Meta AI: Using translations and multilingual features across platforms

The use of translations in messaging, video, and augmented experiences has become a central part of Meta AI’s roadmap. With integrated models such as NLLB-200 and SeamlessStreaming, the system extends support to over 200 languages, focusing on both real-time communication and offline scenarios. These tools are now embedded in WhatsApp, Facebook Reels, and the Ray-Ban smart-glasses platform, making translation a built-in feature rather than a separate utility.

Translation is built into WhatsApp chat and calls.

WhatsApp integrates multilingual support directly into its Meta AI assistant. Users in 22 countries can converse with the assistant in French, German, Hindi, Italian, Portuguese, and Spanish. Live video calls now feature subtitle overlays with support for seven language pairs, where translations appear within one second of speech. The subtitles are embedded in the view, allowing conversations to flow naturally without switching interfaces. Usage is capped at 60 minutes per day per user, and enterprise accounts can adjust quotas.

Real-time translation uses SeamlessStreaming.

Meta AI’s SeamlessStreaming technology delivers end-to-end speech-to-speech and speech-to-text translation in more than 100 language pairs. With an average latency of around two seconds, it is designed for live conversations where speed is critical. Unlike previous models, SeamlessStreaming can handle low-resource languages with improved accuracy, reducing error rates by over 40 percent compared to older translation systems. In enterprise workspaces, sessions are limited to 50 minutes per day, ensuring performance consistency across users.

NLLB-200 expands text translation at scale.

At the core of Meta AI’s multilingual projects is the No Language Left Behind (NLLB-200) model. This model covers 200 languages in a single network, including many underrepresented languages. In benchmarks, it has shown a 44 percent improvement in accuracy for low-resource languages compared to legacy systems. NLLB-200 powers both text translations in Meta’s apps and automatic subtitling in content such as Reels. This allows viewers to see subtitles in their own language regardless of the language of the original audio.

Model	Languages covered	Latency	Strengths
SeamlessStreaming	100+ pairs	~2 s	Real-time voice, live chat
NLLB-200	200	~1.2 s text	Strong on low-resource coverage
WhatsApp subtitles	7 pairs	~1 s	Live overlay in calls
Smart-glasses packs	4 bidirectional	<300 ms	Offline and on-device

Translation is extended to smart-glasses and Reels.

Meta’s Ray-Ban smart-glasses integrate live translation in offline mode for English, Spanish, French, and Italian. The system runs on-device with latency under 300 milliseconds, supporting up to 20-minute sessions without cloud access. Meanwhile, Reels is testing automatic translation of audio tracks, giving global audiences access to content without the barrier of language. Subtitles are styled to match the app’s visual language, making the feature seamless for the end user.

Controls ensure compliance and privacy.

Enterprise settings provide governance for multilingual features. Translated content can be region-locked to data centers in Europe, North America, or Asia-Pacific. A no-train flag prevents conversations and translations from being stored for model training, reducing compliance risks. Audit logs capture metadata such as source and target languages, latency, and token counts, which can be used for operational monitoring or regulatory audits.

Control	Function
Region lock	Keeps processing within EU, US, or APAC clusters
No-train toggle	Excludes translations from storage
Audit logging	Tracks latency, source/target language, and usage
Terminology glossary API	Upcoming feature to enforce consistent brand terms

Roadmap points to greater multilingual integration.

Meta is expanding its multilingual strategy with several upcoming features. The subtitle overlay in WhatsApp will extend from seven to fifteen language pairs. A glossary API is being developed to allow enterprises to lock specific terms and ensure consistent translations for brand or legal requirements. Another planned feature is automatic speaker diarisation in WhatsApp group calls, which would generate separated subtitles for each speaker. These additions point toward a future where translation is more personalised, structured, and adapted to professional as well as personal use.

____________

FOLLOW US FOR MORE.

DATA STUDIOS

datastudios.org