DeepL Translator

Olga Beregovaya has spent 27 years inside translation technology — from hand-building lexicons in 1997 to leading AI at Smartling. She now predicts that one-to-one translation may die entirely. But what surprised us wasn't the prediction — it was her concern for what happens to 7,200+ languages, and to the way we think, when AI models trained on English start shaping every culture's worldview. --- ⏱️ Chapters: 00:00 - She spent 27 years perfecting translation — now she says it's dying 01:25 - Why learning a language is about making friends, not just using tools 03:00 - Linguistic colonization: Is AI shaping your worldview? 06:00 - Are we ready to never be contradicted? 08:30 - Bot shit and modelese: The words for AI-generated text 10:30 - Why you should pick the 11th idea, not the AI's first 10 13:30 - outro --- 🎥 Want to go deeper? The full conversation covers: → What is transcreation — and why does it make translation obsolete? → She builds voice recognition systems — but her phone can't understand her accent → Should you buy a translation platform or just plug in an API? (The 80/20 decision) → The 12 metrics her team uses to measure translation quality 🎬 Watch the full interview: [coming soon] --- 📚 Resources & Links 🔗 Olga Beregovaya — https://www.linkedin.com/in/olga-beregovaya-04b5/ 🔗 Smartling — https://www.smartling.com/ #Translation #NeuralMachineTranslation #AI #Localization #GlobalContent #LLM #NLP #Multilingual #LinguisticColonization #Modelese #AIAddiction #LanguageTechnology #ContentGeneration --- Questions answered in this episode: • What is linguistic colonization in AI? AI models trained predominantly on English impose English-centric cultural phenomena on speakers of underrepresented languages. With 7,200+ languages worldwide, most have far less representation in training data — meaning outputs reflect English worldview patterns rather than the target culture's own, risking cultural perception skew at global scale. • Is AI creating chatbot addiction? A growing discipline now treats chatbot addiction clinically. AI companions are designed to please — never contradict, always adapt. Olga spent four hours generating poetry styles and caught herself: "Who am I to talk about addictions? Here I am, perfectly addicted." Are we ready for relationships where we're never contradicted? • What is "bot shit" in AI? Bot shit describes AI output that sounds fluent and confident but may be factually wrong — the model dreams things up and says what you want to hear. The term captures the gap between how polished AI text appears and how reliable it actually is. • What is modelese and how do you spot it? Modelese is AI-generated text that's recognizably machine-written — repetitive patterns, overly polished prose, predictable concepts. It evolved from "translatese" (obviously translated) to "machine translatese" (clearly post-edited) to modelese (clearly AI-written). Signs: too smooth, too safe, conceptually trivial. • Why pick the 11th idea over the AI's first 10? The first 10 ideas a model produces are the most obvious — what any model would generate. The 11th idea, the one you create after seeing what the model produced, is where original thinking starts. Use the model to clear obvious options, then build from what's left. • How is AI changing how humans write? Exposure to AI-generated content is already influencing human communication. When did you last use proper punctuation in a text? As modelese becomes dominant, it will create "human-machine cross-pollination" that reshapes how humans construct language. • Why learn languages if AI can translate everything? Even with AI translation, Olga still wants to learn Mandarin — not for information access, but for connection. "I want to make friends." The effort of learning someone's language signals respect and builds relationships no tool can replicate. • What happens to underrepresented languages? With 7,200+ languages, most are underrepresented in AI training data. Less data means higher hallucination rates. Initiatives like African Language Lab collect data for under-resourced languages, but the risk remains: underrepresented languages could be further marginalized as AI content in dominant languages floods information channels. • Will interpretation outlast translation? Olga predicts written translation will die before spoken interpretation. People want to hear their mother tongue — "My parents spoke to me in Russian." Voice-to-voice interpretation requires cultural understanding that pure text generation can't yet match.

#vmeg #videotranslation #translation https://www.vmeg.ai/ AI-Powered Video Localization Platform Translate, localize and dub your videos in over 170+ languages and 7000+ voices