Recaps

TAUS New York speakers say translation’s biggest problem isn’t tech — it’s people

Is the CAT tool dead? That’s the question Smartling CEO Jack Welde asked to kick off TAUS Global Content Summit New York, held March 21, 2019. For a company that sells computer-assisted translation, it’s also a controversial query.

To clarify, Welde didn’t mean dead as in gone forever or no longer existing — just changed. “What I’m suggesting is that the CAT tool of the future isn’t a feature war of how can I shove more features into a tool to help translators to do more stuff,” he clarified during Q&A, explaining that as the line between human and machine translation blurs, tools will become more streamlined.

The question fit in well with the event’s larger theme, Fixing the Translation Ecosystem. “Naturally if you’re going to be fixing something, that implies that it’s broken,” he said — which Welde doesn’t think is the case. “I’m not sure that it’s broken, but I do think that there are some real, serious macrochanges that are happening that we as an industry need to, one, be aware of and, number two, be ready and willing and able to adapt to.” Specifically, he mentioned increases in content volume, shorter production cycles and the need to connect with end users speaking languages of lesser diffusion. These growing pressures demand more automation than the localization industry does currently — not just for translation, but across the larger content cycle.

Smartling CEO Jack Welde

“When you say ‘automation’ in the translation industry, people think, oh, that’s machine translation, right?” Welde said, arguing that “Automation is actually much, much broader than that.” There will always be a place for human translation, but process automation is essential to localizing neverending content in real time. This isn’t possible without cloud-based repositories and agile translation. Automation isn’t what translation needs for the future; it’s what we need now.

But Georg Kirchner, globalization technology manager for Dell EMC, indicated the problem that needs fixing isn’t the industry’s technology; rather it’s people. On a panel with Julien Didier, vice president of technology at TransPerfect, Kirchner said translators “think if they’re working in isolation they’re really productive” — something he admits might be true but that isn’t in the buyer’s best interest. Working alone or outside a translation management system creates a black box where neither buyer nor language services provider (LSP) can see the quality and production insights required to produce faster and better translations. “What matters when it comes to productivity is really providing more collaborative tools,” said Didier.

CETRA CEO Jiri Stejskal and BrauerTraining instructor Claudia Brauer defended freelancers, sharing how translators’ attitudes toward technology have changed for the better over the last year. But the LSP itself is where Kirchner laid blame — specifically their and buyers’ refusal to automate project management (PM). At Dell EMC, PM makes up roughly 30% of translation workflow — a figure Kirchner said hasn’t changed in ten years despite multiple translation management systems now on the market. He suggested that LSP’s refuse to automate because they all use the same translators and need PM to differentiate — and that translation buyers are afraid to adopt the automation their companies need because then they’d be “out of a job.”

Meanwhile, nonindustry, tech giants like Facebook, Google and others are redefining translation, with or without the industry. Six days after the conference, Facebook chief AI scientist Yann LeCun received a coveted Turing Award for his work in neural MT and every day Google inches closer to human parity.

When asked if too much industry truculence will render today’s LSP moot, panelist Dominique Goldstein, manager of operational excellence for Google’s global localization, refused to comment.

Defending vendors, though, Didier said modern project complexity requires human oversight. From vendor selection to human vs. machine to which MT engine to use, Welde agrees buyers do make a lot of decisions: it’s not uncommon for a single Smartling customer to automate one billion decisions per year — 32 per second. “Humans can’t do that,” Welde said, “We need machines to do that.” — Terena Bell