Translating Millions of Words per Week Is Just the Start for Meta

Translating Millions of Words per Week Is Just a Start for Meta

Iris Orriss is a localization veteran. Originally from Germany, her career has included a 10-year stint with Microsoft and more than a decade at Meta, where she is now VP of Global Experience and International Marketing.

A geologist by training, she joked during her keynote speech at SlatorCon London 2024 that she chose the subject for a STEM degree with “the best international field trips.”

Orriss’ responsibilities run the gamut and cover Meta’s product offerings, from Facebook to WhatsApp and Instagram.

In her experience, AI is best used to scale localization for Meta’s massive needs, while human language experts are essential to ensuring quality.

According to Orriss, about half of the world’s population currently uses Meta products, with the vast majority using them on a daily basis. Nine out of ten users live outside of Meta’s home base, North America, and more than half of Meta’s revenue is generated internationally.

Meta’s products are currently available in 120 languages, and the company translates, on average, more than a million words every day.

And, perhaps paradoxically, as translation technology develops, demand continues to grow, with a rise in content to be translated and localized. This includes user-generated content.

AI translation is already integrated into Meta’s business workflow, Orriss explained. This technology — including Meta’s own Llama software — enables faster and higher-volume translation into more languages. 

“At the end of the day, though, with all this technology, we still have language experts in our teams and we rely on language experts in your companies and in your teams,” Orriss told attendees. “And I don’t think this will go away because even if we get the best machine translation, the cultural context is really important.”

“At the end of the day, though, with all this technology, we still have language experts in our teams and we rely on language experts in your companies and in your teams” — Iris Oriss, VP of Global Experience and International Marketing, Meta

Orriss cited a few examples of how internationalization, the main focus of her work at Meta, means making a product work in a user’s local context. 

One-third of the population in Indonesia uses only one name, she said. So simply translating the registration page, which requires a first and last name, is not enough, as these users will either drop out of the registration flow or fill in their first name twice to circumvent the system. This is knowledge that a human, rather than a machine, can impart and implement.

“Being local is what gets people to use your products and to engage with your products,” Orriss added.

AI Ahead

Ultimately, Orriss believes AI will enable companies to generate local content in target languages, which she hopes will create a better experience for end-users, compared to translated and localized “global” content. 

This means language professionals might see their work shift from translation to review for cultural fit and appropriateness. 

She also thinks speech-to-speech translation (S2ST) technologies could help Meta and other companies better serve the many Internet users who speak languages other than the 120 already localized — particularly for those whose native language does not have a writing system. 

However, she clarified, “I do think it’s going to move a lot slower than we all think.” OpenAI’s showcase of GPT-4o’s S2ST capabilities was limited to a short video demonstrating Italian-English translation, as opposed to longer conversations in languages not dominant on the internet. 

Comparing this paradigm shift to the evolution of internet usage, Orriss said, “I think that as a company, we all have to experiment, to retool and to actually figure out how to do this in a sensible and deliberate way so that it becomes the best possible way of living for all of us.”

For those who missed SlatorCon London 2024 in real-time, recordings will be available via Slator’s Pro and Enterprise plans in due course.