Google is taking artificial intelligence-powered language technologies beyond text translation, focusing on understanding tone, emotion, context, slang and cultural nuances in conversations. The company’s technologies and products now support the daily interactions of more than 7 billion people in over 300 languages, representing 86% of the world’s population. Google Translate, launched in 2006, has expanded from a few languages to more than 250.
Noting that traditional speech recognition systems lose important elements of communication by first converting speech into text and then back into speech, Google is training models such as Gemini to process audio directly and understand both meaning and manner of speech. The goal is to understand laughter, pauses, overlapping speech and language blends such as Spanglish or Hinglish.
Because dominant languages are disproportionately represented online, the company is working with local communities on open-data partnerships for underrepresented languages. The new Language Explorer tool visualizes LinguaMeta, the world’s largest repository of open-source language data, continuously mapping more than 7,000 spoken, written and signed languages. The Google.org-backed Centre for Digital Language Inclusion and AI Singapore’s Project Aquarium are also helping multilingual tools reach farmers, healthcare workers, teachers and other essential service workers.
TranslateGemma, a family of lightweight open translation models powered by Gemini and trained in 55 languages, was developed for more than 3 billion people who lack access to reliable internet. The on-device system does not require an internet or cloud connection. Google also supports Viamo’s “Ask Viamo Anything” (AVA) service, which brings a Gemini-powered voice assistant to feature phones. In a pilot program in Rwanda, AVA answered more than 2 million questions.
Sign Language-to-Text (SL2T), trained on 50+ sign languages, enables sign-to-text dictation through Gboard and Live Transcribe on the Pixel 11; initially, American Sign Language (ASL) is translated into English. The project aims to improve accessibility for 70 million people who use sign language. In New Zealand, the pronunciation of place names in Google Maps was improved in collaboration with Māori experts. Google’s language technologies also connect more than five billion people across nine platforms, including Search, Android, Chrome, YouTube and Google Play.
Why it matters
This approach is based on translation services going beyond producing word equivalents to also process speech patterns and cultural context, with the aim of reducing what is lost in communication. Lightweight models running on-device remove cloud connectivity as a prerequisite for access for people without reliable internet, while voice assistants brought to standard phones reduce the need for smartphones. Adding sign language support directly to keyboard and transcription tools makes accessibility part of mainstream products rather than a separate service. Open-data partnerships with local communities aim to address the data gap involved in incorporating languages that are underrepresented online into systems. However, the initial availability of sign-to-text translation only from ASL into English leaves open the question of how extensively other sign languages will be supported.