Google’s AI systems now power daily interactions in over 300 languages, covering more than 7 billion people. This reaches 86% of the global population. The milestone marks progress, but it also highlights the work needed to fix decades of bias where technology favoured dominant tongues while ignoring thousands of living dialects.
When Google Translate launched in 2006, the aim was simple: remove barriers between speakers. AI advances have expanded that reach to over 250 languages. Text translation alone does not suffice. Technology must understand how people actually communicate. The focus is on building systems that respect cultural nuance and human language richness. This allows everyone to participate and be understood on their own terms.
Text is no longer enough
Speech recognition systems followed a rigid, multi-step process. Audio became text. That text was processed. Then the system synthesised audio back. This pipeline stripped away the richest parts of human communication. Tone, pacing, emotion, and context were lost.
People do not speak in perfectly neat, grammatical sentences. We laugh, overlap, hesitate, and weave multiple languages together mid-sentence. Spanglish and Hinglish are common examples.
To capture this, the company moved beyond text transcripts to native audio intelligence. Models like Gemini now process audio directly while grasping both sound and intent. The specific efforts include:
- Fluid real-time dialogue tools: Gemini 3.5 Live Translate powers spoken translation across 70 languages and 2,000+ language pairs. It captures code-switching and emotional cues naturally. Gemini 3.5 Transcribe is the most precise speech-to-text model yet. It turns raw audio into polished, formatted text, even in noisy environments or with complex jargon. It powers features like Rambler on Android Gboard. This tool removes filler words, fixes grammar and punctuation, and allows editing or rewriting via voice commands. Users can switch between languages during these tasks.
- The 1,000 Languages Initiative: AI helps break down language barriers at a scale previously unimaginable. Reaching more people in their preferred language means going beyond where AI performs best today. The goal is to support the world’s 1,000 most-spoken languages. The Universal Speech Model, trained on 12 million hours of audio, uses cross-lingual transfer learning. This technique enables models to transfer what they learn from data-rich languages to improve speech understanding in languages with far less training data. Models apply patterns learned from data-rich languages to under-resourced ones.
- Rigorous foundational research: This work builds on 25 years of open research and more than 400 peer-reviewed speech papers. These efforts have helped push the frontier and advance speech models.
Communities guide language data
The web disproportionately represents a few dominant languages. Teaching AI to understand underrepresented languages required a rethink of data gathering. The solution is local grassroots partnerships. This localized approach drove three of the most ambitious open-data partnerships:
- WAXAL (Wolof for “speaking,” pronounced “Wah-hal”): Built with partners including Makerere University and Digital Umuganda, WAXAL is a large-scale, open speech dataset. It covers 27 Sub-Saharan African languages spoken by more than 100 million people across more than 26 countries. The data captures tonal variation and conversational rhythms often missing from traditional datasets.
- Project Vaani: In partnership with the Indian Institute of Science (IISc) and Bhashini, this project maps India’s linguistic diversity. It uses a region-anchored rather than language-anchored approach. This enabled the collection of more than 30,000 hours of speech across 109 languages from more than 155,000 speakers to date.
- Amplify Initiative: The team partnered with more than 1,600 local experts and 20 universities across four continents. Partners included Brazil’s UFMG, India’s IIT Kharagpur, and Uganda’s Makerere University. The group contributed 15,000 multimodal data points capturing local nuance.
Work on open-source language innovation continues through the new tool Language Explorer. This interactive tool visualises LinguaMeta, the world’s largest open-source language data repository. Fast Company recognised the design. The tool continuously maps more than 7,000 spoken, written, and signed languages.
Impact is greatest when innovations reach people who can turn new data and insights into meaningful change in their communities. Google.org-supported efforts, including the Centre for Digital Language Inclusion and AI Singapore’s Project Aquarium, help bring multilingual tools to farmers, healthcare workers, teachers, and other essential community members around the world.
Working without reliable internet
Reliable internet access is still out of reach for more than 3 billion people. Technology is only truly accessible if it works where people live, including areas with limited or intermittent connectivity.
To help address this, the company developed TranslateGemma. This is a family of lightweight open translation models built from Gemini and trained across 55 languages. Because TranslateGemma runs efficiently on-device, high quality translation no longer requires a connection to the cloud or the internet.
Running powerful AI models requires capable hardware, which excludes the hundreds of millions of people still using feature phones in low-resource regions. To bridge this divide, the team supports organisations like Viamo. Viamo powers “Ask Viamo Anything” (AVA), a voice AI assistant that brings the power of Gemini to standard feature phones. Viamo successfully piloted AVA in Rwanda with its existing interactive voice response users. The service has already used Gemini to answer more than 2 million questions.
Designing for accessibility
Language is not just about regional dialects or vocabulary. It is also about the many other ways people communicate. Conventional speech tools frequently fail people with non-standard speech. This forces them to adapt to the technology rather than the other way around.
The work focuses on designing for accessibility from the ground up. One example is Sign Language-to-Text (SL2T). Trained across 50+ sign languages, SL2T powers sign-to-text dictation in Gboard and Live Transcribe on Pixel 11. This starts with American Sign Language (ASL) to English. This is an important first step toward making products more accessible to the 70 million people worldwide who rely on sign language to communicate.
Local pronunciation matters
Details matter in everyday tools like navigation. When a navigation app mispronounces a town or street name, it causes confusion. It can also overlook the cultural heritage of the place.
The belief is that cultural context should be part of how language technology is built. This means working directly with local communities. In New Zealand, for example, the team worked with Māori language experts to improve place name pronunciation in Google Maps. This helped make the experience more accurate and genuinely local. Incorporating culturally authentic pronunciations directly into text-to-speech models ensures technology reflects the language and heritage of the communities it serves.
Expanding human expression
One essential lesson over 20 years of AI language research and development is that technology should never narrow the spectrum of human expression. It should expand it.
Today, language technologies are embedded across the core ecosystem. This connects more than five billion people across nine platforms, including Search, Android, Chrome, YouTube, and Google Play. Scale is only part of the story. The bigger goal is depth and richness. The aim is building systems that grasp context, respect culture, and celebrate the many ways people communicate.
As AI expands to address some of society’s biggest opportunities, the work will continue with local communities. The goal remains to build technology that helps more people communicate, participate, and be understood on their own terms.
What it means
Creators and users in non-dominant language regions gain tools that work offline and on basic hardware. The shift moves from correcting grammar to preserving tone, code-switching, and local pronunciation. Data collection is no longer a top-down process but relies on local partnerships to capture authentic speech patterns.
