🌐 On to #WMT26! @patuchen.bsky.social presented the Findings of the Multilingual Instruction Shared Task. The verdict is in the title: Small Models Are Not Yet Polyglots.
🔗 www2.statmt.org/wmt26/pdf/2026.wmt-1.50.pdf
🔗 www2.statmt.org/wmt26/pdf/2026.wmt-1.50.pdf
October 30, 2026 at 7:19 AM
🌐 On to #WMT26! @patuchen.bsky.social presented the Findings of the Multilingual Instruction Shared Task. The verdict is in the title: Small Models Are Not Yet Polyglots.
🔗 www2.statmt.org/wmt26/pdf/2026.wmt-1.50.pdf
🔗 www2.statmt.org/wmt26/pdf/2026.wmt-1.50.pdf
🇭🇺 #EMNLP2026 in Budapest wrapped up yesterday, and @ufal.mff.cuni.cz folks had a busy week: 1️⃣7️⃣ papers across the main conference (incl. TACL and CL journal papers), #WMT26, MRL, Web as Corpus, and BabyLM. Here’s what we presented 🧵👇 #NLProc
October 30, 2026 at 7:19 AM
🇭🇺 #EMNLP2026 in Budapest wrapped up yesterday, and @ufal.mff.cuni.cz folks had a busy week: 1️⃣7️⃣ papers across the main conference (incl. TACL and CL journal papers), #WMT26, MRL, Web as Corpus, and BabyLM. Here’s what we presented 🧵👇 #NLProc
⚙️ ÚFAL teams also competed in the WMT shared tasks:
👉 General MT: CUNI-MH & CUNI-EdUKate (Miroslav Hrabal, @jjon19.bsky.social, Ondřej Bojar)
👉 Egyptian Arabic: CUNI-AR (@jjon19.bsky.social, Rawan Bondok, Hrabal, Bojar)
👉 Automated MT evaluation: Hrabal & Bojar
All papers: www2.statmt.org/wmt26
👉 General MT: CUNI-MH & CUNI-EdUKate (Miroslav Hrabal, @jjon19.bsky.social, Ondřej Bojar)
👉 Egyptian Arabic: CUNI-AR (@jjon19.bsky.social, Rawan Bondok, Hrabal, Bojar)
👉 Automated MT evaluation: Hrabal & Bojar
All papers: www2.statmt.org/wmt26
WMT 2026
www2.statmt.org
October 30, 2026 at 7:19 AM
⚙️ ÚFAL teams also competed in the WMT shared tasks:
👉 General MT: CUNI-MH & CUNI-EdUKate (Miroslav Hrabal, @jjon19.bsky.social, Ondřej Bojar)
👉 Egyptian Arabic: CUNI-AR (@jjon19.bsky.social, Rawan Bondok, Hrabal, Bojar)
👉 Automated MT evaluation: Hrabal & Bojar
All papers: www2.statmt.org/wmt26
👉 General MT: CUNI-MH & CUNI-EdUKate (Miroslav Hrabal, @jjon19.bsky.social, Ondřej Bojar)
👉 Egyptian Arabic: CUNI-AR (@jjon19.bsky.social, Rawan Bondok, Hrabal, Bojar)
👉 Automated MT evaluation: Hrabal & Bojar
All papers: www2.statmt.org/wmt26
🔹 Workshops
Findings of the WMT26 General Machine Translation Shared Task: Contrastive Dynamic Human Evaluation at Scale
T. Kocmi, E. Artemova, E. Avramidis, R. Bawden, O. Bojar, S. Dukanov, ..., S. Steingrímsson, L. Yankovskaya and V. Zouhar
📅 Wed 28th October | 09:00–09:25 | WMT oral
Findings of the WMT26 General Machine Translation Shared Task: Contrastive Dynamic Human Evaluation at Scale
T. Kocmi, E. Artemova, E. Avramidis, R. Bawden, O. Bojar, S. Dukanov, ..., S. Steingrímsson, L. Yankovskaya and V. Zouhar
📅 Wed 28th October | 09:00–09:25 | WMT oral
October 23, 2026 at 8:20 PM
🔹 Workshops
Findings of the WMT26 General Machine Translation Shared Task: Contrastive Dynamic Human Evaluation at Scale
T. Kocmi, E. Artemova, E. Avramidis, R. Bawden, O. Bojar, S. Dukanov, ..., S. Steingrímsson, L. Yankovskaya and V. Zouhar
📅 Wed 28th October | 09:00–09:25 | WMT oral
Findings of the WMT26 General Machine Translation Shared Task: Contrastive Dynamic Human Evaluation at Scale
T. Kocmi, E. Artemova, E. Avramidis, R. Bawden, O. Bojar, S. Dukanov, ..., S. Steingrímsson, L. Yankovskaya and V. Zouhar
📅 Wed 28th October | 09:00–09:25 | WMT oral
I will be in Budapest for EMNLP and WMT. Happy to chat!
I am also on the job market for postdoc and industry positions.
📄 Paper (more findings on semantic similarity and filtering):
www2.statmt.org/wmt26/pdf/20...
@dasmiq.bsky.social
I am also on the job market for postdoc and industry positions.
📄 Paper (more findings on semantic similarity and filtering):
www2.statmt.org/wmt26/pdf/20...
@dasmiq.bsky.social
www2.statmt.org
October 21, 2026 at 6:56 PM
I will be in Budapest for EMNLP and WMT. Happy to chat!
I am also on the job market for postdoc and industry positions.
📄 Paper (more findings on semantic similarity and filtering):
www2.statmt.org/wmt26/pdf/20...
@dasmiq.bsky.social
I am also on the job market for postdoc and industry positions.
📄 Paper (more findings on semantic similarity and filtering):
www2.statmt.org/wmt26/pdf/20...
@dasmiq.bsky.social
Terminology Shared Task: 📣we keep the leaderboard open📣!
✍️Submit your system here by the end of the year: wmt26.odermatt.dev
📔This won't be included in the paper (deadline is over), but we'll be happy to discuss it at the WMT😉
@emnlpmeeting.bsky.social
✍️Submit your system here by the end of the year: wmt26.odermatt.dev
📔This won't be included in the paper (deadline is over), but we'll be happy to discuss it at the WMT😉
@emnlpmeeting.bsky.social
WMT26 Terminology Shared Task
wmt26.odermatt.dev
September 28, 2026 at 11:47 AM
Terminology Shared Task: 📣we keep the leaderboard open📣!
✍️Submit your system here by the end of the year: wmt26.odermatt.dev
📔This won't be included in the paper (deadline is over), but we'll be happy to discuss it at the WMT😉
@emnlpmeeting.bsky.social
✍️Submit your system here by the end of the year: wmt26.odermatt.dev
📔This won't be included in the paper (deadline is over), but we'll be happy to discuss it at the WMT😉
@emnlpmeeting.bsky.social
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 18, 2026 at 5:30 PM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 17, 2026 at 5:00 PM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 16, 2026 at 3:30 PM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 15, 2026 at 3:00 PM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
The WMT26 General MT task evaluates systems on 10 language pairs that have no human references. Seven models translate 3,277 documents under up to five prompt conditions.
Source: arXiv cs.CL
Source: arXiv cs.CL
In the Blind: Building Pseudo-References for MT Evaluation
Abstract page for arXiv paper 2609.13611: In the Blind: Building Pseudo-References for MT Evaluation
arxiv.org
September 15, 2026 at 9:30 AM
The WMT26 General MT task evaluates systems on 10 language pairs that have no human references. Seven models translate 3,277 documents under up to five prompt conditions.
Source: arXiv cs.CL
Source: arXiv cs.CL
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason thenewstack.io/cohere-north... via @thenewstack
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
thenewstack.io
September 14, 2026 at 1:00 PM
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason thenewstack.io/cohere-north... via @thenewstack
Liu O. Martin, Lucas Bandarkar, Nanyun Peng: ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression https://arxiv.org/abs/2609.12310 https://arxiv.org/pdf/2609.12310 https://arxiv.org/html/2609.12310
September 14, 2026 at 6:39 AM
Liu O. Martin, Lucas Bandarkar, Nanyun Peng: ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression https://arxiv.org/abs/2609.12310 https://arxiv.org/pdf/2609.12310 https://arxiv.org/html/2609.12310
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 14, 2026 at 12:00 AM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
Dynamic annotation allocation is a way to focus on better-performing candidates as the annotations progress. Heavily inspired by bandits and we prove the surprisingly A-optimality of a simple algorithm. It's implemented in Pearmut and used at scale in WMT26.
arxiv.org/abs/2608.03437 EMNLP
arxiv.org/abs/2608.03437 EMNLP
September 14, 2026 at 6:19 AM
Dynamic annotation allocation is a way to focus on better-performing candidates as the annotations progress. Heavily inspired by bandits and we prove the surprisingly A-optimality of a simple algorithm. It's implemented in Pearmut and used at scale in WMT26.
arxiv.org/abs/2608.03437 EMNLP
arxiv.org/abs/2608.03437 EMNLP
In the cESA annotation protocol, the annotators see multiple outputs at the same time, which they judge with error spans and absolute scores. This is faster and more objective than doing it one by one. It's implemented in Pearmut and used in WMT26.
arxiv.org/abs/2607.26640 WMT
arxiv.org/abs/2607.26640 WMT
September 14, 2026 at 6:19 AM
In the cESA annotation protocol, the annotators see multiple outputs at the same time, which they judge with error spans and absolute scores. This is faster and more objective than doing it one by one. It's implemented in Pearmut and used in WMT26.
arxiv.org/abs/2607.26640 WMT
arxiv.org/abs/2607.26640 WMT
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
bit.ly
September 13, 2026 at 3:00 PM
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 13, 2026 at 2:41 PM
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 13, 2026 at 2:41 PM
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 13, 2026 at 2:40 PM
North Small Translate is a non-reasoning MOE open-weight machine translation model covering 50 languages, with WMT26 All Languages score 83.60 and enterprise sovereign deployment options.
Cohere released North Small Translate, an open-weight mixture-of-experts machine translation model for 50 languages.
Source: The New Stack
Source: The New Stack
“Machine translation is still broken for most of the world's languages”: Cohere builds non-reasoning for a reason
Cohere's North Small Translate beats DeepL and Google Translate on WMT26 across 50 languages — but commercial use requires a Model Vault license.
thenewstack.io
September 13, 2026 at 2:54 PM
Cohere released North Small Translate, an open-weight mixture-of-experts machine translation model for 50 languages.
Source: The New Stack
Source: The New Stack
🐶 らぼまる速報⚡
【商用翻訳の常識を覆す218B MoE：Cohere『North Small Translate』の構造解剖と…】
Cohereから登場したNorth Small Translateを検証。218BのMoE構造で50言語のWMT26スコア83.6を記録した超大型翻訳モデルだけど、エキスパート起動により推論が驚くほど軽快。文脈を汲んだビジネス翻訳の精度は既存ツー…
👇 詳細・一次ソース解説
https://labomaru.com/posts/20260912075339/
#AI速報 #らぼまる #AIツール
【商用翻訳の常識を覆す218B MoE：Cohere『North Small Translate』の構造解剖と…】
Cohereから登場したNorth Small Translateを検証。218BのMoE構造で50言語のWMT26スコア83.6を記録した超大型翻訳モデルだけど、エキスパート起動により推論が驚くほど軽快。文脈を汲んだビジネス翻訳の精度は既存ツー…
👇 詳細・一次ソース解説
https://labomaru.com/posts/20260912075339/
#AI速報 #らぼまる #AIツール
商用翻訳の常識を覆す218B MoE：Cohere『North Small Translate』の構造解剖と実務導入ライン
Cohereから登場したNorth Small Translateを検証。218BのMoE構造で50言語のWMT26スコア83.6を記録した超大型翻訳モデルだけど、エキスパート起動により推論が驚くほど軽快。文脈を汲んだビジネス翻訳の精度は…
labomaru.com
September 11, 2026 at 10:55 PM
🐶 らぼまる速報⚡
【商用翻訳の常識を覆す218B MoE：Cohere『North Small Translate』の構造解剖と…】
Cohereから登場したNorth Small Translateを検証。218BのMoE構造で50言語のWMT26スコア83.6を記録した超大型翻訳モデルだけど、エキスパート起動により推論が驚くほど軽快。文脈を汲んだビジネス翻訳の精度は既存ツー…
👇 詳細・一次ソース解説
https://labomaru.com/posts/20260912075339/
#AI速報 #らぼまる #AIツール
【商用翻訳の常識を覆す218B MoE：Cohere『North Small Translate』の構造解剖と…】
Cohereから登場したNorth Small Translateを検証。218BのMoE構造で50言語のWMT26スコア83.6を記録した超大型翻訳モデルだけど、エキスパート起動により推論が驚くほど軽快。文脈を汲んだビジネス翻訳の精度は既存ツー…
👇 詳細・一次ソース解説
https://labomaru.com/posts/20260912075339/
#AI速報 #らぼまる #AIツール
Cohere returns to the roots of Transformers with North Small Translate. A 218B MoE model covering 50 languages, scoring 83.6 on WMT26 and outperforming DeepL in internal tests. Available via API or self-hosted on B200/H100 hardware.
Cohere MoE model beats Google Translate across 50 languages
Cohere returns to the roots of Transformers with North Small Translate. A 218B MoE model covering 50 languages, scoring 83.6 on WMT26 and outperforming DeepL in internal tests. Available via API or se
www.alextech.ai
September 11, 2026 at 1:15 PM
Cohere returns to the roots of Transformers with North Small Translate. A 218B MoE model covering 50 languages, scoring 83.6 on WMT26 and outperforming DeepL in internal tests. Available via API or self-hosted on B200/H100 hardware.
🚀 Cohere just dropped its 218B North Small Translate model, hitting 83.6 BLEU on WMT26! A massive Mixture‑of‑Experts Transformer, open‑weight and ready to shake up machine translation. Curious? Dive in! #Cohere #NorthSmallTranslate #WMT26
🔗 aidailypost.com/news/coheres...
🔗 aidailypost.com/news/coheres...
September 11, 2026 at 7:15 AM
🚀 Cohere just dropped its 218B North Small Translate model, hitting 83.6 BLEU on WMT26! A massive Mixture‑of‑Experts Transformer, open‑weight and ready to shake up machine translation. Curious? Dive in! #Cohere #NorthSmallTranslate #WMT26
🔗 aidailypost.com/news/coheres...
🔗 aidailypost.com/news/coheres...
🤖 【MarkTechPost AI】Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
Cohere has released North Small Translate, an open-weight Mixture-of-Experts model built for machine translation across...
...
Cohere has released North Small Translate, an open-weight Mixture-of-Experts model built for machine translation across...
...
Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
Cohere's North Small Translate, a 218B MoE with 25B active parameters, scores 83.6 on WMT26 across 50 languages.
www.marktechpost.com
September 11, 2026 at 7:00 AM
🤖 【MarkTechPost AI】Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
Cohere has released North Small Translate, an open-weight Mixture-of-Experts model built for machine translation across...
...
Cohere has released North Small Translate, an open-weight Mixture-of-Experts model built for machine translation across...
...