Smartcat+Globalese: Integration case study by Saylon

Saylon, one of Turkey’s leading translation companies, specializes in medical, pharmaceutical and life sciences translations as well as localization of marketing, website and e-learning content into Turkish.

Priding itself as tech-savvy and quality-obsessed, the Saylon team is always on the lookout for the best solutions out there that can help them make managing translation projects as convenient and efficient as possible. It’s been almost six months since Saylon adopted Smartcat as their general translation platform of choice and also became our partner in Turkey. In a recent interview with Smartcat, Saylon’s CEO Kürşat Özel said that Smartcat’s translation automation technologies can help the company “go beyond the current technological trends” in the translation industry, helping to manage suppliers better, save time and increase quality. As the Saylon team continues to incorporate the most forward-thinking ideas into their processes, a little over a month ago they added a powerful productivity tool to their kit. Designed for the special needs of language service providers and content owners, Globalese is an easy-to-use platform for building neural machine translation engines. But what was most important to Saylon was that Globalese could be effortlessly integrated with Smartcat.

Since Saylon has completed numerous projects helping international pharmaceutical companies localize their content into Turkish, they have accumulated extensive text data in English and Turkish related to drug product registration and marketing authorization procedures. According to Orkun Gençoğlu, Project Manager at Saylon, they have tried using Google and Microsoft NMT for their ongoing projects in that field. Both engines are available right in Smartcat, but the post-editing process required too much time. “We needed a more advanced solution and Globalese fit just right, considering we have tons of content stored in our translation memories and this amount continues to grow. The built-in engines cannot make use of that content, nor do they allow you to select a specific terminology for the translation domain. By creating an engine specific to our pharmacy-related projects, specifically files in the CTD (Common Technical Document) format, we had an opportunity to transfer our domain-specific TMs to the engine. By doing so we have managed to train the engine to translate our content more efficiently than you possibly could using any of the publically available MT engines,” says Orkun.

According to Saylon, with Globalese, a series of documents that normally required one week can be translated in two days, including post-editing, which is very cost effective and can give you flexibility on deadlines. However, for now Saylon will only be using Globalese for documents containing more than a thousand words. For smaller projects, it’s easier to translate directly in Smartcat (using the built-in MT engines in some cases) as it takes the same amount of time.

“On average, our translators are capable of translating 2,500–3,500 words per day. After we integrated Smartcat with Globalese, our numbers grew to 5,000–6,000 words per day and we’re working to boost our translators’ efficiency even further by updating the engine with more human-translated data. This would not be possible without integrating Smartcat and Globalese, which was done easily and didn’t cost us a thing.”

Orkun Gençoğlu, Project Manager at Saylon

Also, Smartcat can be used to pre-translate segments with 100% matches as well as those containing only numbers. The MT engine will not affect those segments so you won’t have to post-edit them as they’re already complete. Orkun says that Saylon uses this technique in their workflow and that the rest of the segments will require post-editing; however, through using Smartcat’s advanced CAT editor with its real-time QA mechanism, the glossary feature and other great capabilities the post-editing process is not only simplified, it is much less prone to errors.”

The integration of Smartcat with Globalese did not require any technical competency — the API key was all Saylon needed to get the integration up and running. Both companies’ support teams were ready to help with any issues Saylon might have had. Here’s how it works: The human translator connects to the Smartcat server via Globalese and directly sees the projects created by the project managers in Smartcat. From there, the translator can select the files and use the customized engine to translate them. Once the engine has finished working its magic, the translations are immediately available in Smartcat. In a few weeks, Saylon will begin training their translators in the usage of this new tool.

Gábor Bessenyei is the CEO of MorphoLogic Localisation, the company that developed Globalese. When asked about machine translation, he points out that now is absolutely the time for LSPs to use it in their daily tasks. The cost and time saving factor will differ from user to user and from project to project, he admits, adding that Globalese proves to be most efficient on recurring projects and can even double the productivity of translators in such cases.

“The industry is just at the beginning of the story with Neural MT, and its potential is huge. I believe that neural networks and machine learning are nowadays what steam engines, electricity and internal combustion engines were in the past. They will open new opportunities we couldn’t have imagined before.”

Gábor Bessenyei, CEO of MorphoLogic Localisation

Neural MT opened up machine translation to more languages and dramatically increased the quality of translations in languages like Japanese or Turkish, compared to the days of Statistical MT. On the other hand, SMT engines may still perform better under some circumstances, so it’s not black and white. “Still, it’s pretty clear to me that NMT will take over SMT in the long run”, Gábor argues.

Globalese and Saylon started their relationship in a conventional way, as an MT provider would with an LSP. Smartcat also played a major role, as it can be conveniently used for post-editing the MT output. But soon they decided to expand their cooperation. “We were very happy to hear that Saylon achieved good results, since Turkish has always been among the more poorly-performing languages in the MT world”, he says. “We’re pleased to now have Saylon as our reselling partner in Turkey. The Saylon team aspired to get in touch with other LSPs who are working to add value for their customers instead of simply subcontracting work to freelancers, and we liked that idea a lot.”

Regarding the future of translation in the light of the recent MT advancements, Gábor suggests that the way translators are working will soon change. “Much like in the aviation world, where pilots do not fly planes manually anymore, translators will type less and manage data more, ensuring quality of MT output”, he elaborates.

Terminology work, language data cleanup, post-editing and providing qualified linguistic feedback on the MT output; all of these tasks will eventually transform the translation activities as we know them today. However, these looming changes do not pose a threat to the translating profession.

Instead, automation technologies, such as those available in Smartcat, and the next-level machine translation engines, such as Globalese, will take up routine, repetitive work making the creative part of translating more fun and less stressful. This means that the translator will still be there, although their job will be more about watching the magic happen, while being ready to steer it in the right direction.