 30.07 | 24G | – | Paper | Code | 2019 | linked, not harvested | report | |||
| 15 | MUSE(Parallel Multi-scale Attention) | 29.9 | – | Paper | Code | 2019 | 0 of 1 ran · 1 unverified | report | ||||
| 16 | Evolved Transformer Big | 29.8 | 29.2 | 218M | – | Paper | Code | 2019 | linked, not harvested | report | ||
| 17 | OmniNetP | 29.8 | – | Paper | Code | 2021 | linked, not harvested | report | ||||
| 18 | DynamicConv | 29.7 | 213M | – | Paper | Code | 2019 | linked, not harvested | report | |||
| 19 | Local Joint Self-attention | 29.7 | – | Paper | Code | 2019 | linked, not harvested | report | ||||
| 20 | TaLK Convolutions | 29.6 | 209M | – | Paper | Code | 2020 | 5 of 7 ran · 2 unverified | report | |||
| 21 | Transformer Big + MoS | 29.6 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 22 | AdvAug (aut+adv) | 29.57 | – | Paper | – | 2020 | no code linked | report | ||||
| 23 | PartialFormer | 29.56 | 68M | – | Paper | Code | 2023 | linked, not harvested | report | |||
| 24 | Transformer Big + adversarial MLE | 29.52 | – | Paper | Code | 2019 | linked, not harvested | report | ||||
| 25 | Transformer Big | 29.3 | 210M | 9G | – | Paper | Code | 2018 | linked, not harvested | report | ||
| 26 | Subformer-xlarge | 29.3 | – | Paper | – | 2021 | no code linked | report | ||||
| 27 | SB-NMT | 29.21 | – | Paper | Code | 2019 | linked, not harvested | report | ||||
| 28 | Transformer (big) + Relative Position Representations | 29.2 | – | Paper | Code | 2018 | 13 of 23 ran · 10 unverified | report | ||||
| 29 | FLOATER-large | 29.2 | – | Paper | Code | 2020 | 3 of 6 ran · 3 unverified | report | ||||
| 30 | Local Transformer | 29.2 | – | Paper | – | 2018 | no code linked | report | ||||
| 31 | Transformer Big with FRAGE | 29.11 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 32 | Mask Attention Network (base) | 29.1 | 63M | – | Paper | Code | 2021 | 0 of 3 ran · 3 unverified | report | |||
| 33 | Mega | 29.01 | 27.96 | 67M | – | Paper | Code | 2022 | 12 of 13 ran · 1 unverified | report | ||
| 34 | adequacy-oriented NMT | 28.99 | – | Paper | – | 2018 | no code linked | report | ||||
| 35 | LightConv | 28.9 | 202M | – | Paper | Code | 2019 | linked, not harvested | report | |||
| 36 | Weighted Transformer (large) | 28.9 | – | Paper | Code | 2017 | 2 of 2 ran · 0 unverified | report | ||||
| 37 | universal transformer base | 28.9 | – | Paper | Code | 2018 | 14 of 25 ran · 11 unverified | report | ||||
| 38 | KERMIT | 28.7 | – | Paper | – | 2019 | no code linked | report | ||||
| 39 | T2R + Pretrain | 28.7 | – | Paper | Code | 2021 | 3 of 8 ran · 5 unverified | report | ||||
| 40 | AdvAug (aut) | 28.58 | – | Paper | – | 2020 | no code linked | report | ||||
| 41 | RNMT+ | 28.5 | 44G | 2.81G | – | Paper | Code | 2018 | linked, not harvested | report | ||
| 42 | Synthesizer (Random + Vanilla) | 28.47 | – | Paper | Code | 2020 | 1 of 1 ran · 0 unverified | report | ||||
| 43 | Hardware Aware Transformer | 28.4 | 48M | – | Paper | Code | 2020 | linked, not harvested | report | |||
| 44 | Transformer Big | 28.4 | 871G | 2300000000.0G | – | Paper | Code | 2017 | 600 of 946 ran · 346 unverified | report | ||
| 45 | Transformer + SRU | 28.4 | 34G | – | Paper | Code | 2017 | 1 of 1 ran · 0 unverified | report | |||
| 46 | Evolved Transformer Base | 28.4 | 2488G | – | Paper | Code | 2019 | linked, not harvested | report | |||
| 47 | Rfa-Gate-arccos | 28.2 | – | Paper | – | 2021 | no code linked | report | ||||
| 48 | Transformer-DRILL Base | 28.1 | – | Paper | Code | 2019 | linked, not harvested | report | ||||
| 49 | AdvAug (mixup) | 28.08 | – | Paper | – | 2020 | no code linked | report | ||||
| 50 | CMLM+LAT+4 iterations | 27.35 | – | Paper | Code | 2020 | 3 of 5 ran · 2 unverified | report | ||||
| 51 | Transformer Base | 27.3 | 330000000.0G | – | Paper | Code | 2017 | 600 of 946 ran · 346 unverified | report | |||
| 52 | Levenshtein Transformer (distillation) | 27.27 | – | Paper | Code | 2019 | linked, not harvested | report | ||||
| 53 | DisCo + Mask-Predict (non-autoregressive) | 27.06 | – | Paper | Code | linked, not harvested | report | |||||
| 54 | Adaptively Sparse Transformer (alpha-entmax) | 26.93 | – | Paper | Code | 2019 | 3 of 3 ran · 0 unverified | report | ||||
| 55 | ResMLP-12 | 26.8 | – | Paper | Code | 2021 | 2 of 7 ran · 5 unverified | report | ||||
| 56 | CNAT | 26.6 | – | Paper | Code | 2021 | linked, not harvested | report | ||||
| 57 | Lite Transformer | 26.5 | 17.3M | – | Paper | Code | 2020 | linked, not harvested | report | |||
| 58 | ConvS2S (ensemble) | 26.4 | 54G | – | Paper | Code | 2017 | linked, not harvested | report | |||
| 59 | ResMLP-6 | 26.4 | – | Paper | Code | 2021 | 2 of 7 ran · 5 unverified | report | ||||
| 60 | Average Attention Network | 26.31 | – | Paper | Code | 2018 | 0 of 19 ran · 19 unverified | report | ||||
| 61 | GNMT+RL | 26.3 | – | Paper | Code | 2016 | 23 of 46 ran · 23 unverified | report | ||||
| 62 | SliceNet | 26.1 | – | Paper | Code | 2017 | linked, not harvested | report | ||||
| 63 | Average Attention Network (w/o FFN) | 26.05 | – | Paper | Code | 2018 | 0 of 19 ran · 19 unverified | report | ||||
| 64 | MoE | 26.03 | 24G | – | Paper | Code | 2017 | 4 of 6 ran · 2 unverified | report | |||
| 65 | Average Attention Network (w/o gate) | 25.91 | – | Paper | Code | 2018 | 0 of 19 ran · 19 unverified | report | ||||
| 66 | Adaptively Sparse Transformer (1.5-entmax) | 25.89 | – | Paper | Code | 2019 | 3 of 3 ran · 0 unverified | report | ||||
| 67 | DenseNMT | 25.52 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 68 | GLAT | 25.21 | – | Paper | Code | 2020 | linked, not harvested | report | ||||
| 69 | CMLM+LAT+1 iterations | 25.20 | – | Paper | Code | 2020 | 3 of 5 ran · 2 unverified | report | ||||
| 70 | ConvS2S | 25.16 | 72G | – | Paper | Code | 2017 | linked, not harvested | report | |||
| 71 | ByteNet | 23.75 | – | Paper | Code | 2016 | linked, not harvested | report | ||||
| 72 | FlowSeq-large (NPD n = 30) | 23.64 | – | Paper | Code | 2019 | 1 of 1 ran · 0 unverified | report | ||||
| 73 | FlowSeq-large (NPD n = 15) | 23.14 | – | Paper | Code | 2019 | 1 of 1 ran · 0 unverified | report | ||||
| 74 | FlowSeq-large (IWD n = 15) | 22.94 | – | Paper | Code | 2019 | 1 of 1 ran · 0 unverified | report | ||||
| 75 | Denoising autoencoders (non-autoregressive) | 21.54 | – | Paper | Code | 2018 | 3 of 4 ran · 1 unverified | report | ||||
| 76 | RNN Enc-Dec Att | 20.9 | – | Paper | Code | 2015 | 2 of 8 ran · 6 unverified | report | ||||
| 77 | FlowSeq-large | 20.85 | – | Paper | Code | 2019 | 1 of 1 ran · 0 unverified | report | ||||
| 78 | PBMT | 20.7 | – | – | – | not matched | report | |||||
| 79 | Deep-Att | 20.7 | 119G | – | Paper | Code | 2016 | linked, not harvested | report | |||
| 80 | Phrase Based MT | 20.7 | – | Paper | – | 2015 | no code linked | report | ||||
| 81 | PBSMT + NMT | 20.23 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 82 | NAT +FT + NPD | 19.17 | – | Paper | Code | 2017 | 2 of 2 ran · 0 unverified | report | ||||
| 83 | FlowSeq-base | 18.55 | – | Paper | Code | 2019 | 1 of 1 ran · 0 unverified | report | ||||
| 84 | Seq-KD + Seq-Inter + Word-KD | 18.5 | – | Paper | Code | 2016 | 0 of 2 ran · 2 unverified | report | ||||
| 85 | Unsupervised PBSMT | 17.94 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 86 | NSE-NSE | 17.9 | – | Paper | Code | 2016 | linked, not harvested | report | ||||
| 87 | Unsupervised NMT + Transformer | 17.16 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 88 | SMT + iterative backtranslation (unsupervised) | 14.08 | – | Paper | Code | 2018 | linked, not harvested | report | ||||
| 89 | Reverse RNN Enc-Dec | 14.0 | – | Paper | Code | 2015 | 2 of 8 ran · 6 unverified | report | ||||
| 90 | RNN Enc-Dec | 11.3 | – | Paper | Code | 2015 | 2 of 8 ran · 6 unverified | report | ||||
| 91 | MAT | 29.9 | – | Paper | Code | 2020 | 2 of 3 ran · 1 unverified | report |
All 91 rows shown. 90 link to a paper page on this site; 1 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 45 rows have a graph line, from 32 distinct papers; 38 rows (28 papers) have at least one sample that ran. Counting each paper once: Syntology ran 718 of 1,198 samples; 480 unverified. Separately, 531 of those 1,198 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections