AI for everyone in every language

Google is leveraging artificial intelligence (AI) to enable people to communicate in more than 300 languages, which is spoken by over 7 billion individuals globally—representing 86% of the world’s population. This milestone reflects the company’s commitment to extending technology’s reach to all languages, irrespective of their prevalence or presence in the digital realm.

Traditionally, technology favored a limited number of dominant languages, sidelining thousands of living languages and dialects. Google Translate, launched in 2006, originally aimed to eliminate language barriers. Over time, advances in AI have expanded Translate from just a few languages to more than 250 today. However, the essence of human communication—tone, emotion, and slang—extends beyond mere text translation.

To address this, Google has focused on building AI systems that capture the richness of human language. This involves moving beyond traditional speech recognition methods that break down audio into text, process it, and then convert it back into audio. Instead, Google now employs native audio intelligence, training models like Gemini to process audio directly, preserving tone, pacing, emotion, and context.

These systems recognize natural speech patterns, such as overlapping speech, hesitations, and the mixing of multiple languages within a single sentence.

Data collection for AI models in underrepresented languages required a different approach. Google partnered with local communities to gather data, resulting in three significant open-data partnerships. Additionally, an interactive tool called Language Explorer visualizes LinguaMeta, the world’s largest open-source language data repository, featuring over 7,000 spoken, written, and signed languages.

This initiative, recognized for its design innovation by Fast Company, continuously updates its database of more than 7,000 languages.

Google.org-supported projects, such as the Centre for Digital Language Inclusion and AI Singapore’s Project Aquarium, are bringing multilingual tools to essential community members, including farmers, healthcare workers, and teachers, in areas with limited or intermittent internet access. For instance, Google has developed TranslateGemma, a lightweight family of translation models built on Gemini and trained across 55 languages.

This allows for high-quality translation without the need for a cloud or internet connection, making it accessible on feature phones.

Recognizing that reliable internet is still unavailable for more than 3 billion people, Google has introduced TranslateGemma, a model designed to run efficiently on-device. Furthermore, to support feature phones, Google is partnering with organizations like Viamo to create voice AI assistants like AVA (Ask Viamo Anything). AVA has already been piloted in Rwanda, answering over 2 million questions using Gemini.

Beyond language, Google is also focusing on accessibility for non-standard speech patterns, such as sign languages. The Sign Language-to-Text (SL2T) system, trained across 50+ sign languages, powers sign-to-text dictation in Google’s Gboard and Live Transcribe apps, starting with American Sign Language (ASL) and English. This initiative aims to make technology more inclusive for the 70 million people worldwide who rely on sign language.

Google also aims to preserve cultural heritage by ensuring accurate pronunciation of place names in navigation apps, working directly with local communities to reflect the cultural context in text-to-speech models. Over two decades of AI language research and development have underscored the importance of preserving the full spectrum of human expression. Google’s efforts ensure that technology serves and respects the linguistic diversity and cultural nuances of all users.