This is really a new start of taking our technology and our partner technology and rethinking how software internationalization and localization is performed says Adam Asnes, Founder & CEO of Lingoport.
In March, Ken Behan, VP of Sales and Marketing at SYSTRAN joined Adam Asnes, the Founder & CEO of Lingoport, Olivier Libouban, Product Development Head at Lingoport, and Yuka Kurihara, Globalization Consultant, to discuss the undiscovered potential of internationalization and localization they recently unfurled through SYSTRAN and Lingoport’s partnership.
Localization Now & in the Future
People talk about continuous internationalization and localization when actually, it is never continuous. Even after undertaking an updated approach at Lingoport, a lot of external processes remain disjointed with software development. Amidst all the automation being done, developers still had to wait for localization teams to make appropriate adjustments and localization teams had to wait for development teams to institute those changes. Small changes led to large overheads, which were not ideal for getting information localized efficiently and cost-effectively.
Thus, Lingoport decided to rework their approach and go DevOps — incorporating and automating procedural operations into software development so the entire process of internationalization and localization can happen efficiently.
To ensure localization development wouldn’t fall behind, the team at Lingoport used the available technologies to tie the process together, making it a part of the automated system.
It’s actually integrated into the process so that every time there is a new feature or every time there is a new sprint, release, or instance, it’s always part of the action. It’s not a separate thing. And, we also wanted to help users get inputs from various countries. So that was our goal with this next release. And we had to think a bit differently to make that happen, said Adam.
Where We Will Take the Industry with Our New Solution
Generally, to make a process agile in a sprint or an increment of development, everything needs to be done in a typical waterfall model for a select number of features. From the backlog of features ready for implementation, a specific feature is chosen. Then, plans are made around the feature; the feature is designed, built, tested, reviewed, and finally launched.
This process takes nearly two weeks to complete and by the end of the two week period, there will be a product/feature ready for deployment. Though this process will result in a demonstrable system, the system typically will be done in the native language.
SYSTRAN products, however, allow localization to happen alongside product development, testing, and review. This means the final demonstrable product will be available simultaneously in other target locales.
How SYSTRAN + Lingoport Solution Ensure Complete Translation Automation
When a translation system is deployed with a Translation Management System (TMS), it creates a need for the repository to be analyzed for errors. Typically, a set of people are employed at this stage to manually test and design ad hoc solutions. Time-intensive work in this stage includes scriptwriting, manual steps, or emails. This also relies on a localization manager, for instance, Dev manager, to identify the faults in the repository that needs to be translated.
This process is time-consuming and painful as the people involved will have to push the faults into the right form, branch, and directory after validation. Furthermore, as the people involved shoulder many other responsibilities, this tends to delay the process and drags software engineering.
Nevertheless, with SYSTRAN + Lingoport’s new tool, the Lingoport Resource Manager (LRM) is seamlessly integrated with TMS. As soon as the developer writes the source code, the property files, JS files, or .resx files that need to be translated, they are automatically analyzed by the LRM and passed onto SYSTRAN’s translation engine. SYSTRAN’s MT translation engine sends the translated files back to the LRM where it is analyzed and verified another time. Finally, the files that need to be pushed into the repository are pushed in a way that the faults end up in the right repository, the right directory, and the right branch. So, instead of days and weeks, you can analyze results and send any faults to the TMS in minutes.
Improving the MT Engine
User Dictionaries/ Glossary Management
User dictionaries are used for everything from do-not-translate words to company names, and product names. The user dictionary can be customized to include industry-specific terms and words to make sure that a particular word is always translated the way you intend.
Through domain specialization, translation memory is used as training resources to the engine. Every time we receive new projects, clients choose specific domains in which they want to train their engine. For instance, a large conglomerate trained 84 engines in total, with one domain called the “digital” domain that was solely designed for marketing purposes like documentation and web presence. This is an extremely powerful part of the improvement process and it can significantly impact the overall cost of localization.
Neural Fuzzy Adaptation
The translation memory is used as part of the MT process where it takes both the 100 percent matches and the “fuzzy” matches to use them as part of the MT translation process. This leads to accurate results.
Advancement in Neural Machine Translation
SYSTRAN brought the first NMT engines to the market in December of 2018, and we were also the first of our kind to launch NMT engines commercially. From that time onward, we have witnessed a quality improvement from the engines around 15 points. Similarly, the speed has skyrocketed from 50 to 2000 characters per second (CPS). This means, almost one whole page can be translated within a single second. And with support for over 150 languages, there’s not a market today you’ll enter where NMT can’t make the lives of your developers and translators easier, and your development cycle infinitely more efficient.