How Two Meta Employees Helped Build a Translation Tool That Impacts Billions of People

Imagine you spoke a language that only a small percentage of the world’s population uses, and there was a dearth of digital content—books and news articles, for example—translated so that you could read it. This is the challenge that Meta set out to solve with its recent No Language Left Behind (NLLB) project, which was released to the public in June. It’s the first artificial intelligence (AI) model that can translate across 200 languages—and hopefully more in the future.

Meta employees Safiyyah S. and Vedanuj G. were both key members of the NLLB team: Safiyyah is a technical program manager and Vedanuj is a research engineer and the tech lead responsible for publicizing the launch externally. Despite working closely together, they didn’t meet in person until April 2022.

“Vedanuj and I were actually chatting earlier about how it was perfect timing for us to finally meet,” Safiyyah says. “It was just a few months before we were planning to launch, so it was great to have in-person working sessions. Nothing beats being in the same room.”

NLLB is just one example of the cross-functional work that AI and machine learning (ML) employees like Safiyyah and Vedanuj have a chance to do at Meta. As Safiyyah says, “This tone of inclusivity among different disciplines is not something that always happens in engineering-focused efforts, but it was really critical to this type of project and it was embraced by everyone.”

Here, Safiyyah and Vedanuj share how Meta empowers AI and ML employees to carve their own path, the impact of NLLB and the metaverse, and their shared love of trees.

Vedanuj: I was born and raised in India, and I graduated with a degree in computer science. During my senior year I opted for a course in AI, even though AI and machine learning were not familiar topics back then. When it was time to choose my specialization for my master’s at Georgia Tech, ML was one of the options, and it was really interesting to me.

Safiyyah: I was born and partially raised in Atlanta, GA, and lived in a few different states east of the Mississippi. I was in a graduate program for linguistics at Georgia State, which is where I learned about the intersection between linguistics and computational linguistics. I sought out a program where I could combine these two interests and found one at the University of Washington. After graduating, I worked in Seattle at a startup called Voicebox, which builds voice assistants, mostly for cars.

Safiyyah: We didn’t actually know until fairly recently that we had this connection in our history, but it was interesting talking about it. Vedanuj mentioned he liked the greenery in contrast with the more arid area of California he’s in right now. That really resonated with me because one of the things I love about Seattle is that it has the greenery of Atlanta.

Vedanuj: Yeah, I loved Atlanta. I was there for two years doing my master’s. It was the first U.S. city I lived in before moving to California. I come from the northeastern part of India where it rains quite a lot, and in Atlanta I felt closer to my home because it rains there, too—more than in California.

Vedanuj: I first joined as an intern while I was getting my master’s. Initially, I was a bit apprehensive applying to a big company like Meta, but I submitted my resume nonetheless. After an interview on my school’s campus, I was invited on-site to Menlo Park, where I had two coding interviews and a behavioral interview. The rest of the day was quite fun. They took us around and showed us cool things like the ice cream shop and game station. I really enjoyed the whole experience. I completed the internship successfully and then got a full-time offer as a research engineer focusing on machine translation and multimodal understanding.

Safiyyah: My journey to Meta was a bit different. The company was looking to find people with experience in the voice assistant space, and I was invited to a recruitment dinner in Seattle. I joined right before the pandemic as a people manager. Like Vedanuj, I was invited for a day of on-site interviews in Menlo Park. Everyone was very friendly—and when I arrived, there was a big picture of my Facebook profile to welcome me!

Vedanuj: The great thing about Meta is when you join full-time, new hires go through a boot camp where they can try different teams and choose the one that’s a good fit. I chose the computer vision team, and worked on different types of AI. From there, I just kept following my interests, which led to No Language Left Behind, where I’m the tech lead.

Safiyyah: I was looking for a new type of technical challenge within Meta and decided to transition to the technical program manager (TPM) role. During that process, I was talking with people who manage TPMs, learning about projects that had open roles. One of them was NLLB. It was the perfect match because it was technically interesting and a new challenge that was related to human language. In addition, it focused on low-resource languages versus high-resource languages like English. This idea of not leaving any language behind really resonated with me.

Vedanuj: I connected with the inclusiveness of the project, too, because I speak a low-resource language. I shouldn’t be deprived of learning things or watching videos because it doesn’t exist online in my language. I also believe people should be able to show their culture or literature to those who don’t understand.

Vedanuj: Usually in ML, we work on problems that are more focused on high-resource languages like English. But there are billions of people who don’t speak or understand English. That’s why it’s important to increase the reach of this technology. The most recent model we released covers 200 languages. Our goal is to increase it further, including languages that are not written. NLLB enables researchers to build on top of what we have created and expand it to many more languages that are not currently covered by ML technologies.

Safiyyah: Vedanuj’s description of the impact is fantastic. I would add that one of the ways we distinguish ourselves from other research labs in the industry is by focusing on open sourcing. Vedanuj mentioned that researchers are able to build upon the work we’re doing and that's because we made an effort to ensure the data we mined and the models we created were open sourced. Everyone is able to use that to build the next piece of technology. This really opens the door for additional use cases that have a direct impact on people’s lives.

Vedanuj: One thing I really like about Safiyyah’s style of work is when she organizes meetings, she’ll post an agenda beforehand and then send a summary about what happened afterward. It really helps me to have a detailed summary, especially if I miss the meeting. I try to follow her lead and do that for the meetings I organize. I’ve learned a lot from Safiyyah throughout these interactions.

Safiyyah: I would say that Vedanuj is my go-to person for all things modeling related, particularly when it comes to externalization, or publicizing the work we’re doing. We had an engagement to showcase the models for NLLB, and Vedanuj was the person I went to if there were tech issues. He was always ready with answers, and that kind of responsiveness is super helpful during a launch situation where you have tight time windows in which things can happen. We wouldn’t be successful if we didn’t have people like Vedanuj to rely on. So thanks to Vedanuj for that!

Vedanuj: People do sometimes tell me that I can handle pressure.

Vedanuj: From my perspective, the metaverse is going to be one of the next big computing platforms, just the way we use our mobile devices. That’s why I think it’s extremely important to be able to translate different languages. Translation facilitates an inclusive metaverse so people all over the world can actually enjoy the platform. That’s my perspective. Safiyyah, what’s yours?

Safiyyah: I totally agree. The metaverse is one of the things that made me want to join the company. In building the next computing platform, we need to ensure that everyone can access it in their preferred language. Our work will support this inclusivity.

Safiyyah: Go for it, Vedanuj. I’m going to struggle with this one.

Vedanuj: Think of a teacher in a school, who teaches kids about different subjects. My job is similar but instead of teaching students, I’m teaching computers and the topic I teach them is to translate a sentence. For example, if there’s a sentence in French that I don’t understand, then I can ask the computer to say it in a language that I understand. My job is to teach the computer how to do that.

Safiyyah: I would not have anything to say that would be more easily comprehensible than how Vedanuj put it. For my role, I would say it is more as putting the puzzle pieces together to make sure we have the full picture.

Vedanuj: The satisfaction of working and building a product that reaches more than a billion people is an amazing feeling. Meta is a great place for such opportunities, and we get support and freedom from leadership to work on problems that are important and impact people’s lives in a meaningful way.

Safiyyah: I would add that the variety of tech challenges that Meta is trying to solve means there’s always an interesting problem to work on. AI and ML are core to Meta so it’s good to work in the part of the company that is central to its success.