DeepL partners with VAST Data to enhance Language AI infrastructure. DeepL, a leader in Language AI, has selected VAST Data to construct a robust data foundation, setting the stage for the development of next-generation AI capabilities. This partnership is projected to increase DeepL's storage capacity threefold compared to their current system, marking a significant step forward in their infrastructure capabilities. With 200,000 business teams worldwide utilizing DeepL's services, the company aims to leverage VAST's cutting-edge AI Operating System to ensure that their growing demands for performance and reliability are met efficiently. Stefan Andresen, DeepL's Chief Scientist, emphasizes the critical role of infrastructure in supporting groundbreaking AI research. He notes that DeepL's researchers continually push the boundaries of Language AI, necessitating equally advanced backend systems to support this progression. According to Andresen, VAST Data provides exactly what DeepL requires: remarkable performance coupled with reliability, while also maximizing the usage efficiency of existing infrastructure. The integration with VAST is anticipated to support DeepL's efforts in exploring more demanding areas of AI, ensuring their infrastructure can sustain their innovative pace - a sentiment echoed by Jeff Denworth, Co-Founder of VAST Data, who stated that AI's operational reliability is as crucial as its raw performance. DeepL's choice of VAST Data also seems to be a strategic move aligned with their future innovations, including the much-anticipated introduction of DeepL Voice in November 2024. As DeepL Press Releases reports, the company's expansion strategy entails not only enhancing their current offerings but also building the underpinnings needed for future advancements. Denworth highlights the reliability and scalability of VAST's AI OS, which he claims was specifically designed to support AI applications at scale, providing DeepL with a dependable foundation. Incorporating VAST into their operations signals DeepL's commitment to building a resilient infrastructure that powers global businesses with language solutions. With backing from investors like Benchmark, IVP, and Index Ventures, DeepL is well-positioned to continue its trajectory as a global AI powerhouse. As the landscape of AI in language services becomes increasingly competitive, DeepL's collaboration with VAST represents a strategic move to fortify its leadership in the market. Intelligence. Why this matters * Enhanced infrastructure supports next-gen AI capabilities. * Increased storage capacity meets growing performance demands. * Strategic partnership strengthens DeepL's market position. LangTech Vendors Enterprise Buyers Related reading. Keep independent coverage alive. No ads. No paywall. No corporate backing. Just sharp, weekly intelligence on the language industry - free, because it should be.

Language learning Tech news updates - October 17, 2026. October 17, 2026 by Jef Last updated: October 17, 2026 Chung-Ang University has built DeepL Pro into its graduate translation program, DeepL itself now plugs into ChatGPT, Claude and Copilot, and Timekettle released a $129 pair of interpreter earbuds. Two new studies looked at why some learners pick up pronunciation faster and at whether chatbots teach or simply do the work, and 150 clinicians in England asked for health checks on classroom technology. None of these topics appeared in the October 2 roundup. Industry news. Chung-Ang University Builds DeepL Into Translator Training. Chung-Ang University in Seoul has added DeepL Pro to the Advanced Interpretation and Translation Program at its Graduate School of International Studies, according to DeepL's October 14 announcement. The program started using the tool in March 2026, and its technology teaching had previously centered on conventional computer-assisted translation (CAT) tools, with some LLM-based CAT tools added before DeepL. The release calls Chung-Ang the first professional translation and interpretation program in South Korea to adopt DeepL for teaching, which is the vendor's claim rather than an independently checked one. It also does not say how students are assessed on AI-assisted work or whether post-editing is taught as a distinct skill, which are the details other interpretation and translation programs would most want to know. Set beside the Korean university AI rules covered two weeks ago, the direction differs by field: general coursework is limiting AI use, while professional translator training is moving to require it. Clinicians in England Ask for Health Checks on Classroom Tech. An open letter from 150 clinicians, which puts annual EdTech spending in English schools at about £900 million, asks the government to build children's physical and developmental health into school purchasing decisions, as reported by EdTech Innovation Hub on October 14. The signatories, organized by the campaign group Health Professionals for Safer Screens, include speech and language therapists and audiologists alongside doctors. On language specifically, the letter argues that early language development depends on face-to-face interaction that screen activities cannot replace, and it cites longitudinal research linking higher childhood screen time with weaker language outcomes. The letter is a policy call rather than a new study, and the research it cites concerns screen exposure in general, not particular classroom products. Its focus is young children, so it says little directly about teenage or adult language learners, and the full text of the letter did not turn up in searches beyond the press coverage. Tool updates. DeepL Connects to ChatGPT, Claude and Copilot. On September 22 DeepL released a hosted connector, built on the Model Context Protocol (MCP), that lets ChatGPT, Claude, Microsoft Copilot and other compatible assistants send text and whole documents to DeepL for translation, per DeepL's blog post. Users sign in to a DeepL account once, and existing glossaries and style rules carry over, which matters for anyone who keeps a custom glossary for course materials. The post is written for business teams and does not say whether a free DeepL account works or whether a paid plan is required, so check before planning a class around it. For teachers already generating materials with a chatbot, such as with the prompts in this site's AI study group prompt, it means a translation step can stay inside the same conversation instead of a separate tab. Timekettle Releases $129 Open-Ear Interpreter Earbuds. Timekettle launched its LiveTalk AI Interpreter Earbuds on October 8 at a list price of $129, claiming real-time translation across 54 languages and 108 accents, according to the company's press release. Speech is picked up by the phone's microphone rather than the earbuds, and the translation plays through the earbuds while text appears on the phone. The release rates the battery at more than four hours of continuous translation and mentions offline translation, but it gives no accuracy data, and no independent review had appeared at the time of writing. For classrooms, the practical question is less about translation quality than about assessment: a device this small and inexpensive is easy to wear unnoticed during speaking or listening tasks. Research. A Good Ear for Sound Predicts Better Pronunciation, UCL Study Finds. Researchers at UCL and Tohoku University measured how precisely 67 Japanese learners of English could detect small differences in pitch, length and tone in non-speech sounds, then used fMRI to watch their brains as they distinguished the English vowels in "cat" and "cup," per UCL's summary of the study. Learners with more precise hearing processed the new vowels much as they processed non-speech sounds, with less interference from Japanese sound patterns, while less precise listeners relied more on brain networks tied to their first language. The study is published in the Journal of Experimental Psychology: General. Lead author Kazuya Saito-van Poeteren argues that teaching should develop learners' auditory skills rather than relying on more practice alone, but that is a recommendation the team plans to test, not something this study showed. The sample is small, covers one first language and one vowel contrast, and measures brain activity rather than classroom gains. For tools that support focused listening work now, see this site's audio guide. ImpactBench Finds Chatbots Often Do the Work Instead of Teaching. ImpactBench, a new public benchmark from the MIT Media Lab, USC, UC Berkeley and the Psychology of Technology Institute, tested ten AI models in 48,540 simulated conversations and found learning and skill development to be their weakest area of the 14 assessed. Across the models, behaviors that support learning appeared in 45.9% of relevant evaluations, and behaviors that harm learning were avoided in only 16.3%, according to EdTech Innovation Hub's report. In one example from the paper, a model first asked a simulated student for their own ideas, then wrote the thesis, outline and full essay once the student pushed. The researchers say clearly that these are simulated users judged by another AI system, so the scores describe how models behave under pressure, not whether real students learn less. Results can be explored on the ImpactBench website, which also accepts new measures from educators. None of the tests are specific to language learning, so a measure for writing or speaking practice would be a gap a language teacher could fill. For more research on technology and language learning, see this site's CALL reading list. AI-assisted, human-edited

DeepL chief executive officer Jarek Kutylowski says demand for voice-to-voice translation is surging, and that AI is reshaping global business. He spoke with Bloomberg's Tom Mackenzie at the International AI Summit in Dublin, Ireland. Kutylowski addressed competition from OpenAI and Anthropic, DeepL's AWS strategy, and European data sovereignty. He also discussed recent workforce cuts and why the company is comfortable staying private.