Google is expanding its AI research to better understand the world's diverse languages by moving beyond traditional text translation. By training models like Gemini to process audio directly, the company aims to capture human communication elements such as tone, pacing, and emotion, which are often lost in conventional speech-to-text pipelines.
NewsGoogleGeminiTranslateGemma
Google Expands AI Capabilities to Support Diverse Languages and Cultural Nuance
To support users in low-resource regions or areas with limited connectivity, Google has introduced TranslateGemma. This family of lightweight, on-device translation models is built from Gemini and trained across 55 languages, allowing for high-quality translation without a continuous internet connection.
The company is also addressing accessibility through projects like Sign Language-to-Text (SL2T), which is trained on over 50 sign languages to power dictation features. Additionally, Google is working with local communities to ensure cultural accuracy, such as improving the pronunciation of Māori place names in Google Maps to better reflect local heritage.
Sources
- AI for everyone in every language （Google AI Blog, 2026-09-15）