sponsored content

– Supported by RWS –


Building the Perfect Language Ecosystem


cosystems matter. We live in a world overwhelmed by its constituent elements, and changes in one sector affect everything else downstream. That’s why, more than ever, bringing every consequential ingredient into a single platform saves countless headaches — and that’s exactly what RWS aims to do with its new groundbreaking innovation that combines human and artificial intelligence: Evolve. By bringing all the tools into a single ecosystem, RWS aims to provide clients with the ultimate translation solution, even as the industry resettles following the past few years’ technological earthquakes.   

Andrew Thomas, RWS Vice President of Marketing, has watched language technology struggle to find its footing ever since ChatGPT’s seismic launch. He believes that with Evolve, RWS is setting the foundation for the language solution of the future: the only complete package that bundles every necessary language tool with AI-powered efficiency and human-powered quality assurance (QA).

Putting the System in “Ecosystem”

Evolution means adaptation. And right now, that’s the single defining theme for the language industry. Everyone from individual linguists to the executives of the largest language companies is working overtime to find where large language models (LLMs) fit into the language landscape. According to Thomas, while AI is indeed a game-changer, it’s not a silver-bullet solution — other technologies serve specific purposes better, and without a human expert in the picture, true QA is impossible.

That’s why RWS realized that the language solution of the future isn’t a single product or service — it’s an ecosystem. And that means leveraging the strengths of every technology and linguist under the RWS umbrella.

MultiLingual: Could you tell us more about how you and the team arrived at that conclusion?

Thomas: We rolled out a beta version of Evolve in which we essentially brought together neural machine translation (NMT) and machine translation quality estimation (MTQE), and then added an LLM to do automatic post-editing. The solution translates content, then decides which of those translations are good, bad, or okay. The good is left alone, but the bad or okay is sent to the LLM to generate alternative translations. Once generated, the new translation is reassessed by the MTQE, looping up to three times to achieve the best automatic translation possible.

The clients we were engaging in the beta consistently gave us the feedback, “Wow, this is amazing. But we still need a human translator to ensure quality.” So we evolved Evolve (pardon the pun) to include our language specialists and Trados Enterprise, which provides them with the best user experience for managing large translation volumes.

Assuring Quality at Every Step

It’s no wonder that early Evolve users were concerned about humans staying in the loop. While machine-based language tools have grown leaps and bounds over the past 10 years, and LLMs further expand the possibilities, a human expert is still the ultimate arbiter of QA.

And that’s the beauty of the Evolve solution: It’s perhaps the most elegant distribution of human and machine labor yet achieved for language work. Thanks to RWS’ NMT technology built into Language Weaver, as well as workflow automation from Trados Enterprise, Evolve improves over time at managing clients’ language needs across a staggering array of content types and language pairs. Meanwhile, RWS’ expert linguists bring the creativity, individual expertise, and language nuance to achieve the precise quality needed for any given content type.

The biggest concern many clients have about this new world of AI translation is QA. As you mentioned, they wanted a human in the loop, which Evolve provides. In what other ways does Evolve address the quality concern?

Our Language Weaver NMT has been adaptable for a very long time, which allows organizations to capture quality improvements over time. If content is well-suited for traditional post-editing, then it will translate well in a solution like this. And by the way, you can probably even consider other content types that you wouldn’t necessarily think about for post-editing. Fundamentally, the more you use it, the better it gets, which is why we work with our clients to put all of their translatable content through Evolve.

That said, we quickly realized that people were less interested in buying a product and more interested in buying a solution that combines both the technology and the service, which is basically what we did as we came out of the beta. It started as a product beta; it launched as a service solution.

So now Trados Enterprise supports all of the intake via our customer portal and various content connectors; Language Weaver does all of the machine translation (MT) and MTQE. Then the open-source LLM that we have fine-tuned is leveraged for auto- matic post-editing. 

Finally, of course, our own RWS translators do the remaining review. When human translators are editing in Trados, they use a quality control feature called Smart Review — which is yet another LLM. Rather than emulating a linguist’s decision-making process like our MTQE, it provides general grammatical guidance to translators.

The Human Stays in the Picture

And that goes right back to the most important feedback RWS heard during Evolve’s beta testing: keeping the human in the loop. There’s no doubt that the role of the human translator is shifting — just as it did with the introduction of robust MT tools. However, RWS is committed to deploying AI responsibly, and that means adapting their human linguists to the times. According to Thomas, RWS envisions language professionals shifting toward the role of the language specialist: a trained professional who uses their language and subject matter expertise to deliver the best result for the client.

The upshot? Clients get exactly the translation quality they need with previously unheard-of efficiency. And that’s the magic of working in an ecosystem harmonized to individual needs.

You’ve mentioned that client expectations are shifting from translation quality to outcome quality. How does that fit with the client requests you received to keep humans in the loop?

Clients always want the best of both worlds, and translation quality will always matter. Some of it, but not all, can now be offloaded onto AI. Naturally, businesses then ask how else they can increase quality. The answer is improving whatever outcome was intended for the content originally — for example, call deflection for customer support or increased Net Promoter Score (NPS) for marketing.

At the same time, this shift to outcome quality is exactly analogous to the shift translators must make because of solutions like Evolve. They’re no longer just translators — we’re calling them language specialists. Moving out of pure translation, the new role is about the linguistic expertise they bring to the table, combined with any other subject matter knowledge. In some cases, the subject matter expertise will be more important than the linguistic expertise, and in others, it will be the inverse.

For example, you will want your linguistic experts involved in the training process of any AI initiative. This is clearly more valuable to the business over time, and ought to result in elevating the importance of linguists. But you might not need a full-blown professional translator to do all post-editing work. Instead, you might ask someone who speaks and writes in one language but can read the original language — even if they wouldn’t call themselves a professional translator — to review some content. They’re post-editing because they are experts in the specific content subject matter, not language.

And so, to me, this is another trend. There’s a responsible approach to AI, but the disruption to the market is necessitating the shift of translators’ role from translator to language specialist. And a lot of that is about the quality of the outcome. We’re able to shift to this mindset now because of solutions like Evolve.

Any Client Size, Any Content Type

There are no small fish or apex predators in the ecosystem Evolve enables, and that’s the beauty of it. It’s a remarkably scalable solution, which makes it an appealing choice for companies of all sizes. RWS has always prided itself on its ability to supply language services that fit its customers’ budgets and timetables. With Evolve, the team has doubled down on that commitment to its clients. 

Evolve is advertised as an optimized solution for any organization that has multilingual barriers to overcome. How do you manage that scalability? Does it go right back to the ecosystem’s adaptability?

Exactly, because we can fully optimize this process. Give us everything — don’t worry about what the content type is. We will apply the right combination of technology and people to give you optimized outcomes. Having a solution like this allows us to negotiate volume-based rates with our customers. It’s just like buying in bulk at Costco. We can offer competitive rates that others can’t, as long as there’s that trade-off: you give us the volume, we give you better rates, and we maximize efficiencies for everyone.

I really think this is going to become the new normal. All language service providers that will still be in business a few years from now will have to either adopt third-party solutions — piecing them together and figuring out how to make them work — or they will have to invest in their own technology. Quite bluntly, if you aren’t able to deliver this kind of solution, you’re not competing with us; you’re competing against potential clients using an off-the-shelf LLM that people outside of our industry incorrectly think is good enough.

Shattering the Efficiency Barrier

Depending on the language pair and content type, Evolve pulls off at least a 20% increase in turnaround times and up to 65% efficiency gains. There’s no other way to say it: that’s an amazing achievement. And what makes it all the more impressive is that Evolve ages like fine wine — it only gets better as it learns the client’s content types, requirements, and preferences. What’s more, RWS was among the first to market with its Evolve solution, meaning they’ve also had the most time to refine and expand upon their vision for the future of language work. 

How does RWS do it? According to Thomas, it all goes back to the idea of the ecosystem. An ecosystem is, fundamentally, a complicated network of influences, and when its constituent systems are arranged in balance and working in harmony, new achievements become possible. Because RWS powers Evolve entirely through its in-house tools, it can harmonize its systems to the customer’s exact specifications.

RWS is in the rare position to supply just about any tool needed to get the job done. Can you tell us more about how Evolve’s tools work together to achieve those efficiency numbers you mentioned?

One advantage is that the difference in the quality estimation approach I mentioned before is unique, and we believe it provides greater quality than other approaches.

Second, the solution is delivered via Trados Enterprise. The translator has all the segment history from the automatic post-editing LLM and the NMT, any applicable translation memory or terminology, and AI-powered tools like Smart Review — all within Trados, an editing interface that most translators are already familiar with.

Lastly, we have all the pieces in-house. Everybody else has to go out and put it together, and then they have to maintain those integrations to ensure everything continues to function properly. With Evolve, we believe we’ve found the sweet spot that puts all the pieces together in the right way to create the most optimized outcome.

Related Articles