How a Language Solutions Company Uses Upwork To Scale Multilingual AI

As companies prioritize localization to gain mindshare in a competitive global marketplace, many are doubling down on automated solutions. For Summa Linguae Technologies (SLT), this approach skips an essential piece that machines can’t duplicate.

“Translation is part of the equation, but localization also reflects cultural standards and expectations that are always changing,” said Ian McLaren, marketing content specialist for SLT. “Localization helps products, services, and content feel as though they’ve been created for each individual region, not adapted as an afterthought.”

SLT provides advanced language solutions, leveraging 20 years of experience and global reach made possible by offices in the United States, Canada, Poland, Sweden, Finland, Norway, Denmark, and India into multilingual data solutions that help businesses expand their reach. The company works in 40+ languages, helping clients like Sonos, Nuance, Medallia, and UL collect data to improve AI models, including those used for natural language processing (NLP).

“We deliver high-quality multimodal data—including audio, video, text, and images—to train models, particularly speech-enabled software and devices,” said Katja Krohn, data solutions team lead.

Despite the hype around the machines, it’s actually the humans who are critical to the training cycle—both to generate data and to validate it. Finding the specific talent needed to do the work can be harder than you think.

Fluency is just the beginning

Headquartered in Poland, SLT has seven offices spread across North America, Europe, and Asia. Despite the reach of its distributed team, the company needs to stay flexible. “If we have a collection request for audio in Spanish, for example, we need to find Spanish speakers,” said Krohn. “However, the dialect required could be from Spain or Argentina or Mexico. We can’t have local team members everywhere and we can’t send a team to each location to collect data.”

Instead, SLT uses a proprietary app to record data, which creates a more cohesive process. Even with that consistency, however, quality assurance (QA) is essential. “For all of our data collections—audio, images, whatever else—we need to be confident that the data meets the requirements,” Krohn said.

To start, SLT needs great talent. Over the past 20 years, SLT has built a solid community of independent contractors that they turn to regularly. However, as the nature of client projects has changed, so have SLT’s talent needs—and, sometimes the work description can be pretty granular.

“For example, I don’t speak Hebrew,” said Krohn. “I can’t detect errors, judge whether someone is a native Hebrew speaker or speaks it as a second language, or whether a speaker’s accent is different than the one we need. The QA talent we engage needs to be able to identify this level of detail.”

Language requirements can get even more specific, such as a particular dialect or tone of voice.

“Think about all the differences within languages like English, French, or Arabic—not just accents but common phrases and vocabulary,” Krohn said. “Expand that to include regional dialects, non-native speakers, or people who have a speech impediment. Data points need to encompass all these variations so access can be the same for everyone.”

Leveraging a global work marketplace for multilingual talent

SLT has included Upwork in its talent strategy since 2016. Sometimes, the company engages technical expertise. Other times, it scales translation resources for high-volume projects or to access less-common languages. Most frequently, SLT turns to Upwork for QA and annotation support.

“Upwork’s global aspect, as well as the access to information provided through the site, makes it easier to find talent and get work done,” Krohn said. “For example, I’m in Vancouver and my colleague in Brazil can access the same job post and review the same proposals.”

When engaging talent, Krohn said the team adjusts their approach to meet the needs of each particular project.

“If we’re working in a new language and need a high volume of support, for example, we’ll post an open job post and use the invite feature to share the post and get it moving,” she said. If they already have an independent professional in mind, or if they’re only looking for a couple of people, they’ll create a private job post that’s shared through invite only.

These different strategies help SLT get the volume and caliber of proposals a project calls for.

Human talent is still at the heart of human connection

Even as SLT continues to leverage data and automation, Krohn and McLaren say human involvement remains crucial.

Krohn is proud of the long-term relationships SLT has built with independent professionals on Upwork. “Bringing someone on board and retaining their support over multiple projects—that’s not always a given. These relationships are success stories for me.”

And they’ll continue to be an important part of SLT’s localization strategies. “We know that relying too heavily on machine translation can have an influence in the long run,” McLaren explained. Machines make mistakes and the more obvious and frequent those errors, the less inclined someone is likely to feel toward a product or website.

“In an increasingly automated world, continuing to have those human touch points is a big part of what we’re all about,” McLaren said. SLT is constantly adjusting the balance to deliver effective localization solutions to their clients—and turning to Upwork to find the sweet spot where human talent and automation connect to deliver results.