A native-speaker-curated dataset called MIDI (Multilingual Idiom Dataset) is tracking how large language models interpret idiomatic expressions across 18 high-, medium-, and low-resource languages. The dataset tests AI by placing these idioms into both isolated sentences and multi-turn conversations to see if models can successfully distinguish between a phrase's literal meaning and its culturally grounded, figurative translation. This research exposes a major flaw in how we currently evaluate machine intelligence: our models are largely memorizing linguistic patterns rather than achieving true cultural reasoning. While LLMs are highly proficient at mapping figurative translations in high-resource environments, they struggle significantly with low-resource languages and, ironically, have a much harder time recognizing when an idiom is meant entirely literally within a conversation. For AI to become a truly global, empathetic tool, engineers must move past raw data scaling. We need architectures capable of dynamic, context-aware reasoning that can grasp the nuanced, localized intent behind human speech rather than just regurgitating statistical probabilities. Read full paper >> arxiv.org/abs/2606.02147 @bilalelbouardi @salsabilzahirah @Irina_Akishina @Ashwathrao @Paramkrishnaa @tebe5496 @bao_nguyen38224 @AmirHossein1040 @mukhitson @MennaAttia3 @besherhassan @ahmadfauzanhid4 @ttk_kuribayashi @haonanlp #EqualyzAI #ArtificialIntelligence #LLMs #NLP #MachineTranslation #EthicalAI #TechInnovation #GenerativeAI

Why does it matter if an AI Translation tool stores your text to train its model? If an AI model trains on your confidential data, that information can inadvertently resurface. If a user elsewhere in the world prompts the same AI engine with a highly specific query related to your industry or company, elements of your leaked data could theoretically be generated in their results. This isn’t science fiction, major global tech firms have already banned internal use of public AI tools precisely because proprietary source code was leaked this way. Learn More: bolingoconsult.com/translati… #DataPrivacy #MachineTranslation #InformationSecurity #Compliance #AITranslation #KeepTranslate

*Press Release* EC Innovations in collaboration with @JademondDigital, has announced the release of its joint 2026 English-Chinese Simplified Localization Benchmark Report. @EC_L10n #localization #LLMs #machinetranslation slator.ch/HumanAIEnglishChin…

Research published in the International Journal of Information and Communication Technology has taken inspiration from the hunting behavior of the fish-eating bird… dlvr.it/TSzFH2 #MachineTranslation #AI #NaturalLanguageProcessing #QuantumComputing #TranslationAccuracy

Late update, but we had two great talks last month! #MachineTranslation #FBK #NLProc #GenderBias #SpeechSynthesis

In a retrospective article, Jaap van der Meer and Anne-Maj van der Meer detail how @TAUS_Data developed the EPIC Partner Program to train specialized models for distinct domains like legal, clinical trials, and gaming. Read the full story here: buff.ly/uaIBZkc #MultiLingualMedia #LanguageIndustry #TranslationTech #AI #MachineTranslation #TAUS #MultiLingualMedia

@modelfront has announced a new outcome-based pricing model designed to align AI costs with the value delivered through machine translation automation. Explore more: buff.ly/S1DVCVJ #AI #MachineTranslation #Localization #LanguageServices #MultiLingualMedia

🌍 𝐏𝐢 𝐒𝐜𝐡𝐨𝐨𝐥 𝐩𝐫𝐞𝐬𝐞𝐧𝐭𝐬 𝐌𝐞𝐞𝐭𝐰𝐞𝐞𝐧 𝐚𝐭 𝐓𝐀𝐔𝐒 𝟐𝟎𝟐𝟔 𝐖𝐡𝐚𝐭 𝐢𝐟 𝐲𝐨𝐮 𝐜𝐨𝐮𝐥𝐝 𝐬𝐞𝐧𝐝 𝐚 𝐝𝐢𝐠𝐢𝐭𝐚𝐥 𝐚𝐯𝐚𝐭𝐚𝐫 𝐭𝐨 𝐲𝐨𝐮𝐫 𝐧𝐞𝐱𝐭 𝐜𝐚𝐥𝐥? That's not a hypothetical. It's what the Meetween project is building. On 5 June, Pi School's Managing Director Sébastien Bratières will take the stage at the TAUS Massively Multilingual AI Conference in Rome to present Meetween, the EU Horizon Europe project developing technology for multilingual, multimodal AI-powered meetings, where language and culture no longer become barriers. The consortium behind Meetween includes Academic Computer Centre CYFRONET AGH, Fondazione Bruno Kessler - FBK, İstanbul Teknik Üniversitesi, Karlsruher Institut für Technologie (KIT), Zoom, Translated, and TAUS, the organiser of the event. 📍 Rome | 5 June, 11:15 | Day 2 𝐅𝐨𝐥𝐥𝐨𝐰 @MeetweenEU on LinkedIn and X and visit 👉  pischool.link/Meetween #Meetween #TAUS2026 #MultilingualAI #MachineTranslation #EUHorizon #ArtificialIntelligence #NLP #HorizonEurope

One More Thing. We are thrilled to announce that Tencent Hy is officially collaborating with the WMT26 organizers on the Preliminary Video Subtitle Translation Competition! 👉Join here: statmt.org/wmt26/video-subti… Powered by our advanced Hy-MT series models, you can also take part in the General Machine Translation Competition. 📷 Join here: statmt.org/wmt26/translation… Top participants have the chance to win special grand prizes! We warmly invite researchers, developers, AI enthusiasts, and everyone passionate about machine translation to join us. Together, let’s push the boundaries of AI and drive the future of machine translation technology forward! Don’t miss this incredible opportunity — register and participate today! #WMT26 #Hy #MachineTranslation #AIInnovationWe look forward to seeing your submissions!

One More Thing. We are thrilled to announce that Tencent Hy is officially collaborating with the WMT26 organizers on the Preliminary Video Subtitle Translation Competition! 👉Join here: statmt.org/wmt26/video-subti… Powered by our advanced Hy-MT series models, you can also take part in the General Machine Translation Competition. 📷 Join here: statmt.org/wmt26/translation… Top participants have the chance to win special grand prizes! We warmly invite researchers, developers, AI enthusiasts, and everyone passionate about machine translation to join us. Together, let’s push the boundaries of AI and drive the future of machine translation technology forward! Don’t miss this incredible opportunity — register and participate today! #WMT26 #Hy #MachineTranslation #AIInnovationWe look forward to seeing your submissions!

🏝️ At #LREC2026? @luisabentivogli is there too, come say hi and chat about the positions in person! #PhDPosition #MachineTranslation #NLProc #LREC2026 #FBK

1. Was working on a task related to representation of low-resource Indic language (Punjabi) in NLP. I found two good datasets (parallel corpora) for it namely - Samanantar, BPCC - both curated by @ai4bharat (thx for cool dataset). #NLP #MachineTranslation #Punjabi #AI #CSE

This study by Hiba El Oirghi et al., analyzes 4,335 translation pairs across Hausa, Igbo, Swahili, Yoruba, and Zulu using Wikipedia-based data, comparing raw machine translation (MT) outputs with human-edited (HE) versions. It finds a surprising pattern: in several cases, native annotators preferred machine-generated translations over human-edited ones, including 86% preference for MT in Zulu and 62% in Igbo. The findings point to a key tension in low-resource language workflows: human editing does not always guarantee improvement. In many cases, edits introduced issues such as missing information or unnecessary additions that reduced translation quality. This highlights the importance of designing translation systems that are sensitive to context, editor proficiency, and language structure. For African language AI, the real challenge is not just generating translations, but ensuring that any refinement step preserves meaning, fluency, and cultural intent without distortion. Read full paper >> aclanthology.org/2026.africa… @MarineCarpuat @UMBaltimore @MasakhaneNLP @AfricaAIForum @AfricaAI_Summit @Africa_In_EN #EqualyzAI #AfricanNLP #MachineTranslation #Wikipedia #LowResourceLanguages #NLPResearch #AIforAfrica #DigitalInclusion #LanguageTechnology

#machinetranslation 🤣

*Press Release* One AI model is no longer enough. MachineTranslation(dot)com is the only translation tool that runs 20 AI models and gives you the one most of them agree on. Today, MachineTranslation(dot)com expands its AI pool with two new large language models – Aya Expanse 32B by Cohere and MiniMax M2.7 - with @MTTomedes slator.ch/MTAdds2AIModels

We just upgraded our model, and raised the bar for AI translation. Across 16 language pairs, blind testers commissioned by DeepL preferred our translations in 94% of head-to-head tests: • 100% of language pairs vs Google Translate • 100% vs OpenAI GPT 5.2 (high reasoning) • 100% vs Microsoft Translate • 88% vs Google Gemini 3.1 Pro (high thinking) • 81% vs Claude Opus 4.6 (high reasoning) Translation isn’t just about words. It’s about capturing human meaning precisely. That’s what we test for. That’s what DeepL delivers. Discover more about DeepL's translation quality: deepl.com/en/quality #BestAITranslations #DeepLQuality #MachineTranslation #LanguageAI

Breaking new ground for African language technology, a recent paper from Findings of the Association for Computational Linguistics: EACL 2026 introduces a first‑of‑its‑kind translation resource for East Africa. This study by Naome A. Etomi @netori3 . et al., presents AfriMMT‑EA, a large‑scale, multi‑domain machine translation dataset covering 53 East African languages, many with little or no prior digital presence. The authors also fine‑tune and release adapted models (“Safari‑270M” and “Safari‑1B”) that outperform existing multilingual baselines across these languages, addressing historical under‑representation in NLP benchmarks and MT systems. This work reflects the urgent need and real possibility, to build AI that serves all voices, not just a handful of high‑resource languages. We share this mission: expanding linguistic inclusion in AI so that diverse African communities can truly benefit from NLP technologies. Models like AfriMMT‑EA point toward a future where language barriers no longer limit access to information, services, or digital participation. Read the Full Paper > aclanthology.org/2026.findin… @EzemaKelechi @elishaondieki @alfredkondoro1 @nithrobinson @UMNews @UniversityofColorado @university3466 @blackswandata @TechHanyang @uonbi @sartify_co #EqualyzAI #AfricanLanguages #MachineTranslation #InclusiveAI #NLP

A lot of people are terrified of a full machine replacement, but in this episode, Sébastien Bratières points out that viewing it as Human vs. Machine is completely absurd. Apple Podcast: buff.ly/SNPcDUb Spotify: buff.ly/ZzSwReZ #AI #LanguageTech #Localization #FutureOfWork #MachineTranslation #CareerGrowth #TechTrends #MultiLingualMedia

They started with a language that barely exists in digital systems, built a corpus, aligned it with English and turned it into a working translation pipeline. That’s what this paper does, developing an English–Efik corpus and machine translation system, proving that even low-resource languages can power real AI applications when the right data foundation exists. The challenge was never just the model, it was always the data, and that’s the signal. For languages like Efik and many others across Africa, progress doesn’t begin with scaling models. It begins with creating structured, usable datasets from scratch. This is exactly where the work is, at @equalyz_ai. From collecting voice data across local languages to enabling systems that can train them, the focus is the same: Build the data → unlock the models. Once the data exists, everything else becomes possible. Read the full paper here >> - openreview.net/attachment?id… @mb_awak @AJUniversity @Unical1975 @ml_collective @state95307 #EqualyzAI #LanguageAI #LowResourceLanguages #MachineTranslation #AfricanLanguages #DataInfrastructure

Anyway, Google Translate has pulled quite a few stunts—it's hilariously abstract.#GoogleTranslateFails #TranslationGoneWrong #AbstractHumor #AILaughs #TechFails #MachineTranslation #FunnyMistakes #GrokVsGoogle @Google @gork