{"url":"/sota/machine-translation-on-wmt2016-english-german","task":{"name":"Machine Translation","url":"/task/machine-translation","note":null},"dataset":{"name":"WMT2016 English-German","url":"/dataset/wmt-2016"},"category":"Natural Language Processing","categories":["Natural Language Processing"],"category_note":null,"description":"**Machine translation** is the task of translating a sentence in a source language to a different target language.\r\n\r\nApproaches 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. \r\n\r\nOne 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. \r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [Google seq2seq](https://github.com/google/seq2seq) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["BLEU score","SacreBLEU"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"BLEU score":"higher","SacreBLEU":null}},"counts":{"rows":12,"rows_with_code":9,"rows_with_paper_page":12,"rows_dated":12,"rows_using_additional_data":1},"rows":[{"rank_in_archive_order":1,"model":"MADL","metrics":{"BLEU score":"40.68"},"uses_additional_data":false,"paper_date":"2019-05-01","paper":"/paper/multi-agent-dual-learning","paper_url":"https://openreview.net/forum?id=HyGhN2A5tm","paper_title":"Multi-Agent Dual Learning","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":2,"model":"Attentional encoder-decoder + BPE","metrics":{"BLEU score":"34.2"},"uses_additional_data":false,"paper_date":"2016-06-09","paper":"/paper/edinburgh-neural-machine-translation-systems","paper_url":"http://arxiv.org/abs/1606.02891v2","paper_title":"Edinburgh Neural Machine Translation Systems for WMT 16","code":"https://github.com/rsennrich/wmt16-scripts","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":0}},{"rank_in_archive_order":3,"model":"Linguistic Input Features","metrics":{"BLEU score":"28.4"},"uses_additional_data":false,"paper_date":"2016-06-09","paper":"/paper/linguistic-input-features-improve-neural","paper_url":"http://arxiv.org/abs/1606.02892v2","paper_title":"Linguistic Input Features Improve Neural Machine Translation","code":"https://github.com/rsennrich/wmt16-scripts","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"DeLighT","metrics":{"BLEU score":"28.0"},"uses_additional_data":false,"paper_date":"2020-08-03","paper":"/paper/delight-very-deep-and-light-weight","paper_url":"https://arxiv.org/abs/2008.00623v2","paper_title":"DeLighT: Deep and Light-weight Transformer","code":"https://github.com/sacmehta/delight","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":3,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":5,"model":"FLAN 137B (zero-shot)","metrics":{"BLEU score":"27.0"},"uses_additional_data":false,"paper_date":"2021-09-03","paper":"/paper/finetuned-language-models-are-zero-shot","paper_url":"https://arxiv.org/abs/2109.01652v5","paper_title":"Finetuned Language Models Are Zero-Shot Learners","code":"https://github.com/hiyouga/llama-efficient-tuning","n_code_links":8,"syntology":{"n_ran":0,"n_unverified":1,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":6,"model":"Transformer","metrics":{"BLEU score":"26.7"},"uses_additional_data":false,"paper_date":"2019-10-09","paper":"/paper/on-the-adequacy-of-untuned-warmup-for","paper_url":"https://arxiv.org/abs/1910.04209v3","paper_title":"On the adequacy of untuned warmup for adaptive optimization","code":"https://github.com/Tony-Y/pytorch_warmup","n_code_links":1,"syntology":{"n_ran":4,"n_unverified":0,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":7,"model":"FLAN 137B (few-shot, k=11)","metrics":{"BLEU score":"26.1"},"uses_additional_data":false,"paper_date":"2021-09-03","paper":"/paper/finetuned-language-models-are-zero-shot","paper_url":"https://arxiv.org/abs/2109.01652v5","paper_title":"Finetuned Language Models Are Zero-Shot Learners","code":"https://github.com/hiyouga/llama-efficient-tuning","n_code_links":8,"syntology":{"n_ran":0,"n_unverified":1,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":8,"model":"BiRNN + GCN (Syn + Sem)","metrics":{"BLEU score":"24.9"},"uses_additional_data":false,"paper_date":"2018-04-23","paper":"/paper/exploiting-semantics-in-neural-machine","paper_url":"https://arxiv.org/abs/1804.08313v2","paper_title":"Exploiting Semantics in Neural Machine Translation with Graph Convolutional Networks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":9,"model":"SMT + iterative backtranslation (unsupervised)","metrics":{"BLEU score":" 18.23"},"uses_additional_data":false,"paper_date":"2018-09-04","paper":"/paper/unsupervised-statistical-machine-translation","paper_url":"http://arxiv.org/abs/1809.01272v1","paper_title":"Unsupervised Statistical Machine Translation","code":"https://github.com/artetxem/vecmap","n_code_links":3,"syntology":null},{"rank_in_archive_order":10,"model":"Unsupervised NMT + weight-sharing","metrics":{"BLEU score":"10.86"},"uses_additional_data":false,"paper_date":"2018-04-24","paper":"/paper/unsupervised-neural-machine-translation-with-1","paper_url":"http://arxiv.org/abs/1804.09057v1","paper_title":"Unsupervised Neural Machine Translation with Weight Sharing","code":"https://github.com/ZhenYangIACAS/unsupervised-NMT","n_code_links":1,"syntology":null},{"rank_in_archive_order":11,"model":"Unsupervised S2S with attention","metrics":{"BLEU score":"9.64"},"uses_additional_data":false,"paper_date":"2017-10-31","paper":"/paper/unsupervised-machine-translation-using","paper_url":"http://arxiv.org/abs/1711.00043v2","paper_title":"Unsupervised Machine Translation Using Monolingual Corpora Only","code":"https://github.com/facebookresearch/MUSE","n_code_links":14,"syntology":{"n_ran":3,"n_unverified":11,"n_samples":14,"n_pointer_only_licence":5}},{"rank_in_archive_order":12,"model":"Exploiting Mono at Scale (single)","metrics":{"SacreBLEU":"40.9"},"uses_additional_data":true,"paper_date":"2019-11-01","paper":"/paper/exploiting-monolingual-data-at-scale-for","paper_url":"https://aclanthology.org/D19-1430","paper_title":"Exploiting Monolingual Data at Scale for Neural Machine Translation","code":null,"n_code_links":0,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 7,081 of the 9,623 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9623,"papers_checked":7081,"papers_extracted_not_yet_verified":217,"boards_without_verdict":27,"papers_not_yet_extracted":2325},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-25T09:33:49+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":6,"rows_with_any_sample_ran":3,"distinct_papers_with_graph_line":5,"distinct_papers_with_any_sample_ran":3,"samples_over_distinct_papers":{"n_ran":9,"n_unverified":15,"n_samples":24,"n_pointer_only_licence":5,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":9,"n_unverified":16,"n_samples":25,"n_pointer_only_licence":5,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}