Equal rights for all languages - it’s a phrase about which many working in the translation industry have thought long and hard. It’s the idea that led co-founders Spence Green and John DeNero to decide to partner and create Lilt.
That mission is also something that resonates with TAUS, the translation industry organization that aims to help its members improve efficiency and effectiveness of localization strategies. Recently, both Spence and John spoke with TAUS Founder and Director Jaap van der Meer and Director of Events and Member Services Anne-Maj van der Meer about the industry as a whole.
The conversation touched on a number of important points, from the importance of translation to how critical the translator is to the process. Jaap has been in the industry for decades after starting his first translation company in 1980, and he has seen the ebbs and flows of localization over the years.
Throughout the time that Jaap has been in the industry, the more “traditional” translation workflows, like machine translation plus post editing (MTPE), have become the main processes that many companies relied on for localization efforts. However, in more recent years, newer technologies and more advanced workflows have come into play. Human-in-the-loop, for example, has popped up in the localization industry as a new and improved way to tackle content translation on a level not previously seen. This workflow incorporates the human linguist throughout the translation process, working with the machine translation model instead of simply editing the output.
This process is challenging outdated workflows like MTPE. For John and Spence, it’s not enough to simply build machine translation that works well - their challenge, and the challenge that they’ve continued to overcome at Lilt, is building software around the MT systems to make sure that they are used more efficiently and effectively than ever.
But there’s still one big question on the minds of Jaap and Anne-Maj: will the translator become obsolete? The human-in-the-loop approach uses linguists to train the systems, but will the systems eventually outperform the humans involved?
Emphatically, John and Spence say no.
“Companies these days don’t pay for the correct translation - they pay for the preferred translation. Machine translation systems can generate correct translations. But companies are paying for the translation that is consistent with their tone and voice, and they’re only getting more sophisticated,” Spence said.
The widely available machine translation systems like Google Translate have seen varying levels of “success”. John and Jaap point out that, while the platform saw millions of users early on, the quality of its translations was limited. The technology behind the system is highly advanced, but the goal for the industry is about sharing content around the world in its best translations - not just a quick, machine-translated version with questionable accuracy and no tone or voice.
That premise, according to John and Spence, is a big reason why an adaptive, human-in-the-loop approach to translation is so important. Translators aren’t going anywhere anytime soon. No matter how good a machine translation system is, it’s never going to have the context, domain expertise, or sophistication of a human linguist.
Simply put, content can’t just be adequate - it has to be right and sound right. And the only way for that to truly be the case in translation is to rely on a human-in-the-loop process that combines the ingenuity of a human linguist with the power of machine translation.