趋势解读：Celebrating 20 years of Google Translate，解读最新研究结论

From its beginning as an AI experiment in 2006 to supporting about 250 languages today, Translate has come a long way in two decades. Here’s how 1 billion users use Translate to learn, speak and connect more deeply than ever before. Google Translate is celebrating 20 years of breaking down language barriers for over one billion monthly users. To mark the occasion, you can now use the new "pronunciation practice" tool on Android to refine your speaking skills with instant AI feedback. Try out this feature today to master your delivery in English, Spanish, or Hindi. Google Translate is celebrating 20 years of breaking down language barriers for over one billion monthly users. To mark the occasion, you can now use the new "pronunciation practice" tool on Android to refine your speaking skills with instant AI feedback. Try out this feature today to master your delivery in English, Spanish, or Hindi. Your browser does not support the audio element. Twenty years ago, Google Translate began with a profound mission to help people understand one another, regardless of the language they speak. In the two decades since, we’ve worked to turn the science of language into the magic of connection. What started as a small experiment is now a global tool that helps people every day, from connecting with new people while traveling to learning a new language to support their career. To celebrate two decades of progress, here are 20 things you might not know about Translate and how people are using it around the world. To celebrate our 20th anniversary of Translate, today we’re launching one of our most requested features: pronunciation practice, so you can master your delivery on the Translate app for Android . You can already tap "ask" and "understand" to provide additional context and receive alternatives, and now you can use the new “pronunciation practice” tool, which uses AI to analyze your speech and provide instant feedback — helping you nail the right pronunciation before you start a real-world conversation. This is now available in the U.S. and India in English, Spanish and Hindi. Translate was one of the initial experiments that kickstarted Google’s machine learning work decades ago within Google Research. In 2006, Translate relied on statistical machine learning , and a key part of making more fluent and natural translations was our research into how to maintain much larger-scale and more accurate language models (which capture how often words and short phrases occur) across trillions of words of data. In 2016, we pioneered a massive shift to neural networks to move beyond literal word-for-word translations (building on our research on Sequence-to-Sequence models and our initial work on Tensor Processing Units ), proving that deep learning could work at a global scale to be truly helpful for people. And today we're using AI, our powerful Gemini models and more recent generations of our Tensor Processing Unit hardware to make Translate even more capable and helpful. Translate works for almost 250 languages and more than 60,000 potential language pairs, including endangered and indigenous languages , ensuring more voices are heard as the world becomes more connected. Translation is no longer a standalone task; it is now a fundamental part of how people discover and understand information across the web, and communicate with the world around them. There’s enough text translated across Translate, Search, Lens and Circle to Search every month to keep someone reading out loud 24/7 for the next 12,000 years.