Twenty years ago it was a machine learning experiment, today it is a tool used monthly by over a billion peoplecapable of translating almost 250 languages into more than 60,000 possible combinations. Google Translate celebrates its twentieth birthday with a gift to users, a new function designed for those who want to learn to really speak a language.
How the new service works
The news is called pronunciation practice – we can translate it as pronunciation practice function – and is already available on the Google Translate Android app, for now in the United States and India, in English, Spanish and Hindi. The mechanism is simple: you type the sentence in your language, choose the one to translate it into, select the “Practice” item and then “Pronunciation”. At that point the app shows a microphone and visually breaks down the sentence into its phonemes, the user pronounces it aloud, the AI analyzes his voice and returns an immediate judgment with precise indications on what to correct, or, in the best cases, an encouraging “Excellent”.
It’s not the only feature dedicated to learning. According to Google, about a third of mobile users use Translate not for on-the-fly translations but to study and practice new languages. Almost half of those who use the “Practice” function every week do so to practice speaking, through interactive scenarios designed to build confidence in real conversation.
Twenty years of numbers
For the twenty years of the service, Google has shared some data that gives a measure of the scale on which Translate operates. Each month, enough volume of text is translated that, if read aloud without interruption, would keep a person busy for the next 12,000 years. The most translated language pair remains English-Spanishalthough English-Indonesian, English-Portuguese and three Indian languages — Hindi, Bengali and Malayalam — also appear in the ranking, reflecting significant growth in connectivity across the subcontinent.
Gemini and real-time translation
The new pronunciation function is part of a broader integration process with the Gemini models. Google has already introduced the feature “Live translate“, which allows real-time audio translations through any pair of earphones, preserving the tone and cadence of the original speaker. Over a third of Live translate sessions exceed five minutes — a sign that people are using it not just for orientation on the go, but for real conversations.
THE’AI has also entered the management of nuancesfor which idioms, local slang and implicit context, elements that literal word-for-word translation cannot capture, and which new generation neural models address with an approach closer to that of a native speaker.
The function pronunciation practice it is not yet available for Italy and nothing has been communicated about it, but we are sure that it will arrive soon, because the expansion to other languages and markets is in the logic of a product that has never stopped growing in twenty years.
Source: blog.google
